program(1.0) [buildInfo = dict, tensor>({{"coremlc-component-MIL", "3520.4.1"}, {"coremlc-version", "3520.5.1"}})] { func main(tensor encoder_hidden_states, tensor hidden_states_149_cast_fp16, tensor input_115_cast_fp16, tensor input_117_cast_fp16, tensor input_143_cast_fp16, tensor input_15_cast_fp16, tensor input_169_cast_fp16, tensor input_171_cast_fp16, tensor input_253_cast_fp16, tensor input_35_cast_fp16, tensor input_61_cast_fp16, tensor input_63_cast_fp16, tensor input_7_cast_fp16, tensor input_89_cast_fp16) { tensor cast_2_dtype_0 = const()[name = tensor("cast_2_dtype_0"), val = tensor("fp16")]; tensor add_1_mean_0_to_fp16 = const()[name = tensor("add_1_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; tensor add_1_variance_0_to_fp16 = const()[name = tensor("add_1_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(768)))]; tensor cast_9_dtype_0 = const()[name = tensor("cast_9_dtype_0"), val = tensor("fp16")]; tensor cast_12_dtype_0 = const()[name = tensor("cast_12_dtype_0"), val = tensor("fp16")]; tensor cast_8_dtype_0 = const()[name = tensor("cast_8_dtype_0"), val = tensor("fp16")]; tensor cast_7_dtype_0 = const()[name = tensor("cast_7_dtype_0"), val = tensor("fp16")]; tensor add_15_mean_0_to_fp16 = const()[name = tensor("add_15_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1472)))]; tensor add_15_variance_0_to_fp16 = const()[name = tensor("add_15_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2816)))]; tensor cast_5_dtype_0 = const()[name = tensor("cast_5_dtype_0"), val = tensor("fp16")]; tensor cast_3_dtype_0 = const()[name = tensor("cast_3_dtype_0"), val = tensor("fp16")]; tensor cast_1_dtype_0 = const()[name = tensor("cast_1_dtype_0"), val = tensor("fp16")]; tensor add_27_mean_0_to_fp16 = const()[name = tensor("add_27_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4160)))]; tensor add_27_variance_0_to_fp16 = const()[name = tensor("add_27_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6784)))]; tensor cast_4_dtype_0 = const()[name = tensor("cast_4_dtype_0"), val = tensor("fp16")]; tensor cast_11_dtype_0 = const()[name = tensor("cast_11_dtype_0"), val = tensor("fp16")]; tensor cast_6_dtype_0 = const()[name = tensor("cast_6_dtype_0"), val = tensor("fp16")]; tensor var_6340 = const()[name = tensor("op_6340"), val = tensor(1)]; tensor add_55_mean_0_to_fp16 = const()[name = tensor("add_55_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9408)))]; tensor add_55_variance_0_to_fp16 = const()[name = tensor("add_55_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14592)))]; tensor cast_10_dtype_0 = const()[name = tensor("cast_10_dtype_0"), val = tensor("fp16")]; tensor cast_0_dtype_0 = const()[name = tensor("cast_0_dtype_0"), val = tensor("fp16")]; tensor x_7_pad_type_0 = const()[name = tensor("x_7_pad_type_0"), val = tensor("valid")]; tensor x_7_strides_0 = const()[name = tensor("x_7_strides_0"), val = tensor([1, 1])]; tensor x_7_pad_0 = const()[name = tensor("x_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor x_7_dilations_0 = const()[name = tensor("x_7_dilations_0"), val = tensor([1, 1])]; tensor x_7_groups_0 = const()[name = tensor("x_7_groups_0"), val = tensor(1)]; tensor up_blocks_0_resnets_1_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_1_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19776)))]; tensor up_blocks_0_resnets_1_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_1_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6573440)))]; tensor cast_10 = cast(dtype = cast_10_dtype_0, x = input_253_cast_fp16)[name = tensor("cast_2")]; tensor x_7_cast_fp16 = conv(bias = up_blocks_0_resnets_1_conv_shortcut_bias_to_fp16, dilations = x_7_dilations_0, groups = x_7_groups_0, pad = x_7_pad_0, pad_type = x_7_pad_type_0, strides = x_7_strides_0, weight = up_blocks_0_resnets_1_conv_shortcut_weight_to_fp16, x = cast_10)[name = tensor("x_7_cast_fp16")]; tensor cast_0 = cast(dtype = cast_0_dtype_0, x = hidden_states_149_cast_fp16)[name = tensor("cast_1")]; tensor hidden_states_151_cast_fp16 = add(x = x_7_cast_fp16, y = cast_0)[name = tensor("hidden_states_151_cast_fp16")]; tensor input_267_interleave_0 = const()[name = tensor("input_267_interleave_0"), val = tensor(false)]; tensor cast_6 = cast(dtype = cast_6_dtype_0, x = input_171_cast_fp16)[name = tensor("cast_3")]; tensor input_267_cast_fp16 = concat(axis = var_6340, interleave = input_267_interleave_0, values = (hidden_states_151_cast_fp16, cast_6))[name = tensor("input_267_cast_fp16")]; tensor reshape_124_shape_0 = const()[name = tensor("reshape_124_shape_0"), val = tensor([2, 32, 80, 8, 8])]; tensor reshape_124_cast_fp16 = reshape(shape = reshape_124_shape_0, x = input_267_cast_fp16)[name = tensor("reshape_124_cast_fp16")]; tensor reduce_mean_93_axes_0 = const()[name = tensor("reduce_mean_93_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_93_keep_dims_0 = const()[name = tensor("reduce_mean_93_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_93_cast_fp16 = reduce_mean(axes = reduce_mean_93_axes_0, keep_dims = reduce_mean_93_keep_dims_0, x = reshape_124_cast_fp16)[name = tensor("reduce_mean_93_cast_fp16")]; tensor sub_62_cast_fp16 = sub(x = reshape_124_cast_fp16, y = reduce_mean_93_cast_fp16)[name = tensor("sub_62_cast_fp16")]; tensor square_31_cast_fp16 = square(x = sub_62_cast_fp16)[name = tensor("square_31_cast_fp16")]; tensor reduce_mean_95_axes_0 = const()[name = tensor("reduce_mean_95_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_95_keep_dims_0 = const()[name = tensor("reduce_mean_95_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_95_cast_fp16 = reduce_mean(axes = reduce_mean_95_axes_0, keep_dims = reduce_mean_95_keep_dims_0, x = square_31_cast_fp16)[name = tensor("reduce_mean_95_cast_fp16")]; tensor add_62_y_0_to_fp16 = const()[name = tensor("add_62_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_62_cast_fp16 = add(x = reduce_mean_95_cast_fp16, y = add_62_y_0_to_fp16)[name = tensor("add_62_cast_fp16")]; tensor sqrt_31_cast_fp16 = sqrt(x = add_62_cast_fp16)[name = tensor("sqrt_31_cast_fp16")]; tensor real_div_31_cast_fp16 = real_div(x = sub_62_cast_fp16, y = sqrt_31_cast_fp16)[name = tensor("real_div_31_cast_fp16")]; tensor reshape_125_shape_0 = const()[name = tensor("reshape_125_shape_0"), val = tensor([2, 2560, 8, 8])]; tensor reshape_125_cast_fp16 = reshape(shape = reshape_125_shape_0, x = real_div_31_cast_fp16)[name = tensor("reshape_125_cast_fp16")]; tensor add_63_gamma_0_to_fp16 = const()[name = tensor("add_63_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6576064)))]; tensor add_63_beta_0_to_fp16 = const()[name = tensor("add_63_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6581248)))]; tensor add_63_epsilon_0_to_fp16 = const()[name = tensor("add_63_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_63_cast_fp16 = batch_norm(beta = add_63_beta_0_to_fp16, epsilon = add_63_epsilon_0_to_fp16, gamma = add_63_gamma_0_to_fp16, mean = add_55_mean_0_to_fp16, variance = add_55_variance_0_to_fp16, x = reshape_125_cast_fp16)[name = tensor("add_63_cast_fp16")]; tensor input_271_cast_fp16 = silu(x = add_63_cast_fp16)[name = tensor("input_271_cast_fp16")]; tensor hidden_states_153_pad_type_0 = const()[name = tensor("hidden_states_153_pad_type_0"), val = tensor("custom")]; tensor hidden_states_153_pad_0 = const()[name = tensor("hidden_states_153_pad_0"), val = tensor([1, 1, 1, 1])]; tensor hidden_states_153_strides_0 = const()[name = tensor("hidden_states_153_strides_0"), val = tensor([1, 1])]; tensor hidden_states_153_dilations_0 = const()[name = tensor("hidden_states_153_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_153_groups_0 = const()[name = tensor("hidden_states_153_groups_0"), val = tensor(1)]; tensor up_blocks_0_resnets_2_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6586432)))]; tensor up_blocks_0_resnets_2_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65568896)))]; tensor hidden_states_153_cast_fp16 = conv(bias = up_blocks_0_resnets_2_conv1_bias_to_fp16, dilations = hidden_states_153_dilations_0, groups = hidden_states_153_groups_0, pad = hidden_states_153_pad_0, pad_type = hidden_states_153_pad_type_0, strides = hidden_states_153_strides_0, weight = up_blocks_0_resnets_2_conv1_weight_to_fp16, x = input_271_cast_fp16)[name = tensor("hidden_states_153_cast_fp16")]; tensor temb_25_pad_type_0 = const()[name = tensor("temb_25_pad_type_0"), val = tensor("valid")]; tensor temb_25_strides_0 = const()[name = tensor("temb_25_strides_0"), val = tensor([1, 1])]; tensor temb_25_pad_0 = const()[name = tensor("temb_25_pad_0"), val = tensor([0, 0, 0, 0])]; tensor temb_25_dilations_0 = const()[name = tensor("temb_25_dilations_0"), val = tensor([1, 1])]; tensor temb_25_groups_0 = const()[name = tensor("temb_25_groups_0"), val = tensor(1)]; tensor up_blocks_0_resnets_2_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65571520)))]; tensor up_blocks_0_resnets_2_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68848384)))]; tensor cast_9 = cast(dtype = cast_9_dtype_0, x = input_15_cast_fp16)[name = tensor("cast_12")]; tensor temb_25_cast_fp16 = conv(bias = up_blocks_0_resnets_2_time_emb_proj_bias_to_fp16, dilations = temb_25_dilations_0, groups = temb_25_groups_0, pad = temb_25_pad_0, pad_type = temb_25_pad_type_0, strides = temb_25_strides_0, weight = up_blocks_0_resnets_2_time_emb_proj_weight_to_fp16, x = cast_9)[name = tensor("temb_25_cast_fp16")]; tensor input_275_cast_fp16 = add(x = hidden_states_153_cast_fp16, y = temb_25_cast_fp16)[name = tensor("input_275_cast_fp16")]; tensor reshape_128_shape_0 = const()[name = tensor("reshape_128_shape_0"), val = tensor([2, 32, 40, 8, 8])]; tensor reshape_128_cast_fp16 = reshape(shape = reshape_128_shape_0, x = input_275_cast_fp16)[name = tensor("reshape_128_cast_fp16")]; tensor reduce_mean_96_axes_0 = const()[name = tensor("reduce_mean_96_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_96_keep_dims_0 = const()[name = tensor("reduce_mean_96_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_96_cast_fp16 = reduce_mean(axes = reduce_mean_96_axes_0, keep_dims = reduce_mean_96_keep_dims_0, x = reshape_128_cast_fp16)[name = tensor("reduce_mean_96_cast_fp16")]; tensor sub_64_cast_fp16 = sub(x = reshape_128_cast_fp16, y = reduce_mean_96_cast_fp16)[name = tensor("sub_64_cast_fp16")]; tensor square_32_cast_fp16 = square(x = sub_64_cast_fp16)[name = tensor("square_32_cast_fp16")]; tensor reduce_mean_98_axes_0 = const()[name = tensor("reduce_mean_98_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_98_keep_dims_0 = const()[name = tensor("reduce_mean_98_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_98_cast_fp16 = reduce_mean(axes = reduce_mean_98_axes_0, keep_dims = reduce_mean_98_keep_dims_0, x = square_32_cast_fp16)[name = tensor("reduce_mean_98_cast_fp16")]; tensor add_64_y_0_to_fp16 = const()[name = tensor("add_64_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_64_cast_fp16 = add(x = reduce_mean_98_cast_fp16, y = add_64_y_0_to_fp16)[name = tensor("add_64_cast_fp16")]; tensor sqrt_32_cast_fp16 = sqrt(x = add_64_cast_fp16)[name = tensor("sqrt_32_cast_fp16")]; tensor real_div_32_cast_fp16 = real_div(x = sub_64_cast_fp16, y = sqrt_32_cast_fp16)[name = tensor("real_div_32_cast_fp16")]; tensor reshape_129_shape_0 = const()[name = tensor("reshape_129_shape_0"), val = tensor([2, 1280, 8, 8])]; tensor reshape_129_cast_fp16 = reshape(shape = reshape_129_shape_0, x = real_div_32_cast_fp16)[name = tensor("reshape_129_cast_fp16")]; tensor add_65_gamma_0_to_fp16 = const()[name = tensor("add_65_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68851008)))]; tensor add_65_beta_0_to_fp16 = const()[name = tensor("add_65_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68853632)))]; tensor add_65_epsilon_0_to_fp16 = const()[name = tensor("add_65_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_65_cast_fp16 = batch_norm(beta = add_65_beta_0_to_fp16, epsilon = add_65_epsilon_0_to_fp16, gamma = add_65_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_129_cast_fp16)[name = tensor("add_65_cast_fp16")]; tensor input_279_cast_fp16 = silu(x = add_65_cast_fp16)[name = tensor("input_279_cast_fp16")]; tensor hidden_states_155_pad_type_0 = const()[name = tensor("hidden_states_155_pad_type_0"), val = tensor("custom")]; tensor hidden_states_155_pad_0 = const()[name = tensor("hidden_states_155_pad_0"), val = tensor([1, 1, 1, 1])]; tensor hidden_states_155_strides_0 = const()[name = tensor("hidden_states_155_strides_0"), val = tensor([1, 1])]; tensor hidden_states_155_dilations_0 = const()[name = tensor("hidden_states_155_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_155_groups_0 = const()[name = tensor("hidden_states_155_groups_0"), val = tensor(1)]; tensor up_blocks_0_resnets_2_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68856256)))]; tensor up_blocks_0_resnets_2_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98347520)))]; tensor hidden_states_155_cast_fp16 = conv(bias = up_blocks_0_resnets_2_conv2_bias_to_fp16, dilations = hidden_states_155_dilations_0, groups = hidden_states_155_groups_0, pad = hidden_states_155_pad_0, pad_type = hidden_states_155_pad_type_0, strides = hidden_states_155_strides_0, weight = up_blocks_0_resnets_2_conv2_weight_to_fp16, x = input_279_cast_fp16)[name = tensor("hidden_states_155_cast_fp16")]; tensor x_9_pad_type_0 = const()[name = tensor("x_9_pad_type_0"), val = tensor("valid")]; tensor x_9_strides_0 = const()[name = tensor("x_9_strides_0"), val = tensor([1, 1])]; tensor x_9_pad_0 = const()[name = tensor("x_9_pad_0"), val = tensor([0, 0, 0, 0])]; tensor x_9_dilations_0 = const()[name = tensor("x_9_dilations_0"), val = tensor([1, 1])]; tensor x_9_groups_0 = const()[name = tensor("x_9_groups_0"), val = tensor(1)]; tensor up_blocks_0_resnets_2_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98350144)))]; tensor up_blocks_0_resnets_2_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104903808)))]; tensor x_9_cast_fp16 = conv(bias = up_blocks_0_resnets_2_conv_shortcut_bias_to_fp16, dilations = x_9_dilations_0, groups = x_9_groups_0, pad = x_9_pad_0, pad_type = x_9_pad_type_0, strides = x_9_strides_0, weight = up_blocks_0_resnets_2_conv_shortcut_weight_to_fp16, x = input_267_cast_fp16)[name = tensor("x_9_cast_fp16")]; tensor input_281_cast_fp16 = add(x = x_9_cast_fp16, y = hidden_states_155_cast_fp16)[name = tensor("input_281_cast_fp16")]; tensor input_283_scale_factor_height_0 = const()[name = tensor("input_283_scale_factor_height_0"), val = tensor(0x1p+1)]; tensor input_283_scale_factor_width_0 = const()[name = tensor("input_283_scale_factor_width_0"), val = tensor(0x1p+1)]; tensor input_283_cast_fp16 = upsample_nearest_neighbor(scale_factor_height = input_283_scale_factor_height_0, scale_factor_width = input_283_scale_factor_width_0, x = input_281_cast_fp16)[name = tensor("input_283_cast_fp16")]; tensor hidden_states_157_pad_type_0 = const()[name = tensor("hidden_states_157_pad_type_0"), val = tensor("custom")]; tensor hidden_states_157_pad_0 = const()[name = tensor("hidden_states_157_pad_0"), val = tensor([1, 1, 1, 1])]; tensor hidden_states_157_strides_0 = const()[name = tensor("hidden_states_157_strides_0"), val = tensor([1, 1])]; tensor hidden_states_157_dilations_0 = const()[name = tensor("hidden_states_157_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_157_groups_0 = const()[name = tensor("hidden_states_157_groups_0"), val = tensor(1)]; tensor up_blocks_0_upsamplers_0_conv_weight_to_fp16 = const()[name = tensor("up_blocks_0_upsamplers_0_conv_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104906432)))]; tensor up_blocks_0_upsamplers_0_conv_bias_to_fp16 = const()[name = tensor("up_blocks_0_upsamplers_0_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134397696)))]; tensor hidden_states_157_cast_fp16 = conv(bias = up_blocks_0_upsamplers_0_conv_bias_to_fp16, dilations = hidden_states_157_dilations_0, groups = hidden_states_157_groups_0, pad = hidden_states_157_pad_0, pad_type = hidden_states_157_pad_type_0, strides = hidden_states_157_strides_0, weight = up_blocks_0_upsamplers_0_conv_weight_to_fp16, x = input_283_cast_fp16)[name = tensor("hidden_states_157_cast_fp16")]; tensor var_6522 = const()[name = tensor("op_6522"), val = tensor(1)]; tensor input_285_interleave_0 = const()[name = tensor("input_285_interleave_0"), val = tensor(false)]; tensor cast_11 = cast(dtype = cast_11_dtype_0, x = input_169_cast_fp16)[name = tensor("cast_4")]; tensor input_285_cast_fp16 = concat(axis = var_6522, interleave = input_285_interleave_0, values = (hidden_states_157_cast_fp16, cast_11))[name = tensor("input_285_cast_fp16")]; tensor reshape_132_shape_0 = const()[name = tensor("reshape_132_shape_0"), val = tensor([2, 32, 80, 16, 16])]; tensor reshape_132_cast_fp16 = reshape(shape = reshape_132_shape_0, x = input_285_cast_fp16)[name = tensor("reshape_132_cast_fp16")]; tensor reduce_mean_99_axes_0 = const()[name = tensor("reduce_mean_99_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_99_keep_dims_0 = const()[name = tensor("reduce_mean_99_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_99_cast_fp16 = reduce_mean(axes = reduce_mean_99_axes_0, keep_dims = reduce_mean_99_keep_dims_0, x = reshape_132_cast_fp16)[name = tensor("reduce_mean_99_cast_fp16")]; tensor sub_66_cast_fp16 = sub(x = reshape_132_cast_fp16, y = reduce_mean_99_cast_fp16)[name = tensor("sub_66_cast_fp16")]; tensor square_33_cast_fp16 = square(x = sub_66_cast_fp16)[name = tensor("square_33_cast_fp16")]; tensor reduce_mean_101_axes_0 = const()[name = tensor("reduce_mean_101_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_101_keep_dims_0 = const()[name = tensor("reduce_mean_101_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_101_cast_fp16 = reduce_mean(axes = reduce_mean_101_axes_0, keep_dims = reduce_mean_101_keep_dims_0, x = square_33_cast_fp16)[name = tensor("reduce_mean_101_cast_fp16")]; tensor add_66_y_0_to_fp16 = const()[name = tensor("add_66_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_66_cast_fp16 = add(x = reduce_mean_101_cast_fp16, y = add_66_y_0_to_fp16)[name = tensor("add_66_cast_fp16")]; tensor sqrt_33_cast_fp16 = sqrt(x = add_66_cast_fp16)[name = tensor("sqrt_33_cast_fp16")]; tensor real_div_33_cast_fp16 = real_div(x = sub_66_cast_fp16, y = sqrt_33_cast_fp16)[name = tensor("real_div_33_cast_fp16")]; tensor reshape_133_shape_0 = const()[name = tensor("reshape_133_shape_0"), val = tensor([2, 2560, 16, 16])]; tensor reshape_133_cast_fp16 = reshape(shape = reshape_133_shape_0, x = real_div_33_cast_fp16)[name = tensor("reshape_133_cast_fp16")]; tensor add_67_gamma_0_to_fp16 = const()[name = tensor("add_67_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134400320)))]; tensor add_67_beta_0_to_fp16 = const()[name = tensor("add_67_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134405504)))]; tensor add_67_epsilon_0_to_fp16 = const()[name = tensor("add_67_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_67_cast_fp16 = batch_norm(beta = add_67_beta_0_to_fp16, epsilon = add_67_epsilon_0_to_fp16, gamma = add_67_gamma_0_to_fp16, mean = add_55_mean_0_to_fp16, variance = add_55_variance_0_to_fp16, x = reshape_133_cast_fp16)[name = tensor("add_67_cast_fp16")]; tensor input_289_cast_fp16 = silu(x = add_67_cast_fp16)[name = tensor("input_289_cast_fp16")]; tensor hidden_states_159_pad_type_0 = const()[name = tensor("hidden_states_159_pad_type_0"), val = tensor("custom")]; tensor hidden_states_159_pad_0 = const()[name = tensor("hidden_states_159_pad_0"), val = tensor([1, 1, 1, 1])]; tensor hidden_states_159_strides_0 = const()[name = tensor("hidden_states_159_strides_0"), val = tensor([1, 1])]; tensor hidden_states_159_dilations_0 = const()[name = tensor("hidden_states_159_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_159_groups_0 = const()[name = tensor("hidden_states_159_groups_0"), val = tensor(1)]; tensor up_blocks_1_resnets_0_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134410688)))]; tensor up_blocks_1_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(193393152)))]; tensor hidden_states_159_cast_fp16 = conv(bias = up_blocks_1_resnets_0_conv1_bias_to_fp16, dilations = hidden_states_159_dilations_0, groups = hidden_states_159_groups_0, pad = hidden_states_159_pad_0, pad_type = hidden_states_159_pad_type_0, strides = hidden_states_159_strides_0, weight = up_blocks_1_resnets_0_conv1_weight_to_fp16, x = input_289_cast_fp16)[name = tensor("hidden_states_159_cast_fp16")]; tensor temb_27_pad_type_0 = const()[name = tensor("temb_27_pad_type_0"), val = tensor("valid")]; tensor temb_27_strides_0 = const()[name = tensor("temb_27_strides_0"), val = tensor([1, 1])]; tensor temb_27_pad_0 = const()[name = tensor("temb_27_pad_0"), val = tensor([0, 0, 0, 0])]; tensor temb_27_dilations_0 = const()[name = tensor("temb_27_dilations_0"), val = tensor([1, 1])]; tensor temb_27_groups_0 = const()[name = tensor("temb_27_groups_0"), val = tensor(1)]; tensor up_blocks_1_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(193395776)))]; tensor up_blocks_1_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(196672640)))]; tensor temb_27_cast_fp16 = conv(bias = up_blocks_1_resnets_0_time_emb_proj_bias_to_fp16, dilations = temb_27_dilations_0, groups = temb_27_groups_0, pad = temb_27_pad_0, pad_type = temb_27_pad_type_0, strides = temb_27_strides_0, weight = up_blocks_1_resnets_0_time_emb_proj_weight_to_fp16, x = cast_9)[name = tensor("temb_27_cast_fp16")]; tensor input_293_cast_fp16 = add(x = hidden_states_159_cast_fp16, y = temb_27_cast_fp16)[name = tensor("input_293_cast_fp16")]; tensor reshape_136_shape_0 = const()[name = tensor("reshape_136_shape_0"), val = tensor([2, 32, 40, 16, 16])]; tensor reshape_136_cast_fp16 = reshape(shape = reshape_136_shape_0, x = input_293_cast_fp16)[name = tensor("reshape_136_cast_fp16")]; tensor reduce_mean_102_axes_0 = const()[name = tensor("reduce_mean_102_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_102_keep_dims_0 = const()[name = tensor("reduce_mean_102_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_102_cast_fp16 = reduce_mean(axes = reduce_mean_102_axes_0, keep_dims = reduce_mean_102_keep_dims_0, x = reshape_136_cast_fp16)[name = tensor("reduce_mean_102_cast_fp16")]; tensor sub_68_cast_fp16 = sub(x = reshape_136_cast_fp16, y = reduce_mean_102_cast_fp16)[name = tensor("sub_68_cast_fp16")]; tensor square_34_cast_fp16 = square(x = sub_68_cast_fp16)[name = tensor("square_34_cast_fp16")]; tensor reduce_mean_104_axes_0 = const()[name = tensor("reduce_mean_104_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_104_keep_dims_0 = const()[name = tensor("reduce_mean_104_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_104_cast_fp16 = reduce_mean(axes = reduce_mean_104_axes_0, keep_dims = reduce_mean_104_keep_dims_0, x = square_34_cast_fp16)[name = tensor("reduce_mean_104_cast_fp16")]; tensor add_68_y_0_to_fp16 = const()[name = tensor("add_68_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_68_cast_fp16 = add(x = reduce_mean_104_cast_fp16, y = add_68_y_0_to_fp16)[name = tensor("add_68_cast_fp16")]; tensor sqrt_34_cast_fp16 = sqrt(x = add_68_cast_fp16)[name = tensor("sqrt_34_cast_fp16")]; tensor real_div_34_cast_fp16 = real_div(x = sub_68_cast_fp16, y = sqrt_34_cast_fp16)[name = tensor("real_div_34_cast_fp16")]; tensor reshape_137_shape_0 = const()[name = tensor("reshape_137_shape_0"), val = tensor([2, 1280, 16, 16])]; tensor reshape_137_cast_fp16 = reshape(shape = reshape_137_shape_0, x = real_div_34_cast_fp16)[name = tensor("reshape_137_cast_fp16")]; tensor add_69_gamma_0_to_fp16 = const()[name = tensor("add_69_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(196675264)))]; tensor add_69_beta_0_to_fp16 = const()[name = tensor("add_69_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(196677888)))]; tensor add_69_epsilon_0_to_fp16 = const()[name = tensor("add_69_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_69_cast_fp16 = batch_norm(beta = add_69_beta_0_to_fp16, epsilon = add_69_epsilon_0_to_fp16, gamma = add_69_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_137_cast_fp16)[name = tensor("add_69_cast_fp16")]; tensor input_297_cast_fp16 = silu(x = add_69_cast_fp16)[name = tensor("input_297_cast_fp16")]; tensor hidden_states_161_pad_type_0 = const()[name = tensor("hidden_states_161_pad_type_0"), val = tensor("custom")]; tensor hidden_states_161_pad_0 = const()[name = tensor("hidden_states_161_pad_0"), val = tensor([1, 1, 1, 1])]; tensor hidden_states_161_strides_0 = const()[name = tensor("hidden_states_161_strides_0"), val = tensor([1, 1])]; tensor hidden_states_161_dilations_0 = const()[name = tensor("hidden_states_161_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_161_groups_0 = const()[name = tensor("hidden_states_161_groups_0"), val = tensor(1)]; tensor up_blocks_1_resnets_0_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(196680512)))]; tensor up_blocks_1_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(226171776)))]; tensor hidden_states_161_cast_fp16 = conv(bias = up_blocks_1_resnets_0_conv2_bias_to_fp16, dilations = hidden_states_161_dilations_0, groups = hidden_states_161_groups_0, pad = hidden_states_161_pad_0, pad_type = hidden_states_161_pad_type_0, strides = hidden_states_161_strides_0, weight = up_blocks_1_resnets_0_conv2_weight_to_fp16, x = input_297_cast_fp16)[name = tensor("hidden_states_161_cast_fp16")]; tensor x_11_pad_type_0 = const()[name = tensor("x_11_pad_type_0"), val = tensor("valid")]; tensor x_11_strides_0 = const()[name = tensor("x_11_strides_0"), val = tensor([1, 1])]; tensor x_11_pad_0 = const()[name = tensor("x_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor x_11_dilations_0 = const()[name = tensor("x_11_dilations_0"), val = tensor([1, 1])]; tensor x_11_groups_0 = const()[name = tensor("x_11_groups_0"), val = tensor(1)]; tensor up_blocks_1_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(226174400)))]; tensor up_blocks_1_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(232728064)))]; tensor x_11_cast_fp16 = conv(bias = up_blocks_1_resnets_0_conv_shortcut_bias_to_fp16, dilations = x_11_dilations_0, groups = x_11_groups_0, pad = x_11_pad_0, pad_type = x_11_pad_type_0, strides = x_11_strides_0, weight = up_blocks_1_resnets_0_conv_shortcut_weight_to_fp16, x = input_285_cast_fp16)[name = tensor("x_11_cast_fp16")]; tensor hidden_states_163_cast_fp16 = add(x = x_11_cast_fp16, y = hidden_states_161_cast_fp16)[name = tensor("hidden_states_163_cast_fp16")]; tensor reshape_140_shape_0 = const()[name = tensor("reshape_140_shape_0"), val = tensor([2, 32, 40, 16, 16])]; tensor reshape_140_cast_fp16 = reshape(shape = reshape_140_shape_0, x = hidden_states_163_cast_fp16)[name = tensor("reshape_140_cast_fp16")]; tensor reduce_mean_105_axes_0 = const()[name = tensor("reduce_mean_105_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_105_keep_dims_0 = const()[name = tensor("reduce_mean_105_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_105_cast_fp16 = reduce_mean(axes = reduce_mean_105_axes_0, keep_dims = reduce_mean_105_keep_dims_0, x = reshape_140_cast_fp16)[name = tensor("reduce_mean_105_cast_fp16")]; tensor sub_70_cast_fp16 = sub(x = reshape_140_cast_fp16, y = reduce_mean_105_cast_fp16)[name = tensor("sub_70_cast_fp16")]; tensor square_35_cast_fp16 = square(x = sub_70_cast_fp16)[name = tensor("square_35_cast_fp16")]; tensor reduce_mean_107_axes_0 = const()[name = tensor("reduce_mean_107_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_107_keep_dims_0 = const()[name = tensor("reduce_mean_107_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_107_cast_fp16 = reduce_mean(axes = reduce_mean_107_axes_0, keep_dims = reduce_mean_107_keep_dims_0, x = square_35_cast_fp16)[name = tensor("reduce_mean_107_cast_fp16")]; tensor add_70_y_0_to_fp16 = const()[name = tensor("add_70_y_0_to_fp16"), val = tensor(0x1.1p-20)]; tensor add_70_cast_fp16 = add(x = reduce_mean_107_cast_fp16, y = add_70_y_0_to_fp16)[name = tensor("add_70_cast_fp16")]; tensor sqrt_35_cast_fp16 = sqrt(x = add_70_cast_fp16)[name = tensor("sqrt_35_cast_fp16")]; tensor real_div_35_cast_fp16 = real_div(x = sub_70_cast_fp16, y = sqrt_35_cast_fp16)[name = tensor("real_div_35_cast_fp16")]; tensor reshape_141_shape_0 = const()[name = tensor("reshape_141_shape_0"), val = tensor([2, 1280, 16, 16])]; tensor reshape_141_cast_fp16 = reshape(shape = reshape_141_shape_0, x = real_div_35_cast_fp16)[name = tensor("reshape_141_cast_fp16")]; tensor add_71_gamma_0_to_fp16 = const()[name = tensor("add_71_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(232730688)))]; tensor add_71_beta_0_to_fp16 = const()[name = tensor("add_71_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(232733312)))]; tensor add_71_epsilon_0_to_fp16 = const()[name = tensor("add_71_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_71_cast_fp16 = batch_norm(beta = add_71_beta_0_to_fp16, epsilon = add_71_epsilon_0_to_fp16, gamma = add_71_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_141_cast_fp16)[name = tensor("add_71_cast_fp16")]; tensor hidden_states_165_pad_type_0 = const()[name = tensor("hidden_states_165_pad_type_0"), val = tensor("valid")]; tensor hidden_states_165_strides_0 = const()[name = tensor("hidden_states_165_strides_0"), val = tensor([1, 1])]; tensor hidden_states_165_pad_0 = const()[name = tensor("hidden_states_165_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_165_dilations_0 = const()[name = tensor("hidden_states_165_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_165_groups_0 = const()[name = tensor("hidden_states_165_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(232735936)))]; tensor up_blocks_1_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236012800)))]; tensor hidden_states_165_cast_fp16 = conv(bias = up_blocks_1_attentions_0_proj_in_bias_to_fp16, dilations = hidden_states_165_dilations_0, groups = hidden_states_165_groups_0, pad = hidden_states_165_pad_0, pad_type = hidden_states_165_pad_type_0, strides = hidden_states_165_strides_0, weight = up_blocks_1_attentions_0_proj_in_weight_to_fp16, x = add_71_cast_fp16)[name = tensor("hidden_states_165_cast_fp16")]; tensor var_6607 = const()[name = tensor("op_6607"), val = tensor([2, 1280, 1, 256])]; tensor inputs_43_cast_fp16 = reshape(shape = var_6607, x = hidden_states_165_cast_fp16)[name = tensor("inputs_43_cast_fp16")]; tensor hidden_states_167_axes_0 = const()[name = tensor("hidden_states_167_axes_0"), val = tensor([1])]; tensor hidden_states_167_gamma_0_to_fp16 = const()[name = tensor("hidden_states_167_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236015424)))]; tensor hidden_states_167_beta_0_to_fp16 = const()[name = tensor("hidden_states_167_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236018048)))]; tensor var_6623_to_fp16 = const()[name = tensor("op_6623_to_fp16"), val = tensor(0x1.5p-17)]; tensor hidden_states_167_cast_fp16 = layer_norm(axes = hidden_states_167_axes_0, beta = hidden_states_167_beta_0_to_fp16, epsilon = var_6623_to_fp16, gamma = hidden_states_167_gamma_0_to_fp16, x = inputs_43_cast_fp16)[name = tensor("hidden_states_167_cast_fp16")]; tensor q_29_pad_type_0 = const()[name = tensor("q_29_pad_type_0"), val = tensor("valid")]; tensor q_29_strides_0 = const()[name = tensor("q_29_strides_0"), val = tensor([1, 1])]; tensor q_29_pad_0 = const()[name = tensor("q_29_pad_0"), val = tensor([0, 0, 0, 0])]; tensor q_29_dilations_0 = const()[name = tensor("q_29_dilations_0"), val = tensor([1, 1])]; tensor q_29_groups_0 = const()[name = tensor("q_29_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236020672)))]; tensor q_29_cast_fp16 = conv(dilations = q_29_dilations_0, groups = q_29_groups_0, pad = q_29_pad_0, pad_type = q_29_pad_type_0, strides = q_29_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_167_cast_fp16)[name = tensor("q_29_cast_fp16")]; tensor k_57_pad_type_0 = const()[name = tensor("k_57_pad_type_0"), val = tensor("valid")]; tensor k_57_strides_0 = const()[name = tensor("k_57_strides_0"), val = tensor([1, 1])]; tensor k_57_pad_0 = const()[name = tensor("k_57_pad_0"), val = tensor([0, 0, 0, 0])]; tensor k_57_dilations_0 = const()[name = tensor("k_57_dilations_0"), val = tensor([1, 1])]; tensor k_57_groups_0 = const()[name = tensor("k_57_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(239297536)))]; tensor k_57_cast_fp16 = conv(dilations = k_57_dilations_0, groups = k_57_groups_0, pad = k_57_pad_0, pad_type = k_57_pad_type_0, strides = k_57_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_167_cast_fp16)[name = tensor("k_57_cast_fp16")]; tensor v_29_pad_type_0 = const()[name = tensor("v_29_pad_type_0"), val = tensor("valid")]; tensor v_29_strides_0 = const()[name = tensor("v_29_strides_0"), val = tensor([1, 1])]; tensor v_29_pad_0 = const()[name = tensor("v_29_pad_0"), val = tensor([0, 0, 0, 0])]; tensor v_29_dilations_0 = const()[name = tensor("v_29_dilations_0"), val = tensor([1, 1])]; tensor v_29_groups_0 = const()[name = tensor("v_29_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242574400)))]; tensor v_29_cast_fp16 = conv(dilations = v_29_dilations_0, groups = v_29_groups_0, pad = v_29_pad_0, pad_type = v_29_pad_type_0, strides = v_29_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_167_cast_fp16)[name = tensor("v_29_cast_fp16")]; tensor var_6656_begin_0 = const()[name = tensor("op_6656_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6656_end_0 = const()[name = tensor("op_6656_end_0"), val = tensor([2, 160, 1, 256])]; tensor var_6656_end_mask_0 = const()[name = tensor("op_6656_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6656_cast_fp16 = slice_by_index(begin = var_6656_begin_0, end = var_6656_end_0, end_mask = var_6656_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_6656_cast_fp16")]; tensor var_6660_begin_0 = const()[name = tensor("op_6660_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_6660_end_0 = const()[name = tensor("op_6660_end_0"), val = tensor([2, 320, 1, 256])]; tensor var_6660_end_mask_0 = const()[name = tensor("op_6660_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6660_cast_fp16 = slice_by_index(begin = var_6660_begin_0, end = var_6660_end_0, end_mask = var_6660_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_6660_cast_fp16")]; tensor var_6664_begin_0 = const()[name = tensor("op_6664_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_6664_end_0 = const()[name = tensor("op_6664_end_0"), val = tensor([2, 480, 1, 256])]; tensor var_6664_end_mask_0 = const()[name = tensor("op_6664_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6664_cast_fp16 = slice_by_index(begin = var_6664_begin_0, end = var_6664_end_0, end_mask = var_6664_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_6664_cast_fp16")]; tensor var_6668_begin_0 = const()[name = tensor("op_6668_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_6668_end_0 = const()[name = tensor("op_6668_end_0"), val = tensor([2, 640, 1, 256])]; tensor var_6668_end_mask_0 = const()[name = tensor("op_6668_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6668_cast_fp16 = slice_by_index(begin = var_6668_begin_0, end = var_6668_end_0, end_mask = var_6668_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_6668_cast_fp16")]; tensor var_6672_begin_0 = const()[name = tensor("op_6672_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_6672_end_0 = const()[name = tensor("op_6672_end_0"), val = tensor([2, 800, 1, 256])]; tensor var_6672_end_mask_0 = const()[name = tensor("op_6672_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6672_cast_fp16 = slice_by_index(begin = var_6672_begin_0, end = var_6672_end_0, end_mask = var_6672_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_6672_cast_fp16")]; tensor var_6676_begin_0 = const()[name = tensor("op_6676_begin_0"), val = tensor([0, 800, 0, 0])]; tensor var_6676_end_0 = const()[name = tensor("op_6676_end_0"), val = tensor([2, 960, 1, 256])]; tensor var_6676_end_mask_0 = const()[name = tensor("op_6676_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6676_cast_fp16 = slice_by_index(begin = var_6676_begin_0, end = var_6676_end_0, end_mask = var_6676_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_6676_cast_fp16")]; tensor var_6680_begin_0 = const()[name = tensor("op_6680_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_6680_end_0 = const()[name = tensor("op_6680_end_0"), val = tensor([2, 1120, 1, 256])]; tensor var_6680_end_mask_0 = const()[name = tensor("op_6680_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6680_cast_fp16 = slice_by_index(begin = var_6680_begin_0, end = var_6680_end_0, end_mask = var_6680_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_6680_cast_fp16")]; tensor var_6684_begin_0 = const()[name = tensor("op_6684_begin_0"), val = tensor([0, 1120, 0, 0])]; tensor var_6684_end_0 = const()[name = tensor("op_6684_end_0"), val = tensor([2, 1, 1, 256])]; tensor var_6684_end_mask_0 = const()[name = tensor("op_6684_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_6684_cast_fp16 = slice_by_index(begin = var_6684_begin_0, end = var_6684_end_0, end_mask = var_6684_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_6684_cast_fp16")]; tensor k_59_perm_0 = const()[name = tensor("k_59_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_6691_begin_0 = const()[name = tensor("op_6691_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6691_end_0 = const()[name = tensor("op_6691_end_0"), val = tensor([2, 256, 1, 160])]; tensor var_6691_end_mask_0 = const()[name = tensor("op_6691_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_59_cast_fp16 = transpose(perm = k_59_perm_0, x = k_57_cast_fp16)[name = tensor("transpose_17")]; tensor var_6691_cast_fp16 = slice_by_index(begin = var_6691_begin_0, end = var_6691_end_0, end_mask = var_6691_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_6691_cast_fp16")]; tensor var_6695_begin_0 = const()[name = tensor("op_6695_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_6695_end_0 = const()[name = tensor("op_6695_end_0"), val = tensor([2, 256, 1, 320])]; tensor var_6695_end_mask_0 = const()[name = tensor("op_6695_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6695_cast_fp16 = slice_by_index(begin = var_6695_begin_0, end = var_6695_end_0, end_mask = var_6695_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_6695_cast_fp16")]; tensor var_6699_begin_0 = const()[name = tensor("op_6699_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_6699_end_0 = const()[name = tensor("op_6699_end_0"), val = tensor([2, 256, 1, 480])]; tensor var_6699_end_mask_0 = const()[name = tensor("op_6699_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6699_cast_fp16 = slice_by_index(begin = var_6699_begin_0, end = var_6699_end_0, end_mask = var_6699_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_6699_cast_fp16")]; tensor var_6703_begin_0 = const()[name = tensor("op_6703_begin_0"), val = tensor([0, 0, 0, 480])]; tensor var_6703_end_0 = const()[name = tensor("op_6703_end_0"), val = tensor([2, 256, 1, 640])]; tensor var_6703_end_mask_0 = const()[name = tensor("op_6703_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6703_cast_fp16 = slice_by_index(begin = var_6703_begin_0, end = var_6703_end_0, end_mask = var_6703_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_6703_cast_fp16")]; tensor var_6707_begin_0 = const()[name = tensor("op_6707_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_6707_end_0 = const()[name = tensor("op_6707_end_0"), val = tensor([2, 256, 1, 800])]; tensor var_6707_end_mask_0 = const()[name = tensor("op_6707_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6707_cast_fp16 = slice_by_index(begin = var_6707_begin_0, end = var_6707_end_0, end_mask = var_6707_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_6707_cast_fp16")]; tensor var_6711_begin_0 = const()[name = tensor("op_6711_begin_0"), val = tensor([0, 0, 0, 800])]; tensor var_6711_end_0 = const()[name = tensor("op_6711_end_0"), val = tensor([2, 256, 1, 960])]; tensor var_6711_end_mask_0 = const()[name = tensor("op_6711_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6711_cast_fp16 = slice_by_index(begin = var_6711_begin_0, end = var_6711_end_0, end_mask = var_6711_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_6711_cast_fp16")]; tensor var_6715_begin_0 = const()[name = tensor("op_6715_begin_0"), val = tensor([0, 0, 0, 960])]; tensor var_6715_end_0 = const()[name = tensor("op_6715_end_0"), val = tensor([2, 256, 1, 1120])]; tensor var_6715_end_mask_0 = const()[name = tensor("op_6715_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6715_cast_fp16 = slice_by_index(begin = var_6715_begin_0, end = var_6715_end_0, end_mask = var_6715_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_6715_cast_fp16")]; tensor var_6719_begin_0 = const()[name = tensor("op_6719_begin_0"), val = tensor([0, 0, 0, 1120])]; tensor var_6719_end_0 = const()[name = tensor("op_6719_end_0"), val = tensor([2, 256, 1, 1])]; tensor var_6719_end_mask_0 = const()[name = tensor("op_6719_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_6719_cast_fp16 = slice_by_index(begin = var_6719_begin_0, end = var_6719_end_0, end_mask = var_6719_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_6719_cast_fp16")]; tensor var_6721_begin_0 = const()[name = tensor("op_6721_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6721_end_0 = const()[name = tensor("op_6721_end_0"), val = tensor([2, 160, 1, 256])]; tensor var_6721_end_mask_0 = const()[name = tensor("op_6721_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6721_cast_fp16 = slice_by_index(begin = var_6721_begin_0, end = var_6721_end_0, end_mask = var_6721_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_6721_cast_fp16")]; tensor var_6725_begin_0 = const()[name = tensor("op_6725_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_6725_end_0 = const()[name = tensor("op_6725_end_0"), val = tensor([2, 320, 1, 256])]; tensor var_6725_end_mask_0 = const()[name = tensor("op_6725_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6725_cast_fp16 = slice_by_index(begin = var_6725_begin_0, end = var_6725_end_0, end_mask = var_6725_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_6725_cast_fp16")]; tensor var_6729_begin_0 = const()[name = tensor("op_6729_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_6729_end_0 = const()[name = tensor("op_6729_end_0"), val = tensor([2, 480, 1, 256])]; tensor var_6729_end_mask_0 = const()[name = tensor("op_6729_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6729_cast_fp16 = slice_by_index(begin = var_6729_begin_0, end = var_6729_end_0, end_mask = var_6729_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_6729_cast_fp16")]; tensor var_6733_begin_0 = const()[name = tensor("op_6733_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_6733_end_0 = const()[name = tensor("op_6733_end_0"), val = tensor([2, 640, 1, 256])]; tensor var_6733_end_mask_0 = const()[name = tensor("op_6733_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6733_cast_fp16 = slice_by_index(begin = var_6733_begin_0, end = var_6733_end_0, end_mask = var_6733_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_6733_cast_fp16")]; tensor var_6737_begin_0 = const()[name = tensor("op_6737_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_6737_end_0 = const()[name = tensor("op_6737_end_0"), val = tensor([2, 800, 1, 256])]; tensor var_6737_end_mask_0 = const()[name = tensor("op_6737_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6737_cast_fp16 = slice_by_index(begin = var_6737_begin_0, end = var_6737_end_0, end_mask = var_6737_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_6737_cast_fp16")]; tensor var_6741_begin_0 = const()[name = tensor("op_6741_begin_0"), val = tensor([0, 800, 0, 0])]; tensor var_6741_end_0 = const()[name = tensor("op_6741_end_0"), val = tensor([2, 960, 1, 256])]; tensor var_6741_end_mask_0 = const()[name = tensor("op_6741_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6741_cast_fp16 = slice_by_index(begin = var_6741_begin_0, end = var_6741_end_0, end_mask = var_6741_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_6741_cast_fp16")]; tensor var_6745_begin_0 = const()[name = tensor("op_6745_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_6745_end_0 = const()[name = tensor("op_6745_end_0"), val = tensor([2, 1120, 1, 256])]; tensor var_6745_end_mask_0 = const()[name = tensor("op_6745_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6745_cast_fp16 = slice_by_index(begin = var_6745_begin_0, end = var_6745_end_0, end_mask = var_6745_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_6745_cast_fp16")]; tensor var_6749_begin_0 = const()[name = tensor("op_6749_begin_0"), val = tensor([0, 1120, 0, 0])]; tensor var_6749_end_0 = const()[name = tensor("op_6749_end_0"), val = tensor([2, 1, 1, 256])]; tensor var_6749_end_mask_0 = const()[name = tensor("op_6749_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_6749_cast_fp16 = slice_by_index(begin = var_6749_begin_0, end = var_6749_end_0, end_mask = var_6749_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_6749_cast_fp16")]; tensor var_6753_equation_0 = const()[name = tensor("op_6753_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6753_cast_fp16 = einsum(equation = var_6753_equation_0, values = (var_6691_cast_fp16, var_6656_cast_fp16))[name = tensor("op_6753_cast_fp16")]; tensor var_6754_to_fp16 = const()[name = tensor("op_6754_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_97_cast_fp16 = mul(x = var_6753_cast_fp16, y = var_6754_to_fp16)[name = tensor("aw_97_cast_fp16")]; tensor var_6757_equation_0 = const()[name = tensor("op_6757_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6757_cast_fp16 = einsum(equation = var_6757_equation_0, values = (var_6695_cast_fp16, var_6660_cast_fp16))[name = tensor("op_6757_cast_fp16")]; tensor var_6758_to_fp16 = const()[name = tensor("op_6758_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_99_cast_fp16 = mul(x = var_6757_cast_fp16, y = var_6758_to_fp16)[name = tensor("aw_99_cast_fp16")]; tensor var_6761_equation_0 = const()[name = tensor("op_6761_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6761_cast_fp16 = einsum(equation = var_6761_equation_0, values = (var_6699_cast_fp16, var_6664_cast_fp16))[name = tensor("op_6761_cast_fp16")]; tensor var_6762_to_fp16 = const()[name = tensor("op_6762_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_101_cast_fp16 = mul(x = var_6761_cast_fp16, y = var_6762_to_fp16)[name = tensor("aw_101_cast_fp16")]; tensor var_6765_equation_0 = const()[name = tensor("op_6765_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6765_cast_fp16 = einsum(equation = var_6765_equation_0, values = (var_6703_cast_fp16, var_6668_cast_fp16))[name = tensor("op_6765_cast_fp16")]; tensor var_6766_to_fp16 = const()[name = tensor("op_6766_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_103_cast_fp16 = mul(x = var_6765_cast_fp16, y = var_6766_to_fp16)[name = tensor("aw_103_cast_fp16")]; tensor var_6769_equation_0 = const()[name = tensor("op_6769_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6769_cast_fp16 = einsum(equation = var_6769_equation_0, values = (var_6707_cast_fp16, var_6672_cast_fp16))[name = tensor("op_6769_cast_fp16")]; tensor var_6770_to_fp16 = const()[name = tensor("op_6770_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_105_cast_fp16 = mul(x = var_6769_cast_fp16, y = var_6770_to_fp16)[name = tensor("aw_105_cast_fp16")]; tensor var_6773_equation_0 = const()[name = tensor("op_6773_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6773_cast_fp16 = einsum(equation = var_6773_equation_0, values = (var_6711_cast_fp16, var_6676_cast_fp16))[name = tensor("op_6773_cast_fp16")]; tensor var_6774_to_fp16 = const()[name = tensor("op_6774_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_107_cast_fp16 = mul(x = var_6773_cast_fp16, y = var_6774_to_fp16)[name = tensor("aw_107_cast_fp16")]; tensor var_6777_equation_0 = const()[name = tensor("op_6777_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6777_cast_fp16 = einsum(equation = var_6777_equation_0, values = (var_6715_cast_fp16, var_6680_cast_fp16))[name = tensor("op_6777_cast_fp16")]; tensor var_6778_to_fp16 = const()[name = tensor("op_6778_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_109_cast_fp16 = mul(x = var_6777_cast_fp16, y = var_6778_to_fp16)[name = tensor("aw_109_cast_fp16")]; tensor var_6781_equation_0 = const()[name = tensor("op_6781_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6781_cast_fp16 = einsum(equation = var_6781_equation_0, values = (var_6719_cast_fp16, var_6684_cast_fp16))[name = tensor("op_6781_cast_fp16")]; tensor var_6782_to_fp16 = const()[name = tensor("op_6782_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_111_cast_fp16 = mul(x = var_6781_cast_fp16, y = var_6782_to_fp16)[name = tensor("aw_111_cast_fp16")]; tensor var_6784_cast_fp16 = softmax(axis = var_6522, x = aw_97_cast_fp16)[name = tensor("op_6784_cast_fp16")]; tensor var_6785_cast_fp16 = softmax(axis = var_6522, x = aw_99_cast_fp16)[name = tensor("op_6785_cast_fp16")]; tensor var_6786_cast_fp16 = softmax(axis = var_6522, x = aw_101_cast_fp16)[name = tensor("op_6786_cast_fp16")]; tensor var_6787_cast_fp16 = softmax(axis = var_6522, x = aw_103_cast_fp16)[name = tensor("op_6787_cast_fp16")]; tensor var_6788_cast_fp16 = softmax(axis = var_6522, x = aw_105_cast_fp16)[name = tensor("op_6788_cast_fp16")]; tensor var_6789_cast_fp16 = softmax(axis = var_6522, x = aw_107_cast_fp16)[name = tensor("op_6789_cast_fp16")]; tensor var_6790_cast_fp16 = softmax(axis = var_6522, x = aw_109_cast_fp16)[name = tensor("op_6790_cast_fp16")]; tensor var_6791_cast_fp16 = softmax(axis = var_6522, x = aw_111_cast_fp16)[name = tensor("op_6791_cast_fp16")]; tensor var_6793_equation_0 = const()[name = tensor("op_6793_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6793_cast_fp16 = einsum(equation = var_6793_equation_0, values = (var_6721_cast_fp16, var_6784_cast_fp16))[name = tensor("op_6793_cast_fp16")]; tensor var_6795_equation_0 = const()[name = tensor("op_6795_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6795_cast_fp16 = einsum(equation = var_6795_equation_0, values = (var_6725_cast_fp16, var_6785_cast_fp16))[name = tensor("op_6795_cast_fp16")]; tensor var_6797_equation_0 = const()[name = tensor("op_6797_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6797_cast_fp16 = einsum(equation = var_6797_equation_0, values = (var_6729_cast_fp16, var_6786_cast_fp16))[name = tensor("op_6797_cast_fp16")]; tensor var_6799_equation_0 = const()[name = tensor("op_6799_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6799_cast_fp16 = einsum(equation = var_6799_equation_0, values = (var_6733_cast_fp16, var_6787_cast_fp16))[name = tensor("op_6799_cast_fp16")]; tensor var_6801_equation_0 = const()[name = tensor("op_6801_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6801_cast_fp16 = einsum(equation = var_6801_equation_0, values = (var_6737_cast_fp16, var_6788_cast_fp16))[name = tensor("op_6801_cast_fp16")]; tensor var_6803_equation_0 = const()[name = tensor("op_6803_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6803_cast_fp16 = einsum(equation = var_6803_equation_0, values = (var_6741_cast_fp16, var_6789_cast_fp16))[name = tensor("op_6803_cast_fp16")]; tensor var_6805_equation_0 = const()[name = tensor("op_6805_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6805_cast_fp16 = einsum(equation = var_6805_equation_0, values = (var_6745_cast_fp16, var_6790_cast_fp16))[name = tensor("op_6805_cast_fp16")]; tensor var_6807_equation_0 = const()[name = tensor("op_6807_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6807_cast_fp16 = einsum(equation = var_6807_equation_0, values = (var_6749_cast_fp16, var_6791_cast_fp16))[name = tensor("op_6807_cast_fp16")]; tensor input_301_interleave_0 = const()[name = tensor("input_301_interleave_0"), val = tensor(false)]; tensor input_301_cast_fp16 = concat(axis = var_6522, interleave = input_301_interleave_0, values = (var_6793_cast_fp16, var_6795_cast_fp16, var_6797_cast_fp16, var_6799_cast_fp16, var_6801_cast_fp16, var_6803_cast_fp16, var_6805_cast_fp16, var_6807_cast_fp16))[name = tensor("input_301_cast_fp16")]; tensor var_6817_pad_type_0 = const()[name = tensor("op_6817_pad_type_0"), val = tensor("valid")]; tensor var_6817_strides_0 = const()[name = tensor("op_6817_strides_0"), val = tensor([1, 1])]; tensor var_6817_pad_0 = const()[name = tensor("op_6817_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_6817_dilations_0 = const()[name = tensor("op_6817_dilations_0"), val = tensor([1, 1])]; tensor var_6817_groups_0 = const()[name = tensor("op_6817_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(245851264)))]; tensor up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(249128128)))]; tensor var_6817_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_6817_dilations_0, groups = var_6817_groups_0, pad = var_6817_pad_0, pad_type = var_6817_pad_type_0, strides = var_6817_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_301_cast_fp16)[name = tensor("op_6817_cast_fp16")]; tensor inputs_45_cast_fp16 = add(x = var_6817_cast_fp16, y = inputs_43_cast_fp16)[name = tensor("inputs_45_cast_fp16")]; tensor hidden_states_169_axes_0 = const()[name = tensor("hidden_states_169_axes_0"), val = tensor([1])]; tensor hidden_states_169_gamma_0_to_fp16 = const()[name = tensor("hidden_states_169_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(249130752)))]; tensor hidden_states_169_beta_0_to_fp16 = const()[name = tensor("hidden_states_169_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(249133376)))]; tensor var_6827_to_fp16 = const()[name = tensor("op_6827_to_fp16"), val = tensor(0x1.5p-17)]; tensor hidden_states_169_cast_fp16 = layer_norm(axes = hidden_states_169_axes_0, beta = hidden_states_169_beta_0_to_fp16, epsilon = var_6827_to_fp16, gamma = hidden_states_169_gamma_0_to_fp16, x = inputs_45_cast_fp16)[name = tensor("hidden_states_169_cast_fp16")]; tensor q_31_pad_type_0 = const()[name = tensor("q_31_pad_type_0"), val = tensor("valid")]; tensor q_31_strides_0 = const()[name = tensor("q_31_strides_0"), val = tensor([1, 1])]; tensor q_31_pad_0 = const()[name = tensor("q_31_pad_0"), val = tensor([0, 0, 0, 0])]; tensor q_31_dilations_0 = const()[name = tensor("q_31_dilations_0"), val = tensor([1, 1])]; tensor q_31_groups_0 = const()[name = tensor("q_31_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(249136000)))]; tensor q_31_cast_fp16 = conv(dilations = q_31_dilations_0, groups = q_31_groups_0, pad = q_31_pad_0, pad_type = q_31_pad_type_0, strides = q_31_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_169_cast_fp16)[name = tensor("q_31_cast_fp16")]; tensor k_61_pad_type_0 = const()[name = tensor("k_61_pad_type_0"), val = tensor("valid")]; tensor k_61_strides_0 = const()[name = tensor("k_61_strides_0"), val = tensor([1, 1])]; tensor k_61_pad_0 = const()[name = tensor("k_61_pad_0"), val = tensor([0, 0, 0, 0])]; tensor k_61_dilations_0 = const()[name = tensor("k_61_dilations_0"), val = tensor([1, 1])]; tensor k_61_groups_0 = const()[name = tensor("k_61_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(252412864)))]; tensor k_61_cast_fp16 = conv(dilations = k_61_dilations_0, groups = k_61_groups_0, pad = k_61_pad_0, pad_type = k_61_pad_type_0, strides = k_61_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_61_cast_fp16")]; tensor v_31_pad_type_0 = const()[name = tensor("v_31_pad_type_0"), val = tensor("valid")]; tensor v_31_strides_0 = const()[name = tensor("v_31_strides_0"), val = tensor([1, 1])]; tensor v_31_pad_0 = const()[name = tensor("v_31_pad_0"), val = tensor([0, 0, 0, 0])]; tensor v_31_dilations_0 = const()[name = tensor("v_31_dilations_0"), val = tensor([1, 1])]; tensor v_31_groups_0 = const()[name = tensor("v_31_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(254379008)))]; tensor v_31_cast_fp16 = conv(dilations = v_31_dilations_0, groups = v_31_groups_0, pad = v_31_pad_0, pad_type = v_31_pad_type_0, strides = v_31_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_31_cast_fp16")]; tensor var_6860_begin_0 = const()[name = tensor("op_6860_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6860_end_0 = const()[name = tensor("op_6860_end_0"), val = tensor([2, 160, 1, 256])]; tensor var_6860_end_mask_0 = const()[name = tensor("op_6860_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6860_cast_fp16 = slice_by_index(begin = var_6860_begin_0, end = var_6860_end_0, end_mask = var_6860_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6860_cast_fp16")]; tensor var_6864_begin_0 = const()[name = tensor("op_6864_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_6864_end_0 = const()[name = tensor("op_6864_end_0"), val = tensor([2, 320, 1, 256])]; tensor var_6864_end_mask_0 = const()[name = tensor("op_6864_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6864_cast_fp16 = slice_by_index(begin = var_6864_begin_0, end = var_6864_end_0, end_mask = var_6864_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6864_cast_fp16")]; tensor var_6868_begin_0 = const()[name = tensor("op_6868_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_6868_end_0 = const()[name = tensor("op_6868_end_0"), val = tensor([2, 480, 1, 256])]; tensor var_6868_end_mask_0 = const()[name = tensor("op_6868_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6868_cast_fp16 = slice_by_index(begin = var_6868_begin_0, end = var_6868_end_0, end_mask = var_6868_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6868_cast_fp16")]; tensor var_6872_begin_0 = const()[name = tensor("op_6872_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_6872_end_0 = const()[name = tensor("op_6872_end_0"), val = tensor([2, 640, 1, 256])]; tensor var_6872_end_mask_0 = const()[name = tensor("op_6872_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6872_cast_fp16 = slice_by_index(begin = var_6872_begin_0, end = var_6872_end_0, end_mask = var_6872_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6872_cast_fp16")]; tensor var_6876_begin_0 = const()[name = tensor("op_6876_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_6876_end_0 = const()[name = tensor("op_6876_end_0"), val = tensor([2, 800, 1, 256])]; tensor var_6876_end_mask_0 = const()[name = tensor("op_6876_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6876_cast_fp16 = slice_by_index(begin = var_6876_begin_0, end = var_6876_end_0, end_mask = var_6876_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6876_cast_fp16")]; tensor var_6880_begin_0 = const()[name = tensor("op_6880_begin_0"), val = tensor([0, 800, 0, 0])]; tensor var_6880_end_0 = const()[name = tensor("op_6880_end_0"), val = tensor([2, 960, 1, 256])]; tensor var_6880_end_mask_0 = const()[name = tensor("op_6880_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6880_cast_fp16 = slice_by_index(begin = var_6880_begin_0, end = var_6880_end_0, end_mask = var_6880_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6880_cast_fp16")]; tensor var_6884_begin_0 = const()[name = tensor("op_6884_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_6884_end_0 = const()[name = tensor("op_6884_end_0"), val = tensor([2, 1120, 1, 256])]; tensor var_6884_end_mask_0 = const()[name = tensor("op_6884_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6884_cast_fp16 = slice_by_index(begin = var_6884_begin_0, end = var_6884_end_0, end_mask = var_6884_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6884_cast_fp16")]; tensor var_6888_begin_0 = const()[name = tensor("op_6888_begin_0"), val = tensor([0, 1120, 0, 0])]; tensor var_6888_end_0 = const()[name = tensor("op_6888_end_0"), val = tensor([2, 1, 1, 256])]; tensor var_6888_end_mask_0 = const()[name = tensor("op_6888_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_6888_cast_fp16 = slice_by_index(begin = var_6888_begin_0, end = var_6888_end_0, end_mask = var_6888_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6888_cast_fp16")]; tensor k_63_perm_0 = const()[name = tensor("k_63_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_6895_begin_0 = const()[name = tensor("op_6895_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6895_end_0 = const()[name = tensor("op_6895_end_0"), val = tensor([2, 77, 1, 160])]; tensor var_6895_end_mask_0 = const()[name = tensor("op_6895_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_63_cast_fp16 = transpose(perm = k_63_perm_0, x = k_61_cast_fp16)[name = tensor("transpose_16")]; tensor var_6895_cast_fp16 = slice_by_index(begin = var_6895_begin_0, end = var_6895_end_0, end_mask = var_6895_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6895_cast_fp16")]; tensor var_6899_begin_0 = const()[name = tensor("op_6899_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_6899_end_0 = const()[name = tensor("op_6899_end_0"), val = tensor([2, 77, 1, 320])]; tensor var_6899_end_mask_0 = const()[name = tensor("op_6899_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6899_cast_fp16 = slice_by_index(begin = var_6899_begin_0, end = var_6899_end_0, end_mask = var_6899_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6899_cast_fp16")]; tensor var_6903_begin_0 = const()[name = tensor("op_6903_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_6903_end_0 = const()[name = tensor("op_6903_end_0"), val = tensor([2, 77, 1, 480])]; tensor var_6903_end_mask_0 = const()[name = tensor("op_6903_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6903_cast_fp16 = slice_by_index(begin = var_6903_begin_0, end = var_6903_end_0, end_mask = var_6903_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6903_cast_fp16")]; tensor var_6907_begin_0 = const()[name = tensor("op_6907_begin_0"), val = tensor([0, 0, 0, 480])]; tensor var_6907_end_0 = const()[name = tensor("op_6907_end_0"), val = tensor([2, 77, 1, 640])]; tensor var_6907_end_mask_0 = const()[name = tensor("op_6907_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6907_cast_fp16 = slice_by_index(begin = var_6907_begin_0, end = var_6907_end_0, end_mask = var_6907_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6907_cast_fp16")]; tensor var_6911_begin_0 = const()[name = tensor("op_6911_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_6911_end_0 = const()[name = tensor("op_6911_end_0"), val = tensor([2, 77, 1, 800])]; tensor var_6911_end_mask_0 = const()[name = tensor("op_6911_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6911_cast_fp16 = slice_by_index(begin = var_6911_begin_0, end = var_6911_end_0, end_mask = var_6911_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6911_cast_fp16")]; tensor var_6915_begin_0 = const()[name = tensor("op_6915_begin_0"), val = tensor([0, 0, 0, 800])]; tensor var_6915_end_0 = const()[name = tensor("op_6915_end_0"), val = tensor([2, 77, 1, 960])]; tensor var_6915_end_mask_0 = const()[name = tensor("op_6915_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6915_cast_fp16 = slice_by_index(begin = var_6915_begin_0, end = var_6915_end_0, end_mask = var_6915_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6915_cast_fp16")]; tensor var_6919_begin_0 = const()[name = tensor("op_6919_begin_0"), val = tensor([0, 0, 0, 960])]; tensor var_6919_end_0 = const()[name = tensor("op_6919_end_0"), val = tensor([2, 77, 1, 1120])]; tensor var_6919_end_mask_0 = const()[name = tensor("op_6919_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6919_cast_fp16 = slice_by_index(begin = var_6919_begin_0, end = var_6919_end_0, end_mask = var_6919_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6919_cast_fp16")]; tensor var_6923_begin_0 = const()[name = tensor("op_6923_begin_0"), val = tensor([0, 0, 0, 1120])]; tensor var_6923_end_0 = const()[name = tensor("op_6923_end_0"), val = tensor([2, 77, 1, 1])]; tensor var_6923_end_mask_0 = const()[name = tensor("op_6923_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_6923_cast_fp16 = slice_by_index(begin = var_6923_begin_0, end = var_6923_end_0, end_mask = var_6923_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6923_cast_fp16")]; tensor var_6925_begin_0 = const()[name = tensor("op_6925_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6925_end_0 = const()[name = tensor("op_6925_end_0"), val = tensor([2, 160, 1, 77])]; tensor var_6925_end_mask_0 = const()[name = tensor("op_6925_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6925_cast_fp16 = slice_by_index(begin = var_6925_begin_0, end = var_6925_end_0, end_mask = var_6925_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6925_cast_fp16")]; tensor var_6929_begin_0 = const()[name = tensor("op_6929_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_6929_end_0 = const()[name = tensor("op_6929_end_0"), val = tensor([2, 320, 1, 77])]; tensor var_6929_end_mask_0 = const()[name = tensor("op_6929_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6929_cast_fp16 = slice_by_index(begin = var_6929_begin_0, end = var_6929_end_0, end_mask = var_6929_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6929_cast_fp16")]; tensor var_6933_begin_0 = const()[name = tensor("op_6933_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_6933_end_0 = const()[name = tensor("op_6933_end_0"), val = tensor([2, 480, 1, 77])]; tensor var_6933_end_mask_0 = const()[name = tensor("op_6933_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6933_cast_fp16 = slice_by_index(begin = var_6933_begin_0, end = var_6933_end_0, end_mask = var_6933_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6933_cast_fp16")]; tensor var_6937_begin_0 = const()[name = tensor("op_6937_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_6937_end_0 = const()[name = tensor("op_6937_end_0"), val = tensor([2, 640, 1, 77])]; tensor var_6937_end_mask_0 = const()[name = tensor("op_6937_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6937_cast_fp16 = slice_by_index(begin = var_6937_begin_0, end = var_6937_end_0, end_mask = var_6937_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6937_cast_fp16")]; tensor var_6941_begin_0 = const()[name = tensor("op_6941_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_6941_end_0 = const()[name = tensor("op_6941_end_0"), val = tensor([2, 800, 1, 77])]; tensor var_6941_end_mask_0 = const()[name = tensor("op_6941_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6941_cast_fp16 = slice_by_index(begin = var_6941_begin_0, end = var_6941_end_0, end_mask = var_6941_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6941_cast_fp16")]; tensor var_6945_begin_0 = const()[name = tensor("op_6945_begin_0"), val = tensor([0, 800, 0, 0])]; tensor var_6945_end_0 = const()[name = tensor("op_6945_end_0"), val = tensor([2, 960, 1, 77])]; tensor var_6945_end_mask_0 = const()[name = tensor("op_6945_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6945_cast_fp16 = slice_by_index(begin = var_6945_begin_0, end = var_6945_end_0, end_mask = var_6945_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6945_cast_fp16")]; tensor var_6949_begin_0 = const()[name = tensor("op_6949_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_6949_end_0 = const()[name = tensor("op_6949_end_0"), val = tensor([2, 1120, 1, 77])]; tensor var_6949_end_mask_0 = const()[name = tensor("op_6949_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6949_cast_fp16 = slice_by_index(begin = var_6949_begin_0, end = var_6949_end_0, end_mask = var_6949_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6949_cast_fp16")]; tensor var_6953_begin_0 = const()[name = tensor("op_6953_begin_0"), val = tensor([0, 1120, 0, 0])]; tensor var_6953_end_0 = const()[name = tensor("op_6953_end_0"), val = tensor([2, 1, 1, 77])]; tensor var_6953_end_mask_0 = const()[name = tensor("op_6953_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_6953_cast_fp16 = slice_by_index(begin = var_6953_begin_0, end = var_6953_end_0, end_mask = var_6953_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6953_cast_fp16")]; tensor var_6957_equation_0 = const()[name = tensor("op_6957_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6957_cast_fp16 = einsum(equation = var_6957_equation_0, values = (var_6895_cast_fp16, var_6860_cast_fp16))[name = tensor("op_6957_cast_fp16")]; tensor var_6958_to_fp16 = const()[name = tensor("op_6958_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_113_cast_fp16 = mul(x = var_6957_cast_fp16, y = var_6958_to_fp16)[name = tensor("aw_113_cast_fp16")]; tensor var_6961_equation_0 = const()[name = tensor("op_6961_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6961_cast_fp16 = einsum(equation = var_6961_equation_0, values = (var_6899_cast_fp16, var_6864_cast_fp16))[name = tensor("op_6961_cast_fp16")]; tensor var_6962_to_fp16 = const()[name = tensor("op_6962_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_115_cast_fp16 = mul(x = var_6961_cast_fp16, y = var_6962_to_fp16)[name = tensor("aw_115_cast_fp16")]; tensor var_6965_equation_0 = const()[name = tensor("op_6965_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6965_cast_fp16 = einsum(equation = var_6965_equation_0, values = (var_6903_cast_fp16, var_6868_cast_fp16))[name = tensor("op_6965_cast_fp16")]; tensor var_6966_to_fp16 = const()[name = tensor("op_6966_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_117_cast_fp16 = mul(x = var_6965_cast_fp16, y = var_6966_to_fp16)[name = tensor("aw_117_cast_fp16")]; tensor var_6969_equation_0 = const()[name = tensor("op_6969_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6969_cast_fp16 = einsum(equation = var_6969_equation_0, values = (var_6907_cast_fp16, var_6872_cast_fp16))[name = tensor("op_6969_cast_fp16")]; tensor var_6970_to_fp16 = const()[name = tensor("op_6970_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_119_cast_fp16 = mul(x = var_6969_cast_fp16, y = var_6970_to_fp16)[name = tensor("aw_119_cast_fp16")]; tensor var_6973_equation_0 = const()[name = tensor("op_6973_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6973_cast_fp16 = einsum(equation = var_6973_equation_0, values = (var_6911_cast_fp16, var_6876_cast_fp16))[name = tensor("op_6973_cast_fp16")]; tensor var_6974_to_fp16 = const()[name = tensor("op_6974_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_121_cast_fp16 = mul(x = var_6973_cast_fp16, y = var_6974_to_fp16)[name = tensor("aw_121_cast_fp16")]; tensor var_6977_equation_0 = const()[name = tensor("op_6977_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6977_cast_fp16 = einsum(equation = var_6977_equation_0, values = (var_6915_cast_fp16, var_6880_cast_fp16))[name = tensor("op_6977_cast_fp16")]; tensor var_6978_to_fp16 = const()[name = tensor("op_6978_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_123_cast_fp16 = mul(x = var_6977_cast_fp16, y = var_6978_to_fp16)[name = tensor("aw_123_cast_fp16")]; tensor var_6981_equation_0 = const()[name = tensor("op_6981_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6981_cast_fp16 = einsum(equation = var_6981_equation_0, values = (var_6919_cast_fp16, var_6884_cast_fp16))[name = tensor("op_6981_cast_fp16")]; tensor var_6982_to_fp16 = const()[name = tensor("op_6982_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_125_cast_fp16 = mul(x = var_6981_cast_fp16, y = var_6982_to_fp16)[name = tensor("aw_125_cast_fp16")]; tensor var_6985_equation_0 = const()[name = tensor("op_6985_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6985_cast_fp16 = einsum(equation = var_6985_equation_0, values = (var_6923_cast_fp16, var_6888_cast_fp16))[name = tensor("op_6985_cast_fp16")]; tensor var_6986_to_fp16 = const()[name = tensor("op_6986_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_127_cast_fp16 = mul(x = var_6985_cast_fp16, y = var_6986_to_fp16)[name = tensor("aw_127_cast_fp16")]; tensor var_6988_cast_fp16 = softmax(axis = var_6522, x = aw_113_cast_fp16)[name = tensor("op_6988_cast_fp16")]; tensor var_6989_cast_fp16 = softmax(axis = var_6522, x = aw_115_cast_fp16)[name = tensor("op_6989_cast_fp16")]; tensor var_6990_cast_fp16 = softmax(axis = var_6522, x = aw_117_cast_fp16)[name = tensor("op_6990_cast_fp16")]; tensor var_6991_cast_fp16 = softmax(axis = var_6522, x = aw_119_cast_fp16)[name = tensor("op_6991_cast_fp16")]; tensor var_6992_cast_fp16 = softmax(axis = var_6522, x = aw_121_cast_fp16)[name = tensor("op_6992_cast_fp16")]; tensor var_6993_cast_fp16 = softmax(axis = var_6522, x = aw_123_cast_fp16)[name = tensor("op_6993_cast_fp16")]; tensor var_6994_cast_fp16 = softmax(axis = var_6522, x = aw_125_cast_fp16)[name = tensor("op_6994_cast_fp16")]; tensor var_6995_cast_fp16 = softmax(axis = var_6522, x = aw_127_cast_fp16)[name = tensor("op_6995_cast_fp16")]; tensor var_6997_equation_0 = const()[name = tensor("op_6997_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6997_cast_fp16 = einsum(equation = var_6997_equation_0, values = (var_6925_cast_fp16, var_6988_cast_fp16))[name = tensor("op_6997_cast_fp16")]; tensor var_6999_equation_0 = const()[name = tensor("op_6999_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6999_cast_fp16 = einsum(equation = var_6999_equation_0, values = (var_6929_cast_fp16, var_6989_cast_fp16))[name = tensor("op_6999_cast_fp16")]; tensor var_7001_equation_0 = const()[name = tensor("op_7001_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7001_cast_fp16 = einsum(equation = var_7001_equation_0, values = (var_6933_cast_fp16, var_6990_cast_fp16))[name = tensor("op_7001_cast_fp16")]; tensor var_7003_equation_0 = const()[name = tensor("op_7003_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7003_cast_fp16 = einsum(equation = var_7003_equation_0, values = (var_6937_cast_fp16, var_6991_cast_fp16))[name = tensor("op_7003_cast_fp16")]; tensor var_7005_equation_0 = const()[name = tensor("op_7005_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7005_cast_fp16 = einsum(equation = var_7005_equation_0, values = (var_6941_cast_fp16, var_6992_cast_fp16))[name = tensor("op_7005_cast_fp16")]; tensor var_7007_equation_0 = const()[name = tensor("op_7007_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7007_cast_fp16 = einsum(equation = var_7007_equation_0, values = (var_6945_cast_fp16, var_6993_cast_fp16))[name = tensor("op_7007_cast_fp16")]; tensor var_7009_equation_0 = const()[name = tensor("op_7009_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7009_cast_fp16 = einsum(equation = var_7009_equation_0, values = (var_6949_cast_fp16, var_6994_cast_fp16))[name = tensor("op_7009_cast_fp16")]; tensor var_7011_equation_0 = const()[name = tensor("op_7011_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7011_cast_fp16 = einsum(equation = var_7011_equation_0, values = (var_6953_cast_fp16, var_6995_cast_fp16))[name = tensor("op_7011_cast_fp16")]; tensor input_303_interleave_0 = const()[name = tensor("input_303_interleave_0"), val = tensor(false)]; tensor input_303_cast_fp16 = concat(axis = var_6522, interleave = input_303_interleave_0, values = (var_6997_cast_fp16, var_6999_cast_fp16, var_7001_cast_fp16, var_7003_cast_fp16, var_7005_cast_fp16, var_7007_cast_fp16, var_7009_cast_fp16, var_7011_cast_fp16))[name = tensor("input_303_cast_fp16")]; tensor var_7021_pad_type_0 = const()[name = tensor("op_7021_pad_type_0"), val = tensor("valid")]; tensor var_7021_strides_0 = const()[name = tensor("op_7021_strides_0"), val = tensor([1, 1])]; tensor var_7021_pad_0 = const()[name = tensor("op_7021_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_7021_dilations_0 = const()[name = tensor("op_7021_dilations_0"), val = tensor([1, 1])]; tensor var_7021_groups_0 = const()[name = tensor("op_7021_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(256345152)))]; tensor up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(259622016)))]; tensor var_7021_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_7021_dilations_0, groups = var_7021_groups_0, pad = var_7021_pad_0, pad_type = var_7021_pad_type_0, strides = var_7021_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_303_cast_fp16)[name = tensor("op_7021_cast_fp16")]; tensor inputs_47_cast_fp16 = add(x = var_7021_cast_fp16, y = inputs_45_cast_fp16)[name = tensor("inputs_47_cast_fp16")]; tensor input_305_axes_0 = const()[name = tensor("input_305_axes_0"), val = tensor([1])]; tensor input_305_gamma_0_to_fp16 = const()[name = tensor("input_305_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(259624640)))]; tensor input_305_beta_0_to_fp16 = const()[name = tensor("input_305_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(259627264)))]; tensor var_7031_to_fp16 = const()[name = tensor("op_7031_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_305_cast_fp16 = layer_norm(axes = input_305_axes_0, beta = input_305_beta_0_to_fp16, epsilon = var_7031_to_fp16, gamma = input_305_gamma_0_to_fp16, x = inputs_47_cast_fp16)[name = tensor("input_305_cast_fp16")]; tensor var_7051_pad_type_0 = const()[name = tensor("op_7051_pad_type_0"), val = tensor("valid")]; tensor var_7051_strides_0 = const()[name = tensor("op_7051_strides_0"), val = tensor([1, 1])]; tensor var_7051_pad_0 = const()[name = tensor("op_7051_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_7051_dilations_0 = const()[name = tensor("op_7051_dilations_0"), val = tensor([1, 1])]; tensor var_7051_groups_0 = const()[name = tensor("op_7051_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(259629888)))]; tensor up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(285844352)))]; tensor var_7051_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_7051_dilations_0, groups = var_7051_groups_0, pad = var_7051_pad_0, pad_type = var_7051_pad_type_0, strides = var_7051_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_305_cast_fp16)[name = tensor("op_7051_cast_fp16")]; tensor var_7052_split_sizes_0 = const()[name = tensor("op_7052_split_sizes_0"), val = tensor([5120, 5120])]; tensor var_7052_axis_0 = const()[name = tensor("op_7052_axis_0"), val = tensor(1)]; tensor var_7052_cast_fp16_0, tensor var_7052_cast_fp16_1 = split(axis = var_7052_axis_0, split_sizes = var_7052_split_sizes_0, x = var_7051_cast_fp16)[name = tensor("op_7052_cast_fp16")]; tensor var_7054_mode_0 = const()[name = tensor("op_7054_mode_0"), val = tensor("EXACT")]; tensor var_7054_cast_fp16 = gelu(mode = var_7054_mode_0, x = var_7052_cast_fp16_1)[name = tensor("op_7054_cast_fp16")]; tensor input_307_cast_fp16 = mul(x = var_7052_cast_fp16_0, y = var_7054_cast_fp16)[name = tensor("input_307_cast_fp16")]; tensor var_7062_pad_type_0 = const()[name = tensor("op_7062_pad_type_0"), val = tensor("valid")]; tensor var_7062_strides_0 = const()[name = tensor("op_7062_strides_0"), val = tensor([1, 1])]; tensor var_7062_pad_0 = const()[name = tensor("op_7062_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_7062_dilations_0 = const()[name = tensor("op_7062_dilations_0"), val = tensor([1, 1])]; tensor var_7062_groups_0 = const()[name = tensor("op_7062_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(285864896)))]; tensor up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(298972160)))]; tensor var_7062_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_7062_dilations_0, groups = var_7062_groups_0, pad = var_7062_pad_0, pad_type = var_7062_pad_type_0, strides = var_7062_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_307_cast_fp16)[name = tensor("op_7062_cast_fp16")]; tensor hidden_states_173_cast_fp16 = add(x = var_7062_cast_fp16, y = inputs_47_cast_fp16)[name = tensor("hidden_states_173_cast_fp16")]; tensor var_7064 = const()[name = tensor("op_7064"), val = tensor([2, 1280, 16, 16])]; tensor input_309_cast_fp16 = reshape(shape = var_7064, x = hidden_states_173_cast_fp16)[name = tensor("input_309_cast_fp16")]; tensor hidden_states_175_pad_type_0 = const()[name = tensor("hidden_states_175_pad_type_0"), val = tensor("valid")]; tensor hidden_states_175_strides_0 = const()[name = tensor("hidden_states_175_strides_0"), val = tensor([1, 1])]; tensor hidden_states_175_pad_0 = const()[name = tensor("hidden_states_175_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_175_dilations_0 = const()[name = tensor("hidden_states_175_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_175_groups_0 = const()[name = tensor("hidden_states_175_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(298974784)))]; tensor up_blocks_1_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(302251648)))]; tensor hidden_states_175_cast_fp16 = conv(bias = up_blocks_1_attentions_0_proj_out_bias_to_fp16, dilations = hidden_states_175_dilations_0, groups = hidden_states_175_groups_0, pad = hidden_states_175_pad_0, pad_type = hidden_states_175_pad_type_0, strides = hidden_states_175_strides_0, weight = up_blocks_1_attentions_0_proj_out_weight_to_fp16, x = input_309_cast_fp16)[name = tensor("hidden_states_175_cast_fp16")]; tensor hidden_states_177_cast_fp16 = add(x = hidden_states_175_cast_fp16, y = hidden_states_163_cast_fp16)[name = tensor("hidden_states_177_cast_fp16")]; tensor input_311_interleave_0 = const()[name = tensor("input_311_interleave_0"), val = tensor(false)]; tensor cast_4 = cast(dtype = cast_4_dtype_0, x = input_143_cast_fp16)[name = tensor("cast_5")]; tensor input_311_cast_fp16 = concat(axis = var_6522, interleave = input_311_interleave_0, values = (hidden_states_177_cast_fp16, cast_4))[name = tensor("input_311_cast_fp16")]; tensor reshape_144_shape_0 = const()[name = tensor("reshape_144_shape_0"), val = tensor([2, 32, 80, 16, 16])]; tensor reshape_144_cast_fp16 = reshape(shape = reshape_144_shape_0, x = input_311_cast_fp16)[name = tensor("reshape_144_cast_fp16")]; tensor reduce_mean_108_axes_0 = const()[name = tensor("reduce_mean_108_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_108_keep_dims_0 = const()[name = tensor("reduce_mean_108_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_108_cast_fp16 = reduce_mean(axes = reduce_mean_108_axes_0, keep_dims = reduce_mean_108_keep_dims_0, x = reshape_144_cast_fp16)[name = tensor("reduce_mean_108_cast_fp16")]; tensor sub_72_cast_fp16 = sub(x = reshape_144_cast_fp16, y = reduce_mean_108_cast_fp16)[name = tensor("sub_72_cast_fp16")]; tensor square_36_cast_fp16 = square(x = sub_72_cast_fp16)[name = tensor("square_36_cast_fp16")]; tensor reduce_mean_110_axes_0 = const()[name = tensor("reduce_mean_110_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_110_keep_dims_0 = const()[name = tensor("reduce_mean_110_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_110_cast_fp16 = reduce_mean(axes = reduce_mean_110_axes_0, keep_dims = reduce_mean_110_keep_dims_0, x = square_36_cast_fp16)[name = tensor("reduce_mean_110_cast_fp16")]; tensor add_72_y_0_to_fp16 = const()[name = tensor("add_72_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_72_cast_fp16 = add(x = reduce_mean_110_cast_fp16, y = add_72_y_0_to_fp16)[name = tensor("add_72_cast_fp16")]; tensor sqrt_36_cast_fp16 = sqrt(x = add_72_cast_fp16)[name = tensor("sqrt_36_cast_fp16")]; tensor real_div_36_cast_fp16 = real_div(x = sub_72_cast_fp16, y = sqrt_36_cast_fp16)[name = tensor("real_div_36_cast_fp16")]; tensor reshape_145_shape_0 = const()[name = tensor("reshape_145_shape_0"), val = tensor([2, 2560, 16, 16])]; tensor reshape_145_cast_fp16 = reshape(shape = reshape_145_shape_0, x = real_div_36_cast_fp16)[name = tensor("reshape_145_cast_fp16")]; tensor add_73_gamma_0_to_fp16 = const()[name = tensor("add_73_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(302254272)))]; tensor add_73_beta_0_to_fp16 = const()[name = tensor("add_73_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(302259456)))]; tensor add_73_epsilon_0_to_fp16 = const()[name = tensor("add_73_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_73_cast_fp16 = batch_norm(beta = add_73_beta_0_to_fp16, epsilon = add_73_epsilon_0_to_fp16, gamma = add_73_gamma_0_to_fp16, mean = add_55_mean_0_to_fp16, variance = add_55_variance_0_to_fp16, x = reshape_145_cast_fp16)[name = tensor("add_73_cast_fp16")]; tensor input_315_cast_fp16 = silu(x = add_73_cast_fp16)[name = tensor("input_315_cast_fp16")]; tensor hidden_states_179_pad_type_0 = const()[name = tensor("hidden_states_179_pad_type_0"), val = tensor("custom")]; tensor hidden_states_179_pad_0 = const()[name = tensor("hidden_states_179_pad_0"), val = tensor([1, 1, 1, 1])]; tensor hidden_states_179_strides_0 = const()[name = tensor("hidden_states_179_strides_0"), val = tensor([1, 1])]; tensor hidden_states_179_dilations_0 = const()[name = tensor("hidden_states_179_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_179_groups_0 = const()[name = tensor("hidden_states_179_groups_0"), val = tensor(1)]; tensor up_blocks_1_resnets_1_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(302264640)))]; tensor up_blocks_1_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(361247104)))]; tensor hidden_states_179_cast_fp16 = conv(bias = up_blocks_1_resnets_1_conv1_bias_to_fp16, dilations = hidden_states_179_dilations_0, groups = hidden_states_179_groups_0, pad = hidden_states_179_pad_0, pad_type = hidden_states_179_pad_type_0, strides = hidden_states_179_strides_0, weight = up_blocks_1_resnets_1_conv1_weight_to_fp16, x = input_315_cast_fp16)[name = tensor("hidden_states_179_cast_fp16")]; tensor temb_29_pad_type_0 = const()[name = tensor("temb_29_pad_type_0"), val = tensor("valid")]; tensor temb_29_strides_0 = const()[name = tensor("temb_29_strides_0"), val = tensor([1, 1])]; tensor temb_29_pad_0 = const()[name = tensor("temb_29_pad_0"), val = tensor([0, 0, 0, 0])]; tensor temb_29_dilations_0 = const()[name = tensor("temb_29_dilations_0"), val = tensor([1, 1])]; tensor temb_29_groups_0 = const()[name = tensor("temb_29_groups_0"), val = tensor(1)]; tensor up_blocks_1_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(361249728)))]; tensor up_blocks_1_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364526592)))]; tensor temb_29_cast_fp16 = conv(bias = up_blocks_1_resnets_1_time_emb_proj_bias_to_fp16, dilations = temb_29_dilations_0, groups = temb_29_groups_0, pad = temb_29_pad_0, pad_type = temb_29_pad_type_0, strides = temb_29_strides_0, weight = up_blocks_1_resnets_1_time_emb_proj_weight_to_fp16, x = cast_9)[name = tensor("temb_29_cast_fp16")]; tensor input_319_cast_fp16 = add(x = hidden_states_179_cast_fp16, y = temb_29_cast_fp16)[name = tensor("input_319_cast_fp16")]; tensor reshape_148_shape_0 = const()[name = tensor("reshape_148_shape_0"), val = tensor([2, 32, 40, 16, 16])]; tensor reshape_148_cast_fp16 = reshape(shape = reshape_148_shape_0, x = input_319_cast_fp16)[name = tensor("reshape_148_cast_fp16")]; tensor reduce_mean_111_axes_0 = const()[name = tensor("reduce_mean_111_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_111_keep_dims_0 = const()[name = tensor("reduce_mean_111_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_111_cast_fp16 = reduce_mean(axes = reduce_mean_111_axes_0, keep_dims = reduce_mean_111_keep_dims_0, x = reshape_148_cast_fp16)[name = tensor("reduce_mean_111_cast_fp16")]; tensor sub_74_cast_fp16 = sub(x = reshape_148_cast_fp16, y = reduce_mean_111_cast_fp16)[name = tensor("sub_74_cast_fp16")]; tensor square_37_cast_fp16 = square(x = sub_74_cast_fp16)[name = tensor("square_37_cast_fp16")]; tensor reduce_mean_113_axes_0 = const()[name = tensor("reduce_mean_113_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_113_keep_dims_0 = const()[name = tensor("reduce_mean_113_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_113_cast_fp16 = reduce_mean(axes = reduce_mean_113_axes_0, keep_dims = reduce_mean_113_keep_dims_0, x = square_37_cast_fp16)[name = tensor("reduce_mean_113_cast_fp16")]; tensor add_74_y_0_to_fp16 = const()[name = tensor("add_74_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_74_cast_fp16 = add(x = reduce_mean_113_cast_fp16, y = add_74_y_0_to_fp16)[name = tensor("add_74_cast_fp16")]; tensor sqrt_37_cast_fp16 = sqrt(x = add_74_cast_fp16)[name = tensor("sqrt_37_cast_fp16")]; tensor real_div_37_cast_fp16 = real_div(x = sub_74_cast_fp16, y = sqrt_37_cast_fp16)[name = tensor("real_div_37_cast_fp16")]; tensor reshape_149_shape_0 = const()[name = tensor("reshape_149_shape_0"), val = tensor([2, 1280, 16, 16])]; tensor reshape_149_cast_fp16 = reshape(shape = reshape_149_shape_0, x = real_div_37_cast_fp16)[name = tensor("reshape_149_cast_fp16")]; tensor add_75_gamma_0_to_fp16 = const()[name = tensor("add_75_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364529216)))]; tensor add_75_beta_0_to_fp16 = const()[name = tensor("add_75_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364531840)))]; tensor add_75_epsilon_0_to_fp16 = const()[name = tensor("add_75_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_75_cast_fp16 = batch_norm(beta = add_75_beta_0_to_fp16, epsilon = add_75_epsilon_0_to_fp16, gamma = add_75_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_149_cast_fp16)[name = tensor("add_75_cast_fp16")]; tensor input_323_cast_fp16 = silu(x = add_75_cast_fp16)[name = tensor("input_323_cast_fp16")]; tensor hidden_states_181_pad_type_0 = const()[name = tensor("hidden_states_181_pad_type_0"), val = tensor("custom")]; tensor hidden_states_181_pad_0 = const()[name = tensor("hidden_states_181_pad_0"), val = tensor([1, 1, 1, 1])]; tensor hidden_states_181_strides_0 = const()[name = tensor("hidden_states_181_strides_0"), val = tensor([1, 1])]; tensor hidden_states_181_dilations_0 = const()[name = tensor("hidden_states_181_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_181_groups_0 = const()[name = tensor("hidden_states_181_groups_0"), val = tensor(1)]; tensor up_blocks_1_resnets_1_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364534464)))]; tensor up_blocks_1_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(394025728)))]; tensor hidden_states_181_cast_fp16 = conv(bias = up_blocks_1_resnets_1_conv2_bias_to_fp16, dilations = hidden_states_181_dilations_0, groups = hidden_states_181_groups_0, pad = hidden_states_181_pad_0, pad_type = hidden_states_181_pad_type_0, strides = hidden_states_181_strides_0, weight = up_blocks_1_resnets_1_conv2_weight_to_fp16, x = input_323_cast_fp16)[name = tensor("hidden_states_181_cast_fp16")]; tensor x_13_pad_type_0 = const()[name = tensor("x_13_pad_type_0"), val = tensor("valid")]; tensor x_13_strides_0 = const()[name = tensor("x_13_strides_0"), val = tensor([1, 1])]; tensor x_13_pad_0 = const()[name = tensor("x_13_pad_0"), val = tensor([0, 0, 0, 0])]; tensor x_13_dilations_0 = const()[name = tensor("x_13_dilations_0"), val = tensor([1, 1])]; tensor x_13_groups_0 = const()[name = tensor("x_13_groups_0"), val = tensor(1)]; tensor up_blocks_1_resnets_1_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(394028352)))]; tensor up_blocks_1_resnets_1_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(400582016)))]; tensor x_13_cast_fp16 = conv(bias = up_blocks_1_resnets_1_conv_shortcut_bias_to_fp16, dilations = x_13_dilations_0, groups = x_13_groups_0, pad = x_13_pad_0, pad_type = x_13_pad_type_0, strides = x_13_strides_0, weight = up_blocks_1_resnets_1_conv_shortcut_weight_to_fp16, x = input_311_cast_fp16)[name = tensor("x_13_cast_fp16")]; tensor hidden_states_183_cast_fp16 = add(x = x_13_cast_fp16, y = hidden_states_181_cast_fp16)[name = tensor("hidden_states_183_cast_fp16")]; tensor reshape_152_shape_0 = const()[name = tensor("reshape_152_shape_0"), val = tensor([2, 32, 40, 16, 16])]; tensor reshape_152_cast_fp16 = reshape(shape = reshape_152_shape_0, x = hidden_states_183_cast_fp16)[name = tensor("reshape_152_cast_fp16")]; tensor reduce_mean_114_axes_0 = const()[name = tensor("reduce_mean_114_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_114_keep_dims_0 = const()[name = tensor("reduce_mean_114_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_114_cast_fp16 = reduce_mean(axes = reduce_mean_114_axes_0, keep_dims = reduce_mean_114_keep_dims_0, x = reshape_152_cast_fp16)[name = tensor("reduce_mean_114_cast_fp16")]; tensor sub_76_cast_fp16 = sub(x = reshape_152_cast_fp16, y = reduce_mean_114_cast_fp16)[name = tensor("sub_76_cast_fp16")]; tensor square_38_cast_fp16 = square(x = sub_76_cast_fp16)[name = tensor("square_38_cast_fp16")]; tensor reduce_mean_116_axes_0 = const()[name = tensor("reduce_mean_116_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_116_keep_dims_0 = const()[name = tensor("reduce_mean_116_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_116_cast_fp16 = reduce_mean(axes = reduce_mean_116_axes_0, keep_dims = reduce_mean_116_keep_dims_0, x = square_38_cast_fp16)[name = tensor("reduce_mean_116_cast_fp16")]; tensor add_76_y_0_to_fp16 = const()[name = tensor("add_76_y_0_to_fp16"), val = tensor(0x1.1p-20)]; tensor add_76_cast_fp16 = add(x = reduce_mean_116_cast_fp16, y = add_76_y_0_to_fp16)[name = tensor("add_76_cast_fp16")]; tensor sqrt_38_cast_fp16 = sqrt(x = add_76_cast_fp16)[name = tensor("sqrt_38_cast_fp16")]; tensor real_div_38_cast_fp16 = real_div(x = sub_76_cast_fp16, y = sqrt_38_cast_fp16)[name = tensor("real_div_38_cast_fp16")]; tensor reshape_153_shape_0 = const()[name = tensor("reshape_153_shape_0"), val = tensor([2, 1280, 16, 16])]; tensor reshape_153_cast_fp16 = reshape(shape = reshape_153_shape_0, x = real_div_38_cast_fp16)[name = tensor("reshape_153_cast_fp16")]; tensor add_77_gamma_0_to_fp16 = const()[name = tensor("add_77_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(400584640)))]; tensor add_77_beta_0_to_fp16 = const()[name = tensor("add_77_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(400587264)))]; tensor add_77_epsilon_0_to_fp16 = const()[name = tensor("add_77_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_77_cast_fp16 = batch_norm(beta = add_77_beta_0_to_fp16, epsilon = add_77_epsilon_0_to_fp16, gamma = add_77_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_153_cast_fp16)[name = tensor("add_77_cast_fp16")]; tensor hidden_states_185_pad_type_0 = const()[name = tensor("hidden_states_185_pad_type_0"), val = tensor("valid")]; tensor hidden_states_185_strides_0 = const()[name = tensor("hidden_states_185_strides_0"), val = tensor([1, 1])]; tensor hidden_states_185_pad_0 = const()[name = tensor("hidden_states_185_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_185_dilations_0 = const()[name = tensor("hidden_states_185_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_185_groups_0 = const()[name = tensor("hidden_states_185_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_1_proj_in_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(400589888)))]; tensor up_blocks_1_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(403866752)))]; tensor hidden_states_185_cast_fp16 = conv(bias = up_blocks_1_attentions_1_proj_in_bias_to_fp16, dilations = hidden_states_185_dilations_0, groups = hidden_states_185_groups_0, pad = hidden_states_185_pad_0, pad_type = hidden_states_185_pad_type_0, strides = hidden_states_185_strides_0, weight = up_blocks_1_attentions_1_proj_in_weight_to_fp16, x = add_77_cast_fp16)[name = tensor("hidden_states_185_cast_fp16")]; tensor var_7144 = const()[name = tensor("op_7144"), val = tensor([2, 1280, 1, 256])]; tensor inputs_49_cast_fp16 = reshape(shape = var_7144, x = hidden_states_185_cast_fp16)[name = tensor("inputs_49_cast_fp16")]; tensor hidden_states_187_axes_0 = const()[name = tensor("hidden_states_187_axes_0"), val = tensor([1])]; tensor hidden_states_187_gamma_0_to_fp16 = const()[name = tensor("hidden_states_187_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(403869376)))]; tensor hidden_states_187_beta_0_to_fp16 = const()[name = tensor("hidden_states_187_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(403872000)))]; tensor var_7160_to_fp16 = const()[name = tensor("op_7160_to_fp16"), val = tensor(0x1.5p-17)]; tensor hidden_states_187_cast_fp16 = layer_norm(axes = hidden_states_187_axes_0, beta = hidden_states_187_beta_0_to_fp16, epsilon = var_7160_to_fp16, gamma = hidden_states_187_gamma_0_to_fp16, x = inputs_49_cast_fp16)[name = tensor("hidden_states_187_cast_fp16")]; tensor q_33_pad_type_0 = const()[name = tensor("q_33_pad_type_0"), val = tensor("valid")]; tensor q_33_strides_0 = const()[name = tensor("q_33_strides_0"), val = tensor([1, 1])]; tensor q_33_pad_0 = const()[name = tensor("q_33_pad_0"), val = tensor([0, 0, 0, 0])]; tensor q_33_dilations_0 = const()[name = tensor("q_33_dilations_0"), val = tensor([1, 1])]; tensor q_33_groups_0 = const()[name = tensor("q_33_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(403874624)))]; tensor q_33_cast_fp16 = conv(dilations = q_33_dilations_0, groups = q_33_groups_0, pad = q_33_pad_0, pad_type = q_33_pad_type_0, strides = q_33_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_187_cast_fp16)[name = tensor("q_33_cast_fp16")]; tensor k_65_pad_type_0 = const()[name = tensor("k_65_pad_type_0"), val = tensor("valid")]; tensor k_65_strides_0 = const()[name = tensor("k_65_strides_0"), val = tensor([1, 1])]; tensor k_65_pad_0 = const()[name = tensor("k_65_pad_0"), val = tensor([0, 0, 0, 0])]; tensor k_65_dilations_0 = const()[name = tensor("k_65_dilations_0"), val = tensor([1, 1])]; tensor k_65_groups_0 = const()[name = tensor("k_65_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(407151488)))]; tensor k_65_cast_fp16 = conv(dilations = k_65_dilations_0, groups = k_65_groups_0, pad = k_65_pad_0, pad_type = k_65_pad_type_0, strides = k_65_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_187_cast_fp16)[name = tensor("k_65_cast_fp16")]; tensor v_33_pad_type_0 = const()[name = tensor("v_33_pad_type_0"), val = tensor("valid")]; tensor v_33_strides_0 = const()[name = tensor("v_33_strides_0"), val = tensor([1, 1])]; tensor v_33_pad_0 = const()[name = tensor("v_33_pad_0"), val = tensor([0, 0, 0, 0])]; tensor v_33_dilations_0 = const()[name = tensor("v_33_dilations_0"), val = tensor([1, 1])]; tensor v_33_groups_0 = const()[name = tensor("v_33_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(410428352)))]; tensor v_33_cast_fp16 = conv(dilations = v_33_dilations_0, groups = v_33_groups_0, pad = v_33_pad_0, pad_type = v_33_pad_type_0, strides = v_33_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_187_cast_fp16)[name = tensor("v_33_cast_fp16")]; tensor var_7193_begin_0 = const()[name = tensor("op_7193_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7193_end_0 = const()[name = tensor("op_7193_end_0"), val = tensor([2, 160, 1, 256])]; tensor var_7193_end_mask_0 = const()[name = tensor("op_7193_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7193_cast_fp16 = slice_by_index(begin = var_7193_begin_0, end = var_7193_end_0, end_mask = var_7193_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_7193_cast_fp16")]; tensor var_7197_begin_0 = const()[name = tensor("op_7197_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_7197_end_0 = const()[name = tensor("op_7197_end_0"), val = tensor([2, 320, 1, 256])]; tensor var_7197_end_mask_0 = const()[name = tensor("op_7197_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7197_cast_fp16 = slice_by_index(begin = var_7197_begin_0, end = var_7197_end_0, end_mask = var_7197_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_7197_cast_fp16")]; tensor var_7201_begin_0 = const()[name = tensor("op_7201_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_7201_end_0 = const()[name = tensor("op_7201_end_0"), val = tensor([2, 480, 1, 256])]; tensor var_7201_end_mask_0 = const()[name = tensor("op_7201_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7201_cast_fp16 = slice_by_index(begin = var_7201_begin_0, end = var_7201_end_0, end_mask = var_7201_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_7201_cast_fp16")]; tensor var_7205_begin_0 = const()[name = tensor("op_7205_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_7205_end_0 = const()[name = tensor("op_7205_end_0"), val = tensor([2, 640, 1, 256])]; tensor var_7205_end_mask_0 = const()[name = tensor("op_7205_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7205_cast_fp16 = slice_by_index(begin = var_7205_begin_0, end = var_7205_end_0, end_mask = var_7205_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_7205_cast_fp16")]; tensor var_7209_begin_0 = const()[name = tensor("op_7209_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_7209_end_0 = const()[name = tensor("op_7209_end_0"), val = tensor([2, 800, 1, 256])]; tensor var_7209_end_mask_0 = const()[name = tensor("op_7209_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7209_cast_fp16 = slice_by_index(begin = var_7209_begin_0, end = var_7209_end_0, end_mask = var_7209_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_7209_cast_fp16")]; tensor var_7213_begin_0 = const()[name = tensor("op_7213_begin_0"), val = tensor([0, 800, 0, 0])]; tensor var_7213_end_0 = const()[name = tensor("op_7213_end_0"), val = tensor([2, 960, 1, 256])]; tensor var_7213_end_mask_0 = const()[name = tensor("op_7213_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7213_cast_fp16 = slice_by_index(begin = var_7213_begin_0, end = var_7213_end_0, end_mask = var_7213_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_7213_cast_fp16")]; tensor var_7217_begin_0 = const()[name = tensor("op_7217_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_7217_end_0 = const()[name = tensor("op_7217_end_0"), val = tensor([2, 1120, 1, 256])]; tensor var_7217_end_mask_0 = const()[name = tensor("op_7217_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7217_cast_fp16 = slice_by_index(begin = var_7217_begin_0, end = var_7217_end_0, end_mask = var_7217_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_7217_cast_fp16")]; tensor var_7221_begin_0 = const()[name = tensor("op_7221_begin_0"), val = tensor([0, 1120, 0, 0])]; tensor var_7221_end_0 = const()[name = tensor("op_7221_end_0"), val = tensor([2, 1, 1, 256])]; tensor var_7221_end_mask_0 = const()[name = tensor("op_7221_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_7221_cast_fp16 = slice_by_index(begin = var_7221_begin_0, end = var_7221_end_0, end_mask = var_7221_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_7221_cast_fp16")]; tensor k_67_perm_0 = const()[name = tensor("k_67_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_7228_begin_0 = const()[name = tensor("op_7228_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7228_end_0 = const()[name = tensor("op_7228_end_0"), val = tensor([2, 256, 1, 160])]; tensor var_7228_end_mask_0 = const()[name = tensor("op_7228_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_67_cast_fp16 = transpose(perm = k_67_perm_0, x = k_65_cast_fp16)[name = tensor("transpose_15")]; tensor var_7228_cast_fp16 = slice_by_index(begin = var_7228_begin_0, end = var_7228_end_0, end_mask = var_7228_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_7228_cast_fp16")]; tensor var_7232_begin_0 = const()[name = tensor("op_7232_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_7232_end_0 = const()[name = tensor("op_7232_end_0"), val = tensor([2, 256, 1, 320])]; tensor var_7232_end_mask_0 = const()[name = tensor("op_7232_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7232_cast_fp16 = slice_by_index(begin = var_7232_begin_0, end = var_7232_end_0, end_mask = var_7232_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_7232_cast_fp16")]; tensor var_7236_begin_0 = const()[name = tensor("op_7236_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_7236_end_0 = const()[name = tensor("op_7236_end_0"), val = tensor([2, 256, 1, 480])]; tensor var_7236_end_mask_0 = const()[name = tensor("op_7236_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7236_cast_fp16 = slice_by_index(begin = var_7236_begin_0, end = var_7236_end_0, end_mask = var_7236_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_7236_cast_fp16")]; tensor var_7240_begin_0 = const()[name = tensor("op_7240_begin_0"), val = tensor([0, 0, 0, 480])]; tensor var_7240_end_0 = const()[name = tensor("op_7240_end_0"), val = tensor([2, 256, 1, 640])]; tensor var_7240_end_mask_0 = const()[name = tensor("op_7240_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7240_cast_fp16 = slice_by_index(begin = var_7240_begin_0, end = var_7240_end_0, end_mask = var_7240_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_7240_cast_fp16")]; tensor var_7244_begin_0 = const()[name = tensor("op_7244_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_7244_end_0 = const()[name = tensor("op_7244_end_0"), val = tensor([2, 256, 1, 800])]; tensor var_7244_end_mask_0 = const()[name = tensor("op_7244_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7244_cast_fp16 = slice_by_index(begin = var_7244_begin_0, end = var_7244_end_0, end_mask = var_7244_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_7244_cast_fp16")]; tensor var_7248_begin_0 = const()[name = tensor("op_7248_begin_0"), val = tensor([0, 0, 0, 800])]; tensor var_7248_end_0 = const()[name = tensor("op_7248_end_0"), val = tensor([2, 256, 1, 960])]; tensor var_7248_end_mask_0 = const()[name = tensor("op_7248_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7248_cast_fp16 = slice_by_index(begin = var_7248_begin_0, end = var_7248_end_0, end_mask = var_7248_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_7248_cast_fp16")]; tensor var_7252_begin_0 = const()[name = tensor("op_7252_begin_0"), val = tensor([0, 0, 0, 960])]; tensor var_7252_end_0 = const()[name = tensor("op_7252_end_0"), val = tensor([2, 256, 1, 1120])]; tensor var_7252_end_mask_0 = const()[name = tensor("op_7252_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7252_cast_fp16 = slice_by_index(begin = var_7252_begin_0, end = var_7252_end_0, end_mask = var_7252_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_7252_cast_fp16")]; tensor var_7256_begin_0 = const()[name = tensor("op_7256_begin_0"), val = tensor([0, 0, 0, 1120])]; tensor var_7256_end_0 = const()[name = tensor("op_7256_end_0"), val = tensor([2, 256, 1, 1])]; tensor var_7256_end_mask_0 = const()[name = tensor("op_7256_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_7256_cast_fp16 = slice_by_index(begin = var_7256_begin_0, end = var_7256_end_0, end_mask = var_7256_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_7256_cast_fp16")]; tensor var_7258_begin_0 = const()[name = tensor("op_7258_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7258_end_0 = const()[name = tensor("op_7258_end_0"), val = tensor([2, 160, 1, 256])]; tensor var_7258_end_mask_0 = const()[name = tensor("op_7258_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7258_cast_fp16 = slice_by_index(begin = var_7258_begin_0, end = var_7258_end_0, end_mask = var_7258_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_7258_cast_fp16")]; tensor var_7262_begin_0 = const()[name = tensor("op_7262_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_7262_end_0 = const()[name = tensor("op_7262_end_0"), val = tensor([2, 320, 1, 256])]; tensor var_7262_end_mask_0 = const()[name = tensor("op_7262_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7262_cast_fp16 = slice_by_index(begin = var_7262_begin_0, end = var_7262_end_0, end_mask = var_7262_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_7262_cast_fp16")]; tensor var_7266_begin_0 = const()[name = tensor("op_7266_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_7266_end_0 = const()[name = tensor("op_7266_end_0"), val = tensor([2, 480, 1, 256])]; tensor var_7266_end_mask_0 = const()[name = tensor("op_7266_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7266_cast_fp16 = slice_by_index(begin = var_7266_begin_0, end = var_7266_end_0, end_mask = var_7266_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_7266_cast_fp16")]; tensor var_7270_begin_0 = const()[name = tensor("op_7270_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_7270_end_0 = const()[name = tensor("op_7270_end_0"), val = tensor([2, 640, 1, 256])]; tensor var_7270_end_mask_0 = const()[name = tensor("op_7270_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7270_cast_fp16 = slice_by_index(begin = var_7270_begin_0, end = var_7270_end_0, end_mask = var_7270_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_7270_cast_fp16")]; tensor var_7274_begin_0 = const()[name = tensor("op_7274_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_7274_end_0 = const()[name = tensor("op_7274_end_0"), val = tensor([2, 800, 1, 256])]; tensor var_7274_end_mask_0 = const()[name = tensor("op_7274_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7274_cast_fp16 = slice_by_index(begin = var_7274_begin_0, end = var_7274_end_0, end_mask = var_7274_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_7274_cast_fp16")]; tensor var_7278_begin_0 = const()[name = tensor("op_7278_begin_0"), val = tensor([0, 800, 0, 0])]; tensor var_7278_end_0 = const()[name = tensor("op_7278_end_0"), val = tensor([2, 960, 1, 256])]; tensor var_7278_end_mask_0 = const()[name = tensor("op_7278_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7278_cast_fp16 = slice_by_index(begin = var_7278_begin_0, end = var_7278_end_0, end_mask = var_7278_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_7278_cast_fp16")]; tensor var_7282_begin_0 = const()[name = tensor("op_7282_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_7282_end_0 = const()[name = tensor("op_7282_end_0"), val = tensor([2, 1120, 1, 256])]; tensor var_7282_end_mask_0 = const()[name = tensor("op_7282_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7282_cast_fp16 = slice_by_index(begin = var_7282_begin_0, end = var_7282_end_0, end_mask = var_7282_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_7282_cast_fp16")]; tensor var_7286_begin_0 = const()[name = tensor("op_7286_begin_0"), val = tensor([0, 1120, 0, 0])]; tensor var_7286_end_0 = const()[name = tensor("op_7286_end_0"), val = tensor([2, 1, 1, 256])]; tensor var_7286_end_mask_0 = const()[name = tensor("op_7286_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_7286_cast_fp16 = slice_by_index(begin = var_7286_begin_0, end = var_7286_end_0, end_mask = var_7286_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_7286_cast_fp16")]; tensor var_7290_equation_0 = const()[name = tensor("op_7290_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7290_cast_fp16 = einsum(equation = var_7290_equation_0, values = (var_7228_cast_fp16, var_7193_cast_fp16))[name = tensor("op_7290_cast_fp16")]; tensor var_7291_to_fp16 = const()[name = tensor("op_7291_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_129_cast_fp16 = mul(x = var_7290_cast_fp16, y = var_7291_to_fp16)[name = tensor("aw_129_cast_fp16")]; tensor var_7294_equation_0 = const()[name = tensor("op_7294_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7294_cast_fp16 = einsum(equation = var_7294_equation_0, values = (var_7232_cast_fp16, var_7197_cast_fp16))[name = tensor("op_7294_cast_fp16")]; tensor var_7295_to_fp16 = const()[name = tensor("op_7295_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_131_cast_fp16 = mul(x = var_7294_cast_fp16, y = var_7295_to_fp16)[name = tensor("aw_131_cast_fp16")]; tensor var_7298_equation_0 = const()[name = tensor("op_7298_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7298_cast_fp16 = einsum(equation = var_7298_equation_0, values = (var_7236_cast_fp16, var_7201_cast_fp16))[name = tensor("op_7298_cast_fp16")]; tensor var_7299_to_fp16 = const()[name = tensor("op_7299_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_133_cast_fp16 = mul(x = var_7298_cast_fp16, y = var_7299_to_fp16)[name = tensor("aw_133_cast_fp16")]; tensor var_7302_equation_0 = const()[name = tensor("op_7302_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7302_cast_fp16 = einsum(equation = var_7302_equation_0, values = (var_7240_cast_fp16, var_7205_cast_fp16))[name = tensor("op_7302_cast_fp16")]; tensor var_7303_to_fp16 = const()[name = tensor("op_7303_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_135_cast_fp16 = mul(x = var_7302_cast_fp16, y = var_7303_to_fp16)[name = tensor("aw_135_cast_fp16")]; tensor var_7306_equation_0 = const()[name = tensor("op_7306_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7306_cast_fp16 = einsum(equation = var_7306_equation_0, values = (var_7244_cast_fp16, var_7209_cast_fp16))[name = tensor("op_7306_cast_fp16")]; tensor var_7307_to_fp16 = const()[name = tensor("op_7307_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_137_cast_fp16 = mul(x = var_7306_cast_fp16, y = var_7307_to_fp16)[name = tensor("aw_137_cast_fp16")]; tensor var_7310_equation_0 = const()[name = tensor("op_7310_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7310_cast_fp16 = einsum(equation = var_7310_equation_0, values = (var_7248_cast_fp16, var_7213_cast_fp16))[name = tensor("op_7310_cast_fp16")]; tensor var_7311_to_fp16 = const()[name = tensor("op_7311_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_139_cast_fp16 = mul(x = var_7310_cast_fp16, y = var_7311_to_fp16)[name = tensor("aw_139_cast_fp16")]; tensor var_7314_equation_0 = const()[name = tensor("op_7314_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7314_cast_fp16 = einsum(equation = var_7314_equation_0, values = (var_7252_cast_fp16, var_7217_cast_fp16))[name = tensor("op_7314_cast_fp16")]; tensor var_7315_to_fp16 = const()[name = tensor("op_7315_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_141_cast_fp16 = mul(x = var_7314_cast_fp16, y = var_7315_to_fp16)[name = tensor("aw_141_cast_fp16")]; tensor var_7318_equation_0 = const()[name = tensor("op_7318_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7318_cast_fp16 = einsum(equation = var_7318_equation_0, values = (var_7256_cast_fp16, var_7221_cast_fp16))[name = tensor("op_7318_cast_fp16")]; tensor var_7319_to_fp16 = const()[name = tensor("op_7319_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_143_cast_fp16 = mul(x = var_7318_cast_fp16, y = var_7319_to_fp16)[name = tensor("aw_143_cast_fp16")]; tensor var_7321_cast_fp16 = softmax(axis = var_6522, x = aw_129_cast_fp16)[name = tensor("op_7321_cast_fp16")]; tensor var_7322_cast_fp16 = softmax(axis = var_6522, x = aw_131_cast_fp16)[name = tensor("op_7322_cast_fp16")]; tensor var_7323_cast_fp16 = softmax(axis = var_6522, x = aw_133_cast_fp16)[name = tensor("op_7323_cast_fp16")]; tensor var_7324_cast_fp16 = softmax(axis = var_6522, x = aw_135_cast_fp16)[name = tensor("op_7324_cast_fp16")]; tensor var_7325_cast_fp16 = softmax(axis = var_6522, x = aw_137_cast_fp16)[name = tensor("op_7325_cast_fp16")]; tensor var_7326_cast_fp16 = softmax(axis = var_6522, x = aw_139_cast_fp16)[name = tensor("op_7326_cast_fp16")]; tensor var_7327_cast_fp16 = softmax(axis = var_6522, x = aw_141_cast_fp16)[name = tensor("op_7327_cast_fp16")]; tensor var_7328_cast_fp16 = softmax(axis = var_6522, x = aw_143_cast_fp16)[name = tensor("op_7328_cast_fp16")]; tensor var_7330_equation_0 = const()[name = tensor("op_7330_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7330_cast_fp16 = einsum(equation = var_7330_equation_0, values = (var_7258_cast_fp16, var_7321_cast_fp16))[name = tensor("op_7330_cast_fp16")]; tensor var_7332_equation_0 = const()[name = tensor("op_7332_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7332_cast_fp16 = einsum(equation = var_7332_equation_0, values = (var_7262_cast_fp16, var_7322_cast_fp16))[name = tensor("op_7332_cast_fp16")]; tensor var_7334_equation_0 = const()[name = tensor("op_7334_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7334_cast_fp16 = einsum(equation = var_7334_equation_0, values = (var_7266_cast_fp16, var_7323_cast_fp16))[name = tensor("op_7334_cast_fp16")]; tensor var_7336_equation_0 = const()[name = tensor("op_7336_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7336_cast_fp16 = einsum(equation = var_7336_equation_0, values = (var_7270_cast_fp16, var_7324_cast_fp16))[name = tensor("op_7336_cast_fp16")]; tensor var_7338_equation_0 = const()[name = tensor("op_7338_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7338_cast_fp16 = einsum(equation = var_7338_equation_0, values = (var_7274_cast_fp16, var_7325_cast_fp16))[name = tensor("op_7338_cast_fp16")]; tensor var_7340_equation_0 = const()[name = tensor("op_7340_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7340_cast_fp16 = einsum(equation = var_7340_equation_0, values = (var_7278_cast_fp16, var_7326_cast_fp16))[name = tensor("op_7340_cast_fp16")]; tensor var_7342_equation_0 = const()[name = tensor("op_7342_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7342_cast_fp16 = einsum(equation = var_7342_equation_0, values = (var_7282_cast_fp16, var_7327_cast_fp16))[name = tensor("op_7342_cast_fp16")]; tensor var_7344_equation_0 = const()[name = tensor("op_7344_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7344_cast_fp16 = einsum(equation = var_7344_equation_0, values = (var_7286_cast_fp16, var_7328_cast_fp16))[name = tensor("op_7344_cast_fp16")]; tensor input_327_interleave_0 = const()[name = tensor("input_327_interleave_0"), val = tensor(false)]; tensor input_327_cast_fp16 = concat(axis = var_6522, interleave = input_327_interleave_0, values = (var_7330_cast_fp16, var_7332_cast_fp16, var_7334_cast_fp16, var_7336_cast_fp16, var_7338_cast_fp16, var_7340_cast_fp16, var_7342_cast_fp16, var_7344_cast_fp16))[name = tensor("input_327_cast_fp16")]; tensor var_7354_pad_type_0 = const()[name = tensor("op_7354_pad_type_0"), val = tensor("valid")]; tensor var_7354_strides_0 = const()[name = tensor("op_7354_strides_0"), val = tensor([1, 1])]; tensor var_7354_pad_0 = const()[name = tensor("op_7354_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_7354_dilations_0 = const()[name = tensor("op_7354_dilations_0"), val = tensor([1, 1])]; tensor var_7354_groups_0 = const()[name = tensor("op_7354_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(413705216)))]; tensor up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(416982080)))]; tensor var_7354_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_7354_dilations_0, groups = var_7354_groups_0, pad = var_7354_pad_0, pad_type = var_7354_pad_type_0, strides = var_7354_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_327_cast_fp16)[name = tensor("op_7354_cast_fp16")]; tensor inputs_51_cast_fp16 = add(x = var_7354_cast_fp16, y = inputs_49_cast_fp16)[name = tensor("inputs_51_cast_fp16")]; tensor hidden_states_189_axes_0 = const()[name = tensor("hidden_states_189_axes_0"), val = tensor([1])]; tensor hidden_states_189_gamma_0_to_fp16 = const()[name = tensor("hidden_states_189_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(416984704)))]; tensor hidden_states_189_beta_0_to_fp16 = const()[name = tensor("hidden_states_189_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(416987328)))]; tensor var_7364_to_fp16 = const()[name = tensor("op_7364_to_fp16"), val = tensor(0x1.5p-17)]; tensor hidden_states_189_cast_fp16 = layer_norm(axes = hidden_states_189_axes_0, beta = hidden_states_189_beta_0_to_fp16, epsilon = var_7364_to_fp16, gamma = hidden_states_189_gamma_0_to_fp16, x = inputs_51_cast_fp16)[name = tensor("hidden_states_189_cast_fp16")]; tensor q_35_pad_type_0 = const()[name = tensor("q_35_pad_type_0"), val = tensor("valid")]; tensor q_35_strides_0 = const()[name = tensor("q_35_strides_0"), val = tensor([1, 1])]; tensor q_35_pad_0 = const()[name = tensor("q_35_pad_0"), val = tensor([0, 0, 0, 0])]; tensor q_35_dilations_0 = const()[name = tensor("q_35_dilations_0"), val = tensor([1, 1])]; tensor q_35_groups_0 = const()[name = tensor("q_35_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(416989952)))]; tensor q_35_cast_fp16 = conv(dilations = q_35_dilations_0, groups = q_35_groups_0, pad = q_35_pad_0, pad_type = q_35_pad_type_0, strides = q_35_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_189_cast_fp16)[name = tensor("q_35_cast_fp16")]; tensor k_69_pad_type_0 = const()[name = tensor("k_69_pad_type_0"), val = tensor("valid")]; tensor k_69_strides_0 = const()[name = tensor("k_69_strides_0"), val = tensor([1, 1])]; tensor k_69_pad_0 = const()[name = tensor("k_69_pad_0"), val = tensor([0, 0, 0, 0])]; tensor k_69_dilations_0 = const()[name = tensor("k_69_dilations_0"), val = tensor([1, 1])]; tensor k_69_groups_0 = const()[name = tensor("k_69_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(420266816)))]; tensor k_69_cast_fp16 = conv(dilations = k_69_dilations_0, groups = k_69_groups_0, pad = k_69_pad_0, pad_type = k_69_pad_type_0, strides = k_69_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_69_cast_fp16")]; tensor v_35_pad_type_0 = const()[name = tensor("v_35_pad_type_0"), val = tensor("valid")]; tensor v_35_strides_0 = const()[name = tensor("v_35_strides_0"), val = tensor([1, 1])]; tensor v_35_pad_0 = const()[name = tensor("v_35_pad_0"), val = tensor([0, 0, 0, 0])]; tensor v_35_dilations_0 = const()[name = tensor("v_35_dilations_0"), val = tensor([1, 1])]; tensor v_35_groups_0 = const()[name = tensor("v_35_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(422232960)))]; tensor v_35_cast_fp16 = conv(dilations = v_35_dilations_0, groups = v_35_groups_0, pad = v_35_pad_0, pad_type = v_35_pad_type_0, strides = v_35_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_35_cast_fp16")]; tensor var_7397_begin_0 = const()[name = tensor("op_7397_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7397_end_0 = const()[name = tensor("op_7397_end_0"), val = tensor([2, 160, 1, 256])]; tensor var_7397_end_mask_0 = const()[name = tensor("op_7397_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7397_cast_fp16 = slice_by_index(begin = var_7397_begin_0, end = var_7397_end_0, end_mask = var_7397_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7397_cast_fp16")]; tensor var_7401_begin_0 = const()[name = tensor("op_7401_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_7401_end_0 = const()[name = tensor("op_7401_end_0"), val = tensor([2, 320, 1, 256])]; tensor var_7401_end_mask_0 = const()[name = tensor("op_7401_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7401_cast_fp16 = slice_by_index(begin = var_7401_begin_0, end = var_7401_end_0, end_mask = var_7401_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7401_cast_fp16")]; tensor var_7405_begin_0 = const()[name = tensor("op_7405_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_7405_end_0 = const()[name = tensor("op_7405_end_0"), val = tensor([2, 480, 1, 256])]; tensor var_7405_end_mask_0 = const()[name = tensor("op_7405_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7405_cast_fp16 = slice_by_index(begin = var_7405_begin_0, end = var_7405_end_0, end_mask = var_7405_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7405_cast_fp16")]; tensor var_7409_begin_0 = const()[name = tensor("op_7409_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_7409_end_0 = const()[name = tensor("op_7409_end_0"), val = tensor([2, 640, 1, 256])]; tensor var_7409_end_mask_0 = const()[name = tensor("op_7409_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7409_cast_fp16 = slice_by_index(begin = var_7409_begin_0, end = var_7409_end_0, end_mask = var_7409_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7409_cast_fp16")]; tensor var_7413_begin_0 = const()[name = tensor("op_7413_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_7413_end_0 = const()[name = tensor("op_7413_end_0"), val = tensor([2, 800, 1, 256])]; tensor var_7413_end_mask_0 = const()[name = tensor("op_7413_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7413_cast_fp16 = slice_by_index(begin = var_7413_begin_0, end = var_7413_end_0, end_mask = var_7413_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7413_cast_fp16")]; tensor var_7417_begin_0 = const()[name = tensor("op_7417_begin_0"), val = tensor([0, 800, 0, 0])]; tensor var_7417_end_0 = const()[name = tensor("op_7417_end_0"), val = tensor([2, 960, 1, 256])]; tensor var_7417_end_mask_0 = const()[name = tensor("op_7417_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7417_cast_fp16 = slice_by_index(begin = var_7417_begin_0, end = var_7417_end_0, end_mask = var_7417_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7417_cast_fp16")]; tensor var_7421_begin_0 = const()[name = tensor("op_7421_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_7421_end_0 = const()[name = tensor("op_7421_end_0"), val = tensor([2, 1120, 1, 256])]; tensor var_7421_end_mask_0 = const()[name = tensor("op_7421_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7421_cast_fp16 = slice_by_index(begin = var_7421_begin_0, end = var_7421_end_0, end_mask = var_7421_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7421_cast_fp16")]; tensor var_7425_begin_0 = const()[name = tensor("op_7425_begin_0"), val = tensor([0, 1120, 0, 0])]; tensor var_7425_end_0 = const()[name = tensor("op_7425_end_0"), val = tensor([2, 1, 1, 256])]; tensor var_7425_end_mask_0 = const()[name = tensor("op_7425_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_7425_cast_fp16 = slice_by_index(begin = var_7425_begin_0, end = var_7425_end_0, end_mask = var_7425_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7425_cast_fp16")]; tensor k_71_perm_0 = const()[name = tensor("k_71_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_7432_begin_0 = const()[name = tensor("op_7432_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7432_end_0 = const()[name = tensor("op_7432_end_0"), val = tensor([2, 77, 1, 160])]; tensor var_7432_end_mask_0 = const()[name = tensor("op_7432_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_71_cast_fp16 = transpose(perm = k_71_perm_0, x = k_69_cast_fp16)[name = tensor("transpose_14")]; tensor var_7432_cast_fp16 = slice_by_index(begin = var_7432_begin_0, end = var_7432_end_0, end_mask = var_7432_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_7432_cast_fp16")]; tensor var_7436_begin_0 = const()[name = tensor("op_7436_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_7436_end_0 = const()[name = tensor("op_7436_end_0"), val = tensor([2, 77, 1, 320])]; tensor var_7436_end_mask_0 = const()[name = tensor("op_7436_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7436_cast_fp16 = slice_by_index(begin = var_7436_begin_0, end = var_7436_end_0, end_mask = var_7436_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_7436_cast_fp16")]; tensor var_7440_begin_0 = const()[name = tensor("op_7440_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_7440_end_0 = const()[name = tensor("op_7440_end_0"), val = tensor([2, 77, 1, 480])]; tensor var_7440_end_mask_0 = const()[name = tensor("op_7440_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7440_cast_fp16 = slice_by_index(begin = var_7440_begin_0, end = var_7440_end_0, end_mask = var_7440_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_7440_cast_fp16")]; tensor var_7444_begin_0 = const()[name = tensor("op_7444_begin_0"), val = tensor([0, 0, 0, 480])]; tensor var_7444_end_0 = const()[name = tensor("op_7444_end_0"), val = tensor([2, 77, 1, 640])]; tensor var_7444_end_mask_0 = const()[name = tensor("op_7444_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7444_cast_fp16 = slice_by_index(begin = var_7444_begin_0, end = var_7444_end_0, end_mask = var_7444_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_7444_cast_fp16")]; tensor var_7448_begin_0 = const()[name = tensor("op_7448_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_7448_end_0 = const()[name = tensor("op_7448_end_0"), val = tensor([2, 77, 1, 800])]; tensor var_7448_end_mask_0 = const()[name = tensor("op_7448_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7448_cast_fp16 = slice_by_index(begin = var_7448_begin_0, end = var_7448_end_0, end_mask = var_7448_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_7448_cast_fp16")]; tensor var_7452_begin_0 = const()[name = tensor("op_7452_begin_0"), val = tensor([0, 0, 0, 800])]; tensor var_7452_end_0 = const()[name = tensor("op_7452_end_0"), val = tensor([2, 77, 1, 960])]; tensor var_7452_end_mask_0 = const()[name = tensor("op_7452_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7452_cast_fp16 = slice_by_index(begin = var_7452_begin_0, end = var_7452_end_0, end_mask = var_7452_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_7452_cast_fp16")]; tensor var_7456_begin_0 = const()[name = tensor("op_7456_begin_0"), val = tensor([0, 0, 0, 960])]; tensor var_7456_end_0 = const()[name = tensor("op_7456_end_0"), val = tensor([2, 77, 1, 1120])]; tensor var_7456_end_mask_0 = const()[name = tensor("op_7456_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7456_cast_fp16 = slice_by_index(begin = var_7456_begin_0, end = var_7456_end_0, end_mask = var_7456_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_7456_cast_fp16")]; tensor var_7460_begin_0 = const()[name = tensor("op_7460_begin_0"), val = tensor([0, 0, 0, 1120])]; tensor var_7460_end_0 = const()[name = tensor("op_7460_end_0"), val = tensor([2, 77, 1, 1])]; tensor var_7460_end_mask_0 = const()[name = tensor("op_7460_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_7460_cast_fp16 = slice_by_index(begin = var_7460_begin_0, end = var_7460_end_0, end_mask = var_7460_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_7460_cast_fp16")]; tensor var_7462_begin_0 = const()[name = tensor("op_7462_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7462_end_0 = const()[name = tensor("op_7462_end_0"), val = tensor([2, 160, 1, 77])]; tensor var_7462_end_mask_0 = const()[name = tensor("op_7462_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7462_cast_fp16 = slice_by_index(begin = var_7462_begin_0, end = var_7462_end_0, end_mask = var_7462_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7462_cast_fp16")]; tensor var_7466_begin_0 = const()[name = tensor("op_7466_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_7466_end_0 = const()[name = tensor("op_7466_end_0"), val = tensor([2, 320, 1, 77])]; tensor var_7466_end_mask_0 = const()[name = tensor("op_7466_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7466_cast_fp16 = slice_by_index(begin = var_7466_begin_0, end = var_7466_end_0, end_mask = var_7466_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7466_cast_fp16")]; tensor var_7470_begin_0 = const()[name = tensor("op_7470_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_7470_end_0 = const()[name = tensor("op_7470_end_0"), val = tensor([2, 480, 1, 77])]; tensor var_7470_end_mask_0 = const()[name = tensor("op_7470_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7470_cast_fp16 = slice_by_index(begin = var_7470_begin_0, end = var_7470_end_0, end_mask = var_7470_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7470_cast_fp16")]; tensor var_7474_begin_0 = const()[name = tensor("op_7474_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_7474_end_0 = const()[name = tensor("op_7474_end_0"), val = tensor([2, 640, 1, 77])]; tensor var_7474_end_mask_0 = const()[name = tensor("op_7474_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7474_cast_fp16 = slice_by_index(begin = var_7474_begin_0, end = var_7474_end_0, end_mask = var_7474_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7474_cast_fp16")]; tensor var_7478_begin_0 = const()[name = tensor("op_7478_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_7478_end_0 = const()[name = tensor("op_7478_end_0"), val = tensor([2, 800, 1, 77])]; tensor var_7478_end_mask_0 = const()[name = tensor("op_7478_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7478_cast_fp16 = slice_by_index(begin = var_7478_begin_0, end = var_7478_end_0, end_mask = var_7478_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7478_cast_fp16")]; tensor var_7482_begin_0 = const()[name = tensor("op_7482_begin_0"), val = tensor([0, 800, 0, 0])]; tensor var_7482_end_0 = const()[name = tensor("op_7482_end_0"), val = tensor([2, 960, 1, 77])]; tensor var_7482_end_mask_0 = const()[name = tensor("op_7482_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7482_cast_fp16 = slice_by_index(begin = var_7482_begin_0, end = var_7482_end_0, end_mask = var_7482_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7482_cast_fp16")]; tensor var_7486_begin_0 = const()[name = tensor("op_7486_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_7486_end_0 = const()[name = tensor("op_7486_end_0"), val = tensor([2, 1120, 1, 77])]; tensor var_7486_end_mask_0 = const()[name = tensor("op_7486_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7486_cast_fp16 = slice_by_index(begin = var_7486_begin_0, end = var_7486_end_0, end_mask = var_7486_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7486_cast_fp16")]; tensor var_7490_begin_0 = const()[name = tensor("op_7490_begin_0"), val = tensor([0, 1120, 0, 0])]; tensor var_7490_end_0 = const()[name = tensor("op_7490_end_0"), val = tensor([2, 1, 1, 77])]; tensor var_7490_end_mask_0 = const()[name = tensor("op_7490_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_7490_cast_fp16 = slice_by_index(begin = var_7490_begin_0, end = var_7490_end_0, end_mask = var_7490_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7490_cast_fp16")]; tensor var_7494_equation_0 = const()[name = tensor("op_7494_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7494_cast_fp16 = einsum(equation = var_7494_equation_0, values = (var_7432_cast_fp16, var_7397_cast_fp16))[name = tensor("op_7494_cast_fp16")]; tensor var_7495_to_fp16 = const()[name = tensor("op_7495_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_145_cast_fp16 = mul(x = var_7494_cast_fp16, y = var_7495_to_fp16)[name = tensor("aw_145_cast_fp16")]; tensor var_7498_equation_0 = const()[name = tensor("op_7498_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7498_cast_fp16 = einsum(equation = var_7498_equation_0, values = (var_7436_cast_fp16, var_7401_cast_fp16))[name = tensor("op_7498_cast_fp16")]; tensor var_7499_to_fp16 = const()[name = tensor("op_7499_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_147_cast_fp16 = mul(x = var_7498_cast_fp16, y = var_7499_to_fp16)[name = tensor("aw_147_cast_fp16")]; tensor var_7502_equation_0 = const()[name = tensor("op_7502_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7502_cast_fp16 = einsum(equation = var_7502_equation_0, values = (var_7440_cast_fp16, var_7405_cast_fp16))[name = tensor("op_7502_cast_fp16")]; tensor var_7503_to_fp16 = const()[name = tensor("op_7503_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_149_cast_fp16 = mul(x = var_7502_cast_fp16, y = var_7503_to_fp16)[name = tensor("aw_149_cast_fp16")]; tensor var_7506_equation_0 = const()[name = tensor("op_7506_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7506_cast_fp16 = einsum(equation = var_7506_equation_0, values = (var_7444_cast_fp16, var_7409_cast_fp16))[name = tensor("op_7506_cast_fp16")]; tensor var_7507_to_fp16 = const()[name = tensor("op_7507_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_151_cast_fp16 = mul(x = var_7506_cast_fp16, y = var_7507_to_fp16)[name = tensor("aw_151_cast_fp16")]; tensor var_7510_equation_0 = const()[name = tensor("op_7510_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7510_cast_fp16 = einsum(equation = var_7510_equation_0, values = (var_7448_cast_fp16, var_7413_cast_fp16))[name = tensor("op_7510_cast_fp16")]; tensor var_7511_to_fp16 = const()[name = tensor("op_7511_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_153_cast_fp16 = mul(x = var_7510_cast_fp16, y = var_7511_to_fp16)[name = tensor("aw_153_cast_fp16")]; tensor var_7514_equation_0 = const()[name = tensor("op_7514_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7514_cast_fp16 = einsum(equation = var_7514_equation_0, values = (var_7452_cast_fp16, var_7417_cast_fp16))[name = tensor("op_7514_cast_fp16")]; tensor var_7515_to_fp16 = const()[name = tensor("op_7515_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_155_cast_fp16 = mul(x = var_7514_cast_fp16, y = var_7515_to_fp16)[name = tensor("aw_155_cast_fp16")]; tensor var_7518_equation_0 = const()[name = tensor("op_7518_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7518_cast_fp16 = einsum(equation = var_7518_equation_0, values = (var_7456_cast_fp16, var_7421_cast_fp16))[name = tensor("op_7518_cast_fp16")]; tensor var_7519_to_fp16 = const()[name = tensor("op_7519_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_157_cast_fp16 = mul(x = var_7518_cast_fp16, y = var_7519_to_fp16)[name = tensor("aw_157_cast_fp16")]; tensor var_7522_equation_0 = const()[name = tensor("op_7522_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7522_cast_fp16 = einsum(equation = var_7522_equation_0, values = (var_7460_cast_fp16, var_7425_cast_fp16))[name = tensor("op_7522_cast_fp16")]; tensor var_7523_to_fp16 = const()[name = tensor("op_7523_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_159_cast_fp16 = mul(x = var_7522_cast_fp16, y = var_7523_to_fp16)[name = tensor("aw_159_cast_fp16")]; tensor var_7525_cast_fp16 = softmax(axis = var_6522, x = aw_145_cast_fp16)[name = tensor("op_7525_cast_fp16")]; tensor var_7526_cast_fp16 = softmax(axis = var_6522, x = aw_147_cast_fp16)[name = tensor("op_7526_cast_fp16")]; tensor var_7527_cast_fp16 = softmax(axis = var_6522, x = aw_149_cast_fp16)[name = tensor("op_7527_cast_fp16")]; tensor var_7528_cast_fp16 = softmax(axis = var_6522, x = aw_151_cast_fp16)[name = tensor("op_7528_cast_fp16")]; tensor var_7529_cast_fp16 = softmax(axis = var_6522, x = aw_153_cast_fp16)[name = tensor("op_7529_cast_fp16")]; tensor var_7530_cast_fp16 = softmax(axis = var_6522, x = aw_155_cast_fp16)[name = tensor("op_7530_cast_fp16")]; tensor var_7531_cast_fp16 = softmax(axis = var_6522, x = aw_157_cast_fp16)[name = tensor("op_7531_cast_fp16")]; tensor var_7532_cast_fp16 = softmax(axis = var_6522, x = aw_159_cast_fp16)[name = tensor("op_7532_cast_fp16")]; tensor var_7534_equation_0 = const()[name = tensor("op_7534_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7534_cast_fp16 = einsum(equation = var_7534_equation_0, values = (var_7462_cast_fp16, var_7525_cast_fp16))[name = tensor("op_7534_cast_fp16")]; tensor var_7536_equation_0 = const()[name = tensor("op_7536_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7536_cast_fp16 = einsum(equation = var_7536_equation_0, values = (var_7466_cast_fp16, var_7526_cast_fp16))[name = tensor("op_7536_cast_fp16")]; tensor var_7538_equation_0 = const()[name = tensor("op_7538_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7538_cast_fp16 = einsum(equation = var_7538_equation_0, values = (var_7470_cast_fp16, var_7527_cast_fp16))[name = tensor("op_7538_cast_fp16")]; tensor var_7540_equation_0 = const()[name = tensor("op_7540_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7540_cast_fp16 = einsum(equation = var_7540_equation_0, values = (var_7474_cast_fp16, var_7528_cast_fp16))[name = tensor("op_7540_cast_fp16")]; tensor var_7542_equation_0 = const()[name = tensor("op_7542_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7542_cast_fp16 = einsum(equation = var_7542_equation_0, values = (var_7478_cast_fp16, var_7529_cast_fp16))[name = tensor("op_7542_cast_fp16")]; tensor var_7544_equation_0 = const()[name = tensor("op_7544_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7544_cast_fp16 = einsum(equation = var_7544_equation_0, values = (var_7482_cast_fp16, var_7530_cast_fp16))[name = tensor("op_7544_cast_fp16")]; tensor var_7546_equation_0 = const()[name = tensor("op_7546_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7546_cast_fp16 = einsum(equation = var_7546_equation_0, values = (var_7486_cast_fp16, var_7531_cast_fp16))[name = tensor("op_7546_cast_fp16")]; tensor var_7548_equation_0 = const()[name = tensor("op_7548_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7548_cast_fp16 = einsum(equation = var_7548_equation_0, values = (var_7490_cast_fp16, var_7532_cast_fp16))[name = tensor("op_7548_cast_fp16")]; tensor input_329_interleave_0 = const()[name = tensor("input_329_interleave_0"), val = tensor(false)]; tensor input_329_cast_fp16 = concat(axis = var_6522, interleave = input_329_interleave_0, values = (var_7534_cast_fp16, var_7536_cast_fp16, var_7538_cast_fp16, var_7540_cast_fp16, var_7542_cast_fp16, var_7544_cast_fp16, var_7546_cast_fp16, var_7548_cast_fp16))[name = tensor("input_329_cast_fp16")]; tensor var_7558_pad_type_0 = const()[name = tensor("op_7558_pad_type_0"), val = tensor("valid")]; tensor var_7558_strides_0 = const()[name = tensor("op_7558_strides_0"), val = tensor([1, 1])]; tensor var_7558_pad_0 = const()[name = tensor("op_7558_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_7558_dilations_0 = const()[name = tensor("op_7558_dilations_0"), val = tensor([1, 1])]; tensor var_7558_groups_0 = const()[name = tensor("op_7558_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(424199104)))]; tensor up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(427475968)))]; tensor var_7558_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_7558_dilations_0, groups = var_7558_groups_0, pad = var_7558_pad_0, pad_type = var_7558_pad_type_0, strides = var_7558_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_329_cast_fp16)[name = tensor("op_7558_cast_fp16")]; tensor inputs_53_cast_fp16 = add(x = var_7558_cast_fp16, y = inputs_51_cast_fp16)[name = tensor("inputs_53_cast_fp16")]; tensor input_331_axes_0 = const()[name = tensor("input_331_axes_0"), val = tensor([1])]; tensor input_331_gamma_0_to_fp16 = const()[name = tensor("input_331_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(427478592)))]; tensor input_331_beta_0_to_fp16 = const()[name = tensor("input_331_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(427481216)))]; tensor var_7568_to_fp16 = const()[name = tensor("op_7568_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_331_cast_fp16 = layer_norm(axes = input_331_axes_0, beta = input_331_beta_0_to_fp16, epsilon = var_7568_to_fp16, gamma = input_331_gamma_0_to_fp16, x = inputs_53_cast_fp16)[name = tensor("input_331_cast_fp16")]; tensor var_7588_pad_type_0 = const()[name = tensor("op_7588_pad_type_0"), val = tensor("valid")]; tensor var_7588_strides_0 = const()[name = tensor("op_7588_strides_0"), val = tensor([1, 1])]; tensor var_7588_pad_0 = const()[name = tensor("op_7588_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_7588_dilations_0 = const()[name = tensor("op_7588_dilations_0"), val = tensor([1, 1])]; tensor var_7588_groups_0 = const()[name = tensor("op_7588_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(427483840)))]; tensor up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(453698304)))]; tensor var_7588_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_7588_dilations_0, groups = var_7588_groups_0, pad = var_7588_pad_0, pad_type = var_7588_pad_type_0, strides = var_7588_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_331_cast_fp16)[name = tensor("op_7588_cast_fp16")]; tensor var_7589_split_sizes_0 = const()[name = tensor("op_7589_split_sizes_0"), val = tensor([5120, 5120])]; tensor var_7589_axis_0 = const()[name = tensor("op_7589_axis_0"), val = tensor(1)]; tensor var_7589_cast_fp16_0, tensor var_7589_cast_fp16_1 = split(axis = var_7589_axis_0, split_sizes = var_7589_split_sizes_0, x = var_7588_cast_fp16)[name = tensor("op_7589_cast_fp16")]; tensor var_7591_mode_0 = const()[name = tensor("op_7591_mode_0"), val = tensor("EXACT")]; tensor var_7591_cast_fp16 = gelu(mode = var_7591_mode_0, x = var_7589_cast_fp16_1)[name = tensor("op_7591_cast_fp16")]; tensor input_333_cast_fp16 = mul(x = var_7589_cast_fp16_0, y = var_7591_cast_fp16)[name = tensor("input_333_cast_fp16")]; tensor var_7599_pad_type_0 = const()[name = tensor("op_7599_pad_type_0"), val = tensor("valid")]; tensor var_7599_strides_0 = const()[name = tensor("op_7599_strides_0"), val = tensor([1, 1])]; tensor var_7599_pad_0 = const()[name = tensor("op_7599_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_7599_dilations_0 = const()[name = tensor("op_7599_dilations_0"), val = tensor([1, 1])]; tensor var_7599_groups_0 = const()[name = tensor("op_7599_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(453718848)))]; tensor up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(466826112)))]; tensor var_7599_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_7599_dilations_0, groups = var_7599_groups_0, pad = var_7599_pad_0, pad_type = var_7599_pad_type_0, strides = var_7599_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_333_cast_fp16)[name = tensor("op_7599_cast_fp16")]; tensor hidden_states_193_cast_fp16 = add(x = var_7599_cast_fp16, y = inputs_53_cast_fp16)[name = tensor("hidden_states_193_cast_fp16")]; tensor var_7601 = const()[name = tensor("op_7601"), val = tensor([2, 1280, 16, 16])]; tensor input_335_cast_fp16 = reshape(shape = var_7601, x = hidden_states_193_cast_fp16)[name = tensor("input_335_cast_fp16")]; tensor hidden_states_195_pad_type_0 = const()[name = tensor("hidden_states_195_pad_type_0"), val = tensor("valid")]; tensor hidden_states_195_strides_0 = const()[name = tensor("hidden_states_195_strides_0"), val = tensor([1, 1])]; tensor hidden_states_195_pad_0 = const()[name = tensor("hidden_states_195_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_195_dilations_0 = const()[name = tensor("hidden_states_195_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_195_groups_0 = const()[name = tensor("hidden_states_195_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_1_proj_out_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(466828736)))]; tensor up_blocks_1_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(470105600)))]; tensor hidden_states_195_cast_fp16 = conv(bias = up_blocks_1_attentions_1_proj_out_bias_to_fp16, dilations = hidden_states_195_dilations_0, groups = hidden_states_195_groups_0, pad = hidden_states_195_pad_0, pad_type = hidden_states_195_pad_type_0, strides = hidden_states_195_strides_0, weight = up_blocks_1_attentions_1_proj_out_weight_to_fp16, x = input_335_cast_fp16)[name = tensor("hidden_states_195_cast_fp16")]; tensor hidden_states_197_cast_fp16 = add(x = hidden_states_195_cast_fp16, y = hidden_states_183_cast_fp16)[name = tensor("hidden_states_197_cast_fp16")]; tensor input_337_interleave_0 = const()[name = tensor("input_337_interleave_0"), val = tensor(false)]; tensor cast_1 = cast(dtype = cast_1_dtype_0, x = input_117_cast_fp16)[name = tensor("cast_6")]; tensor input_337_cast_fp16 = concat(axis = var_6522, interleave = input_337_interleave_0, values = (hidden_states_197_cast_fp16, cast_1))[name = tensor("input_337_cast_fp16")]; tensor reshape_156_shape_0 = const()[name = tensor("reshape_156_shape_0"), val = tensor([2, 32, 60, 16, 16])]; tensor reshape_156_cast_fp16 = reshape(shape = reshape_156_shape_0, x = input_337_cast_fp16)[name = tensor("reshape_156_cast_fp16")]; tensor reduce_mean_117_axes_0 = const()[name = tensor("reduce_mean_117_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_117_keep_dims_0 = const()[name = tensor("reduce_mean_117_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_117_cast_fp16 = reduce_mean(axes = reduce_mean_117_axes_0, keep_dims = reduce_mean_117_keep_dims_0, x = reshape_156_cast_fp16)[name = tensor("reduce_mean_117_cast_fp16")]; tensor sub_78_cast_fp16 = sub(x = reshape_156_cast_fp16, y = reduce_mean_117_cast_fp16)[name = tensor("sub_78_cast_fp16")]; tensor square_39_cast_fp16 = square(x = sub_78_cast_fp16)[name = tensor("square_39_cast_fp16")]; tensor reduce_mean_119_axes_0 = const()[name = tensor("reduce_mean_119_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_119_keep_dims_0 = const()[name = tensor("reduce_mean_119_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_119_cast_fp16 = reduce_mean(axes = reduce_mean_119_axes_0, keep_dims = reduce_mean_119_keep_dims_0, x = square_39_cast_fp16)[name = tensor("reduce_mean_119_cast_fp16")]; tensor add_78_y_0_to_fp16 = const()[name = tensor("add_78_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_78_cast_fp16 = add(x = reduce_mean_119_cast_fp16, y = add_78_y_0_to_fp16)[name = tensor("add_78_cast_fp16")]; tensor sqrt_39_cast_fp16 = sqrt(x = add_78_cast_fp16)[name = tensor("sqrt_39_cast_fp16")]; tensor real_div_39_cast_fp16 = real_div(x = sub_78_cast_fp16, y = sqrt_39_cast_fp16)[name = tensor("real_div_39_cast_fp16")]; tensor reshape_157_shape_0 = const()[name = tensor("reshape_157_shape_0"), val = tensor([2, 1920, 16, 16])]; tensor reshape_157_cast_fp16 = reshape(shape = reshape_157_shape_0, x = real_div_39_cast_fp16)[name = tensor("reshape_157_cast_fp16")]; tensor add_79_mean_0_to_fp16 = const()[name = tensor("add_79_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(470108224)))]; tensor add_79_variance_0_to_fp16 = const()[name = tensor("add_79_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(470112128)))]; tensor add_79_gamma_0_to_fp16 = const()[name = tensor("add_79_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(470116032)))]; tensor add_79_beta_0_to_fp16 = const()[name = tensor("add_79_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(470119936)))]; tensor add_79_epsilon_0_to_fp16 = const()[name = tensor("add_79_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_79_cast_fp16 = batch_norm(beta = add_79_beta_0_to_fp16, epsilon = add_79_epsilon_0_to_fp16, gamma = add_79_gamma_0_to_fp16, mean = add_79_mean_0_to_fp16, variance = add_79_variance_0_to_fp16, x = reshape_157_cast_fp16)[name = tensor("add_79_cast_fp16")]; tensor input_341_cast_fp16 = silu(x = add_79_cast_fp16)[name = tensor("input_341_cast_fp16")]; tensor hidden_states_199_pad_type_0 = const()[name = tensor("hidden_states_199_pad_type_0"), val = tensor("custom")]; tensor hidden_states_199_pad_0 = const()[name = tensor("hidden_states_199_pad_0"), val = tensor([1, 1, 1, 1])]; tensor hidden_states_199_strides_0 = const()[name = tensor("hidden_states_199_strides_0"), val = tensor([1, 1])]; tensor hidden_states_199_dilations_0 = const()[name = tensor("hidden_states_199_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_199_groups_0 = const()[name = tensor("hidden_states_199_groups_0"), val = tensor(1)]; tensor up_blocks_1_resnets_2_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(470123840)))]; tensor up_blocks_1_resnets_2_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(514360704)))]; tensor hidden_states_199_cast_fp16 = conv(bias = up_blocks_1_resnets_2_conv1_bias_to_fp16, dilations = hidden_states_199_dilations_0, groups = hidden_states_199_groups_0, pad = hidden_states_199_pad_0, pad_type = hidden_states_199_pad_type_0, strides = hidden_states_199_strides_0, weight = up_blocks_1_resnets_2_conv1_weight_to_fp16, x = input_341_cast_fp16)[name = tensor("hidden_states_199_cast_fp16")]; tensor temb_31_pad_type_0 = const()[name = tensor("temb_31_pad_type_0"), val = tensor("valid")]; tensor temb_31_strides_0 = const()[name = tensor("temb_31_strides_0"), val = tensor([1, 1])]; tensor temb_31_pad_0 = const()[name = tensor("temb_31_pad_0"), val = tensor([0, 0, 0, 0])]; tensor temb_31_dilations_0 = const()[name = tensor("temb_31_dilations_0"), val = tensor([1, 1])]; tensor temb_31_groups_0 = const()[name = tensor("temb_31_groups_0"), val = tensor(1)]; tensor up_blocks_1_resnets_2_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(514363328)))]; tensor up_blocks_1_resnets_2_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(517640192)))]; tensor temb_31_cast_fp16 = conv(bias = up_blocks_1_resnets_2_time_emb_proj_bias_to_fp16, dilations = temb_31_dilations_0, groups = temb_31_groups_0, pad = temb_31_pad_0, pad_type = temb_31_pad_type_0, strides = temb_31_strides_0, weight = up_blocks_1_resnets_2_time_emb_proj_weight_to_fp16, x = cast_9)[name = tensor("temb_31_cast_fp16")]; tensor input_345_cast_fp16 = add(x = hidden_states_199_cast_fp16, y = temb_31_cast_fp16)[name = tensor("input_345_cast_fp16")]; tensor reshape_160_shape_0 = const()[name = tensor("reshape_160_shape_0"), val = tensor([2, 32, 40, 16, 16])]; tensor reshape_160_cast_fp16 = reshape(shape = reshape_160_shape_0, x = input_345_cast_fp16)[name = tensor("reshape_160_cast_fp16")]; tensor reduce_mean_120_axes_0 = const()[name = tensor("reduce_mean_120_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_120_keep_dims_0 = const()[name = tensor("reduce_mean_120_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_120_cast_fp16 = reduce_mean(axes = reduce_mean_120_axes_0, keep_dims = reduce_mean_120_keep_dims_0, x = reshape_160_cast_fp16)[name = tensor("reduce_mean_120_cast_fp16")]; tensor sub_80_cast_fp16 = sub(x = reshape_160_cast_fp16, y = reduce_mean_120_cast_fp16)[name = tensor("sub_80_cast_fp16")]; tensor square_40_cast_fp16 = square(x = sub_80_cast_fp16)[name = tensor("square_40_cast_fp16")]; tensor reduce_mean_122_axes_0 = const()[name = tensor("reduce_mean_122_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_122_keep_dims_0 = const()[name = tensor("reduce_mean_122_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_122_cast_fp16 = reduce_mean(axes = reduce_mean_122_axes_0, keep_dims = reduce_mean_122_keep_dims_0, x = square_40_cast_fp16)[name = tensor("reduce_mean_122_cast_fp16")]; tensor add_80_y_0_to_fp16 = const()[name = tensor("add_80_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_80_cast_fp16 = add(x = reduce_mean_122_cast_fp16, y = add_80_y_0_to_fp16)[name = tensor("add_80_cast_fp16")]; tensor sqrt_40_cast_fp16 = sqrt(x = add_80_cast_fp16)[name = tensor("sqrt_40_cast_fp16")]; tensor real_div_40_cast_fp16 = real_div(x = sub_80_cast_fp16, y = sqrt_40_cast_fp16)[name = tensor("real_div_40_cast_fp16")]; tensor reshape_161_shape_0 = const()[name = tensor("reshape_161_shape_0"), val = tensor([2, 1280, 16, 16])]; tensor reshape_161_cast_fp16 = reshape(shape = reshape_161_shape_0, x = real_div_40_cast_fp16)[name = tensor("reshape_161_cast_fp16")]; tensor add_81_gamma_0_to_fp16 = const()[name = tensor("add_81_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(517642816)))]; tensor add_81_beta_0_to_fp16 = const()[name = tensor("add_81_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(517645440)))]; tensor add_81_epsilon_0_to_fp16 = const()[name = tensor("add_81_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_81_cast_fp16 = batch_norm(beta = add_81_beta_0_to_fp16, epsilon = add_81_epsilon_0_to_fp16, gamma = add_81_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_161_cast_fp16)[name = tensor("add_81_cast_fp16")]; tensor input_349_cast_fp16 = silu(x = add_81_cast_fp16)[name = tensor("input_349_cast_fp16")]; tensor hidden_states_201_pad_type_0 = const()[name = tensor("hidden_states_201_pad_type_0"), val = tensor("custom")]; tensor hidden_states_201_pad_0 = const()[name = tensor("hidden_states_201_pad_0"), val = tensor([1, 1, 1, 1])]; tensor hidden_states_201_strides_0 = const()[name = tensor("hidden_states_201_strides_0"), val = tensor([1, 1])]; tensor hidden_states_201_dilations_0 = const()[name = tensor("hidden_states_201_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_201_groups_0 = const()[name = tensor("hidden_states_201_groups_0"), val = tensor(1)]; tensor up_blocks_1_resnets_2_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(517648064)))]; tensor up_blocks_1_resnets_2_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(547139328)))]; tensor hidden_states_201_cast_fp16 = conv(bias = up_blocks_1_resnets_2_conv2_bias_to_fp16, dilations = hidden_states_201_dilations_0, groups = hidden_states_201_groups_0, pad = hidden_states_201_pad_0, pad_type = hidden_states_201_pad_type_0, strides = hidden_states_201_strides_0, weight = up_blocks_1_resnets_2_conv2_weight_to_fp16, x = input_349_cast_fp16)[name = tensor("hidden_states_201_cast_fp16")]; tensor x_15_pad_type_0 = const()[name = tensor("x_15_pad_type_0"), val = tensor("valid")]; tensor x_15_strides_0 = const()[name = tensor("x_15_strides_0"), val = tensor([1, 1])]; tensor x_15_pad_0 = const()[name = tensor("x_15_pad_0"), val = tensor([0, 0, 0, 0])]; tensor x_15_dilations_0 = const()[name = tensor("x_15_dilations_0"), val = tensor([1, 1])]; tensor x_15_groups_0 = const()[name = tensor("x_15_groups_0"), val = tensor(1)]; tensor up_blocks_1_resnets_2_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(547141952)))]; tensor up_blocks_1_resnets_2_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(552057216)))]; tensor x_15_cast_fp16 = conv(bias = up_blocks_1_resnets_2_conv_shortcut_bias_to_fp16, dilations = x_15_dilations_0, groups = x_15_groups_0, pad = x_15_pad_0, pad_type = x_15_pad_type_0, strides = x_15_strides_0, weight = up_blocks_1_resnets_2_conv_shortcut_weight_to_fp16, x = input_337_cast_fp16)[name = tensor("x_15_cast_fp16")]; tensor hidden_states_203_cast_fp16 = add(x = x_15_cast_fp16, y = hidden_states_201_cast_fp16)[name = tensor("hidden_states_203_cast_fp16")]; tensor reshape_164_shape_0 = const()[name = tensor("reshape_164_shape_0"), val = tensor([2, 32, 40, 16, 16])]; tensor reshape_164_cast_fp16 = reshape(shape = reshape_164_shape_0, x = hidden_states_203_cast_fp16)[name = tensor("reshape_164_cast_fp16")]; tensor reduce_mean_123_axes_0 = const()[name = tensor("reduce_mean_123_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_123_keep_dims_0 = const()[name = tensor("reduce_mean_123_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_123_cast_fp16 = reduce_mean(axes = reduce_mean_123_axes_0, keep_dims = reduce_mean_123_keep_dims_0, x = reshape_164_cast_fp16)[name = tensor("reduce_mean_123_cast_fp16")]; tensor sub_82_cast_fp16 = sub(x = reshape_164_cast_fp16, y = reduce_mean_123_cast_fp16)[name = tensor("sub_82_cast_fp16")]; tensor square_41_cast_fp16 = square(x = sub_82_cast_fp16)[name = tensor("square_41_cast_fp16")]; tensor reduce_mean_125_axes_0 = const()[name = tensor("reduce_mean_125_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_125_keep_dims_0 = const()[name = tensor("reduce_mean_125_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_125_cast_fp16 = reduce_mean(axes = reduce_mean_125_axes_0, keep_dims = reduce_mean_125_keep_dims_0, x = square_41_cast_fp16)[name = tensor("reduce_mean_125_cast_fp16")]; tensor add_82_y_0_to_fp16 = const()[name = tensor("add_82_y_0_to_fp16"), val = tensor(0x1.1p-20)]; tensor add_82_cast_fp16 = add(x = reduce_mean_125_cast_fp16, y = add_82_y_0_to_fp16)[name = tensor("add_82_cast_fp16")]; tensor sqrt_41_cast_fp16 = sqrt(x = add_82_cast_fp16)[name = tensor("sqrt_41_cast_fp16")]; tensor real_div_41_cast_fp16 = real_div(x = sub_82_cast_fp16, y = sqrt_41_cast_fp16)[name = tensor("real_div_41_cast_fp16")]; tensor reshape_165_shape_0 = const()[name = tensor("reshape_165_shape_0"), val = tensor([2, 1280, 16, 16])]; tensor reshape_165_cast_fp16 = reshape(shape = reshape_165_shape_0, x = real_div_41_cast_fp16)[name = tensor("reshape_165_cast_fp16")]; tensor add_83_gamma_0_to_fp16 = const()[name = tensor("add_83_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(552059840)))]; tensor add_83_beta_0_to_fp16 = const()[name = tensor("add_83_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(552062464)))]; tensor add_83_epsilon_0_to_fp16 = const()[name = tensor("add_83_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_83_cast_fp16 = batch_norm(beta = add_83_beta_0_to_fp16, epsilon = add_83_epsilon_0_to_fp16, gamma = add_83_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_165_cast_fp16)[name = tensor("add_83_cast_fp16")]; tensor hidden_states_205_pad_type_0 = const()[name = tensor("hidden_states_205_pad_type_0"), val = tensor("valid")]; tensor hidden_states_205_strides_0 = const()[name = tensor("hidden_states_205_strides_0"), val = tensor([1, 1])]; tensor hidden_states_205_pad_0 = const()[name = tensor("hidden_states_205_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_205_dilations_0 = const()[name = tensor("hidden_states_205_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_205_groups_0 = const()[name = tensor("hidden_states_205_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_2_proj_in_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(552065088)))]; tensor up_blocks_1_attentions_2_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(555341952)))]; tensor hidden_states_205_cast_fp16 = conv(bias = up_blocks_1_attentions_2_proj_in_bias_to_fp16, dilations = hidden_states_205_dilations_0, groups = hidden_states_205_groups_0, pad = hidden_states_205_pad_0, pad_type = hidden_states_205_pad_type_0, strides = hidden_states_205_strides_0, weight = up_blocks_1_attentions_2_proj_in_weight_to_fp16, x = add_83_cast_fp16)[name = tensor("hidden_states_205_cast_fp16")]; tensor var_7681 = const()[name = tensor("op_7681"), val = tensor([2, 1280, 1, 256])]; tensor inputs_55_cast_fp16 = reshape(shape = var_7681, x = hidden_states_205_cast_fp16)[name = tensor("inputs_55_cast_fp16")]; tensor hidden_states_207_axes_0 = const()[name = tensor("hidden_states_207_axes_0"), val = tensor([1])]; tensor hidden_states_207_gamma_0_to_fp16 = const()[name = tensor("hidden_states_207_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(555344576)))]; tensor hidden_states_207_beta_0_to_fp16 = const()[name = tensor("hidden_states_207_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(555347200)))]; tensor var_7697_to_fp16 = const()[name = tensor("op_7697_to_fp16"), val = tensor(0x1.5p-17)]; tensor hidden_states_207_cast_fp16 = layer_norm(axes = hidden_states_207_axes_0, beta = hidden_states_207_beta_0_to_fp16, epsilon = var_7697_to_fp16, gamma = hidden_states_207_gamma_0_to_fp16, x = inputs_55_cast_fp16)[name = tensor("hidden_states_207_cast_fp16")]; tensor q_37_pad_type_0 = const()[name = tensor("q_37_pad_type_0"), val = tensor("valid")]; tensor q_37_strides_0 = const()[name = tensor("q_37_strides_0"), val = tensor([1, 1])]; tensor q_37_pad_0 = const()[name = tensor("q_37_pad_0"), val = tensor([0, 0, 0, 0])]; tensor q_37_dilations_0 = const()[name = tensor("q_37_dilations_0"), val = tensor([1, 1])]; tensor q_37_groups_0 = const()[name = tensor("q_37_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(555349824)))]; tensor q_37_cast_fp16 = conv(dilations = q_37_dilations_0, groups = q_37_groups_0, pad = q_37_pad_0, pad_type = q_37_pad_type_0, strides = q_37_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_207_cast_fp16)[name = tensor("q_37_cast_fp16")]; tensor k_73_pad_type_0 = const()[name = tensor("k_73_pad_type_0"), val = tensor("valid")]; tensor k_73_strides_0 = const()[name = tensor("k_73_strides_0"), val = tensor([1, 1])]; tensor k_73_pad_0 = const()[name = tensor("k_73_pad_0"), val = tensor([0, 0, 0, 0])]; tensor k_73_dilations_0 = const()[name = tensor("k_73_dilations_0"), val = tensor([1, 1])]; tensor k_73_groups_0 = const()[name = tensor("k_73_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(558626688)))]; tensor k_73_cast_fp16 = conv(dilations = k_73_dilations_0, groups = k_73_groups_0, pad = k_73_pad_0, pad_type = k_73_pad_type_0, strides = k_73_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_207_cast_fp16)[name = tensor("k_73_cast_fp16")]; tensor v_37_pad_type_0 = const()[name = tensor("v_37_pad_type_0"), val = tensor("valid")]; tensor v_37_strides_0 = const()[name = tensor("v_37_strides_0"), val = tensor([1, 1])]; tensor v_37_pad_0 = const()[name = tensor("v_37_pad_0"), val = tensor([0, 0, 0, 0])]; tensor v_37_dilations_0 = const()[name = tensor("v_37_dilations_0"), val = tensor([1, 1])]; tensor v_37_groups_0 = const()[name = tensor("v_37_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(561903552)))]; tensor v_37_cast_fp16 = conv(dilations = v_37_dilations_0, groups = v_37_groups_0, pad = v_37_pad_0, pad_type = v_37_pad_type_0, strides = v_37_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_207_cast_fp16)[name = tensor("v_37_cast_fp16")]; tensor var_7730_begin_0 = const()[name = tensor("op_7730_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7730_end_0 = const()[name = tensor("op_7730_end_0"), val = tensor([2, 160, 1, 256])]; tensor var_7730_end_mask_0 = const()[name = tensor("op_7730_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7730_cast_fp16 = slice_by_index(begin = var_7730_begin_0, end = var_7730_end_0, end_mask = var_7730_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7730_cast_fp16")]; tensor var_7734_begin_0 = const()[name = tensor("op_7734_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_7734_end_0 = const()[name = tensor("op_7734_end_0"), val = tensor([2, 320, 1, 256])]; tensor var_7734_end_mask_0 = const()[name = tensor("op_7734_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7734_cast_fp16 = slice_by_index(begin = var_7734_begin_0, end = var_7734_end_0, end_mask = var_7734_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7734_cast_fp16")]; tensor var_7738_begin_0 = const()[name = tensor("op_7738_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_7738_end_0 = const()[name = tensor("op_7738_end_0"), val = tensor([2, 480, 1, 256])]; tensor var_7738_end_mask_0 = const()[name = tensor("op_7738_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7738_cast_fp16 = slice_by_index(begin = var_7738_begin_0, end = var_7738_end_0, end_mask = var_7738_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7738_cast_fp16")]; tensor var_7742_begin_0 = const()[name = tensor("op_7742_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_7742_end_0 = const()[name = tensor("op_7742_end_0"), val = tensor([2, 640, 1, 256])]; tensor var_7742_end_mask_0 = const()[name = tensor("op_7742_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7742_cast_fp16 = slice_by_index(begin = var_7742_begin_0, end = var_7742_end_0, end_mask = var_7742_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7742_cast_fp16")]; tensor var_7746_begin_0 = const()[name = tensor("op_7746_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_7746_end_0 = const()[name = tensor("op_7746_end_0"), val = tensor([2, 800, 1, 256])]; tensor var_7746_end_mask_0 = const()[name = tensor("op_7746_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7746_cast_fp16 = slice_by_index(begin = var_7746_begin_0, end = var_7746_end_0, end_mask = var_7746_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7746_cast_fp16")]; tensor var_7750_begin_0 = const()[name = tensor("op_7750_begin_0"), val = tensor([0, 800, 0, 0])]; tensor var_7750_end_0 = const()[name = tensor("op_7750_end_0"), val = tensor([2, 960, 1, 256])]; tensor var_7750_end_mask_0 = const()[name = tensor("op_7750_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7750_cast_fp16 = slice_by_index(begin = var_7750_begin_0, end = var_7750_end_0, end_mask = var_7750_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7750_cast_fp16")]; tensor var_7754_begin_0 = const()[name = tensor("op_7754_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_7754_end_0 = const()[name = tensor("op_7754_end_0"), val = tensor([2, 1120, 1, 256])]; tensor var_7754_end_mask_0 = const()[name = tensor("op_7754_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7754_cast_fp16 = slice_by_index(begin = var_7754_begin_0, end = var_7754_end_0, end_mask = var_7754_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7754_cast_fp16")]; tensor var_7758_begin_0 = const()[name = tensor("op_7758_begin_0"), val = tensor([0, 1120, 0, 0])]; tensor var_7758_end_0 = const()[name = tensor("op_7758_end_0"), val = tensor([2, 1, 1, 256])]; tensor var_7758_end_mask_0 = const()[name = tensor("op_7758_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_7758_cast_fp16 = slice_by_index(begin = var_7758_begin_0, end = var_7758_end_0, end_mask = var_7758_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7758_cast_fp16")]; tensor k_75_perm_0 = const()[name = tensor("k_75_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_7765_begin_0 = const()[name = tensor("op_7765_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7765_end_0 = const()[name = tensor("op_7765_end_0"), val = tensor([2, 256, 1, 160])]; tensor var_7765_end_mask_0 = const()[name = tensor("op_7765_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_75_cast_fp16 = transpose(perm = k_75_perm_0, x = k_73_cast_fp16)[name = tensor("transpose_13")]; tensor var_7765_cast_fp16 = slice_by_index(begin = var_7765_begin_0, end = var_7765_end_0, end_mask = var_7765_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7765_cast_fp16")]; tensor var_7769_begin_0 = const()[name = tensor("op_7769_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_7769_end_0 = const()[name = tensor("op_7769_end_0"), val = tensor([2, 256, 1, 320])]; tensor var_7769_end_mask_0 = const()[name = tensor("op_7769_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7769_cast_fp16 = slice_by_index(begin = var_7769_begin_0, end = var_7769_end_0, end_mask = var_7769_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7769_cast_fp16")]; tensor var_7773_begin_0 = const()[name = tensor("op_7773_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_7773_end_0 = const()[name = tensor("op_7773_end_0"), val = tensor([2, 256, 1, 480])]; tensor var_7773_end_mask_0 = const()[name = tensor("op_7773_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7773_cast_fp16 = slice_by_index(begin = var_7773_begin_0, end = var_7773_end_0, end_mask = var_7773_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7773_cast_fp16")]; tensor var_7777_begin_0 = const()[name = tensor("op_7777_begin_0"), val = tensor([0, 0, 0, 480])]; tensor var_7777_end_0 = const()[name = tensor("op_7777_end_0"), val = tensor([2, 256, 1, 640])]; tensor var_7777_end_mask_0 = const()[name = tensor("op_7777_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7777_cast_fp16 = slice_by_index(begin = var_7777_begin_0, end = var_7777_end_0, end_mask = var_7777_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7777_cast_fp16")]; tensor var_7781_begin_0 = const()[name = tensor("op_7781_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_7781_end_0 = const()[name = tensor("op_7781_end_0"), val = tensor([2, 256, 1, 800])]; tensor var_7781_end_mask_0 = const()[name = tensor("op_7781_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7781_cast_fp16 = slice_by_index(begin = var_7781_begin_0, end = var_7781_end_0, end_mask = var_7781_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7781_cast_fp16")]; tensor var_7785_begin_0 = const()[name = tensor("op_7785_begin_0"), val = tensor([0, 0, 0, 800])]; tensor var_7785_end_0 = const()[name = tensor("op_7785_end_0"), val = tensor([2, 256, 1, 960])]; tensor var_7785_end_mask_0 = const()[name = tensor("op_7785_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7785_cast_fp16 = slice_by_index(begin = var_7785_begin_0, end = var_7785_end_0, end_mask = var_7785_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7785_cast_fp16")]; tensor var_7789_begin_0 = const()[name = tensor("op_7789_begin_0"), val = tensor([0, 0, 0, 960])]; tensor var_7789_end_0 = const()[name = tensor("op_7789_end_0"), val = tensor([2, 256, 1, 1120])]; tensor var_7789_end_mask_0 = const()[name = tensor("op_7789_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7789_cast_fp16 = slice_by_index(begin = var_7789_begin_0, end = var_7789_end_0, end_mask = var_7789_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7789_cast_fp16")]; tensor var_7793_begin_0 = const()[name = tensor("op_7793_begin_0"), val = tensor([0, 0, 0, 1120])]; tensor var_7793_end_0 = const()[name = tensor("op_7793_end_0"), val = tensor([2, 256, 1, 1])]; tensor var_7793_end_mask_0 = const()[name = tensor("op_7793_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_7793_cast_fp16 = slice_by_index(begin = var_7793_begin_0, end = var_7793_end_0, end_mask = var_7793_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7793_cast_fp16")]; tensor var_7795_begin_0 = const()[name = tensor("op_7795_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7795_end_0 = const()[name = tensor("op_7795_end_0"), val = tensor([2, 160, 1, 256])]; tensor var_7795_end_mask_0 = const()[name = tensor("op_7795_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7795_cast_fp16 = slice_by_index(begin = var_7795_begin_0, end = var_7795_end_0, end_mask = var_7795_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7795_cast_fp16")]; tensor var_7799_begin_0 = const()[name = tensor("op_7799_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_7799_end_0 = const()[name = tensor("op_7799_end_0"), val = tensor([2, 320, 1, 256])]; tensor var_7799_end_mask_0 = const()[name = tensor("op_7799_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7799_cast_fp16 = slice_by_index(begin = var_7799_begin_0, end = var_7799_end_0, end_mask = var_7799_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7799_cast_fp16")]; tensor var_7803_begin_0 = const()[name = tensor("op_7803_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_7803_end_0 = const()[name = tensor("op_7803_end_0"), val = tensor([2, 480, 1, 256])]; tensor var_7803_end_mask_0 = const()[name = tensor("op_7803_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7803_cast_fp16 = slice_by_index(begin = var_7803_begin_0, end = var_7803_end_0, end_mask = var_7803_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7803_cast_fp16")]; tensor var_7807_begin_0 = const()[name = tensor("op_7807_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_7807_end_0 = const()[name = tensor("op_7807_end_0"), val = tensor([2, 640, 1, 256])]; tensor var_7807_end_mask_0 = const()[name = tensor("op_7807_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7807_cast_fp16 = slice_by_index(begin = var_7807_begin_0, end = var_7807_end_0, end_mask = var_7807_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7807_cast_fp16")]; tensor var_7811_begin_0 = const()[name = tensor("op_7811_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_7811_end_0 = const()[name = tensor("op_7811_end_0"), val = tensor([2, 800, 1, 256])]; tensor var_7811_end_mask_0 = const()[name = tensor("op_7811_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7811_cast_fp16 = slice_by_index(begin = var_7811_begin_0, end = var_7811_end_0, end_mask = var_7811_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7811_cast_fp16")]; tensor var_7815_begin_0 = const()[name = tensor("op_7815_begin_0"), val = tensor([0, 800, 0, 0])]; tensor var_7815_end_0 = const()[name = tensor("op_7815_end_0"), val = tensor([2, 960, 1, 256])]; tensor var_7815_end_mask_0 = const()[name = tensor("op_7815_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7815_cast_fp16 = slice_by_index(begin = var_7815_begin_0, end = var_7815_end_0, end_mask = var_7815_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7815_cast_fp16")]; tensor var_7819_begin_0 = const()[name = tensor("op_7819_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_7819_end_0 = const()[name = tensor("op_7819_end_0"), val = tensor([2, 1120, 1, 256])]; tensor var_7819_end_mask_0 = const()[name = tensor("op_7819_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7819_cast_fp16 = slice_by_index(begin = var_7819_begin_0, end = var_7819_end_0, end_mask = var_7819_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7819_cast_fp16")]; tensor var_7823_begin_0 = const()[name = tensor("op_7823_begin_0"), val = tensor([0, 1120, 0, 0])]; tensor var_7823_end_0 = const()[name = tensor("op_7823_end_0"), val = tensor([2, 1, 1, 256])]; tensor var_7823_end_mask_0 = const()[name = tensor("op_7823_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_7823_cast_fp16 = slice_by_index(begin = var_7823_begin_0, end = var_7823_end_0, end_mask = var_7823_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7823_cast_fp16")]; tensor var_7827_equation_0 = const()[name = tensor("op_7827_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7827_cast_fp16 = einsum(equation = var_7827_equation_0, values = (var_7765_cast_fp16, var_7730_cast_fp16))[name = tensor("op_7827_cast_fp16")]; tensor var_7828_to_fp16 = const()[name = tensor("op_7828_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_161_cast_fp16 = mul(x = var_7827_cast_fp16, y = var_7828_to_fp16)[name = tensor("aw_161_cast_fp16")]; tensor var_7831_equation_0 = const()[name = tensor("op_7831_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7831_cast_fp16 = einsum(equation = var_7831_equation_0, values = (var_7769_cast_fp16, var_7734_cast_fp16))[name = tensor("op_7831_cast_fp16")]; tensor var_7832_to_fp16 = const()[name = tensor("op_7832_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_163_cast_fp16 = mul(x = var_7831_cast_fp16, y = var_7832_to_fp16)[name = tensor("aw_163_cast_fp16")]; tensor var_7835_equation_0 = const()[name = tensor("op_7835_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7835_cast_fp16 = einsum(equation = var_7835_equation_0, values = (var_7773_cast_fp16, var_7738_cast_fp16))[name = tensor("op_7835_cast_fp16")]; tensor var_7836_to_fp16 = const()[name = tensor("op_7836_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_165_cast_fp16 = mul(x = var_7835_cast_fp16, y = var_7836_to_fp16)[name = tensor("aw_165_cast_fp16")]; tensor var_7839_equation_0 = const()[name = tensor("op_7839_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7839_cast_fp16 = einsum(equation = var_7839_equation_0, values = (var_7777_cast_fp16, var_7742_cast_fp16))[name = tensor("op_7839_cast_fp16")]; tensor var_7840_to_fp16 = const()[name = tensor("op_7840_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_167_cast_fp16 = mul(x = var_7839_cast_fp16, y = var_7840_to_fp16)[name = tensor("aw_167_cast_fp16")]; tensor var_7843_equation_0 = const()[name = tensor("op_7843_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7843_cast_fp16 = einsum(equation = var_7843_equation_0, values = (var_7781_cast_fp16, var_7746_cast_fp16))[name = tensor("op_7843_cast_fp16")]; tensor var_7844_to_fp16 = const()[name = tensor("op_7844_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_169_cast_fp16 = mul(x = var_7843_cast_fp16, y = var_7844_to_fp16)[name = tensor("aw_169_cast_fp16")]; tensor var_7847_equation_0 = const()[name = tensor("op_7847_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7847_cast_fp16 = einsum(equation = var_7847_equation_0, values = (var_7785_cast_fp16, var_7750_cast_fp16))[name = tensor("op_7847_cast_fp16")]; tensor var_7848_to_fp16 = const()[name = tensor("op_7848_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_171_cast_fp16 = mul(x = var_7847_cast_fp16, y = var_7848_to_fp16)[name = tensor("aw_171_cast_fp16")]; tensor var_7851_equation_0 = const()[name = tensor("op_7851_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7851_cast_fp16 = einsum(equation = var_7851_equation_0, values = (var_7789_cast_fp16, var_7754_cast_fp16))[name = tensor("op_7851_cast_fp16")]; tensor var_7852_to_fp16 = const()[name = tensor("op_7852_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_173_cast_fp16 = mul(x = var_7851_cast_fp16, y = var_7852_to_fp16)[name = tensor("aw_173_cast_fp16")]; tensor var_7855_equation_0 = const()[name = tensor("op_7855_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7855_cast_fp16 = einsum(equation = var_7855_equation_0, values = (var_7793_cast_fp16, var_7758_cast_fp16))[name = tensor("op_7855_cast_fp16")]; tensor var_7856_to_fp16 = const()[name = tensor("op_7856_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_175_cast_fp16 = mul(x = var_7855_cast_fp16, y = var_7856_to_fp16)[name = tensor("aw_175_cast_fp16")]; tensor var_7858_cast_fp16 = softmax(axis = var_6522, x = aw_161_cast_fp16)[name = tensor("op_7858_cast_fp16")]; tensor var_7859_cast_fp16 = softmax(axis = var_6522, x = aw_163_cast_fp16)[name = tensor("op_7859_cast_fp16")]; tensor var_7860_cast_fp16 = softmax(axis = var_6522, x = aw_165_cast_fp16)[name = tensor("op_7860_cast_fp16")]; tensor var_7861_cast_fp16 = softmax(axis = var_6522, x = aw_167_cast_fp16)[name = tensor("op_7861_cast_fp16")]; tensor var_7862_cast_fp16 = softmax(axis = var_6522, x = aw_169_cast_fp16)[name = tensor("op_7862_cast_fp16")]; tensor var_7863_cast_fp16 = softmax(axis = var_6522, x = aw_171_cast_fp16)[name = tensor("op_7863_cast_fp16")]; tensor var_7864_cast_fp16 = softmax(axis = var_6522, x = aw_173_cast_fp16)[name = tensor("op_7864_cast_fp16")]; tensor var_7865_cast_fp16 = softmax(axis = var_6522, x = aw_175_cast_fp16)[name = tensor("op_7865_cast_fp16")]; tensor var_7867_equation_0 = const()[name = tensor("op_7867_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7867_cast_fp16 = einsum(equation = var_7867_equation_0, values = (var_7795_cast_fp16, var_7858_cast_fp16))[name = tensor("op_7867_cast_fp16")]; tensor var_7869_equation_0 = const()[name = tensor("op_7869_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7869_cast_fp16 = einsum(equation = var_7869_equation_0, values = (var_7799_cast_fp16, var_7859_cast_fp16))[name = tensor("op_7869_cast_fp16")]; tensor var_7871_equation_0 = const()[name = tensor("op_7871_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7871_cast_fp16 = einsum(equation = var_7871_equation_0, values = (var_7803_cast_fp16, var_7860_cast_fp16))[name = tensor("op_7871_cast_fp16")]; tensor var_7873_equation_0 = const()[name = tensor("op_7873_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7873_cast_fp16 = einsum(equation = var_7873_equation_0, values = (var_7807_cast_fp16, var_7861_cast_fp16))[name = tensor("op_7873_cast_fp16")]; tensor var_7875_equation_0 = const()[name = tensor("op_7875_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7875_cast_fp16 = einsum(equation = var_7875_equation_0, values = (var_7811_cast_fp16, var_7862_cast_fp16))[name = tensor("op_7875_cast_fp16")]; tensor var_7877_equation_0 = const()[name = tensor("op_7877_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7877_cast_fp16 = einsum(equation = var_7877_equation_0, values = (var_7815_cast_fp16, var_7863_cast_fp16))[name = tensor("op_7877_cast_fp16")]; tensor var_7879_equation_0 = const()[name = tensor("op_7879_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7879_cast_fp16 = einsum(equation = var_7879_equation_0, values = (var_7819_cast_fp16, var_7864_cast_fp16))[name = tensor("op_7879_cast_fp16")]; tensor var_7881_equation_0 = const()[name = tensor("op_7881_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7881_cast_fp16 = einsum(equation = var_7881_equation_0, values = (var_7823_cast_fp16, var_7865_cast_fp16))[name = tensor("op_7881_cast_fp16")]; tensor input_353_interleave_0 = const()[name = tensor("input_353_interleave_0"), val = tensor(false)]; tensor input_353_cast_fp16 = concat(axis = var_6522, interleave = input_353_interleave_0, values = (var_7867_cast_fp16, var_7869_cast_fp16, var_7871_cast_fp16, var_7873_cast_fp16, var_7875_cast_fp16, var_7877_cast_fp16, var_7879_cast_fp16, var_7881_cast_fp16))[name = tensor("input_353_cast_fp16")]; tensor var_7891_pad_type_0 = const()[name = tensor("op_7891_pad_type_0"), val = tensor("valid")]; tensor var_7891_strides_0 = const()[name = tensor("op_7891_strides_0"), val = tensor([1, 1])]; tensor var_7891_pad_0 = const()[name = tensor("op_7891_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_7891_dilations_0 = const()[name = tensor("op_7891_dilations_0"), val = tensor([1, 1])]; tensor var_7891_groups_0 = const()[name = tensor("op_7891_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565180416)))]; tensor up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568457280)))]; tensor var_7891_cast_fp16 = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_7891_dilations_0, groups = var_7891_groups_0, pad = var_7891_pad_0, pad_type = var_7891_pad_type_0, strides = var_7891_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_353_cast_fp16)[name = tensor("op_7891_cast_fp16")]; tensor inputs_57_cast_fp16 = add(x = var_7891_cast_fp16, y = inputs_55_cast_fp16)[name = tensor("inputs_57_cast_fp16")]; tensor hidden_states_209_axes_0 = const()[name = tensor("hidden_states_209_axes_0"), val = tensor([1])]; tensor hidden_states_209_gamma_0_to_fp16 = const()[name = tensor("hidden_states_209_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568459904)))]; tensor hidden_states_209_beta_0_to_fp16 = const()[name = tensor("hidden_states_209_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568462528)))]; tensor var_7901_to_fp16 = const()[name = tensor("op_7901_to_fp16"), val = tensor(0x1.5p-17)]; tensor hidden_states_209_cast_fp16 = layer_norm(axes = hidden_states_209_axes_0, beta = hidden_states_209_beta_0_to_fp16, epsilon = var_7901_to_fp16, gamma = hidden_states_209_gamma_0_to_fp16, x = inputs_57_cast_fp16)[name = tensor("hidden_states_209_cast_fp16")]; tensor q_39_pad_type_0 = const()[name = tensor("q_39_pad_type_0"), val = tensor("valid")]; tensor q_39_strides_0 = const()[name = tensor("q_39_strides_0"), val = tensor([1, 1])]; tensor q_39_pad_0 = const()[name = tensor("q_39_pad_0"), val = tensor([0, 0, 0, 0])]; tensor q_39_dilations_0 = const()[name = tensor("q_39_dilations_0"), val = tensor([1, 1])]; tensor q_39_groups_0 = const()[name = tensor("q_39_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568465152)))]; tensor q_39_cast_fp16 = conv(dilations = q_39_dilations_0, groups = q_39_groups_0, pad = q_39_pad_0, pad_type = q_39_pad_type_0, strides = q_39_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_209_cast_fp16)[name = tensor("q_39_cast_fp16")]; tensor k_77_pad_type_0 = const()[name = tensor("k_77_pad_type_0"), val = tensor("valid")]; tensor k_77_strides_0 = const()[name = tensor("k_77_strides_0"), val = tensor([1, 1])]; tensor k_77_pad_0 = const()[name = tensor("k_77_pad_0"), val = tensor([0, 0, 0, 0])]; tensor k_77_dilations_0 = const()[name = tensor("k_77_dilations_0"), val = tensor([1, 1])]; tensor k_77_groups_0 = const()[name = tensor("k_77_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(571742016)))]; tensor k_77_cast_fp16 = conv(dilations = k_77_dilations_0, groups = k_77_groups_0, pad = k_77_pad_0, pad_type = k_77_pad_type_0, strides = k_77_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_77_cast_fp16")]; tensor v_39_pad_type_0 = const()[name = tensor("v_39_pad_type_0"), val = tensor("valid")]; tensor v_39_strides_0 = const()[name = tensor("v_39_strides_0"), val = tensor([1, 1])]; tensor v_39_pad_0 = const()[name = tensor("v_39_pad_0"), val = tensor([0, 0, 0, 0])]; tensor v_39_dilations_0 = const()[name = tensor("v_39_dilations_0"), val = tensor([1, 1])]; tensor v_39_groups_0 = const()[name = tensor("v_39_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(573708160)))]; tensor v_39_cast_fp16 = conv(dilations = v_39_dilations_0, groups = v_39_groups_0, pad = v_39_pad_0, pad_type = v_39_pad_type_0, strides = v_39_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_39_cast_fp16")]; tensor var_7934_begin_0 = const()[name = tensor("op_7934_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7934_end_0 = const()[name = tensor("op_7934_end_0"), val = tensor([2, 160, 1, 256])]; tensor var_7934_end_mask_0 = const()[name = tensor("op_7934_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7934_cast_fp16 = slice_by_index(begin = var_7934_begin_0, end = var_7934_end_0, end_mask = var_7934_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7934_cast_fp16")]; tensor var_7938_begin_0 = const()[name = tensor("op_7938_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_7938_end_0 = const()[name = tensor("op_7938_end_0"), val = tensor([2, 320, 1, 256])]; tensor var_7938_end_mask_0 = const()[name = tensor("op_7938_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7938_cast_fp16 = slice_by_index(begin = var_7938_begin_0, end = var_7938_end_0, end_mask = var_7938_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7938_cast_fp16")]; tensor var_7942_begin_0 = const()[name = tensor("op_7942_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_7942_end_0 = const()[name = tensor("op_7942_end_0"), val = tensor([2, 480, 1, 256])]; tensor var_7942_end_mask_0 = const()[name = tensor("op_7942_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7942_cast_fp16 = slice_by_index(begin = var_7942_begin_0, end = var_7942_end_0, end_mask = var_7942_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7942_cast_fp16")]; tensor var_7946_begin_0 = const()[name = tensor("op_7946_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_7946_end_0 = const()[name = tensor("op_7946_end_0"), val = tensor([2, 640, 1, 256])]; tensor var_7946_end_mask_0 = const()[name = tensor("op_7946_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7946_cast_fp16 = slice_by_index(begin = var_7946_begin_0, end = var_7946_end_0, end_mask = var_7946_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7946_cast_fp16")]; tensor var_7950_begin_0 = const()[name = tensor("op_7950_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_7950_end_0 = const()[name = tensor("op_7950_end_0"), val = tensor([2, 800, 1, 256])]; tensor var_7950_end_mask_0 = const()[name = tensor("op_7950_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7950_cast_fp16 = slice_by_index(begin = var_7950_begin_0, end = var_7950_end_0, end_mask = var_7950_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7950_cast_fp16")]; tensor var_7954_begin_0 = const()[name = tensor("op_7954_begin_0"), val = tensor([0, 800, 0, 0])]; tensor var_7954_end_0 = const()[name = tensor("op_7954_end_0"), val = tensor([2, 960, 1, 256])]; tensor var_7954_end_mask_0 = const()[name = tensor("op_7954_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7954_cast_fp16 = slice_by_index(begin = var_7954_begin_0, end = var_7954_end_0, end_mask = var_7954_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7954_cast_fp16")]; tensor var_7958_begin_0 = const()[name = tensor("op_7958_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_7958_end_0 = const()[name = tensor("op_7958_end_0"), val = tensor([2, 1120, 1, 256])]; tensor var_7958_end_mask_0 = const()[name = tensor("op_7958_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7958_cast_fp16 = slice_by_index(begin = var_7958_begin_0, end = var_7958_end_0, end_mask = var_7958_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7958_cast_fp16")]; tensor var_7962_begin_0 = const()[name = tensor("op_7962_begin_0"), val = tensor([0, 1120, 0, 0])]; tensor var_7962_end_0 = const()[name = tensor("op_7962_end_0"), val = tensor([2, 1, 1, 256])]; tensor var_7962_end_mask_0 = const()[name = tensor("op_7962_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_7962_cast_fp16 = slice_by_index(begin = var_7962_begin_0, end = var_7962_end_0, end_mask = var_7962_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7962_cast_fp16")]; tensor k_79_perm_0 = const()[name = tensor("k_79_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_7969_begin_0 = const()[name = tensor("op_7969_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7969_end_0 = const()[name = tensor("op_7969_end_0"), val = tensor([2, 77, 1, 160])]; tensor var_7969_end_mask_0 = const()[name = tensor("op_7969_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_79_cast_fp16 = transpose(perm = k_79_perm_0, x = k_77_cast_fp16)[name = tensor("transpose_12")]; tensor var_7969_cast_fp16 = slice_by_index(begin = var_7969_begin_0, end = var_7969_end_0, end_mask = var_7969_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7969_cast_fp16")]; tensor var_7973_begin_0 = const()[name = tensor("op_7973_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_7973_end_0 = const()[name = tensor("op_7973_end_0"), val = tensor([2, 77, 1, 320])]; tensor var_7973_end_mask_0 = const()[name = tensor("op_7973_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7973_cast_fp16 = slice_by_index(begin = var_7973_begin_0, end = var_7973_end_0, end_mask = var_7973_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7973_cast_fp16")]; tensor var_7977_begin_0 = const()[name = tensor("op_7977_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_7977_end_0 = const()[name = tensor("op_7977_end_0"), val = tensor([2, 77, 1, 480])]; tensor var_7977_end_mask_0 = const()[name = tensor("op_7977_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7977_cast_fp16 = slice_by_index(begin = var_7977_begin_0, end = var_7977_end_0, end_mask = var_7977_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7977_cast_fp16")]; tensor var_7981_begin_0 = const()[name = tensor("op_7981_begin_0"), val = tensor([0, 0, 0, 480])]; tensor var_7981_end_0 = const()[name = tensor("op_7981_end_0"), val = tensor([2, 77, 1, 640])]; tensor var_7981_end_mask_0 = const()[name = tensor("op_7981_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7981_cast_fp16 = slice_by_index(begin = var_7981_begin_0, end = var_7981_end_0, end_mask = var_7981_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7981_cast_fp16")]; tensor var_7985_begin_0 = const()[name = tensor("op_7985_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_7985_end_0 = const()[name = tensor("op_7985_end_0"), val = tensor([2, 77, 1, 800])]; tensor var_7985_end_mask_0 = const()[name = tensor("op_7985_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7985_cast_fp16 = slice_by_index(begin = var_7985_begin_0, end = var_7985_end_0, end_mask = var_7985_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7985_cast_fp16")]; tensor var_7989_begin_0 = const()[name = tensor("op_7989_begin_0"), val = tensor([0, 0, 0, 800])]; tensor var_7989_end_0 = const()[name = tensor("op_7989_end_0"), val = tensor([2, 77, 1, 960])]; tensor var_7989_end_mask_0 = const()[name = tensor("op_7989_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7989_cast_fp16 = slice_by_index(begin = var_7989_begin_0, end = var_7989_end_0, end_mask = var_7989_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7989_cast_fp16")]; tensor var_7993_begin_0 = const()[name = tensor("op_7993_begin_0"), val = tensor([0, 0, 0, 960])]; tensor var_7993_end_0 = const()[name = tensor("op_7993_end_0"), val = tensor([2, 77, 1, 1120])]; tensor var_7993_end_mask_0 = const()[name = tensor("op_7993_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7993_cast_fp16 = slice_by_index(begin = var_7993_begin_0, end = var_7993_end_0, end_mask = var_7993_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7993_cast_fp16")]; tensor var_7997_begin_0 = const()[name = tensor("op_7997_begin_0"), val = tensor([0, 0, 0, 1120])]; tensor var_7997_end_0 = const()[name = tensor("op_7997_end_0"), val = tensor([2, 77, 1, 1])]; tensor var_7997_end_mask_0 = const()[name = tensor("op_7997_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_7997_cast_fp16 = slice_by_index(begin = var_7997_begin_0, end = var_7997_end_0, end_mask = var_7997_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7997_cast_fp16")]; tensor var_7999_begin_0 = const()[name = tensor("op_7999_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7999_end_0 = const()[name = tensor("op_7999_end_0"), val = tensor([2, 160, 1, 77])]; tensor var_7999_end_mask_0 = const()[name = tensor("op_7999_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7999_cast_fp16 = slice_by_index(begin = var_7999_begin_0, end = var_7999_end_0, end_mask = var_7999_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7999_cast_fp16")]; tensor var_8003_begin_0 = const()[name = tensor("op_8003_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_8003_end_0 = const()[name = tensor("op_8003_end_0"), val = tensor([2, 320, 1, 77])]; tensor var_8003_end_mask_0 = const()[name = tensor("op_8003_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8003_cast_fp16 = slice_by_index(begin = var_8003_begin_0, end = var_8003_end_0, end_mask = var_8003_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8003_cast_fp16")]; tensor var_8007_begin_0 = const()[name = tensor("op_8007_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_8007_end_0 = const()[name = tensor("op_8007_end_0"), val = tensor([2, 480, 1, 77])]; tensor var_8007_end_mask_0 = const()[name = tensor("op_8007_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8007_cast_fp16 = slice_by_index(begin = var_8007_begin_0, end = var_8007_end_0, end_mask = var_8007_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8007_cast_fp16")]; tensor var_8011_begin_0 = const()[name = tensor("op_8011_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_8011_end_0 = const()[name = tensor("op_8011_end_0"), val = tensor([2, 640, 1, 77])]; tensor var_8011_end_mask_0 = const()[name = tensor("op_8011_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8011_cast_fp16 = slice_by_index(begin = var_8011_begin_0, end = var_8011_end_0, end_mask = var_8011_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8011_cast_fp16")]; tensor var_8015_begin_0 = const()[name = tensor("op_8015_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_8015_end_0 = const()[name = tensor("op_8015_end_0"), val = tensor([2, 800, 1, 77])]; tensor var_8015_end_mask_0 = const()[name = tensor("op_8015_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8015_cast_fp16 = slice_by_index(begin = var_8015_begin_0, end = var_8015_end_0, end_mask = var_8015_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8015_cast_fp16")]; tensor var_8019_begin_0 = const()[name = tensor("op_8019_begin_0"), val = tensor([0, 800, 0, 0])]; tensor var_8019_end_0 = const()[name = tensor("op_8019_end_0"), val = tensor([2, 960, 1, 77])]; tensor var_8019_end_mask_0 = const()[name = tensor("op_8019_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8019_cast_fp16 = slice_by_index(begin = var_8019_begin_0, end = var_8019_end_0, end_mask = var_8019_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8019_cast_fp16")]; tensor var_8023_begin_0 = const()[name = tensor("op_8023_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_8023_end_0 = const()[name = tensor("op_8023_end_0"), val = tensor([2, 1120, 1, 77])]; tensor var_8023_end_mask_0 = const()[name = tensor("op_8023_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8023_cast_fp16 = slice_by_index(begin = var_8023_begin_0, end = var_8023_end_0, end_mask = var_8023_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8023_cast_fp16")]; tensor var_8027_begin_0 = const()[name = tensor("op_8027_begin_0"), val = tensor([0, 1120, 0, 0])]; tensor var_8027_end_0 = const()[name = tensor("op_8027_end_0"), val = tensor([2, 1, 1, 77])]; tensor var_8027_end_mask_0 = const()[name = tensor("op_8027_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_8027_cast_fp16 = slice_by_index(begin = var_8027_begin_0, end = var_8027_end_0, end_mask = var_8027_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8027_cast_fp16")]; tensor var_8031_equation_0 = const()[name = tensor("op_8031_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8031_cast_fp16 = einsum(equation = var_8031_equation_0, values = (var_7969_cast_fp16, var_7934_cast_fp16))[name = tensor("op_8031_cast_fp16")]; tensor var_8032_to_fp16 = const()[name = tensor("op_8032_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_177_cast_fp16 = mul(x = var_8031_cast_fp16, y = var_8032_to_fp16)[name = tensor("aw_177_cast_fp16")]; tensor var_8035_equation_0 = const()[name = tensor("op_8035_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8035_cast_fp16 = einsum(equation = var_8035_equation_0, values = (var_7973_cast_fp16, var_7938_cast_fp16))[name = tensor("op_8035_cast_fp16")]; tensor var_8036_to_fp16 = const()[name = tensor("op_8036_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_179_cast_fp16 = mul(x = var_8035_cast_fp16, y = var_8036_to_fp16)[name = tensor("aw_179_cast_fp16")]; tensor var_8039_equation_0 = const()[name = tensor("op_8039_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8039_cast_fp16 = einsum(equation = var_8039_equation_0, values = (var_7977_cast_fp16, var_7942_cast_fp16))[name = tensor("op_8039_cast_fp16")]; tensor var_8040_to_fp16 = const()[name = tensor("op_8040_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_181_cast_fp16 = mul(x = var_8039_cast_fp16, y = var_8040_to_fp16)[name = tensor("aw_181_cast_fp16")]; tensor var_8043_equation_0 = const()[name = tensor("op_8043_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8043_cast_fp16 = einsum(equation = var_8043_equation_0, values = (var_7981_cast_fp16, var_7946_cast_fp16))[name = tensor("op_8043_cast_fp16")]; tensor var_8044_to_fp16 = const()[name = tensor("op_8044_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_183_cast_fp16 = mul(x = var_8043_cast_fp16, y = var_8044_to_fp16)[name = tensor("aw_183_cast_fp16")]; tensor var_8047_equation_0 = const()[name = tensor("op_8047_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8047_cast_fp16 = einsum(equation = var_8047_equation_0, values = (var_7985_cast_fp16, var_7950_cast_fp16))[name = tensor("op_8047_cast_fp16")]; tensor var_8048_to_fp16 = const()[name = tensor("op_8048_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_185_cast_fp16 = mul(x = var_8047_cast_fp16, y = var_8048_to_fp16)[name = tensor("aw_185_cast_fp16")]; tensor var_8051_equation_0 = const()[name = tensor("op_8051_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8051_cast_fp16 = einsum(equation = var_8051_equation_0, values = (var_7989_cast_fp16, var_7954_cast_fp16))[name = tensor("op_8051_cast_fp16")]; tensor var_8052_to_fp16 = const()[name = tensor("op_8052_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_187_cast_fp16 = mul(x = var_8051_cast_fp16, y = var_8052_to_fp16)[name = tensor("aw_187_cast_fp16")]; tensor var_8055_equation_0 = const()[name = tensor("op_8055_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8055_cast_fp16 = einsum(equation = var_8055_equation_0, values = (var_7993_cast_fp16, var_7958_cast_fp16))[name = tensor("op_8055_cast_fp16")]; tensor var_8056_to_fp16 = const()[name = tensor("op_8056_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_189_cast_fp16 = mul(x = var_8055_cast_fp16, y = var_8056_to_fp16)[name = tensor("aw_189_cast_fp16")]; tensor var_8059_equation_0 = const()[name = tensor("op_8059_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8059_cast_fp16 = einsum(equation = var_8059_equation_0, values = (var_7997_cast_fp16, var_7962_cast_fp16))[name = tensor("op_8059_cast_fp16")]; tensor var_8060_to_fp16 = const()[name = tensor("op_8060_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_cast_fp16 = mul(x = var_8059_cast_fp16, y = var_8060_to_fp16)[name = tensor("aw_cast_fp16")]; tensor var_8062_cast_fp16 = softmax(axis = var_6522, x = aw_177_cast_fp16)[name = tensor("op_8062_cast_fp16")]; tensor var_8063_cast_fp16 = softmax(axis = var_6522, x = aw_179_cast_fp16)[name = tensor("op_8063_cast_fp16")]; tensor var_8064_cast_fp16 = softmax(axis = var_6522, x = aw_181_cast_fp16)[name = tensor("op_8064_cast_fp16")]; tensor var_8065_cast_fp16 = softmax(axis = var_6522, x = aw_183_cast_fp16)[name = tensor("op_8065_cast_fp16")]; tensor var_8066_cast_fp16 = softmax(axis = var_6522, x = aw_185_cast_fp16)[name = tensor("op_8066_cast_fp16")]; tensor var_8067_cast_fp16 = softmax(axis = var_6522, x = aw_187_cast_fp16)[name = tensor("op_8067_cast_fp16")]; tensor var_8068_cast_fp16 = softmax(axis = var_6522, x = aw_189_cast_fp16)[name = tensor("op_8068_cast_fp16")]; tensor var_8069_cast_fp16 = softmax(axis = var_6522, x = aw_cast_fp16)[name = tensor("op_8069_cast_fp16")]; tensor var_8071_equation_0 = const()[name = tensor("op_8071_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8071_cast_fp16 = einsum(equation = var_8071_equation_0, values = (var_7999_cast_fp16, var_8062_cast_fp16))[name = tensor("op_8071_cast_fp16")]; tensor var_8073_equation_0 = const()[name = tensor("op_8073_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8073_cast_fp16 = einsum(equation = var_8073_equation_0, values = (var_8003_cast_fp16, var_8063_cast_fp16))[name = tensor("op_8073_cast_fp16")]; tensor var_8075_equation_0 = const()[name = tensor("op_8075_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8075_cast_fp16 = einsum(equation = var_8075_equation_0, values = (var_8007_cast_fp16, var_8064_cast_fp16))[name = tensor("op_8075_cast_fp16")]; tensor var_8077_equation_0 = const()[name = tensor("op_8077_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8077_cast_fp16 = einsum(equation = var_8077_equation_0, values = (var_8011_cast_fp16, var_8065_cast_fp16))[name = tensor("op_8077_cast_fp16")]; tensor var_8079_equation_0 = const()[name = tensor("op_8079_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8079_cast_fp16 = einsum(equation = var_8079_equation_0, values = (var_8015_cast_fp16, var_8066_cast_fp16))[name = tensor("op_8079_cast_fp16")]; tensor var_8081_equation_0 = const()[name = tensor("op_8081_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8081_cast_fp16 = einsum(equation = var_8081_equation_0, values = (var_8019_cast_fp16, var_8067_cast_fp16))[name = tensor("op_8081_cast_fp16")]; tensor var_8083_equation_0 = const()[name = tensor("op_8083_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8083_cast_fp16 = einsum(equation = var_8083_equation_0, values = (var_8023_cast_fp16, var_8068_cast_fp16))[name = tensor("op_8083_cast_fp16")]; tensor var_8085_equation_0 = const()[name = tensor("op_8085_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8085_cast_fp16 = einsum(equation = var_8085_equation_0, values = (var_8027_cast_fp16, var_8069_cast_fp16))[name = tensor("op_8085_cast_fp16")]; tensor input_355_interleave_0 = const()[name = tensor("input_355_interleave_0"), val = tensor(false)]; tensor input_355_cast_fp16 = concat(axis = var_6522, interleave = input_355_interleave_0, values = (var_8071_cast_fp16, var_8073_cast_fp16, var_8075_cast_fp16, var_8077_cast_fp16, var_8079_cast_fp16, var_8081_cast_fp16, var_8083_cast_fp16, var_8085_cast_fp16))[name = tensor("input_355_cast_fp16")]; tensor var_8095_pad_type_0 = const()[name = tensor("op_8095_pad_type_0"), val = tensor("valid")]; tensor var_8095_strides_0 = const()[name = tensor("op_8095_strides_0"), val = tensor([1, 1])]; tensor var_8095_pad_0 = const()[name = tensor("op_8095_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_8095_dilations_0 = const()[name = tensor("op_8095_dilations_0"), val = tensor([1, 1])]; tensor var_8095_groups_0 = const()[name = tensor("op_8095_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(575674304)))]; tensor up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578951168)))]; tensor var_8095_cast_fp16 = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_8095_dilations_0, groups = var_8095_groups_0, pad = var_8095_pad_0, pad_type = var_8095_pad_type_0, strides = var_8095_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_355_cast_fp16)[name = tensor("op_8095_cast_fp16")]; tensor inputs_59_cast_fp16 = add(x = var_8095_cast_fp16, y = inputs_57_cast_fp16)[name = tensor("inputs_59_cast_fp16")]; tensor input_357_axes_0 = const()[name = tensor("input_357_axes_0"), val = tensor([1])]; tensor input_357_gamma_0_to_fp16 = const()[name = tensor("input_357_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578953792)))]; tensor input_357_beta_0_to_fp16 = const()[name = tensor("input_357_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578956416)))]; tensor var_8105_to_fp16 = const()[name = tensor("op_8105_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_357_cast_fp16 = layer_norm(axes = input_357_axes_0, beta = input_357_beta_0_to_fp16, epsilon = var_8105_to_fp16, gamma = input_357_gamma_0_to_fp16, x = inputs_59_cast_fp16)[name = tensor("input_357_cast_fp16")]; tensor var_8125_pad_type_0 = const()[name = tensor("op_8125_pad_type_0"), val = tensor("valid")]; tensor var_8125_strides_0 = const()[name = tensor("op_8125_strides_0"), val = tensor([1, 1])]; tensor var_8125_pad_0 = const()[name = tensor("op_8125_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_8125_dilations_0 = const()[name = tensor("op_8125_dilations_0"), val = tensor([1, 1])]; tensor var_8125_groups_0 = const()[name = tensor("op_8125_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578959040)))]; tensor up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(605173504)))]; tensor var_8125_cast_fp16 = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_8125_dilations_0, groups = var_8125_groups_0, pad = var_8125_pad_0, pad_type = var_8125_pad_type_0, strides = var_8125_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_357_cast_fp16)[name = tensor("op_8125_cast_fp16")]; tensor var_8126_split_sizes_0 = const()[name = tensor("op_8126_split_sizes_0"), val = tensor([5120, 5120])]; tensor var_8126_axis_0 = const()[name = tensor("op_8126_axis_0"), val = tensor(1)]; tensor var_8126_cast_fp16_0, tensor var_8126_cast_fp16_1 = split(axis = var_8126_axis_0, split_sizes = var_8126_split_sizes_0, x = var_8125_cast_fp16)[name = tensor("op_8126_cast_fp16")]; tensor var_8128_mode_0 = const()[name = tensor("op_8128_mode_0"), val = tensor("EXACT")]; tensor var_8128_cast_fp16 = gelu(mode = var_8128_mode_0, x = var_8126_cast_fp16_1)[name = tensor("op_8128_cast_fp16")]; tensor input_359_cast_fp16 = mul(x = var_8126_cast_fp16_0, y = var_8128_cast_fp16)[name = tensor("input_359_cast_fp16")]; tensor var_8136_pad_type_0 = const()[name = tensor("op_8136_pad_type_0"), val = tensor("valid")]; tensor var_8136_strides_0 = const()[name = tensor("op_8136_strides_0"), val = tensor([1, 1])]; tensor var_8136_pad_0 = const()[name = tensor("op_8136_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_8136_dilations_0 = const()[name = tensor("op_8136_dilations_0"), val = tensor([1, 1])]; tensor var_8136_groups_0 = const()[name = tensor("op_8136_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(605194048)))]; tensor up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(618301312)))]; tensor var_8136_cast_fp16 = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_8136_dilations_0, groups = var_8136_groups_0, pad = var_8136_pad_0, pad_type = var_8136_pad_type_0, strides = var_8136_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_359_cast_fp16)[name = tensor("op_8136_cast_fp16")]; tensor hidden_states_213_cast_fp16 = add(x = var_8136_cast_fp16, y = inputs_59_cast_fp16)[name = tensor("hidden_states_213_cast_fp16")]; tensor var_8138 = const()[name = tensor("op_8138"), val = tensor([2, 1280, 16, 16])]; tensor input_361_cast_fp16 = reshape(shape = var_8138, x = hidden_states_213_cast_fp16)[name = tensor("input_361_cast_fp16")]; tensor hidden_states_215_pad_type_0 = const()[name = tensor("hidden_states_215_pad_type_0"), val = tensor("valid")]; tensor hidden_states_215_strides_0 = const()[name = tensor("hidden_states_215_strides_0"), val = tensor([1, 1])]; tensor hidden_states_215_pad_0 = const()[name = tensor("hidden_states_215_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_215_dilations_0 = const()[name = tensor("hidden_states_215_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_215_groups_0 = const()[name = tensor("hidden_states_215_groups_0"), val = tensor(1)]; tensor up_blocks_1_attentions_2_proj_out_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(618303936)))]; tensor up_blocks_1_attentions_2_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(621580800)))]; tensor hidden_states_215_cast_fp16 = conv(bias = up_blocks_1_attentions_2_proj_out_bias_to_fp16, dilations = hidden_states_215_dilations_0, groups = hidden_states_215_groups_0, pad = hidden_states_215_pad_0, pad_type = hidden_states_215_pad_type_0, strides = hidden_states_215_strides_0, weight = up_blocks_1_attentions_2_proj_out_weight_to_fp16, x = input_361_cast_fp16)[name = tensor("hidden_states_215_cast_fp16")]; tensor input_363_cast_fp16 = add(x = hidden_states_215_cast_fp16, y = hidden_states_203_cast_fp16)[name = tensor("input_363_cast_fp16")]; tensor input_365_scale_factor_height_0 = const()[name = tensor("input_365_scale_factor_height_0"), val = tensor(0x1p+1)]; tensor input_365_scale_factor_width_0 = const()[name = tensor("input_365_scale_factor_width_0"), val = tensor(0x1p+1)]; tensor input_365_cast_fp16 = upsample_nearest_neighbor(scale_factor_height = input_365_scale_factor_height_0, scale_factor_width = input_365_scale_factor_width_0, x = input_363_cast_fp16)[name = tensor("input_365_cast_fp16")]; tensor hidden_states_217_pad_type_0 = const()[name = tensor("hidden_states_217_pad_type_0"), val = tensor("custom")]; tensor hidden_states_217_pad_0 = const()[name = tensor("hidden_states_217_pad_0"), val = tensor([1, 1, 1, 1])]; tensor hidden_states_217_strides_0 = const()[name = tensor("hidden_states_217_strides_0"), val = tensor([1, 1])]; tensor hidden_states_217_dilations_0 = const()[name = tensor("hidden_states_217_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_217_groups_0 = const()[name = tensor("hidden_states_217_groups_0"), val = tensor(1)]; tensor up_blocks_1_upsamplers_0_conv_weight_to_fp16 = const()[name = tensor("up_blocks_1_upsamplers_0_conv_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(621583424)))]; tensor up_blocks_1_upsamplers_0_conv_bias_to_fp16 = const()[name = tensor("up_blocks_1_upsamplers_0_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(651074688)))]; tensor hidden_states_217_cast_fp16 = conv(bias = up_blocks_1_upsamplers_0_conv_bias_to_fp16, dilations = hidden_states_217_dilations_0, groups = hidden_states_217_groups_0, pad = hidden_states_217_pad_0, pad_type = hidden_states_217_pad_type_0, strides = hidden_states_217_strides_0, weight = up_blocks_1_upsamplers_0_conv_weight_to_fp16, x = input_365_cast_fp16)[name = tensor("hidden_states_217_cast_fp16")]; tensor var_8160 = const()[name = tensor("op_8160"), val = tensor(3)]; tensor var_8182 = const()[name = tensor("op_8182"), val = tensor(1)]; tensor input_367_interleave_0 = const()[name = tensor("input_367_interleave_0"), val = tensor(false)]; tensor cast_3 = cast(dtype = cast_3_dtype_0, x = input_115_cast_fp16)[name = tensor("cast_7")]; tensor input_367_cast_fp16 = concat(axis = var_8182, interleave = input_367_interleave_0, values = (hidden_states_217_cast_fp16, cast_3))[name = tensor("input_367_cast_fp16")]; tensor reshape_168_shape_0 = const()[name = tensor("reshape_168_shape_0"), val = tensor([2, 32, 60, 32, 32])]; tensor reshape_168_cast_fp16 = reshape(shape = reshape_168_shape_0, x = input_367_cast_fp16)[name = tensor("reshape_168_cast_fp16")]; tensor reduce_mean_126_axes_0 = const()[name = tensor("reduce_mean_126_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_126_keep_dims_0 = const()[name = tensor("reduce_mean_126_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_126_cast_fp16 = reduce_mean(axes = reduce_mean_126_axes_0, keep_dims = reduce_mean_126_keep_dims_0, x = reshape_168_cast_fp16)[name = tensor("reduce_mean_126_cast_fp16")]; tensor sub_84_cast_fp16 = sub(x = reshape_168_cast_fp16, y = reduce_mean_126_cast_fp16)[name = tensor("sub_84_cast_fp16")]; tensor square_42_cast_fp16 = square(x = sub_84_cast_fp16)[name = tensor("square_42_cast_fp16")]; tensor reduce_mean_128_axes_0 = const()[name = tensor("reduce_mean_128_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_128_keep_dims_0 = const()[name = tensor("reduce_mean_128_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_128_cast_fp16 = reduce_mean(axes = reduce_mean_128_axes_0, keep_dims = reduce_mean_128_keep_dims_0, x = square_42_cast_fp16)[name = tensor("reduce_mean_128_cast_fp16")]; tensor add_84_y_0_to_fp16 = const()[name = tensor("add_84_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_84_cast_fp16 = add(x = reduce_mean_128_cast_fp16, y = add_84_y_0_to_fp16)[name = tensor("add_84_cast_fp16")]; tensor sqrt_42_cast_fp16 = sqrt(x = add_84_cast_fp16)[name = tensor("sqrt_42_cast_fp16")]; tensor real_div_42_cast_fp16 = real_div(x = sub_84_cast_fp16, y = sqrt_42_cast_fp16)[name = tensor("real_div_42_cast_fp16")]; tensor reshape_169_shape_0 = const()[name = tensor("reshape_169_shape_0"), val = tensor([2, 1920, 32, 32])]; tensor reshape_169_cast_fp16 = reshape(shape = reshape_169_shape_0, x = real_div_42_cast_fp16)[name = tensor("reshape_169_cast_fp16")]; tensor add_85_gamma_0_to_fp16 = const()[name = tensor("add_85_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(651077312)))]; tensor add_85_beta_0_to_fp16 = const()[name = tensor("add_85_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(651081216)))]; tensor add_85_epsilon_0_to_fp16 = const()[name = tensor("add_85_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_85_cast_fp16 = batch_norm(beta = add_85_beta_0_to_fp16, epsilon = add_85_epsilon_0_to_fp16, gamma = add_85_gamma_0_to_fp16, mean = add_79_mean_0_to_fp16, variance = add_79_variance_0_to_fp16, x = reshape_169_cast_fp16)[name = tensor("add_85_cast_fp16")]; tensor input_371_cast_fp16 = silu(x = add_85_cast_fp16)[name = tensor("input_371_cast_fp16")]; tensor hidden_states_219_pad_type_0 = const()[name = tensor("hidden_states_219_pad_type_0"), val = tensor("custom")]; tensor hidden_states_219_pad_0 = const()[name = tensor("hidden_states_219_pad_0"), val = tensor([1, 1, 1, 1])]; tensor hidden_states_219_strides_0 = const()[name = tensor("hidden_states_219_strides_0"), val = tensor([1, 1])]; tensor hidden_states_219_dilations_0 = const()[name = tensor("hidden_states_219_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_219_groups_0 = const()[name = tensor("hidden_states_219_groups_0"), val = tensor(1)]; tensor up_blocks_2_resnets_0_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(651085120)))]; tensor up_blocks_2_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(673203584)))]; tensor hidden_states_219_cast_fp16 = conv(bias = up_blocks_2_resnets_0_conv1_bias_to_fp16, dilations = hidden_states_219_dilations_0, groups = hidden_states_219_groups_0, pad = hidden_states_219_pad_0, pad_type = hidden_states_219_pad_type_0, strides = hidden_states_219_strides_0, weight = up_blocks_2_resnets_0_conv1_weight_to_fp16, x = input_371_cast_fp16)[name = tensor("hidden_states_219_cast_fp16")]; tensor temb_33_pad_type_0 = const()[name = tensor("temb_33_pad_type_0"), val = tensor("valid")]; tensor temb_33_strides_0 = const()[name = tensor("temb_33_strides_0"), val = tensor([1, 1])]; tensor temb_33_pad_0 = const()[name = tensor("temb_33_pad_0"), val = tensor([0, 0, 0, 0])]; tensor temb_33_dilations_0 = const()[name = tensor("temb_33_dilations_0"), val = tensor([1, 1])]; tensor temb_33_groups_0 = const()[name = tensor("temb_33_groups_0"), val = tensor(1)]; tensor up_blocks_2_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(673204928)))]; tensor up_blocks_2_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(674843392)))]; tensor temb_33_cast_fp16 = conv(bias = up_blocks_2_resnets_0_time_emb_proj_bias_to_fp16, dilations = temb_33_dilations_0, groups = temb_33_groups_0, pad = temb_33_pad_0, pad_type = temb_33_pad_type_0, strides = temb_33_strides_0, weight = up_blocks_2_resnets_0_time_emb_proj_weight_to_fp16, x = cast_9)[name = tensor("temb_33_cast_fp16")]; tensor input_375_cast_fp16 = add(x = hidden_states_219_cast_fp16, y = temb_33_cast_fp16)[name = tensor("input_375_cast_fp16")]; tensor reshape_172_shape_0 = const()[name = tensor("reshape_172_shape_0"), val = tensor([2, 32, 20, 32, 32])]; tensor reshape_172_cast_fp16 = reshape(shape = reshape_172_shape_0, x = input_375_cast_fp16)[name = tensor("reshape_172_cast_fp16")]; tensor reduce_mean_129_axes_0 = const()[name = tensor("reduce_mean_129_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_129_keep_dims_0 = const()[name = tensor("reduce_mean_129_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_129_cast_fp16 = reduce_mean(axes = reduce_mean_129_axes_0, keep_dims = reduce_mean_129_keep_dims_0, x = reshape_172_cast_fp16)[name = tensor("reduce_mean_129_cast_fp16")]; tensor sub_86_cast_fp16 = sub(x = reshape_172_cast_fp16, y = reduce_mean_129_cast_fp16)[name = tensor("sub_86_cast_fp16")]; tensor square_43_cast_fp16 = square(x = sub_86_cast_fp16)[name = tensor("square_43_cast_fp16")]; tensor reduce_mean_131_axes_0 = const()[name = tensor("reduce_mean_131_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_131_keep_dims_0 = const()[name = tensor("reduce_mean_131_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_131_cast_fp16 = reduce_mean(axes = reduce_mean_131_axes_0, keep_dims = reduce_mean_131_keep_dims_0, x = square_43_cast_fp16)[name = tensor("reduce_mean_131_cast_fp16")]; tensor add_86_y_0_to_fp16 = const()[name = tensor("add_86_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_86_cast_fp16 = add(x = reduce_mean_131_cast_fp16, y = add_86_y_0_to_fp16)[name = tensor("add_86_cast_fp16")]; tensor sqrt_43_cast_fp16 = sqrt(x = add_86_cast_fp16)[name = tensor("sqrt_43_cast_fp16")]; tensor real_div_43_cast_fp16 = real_div(x = sub_86_cast_fp16, y = sqrt_43_cast_fp16)[name = tensor("real_div_43_cast_fp16")]; tensor reshape_173_shape_0 = const()[name = tensor("reshape_173_shape_0"), val = tensor([2, 640, 32, 32])]; tensor reshape_173_cast_fp16 = reshape(shape = reshape_173_shape_0, x = real_div_43_cast_fp16)[name = tensor("reshape_173_cast_fp16")]; tensor add_87_gamma_0_to_fp16 = const()[name = tensor("add_87_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(674844736)))]; tensor add_87_beta_0_to_fp16 = const()[name = tensor("add_87_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(674846080)))]; tensor add_87_epsilon_0_to_fp16 = const()[name = tensor("add_87_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_87_cast_fp16 = batch_norm(beta = add_87_beta_0_to_fp16, epsilon = add_87_epsilon_0_to_fp16, gamma = add_87_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_173_cast_fp16)[name = tensor("add_87_cast_fp16")]; tensor input_379_cast_fp16 = silu(x = add_87_cast_fp16)[name = tensor("input_379_cast_fp16")]; tensor hidden_states_221_pad_type_0 = const()[name = tensor("hidden_states_221_pad_type_0"), val = tensor("custom")]; tensor hidden_states_221_pad_0 = const()[name = tensor("hidden_states_221_pad_0"), val = tensor([1, 1, 1, 1])]; tensor hidden_states_221_strides_0 = const()[name = tensor("hidden_states_221_strides_0"), val = tensor([1, 1])]; tensor hidden_states_221_dilations_0 = const()[name = tensor("hidden_states_221_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_221_groups_0 = const()[name = tensor("hidden_states_221_groups_0"), val = tensor(1)]; tensor up_blocks_2_resnets_0_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(674847424)))]; tensor up_blocks_2_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(682220288)))]; tensor hidden_states_221_cast_fp16 = conv(bias = up_blocks_2_resnets_0_conv2_bias_to_fp16, dilations = hidden_states_221_dilations_0, groups = hidden_states_221_groups_0, pad = hidden_states_221_pad_0, pad_type = hidden_states_221_pad_type_0, strides = hidden_states_221_strides_0, weight = up_blocks_2_resnets_0_conv2_weight_to_fp16, x = input_379_cast_fp16)[name = tensor("hidden_states_221_cast_fp16")]; tensor x_17_pad_type_0 = const()[name = tensor("x_17_pad_type_0"), val = tensor("valid")]; tensor x_17_strides_0 = const()[name = tensor("x_17_strides_0"), val = tensor([1, 1])]; tensor x_17_pad_0 = const()[name = tensor("x_17_pad_0"), val = tensor([0, 0, 0, 0])]; tensor x_17_dilations_0 = const()[name = tensor("x_17_dilations_0"), val = tensor([1, 1])]; tensor x_17_groups_0 = const()[name = tensor("x_17_groups_0"), val = tensor(1)]; tensor up_blocks_2_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(682221632)))]; tensor up_blocks_2_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(684679296)))]; tensor x_17_cast_fp16 = conv(bias = up_blocks_2_resnets_0_conv_shortcut_bias_to_fp16, dilations = x_17_dilations_0, groups = x_17_groups_0, pad = x_17_pad_0, pad_type = x_17_pad_type_0, strides = x_17_strides_0, weight = up_blocks_2_resnets_0_conv_shortcut_weight_to_fp16, x = input_367_cast_fp16)[name = tensor("x_17_cast_fp16")]; tensor hidden_states_223_cast_fp16 = add(x = x_17_cast_fp16, y = hidden_states_221_cast_fp16)[name = tensor("hidden_states_223_cast_fp16")]; tensor reshape_176_shape_0 = const()[name = tensor("reshape_176_shape_0"), val = tensor([2, 32, 20, 32, 32])]; tensor reshape_176_cast_fp16 = reshape(shape = reshape_176_shape_0, x = hidden_states_223_cast_fp16)[name = tensor("reshape_176_cast_fp16")]; tensor reduce_mean_132_axes_0 = const()[name = tensor("reduce_mean_132_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_132_keep_dims_0 = const()[name = tensor("reduce_mean_132_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_132_cast_fp16 = reduce_mean(axes = reduce_mean_132_axes_0, keep_dims = reduce_mean_132_keep_dims_0, x = reshape_176_cast_fp16)[name = tensor("reduce_mean_132_cast_fp16")]; tensor sub_88_cast_fp16 = sub(x = reshape_176_cast_fp16, y = reduce_mean_132_cast_fp16)[name = tensor("sub_88_cast_fp16")]; tensor square_44_cast_fp16 = square(x = sub_88_cast_fp16)[name = tensor("square_44_cast_fp16")]; tensor reduce_mean_134_axes_0 = const()[name = tensor("reduce_mean_134_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_134_keep_dims_0 = const()[name = tensor("reduce_mean_134_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_134_cast_fp16 = reduce_mean(axes = reduce_mean_134_axes_0, keep_dims = reduce_mean_134_keep_dims_0, x = square_44_cast_fp16)[name = tensor("reduce_mean_134_cast_fp16")]; tensor add_88_y_0_to_fp16 = const()[name = tensor("add_88_y_0_to_fp16"), val = tensor(0x1.1p-20)]; tensor add_88_cast_fp16 = add(x = reduce_mean_134_cast_fp16, y = add_88_y_0_to_fp16)[name = tensor("add_88_cast_fp16")]; tensor sqrt_44_cast_fp16 = sqrt(x = add_88_cast_fp16)[name = tensor("sqrt_44_cast_fp16")]; tensor real_div_44_cast_fp16 = real_div(x = sub_88_cast_fp16, y = sqrt_44_cast_fp16)[name = tensor("real_div_44_cast_fp16")]; tensor reshape_177_shape_0 = const()[name = tensor("reshape_177_shape_0"), val = tensor([2, 640, 32, 32])]; tensor reshape_177_cast_fp16 = reshape(shape = reshape_177_shape_0, x = real_div_44_cast_fp16)[name = tensor("reshape_177_cast_fp16")]; tensor add_89_gamma_0_to_fp16 = const()[name = tensor("add_89_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(684680640)))]; tensor add_89_beta_0_to_fp16 = const()[name = tensor("add_89_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(684681984)))]; tensor add_89_epsilon_0_to_fp16 = const()[name = tensor("add_89_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_89_cast_fp16 = batch_norm(beta = add_89_beta_0_to_fp16, epsilon = add_89_epsilon_0_to_fp16, gamma = add_89_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_177_cast_fp16)[name = tensor("add_89_cast_fp16")]; tensor hidden_states_225_pad_type_0 = const()[name = tensor("hidden_states_225_pad_type_0"), val = tensor("valid")]; tensor hidden_states_225_strides_0 = const()[name = tensor("hidden_states_225_strides_0"), val = tensor([1, 1])]; tensor hidden_states_225_pad_0 = const()[name = tensor("hidden_states_225_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_225_dilations_0 = const()[name = tensor("hidden_states_225_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_225_groups_0 = const()[name = tensor("hidden_states_225_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(684683328)))]; tensor up_blocks_2_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(685502592)))]; tensor hidden_states_225_cast_fp16 = conv(bias = up_blocks_2_attentions_0_proj_in_bias_to_fp16, dilations = hidden_states_225_dilations_0, groups = hidden_states_225_groups_0, pad = hidden_states_225_pad_0, pad_type = hidden_states_225_pad_type_0, strides = hidden_states_225_strides_0, weight = up_blocks_2_attentions_0_proj_in_weight_to_fp16, x = add_89_cast_fp16)[name = tensor("hidden_states_225_cast_fp16")]; tensor var_8267 = const()[name = tensor("op_8267"), val = tensor([2, 640, 1, 1024])]; tensor inputs_61_cast_fp16 = reshape(shape = var_8267, x = hidden_states_225_cast_fp16)[name = tensor("inputs_61_cast_fp16")]; tensor hidden_states_227_axes_0 = const()[name = tensor("hidden_states_227_axes_0"), val = tensor([1])]; tensor hidden_states_227_gamma_0_to_fp16 = const()[name = tensor("hidden_states_227_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(685503936)))]; tensor hidden_states_227_beta_0_to_fp16 = const()[name = tensor("hidden_states_227_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(685505280)))]; tensor var_8283_to_fp16 = const()[name = tensor("op_8283_to_fp16"), val = tensor(0x1.5p-17)]; tensor hidden_states_227_cast_fp16 = layer_norm(axes = hidden_states_227_axes_0, beta = hidden_states_227_beta_0_to_fp16, epsilon = var_8283_to_fp16, gamma = hidden_states_227_gamma_0_to_fp16, x = inputs_61_cast_fp16)[name = tensor("hidden_states_227_cast_fp16")]; tensor q_41_pad_type_0 = const()[name = tensor("q_41_pad_type_0"), val = tensor("valid")]; tensor q_41_strides_0 = const()[name = tensor("q_41_strides_0"), val = tensor([1, 1])]; tensor q_41_pad_0 = const()[name = tensor("q_41_pad_0"), val = tensor([0, 0, 0, 0])]; tensor q_41_dilations_0 = const()[name = tensor("q_41_dilations_0"), val = tensor([1, 1])]; tensor q_41_groups_0 = const()[name = tensor("q_41_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(685506624)))]; tensor q_41_cast_fp16 = conv(dilations = q_41_dilations_0, groups = q_41_groups_0, pad = q_41_pad_0, pad_type = q_41_pad_type_0, strides = q_41_strides_0, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_227_cast_fp16)[name = tensor("q_41_cast_fp16")]; tensor k_81_pad_type_0 = const()[name = tensor("k_81_pad_type_0"), val = tensor("valid")]; tensor k_81_strides_0 = const()[name = tensor("k_81_strides_0"), val = tensor([1, 1])]; tensor k_81_pad_0 = const()[name = tensor("k_81_pad_0"), val = tensor([0, 0, 0, 0])]; tensor k_81_dilations_0 = const()[name = tensor("k_81_dilations_0"), val = tensor([1, 1])]; tensor k_81_groups_0 = const()[name = tensor("k_81_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(686325888)))]; tensor k_81_cast_fp16 = conv(dilations = k_81_dilations_0, groups = k_81_groups_0, pad = k_81_pad_0, pad_type = k_81_pad_type_0, strides = k_81_strides_0, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_227_cast_fp16)[name = tensor("k_81_cast_fp16")]; tensor v_41_pad_type_0 = const()[name = tensor("v_41_pad_type_0"), val = tensor("valid")]; tensor v_41_strides_0 = const()[name = tensor("v_41_strides_0"), val = tensor([1, 1])]; tensor v_41_pad_0 = const()[name = tensor("v_41_pad_0"), val = tensor([0, 0, 0, 0])]; tensor v_41_dilations_0 = const()[name = tensor("v_41_dilations_0"), val = tensor([1, 1])]; tensor v_41_groups_0 = const()[name = tensor("v_41_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(687145152)))]; tensor v_41_cast_fp16 = conv(dilations = v_41_dilations_0, groups = v_41_groups_0, pad = v_41_pad_0, pad_type = v_41_pad_type_0, strides = v_41_strides_0, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_227_cast_fp16)[name = tensor("v_41_cast_fp16")]; tensor var_8316_begin_0 = const()[name = tensor("op_8316_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8316_end_0 = const()[name = tensor("op_8316_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_8316_end_mask_0 = const()[name = tensor("op_8316_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8316_cast_fp16 = slice_by_index(begin = var_8316_begin_0, end = var_8316_end_0, end_mask = var_8316_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8316_cast_fp16")]; tensor var_8320_begin_0 = const()[name = tensor("op_8320_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_8320_end_0 = const()[name = tensor("op_8320_end_0"), val = tensor([2, 160, 1, 1024])]; tensor var_8320_end_mask_0 = const()[name = tensor("op_8320_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8320_cast_fp16 = slice_by_index(begin = var_8320_begin_0, end = var_8320_end_0, end_mask = var_8320_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8320_cast_fp16")]; tensor var_8324_begin_0 = const()[name = tensor("op_8324_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_8324_end_0 = const()[name = tensor("op_8324_end_0"), val = tensor([2, 240, 1, 1024])]; tensor var_8324_end_mask_0 = const()[name = tensor("op_8324_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8324_cast_fp16 = slice_by_index(begin = var_8324_begin_0, end = var_8324_end_0, end_mask = var_8324_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8324_cast_fp16")]; tensor var_8328_begin_0 = const()[name = tensor("op_8328_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_8328_end_0 = const()[name = tensor("op_8328_end_0"), val = tensor([2, 320, 1, 1024])]; tensor var_8328_end_mask_0 = const()[name = tensor("op_8328_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8328_cast_fp16 = slice_by_index(begin = var_8328_begin_0, end = var_8328_end_0, end_mask = var_8328_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8328_cast_fp16")]; tensor var_8332_begin_0 = const()[name = tensor("op_8332_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_8332_end_0 = const()[name = tensor("op_8332_end_0"), val = tensor([2, 400, 1, 1024])]; tensor var_8332_end_mask_0 = const()[name = tensor("op_8332_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8332_cast_fp16 = slice_by_index(begin = var_8332_begin_0, end = var_8332_end_0, end_mask = var_8332_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8332_cast_fp16")]; tensor var_8336_begin_0 = const()[name = tensor("op_8336_begin_0"), val = tensor([0, 400, 0, 0])]; tensor var_8336_end_0 = const()[name = tensor("op_8336_end_0"), val = tensor([2, 480, 1, 1024])]; tensor var_8336_end_mask_0 = const()[name = tensor("op_8336_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8336_cast_fp16 = slice_by_index(begin = var_8336_begin_0, end = var_8336_end_0, end_mask = var_8336_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8336_cast_fp16")]; tensor var_8340_begin_0 = const()[name = tensor("op_8340_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_8340_end_0 = const()[name = tensor("op_8340_end_0"), val = tensor([2, 560, 1, 1024])]; tensor var_8340_end_mask_0 = const()[name = tensor("op_8340_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8340_cast_fp16 = slice_by_index(begin = var_8340_begin_0, end = var_8340_end_0, end_mask = var_8340_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8340_cast_fp16")]; tensor var_8344_begin_0 = const()[name = tensor("op_8344_begin_0"), val = tensor([0, 560, 0, 0])]; tensor var_8344_end_0 = const()[name = tensor("op_8344_end_0"), val = tensor([2, 1, 1, 1024])]; tensor var_8344_end_mask_0 = const()[name = tensor("op_8344_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_8344_cast_fp16 = slice_by_index(begin = var_8344_begin_0, end = var_8344_end_0, end_mask = var_8344_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8344_cast_fp16")]; tensor var_8347_begin_0 = const()[name = tensor("op_8347_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8347_end_0 = const()[name = tensor("op_8347_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_8347_end_mask_0 = const()[name = tensor("op_8347_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8347_cast_fp16 = slice_by_index(begin = var_8347_begin_0, end = var_8347_end_0, end_mask = var_8347_end_mask_0, x = var_8316_cast_fp16)[name = tensor("op_8347_cast_fp16")]; tensor var_8348_begin_0 = const()[name = tensor("op_8348_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_8348_end_0 = const()[name = tensor("op_8348_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_8348_end_mask_0 = const()[name = tensor("op_8348_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_8348_cast_fp16 = slice_by_index(begin = var_8348_begin_0, end = var_8348_end_0, end_mask = var_8348_end_mask_0, x = var_8316_cast_fp16)[name = tensor("op_8348_cast_fp16")]; tensor var_8349_begin_0 = const()[name = tensor("op_8349_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8349_end_0 = const()[name = tensor("op_8349_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_8349_end_mask_0 = const()[name = tensor("op_8349_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8349_cast_fp16 = slice_by_index(begin = var_8349_begin_0, end = var_8349_end_0, end_mask = var_8349_end_mask_0, x = var_8320_cast_fp16)[name = tensor("op_8349_cast_fp16")]; tensor var_8350_begin_0 = const()[name = tensor("op_8350_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_8350_end_0 = const()[name = tensor("op_8350_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_8350_end_mask_0 = const()[name = tensor("op_8350_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_8350_cast_fp16 = slice_by_index(begin = var_8350_begin_0, end = var_8350_end_0, end_mask = var_8350_end_mask_0, x = var_8320_cast_fp16)[name = tensor("op_8350_cast_fp16")]; tensor var_8351_begin_0 = const()[name = tensor("op_8351_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8351_end_0 = const()[name = tensor("op_8351_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_8351_end_mask_0 = const()[name = tensor("op_8351_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8351_cast_fp16 = slice_by_index(begin = var_8351_begin_0, end = var_8351_end_0, end_mask = var_8351_end_mask_0, x = var_8324_cast_fp16)[name = tensor("op_8351_cast_fp16")]; tensor var_8352_begin_0 = const()[name = tensor("op_8352_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_8352_end_0 = const()[name = tensor("op_8352_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_8352_end_mask_0 = const()[name = tensor("op_8352_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_8352_cast_fp16 = slice_by_index(begin = var_8352_begin_0, end = var_8352_end_0, end_mask = var_8352_end_mask_0, x = var_8324_cast_fp16)[name = tensor("op_8352_cast_fp16")]; tensor var_8353_begin_0 = const()[name = tensor("op_8353_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8353_end_0 = const()[name = tensor("op_8353_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_8353_end_mask_0 = const()[name = tensor("op_8353_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8353_cast_fp16 = slice_by_index(begin = var_8353_begin_0, end = var_8353_end_0, end_mask = var_8353_end_mask_0, x = var_8328_cast_fp16)[name = tensor("op_8353_cast_fp16")]; tensor var_8354_begin_0 = const()[name = tensor("op_8354_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_8354_end_0 = const()[name = tensor("op_8354_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_8354_end_mask_0 = const()[name = tensor("op_8354_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_8354_cast_fp16 = slice_by_index(begin = var_8354_begin_0, end = var_8354_end_0, end_mask = var_8354_end_mask_0, x = var_8328_cast_fp16)[name = tensor("op_8354_cast_fp16")]; tensor var_8355_begin_0 = const()[name = tensor("op_8355_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8355_end_0 = const()[name = tensor("op_8355_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_8355_end_mask_0 = const()[name = tensor("op_8355_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8355_cast_fp16 = slice_by_index(begin = var_8355_begin_0, end = var_8355_end_0, end_mask = var_8355_end_mask_0, x = var_8332_cast_fp16)[name = tensor("op_8355_cast_fp16")]; tensor var_8356_begin_0 = const()[name = tensor("op_8356_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_8356_end_0 = const()[name = tensor("op_8356_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_8356_end_mask_0 = const()[name = tensor("op_8356_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_8356_cast_fp16 = slice_by_index(begin = var_8356_begin_0, end = var_8356_end_0, end_mask = var_8356_end_mask_0, x = var_8332_cast_fp16)[name = tensor("op_8356_cast_fp16")]; tensor var_8357_begin_0 = const()[name = tensor("op_8357_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8357_end_0 = const()[name = tensor("op_8357_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_8357_end_mask_0 = const()[name = tensor("op_8357_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8357_cast_fp16 = slice_by_index(begin = var_8357_begin_0, end = var_8357_end_0, end_mask = var_8357_end_mask_0, x = var_8336_cast_fp16)[name = tensor("op_8357_cast_fp16")]; tensor var_8358_begin_0 = const()[name = tensor("op_8358_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_8358_end_0 = const()[name = tensor("op_8358_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_8358_end_mask_0 = const()[name = tensor("op_8358_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_8358_cast_fp16 = slice_by_index(begin = var_8358_begin_0, end = var_8358_end_0, end_mask = var_8358_end_mask_0, x = var_8336_cast_fp16)[name = tensor("op_8358_cast_fp16")]; tensor var_8359_begin_0 = const()[name = tensor("op_8359_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8359_end_0 = const()[name = tensor("op_8359_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_8359_end_mask_0 = const()[name = tensor("op_8359_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8359_cast_fp16 = slice_by_index(begin = var_8359_begin_0, end = var_8359_end_0, end_mask = var_8359_end_mask_0, x = var_8340_cast_fp16)[name = tensor("op_8359_cast_fp16")]; tensor var_8360_begin_0 = const()[name = tensor("op_8360_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_8360_end_0 = const()[name = tensor("op_8360_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_8360_end_mask_0 = const()[name = tensor("op_8360_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_8360_cast_fp16 = slice_by_index(begin = var_8360_begin_0, end = var_8360_end_0, end_mask = var_8360_end_mask_0, x = var_8340_cast_fp16)[name = tensor("op_8360_cast_fp16")]; tensor var_8361_begin_0 = const()[name = tensor("op_8361_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8361_end_0 = const()[name = tensor("op_8361_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_8361_end_mask_0 = const()[name = tensor("op_8361_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8361_cast_fp16 = slice_by_index(begin = var_8361_begin_0, end = var_8361_end_0, end_mask = var_8361_end_mask_0, x = var_8344_cast_fp16)[name = tensor("op_8361_cast_fp16")]; tensor var_8362_begin_0 = const()[name = tensor("op_8362_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_8362_end_0 = const()[name = tensor("op_8362_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_8362_end_mask_0 = const()[name = tensor("op_8362_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_8362_cast_fp16 = slice_by_index(begin = var_8362_begin_0, end = var_8362_end_0, end_mask = var_8362_end_mask_0, x = var_8344_cast_fp16)[name = tensor("op_8362_cast_fp16")]; tensor k_83_perm_0 = const()[name = tensor("k_83_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_8367_begin_0 = const()[name = tensor("op_8367_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8367_end_0 = const()[name = tensor("op_8367_end_0"), val = tensor([2, 1024, 1, 80])]; tensor var_8367_end_mask_0 = const()[name = tensor("op_8367_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_83_cast_fp16 = transpose(perm = k_83_perm_0, x = k_81_cast_fp16)[name = tensor("transpose_11")]; tensor var_8367_cast_fp16 = slice_by_index(begin = var_8367_begin_0, end = var_8367_end_0, end_mask = var_8367_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8367_cast_fp16")]; tensor var_8371_begin_0 = const()[name = tensor("op_8371_begin_0"), val = tensor([0, 0, 0, 80])]; tensor var_8371_end_0 = const()[name = tensor("op_8371_end_0"), val = tensor([2, 1024, 1, 160])]; tensor var_8371_end_mask_0 = const()[name = tensor("op_8371_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8371_cast_fp16 = slice_by_index(begin = var_8371_begin_0, end = var_8371_end_0, end_mask = var_8371_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8371_cast_fp16")]; tensor var_8375_begin_0 = const()[name = tensor("op_8375_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_8375_end_0 = const()[name = tensor("op_8375_end_0"), val = tensor([2, 1024, 1, 240])]; tensor var_8375_end_mask_0 = const()[name = tensor("op_8375_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8375_cast_fp16 = slice_by_index(begin = var_8375_begin_0, end = var_8375_end_0, end_mask = var_8375_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8375_cast_fp16")]; tensor var_8379_begin_0 = const()[name = tensor("op_8379_begin_0"), val = tensor([0, 0, 0, 240])]; tensor var_8379_end_0 = const()[name = tensor("op_8379_end_0"), val = tensor([2, 1024, 1, 320])]; tensor var_8379_end_mask_0 = const()[name = tensor("op_8379_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8379_cast_fp16 = slice_by_index(begin = var_8379_begin_0, end = var_8379_end_0, end_mask = var_8379_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8379_cast_fp16")]; tensor var_8383_begin_0 = const()[name = tensor("op_8383_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_8383_end_0 = const()[name = tensor("op_8383_end_0"), val = tensor([2, 1024, 1, 400])]; tensor var_8383_end_mask_0 = const()[name = tensor("op_8383_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8383_cast_fp16 = slice_by_index(begin = var_8383_begin_0, end = var_8383_end_0, end_mask = var_8383_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8383_cast_fp16")]; tensor var_8387_begin_0 = const()[name = tensor("op_8387_begin_0"), val = tensor([0, 0, 0, 400])]; tensor var_8387_end_0 = const()[name = tensor("op_8387_end_0"), val = tensor([2, 1024, 1, 480])]; tensor var_8387_end_mask_0 = const()[name = tensor("op_8387_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8387_cast_fp16 = slice_by_index(begin = var_8387_begin_0, end = var_8387_end_0, end_mask = var_8387_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8387_cast_fp16")]; tensor var_8391_begin_0 = const()[name = tensor("op_8391_begin_0"), val = tensor([0, 0, 0, 480])]; tensor var_8391_end_0 = const()[name = tensor("op_8391_end_0"), val = tensor([2, 1024, 1, 560])]; tensor var_8391_end_mask_0 = const()[name = tensor("op_8391_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8391_cast_fp16 = slice_by_index(begin = var_8391_begin_0, end = var_8391_end_0, end_mask = var_8391_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8391_cast_fp16")]; tensor var_8395_begin_0 = const()[name = tensor("op_8395_begin_0"), val = tensor([0, 0, 0, 560])]; tensor var_8395_end_0 = const()[name = tensor("op_8395_end_0"), val = tensor([2, 1024, 1, 1])]; tensor var_8395_end_mask_0 = const()[name = tensor("op_8395_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_8395_cast_fp16 = slice_by_index(begin = var_8395_begin_0, end = var_8395_end_0, end_mask = var_8395_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8395_cast_fp16")]; tensor var_8397_begin_0 = const()[name = tensor("op_8397_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8397_end_0 = const()[name = tensor("op_8397_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_8397_end_mask_0 = const()[name = tensor("op_8397_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8397_cast_fp16 = slice_by_index(begin = var_8397_begin_0, end = var_8397_end_0, end_mask = var_8397_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8397_cast_fp16")]; tensor var_8401_begin_0 = const()[name = tensor("op_8401_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_8401_end_0 = const()[name = tensor("op_8401_end_0"), val = tensor([2, 160, 1, 1024])]; tensor var_8401_end_mask_0 = const()[name = tensor("op_8401_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8401_cast_fp16 = slice_by_index(begin = var_8401_begin_0, end = var_8401_end_0, end_mask = var_8401_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8401_cast_fp16")]; tensor var_8405_begin_0 = const()[name = tensor("op_8405_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_8405_end_0 = const()[name = tensor("op_8405_end_0"), val = tensor([2, 240, 1, 1024])]; tensor var_8405_end_mask_0 = const()[name = tensor("op_8405_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8405_cast_fp16 = slice_by_index(begin = var_8405_begin_0, end = var_8405_end_0, end_mask = var_8405_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8405_cast_fp16")]; tensor var_8409_begin_0 = const()[name = tensor("op_8409_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_8409_end_0 = const()[name = tensor("op_8409_end_0"), val = tensor([2, 320, 1, 1024])]; tensor var_8409_end_mask_0 = const()[name = tensor("op_8409_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8409_cast_fp16 = slice_by_index(begin = var_8409_begin_0, end = var_8409_end_0, end_mask = var_8409_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8409_cast_fp16")]; tensor var_8413_begin_0 = const()[name = tensor("op_8413_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_8413_end_0 = const()[name = tensor("op_8413_end_0"), val = tensor([2, 400, 1, 1024])]; tensor var_8413_end_mask_0 = const()[name = tensor("op_8413_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8413_cast_fp16 = slice_by_index(begin = var_8413_begin_0, end = var_8413_end_0, end_mask = var_8413_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8413_cast_fp16")]; tensor var_8417_begin_0 = const()[name = tensor("op_8417_begin_0"), val = tensor([0, 400, 0, 0])]; tensor var_8417_end_0 = const()[name = tensor("op_8417_end_0"), val = tensor([2, 480, 1, 1024])]; tensor var_8417_end_mask_0 = const()[name = tensor("op_8417_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8417_cast_fp16 = slice_by_index(begin = var_8417_begin_0, end = var_8417_end_0, end_mask = var_8417_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8417_cast_fp16")]; tensor var_8421_begin_0 = const()[name = tensor("op_8421_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_8421_end_0 = const()[name = tensor("op_8421_end_0"), val = tensor([2, 560, 1, 1024])]; tensor var_8421_end_mask_0 = const()[name = tensor("op_8421_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8421_cast_fp16 = slice_by_index(begin = var_8421_begin_0, end = var_8421_end_0, end_mask = var_8421_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8421_cast_fp16")]; tensor var_8425_begin_0 = const()[name = tensor("op_8425_begin_0"), val = tensor([0, 560, 0, 0])]; tensor var_8425_end_0 = const()[name = tensor("op_8425_end_0"), val = tensor([2, 1, 1, 1024])]; tensor var_8425_end_mask_0 = const()[name = tensor("op_8425_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_8425_cast_fp16 = slice_by_index(begin = var_8425_begin_0, end = var_8425_end_0, end_mask = var_8425_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8425_cast_fp16")]; tensor var_8429_equation_0 = const()[name = tensor("op_8429_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8429_cast_fp16 = einsum(equation = var_8429_equation_0, values = (var_8367_cast_fp16, var_8347_cast_fp16))[name = tensor("op_8429_cast_fp16")]; tensor var_8430_to_fp16 = const()[name = tensor("op_8430_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_641_cast_fp16 = mul(x = var_8429_cast_fp16, y = var_8430_to_fp16)[name = tensor("aw_chunk_641_cast_fp16")]; tensor var_8433_equation_0 = const()[name = tensor("op_8433_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8433_cast_fp16 = einsum(equation = var_8433_equation_0, values = (var_8367_cast_fp16, var_8348_cast_fp16))[name = tensor("op_8433_cast_fp16")]; tensor var_8434_to_fp16 = const()[name = tensor("op_8434_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_643_cast_fp16 = mul(x = var_8433_cast_fp16, y = var_8434_to_fp16)[name = tensor("aw_chunk_643_cast_fp16")]; tensor var_8437_equation_0 = const()[name = tensor("op_8437_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8437_cast_fp16 = einsum(equation = var_8437_equation_0, values = (var_8371_cast_fp16, var_8349_cast_fp16))[name = tensor("op_8437_cast_fp16")]; tensor var_8438_to_fp16 = const()[name = tensor("op_8438_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_645_cast_fp16 = mul(x = var_8437_cast_fp16, y = var_8438_to_fp16)[name = tensor("aw_chunk_645_cast_fp16")]; tensor var_8441_equation_0 = const()[name = tensor("op_8441_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8441_cast_fp16 = einsum(equation = var_8441_equation_0, values = (var_8371_cast_fp16, var_8350_cast_fp16))[name = tensor("op_8441_cast_fp16")]; tensor var_8442_to_fp16 = const()[name = tensor("op_8442_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_647_cast_fp16 = mul(x = var_8441_cast_fp16, y = var_8442_to_fp16)[name = tensor("aw_chunk_647_cast_fp16")]; tensor var_8445_equation_0 = const()[name = tensor("op_8445_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8445_cast_fp16 = einsum(equation = var_8445_equation_0, values = (var_8375_cast_fp16, var_8351_cast_fp16))[name = tensor("op_8445_cast_fp16")]; tensor var_8446_to_fp16 = const()[name = tensor("op_8446_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_649_cast_fp16 = mul(x = var_8445_cast_fp16, y = var_8446_to_fp16)[name = tensor("aw_chunk_649_cast_fp16")]; tensor var_8449_equation_0 = const()[name = tensor("op_8449_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8449_cast_fp16 = einsum(equation = var_8449_equation_0, values = (var_8375_cast_fp16, var_8352_cast_fp16))[name = tensor("op_8449_cast_fp16")]; tensor var_8450_to_fp16 = const()[name = tensor("op_8450_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_651_cast_fp16 = mul(x = var_8449_cast_fp16, y = var_8450_to_fp16)[name = tensor("aw_chunk_651_cast_fp16")]; tensor var_8453_equation_0 = const()[name = tensor("op_8453_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8453_cast_fp16 = einsum(equation = var_8453_equation_0, values = (var_8379_cast_fp16, var_8353_cast_fp16))[name = tensor("op_8453_cast_fp16")]; tensor var_8454_to_fp16 = const()[name = tensor("op_8454_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_653_cast_fp16 = mul(x = var_8453_cast_fp16, y = var_8454_to_fp16)[name = tensor("aw_chunk_653_cast_fp16")]; tensor var_8457_equation_0 = const()[name = tensor("op_8457_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8457_cast_fp16 = einsum(equation = var_8457_equation_0, values = (var_8379_cast_fp16, var_8354_cast_fp16))[name = tensor("op_8457_cast_fp16")]; tensor var_8458_to_fp16 = const()[name = tensor("op_8458_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_655_cast_fp16 = mul(x = var_8457_cast_fp16, y = var_8458_to_fp16)[name = tensor("aw_chunk_655_cast_fp16")]; tensor var_8461_equation_0 = const()[name = tensor("op_8461_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8461_cast_fp16 = einsum(equation = var_8461_equation_0, values = (var_8383_cast_fp16, var_8355_cast_fp16))[name = tensor("op_8461_cast_fp16")]; tensor var_8462_to_fp16 = const()[name = tensor("op_8462_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_657_cast_fp16 = mul(x = var_8461_cast_fp16, y = var_8462_to_fp16)[name = tensor("aw_chunk_657_cast_fp16")]; tensor var_8465_equation_0 = const()[name = tensor("op_8465_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8465_cast_fp16 = einsum(equation = var_8465_equation_0, values = (var_8383_cast_fp16, var_8356_cast_fp16))[name = tensor("op_8465_cast_fp16")]; tensor var_8466_to_fp16 = const()[name = tensor("op_8466_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_659_cast_fp16 = mul(x = var_8465_cast_fp16, y = var_8466_to_fp16)[name = tensor("aw_chunk_659_cast_fp16")]; tensor var_8469_equation_0 = const()[name = tensor("op_8469_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8469_cast_fp16 = einsum(equation = var_8469_equation_0, values = (var_8387_cast_fp16, var_8357_cast_fp16))[name = tensor("op_8469_cast_fp16")]; tensor var_8470_to_fp16 = const()[name = tensor("op_8470_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_661_cast_fp16 = mul(x = var_8469_cast_fp16, y = var_8470_to_fp16)[name = tensor("aw_chunk_661_cast_fp16")]; tensor var_8473_equation_0 = const()[name = tensor("op_8473_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8473_cast_fp16 = einsum(equation = var_8473_equation_0, values = (var_8387_cast_fp16, var_8358_cast_fp16))[name = tensor("op_8473_cast_fp16")]; tensor var_8474_to_fp16 = const()[name = tensor("op_8474_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_663_cast_fp16 = mul(x = var_8473_cast_fp16, y = var_8474_to_fp16)[name = tensor("aw_chunk_663_cast_fp16")]; tensor var_8477_equation_0 = const()[name = tensor("op_8477_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8477_cast_fp16 = einsum(equation = var_8477_equation_0, values = (var_8391_cast_fp16, var_8359_cast_fp16))[name = tensor("op_8477_cast_fp16")]; tensor var_8478_to_fp16 = const()[name = tensor("op_8478_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_665_cast_fp16 = mul(x = var_8477_cast_fp16, y = var_8478_to_fp16)[name = tensor("aw_chunk_665_cast_fp16")]; tensor var_8481_equation_0 = const()[name = tensor("op_8481_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8481_cast_fp16 = einsum(equation = var_8481_equation_0, values = (var_8391_cast_fp16, var_8360_cast_fp16))[name = tensor("op_8481_cast_fp16")]; tensor var_8482_to_fp16 = const()[name = tensor("op_8482_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_667_cast_fp16 = mul(x = var_8481_cast_fp16, y = var_8482_to_fp16)[name = tensor("aw_chunk_667_cast_fp16")]; tensor var_8485_equation_0 = const()[name = tensor("op_8485_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8485_cast_fp16 = einsum(equation = var_8485_equation_0, values = (var_8395_cast_fp16, var_8361_cast_fp16))[name = tensor("op_8485_cast_fp16")]; tensor var_8486_to_fp16 = const()[name = tensor("op_8486_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_669_cast_fp16 = mul(x = var_8485_cast_fp16, y = var_8486_to_fp16)[name = tensor("aw_chunk_669_cast_fp16")]; tensor var_8489_equation_0 = const()[name = tensor("op_8489_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8489_cast_fp16 = einsum(equation = var_8489_equation_0, values = (var_8395_cast_fp16, var_8362_cast_fp16))[name = tensor("op_8489_cast_fp16")]; tensor var_8490_to_fp16 = const()[name = tensor("op_8490_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_671_cast_fp16 = mul(x = var_8489_cast_fp16, y = var_8490_to_fp16)[name = tensor("aw_chunk_671_cast_fp16")]; tensor var_8492_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_641_cast_fp16)[name = tensor("op_8492_cast_fp16")]; tensor var_8493_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_643_cast_fp16)[name = tensor("op_8493_cast_fp16")]; tensor var_8494_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_645_cast_fp16)[name = tensor("op_8494_cast_fp16")]; tensor var_8495_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_647_cast_fp16)[name = tensor("op_8495_cast_fp16")]; tensor var_8496_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_649_cast_fp16)[name = tensor("op_8496_cast_fp16")]; tensor var_8497_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_651_cast_fp16)[name = tensor("op_8497_cast_fp16")]; tensor var_8498_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_653_cast_fp16)[name = tensor("op_8498_cast_fp16")]; tensor var_8499_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_655_cast_fp16)[name = tensor("op_8499_cast_fp16")]; tensor var_8500_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_657_cast_fp16)[name = tensor("op_8500_cast_fp16")]; tensor var_8501_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_659_cast_fp16)[name = tensor("op_8501_cast_fp16")]; tensor var_8502_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_661_cast_fp16)[name = tensor("op_8502_cast_fp16")]; tensor var_8503_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_663_cast_fp16)[name = tensor("op_8503_cast_fp16")]; tensor var_8504_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_665_cast_fp16)[name = tensor("op_8504_cast_fp16")]; tensor var_8505_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_667_cast_fp16)[name = tensor("op_8505_cast_fp16")]; tensor var_8506_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_669_cast_fp16)[name = tensor("op_8506_cast_fp16")]; tensor var_8507_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_671_cast_fp16)[name = tensor("op_8507_cast_fp16")]; tensor var_8509_equation_0 = const()[name = tensor("op_8509_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8509_cast_fp16 = einsum(equation = var_8509_equation_0, values = (var_8397_cast_fp16, var_8492_cast_fp16))[name = tensor("op_8509_cast_fp16")]; tensor var_8511_equation_0 = const()[name = tensor("op_8511_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8511_cast_fp16 = einsum(equation = var_8511_equation_0, values = (var_8397_cast_fp16, var_8493_cast_fp16))[name = tensor("op_8511_cast_fp16")]; tensor var_8513_equation_0 = const()[name = tensor("op_8513_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8513_cast_fp16 = einsum(equation = var_8513_equation_0, values = (var_8401_cast_fp16, var_8494_cast_fp16))[name = tensor("op_8513_cast_fp16")]; tensor var_8515_equation_0 = const()[name = tensor("op_8515_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8515_cast_fp16 = einsum(equation = var_8515_equation_0, values = (var_8401_cast_fp16, var_8495_cast_fp16))[name = tensor("op_8515_cast_fp16")]; tensor var_8517_equation_0 = const()[name = tensor("op_8517_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8517_cast_fp16 = einsum(equation = var_8517_equation_0, values = (var_8405_cast_fp16, var_8496_cast_fp16))[name = tensor("op_8517_cast_fp16")]; tensor var_8519_equation_0 = const()[name = tensor("op_8519_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8519_cast_fp16 = einsum(equation = var_8519_equation_0, values = (var_8405_cast_fp16, var_8497_cast_fp16))[name = tensor("op_8519_cast_fp16")]; tensor var_8521_equation_0 = const()[name = tensor("op_8521_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8521_cast_fp16 = einsum(equation = var_8521_equation_0, values = (var_8409_cast_fp16, var_8498_cast_fp16))[name = tensor("op_8521_cast_fp16")]; tensor var_8523_equation_0 = const()[name = tensor("op_8523_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8523_cast_fp16 = einsum(equation = var_8523_equation_0, values = (var_8409_cast_fp16, var_8499_cast_fp16))[name = tensor("op_8523_cast_fp16")]; tensor var_8525_equation_0 = const()[name = tensor("op_8525_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8525_cast_fp16 = einsum(equation = var_8525_equation_0, values = (var_8413_cast_fp16, var_8500_cast_fp16))[name = tensor("op_8525_cast_fp16")]; tensor var_8527_equation_0 = const()[name = tensor("op_8527_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8527_cast_fp16 = einsum(equation = var_8527_equation_0, values = (var_8413_cast_fp16, var_8501_cast_fp16))[name = tensor("op_8527_cast_fp16")]; tensor var_8529_equation_0 = const()[name = tensor("op_8529_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8529_cast_fp16 = einsum(equation = var_8529_equation_0, values = (var_8417_cast_fp16, var_8502_cast_fp16))[name = tensor("op_8529_cast_fp16")]; tensor var_8531_equation_0 = const()[name = tensor("op_8531_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8531_cast_fp16 = einsum(equation = var_8531_equation_0, values = (var_8417_cast_fp16, var_8503_cast_fp16))[name = tensor("op_8531_cast_fp16")]; tensor var_8533_equation_0 = const()[name = tensor("op_8533_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8533_cast_fp16 = einsum(equation = var_8533_equation_0, values = (var_8421_cast_fp16, var_8504_cast_fp16))[name = tensor("op_8533_cast_fp16")]; tensor var_8535_equation_0 = const()[name = tensor("op_8535_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8535_cast_fp16 = einsum(equation = var_8535_equation_0, values = (var_8421_cast_fp16, var_8505_cast_fp16))[name = tensor("op_8535_cast_fp16")]; tensor var_8537_equation_0 = const()[name = tensor("op_8537_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8537_cast_fp16 = einsum(equation = var_8537_equation_0, values = (var_8425_cast_fp16, var_8506_cast_fp16))[name = tensor("op_8537_cast_fp16")]; tensor var_8539_equation_0 = const()[name = tensor("op_8539_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8539_cast_fp16 = einsum(equation = var_8539_equation_0, values = (var_8425_cast_fp16, var_8507_cast_fp16))[name = tensor("op_8539_cast_fp16")]; tensor var_8541_interleave_0 = const()[name = tensor("op_8541_interleave_0"), val = tensor(false)]; tensor var_8541_cast_fp16 = concat(axis = var_8160, interleave = var_8541_interleave_0, values = (var_8509_cast_fp16, var_8511_cast_fp16))[name = tensor("op_8541_cast_fp16")]; tensor var_8543_interleave_0 = const()[name = tensor("op_8543_interleave_0"), val = tensor(false)]; tensor var_8543_cast_fp16 = concat(axis = var_8160, interleave = var_8543_interleave_0, values = (var_8513_cast_fp16, var_8515_cast_fp16))[name = tensor("op_8543_cast_fp16")]; tensor var_8545_interleave_0 = const()[name = tensor("op_8545_interleave_0"), val = tensor(false)]; tensor var_8545_cast_fp16 = concat(axis = var_8160, interleave = var_8545_interleave_0, values = (var_8517_cast_fp16, var_8519_cast_fp16))[name = tensor("op_8545_cast_fp16")]; tensor var_8547_interleave_0 = const()[name = tensor("op_8547_interleave_0"), val = tensor(false)]; tensor var_8547_cast_fp16 = concat(axis = var_8160, interleave = var_8547_interleave_0, values = (var_8521_cast_fp16, var_8523_cast_fp16))[name = tensor("op_8547_cast_fp16")]; tensor var_8549_interleave_0 = const()[name = tensor("op_8549_interleave_0"), val = tensor(false)]; tensor var_8549_cast_fp16 = concat(axis = var_8160, interleave = var_8549_interleave_0, values = (var_8525_cast_fp16, var_8527_cast_fp16))[name = tensor("op_8549_cast_fp16")]; tensor var_8551_interleave_0 = const()[name = tensor("op_8551_interleave_0"), val = tensor(false)]; tensor var_8551_cast_fp16 = concat(axis = var_8160, interleave = var_8551_interleave_0, values = (var_8529_cast_fp16, var_8531_cast_fp16))[name = tensor("op_8551_cast_fp16")]; tensor var_8553_interleave_0 = const()[name = tensor("op_8553_interleave_0"), val = tensor(false)]; tensor var_8553_cast_fp16 = concat(axis = var_8160, interleave = var_8553_interleave_0, values = (var_8533_cast_fp16, var_8535_cast_fp16))[name = tensor("op_8553_cast_fp16")]; tensor var_8555_interleave_0 = const()[name = tensor("op_8555_interleave_0"), val = tensor(false)]; tensor var_8555_cast_fp16 = concat(axis = var_8160, interleave = var_8555_interleave_0, values = (var_8537_cast_fp16, var_8539_cast_fp16))[name = tensor("op_8555_cast_fp16")]; tensor input_383_interleave_0 = const()[name = tensor("input_383_interleave_0"), val = tensor(false)]; tensor input_383_cast_fp16 = concat(axis = var_8182, interleave = input_383_interleave_0, values = (var_8541_cast_fp16, var_8543_cast_fp16, var_8545_cast_fp16, var_8547_cast_fp16, var_8549_cast_fp16, var_8551_cast_fp16, var_8553_cast_fp16, var_8555_cast_fp16))[name = tensor("input_383_cast_fp16")]; tensor var_8565_pad_type_0 = const()[name = tensor("op_8565_pad_type_0"), val = tensor("valid")]; tensor var_8565_strides_0 = const()[name = tensor("op_8565_strides_0"), val = tensor([1, 1])]; tensor var_8565_pad_0 = const()[name = tensor("op_8565_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_8565_dilations_0 = const()[name = tensor("op_8565_dilations_0"), val = tensor([1, 1])]; tensor var_8565_groups_0 = const()[name = tensor("op_8565_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(687964416)))]; tensor up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(688783680)))]; tensor var_8565_cast_fp16 = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_8565_dilations_0, groups = var_8565_groups_0, pad = var_8565_pad_0, pad_type = var_8565_pad_type_0, strides = var_8565_strides_0, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_383_cast_fp16)[name = tensor("op_8565_cast_fp16")]; tensor inputs_63_cast_fp16 = add(x = var_8565_cast_fp16, y = inputs_61_cast_fp16)[name = tensor("inputs_63_cast_fp16")]; tensor hidden_states_229_axes_0 = const()[name = tensor("hidden_states_229_axes_0"), val = tensor([1])]; tensor hidden_states_229_gamma_0_to_fp16 = const()[name = tensor("hidden_states_229_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(688785024)))]; tensor hidden_states_229_beta_0_to_fp16 = const()[name = tensor("hidden_states_229_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(688786368)))]; tensor var_8575_to_fp16 = const()[name = tensor("op_8575_to_fp16"), val = tensor(0x1.5p-17)]; tensor hidden_states_229_cast_fp16 = layer_norm(axes = hidden_states_229_axes_0, beta = hidden_states_229_beta_0_to_fp16, epsilon = var_8575_to_fp16, gamma = hidden_states_229_gamma_0_to_fp16, x = inputs_63_cast_fp16)[name = tensor("hidden_states_229_cast_fp16")]; tensor q_43_pad_type_0 = const()[name = tensor("q_43_pad_type_0"), val = tensor("valid")]; tensor q_43_strides_0 = const()[name = tensor("q_43_strides_0"), val = tensor([1, 1])]; tensor q_43_pad_0 = const()[name = tensor("q_43_pad_0"), val = tensor([0, 0, 0, 0])]; tensor q_43_dilations_0 = const()[name = tensor("q_43_dilations_0"), val = tensor([1, 1])]; tensor q_43_groups_0 = const()[name = tensor("q_43_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(688787712)))]; tensor q_43_cast_fp16 = conv(dilations = q_43_dilations_0, groups = q_43_groups_0, pad = q_43_pad_0, pad_type = q_43_pad_type_0, strides = q_43_strides_0, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_229_cast_fp16)[name = tensor("q_43_cast_fp16")]; tensor k_85_pad_type_0 = const()[name = tensor("k_85_pad_type_0"), val = tensor("valid")]; tensor k_85_strides_0 = const()[name = tensor("k_85_strides_0"), val = tensor([1, 1])]; tensor k_85_pad_0 = const()[name = tensor("k_85_pad_0"), val = tensor([0, 0, 0, 0])]; tensor k_85_dilations_0 = const()[name = tensor("k_85_dilations_0"), val = tensor([1, 1])]; tensor k_85_groups_0 = const()[name = tensor("k_85_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(689606976)))]; tensor k_85_cast_fp16 = conv(dilations = k_85_dilations_0, groups = k_85_groups_0, pad = k_85_pad_0, pad_type = k_85_pad_type_0, strides = k_85_strides_0, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_85_cast_fp16")]; tensor v_43_pad_type_0 = const()[name = tensor("v_43_pad_type_0"), val = tensor("valid")]; tensor v_43_strides_0 = const()[name = tensor("v_43_strides_0"), val = tensor([1, 1])]; tensor v_43_pad_0 = const()[name = tensor("v_43_pad_0"), val = tensor([0, 0, 0, 0])]; tensor v_43_dilations_0 = const()[name = tensor("v_43_dilations_0"), val = tensor([1, 1])]; tensor v_43_groups_0 = const()[name = tensor("v_43_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(690590080)))]; tensor v_43_cast_fp16 = conv(dilations = v_43_dilations_0, groups = v_43_groups_0, pad = v_43_pad_0, pad_type = v_43_pad_type_0, strides = v_43_strides_0, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_43_cast_fp16")]; tensor var_8608_begin_0 = const()[name = tensor("op_8608_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8608_end_0 = const()[name = tensor("op_8608_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_8608_end_mask_0 = const()[name = tensor("op_8608_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8608_cast_fp16 = slice_by_index(begin = var_8608_begin_0, end = var_8608_end_0, end_mask = var_8608_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8608_cast_fp16")]; tensor var_8612_begin_0 = const()[name = tensor("op_8612_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_8612_end_0 = const()[name = tensor("op_8612_end_0"), val = tensor([2, 160, 1, 1024])]; tensor var_8612_end_mask_0 = const()[name = tensor("op_8612_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8612_cast_fp16 = slice_by_index(begin = var_8612_begin_0, end = var_8612_end_0, end_mask = var_8612_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8612_cast_fp16")]; tensor var_8616_begin_0 = const()[name = tensor("op_8616_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_8616_end_0 = const()[name = tensor("op_8616_end_0"), val = tensor([2, 240, 1, 1024])]; tensor var_8616_end_mask_0 = const()[name = tensor("op_8616_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8616_cast_fp16 = slice_by_index(begin = var_8616_begin_0, end = var_8616_end_0, end_mask = var_8616_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8616_cast_fp16")]; tensor var_8620_begin_0 = const()[name = tensor("op_8620_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_8620_end_0 = const()[name = tensor("op_8620_end_0"), val = tensor([2, 320, 1, 1024])]; tensor var_8620_end_mask_0 = const()[name = tensor("op_8620_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8620_cast_fp16 = slice_by_index(begin = var_8620_begin_0, end = var_8620_end_0, end_mask = var_8620_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8620_cast_fp16")]; tensor var_8624_begin_0 = const()[name = tensor("op_8624_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_8624_end_0 = const()[name = tensor("op_8624_end_0"), val = tensor([2, 400, 1, 1024])]; tensor var_8624_end_mask_0 = const()[name = tensor("op_8624_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8624_cast_fp16 = slice_by_index(begin = var_8624_begin_0, end = var_8624_end_0, end_mask = var_8624_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8624_cast_fp16")]; tensor var_8628_begin_0 = const()[name = tensor("op_8628_begin_0"), val = tensor([0, 400, 0, 0])]; tensor var_8628_end_0 = const()[name = tensor("op_8628_end_0"), val = tensor([2, 480, 1, 1024])]; tensor var_8628_end_mask_0 = const()[name = tensor("op_8628_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8628_cast_fp16 = slice_by_index(begin = var_8628_begin_0, end = var_8628_end_0, end_mask = var_8628_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8628_cast_fp16")]; tensor var_8632_begin_0 = const()[name = tensor("op_8632_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_8632_end_0 = const()[name = tensor("op_8632_end_0"), val = tensor([2, 560, 1, 1024])]; tensor var_8632_end_mask_0 = const()[name = tensor("op_8632_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8632_cast_fp16 = slice_by_index(begin = var_8632_begin_0, end = var_8632_end_0, end_mask = var_8632_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8632_cast_fp16")]; tensor var_8636_begin_0 = const()[name = tensor("op_8636_begin_0"), val = tensor([0, 560, 0, 0])]; tensor var_8636_end_0 = const()[name = tensor("op_8636_end_0"), val = tensor([2, 1, 1, 1024])]; tensor var_8636_end_mask_0 = const()[name = tensor("op_8636_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_8636_cast_fp16 = slice_by_index(begin = var_8636_begin_0, end = var_8636_end_0, end_mask = var_8636_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8636_cast_fp16")]; tensor var_8639_begin_0 = const()[name = tensor("op_8639_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8639_end_0 = const()[name = tensor("op_8639_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_8639_end_mask_0 = const()[name = tensor("op_8639_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8639_cast_fp16 = slice_by_index(begin = var_8639_begin_0, end = var_8639_end_0, end_mask = var_8639_end_mask_0, x = var_8608_cast_fp16)[name = tensor("op_8639_cast_fp16")]; tensor var_8640_begin_0 = const()[name = tensor("op_8640_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_8640_end_0 = const()[name = tensor("op_8640_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_8640_end_mask_0 = const()[name = tensor("op_8640_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_8640_cast_fp16 = slice_by_index(begin = var_8640_begin_0, end = var_8640_end_0, end_mask = var_8640_end_mask_0, x = var_8608_cast_fp16)[name = tensor("op_8640_cast_fp16")]; tensor var_8641_begin_0 = const()[name = tensor("op_8641_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8641_end_0 = const()[name = tensor("op_8641_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_8641_end_mask_0 = const()[name = tensor("op_8641_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8641_cast_fp16 = slice_by_index(begin = var_8641_begin_0, end = var_8641_end_0, end_mask = var_8641_end_mask_0, x = var_8612_cast_fp16)[name = tensor("op_8641_cast_fp16")]; tensor var_8642_begin_0 = const()[name = tensor("op_8642_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_8642_end_0 = const()[name = tensor("op_8642_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_8642_end_mask_0 = const()[name = tensor("op_8642_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_8642_cast_fp16 = slice_by_index(begin = var_8642_begin_0, end = var_8642_end_0, end_mask = var_8642_end_mask_0, x = var_8612_cast_fp16)[name = tensor("op_8642_cast_fp16")]; tensor var_8643_begin_0 = const()[name = tensor("op_8643_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8643_end_0 = const()[name = tensor("op_8643_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_8643_end_mask_0 = const()[name = tensor("op_8643_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8643_cast_fp16 = slice_by_index(begin = var_8643_begin_0, end = var_8643_end_0, end_mask = var_8643_end_mask_0, x = var_8616_cast_fp16)[name = tensor("op_8643_cast_fp16")]; tensor var_8644_begin_0 = const()[name = tensor("op_8644_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_8644_end_0 = const()[name = tensor("op_8644_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_8644_end_mask_0 = const()[name = tensor("op_8644_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_8644_cast_fp16 = slice_by_index(begin = var_8644_begin_0, end = var_8644_end_0, end_mask = var_8644_end_mask_0, x = var_8616_cast_fp16)[name = tensor("op_8644_cast_fp16")]; tensor var_8645_begin_0 = const()[name = tensor("op_8645_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8645_end_0 = const()[name = tensor("op_8645_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_8645_end_mask_0 = const()[name = tensor("op_8645_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8645_cast_fp16 = slice_by_index(begin = var_8645_begin_0, end = var_8645_end_0, end_mask = var_8645_end_mask_0, x = var_8620_cast_fp16)[name = tensor("op_8645_cast_fp16")]; tensor var_8646_begin_0 = const()[name = tensor("op_8646_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_8646_end_0 = const()[name = tensor("op_8646_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_8646_end_mask_0 = const()[name = tensor("op_8646_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_8646_cast_fp16 = slice_by_index(begin = var_8646_begin_0, end = var_8646_end_0, end_mask = var_8646_end_mask_0, x = var_8620_cast_fp16)[name = tensor("op_8646_cast_fp16")]; tensor var_8647_begin_0 = const()[name = tensor("op_8647_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8647_end_0 = const()[name = tensor("op_8647_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_8647_end_mask_0 = const()[name = tensor("op_8647_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8647_cast_fp16 = slice_by_index(begin = var_8647_begin_0, end = var_8647_end_0, end_mask = var_8647_end_mask_0, x = var_8624_cast_fp16)[name = tensor("op_8647_cast_fp16")]; tensor var_8648_begin_0 = const()[name = tensor("op_8648_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_8648_end_0 = const()[name = tensor("op_8648_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_8648_end_mask_0 = const()[name = tensor("op_8648_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_8648_cast_fp16 = slice_by_index(begin = var_8648_begin_0, end = var_8648_end_0, end_mask = var_8648_end_mask_0, x = var_8624_cast_fp16)[name = tensor("op_8648_cast_fp16")]; tensor var_8649_begin_0 = const()[name = tensor("op_8649_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8649_end_0 = const()[name = tensor("op_8649_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_8649_end_mask_0 = const()[name = tensor("op_8649_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8649_cast_fp16 = slice_by_index(begin = var_8649_begin_0, end = var_8649_end_0, end_mask = var_8649_end_mask_0, x = var_8628_cast_fp16)[name = tensor("op_8649_cast_fp16")]; tensor var_8650_begin_0 = const()[name = tensor("op_8650_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_8650_end_0 = const()[name = tensor("op_8650_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_8650_end_mask_0 = const()[name = tensor("op_8650_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_8650_cast_fp16 = slice_by_index(begin = var_8650_begin_0, end = var_8650_end_0, end_mask = var_8650_end_mask_0, x = var_8628_cast_fp16)[name = tensor("op_8650_cast_fp16")]; tensor var_8651_begin_0 = const()[name = tensor("op_8651_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8651_end_0 = const()[name = tensor("op_8651_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_8651_end_mask_0 = const()[name = tensor("op_8651_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8651_cast_fp16 = slice_by_index(begin = var_8651_begin_0, end = var_8651_end_0, end_mask = var_8651_end_mask_0, x = var_8632_cast_fp16)[name = tensor("op_8651_cast_fp16")]; tensor var_8652_begin_0 = const()[name = tensor("op_8652_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_8652_end_0 = const()[name = tensor("op_8652_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_8652_end_mask_0 = const()[name = tensor("op_8652_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_8652_cast_fp16 = slice_by_index(begin = var_8652_begin_0, end = var_8652_end_0, end_mask = var_8652_end_mask_0, x = var_8632_cast_fp16)[name = tensor("op_8652_cast_fp16")]; tensor var_8653_begin_0 = const()[name = tensor("op_8653_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8653_end_0 = const()[name = tensor("op_8653_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_8653_end_mask_0 = const()[name = tensor("op_8653_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8653_cast_fp16 = slice_by_index(begin = var_8653_begin_0, end = var_8653_end_0, end_mask = var_8653_end_mask_0, x = var_8636_cast_fp16)[name = tensor("op_8653_cast_fp16")]; tensor var_8654_begin_0 = const()[name = tensor("op_8654_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_8654_end_0 = const()[name = tensor("op_8654_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_8654_end_mask_0 = const()[name = tensor("op_8654_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_8654_cast_fp16 = slice_by_index(begin = var_8654_begin_0, end = var_8654_end_0, end_mask = var_8654_end_mask_0, x = var_8636_cast_fp16)[name = tensor("op_8654_cast_fp16")]; tensor k_87_perm_0 = const()[name = tensor("k_87_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_8659_begin_0 = const()[name = tensor("op_8659_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8659_end_0 = const()[name = tensor("op_8659_end_0"), val = tensor([2, 77, 1, 80])]; tensor var_8659_end_mask_0 = const()[name = tensor("op_8659_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_87_cast_fp16 = transpose(perm = k_87_perm_0, x = k_85_cast_fp16)[name = tensor("transpose_10")]; tensor var_8659_cast_fp16 = slice_by_index(begin = var_8659_begin_0, end = var_8659_end_0, end_mask = var_8659_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8659_cast_fp16")]; tensor var_8663_begin_0 = const()[name = tensor("op_8663_begin_0"), val = tensor([0, 0, 0, 80])]; tensor var_8663_end_0 = const()[name = tensor("op_8663_end_0"), val = tensor([2, 77, 1, 160])]; tensor var_8663_end_mask_0 = const()[name = tensor("op_8663_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8663_cast_fp16 = slice_by_index(begin = var_8663_begin_0, end = var_8663_end_0, end_mask = var_8663_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8663_cast_fp16")]; tensor var_8667_begin_0 = const()[name = tensor("op_8667_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_8667_end_0 = const()[name = tensor("op_8667_end_0"), val = tensor([2, 77, 1, 240])]; tensor var_8667_end_mask_0 = const()[name = tensor("op_8667_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8667_cast_fp16 = slice_by_index(begin = var_8667_begin_0, end = var_8667_end_0, end_mask = var_8667_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8667_cast_fp16")]; tensor var_8671_begin_0 = const()[name = tensor("op_8671_begin_0"), val = tensor([0, 0, 0, 240])]; tensor var_8671_end_0 = const()[name = tensor("op_8671_end_0"), val = tensor([2, 77, 1, 320])]; tensor var_8671_end_mask_0 = const()[name = tensor("op_8671_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8671_cast_fp16 = slice_by_index(begin = var_8671_begin_0, end = var_8671_end_0, end_mask = var_8671_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8671_cast_fp16")]; tensor var_8675_begin_0 = const()[name = tensor("op_8675_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_8675_end_0 = const()[name = tensor("op_8675_end_0"), val = tensor([2, 77, 1, 400])]; tensor var_8675_end_mask_0 = const()[name = tensor("op_8675_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8675_cast_fp16 = slice_by_index(begin = var_8675_begin_0, end = var_8675_end_0, end_mask = var_8675_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8675_cast_fp16")]; tensor var_8679_begin_0 = const()[name = tensor("op_8679_begin_0"), val = tensor([0, 0, 0, 400])]; tensor var_8679_end_0 = const()[name = tensor("op_8679_end_0"), val = tensor([2, 77, 1, 480])]; tensor var_8679_end_mask_0 = const()[name = tensor("op_8679_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8679_cast_fp16 = slice_by_index(begin = var_8679_begin_0, end = var_8679_end_0, end_mask = var_8679_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8679_cast_fp16")]; tensor var_8683_begin_0 = const()[name = tensor("op_8683_begin_0"), val = tensor([0, 0, 0, 480])]; tensor var_8683_end_0 = const()[name = tensor("op_8683_end_0"), val = tensor([2, 77, 1, 560])]; tensor var_8683_end_mask_0 = const()[name = tensor("op_8683_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8683_cast_fp16 = slice_by_index(begin = var_8683_begin_0, end = var_8683_end_0, end_mask = var_8683_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8683_cast_fp16")]; tensor var_8687_begin_0 = const()[name = tensor("op_8687_begin_0"), val = tensor([0, 0, 0, 560])]; tensor var_8687_end_0 = const()[name = tensor("op_8687_end_0"), val = tensor([2, 77, 1, 1])]; tensor var_8687_end_mask_0 = const()[name = tensor("op_8687_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_8687_cast_fp16 = slice_by_index(begin = var_8687_begin_0, end = var_8687_end_0, end_mask = var_8687_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8687_cast_fp16")]; tensor var_8689_begin_0 = const()[name = tensor("op_8689_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8689_end_0 = const()[name = tensor("op_8689_end_0"), val = tensor([2, 80, 1, 77])]; tensor var_8689_end_mask_0 = const()[name = tensor("op_8689_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8689_cast_fp16 = slice_by_index(begin = var_8689_begin_0, end = var_8689_end_0, end_mask = var_8689_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8689_cast_fp16")]; tensor var_8693_begin_0 = const()[name = tensor("op_8693_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_8693_end_0 = const()[name = tensor("op_8693_end_0"), val = tensor([2, 160, 1, 77])]; tensor var_8693_end_mask_0 = const()[name = tensor("op_8693_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8693_cast_fp16 = slice_by_index(begin = var_8693_begin_0, end = var_8693_end_0, end_mask = var_8693_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8693_cast_fp16")]; tensor var_8697_begin_0 = const()[name = tensor("op_8697_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_8697_end_0 = const()[name = tensor("op_8697_end_0"), val = tensor([2, 240, 1, 77])]; tensor var_8697_end_mask_0 = const()[name = tensor("op_8697_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8697_cast_fp16 = slice_by_index(begin = var_8697_begin_0, end = var_8697_end_0, end_mask = var_8697_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8697_cast_fp16")]; tensor var_8701_begin_0 = const()[name = tensor("op_8701_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_8701_end_0 = const()[name = tensor("op_8701_end_0"), val = tensor([2, 320, 1, 77])]; tensor var_8701_end_mask_0 = const()[name = tensor("op_8701_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8701_cast_fp16 = slice_by_index(begin = var_8701_begin_0, end = var_8701_end_0, end_mask = var_8701_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8701_cast_fp16")]; tensor var_8705_begin_0 = const()[name = tensor("op_8705_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_8705_end_0 = const()[name = tensor("op_8705_end_0"), val = tensor([2, 400, 1, 77])]; tensor var_8705_end_mask_0 = const()[name = tensor("op_8705_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8705_cast_fp16 = slice_by_index(begin = var_8705_begin_0, end = var_8705_end_0, end_mask = var_8705_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8705_cast_fp16")]; tensor var_8709_begin_0 = const()[name = tensor("op_8709_begin_0"), val = tensor([0, 400, 0, 0])]; tensor var_8709_end_0 = const()[name = tensor("op_8709_end_0"), val = tensor([2, 480, 1, 77])]; tensor var_8709_end_mask_0 = const()[name = tensor("op_8709_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8709_cast_fp16 = slice_by_index(begin = var_8709_begin_0, end = var_8709_end_0, end_mask = var_8709_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8709_cast_fp16")]; tensor var_8713_begin_0 = const()[name = tensor("op_8713_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_8713_end_0 = const()[name = tensor("op_8713_end_0"), val = tensor([2, 560, 1, 77])]; tensor var_8713_end_mask_0 = const()[name = tensor("op_8713_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8713_cast_fp16 = slice_by_index(begin = var_8713_begin_0, end = var_8713_end_0, end_mask = var_8713_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8713_cast_fp16")]; tensor var_8717_begin_0 = const()[name = tensor("op_8717_begin_0"), val = tensor([0, 560, 0, 0])]; tensor var_8717_end_0 = const()[name = tensor("op_8717_end_0"), val = tensor([2, 1, 1, 77])]; tensor var_8717_end_mask_0 = const()[name = tensor("op_8717_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_8717_cast_fp16 = slice_by_index(begin = var_8717_begin_0, end = var_8717_end_0, end_mask = var_8717_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8717_cast_fp16")]; tensor var_8721_equation_0 = const()[name = tensor("op_8721_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8721_cast_fp16 = einsum(equation = var_8721_equation_0, values = (var_8659_cast_fp16, var_8639_cast_fp16))[name = tensor("op_8721_cast_fp16")]; tensor var_8722_to_fp16 = const()[name = tensor("op_8722_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_673_cast_fp16 = mul(x = var_8721_cast_fp16, y = var_8722_to_fp16)[name = tensor("aw_chunk_673_cast_fp16")]; tensor var_8725_equation_0 = const()[name = tensor("op_8725_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8725_cast_fp16 = einsum(equation = var_8725_equation_0, values = (var_8659_cast_fp16, var_8640_cast_fp16))[name = tensor("op_8725_cast_fp16")]; tensor var_8726_to_fp16 = const()[name = tensor("op_8726_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_675_cast_fp16 = mul(x = var_8725_cast_fp16, y = var_8726_to_fp16)[name = tensor("aw_chunk_675_cast_fp16")]; tensor var_8729_equation_0 = const()[name = tensor("op_8729_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8729_cast_fp16 = einsum(equation = var_8729_equation_0, values = (var_8663_cast_fp16, var_8641_cast_fp16))[name = tensor("op_8729_cast_fp16")]; tensor var_8730_to_fp16 = const()[name = tensor("op_8730_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_677_cast_fp16 = mul(x = var_8729_cast_fp16, y = var_8730_to_fp16)[name = tensor("aw_chunk_677_cast_fp16")]; tensor var_8733_equation_0 = const()[name = tensor("op_8733_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8733_cast_fp16 = einsum(equation = var_8733_equation_0, values = (var_8663_cast_fp16, var_8642_cast_fp16))[name = tensor("op_8733_cast_fp16")]; tensor var_8734_to_fp16 = const()[name = tensor("op_8734_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_679_cast_fp16 = mul(x = var_8733_cast_fp16, y = var_8734_to_fp16)[name = tensor("aw_chunk_679_cast_fp16")]; tensor var_8737_equation_0 = const()[name = tensor("op_8737_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8737_cast_fp16 = einsum(equation = var_8737_equation_0, values = (var_8667_cast_fp16, var_8643_cast_fp16))[name = tensor("op_8737_cast_fp16")]; tensor var_8738_to_fp16 = const()[name = tensor("op_8738_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_681_cast_fp16 = mul(x = var_8737_cast_fp16, y = var_8738_to_fp16)[name = tensor("aw_chunk_681_cast_fp16")]; tensor var_8741_equation_0 = const()[name = tensor("op_8741_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8741_cast_fp16 = einsum(equation = var_8741_equation_0, values = (var_8667_cast_fp16, var_8644_cast_fp16))[name = tensor("op_8741_cast_fp16")]; tensor var_8742_to_fp16 = const()[name = tensor("op_8742_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_683_cast_fp16 = mul(x = var_8741_cast_fp16, y = var_8742_to_fp16)[name = tensor("aw_chunk_683_cast_fp16")]; tensor var_8745_equation_0 = const()[name = tensor("op_8745_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8745_cast_fp16 = einsum(equation = var_8745_equation_0, values = (var_8671_cast_fp16, var_8645_cast_fp16))[name = tensor("op_8745_cast_fp16")]; tensor var_8746_to_fp16 = const()[name = tensor("op_8746_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_685_cast_fp16 = mul(x = var_8745_cast_fp16, y = var_8746_to_fp16)[name = tensor("aw_chunk_685_cast_fp16")]; tensor var_8749_equation_0 = const()[name = tensor("op_8749_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8749_cast_fp16 = einsum(equation = var_8749_equation_0, values = (var_8671_cast_fp16, var_8646_cast_fp16))[name = tensor("op_8749_cast_fp16")]; tensor var_8750_to_fp16 = const()[name = tensor("op_8750_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_687_cast_fp16 = mul(x = var_8749_cast_fp16, y = var_8750_to_fp16)[name = tensor("aw_chunk_687_cast_fp16")]; tensor var_8753_equation_0 = const()[name = tensor("op_8753_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8753_cast_fp16 = einsum(equation = var_8753_equation_0, values = (var_8675_cast_fp16, var_8647_cast_fp16))[name = tensor("op_8753_cast_fp16")]; tensor var_8754_to_fp16 = const()[name = tensor("op_8754_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_689_cast_fp16 = mul(x = var_8753_cast_fp16, y = var_8754_to_fp16)[name = tensor("aw_chunk_689_cast_fp16")]; tensor var_8757_equation_0 = const()[name = tensor("op_8757_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8757_cast_fp16 = einsum(equation = var_8757_equation_0, values = (var_8675_cast_fp16, var_8648_cast_fp16))[name = tensor("op_8757_cast_fp16")]; tensor var_8758_to_fp16 = const()[name = tensor("op_8758_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_691_cast_fp16 = mul(x = var_8757_cast_fp16, y = var_8758_to_fp16)[name = tensor("aw_chunk_691_cast_fp16")]; tensor var_8761_equation_0 = const()[name = tensor("op_8761_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8761_cast_fp16 = einsum(equation = var_8761_equation_0, values = (var_8679_cast_fp16, var_8649_cast_fp16))[name = tensor("op_8761_cast_fp16")]; tensor var_8762_to_fp16 = const()[name = tensor("op_8762_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_693_cast_fp16 = mul(x = var_8761_cast_fp16, y = var_8762_to_fp16)[name = tensor("aw_chunk_693_cast_fp16")]; tensor var_8765_equation_0 = const()[name = tensor("op_8765_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8765_cast_fp16 = einsum(equation = var_8765_equation_0, values = (var_8679_cast_fp16, var_8650_cast_fp16))[name = tensor("op_8765_cast_fp16")]; tensor var_8766_to_fp16 = const()[name = tensor("op_8766_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_695_cast_fp16 = mul(x = var_8765_cast_fp16, y = var_8766_to_fp16)[name = tensor("aw_chunk_695_cast_fp16")]; tensor var_8769_equation_0 = const()[name = tensor("op_8769_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8769_cast_fp16 = einsum(equation = var_8769_equation_0, values = (var_8683_cast_fp16, var_8651_cast_fp16))[name = tensor("op_8769_cast_fp16")]; tensor var_8770_to_fp16 = const()[name = tensor("op_8770_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_697_cast_fp16 = mul(x = var_8769_cast_fp16, y = var_8770_to_fp16)[name = tensor("aw_chunk_697_cast_fp16")]; tensor var_8773_equation_0 = const()[name = tensor("op_8773_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8773_cast_fp16 = einsum(equation = var_8773_equation_0, values = (var_8683_cast_fp16, var_8652_cast_fp16))[name = tensor("op_8773_cast_fp16")]; tensor var_8774_to_fp16 = const()[name = tensor("op_8774_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_699_cast_fp16 = mul(x = var_8773_cast_fp16, y = var_8774_to_fp16)[name = tensor("aw_chunk_699_cast_fp16")]; tensor var_8777_equation_0 = const()[name = tensor("op_8777_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8777_cast_fp16 = einsum(equation = var_8777_equation_0, values = (var_8687_cast_fp16, var_8653_cast_fp16))[name = tensor("op_8777_cast_fp16")]; tensor var_8778_to_fp16 = const()[name = tensor("op_8778_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_701_cast_fp16 = mul(x = var_8777_cast_fp16, y = var_8778_to_fp16)[name = tensor("aw_chunk_701_cast_fp16")]; tensor var_8781_equation_0 = const()[name = tensor("op_8781_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8781_cast_fp16 = einsum(equation = var_8781_equation_0, values = (var_8687_cast_fp16, var_8654_cast_fp16))[name = tensor("op_8781_cast_fp16")]; tensor var_8782_to_fp16 = const()[name = tensor("op_8782_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_703_cast_fp16 = mul(x = var_8781_cast_fp16, y = var_8782_to_fp16)[name = tensor("aw_chunk_703_cast_fp16")]; tensor var_8784_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_673_cast_fp16)[name = tensor("op_8784_cast_fp16")]; tensor var_8785_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_675_cast_fp16)[name = tensor("op_8785_cast_fp16")]; tensor var_8786_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_677_cast_fp16)[name = tensor("op_8786_cast_fp16")]; tensor var_8787_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_679_cast_fp16)[name = tensor("op_8787_cast_fp16")]; tensor var_8788_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_681_cast_fp16)[name = tensor("op_8788_cast_fp16")]; tensor var_8789_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_683_cast_fp16)[name = tensor("op_8789_cast_fp16")]; tensor var_8790_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_685_cast_fp16)[name = tensor("op_8790_cast_fp16")]; tensor var_8791_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_687_cast_fp16)[name = tensor("op_8791_cast_fp16")]; tensor var_8792_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_689_cast_fp16)[name = tensor("op_8792_cast_fp16")]; tensor var_8793_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_691_cast_fp16)[name = tensor("op_8793_cast_fp16")]; tensor var_8794_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_693_cast_fp16)[name = tensor("op_8794_cast_fp16")]; tensor var_8795_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_695_cast_fp16)[name = tensor("op_8795_cast_fp16")]; tensor var_8796_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_697_cast_fp16)[name = tensor("op_8796_cast_fp16")]; tensor var_8797_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_699_cast_fp16)[name = tensor("op_8797_cast_fp16")]; tensor var_8798_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_701_cast_fp16)[name = tensor("op_8798_cast_fp16")]; tensor var_8799_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_703_cast_fp16)[name = tensor("op_8799_cast_fp16")]; tensor var_8801_equation_0 = const()[name = tensor("op_8801_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8801_cast_fp16 = einsum(equation = var_8801_equation_0, values = (var_8689_cast_fp16, var_8784_cast_fp16))[name = tensor("op_8801_cast_fp16")]; tensor var_8803_equation_0 = const()[name = tensor("op_8803_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8803_cast_fp16 = einsum(equation = var_8803_equation_0, values = (var_8689_cast_fp16, var_8785_cast_fp16))[name = tensor("op_8803_cast_fp16")]; tensor var_8805_equation_0 = const()[name = tensor("op_8805_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8805_cast_fp16 = einsum(equation = var_8805_equation_0, values = (var_8693_cast_fp16, var_8786_cast_fp16))[name = tensor("op_8805_cast_fp16")]; tensor var_8807_equation_0 = const()[name = tensor("op_8807_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8807_cast_fp16 = einsum(equation = var_8807_equation_0, values = (var_8693_cast_fp16, var_8787_cast_fp16))[name = tensor("op_8807_cast_fp16")]; tensor var_8809_equation_0 = const()[name = tensor("op_8809_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8809_cast_fp16 = einsum(equation = var_8809_equation_0, values = (var_8697_cast_fp16, var_8788_cast_fp16))[name = tensor("op_8809_cast_fp16")]; tensor var_8811_equation_0 = const()[name = tensor("op_8811_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8811_cast_fp16 = einsum(equation = var_8811_equation_0, values = (var_8697_cast_fp16, var_8789_cast_fp16))[name = tensor("op_8811_cast_fp16")]; tensor var_8813_equation_0 = const()[name = tensor("op_8813_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8813_cast_fp16 = einsum(equation = var_8813_equation_0, values = (var_8701_cast_fp16, var_8790_cast_fp16))[name = tensor("op_8813_cast_fp16")]; tensor var_8815_equation_0 = const()[name = tensor("op_8815_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8815_cast_fp16 = einsum(equation = var_8815_equation_0, values = (var_8701_cast_fp16, var_8791_cast_fp16))[name = tensor("op_8815_cast_fp16")]; tensor var_8817_equation_0 = const()[name = tensor("op_8817_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8817_cast_fp16 = einsum(equation = var_8817_equation_0, values = (var_8705_cast_fp16, var_8792_cast_fp16))[name = tensor("op_8817_cast_fp16")]; tensor var_8819_equation_0 = const()[name = tensor("op_8819_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8819_cast_fp16 = einsum(equation = var_8819_equation_0, values = (var_8705_cast_fp16, var_8793_cast_fp16))[name = tensor("op_8819_cast_fp16")]; tensor var_8821_equation_0 = const()[name = tensor("op_8821_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8821_cast_fp16 = einsum(equation = var_8821_equation_0, values = (var_8709_cast_fp16, var_8794_cast_fp16))[name = tensor("op_8821_cast_fp16")]; tensor var_8823_equation_0 = const()[name = tensor("op_8823_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8823_cast_fp16 = einsum(equation = var_8823_equation_0, values = (var_8709_cast_fp16, var_8795_cast_fp16))[name = tensor("op_8823_cast_fp16")]; tensor var_8825_equation_0 = const()[name = tensor("op_8825_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8825_cast_fp16 = einsum(equation = var_8825_equation_0, values = (var_8713_cast_fp16, var_8796_cast_fp16))[name = tensor("op_8825_cast_fp16")]; tensor var_8827_equation_0 = const()[name = tensor("op_8827_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8827_cast_fp16 = einsum(equation = var_8827_equation_0, values = (var_8713_cast_fp16, var_8797_cast_fp16))[name = tensor("op_8827_cast_fp16")]; tensor var_8829_equation_0 = const()[name = tensor("op_8829_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8829_cast_fp16 = einsum(equation = var_8829_equation_0, values = (var_8717_cast_fp16, var_8798_cast_fp16))[name = tensor("op_8829_cast_fp16")]; tensor var_8831_equation_0 = const()[name = tensor("op_8831_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8831_cast_fp16 = einsum(equation = var_8831_equation_0, values = (var_8717_cast_fp16, var_8799_cast_fp16))[name = tensor("op_8831_cast_fp16")]; tensor var_8833_interleave_0 = const()[name = tensor("op_8833_interleave_0"), val = tensor(false)]; tensor var_8833_cast_fp16 = concat(axis = var_8160, interleave = var_8833_interleave_0, values = (var_8801_cast_fp16, var_8803_cast_fp16))[name = tensor("op_8833_cast_fp16")]; tensor var_8835_interleave_0 = const()[name = tensor("op_8835_interleave_0"), val = tensor(false)]; tensor var_8835_cast_fp16 = concat(axis = var_8160, interleave = var_8835_interleave_0, values = (var_8805_cast_fp16, var_8807_cast_fp16))[name = tensor("op_8835_cast_fp16")]; tensor var_8837_interleave_0 = const()[name = tensor("op_8837_interleave_0"), val = tensor(false)]; tensor var_8837_cast_fp16 = concat(axis = var_8160, interleave = var_8837_interleave_0, values = (var_8809_cast_fp16, var_8811_cast_fp16))[name = tensor("op_8837_cast_fp16")]; tensor var_8839_interleave_0 = const()[name = tensor("op_8839_interleave_0"), val = tensor(false)]; tensor var_8839_cast_fp16 = concat(axis = var_8160, interleave = var_8839_interleave_0, values = (var_8813_cast_fp16, var_8815_cast_fp16))[name = tensor("op_8839_cast_fp16")]; tensor var_8841_interleave_0 = const()[name = tensor("op_8841_interleave_0"), val = tensor(false)]; tensor var_8841_cast_fp16 = concat(axis = var_8160, interleave = var_8841_interleave_0, values = (var_8817_cast_fp16, var_8819_cast_fp16))[name = tensor("op_8841_cast_fp16")]; tensor var_8843_interleave_0 = const()[name = tensor("op_8843_interleave_0"), val = tensor(false)]; tensor var_8843_cast_fp16 = concat(axis = var_8160, interleave = var_8843_interleave_0, values = (var_8821_cast_fp16, var_8823_cast_fp16))[name = tensor("op_8843_cast_fp16")]; tensor var_8845_interleave_0 = const()[name = tensor("op_8845_interleave_0"), val = tensor(false)]; tensor var_8845_cast_fp16 = concat(axis = var_8160, interleave = var_8845_interleave_0, values = (var_8825_cast_fp16, var_8827_cast_fp16))[name = tensor("op_8845_cast_fp16")]; tensor var_8847_interleave_0 = const()[name = tensor("op_8847_interleave_0"), val = tensor(false)]; tensor var_8847_cast_fp16 = concat(axis = var_8160, interleave = var_8847_interleave_0, values = (var_8829_cast_fp16, var_8831_cast_fp16))[name = tensor("op_8847_cast_fp16")]; tensor input_385_interleave_0 = const()[name = tensor("input_385_interleave_0"), val = tensor(false)]; tensor input_385_cast_fp16 = concat(axis = var_8182, interleave = input_385_interleave_0, values = (var_8833_cast_fp16, var_8835_cast_fp16, var_8837_cast_fp16, var_8839_cast_fp16, var_8841_cast_fp16, var_8843_cast_fp16, var_8845_cast_fp16, var_8847_cast_fp16))[name = tensor("input_385_cast_fp16")]; tensor var_8857_pad_type_0 = const()[name = tensor("op_8857_pad_type_0"), val = tensor("valid")]; tensor var_8857_strides_0 = const()[name = tensor("op_8857_strides_0"), val = tensor([1, 1])]; tensor var_8857_pad_0 = const()[name = tensor("op_8857_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_8857_dilations_0 = const()[name = tensor("op_8857_dilations_0"), val = tensor([1, 1])]; tensor var_8857_groups_0 = const()[name = tensor("op_8857_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(691573184)))]; tensor up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(692392448)))]; tensor var_8857_cast_fp16 = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_8857_dilations_0, groups = var_8857_groups_0, pad = var_8857_pad_0, pad_type = var_8857_pad_type_0, strides = var_8857_strides_0, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_385_cast_fp16)[name = tensor("op_8857_cast_fp16")]; tensor inputs_65_cast_fp16 = add(x = var_8857_cast_fp16, y = inputs_63_cast_fp16)[name = tensor("inputs_65_cast_fp16")]; tensor input_387_axes_0 = const()[name = tensor("input_387_axes_0"), val = tensor([1])]; tensor input_387_gamma_0_to_fp16 = const()[name = tensor("input_387_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(692393792)))]; tensor input_387_beta_0_to_fp16 = const()[name = tensor("input_387_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(692395136)))]; tensor var_8867_to_fp16 = const()[name = tensor("op_8867_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_387_cast_fp16 = layer_norm(axes = input_387_axes_0, beta = input_387_beta_0_to_fp16, epsilon = var_8867_to_fp16, gamma = input_387_gamma_0_to_fp16, x = inputs_65_cast_fp16)[name = tensor("input_387_cast_fp16")]; tensor var_8887_pad_type_0 = const()[name = tensor("op_8887_pad_type_0"), val = tensor("valid")]; tensor var_8887_strides_0 = const()[name = tensor("op_8887_strides_0"), val = tensor([1, 1])]; tensor var_8887_pad_0 = const()[name = tensor("op_8887_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_8887_dilations_0 = const()[name = tensor("op_8887_dilations_0"), val = tensor([1, 1])]; tensor var_8887_groups_0 = const()[name = tensor("op_8887_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(692396480)))]; tensor up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(698950144)))]; tensor var_8887_cast_fp16 = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_8887_dilations_0, groups = var_8887_groups_0, pad = var_8887_pad_0, pad_type = var_8887_pad_type_0, strides = var_8887_strides_0, weight = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_387_cast_fp16)[name = tensor("op_8887_cast_fp16")]; tensor var_8888_split_sizes_0 = const()[name = tensor("op_8888_split_sizes_0"), val = tensor([2560, 2560])]; tensor var_8888_axis_0 = const()[name = tensor("op_8888_axis_0"), val = tensor(1)]; tensor var_8888_cast_fp16_0, tensor var_8888_cast_fp16_1 = split(axis = var_8888_axis_0, split_sizes = var_8888_split_sizes_0, x = var_8887_cast_fp16)[name = tensor("op_8888_cast_fp16")]; tensor var_8890_mode_0 = const()[name = tensor("op_8890_mode_0"), val = tensor("EXACT")]; tensor var_8890_cast_fp16 = gelu(mode = var_8890_mode_0, x = var_8888_cast_fp16_1)[name = tensor("op_8890_cast_fp16")]; tensor input_389_cast_fp16 = mul(x = var_8888_cast_fp16_0, y = var_8890_cast_fp16)[name = tensor("input_389_cast_fp16")]; tensor var_8898_pad_type_0 = const()[name = tensor("op_8898_pad_type_0"), val = tensor("valid")]; tensor var_8898_strides_0 = const()[name = tensor("op_8898_strides_0"), val = tensor([1, 1])]; tensor var_8898_pad_0 = const()[name = tensor("op_8898_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_8898_dilations_0 = const()[name = tensor("op_8898_dilations_0"), val = tensor([1, 1])]; tensor var_8898_groups_0 = const()[name = tensor("op_8898_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(698960448)))]; tensor up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(702237312)))]; tensor var_8898_cast_fp16 = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_8898_dilations_0, groups = var_8898_groups_0, pad = var_8898_pad_0, pad_type = var_8898_pad_type_0, strides = var_8898_strides_0, weight = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_389_cast_fp16)[name = tensor("op_8898_cast_fp16")]; tensor hidden_states_233_cast_fp16 = add(x = var_8898_cast_fp16, y = inputs_65_cast_fp16)[name = tensor("hidden_states_233_cast_fp16")]; tensor var_8900 = const()[name = tensor("op_8900"), val = tensor([2, 640, 32, 32])]; tensor input_391_cast_fp16 = reshape(shape = var_8900, x = hidden_states_233_cast_fp16)[name = tensor("input_391_cast_fp16")]; tensor hidden_states_235_pad_type_0 = const()[name = tensor("hidden_states_235_pad_type_0"), val = tensor("valid")]; tensor hidden_states_235_strides_0 = const()[name = tensor("hidden_states_235_strides_0"), val = tensor([1, 1])]; tensor hidden_states_235_pad_0 = const()[name = tensor("hidden_states_235_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_235_dilations_0 = const()[name = tensor("hidden_states_235_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_235_groups_0 = const()[name = tensor("hidden_states_235_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(702238656)))]; tensor up_blocks_2_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(703057920)))]; tensor hidden_states_235_cast_fp16 = conv(bias = up_blocks_2_attentions_0_proj_out_bias_to_fp16, dilations = hidden_states_235_dilations_0, groups = hidden_states_235_groups_0, pad = hidden_states_235_pad_0, pad_type = hidden_states_235_pad_type_0, strides = hidden_states_235_strides_0, weight = up_blocks_2_attentions_0_proj_out_weight_to_fp16, x = input_391_cast_fp16)[name = tensor("hidden_states_235_cast_fp16")]; tensor hidden_states_237_cast_fp16 = add(x = hidden_states_235_cast_fp16, y = hidden_states_223_cast_fp16)[name = tensor("hidden_states_237_cast_fp16")]; tensor input_393_interleave_0 = const()[name = tensor("input_393_interleave_0"), val = tensor(false)]; tensor cast_5 = cast(dtype = cast_5_dtype_0, x = input_89_cast_fp16)[name = tensor("cast_8")]; tensor input_393_cast_fp16 = concat(axis = var_8182, interleave = input_393_interleave_0, values = (hidden_states_237_cast_fp16, cast_5))[name = tensor("input_393_cast_fp16")]; tensor reshape_180_shape_0 = const()[name = tensor("reshape_180_shape_0"), val = tensor([2, 32, 40, 32, 32])]; tensor reshape_180_cast_fp16 = reshape(shape = reshape_180_shape_0, x = input_393_cast_fp16)[name = tensor("reshape_180_cast_fp16")]; tensor reduce_mean_135_axes_0 = const()[name = tensor("reduce_mean_135_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_135_keep_dims_0 = const()[name = tensor("reduce_mean_135_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_135_cast_fp16 = reduce_mean(axes = reduce_mean_135_axes_0, keep_dims = reduce_mean_135_keep_dims_0, x = reshape_180_cast_fp16)[name = tensor("reduce_mean_135_cast_fp16")]; tensor sub_90_cast_fp16 = sub(x = reshape_180_cast_fp16, y = reduce_mean_135_cast_fp16)[name = tensor("sub_90_cast_fp16")]; tensor square_45_cast_fp16 = square(x = sub_90_cast_fp16)[name = tensor("square_45_cast_fp16")]; tensor reduce_mean_137_axes_0 = const()[name = tensor("reduce_mean_137_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_137_keep_dims_0 = const()[name = tensor("reduce_mean_137_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_137_cast_fp16 = reduce_mean(axes = reduce_mean_137_axes_0, keep_dims = reduce_mean_137_keep_dims_0, x = square_45_cast_fp16)[name = tensor("reduce_mean_137_cast_fp16")]; tensor add_90_y_0_to_fp16 = const()[name = tensor("add_90_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_90_cast_fp16 = add(x = reduce_mean_137_cast_fp16, y = add_90_y_0_to_fp16)[name = tensor("add_90_cast_fp16")]; tensor sqrt_45_cast_fp16 = sqrt(x = add_90_cast_fp16)[name = tensor("sqrt_45_cast_fp16")]; tensor real_div_45_cast_fp16 = real_div(x = sub_90_cast_fp16, y = sqrt_45_cast_fp16)[name = tensor("real_div_45_cast_fp16")]; tensor reshape_181_shape_0 = const()[name = tensor("reshape_181_shape_0"), val = tensor([2, 1280, 32, 32])]; tensor reshape_181_cast_fp16 = reshape(shape = reshape_181_shape_0, x = real_div_45_cast_fp16)[name = tensor("reshape_181_cast_fp16")]; tensor add_91_gamma_0_to_fp16 = const()[name = tensor("add_91_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(703059264)))]; tensor add_91_beta_0_to_fp16 = const()[name = tensor("add_91_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(703061888)))]; tensor add_91_epsilon_0_to_fp16 = const()[name = tensor("add_91_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_91_cast_fp16 = batch_norm(beta = add_91_beta_0_to_fp16, epsilon = add_91_epsilon_0_to_fp16, gamma = add_91_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_181_cast_fp16)[name = tensor("add_91_cast_fp16")]; tensor input_397_cast_fp16 = silu(x = add_91_cast_fp16)[name = tensor("input_397_cast_fp16")]; tensor hidden_states_239_pad_type_0 = const()[name = tensor("hidden_states_239_pad_type_0"), val = tensor("custom")]; tensor hidden_states_239_pad_0 = const()[name = tensor("hidden_states_239_pad_0"), val = tensor([1, 1, 1, 1])]; tensor hidden_states_239_strides_0 = const()[name = tensor("hidden_states_239_strides_0"), val = tensor([1, 1])]; tensor hidden_states_239_dilations_0 = const()[name = tensor("hidden_states_239_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_239_groups_0 = const()[name = tensor("hidden_states_239_groups_0"), val = tensor(1)]; tensor up_blocks_2_resnets_1_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(703064512)))]; tensor up_blocks_2_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(717810176)))]; tensor hidden_states_239_cast_fp16 = conv(bias = up_blocks_2_resnets_1_conv1_bias_to_fp16, dilations = hidden_states_239_dilations_0, groups = hidden_states_239_groups_0, pad = hidden_states_239_pad_0, pad_type = hidden_states_239_pad_type_0, strides = hidden_states_239_strides_0, weight = up_blocks_2_resnets_1_conv1_weight_to_fp16, x = input_397_cast_fp16)[name = tensor("hidden_states_239_cast_fp16")]; tensor temb_35_pad_type_0 = const()[name = tensor("temb_35_pad_type_0"), val = tensor("valid")]; tensor temb_35_strides_0 = const()[name = tensor("temb_35_strides_0"), val = tensor([1, 1])]; tensor temb_35_pad_0 = const()[name = tensor("temb_35_pad_0"), val = tensor([0, 0, 0, 0])]; tensor temb_35_dilations_0 = const()[name = tensor("temb_35_dilations_0"), val = tensor([1, 1])]; tensor temb_35_groups_0 = const()[name = tensor("temb_35_groups_0"), val = tensor(1)]; tensor up_blocks_2_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(717811520)))]; tensor up_blocks_2_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(719449984)))]; tensor temb_35_cast_fp16 = conv(bias = up_blocks_2_resnets_1_time_emb_proj_bias_to_fp16, dilations = temb_35_dilations_0, groups = temb_35_groups_0, pad = temb_35_pad_0, pad_type = temb_35_pad_type_0, strides = temb_35_strides_0, weight = up_blocks_2_resnets_1_time_emb_proj_weight_to_fp16, x = cast_9)[name = tensor("temb_35_cast_fp16")]; tensor input_401_cast_fp16 = add(x = hidden_states_239_cast_fp16, y = temb_35_cast_fp16)[name = tensor("input_401_cast_fp16")]; tensor reshape_184_shape_0 = const()[name = tensor("reshape_184_shape_0"), val = tensor([2, 32, 20, 32, 32])]; tensor reshape_184_cast_fp16 = reshape(shape = reshape_184_shape_0, x = input_401_cast_fp16)[name = tensor("reshape_184_cast_fp16")]; tensor reduce_mean_138_axes_0 = const()[name = tensor("reduce_mean_138_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_138_keep_dims_0 = const()[name = tensor("reduce_mean_138_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_138_cast_fp16 = reduce_mean(axes = reduce_mean_138_axes_0, keep_dims = reduce_mean_138_keep_dims_0, x = reshape_184_cast_fp16)[name = tensor("reduce_mean_138_cast_fp16")]; tensor sub_92_cast_fp16 = sub(x = reshape_184_cast_fp16, y = reduce_mean_138_cast_fp16)[name = tensor("sub_92_cast_fp16")]; tensor square_46_cast_fp16 = square(x = sub_92_cast_fp16)[name = tensor("square_46_cast_fp16")]; tensor reduce_mean_140_axes_0 = const()[name = tensor("reduce_mean_140_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_140_keep_dims_0 = const()[name = tensor("reduce_mean_140_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_140_cast_fp16 = reduce_mean(axes = reduce_mean_140_axes_0, keep_dims = reduce_mean_140_keep_dims_0, x = square_46_cast_fp16)[name = tensor("reduce_mean_140_cast_fp16")]; tensor add_92_y_0_to_fp16 = const()[name = tensor("add_92_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_92_cast_fp16 = add(x = reduce_mean_140_cast_fp16, y = add_92_y_0_to_fp16)[name = tensor("add_92_cast_fp16")]; tensor sqrt_46_cast_fp16 = sqrt(x = add_92_cast_fp16)[name = tensor("sqrt_46_cast_fp16")]; tensor real_div_46_cast_fp16 = real_div(x = sub_92_cast_fp16, y = sqrt_46_cast_fp16)[name = tensor("real_div_46_cast_fp16")]; tensor reshape_185_shape_0 = const()[name = tensor("reshape_185_shape_0"), val = tensor([2, 640, 32, 32])]; tensor reshape_185_cast_fp16 = reshape(shape = reshape_185_shape_0, x = real_div_46_cast_fp16)[name = tensor("reshape_185_cast_fp16")]; tensor add_93_gamma_0_to_fp16 = const()[name = tensor("add_93_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(719451328)))]; tensor add_93_beta_0_to_fp16 = const()[name = tensor("add_93_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(719452672)))]; tensor add_93_epsilon_0_to_fp16 = const()[name = tensor("add_93_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_93_cast_fp16 = batch_norm(beta = add_93_beta_0_to_fp16, epsilon = add_93_epsilon_0_to_fp16, gamma = add_93_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_185_cast_fp16)[name = tensor("add_93_cast_fp16")]; tensor input_405_cast_fp16 = silu(x = add_93_cast_fp16)[name = tensor("input_405_cast_fp16")]; tensor hidden_states_241_pad_type_0 = const()[name = tensor("hidden_states_241_pad_type_0"), val = tensor("custom")]; tensor hidden_states_241_pad_0 = const()[name = tensor("hidden_states_241_pad_0"), val = tensor([1, 1, 1, 1])]; tensor hidden_states_241_strides_0 = const()[name = tensor("hidden_states_241_strides_0"), val = tensor([1, 1])]; tensor hidden_states_241_dilations_0 = const()[name = tensor("hidden_states_241_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_241_groups_0 = const()[name = tensor("hidden_states_241_groups_0"), val = tensor(1)]; tensor up_blocks_2_resnets_1_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(719454016)))]; tensor up_blocks_2_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(726826880)))]; tensor hidden_states_241_cast_fp16 = conv(bias = up_blocks_2_resnets_1_conv2_bias_to_fp16, dilations = hidden_states_241_dilations_0, groups = hidden_states_241_groups_0, pad = hidden_states_241_pad_0, pad_type = hidden_states_241_pad_type_0, strides = hidden_states_241_strides_0, weight = up_blocks_2_resnets_1_conv2_weight_to_fp16, x = input_405_cast_fp16)[name = tensor("hidden_states_241_cast_fp16")]; tensor x_19_pad_type_0 = const()[name = tensor("x_19_pad_type_0"), val = tensor("valid")]; tensor x_19_strides_0 = const()[name = tensor("x_19_strides_0"), val = tensor([1, 1])]; tensor x_19_pad_0 = const()[name = tensor("x_19_pad_0"), val = tensor([0, 0, 0, 0])]; tensor x_19_dilations_0 = const()[name = tensor("x_19_dilations_0"), val = tensor([1, 1])]; tensor x_19_groups_0 = const()[name = tensor("x_19_groups_0"), val = tensor(1)]; tensor up_blocks_2_resnets_1_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(726828224)))]; tensor up_blocks_2_resnets_1_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(728466688)))]; tensor x_19_cast_fp16 = conv(bias = up_blocks_2_resnets_1_conv_shortcut_bias_to_fp16, dilations = x_19_dilations_0, groups = x_19_groups_0, pad = x_19_pad_0, pad_type = x_19_pad_type_0, strides = x_19_strides_0, weight = up_blocks_2_resnets_1_conv_shortcut_weight_to_fp16, x = input_393_cast_fp16)[name = tensor("x_19_cast_fp16")]; tensor hidden_states_243_cast_fp16 = add(x = x_19_cast_fp16, y = hidden_states_241_cast_fp16)[name = tensor("hidden_states_243_cast_fp16")]; tensor reshape_188_shape_0 = const()[name = tensor("reshape_188_shape_0"), val = tensor([2, 32, 20, 32, 32])]; tensor reshape_188_cast_fp16 = reshape(shape = reshape_188_shape_0, x = hidden_states_243_cast_fp16)[name = tensor("reshape_188_cast_fp16")]; tensor reduce_mean_141_axes_0 = const()[name = tensor("reduce_mean_141_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_141_keep_dims_0 = const()[name = tensor("reduce_mean_141_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_141_cast_fp16 = reduce_mean(axes = reduce_mean_141_axes_0, keep_dims = reduce_mean_141_keep_dims_0, x = reshape_188_cast_fp16)[name = tensor("reduce_mean_141_cast_fp16")]; tensor sub_94_cast_fp16 = sub(x = reshape_188_cast_fp16, y = reduce_mean_141_cast_fp16)[name = tensor("sub_94_cast_fp16")]; tensor square_47_cast_fp16 = square(x = sub_94_cast_fp16)[name = tensor("square_47_cast_fp16")]; tensor reduce_mean_143_axes_0 = const()[name = tensor("reduce_mean_143_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_143_keep_dims_0 = const()[name = tensor("reduce_mean_143_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_143_cast_fp16 = reduce_mean(axes = reduce_mean_143_axes_0, keep_dims = reduce_mean_143_keep_dims_0, x = square_47_cast_fp16)[name = tensor("reduce_mean_143_cast_fp16")]; tensor add_94_y_0_to_fp16 = const()[name = tensor("add_94_y_0_to_fp16"), val = tensor(0x1.1p-20)]; tensor add_94_cast_fp16 = add(x = reduce_mean_143_cast_fp16, y = add_94_y_0_to_fp16)[name = tensor("add_94_cast_fp16")]; tensor sqrt_47_cast_fp16 = sqrt(x = add_94_cast_fp16)[name = tensor("sqrt_47_cast_fp16")]; tensor real_div_47_cast_fp16 = real_div(x = sub_94_cast_fp16, y = sqrt_47_cast_fp16)[name = tensor("real_div_47_cast_fp16")]; tensor reshape_189_shape_0 = const()[name = tensor("reshape_189_shape_0"), val = tensor([2, 640, 32, 32])]; tensor reshape_189_cast_fp16 = reshape(shape = reshape_189_shape_0, x = real_div_47_cast_fp16)[name = tensor("reshape_189_cast_fp16")]; tensor add_95_gamma_0_to_fp16 = const()[name = tensor("add_95_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(728468032)))]; tensor add_95_beta_0_to_fp16 = const()[name = tensor("add_95_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(728469376)))]; tensor add_95_epsilon_0_to_fp16 = const()[name = tensor("add_95_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_95_cast_fp16 = batch_norm(beta = add_95_beta_0_to_fp16, epsilon = add_95_epsilon_0_to_fp16, gamma = add_95_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_189_cast_fp16)[name = tensor("add_95_cast_fp16")]; tensor hidden_states_245_pad_type_0 = const()[name = tensor("hidden_states_245_pad_type_0"), val = tensor("valid")]; tensor hidden_states_245_strides_0 = const()[name = tensor("hidden_states_245_strides_0"), val = tensor([1, 1])]; tensor hidden_states_245_pad_0 = const()[name = tensor("hidden_states_245_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_245_dilations_0 = const()[name = tensor("hidden_states_245_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_245_groups_0 = const()[name = tensor("hidden_states_245_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_1_proj_in_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(728470720)))]; tensor up_blocks_2_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(729289984)))]; tensor hidden_states_245_cast_fp16 = conv(bias = up_blocks_2_attentions_1_proj_in_bias_to_fp16, dilations = hidden_states_245_dilations_0, groups = hidden_states_245_groups_0, pad = hidden_states_245_pad_0, pad_type = hidden_states_245_pad_type_0, strides = hidden_states_245_strides_0, weight = up_blocks_2_attentions_1_proj_in_weight_to_fp16, x = add_95_cast_fp16)[name = tensor("hidden_states_245_cast_fp16")]; tensor var_8980 = const()[name = tensor("op_8980"), val = tensor([2, 640, 1, 1024])]; tensor inputs_67_cast_fp16 = reshape(shape = var_8980, x = hidden_states_245_cast_fp16)[name = tensor("inputs_67_cast_fp16")]; tensor hidden_states_247_axes_0 = const()[name = tensor("hidden_states_247_axes_0"), val = tensor([1])]; tensor hidden_states_247_gamma_0_to_fp16 = const()[name = tensor("hidden_states_247_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(729291328)))]; tensor hidden_states_247_beta_0_to_fp16 = const()[name = tensor("hidden_states_247_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(729292672)))]; tensor var_8996_to_fp16 = const()[name = tensor("op_8996_to_fp16"), val = tensor(0x1.5p-17)]; tensor hidden_states_247_cast_fp16 = layer_norm(axes = hidden_states_247_axes_0, beta = hidden_states_247_beta_0_to_fp16, epsilon = var_8996_to_fp16, gamma = hidden_states_247_gamma_0_to_fp16, x = inputs_67_cast_fp16)[name = tensor("hidden_states_247_cast_fp16")]; tensor q_45_pad_type_0 = const()[name = tensor("q_45_pad_type_0"), val = tensor("valid")]; tensor q_45_strides_0 = const()[name = tensor("q_45_strides_0"), val = tensor([1, 1])]; tensor q_45_pad_0 = const()[name = tensor("q_45_pad_0"), val = tensor([0, 0, 0, 0])]; tensor q_45_dilations_0 = const()[name = tensor("q_45_dilations_0"), val = tensor([1, 1])]; tensor q_45_groups_0 = const()[name = tensor("q_45_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(729294016)))]; tensor q_45_cast_fp16 = conv(dilations = q_45_dilations_0, groups = q_45_groups_0, pad = q_45_pad_0, pad_type = q_45_pad_type_0, strides = q_45_strides_0, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_247_cast_fp16)[name = tensor("q_45_cast_fp16")]; tensor k_89_pad_type_0 = const()[name = tensor("k_89_pad_type_0"), val = tensor("valid")]; tensor k_89_strides_0 = const()[name = tensor("k_89_strides_0"), val = tensor([1, 1])]; tensor k_89_pad_0 = const()[name = tensor("k_89_pad_0"), val = tensor([0, 0, 0, 0])]; tensor k_89_dilations_0 = const()[name = tensor("k_89_dilations_0"), val = tensor([1, 1])]; tensor k_89_groups_0 = const()[name = tensor("k_89_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(730113280)))]; tensor k_89_cast_fp16 = conv(dilations = k_89_dilations_0, groups = k_89_groups_0, pad = k_89_pad_0, pad_type = k_89_pad_type_0, strides = k_89_strides_0, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_247_cast_fp16)[name = tensor("k_89_cast_fp16")]; tensor v_45_pad_type_0 = const()[name = tensor("v_45_pad_type_0"), val = tensor("valid")]; tensor v_45_strides_0 = const()[name = tensor("v_45_strides_0"), val = tensor([1, 1])]; tensor v_45_pad_0 = const()[name = tensor("v_45_pad_0"), val = tensor([0, 0, 0, 0])]; tensor v_45_dilations_0 = const()[name = tensor("v_45_dilations_0"), val = tensor([1, 1])]; tensor v_45_groups_0 = const()[name = tensor("v_45_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(730932544)))]; tensor v_45_cast_fp16 = conv(dilations = v_45_dilations_0, groups = v_45_groups_0, pad = v_45_pad_0, pad_type = v_45_pad_type_0, strides = v_45_strides_0, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_247_cast_fp16)[name = tensor("v_45_cast_fp16")]; tensor var_9029_begin_0 = const()[name = tensor("op_9029_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9029_end_0 = const()[name = tensor("op_9029_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_9029_end_mask_0 = const()[name = tensor("op_9029_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9029_cast_fp16 = slice_by_index(begin = var_9029_begin_0, end = var_9029_end_0, end_mask = var_9029_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9029_cast_fp16")]; tensor var_9033_begin_0 = const()[name = tensor("op_9033_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_9033_end_0 = const()[name = tensor("op_9033_end_0"), val = tensor([2, 160, 1, 1024])]; tensor var_9033_end_mask_0 = const()[name = tensor("op_9033_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9033_cast_fp16 = slice_by_index(begin = var_9033_begin_0, end = var_9033_end_0, end_mask = var_9033_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9033_cast_fp16")]; tensor var_9037_begin_0 = const()[name = tensor("op_9037_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_9037_end_0 = const()[name = tensor("op_9037_end_0"), val = tensor([2, 240, 1, 1024])]; tensor var_9037_end_mask_0 = const()[name = tensor("op_9037_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9037_cast_fp16 = slice_by_index(begin = var_9037_begin_0, end = var_9037_end_0, end_mask = var_9037_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9037_cast_fp16")]; tensor var_9041_begin_0 = const()[name = tensor("op_9041_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_9041_end_0 = const()[name = tensor("op_9041_end_0"), val = tensor([2, 320, 1, 1024])]; tensor var_9041_end_mask_0 = const()[name = tensor("op_9041_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9041_cast_fp16 = slice_by_index(begin = var_9041_begin_0, end = var_9041_end_0, end_mask = var_9041_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9041_cast_fp16")]; tensor var_9045_begin_0 = const()[name = tensor("op_9045_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_9045_end_0 = const()[name = tensor("op_9045_end_0"), val = tensor([2, 400, 1, 1024])]; tensor var_9045_end_mask_0 = const()[name = tensor("op_9045_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9045_cast_fp16 = slice_by_index(begin = var_9045_begin_0, end = var_9045_end_0, end_mask = var_9045_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9045_cast_fp16")]; tensor var_9049_begin_0 = const()[name = tensor("op_9049_begin_0"), val = tensor([0, 400, 0, 0])]; tensor var_9049_end_0 = const()[name = tensor("op_9049_end_0"), val = tensor([2, 480, 1, 1024])]; tensor var_9049_end_mask_0 = const()[name = tensor("op_9049_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9049_cast_fp16 = slice_by_index(begin = var_9049_begin_0, end = var_9049_end_0, end_mask = var_9049_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9049_cast_fp16")]; tensor var_9053_begin_0 = const()[name = tensor("op_9053_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_9053_end_0 = const()[name = tensor("op_9053_end_0"), val = tensor([2, 560, 1, 1024])]; tensor var_9053_end_mask_0 = const()[name = tensor("op_9053_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9053_cast_fp16 = slice_by_index(begin = var_9053_begin_0, end = var_9053_end_0, end_mask = var_9053_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9053_cast_fp16")]; tensor var_9057_begin_0 = const()[name = tensor("op_9057_begin_0"), val = tensor([0, 560, 0, 0])]; tensor var_9057_end_0 = const()[name = tensor("op_9057_end_0"), val = tensor([2, 1, 1, 1024])]; tensor var_9057_end_mask_0 = const()[name = tensor("op_9057_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9057_cast_fp16 = slice_by_index(begin = var_9057_begin_0, end = var_9057_end_0, end_mask = var_9057_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9057_cast_fp16")]; tensor var_9060_begin_0 = const()[name = tensor("op_9060_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9060_end_0 = const()[name = tensor("op_9060_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_9060_end_mask_0 = const()[name = tensor("op_9060_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9060_cast_fp16 = slice_by_index(begin = var_9060_begin_0, end = var_9060_end_0, end_mask = var_9060_end_mask_0, x = var_9029_cast_fp16)[name = tensor("op_9060_cast_fp16")]; tensor var_9061_begin_0 = const()[name = tensor("op_9061_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9061_end_0 = const()[name = tensor("op_9061_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_9061_end_mask_0 = const()[name = tensor("op_9061_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9061_cast_fp16 = slice_by_index(begin = var_9061_begin_0, end = var_9061_end_0, end_mask = var_9061_end_mask_0, x = var_9029_cast_fp16)[name = tensor("op_9061_cast_fp16")]; tensor var_9062_begin_0 = const()[name = tensor("op_9062_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9062_end_0 = const()[name = tensor("op_9062_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_9062_end_mask_0 = const()[name = tensor("op_9062_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9062_cast_fp16 = slice_by_index(begin = var_9062_begin_0, end = var_9062_end_0, end_mask = var_9062_end_mask_0, x = var_9033_cast_fp16)[name = tensor("op_9062_cast_fp16")]; tensor var_9063_begin_0 = const()[name = tensor("op_9063_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9063_end_0 = const()[name = tensor("op_9063_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_9063_end_mask_0 = const()[name = tensor("op_9063_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9063_cast_fp16 = slice_by_index(begin = var_9063_begin_0, end = var_9063_end_0, end_mask = var_9063_end_mask_0, x = var_9033_cast_fp16)[name = tensor("op_9063_cast_fp16")]; tensor var_9064_begin_0 = const()[name = tensor("op_9064_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9064_end_0 = const()[name = tensor("op_9064_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_9064_end_mask_0 = const()[name = tensor("op_9064_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9064_cast_fp16 = slice_by_index(begin = var_9064_begin_0, end = var_9064_end_0, end_mask = var_9064_end_mask_0, x = var_9037_cast_fp16)[name = tensor("op_9064_cast_fp16")]; tensor var_9065_begin_0 = const()[name = tensor("op_9065_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9065_end_0 = const()[name = tensor("op_9065_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_9065_end_mask_0 = const()[name = tensor("op_9065_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9065_cast_fp16 = slice_by_index(begin = var_9065_begin_0, end = var_9065_end_0, end_mask = var_9065_end_mask_0, x = var_9037_cast_fp16)[name = tensor("op_9065_cast_fp16")]; tensor var_9066_begin_0 = const()[name = tensor("op_9066_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9066_end_0 = const()[name = tensor("op_9066_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_9066_end_mask_0 = const()[name = tensor("op_9066_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9066_cast_fp16 = slice_by_index(begin = var_9066_begin_0, end = var_9066_end_0, end_mask = var_9066_end_mask_0, x = var_9041_cast_fp16)[name = tensor("op_9066_cast_fp16")]; tensor var_9067_begin_0 = const()[name = tensor("op_9067_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9067_end_0 = const()[name = tensor("op_9067_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_9067_end_mask_0 = const()[name = tensor("op_9067_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9067_cast_fp16 = slice_by_index(begin = var_9067_begin_0, end = var_9067_end_0, end_mask = var_9067_end_mask_0, x = var_9041_cast_fp16)[name = tensor("op_9067_cast_fp16")]; tensor var_9068_begin_0 = const()[name = tensor("op_9068_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9068_end_0 = const()[name = tensor("op_9068_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_9068_end_mask_0 = const()[name = tensor("op_9068_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9068_cast_fp16 = slice_by_index(begin = var_9068_begin_0, end = var_9068_end_0, end_mask = var_9068_end_mask_0, x = var_9045_cast_fp16)[name = tensor("op_9068_cast_fp16")]; tensor var_9069_begin_0 = const()[name = tensor("op_9069_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9069_end_0 = const()[name = tensor("op_9069_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_9069_end_mask_0 = const()[name = tensor("op_9069_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9069_cast_fp16 = slice_by_index(begin = var_9069_begin_0, end = var_9069_end_0, end_mask = var_9069_end_mask_0, x = var_9045_cast_fp16)[name = tensor("op_9069_cast_fp16")]; tensor var_9070_begin_0 = const()[name = tensor("op_9070_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9070_end_0 = const()[name = tensor("op_9070_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_9070_end_mask_0 = const()[name = tensor("op_9070_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9070_cast_fp16 = slice_by_index(begin = var_9070_begin_0, end = var_9070_end_0, end_mask = var_9070_end_mask_0, x = var_9049_cast_fp16)[name = tensor("op_9070_cast_fp16")]; tensor var_9071_begin_0 = const()[name = tensor("op_9071_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9071_end_0 = const()[name = tensor("op_9071_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_9071_end_mask_0 = const()[name = tensor("op_9071_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9071_cast_fp16 = slice_by_index(begin = var_9071_begin_0, end = var_9071_end_0, end_mask = var_9071_end_mask_0, x = var_9049_cast_fp16)[name = tensor("op_9071_cast_fp16")]; tensor var_9072_begin_0 = const()[name = tensor("op_9072_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9072_end_0 = const()[name = tensor("op_9072_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_9072_end_mask_0 = const()[name = tensor("op_9072_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9072_cast_fp16 = slice_by_index(begin = var_9072_begin_0, end = var_9072_end_0, end_mask = var_9072_end_mask_0, x = var_9053_cast_fp16)[name = tensor("op_9072_cast_fp16")]; tensor var_9073_begin_0 = const()[name = tensor("op_9073_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9073_end_0 = const()[name = tensor("op_9073_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_9073_end_mask_0 = const()[name = tensor("op_9073_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9073_cast_fp16 = slice_by_index(begin = var_9073_begin_0, end = var_9073_end_0, end_mask = var_9073_end_mask_0, x = var_9053_cast_fp16)[name = tensor("op_9073_cast_fp16")]; tensor var_9074_begin_0 = const()[name = tensor("op_9074_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9074_end_0 = const()[name = tensor("op_9074_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_9074_end_mask_0 = const()[name = tensor("op_9074_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9074_cast_fp16 = slice_by_index(begin = var_9074_begin_0, end = var_9074_end_0, end_mask = var_9074_end_mask_0, x = var_9057_cast_fp16)[name = tensor("op_9074_cast_fp16")]; tensor var_9075_begin_0 = const()[name = tensor("op_9075_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9075_end_0 = const()[name = tensor("op_9075_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_9075_end_mask_0 = const()[name = tensor("op_9075_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9075_cast_fp16 = slice_by_index(begin = var_9075_begin_0, end = var_9075_end_0, end_mask = var_9075_end_mask_0, x = var_9057_cast_fp16)[name = tensor("op_9075_cast_fp16")]; tensor k_91_perm_0 = const()[name = tensor("k_91_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_9080_begin_0 = const()[name = tensor("op_9080_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9080_end_0 = const()[name = tensor("op_9080_end_0"), val = tensor([2, 1024, 1, 80])]; tensor var_9080_end_mask_0 = const()[name = tensor("op_9080_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_91_cast_fp16 = transpose(perm = k_91_perm_0, x = k_89_cast_fp16)[name = tensor("transpose_9")]; tensor var_9080_cast_fp16 = slice_by_index(begin = var_9080_begin_0, end = var_9080_end_0, end_mask = var_9080_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9080_cast_fp16")]; tensor var_9084_begin_0 = const()[name = tensor("op_9084_begin_0"), val = tensor([0, 0, 0, 80])]; tensor var_9084_end_0 = const()[name = tensor("op_9084_end_0"), val = tensor([2, 1024, 1, 160])]; tensor var_9084_end_mask_0 = const()[name = tensor("op_9084_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9084_cast_fp16 = slice_by_index(begin = var_9084_begin_0, end = var_9084_end_0, end_mask = var_9084_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9084_cast_fp16")]; tensor var_9088_begin_0 = const()[name = tensor("op_9088_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_9088_end_0 = const()[name = tensor("op_9088_end_0"), val = tensor([2, 1024, 1, 240])]; tensor var_9088_end_mask_0 = const()[name = tensor("op_9088_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9088_cast_fp16 = slice_by_index(begin = var_9088_begin_0, end = var_9088_end_0, end_mask = var_9088_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9088_cast_fp16")]; tensor var_9092_begin_0 = const()[name = tensor("op_9092_begin_0"), val = tensor([0, 0, 0, 240])]; tensor var_9092_end_0 = const()[name = tensor("op_9092_end_0"), val = tensor([2, 1024, 1, 320])]; tensor var_9092_end_mask_0 = const()[name = tensor("op_9092_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9092_cast_fp16 = slice_by_index(begin = var_9092_begin_0, end = var_9092_end_0, end_mask = var_9092_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9092_cast_fp16")]; tensor var_9096_begin_0 = const()[name = tensor("op_9096_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_9096_end_0 = const()[name = tensor("op_9096_end_0"), val = tensor([2, 1024, 1, 400])]; tensor var_9096_end_mask_0 = const()[name = tensor("op_9096_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9096_cast_fp16 = slice_by_index(begin = var_9096_begin_0, end = var_9096_end_0, end_mask = var_9096_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9096_cast_fp16")]; tensor var_9100_begin_0 = const()[name = tensor("op_9100_begin_0"), val = tensor([0, 0, 0, 400])]; tensor var_9100_end_0 = const()[name = tensor("op_9100_end_0"), val = tensor([2, 1024, 1, 480])]; tensor var_9100_end_mask_0 = const()[name = tensor("op_9100_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9100_cast_fp16 = slice_by_index(begin = var_9100_begin_0, end = var_9100_end_0, end_mask = var_9100_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9100_cast_fp16")]; tensor var_9104_begin_0 = const()[name = tensor("op_9104_begin_0"), val = tensor([0, 0, 0, 480])]; tensor var_9104_end_0 = const()[name = tensor("op_9104_end_0"), val = tensor([2, 1024, 1, 560])]; tensor var_9104_end_mask_0 = const()[name = tensor("op_9104_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9104_cast_fp16 = slice_by_index(begin = var_9104_begin_0, end = var_9104_end_0, end_mask = var_9104_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9104_cast_fp16")]; tensor var_9108_begin_0 = const()[name = tensor("op_9108_begin_0"), val = tensor([0, 0, 0, 560])]; tensor var_9108_end_0 = const()[name = tensor("op_9108_end_0"), val = tensor([2, 1024, 1, 1])]; tensor var_9108_end_mask_0 = const()[name = tensor("op_9108_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9108_cast_fp16 = slice_by_index(begin = var_9108_begin_0, end = var_9108_end_0, end_mask = var_9108_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9108_cast_fp16")]; tensor var_9110_begin_0 = const()[name = tensor("op_9110_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9110_end_0 = const()[name = tensor("op_9110_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_9110_end_mask_0 = const()[name = tensor("op_9110_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9110_cast_fp16 = slice_by_index(begin = var_9110_begin_0, end = var_9110_end_0, end_mask = var_9110_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9110_cast_fp16")]; tensor var_9114_begin_0 = const()[name = tensor("op_9114_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_9114_end_0 = const()[name = tensor("op_9114_end_0"), val = tensor([2, 160, 1, 1024])]; tensor var_9114_end_mask_0 = const()[name = tensor("op_9114_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9114_cast_fp16 = slice_by_index(begin = var_9114_begin_0, end = var_9114_end_0, end_mask = var_9114_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9114_cast_fp16")]; tensor var_9118_begin_0 = const()[name = tensor("op_9118_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_9118_end_0 = const()[name = tensor("op_9118_end_0"), val = tensor([2, 240, 1, 1024])]; tensor var_9118_end_mask_0 = const()[name = tensor("op_9118_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9118_cast_fp16 = slice_by_index(begin = var_9118_begin_0, end = var_9118_end_0, end_mask = var_9118_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9118_cast_fp16")]; tensor var_9122_begin_0 = const()[name = tensor("op_9122_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_9122_end_0 = const()[name = tensor("op_9122_end_0"), val = tensor([2, 320, 1, 1024])]; tensor var_9122_end_mask_0 = const()[name = tensor("op_9122_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9122_cast_fp16 = slice_by_index(begin = var_9122_begin_0, end = var_9122_end_0, end_mask = var_9122_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9122_cast_fp16")]; tensor var_9126_begin_0 = const()[name = tensor("op_9126_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_9126_end_0 = const()[name = tensor("op_9126_end_0"), val = tensor([2, 400, 1, 1024])]; tensor var_9126_end_mask_0 = const()[name = tensor("op_9126_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9126_cast_fp16 = slice_by_index(begin = var_9126_begin_0, end = var_9126_end_0, end_mask = var_9126_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9126_cast_fp16")]; tensor var_9130_begin_0 = const()[name = tensor("op_9130_begin_0"), val = tensor([0, 400, 0, 0])]; tensor var_9130_end_0 = const()[name = tensor("op_9130_end_0"), val = tensor([2, 480, 1, 1024])]; tensor var_9130_end_mask_0 = const()[name = tensor("op_9130_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9130_cast_fp16 = slice_by_index(begin = var_9130_begin_0, end = var_9130_end_0, end_mask = var_9130_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9130_cast_fp16")]; tensor var_9134_begin_0 = const()[name = tensor("op_9134_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_9134_end_0 = const()[name = tensor("op_9134_end_0"), val = tensor([2, 560, 1, 1024])]; tensor var_9134_end_mask_0 = const()[name = tensor("op_9134_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9134_cast_fp16 = slice_by_index(begin = var_9134_begin_0, end = var_9134_end_0, end_mask = var_9134_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9134_cast_fp16")]; tensor var_9138_begin_0 = const()[name = tensor("op_9138_begin_0"), val = tensor([0, 560, 0, 0])]; tensor var_9138_end_0 = const()[name = tensor("op_9138_end_0"), val = tensor([2, 1, 1, 1024])]; tensor var_9138_end_mask_0 = const()[name = tensor("op_9138_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9138_cast_fp16 = slice_by_index(begin = var_9138_begin_0, end = var_9138_end_0, end_mask = var_9138_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9138_cast_fp16")]; tensor var_9142_equation_0 = const()[name = tensor("op_9142_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9142_cast_fp16 = einsum(equation = var_9142_equation_0, values = (var_9080_cast_fp16, var_9060_cast_fp16))[name = tensor("op_9142_cast_fp16")]; tensor var_9143_to_fp16 = const()[name = tensor("op_9143_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_705_cast_fp16 = mul(x = var_9142_cast_fp16, y = var_9143_to_fp16)[name = tensor("aw_chunk_705_cast_fp16")]; tensor var_9146_equation_0 = const()[name = tensor("op_9146_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9146_cast_fp16 = einsum(equation = var_9146_equation_0, values = (var_9080_cast_fp16, var_9061_cast_fp16))[name = tensor("op_9146_cast_fp16")]; tensor var_9147_to_fp16 = const()[name = tensor("op_9147_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_707_cast_fp16 = mul(x = var_9146_cast_fp16, y = var_9147_to_fp16)[name = tensor("aw_chunk_707_cast_fp16")]; tensor var_9150_equation_0 = const()[name = tensor("op_9150_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9150_cast_fp16 = einsum(equation = var_9150_equation_0, values = (var_9084_cast_fp16, var_9062_cast_fp16))[name = tensor("op_9150_cast_fp16")]; tensor var_9151_to_fp16 = const()[name = tensor("op_9151_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_709_cast_fp16 = mul(x = var_9150_cast_fp16, y = var_9151_to_fp16)[name = tensor("aw_chunk_709_cast_fp16")]; tensor var_9154_equation_0 = const()[name = tensor("op_9154_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9154_cast_fp16 = einsum(equation = var_9154_equation_0, values = (var_9084_cast_fp16, var_9063_cast_fp16))[name = tensor("op_9154_cast_fp16")]; tensor var_9155_to_fp16 = const()[name = tensor("op_9155_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_711_cast_fp16 = mul(x = var_9154_cast_fp16, y = var_9155_to_fp16)[name = tensor("aw_chunk_711_cast_fp16")]; tensor var_9158_equation_0 = const()[name = tensor("op_9158_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9158_cast_fp16 = einsum(equation = var_9158_equation_0, values = (var_9088_cast_fp16, var_9064_cast_fp16))[name = tensor("op_9158_cast_fp16")]; tensor var_9159_to_fp16 = const()[name = tensor("op_9159_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_713_cast_fp16 = mul(x = var_9158_cast_fp16, y = var_9159_to_fp16)[name = tensor("aw_chunk_713_cast_fp16")]; tensor var_9162_equation_0 = const()[name = tensor("op_9162_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9162_cast_fp16 = einsum(equation = var_9162_equation_0, values = (var_9088_cast_fp16, var_9065_cast_fp16))[name = tensor("op_9162_cast_fp16")]; tensor var_9163_to_fp16 = const()[name = tensor("op_9163_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_715_cast_fp16 = mul(x = var_9162_cast_fp16, y = var_9163_to_fp16)[name = tensor("aw_chunk_715_cast_fp16")]; tensor var_9166_equation_0 = const()[name = tensor("op_9166_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9166_cast_fp16 = einsum(equation = var_9166_equation_0, values = (var_9092_cast_fp16, var_9066_cast_fp16))[name = tensor("op_9166_cast_fp16")]; tensor var_9167_to_fp16 = const()[name = tensor("op_9167_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_717_cast_fp16 = mul(x = var_9166_cast_fp16, y = var_9167_to_fp16)[name = tensor("aw_chunk_717_cast_fp16")]; tensor var_9170_equation_0 = const()[name = tensor("op_9170_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9170_cast_fp16 = einsum(equation = var_9170_equation_0, values = (var_9092_cast_fp16, var_9067_cast_fp16))[name = tensor("op_9170_cast_fp16")]; tensor var_9171_to_fp16 = const()[name = tensor("op_9171_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_719_cast_fp16 = mul(x = var_9170_cast_fp16, y = var_9171_to_fp16)[name = tensor("aw_chunk_719_cast_fp16")]; tensor var_9174_equation_0 = const()[name = tensor("op_9174_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9174_cast_fp16 = einsum(equation = var_9174_equation_0, values = (var_9096_cast_fp16, var_9068_cast_fp16))[name = tensor("op_9174_cast_fp16")]; tensor var_9175_to_fp16 = const()[name = tensor("op_9175_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_721_cast_fp16 = mul(x = var_9174_cast_fp16, y = var_9175_to_fp16)[name = tensor("aw_chunk_721_cast_fp16")]; tensor var_9178_equation_0 = const()[name = tensor("op_9178_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9178_cast_fp16 = einsum(equation = var_9178_equation_0, values = (var_9096_cast_fp16, var_9069_cast_fp16))[name = tensor("op_9178_cast_fp16")]; tensor var_9179_to_fp16 = const()[name = tensor("op_9179_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_723_cast_fp16 = mul(x = var_9178_cast_fp16, y = var_9179_to_fp16)[name = tensor("aw_chunk_723_cast_fp16")]; tensor var_9182_equation_0 = const()[name = tensor("op_9182_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9182_cast_fp16 = einsum(equation = var_9182_equation_0, values = (var_9100_cast_fp16, var_9070_cast_fp16))[name = tensor("op_9182_cast_fp16")]; tensor var_9183_to_fp16 = const()[name = tensor("op_9183_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_725_cast_fp16 = mul(x = var_9182_cast_fp16, y = var_9183_to_fp16)[name = tensor("aw_chunk_725_cast_fp16")]; tensor var_9186_equation_0 = const()[name = tensor("op_9186_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9186_cast_fp16 = einsum(equation = var_9186_equation_0, values = (var_9100_cast_fp16, var_9071_cast_fp16))[name = tensor("op_9186_cast_fp16")]; tensor var_9187_to_fp16 = const()[name = tensor("op_9187_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_727_cast_fp16 = mul(x = var_9186_cast_fp16, y = var_9187_to_fp16)[name = tensor("aw_chunk_727_cast_fp16")]; tensor var_9190_equation_0 = const()[name = tensor("op_9190_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9190_cast_fp16 = einsum(equation = var_9190_equation_0, values = (var_9104_cast_fp16, var_9072_cast_fp16))[name = tensor("op_9190_cast_fp16")]; tensor var_9191_to_fp16 = const()[name = tensor("op_9191_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_729_cast_fp16 = mul(x = var_9190_cast_fp16, y = var_9191_to_fp16)[name = tensor("aw_chunk_729_cast_fp16")]; tensor var_9194_equation_0 = const()[name = tensor("op_9194_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9194_cast_fp16 = einsum(equation = var_9194_equation_0, values = (var_9104_cast_fp16, var_9073_cast_fp16))[name = tensor("op_9194_cast_fp16")]; tensor var_9195_to_fp16 = const()[name = tensor("op_9195_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_731_cast_fp16 = mul(x = var_9194_cast_fp16, y = var_9195_to_fp16)[name = tensor("aw_chunk_731_cast_fp16")]; tensor var_9198_equation_0 = const()[name = tensor("op_9198_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9198_cast_fp16 = einsum(equation = var_9198_equation_0, values = (var_9108_cast_fp16, var_9074_cast_fp16))[name = tensor("op_9198_cast_fp16")]; tensor var_9199_to_fp16 = const()[name = tensor("op_9199_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_733_cast_fp16 = mul(x = var_9198_cast_fp16, y = var_9199_to_fp16)[name = tensor("aw_chunk_733_cast_fp16")]; tensor var_9202_equation_0 = const()[name = tensor("op_9202_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9202_cast_fp16 = einsum(equation = var_9202_equation_0, values = (var_9108_cast_fp16, var_9075_cast_fp16))[name = tensor("op_9202_cast_fp16")]; tensor var_9203_to_fp16 = const()[name = tensor("op_9203_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_735_cast_fp16 = mul(x = var_9202_cast_fp16, y = var_9203_to_fp16)[name = tensor("aw_chunk_735_cast_fp16")]; tensor var_9205_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_705_cast_fp16)[name = tensor("op_9205_cast_fp16")]; tensor var_9206_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_707_cast_fp16)[name = tensor("op_9206_cast_fp16")]; tensor var_9207_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_709_cast_fp16)[name = tensor("op_9207_cast_fp16")]; tensor var_9208_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_711_cast_fp16)[name = tensor("op_9208_cast_fp16")]; tensor var_9209_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_713_cast_fp16)[name = tensor("op_9209_cast_fp16")]; tensor var_9210_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_715_cast_fp16)[name = tensor("op_9210_cast_fp16")]; tensor var_9211_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_717_cast_fp16)[name = tensor("op_9211_cast_fp16")]; tensor var_9212_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_719_cast_fp16)[name = tensor("op_9212_cast_fp16")]; tensor var_9213_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_721_cast_fp16)[name = tensor("op_9213_cast_fp16")]; tensor var_9214_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_723_cast_fp16)[name = tensor("op_9214_cast_fp16")]; tensor var_9215_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_725_cast_fp16)[name = tensor("op_9215_cast_fp16")]; tensor var_9216_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_727_cast_fp16)[name = tensor("op_9216_cast_fp16")]; tensor var_9217_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_729_cast_fp16)[name = tensor("op_9217_cast_fp16")]; tensor var_9218_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_731_cast_fp16)[name = tensor("op_9218_cast_fp16")]; tensor var_9219_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_733_cast_fp16)[name = tensor("op_9219_cast_fp16")]; tensor var_9220_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_735_cast_fp16)[name = tensor("op_9220_cast_fp16")]; tensor var_9222_equation_0 = const()[name = tensor("op_9222_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9222_cast_fp16 = einsum(equation = var_9222_equation_0, values = (var_9110_cast_fp16, var_9205_cast_fp16))[name = tensor("op_9222_cast_fp16")]; tensor var_9224_equation_0 = const()[name = tensor("op_9224_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9224_cast_fp16 = einsum(equation = var_9224_equation_0, values = (var_9110_cast_fp16, var_9206_cast_fp16))[name = tensor("op_9224_cast_fp16")]; tensor var_9226_equation_0 = const()[name = tensor("op_9226_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9226_cast_fp16 = einsum(equation = var_9226_equation_0, values = (var_9114_cast_fp16, var_9207_cast_fp16))[name = tensor("op_9226_cast_fp16")]; tensor var_9228_equation_0 = const()[name = tensor("op_9228_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9228_cast_fp16 = einsum(equation = var_9228_equation_0, values = (var_9114_cast_fp16, var_9208_cast_fp16))[name = tensor("op_9228_cast_fp16")]; tensor var_9230_equation_0 = const()[name = tensor("op_9230_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9230_cast_fp16 = einsum(equation = var_9230_equation_0, values = (var_9118_cast_fp16, var_9209_cast_fp16))[name = tensor("op_9230_cast_fp16")]; tensor var_9232_equation_0 = const()[name = tensor("op_9232_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9232_cast_fp16 = einsum(equation = var_9232_equation_0, values = (var_9118_cast_fp16, var_9210_cast_fp16))[name = tensor("op_9232_cast_fp16")]; tensor var_9234_equation_0 = const()[name = tensor("op_9234_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9234_cast_fp16 = einsum(equation = var_9234_equation_0, values = (var_9122_cast_fp16, var_9211_cast_fp16))[name = tensor("op_9234_cast_fp16")]; tensor var_9236_equation_0 = const()[name = tensor("op_9236_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9236_cast_fp16 = einsum(equation = var_9236_equation_0, values = (var_9122_cast_fp16, var_9212_cast_fp16))[name = tensor("op_9236_cast_fp16")]; tensor var_9238_equation_0 = const()[name = tensor("op_9238_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9238_cast_fp16 = einsum(equation = var_9238_equation_0, values = (var_9126_cast_fp16, var_9213_cast_fp16))[name = tensor("op_9238_cast_fp16")]; tensor var_9240_equation_0 = const()[name = tensor("op_9240_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9240_cast_fp16 = einsum(equation = var_9240_equation_0, values = (var_9126_cast_fp16, var_9214_cast_fp16))[name = tensor("op_9240_cast_fp16")]; tensor var_9242_equation_0 = const()[name = tensor("op_9242_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9242_cast_fp16 = einsum(equation = var_9242_equation_0, values = (var_9130_cast_fp16, var_9215_cast_fp16))[name = tensor("op_9242_cast_fp16")]; tensor var_9244_equation_0 = const()[name = tensor("op_9244_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9244_cast_fp16 = einsum(equation = var_9244_equation_0, values = (var_9130_cast_fp16, var_9216_cast_fp16))[name = tensor("op_9244_cast_fp16")]; tensor var_9246_equation_0 = const()[name = tensor("op_9246_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9246_cast_fp16 = einsum(equation = var_9246_equation_0, values = (var_9134_cast_fp16, var_9217_cast_fp16))[name = tensor("op_9246_cast_fp16")]; tensor var_9248_equation_0 = const()[name = tensor("op_9248_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9248_cast_fp16 = einsum(equation = var_9248_equation_0, values = (var_9134_cast_fp16, var_9218_cast_fp16))[name = tensor("op_9248_cast_fp16")]; tensor var_9250_equation_0 = const()[name = tensor("op_9250_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9250_cast_fp16 = einsum(equation = var_9250_equation_0, values = (var_9138_cast_fp16, var_9219_cast_fp16))[name = tensor("op_9250_cast_fp16")]; tensor var_9252_equation_0 = const()[name = tensor("op_9252_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9252_cast_fp16 = einsum(equation = var_9252_equation_0, values = (var_9138_cast_fp16, var_9220_cast_fp16))[name = tensor("op_9252_cast_fp16")]; tensor var_9254_interleave_0 = const()[name = tensor("op_9254_interleave_0"), val = tensor(false)]; tensor var_9254_cast_fp16 = concat(axis = var_8160, interleave = var_9254_interleave_0, values = (var_9222_cast_fp16, var_9224_cast_fp16))[name = tensor("op_9254_cast_fp16")]; tensor var_9256_interleave_0 = const()[name = tensor("op_9256_interleave_0"), val = tensor(false)]; tensor var_9256_cast_fp16 = concat(axis = var_8160, interleave = var_9256_interleave_0, values = (var_9226_cast_fp16, var_9228_cast_fp16))[name = tensor("op_9256_cast_fp16")]; tensor var_9258_interleave_0 = const()[name = tensor("op_9258_interleave_0"), val = tensor(false)]; tensor var_9258_cast_fp16 = concat(axis = var_8160, interleave = var_9258_interleave_0, values = (var_9230_cast_fp16, var_9232_cast_fp16))[name = tensor("op_9258_cast_fp16")]; tensor var_9260_interleave_0 = const()[name = tensor("op_9260_interleave_0"), val = tensor(false)]; tensor var_9260_cast_fp16 = concat(axis = var_8160, interleave = var_9260_interleave_0, values = (var_9234_cast_fp16, var_9236_cast_fp16))[name = tensor("op_9260_cast_fp16")]; tensor var_9262_interleave_0 = const()[name = tensor("op_9262_interleave_0"), val = tensor(false)]; tensor var_9262_cast_fp16 = concat(axis = var_8160, interleave = var_9262_interleave_0, values = (var_9238_cast_fp16, var_9240_cast_fp16))[name = tensor("op_9262_cast_fp16")]; tensor var_9264_interleave_0 = const()[name = tensor("op_9264_interleave_0"), val = tensor(false)]; tensor var_9264_cast_fp16 = concat(axis = var_8160, interleave = var_9264_interleave_0, values = (var_9242_cast_fp16, var_9244_cast_fp16))[name = tensor("op_9264_cast_fp16")]; tensor var_9266_interleave_0 = const()[name = tensor("op_9266_interleave_0"), val = tensor(false)]; tensor var_9266_cast_fp16 = concat(axis = var_8160, interleave = var_9266_interleave_0, values = (var_9246_cast_fp16, var_9248_cast_fp16))[name = tensor("op_9266_cast_fp16")]; tensor var_9268_interleave_0 = const()[name = tensor("op_9268_interleave_0"), val = tensor(false)]; tensor var_9268_cast_fp16 = concat(axis = var_8160, interleave = var_9268_interleave_0, values = (var_9250_cast_fp16, var_9252_cast_fp16))[name = tensor("op_9268_cast_fp16")]; tensor input_409_interleave_0 = const()[name = tensor("input_409_interleave_0"), val = tensor(false)]; tensor input_409_cast_fp16 = concat(axis = var_8182, interleave = input_409_interleave_0, values = (var_9254_cast_fp16, var_9256_cast_fp16, var_9258_cast_fp16, var_9260_cast_fp16, var_9262_cast_fp16, var_9264_cast_fp16, var_9266_cast_fp16, var_9268_cast_fp16))[name = tensor("input_409_cast_fp16")]; tensor var_9278_pad_type_0 = const()[name = tensor("op_9278_pad_type_0"), val = tensor("valid")]; tensor var_9278_strides_0 = const()[name = tensor("op_9278_strides_0"), val = tensor([1, 1])]; tensor var_9278_pad_0 = const()[name = tensor("op_9278_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_9278_dilations_0 = const()[name = tensor("op_9278_dilations_0"), val = tensor([1, 1])]; tensor var_9278_groups_0 = const()[name = tensor("op_9278_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(731751808)))]; tensor up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(732571072)))]; tensor var_9278_cast_fp16 = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_9278_dilations_0, groups = var_9278_groups_0, pad = var_9278_pad_0, pad_type = var_9278_pad_type_0, strides = var_9278_strides_0, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_409_cast_fp16)[name = tensor("op_9278_cast_fp16")]; tensor inputs_69_cast_fp16 = add(x = var_9278_cast_fp16, y = inputs_67_cast_fp16)[name = tensor("inputs_69_cast_fp16")]; tensor hidden_states_249_axes_0 = const()[name = tensor("hidden_states_249_axes_0"), val = tensor([1])]; tensor hidden_states_249_gamma_0_to_fp16 = const()[name = tensor("hidden_states_249_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(732572416)))]; tensor hidden_states_249_beta_0_to_fp16 = const()[name = tensor("hidden_states_249_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(732573760)))]; tensor var_9288_to_fp16 = const()[name = tensor("op_9288_to_fp16"), val = tensor(0x1.5p-17)]; tensor hidden_states_249_cast_fp16 = layer_norm(axes = hidden_states_249_axes_0, beta = hidden_states_249_beta_0_to_fp16, epsilon = var_9288_to_fp16, gamma = hidden_states_249_gamma_0_to_fp16, x = inputs_69_cast_fp16)[name = tensor("hidden_states_249_cast_fp16")]; tensor q_47_pad_type_0 = const()[name = tensor("q_47_pad_type_0"), val = tensor("valid")]; tensor q_47_strides_0 = const()[name = tensor("q_47_strides_0"), val = tensor([1, 1])]; tensor q_47_pad_0 = const()[name = tensor("q_47_pad_0"), val = tensor([0, 0, 0, 0])]; tensor q_47_dilations_0 = const()[name = tensor("q_47_dilations_0"), val = tensor([1, 1])]; tensor q_47_groups_0 = const()[name = tensor("q_47_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(732575104)))]; tensor q_47_cast_fp16 = conv(dilations = q_47_dilations_0, groups = q_47_groups_0, pad = q_47_pad_0, pad_type = q_47_pad_type_0, strides = q_47_strides_0, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_249_cast_fp16)[name = tensor("q_47_cast_fp16")]; tensor k_93_pad_type_0 = const()[name = tensor("k_93_pad_type_0"), val = tensor("valid")]; tensor k_93_strides_0 = const()[name = tensor("k_93_strides_0"), val = tensor([1, 1])]; tensor k_93_pad_0 = const()[name = tensor("k_93_pad_0"), val = tensor([0, 0, 0, 0])]; tensor k_93_dilations_0 = const()[name = tensor("k_93_dilations_0"), val = tensor([1, 1])]; tensor k_93_groups_0 = const()[name = tensor("k_93_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(733394368)))]; tensor k_93_cast_fp16 = conv(dilations = k_93_dilations_0, groups = k_93_groups_0, pad = k_93_pad_0, pad_type = k_93_pad_type_0, strides = k_93_strides_0, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_93_cast_fp16")]; tensor v_47_pad_type_0 = const()[name = tensor("v_47_pad_type_0"), val = tensor("valid")]; tensor v_47_strides_0 = const()[name = tensor("v_47_strides_0"), val = tensor([1, 1])]; tensor v_47_pad_0 = const()[name = tensor("v_47_pad_0"), val = tensor([0, 0, 0, 0])]; tensor v_47_dilations_0 = const()[name = tensor("v_47_dilations_0"), val = tensor([1, 1])]; tensor v_47_groups_0 = const()[name = tensor("v_47_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(734377472)))]; tensor v_47_cast_fp16 = conv(dilations = v_47_dilations_0, groups = v_47_groups_0, pad = v_47_pad_0, pad_type = v_47_pad_type_0, strides = v_47_strides_0, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_47_cast_fp16")]; tensor var_9321_begin_0 = const()[name = tensor("op_9321_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9321_end_0 = const()[name = tensor("op_9321_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_9321_end_mask_0 = const()[name = tensor("op_9321_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9321_cast_fp16 = slice_by_index(begin = var_9321_begin_0, end = var_9321_end_0, end_mask = var_9321_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9321_cast_fp16")]; tensor var_9325_begin_0 = const()[name = tensor("op_9325_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_9325_end_0 = const()[name = tensor("op_9325_end_0"), val = tensor([2, 160, 1, 1024])]; tensor var_9325_end_mask_0 = const()[name = tensor("op_9325_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9325_cast_fp16 = slice_by_index(begin = var_9325_begin_0, end = var_9325_end_0, end_mask = var_9325_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9325_cast_fp16")]; tensor var_9329_begin_0 = const()[name = tensor("op_9329_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_9329_end_0 = const()[name = tensor("op_9329_end_0"), val = tensor([2, 240, 1, 1024])]; tensor var_9329_end_mask_0 = const()[name = tensor("op_9329_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9329_cast_fp16 = slice_by_index(begin = var_9329_begin_0, end = var_9329_end_0, end_mask = var_9329_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9329_cast_fp16")]; tensor var_9333_begin_0 = const()[name = tensor("op_9333_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_9333_end_0 = const()[name = tensor("op_9333_end_0"), val = tensor([2, 320, 1, 1024])]; tensor var_9333_end_mask_0 = const()[name = tensor("op_9333_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9333_cast_fp16 = slice_by_index(begin = var_9333_begin_0, end = var_9333_end_0, end_mask = var_9333_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9333_cast_fp16")]; tensor var_9337_begin_0 = const()[name = tensor("op_9337_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_9337_end_0 = const()[name = tensor("op_9337_end_0"), val = tensor([2, 400, 1, 1024])]; tensor var_9337_end_mask_0 = const()[name = tensor("op_9337_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9337_cast_fp16 = slice_by_index(begin = var_9337_begin_0, end = var_9337_end_0, end_mask = var_9337_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9337_cast_fp16")]; tensor var_9341_begin_0 = const()[name = tensor("op_9341_begin_0"), val = tensor([0, 400, 0, 0])]; tensor var_9341_end_0 = const()[name = tensor("op_9341_end_0"), val = tensor([2, 480, 1, 1024])]; tensor var_9341_end_mask_0 = const()[name = tensor("op_9341_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9341_cast_fp16 = slice_by_index(begin = var_9341_begin_0, end = var_9341_end_0, end_mask = var_9341_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9341_cast_fp16")]; tensor var_9345_begin_0 = const()[name = tensor("op_9345_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_9345_end_0 = const()[name = tensor("op_9345_end_0"), val = tensor([2, 560, 1, 1024])]; tensor var_9345_end_mask_0 = const()[name = tensor("op_9345_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9345_cast_fp16 = slice_by_index(begin = var_9345_begin_0, end = var_9345_end_0, end_mask = var_9345_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9345_cast_fp16")]; tensor var_9349_begin_0 = const()[name = tensor("op_9349_begin_0"), val = tensor([0, 560, 0, 0])]; tensor var_9349_end_0 = const()[name = tensor("op_9349_end_0"), val = tensor([2, 1, 1, 1024])]; tensor var_9349_end_mask_0 = const()[name = tensor("op_9349_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9349_cast_fp16 = slice_by_index(begin = var_9349_begin_0, end = var_9349_end_0, end_mask = var_9349_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9349_cast_fp16")]; tensor var_9352_begin_0 = const()[name = tensor("op_9352_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9352_end_0 = const()[name = tensor("op_9352_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_9352_end_mask_0 = const()[name = tensor("op_9352_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9352_cast_fp16 = slice_by_index(begin = var_9352_begin_0, end = var_9352_end_0, end_mask = var_9352_end_mask_0, x = var_9321_cast_fp16)[name = tensor("op_9352_cast_fp16")]; tensor var_9353_begin_0 = const()[name = tensor("op_9353_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9353_end_0 = const()[name = tensor("op_9353_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_9353_end_mask_0 = const()[name = tensor("op_9353_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9353_cast_fp16 = slice_by_index(begin = var_9353_begin_0, end = var_9353_end_0, end_mask = var_9353_end_mask_0, x = var_9321_cast_fp16)[name = tensor("op_9353_cast_fp16")]; tensor var_9354_begin_0 = const()[name = tensor("op_9354_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9354_end_0 = const()[name = tensor("op_9354_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_9354_end_mask_0 = const()[name = tensor("op_9354_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9354_cast_fp16 = slice_by_index(begin = var_9354_begin_0, end = var_9354_end_0, end_mask = var_9354_end_mask_0, x = var_9325_cast_fp16)[name = tensor("op_9354_cast_fp16")]; tensor var_9355_begin_0 = const()[name = tensor("op_9355_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9355_end_0 = const()[name = tensor("op_9355_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_9355_end_mask_0 = const()[name = tensor("op_9355_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9355_cast_fp16 = slice_by_index(begin = var_9355_begin_0, end = var_9355_end_0, end_mask = var_9355_end_mask_0, x = var_9325_cast_fp16)[name = tensor("op_9355_cast_fp16")]; tensor var_9356_begin_0 = const()[name = tensor("op_9356_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9356_end_0 = const()[name = tensor("op_9356_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_9356_end_mask_0 = const()[name = tensor("op_9356_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9356_cast_fp16 = slice_by_index(begin = var_9356_begin_0, end = var_9356_end_0, end_mask = var_9356_end_mask_0, x = var_9329_cast_fp16)[name = tensor("op_9356_cast_fp16")]; tensor var_9357_begin_0 = const()[name = tensor("op_9357_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9357_end_0 = const()[name = tensor("op_9357_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_9357_end_mask_0 = const()[name = tensor("op_9357_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9357_cast_fp16 = slice_by_index(begin = var_9357_begin_0, end = var_9357_end_0, end_mask = var_9357_end_mask_0, x = var_9329_cast_fp16)[name = tensor("op_9357_cast_fp16")]; tensor var_9358_begin_0 = const()[name = tensor("op_9358_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9358_end_0 = const()[name = tensor("op_9358_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_9358_end_mask_0 = const()[name = tensor("op_9358_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9358_cast_fp16 = slice_by_index(begin = var_9358_begin_0, end = var_9358_end_0, end_mask = var_9358_end_mask_0, x = var_9333_cast_fp16)[name = tensor("op_9358_cast_fp16")]; tensor var_9359_begin_0 = const()[name = tensor("op_9359_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9359_end_0 = const()[name = tensor("op_9359_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_9359_end_mask_0 = const()[name = tensor("op_9359_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9359_cast_fp16 = slice_by_index(begin = var_9359_begin_0, end = var_9359_end_0, end_mask = var_9359_end_mask_0, x = var_9333_cast_fp16)[name = tensor("op_9359_cast_fp16")]; tensor var_9360_begin_0 = const()[name = tensor("op_9360_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9360_end_0 = const()[name = tensor("op_9360_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_9360_end_mask_0 = const()[name = tensor("op_9360_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9360_cast_fp16 = slice_by_index(begin = var_9360_begin_0, end = var_9360_end_0, end_mask = var_9360_end_mask_0, x = var_9337_cast_fp16)[name = tensor("op_9360_cast_fp16")]; tensor var_9361_begin_0 = const()[name = tensor("op_9361_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9361_end_0 = const()[name = tensor("op_9361_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_9361_end_mask_0 = const()[name = tensor("op_9361_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9361_cast_fp16 = slice_by_index(begin = var_9361_begin_0, end = var_9361_end_0, end_mask = var_9361_end_mask_0, x = var_9337_cast_fp16)[name = tensor("op_9361_cast_fp16")]; tensor var_9362_begin_0 = const()[name = tensor("op_9362_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9362_end_0 = const()[name = tensor("op_9362_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_9362_end_mask_0 = const()[name = tensor("op_9362_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9362_cast_fp16 = slice_by_index(begin = var_9362_begin_0, end = var_9362_end_0, end_mask = var_9362_end_mask_0, x = var_9341_cast_fp16)[name = tensor("op_9362_cast_fp16")]; tensor var_9363_begin_0 = const()[name = tensor("op_9363_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9363_end_0 = const()[name = tensor("op_9363_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_9363_end_mask_0 = const()[name = tensor("op_9363_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9363_cast_fp16 = slice_by_index(begin = var_9363_begin_0, end = var_9363_end_0, end_mask = var_9363_end_mask_0, x = var_9341_cast_fp16)[name = tensor("op_9363_cast_fp16")]; tensor var_9364_begin_0 = const()[name = tensor("op_9364_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9364_end_0 = const()[name = tensor("op_9364_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_9364_end_mask_0 = const()[name = tensor("op_9364_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9364_cast_fp16 = slice_by_index(begin = var_9364_begin_0, end = var_9364_end_0, end_mask = var_9364_end_mask_0, x = var_9345_cast_fp16)[name = tensor("op_9364_cast_fp16")]; tensor var_9365_begin_0 = const()[name = tensor("op_9365_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9365_end_0 = const()[name = tensor("op_9365_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_9365_end_mask_0 = const()[name = tensor("op_9365_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9365_cast_fp16 = slice_by_index(begin = var_9365_begin_0, end = var_9365_end_0, end_mask = var_9365_end_mask_0, x = var_9345_cast_fp16)[name = tensor("op_9365_cast_fp16")]; tensor var_9366_begin_0 = const()[name = tensor("op_9366_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9366_end_0 = const()[name = tensor("op_9366_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_9366_end_mask_0 = const()[name = tensor("op_9366_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9366_cast_fp16 = slice_by_index(begin = var_9366_begin_0, end = var_9366_end_0, end_mask = var_9366_end_mask_0, x = var_9349_cast_fp16)[name = tensor("op_9366_cast_fp16")]; tensor var_9367_begin_0 = const()[name = tensor("op_9367_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9367_end_0 = const()[name = tensor("op_9367_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_9367_end_mask_0 = const()[name = tensor("op_9367_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9367_cast_fp16 = slice_by_index(begin = var_9367_begin_0, end = var_9367_end_0, end_mask = var_9367_end_mask_0, x = var_9349_cast_fp16)[name = tensor("op_9367_cast_fp16")]; tensor k_95_perm_0 = const()[name = tensor("k_95_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_9372_begin_0 = const()[name = tensor("op_9372_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9372_end_0 = const()[name = tensor("op_9372_end_0"), val = tensor([2, 77, 1, 80])]; tensor var_9372_end_mask_0 = const()[name = tensor("op_9372_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_95_cast_fp16 = transpose(perm = k_95_perm_0, x = k_93_cast_fp16)[name = tensor("transpose_8")]; tensor var_9372_cast_fp16 = slice_by_index(begin = var_9372_begin_0, end = var_9372_end_0, end_mask = var_9372_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9372_cast_fp16")]; tensor var_9376_begin_0 = const()[name = tensor("op_9376_begin_0"), val = tensor([0, 0, 0, 80])]; tensor var_9376_end_0 = const()[name = tensor("op_9376_end_0"), val = tensor([2, 77, 1, 160])]; tensor var_9376_end_mask_0 = const()[name = tensor("op_9376_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9376_cast_fp16 = slice_by_index(begin = var_9376_begin_0, end = var_9376_end_0, end_mask = var_9376_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9376_cast_fp16")]; tensor var_9380_begin_0 = const()[name = tensor("op_9380_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_9380_end_0 = const()[name = tensor("op_9380_end_0"), val = tensor([2, 77, 1, 240])]; tensor var_9380_end_mask_0 = const()[name = tensor("op_9380_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9380_cast_fp16 = slice_by_index(begin = var_9380_begin_0, end = var_9380_end_0, end_mask = var_9380_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9380_cast_fp16")]; tensor var_9384_begin_0 = const()[name = tensor("op_9384_begin_0"), val = tensor([0, 0, 0, 240])]; tensor var_9384_end_0 = const()[name = tensor("op_9384_end_0"), val = tensor([2, 77, 1, 320])]; tensor var_9384_end_mask_0 = const()[name = tensor("op_9384_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9384_cast_fp16 = slice_by_index(begin = var_9384_begin_0, end = var_9384_end_0, end_mask = var_9384_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9384_cast_fp16")]; tensor var_9388_begin_0 = const()[name = tensor("op_9388_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_9388_end_0 = const()[name = tensor("op_9388_end_0"), val = tensor([2, 77, 1, 400])]; tensor var_9388_end_mask_0 = const()[name = tensor("op_9388_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9388_cast_fp16 = slice_by_index(begin = var_9388_begin_0, end = var_9388_end_0, end_mask = var_9388_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9388_cast_fp16")]; tensor var_9392_begin_0 = const()[name = tensor("op_9392_begin_0"), val = tensor([0, 0, 0, 400])]; tensor var_9392_end_0 = const()[name = tensor("op_9392_end_0"), val = tensor([2, 77, 1, 480])]; tensor var_9392_end_mask_0 = const()[name = tensor("op_9392_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9392_cast_fp16 = slice_by_index(begin = var_9392_begin_0, end = var_9392_end_0, end_mask = var_9392_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9392_cast_fp16")]; tensor var_9396_begin_0 = const()[name = tensor("op_9396_begin_0"), val = tensor([0, 0, 0, 480])]; tensor var_9396_end_0 = const()[name = tensor("op_9396_end_0"), val = tensor([2, 77, 1, 560])]; tensor var_9396_end_mask_0 = const()[name = tensor("op_9396_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9396_cast_fp16 = slice_by_index(begin = var_9396_begin_0, end = var_9396_end_0, end_mask = var_9396_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9396_cast_fp16")]; tensor var_9400_begin_0 = const()[name = tensor("op_9400_begin_0"), val = tensor([0, 0, 0, 560])]; tensor var_9400_end_0 = const()[name = tensor("op_9400_end_0"), val = tensor([2, 77, 1, 1])]; tensor var_9400_end_mask_0 = const()[name = tensor("op_9400_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9400_cast_fp16 = slice_by_index(begin = var_9400_begin_0, end = var_9400_end_0, end_mask = var_9400_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9400_cast_fp16")]; tensor var_9402_begin_0 = const()[name = tensor("op_9402_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9402_end_0 = const()[name = tensor("op_9402_end_0"), val = tensor([2, 80, 1, 77])]; tensor var_9402_end_mask_0 = const()[name = tensor("op_9402_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9402_cast_fp16 = slice_by_index(begin = var_9402_begin_0, end = var_9402_end_0, end_mask = var_9402_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9402_cast_fp16")]; tensor var_9406_begin_0 = const()[name = tensor("op_9406_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_9406_end_0 = const()[name = tensor("op_9406_end_0"), val = tensor([2, 160, 1, 77])]; tensor var_9406_end_mask_0 = const()[name = tensor("op_9406_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9406_cast_fp16 = slice_by_index(begin = var_9406_begin_0, end = var_9406_end_0, end_mask = var_9406_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9406_cast_fp16")]; tensor var_9410_begin_0 = const()[name = tensor("op_9410_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_9410_end_0 = const()[name = tensor("op_9410_end_0"), val = tensor([2, 240, 1, 77])]; tensor var_9410_end_mask_0 = const()[name = tensor("op_9410_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9410_cast_fp16 = slice_by_index(begin = var_9410_begin_0, end = var_9410_end_0, end_mask = var_9410_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9410_cast_fp16")]; tensor var_9414_begin_0 = const()[name = tensor("op_9414_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_9414_end_0 = const()[name = tensor("op_9414_end_0"), val = tensor([2, 320, 1, 77])]; tensor var_9414_end_mask_0 = const()[name = tensor("op_9414_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9414_cast_fp16 = slice_by_index(begin = var_9414_begin_0, end = var_9414_end_0, end_mask = var_9414_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9414_cast_fp16")]; tensor var_9418_begin_0 = const()[name = tensor("op_9418_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_9418_end_0 = const()[name = tensor("op_9418_end_0"), val = tensor([2, 400, 1, 77])]; tensor var_9418_end_mask_0 = const()[name = tensor("op_9418_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9418_cast_fp16 = slice_by_index(begin = var_9418_begin_0, end = var_9418_end_0, end_mask = var_9418_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9418_cast_fp16")]; tensor var_9422_begin_0 = const()[name = tensor("op_9422_begin_0"), val = tensor([0, 400, 0, 0])]; tensor var_9422_end_0 = const()[name = tensor("op_9422_end_0"), val = tensor([2, 480, 1, 77])]; tensor var_9422_end_mask_0 = const()[name = tensor("op_9422_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9422_cast_fp16 = slice_by_index(begin = var_9422_begin_0, end = var_9422_end_0, end_mask = var_9422_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9422_cast_fp16")]; tensor var_9426_begin_0 = const()[name = tensor("op_9426_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_9426_end_0 = const()[name = tensor("op_9426_end_0"), val = tensor([2, 560, 1, 77])]; tensor var_9426_end_mask_0 = const()[name = tensor("op_9426_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9426_cast_fp16 = slice_by_index(begin = var_9426_begin_0, end = var_9426_end_0, end_mask = var_9426_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9426_cast_fp16")]; tensor var_9430_begin_0 = const()[name = tensor("op_9430_begin_0"), val = tensor([0, 560, 0, 0])]; tensor var_9430_end_0 = const()[name = tensor("op_9430_end_0"), val = tensor([2, 1, 1, 77])]; tensor var_9430_end_mask_0 = const()[name = tensor("op_9430_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9430_cast_fp16 = slice_by_index(begin = var_9430_begin_0, end = var_9430_end_0, end_mask = var_9430_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9430_cast_fp16")]; tensor var_9434_equation_0 = const()[name = tensor("op_9434_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9434_cast_fp16 = einsum(equation = var_9434_equation_0, values = (var_9372_cast_fp16, var_9352_cast_fp16))[name = tensor("op_9434_cast_fp16")]; tensor var_9435_to_fp16 = const()[name = tensor("op_9435_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_737_cast_fp16 = mul(x = var_9434_cast_fp16, y = var_9435_to_fp16)[name = tensor("aw_chunk_737_cast_fp16")]; tensor var_9438_equation_0 = const()[name = tensor("op_9438_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9438_cast_fp16 = einsum(equation = var_9438_equation_0, values = (var_9372_cast_fp16, var_9353_cast_fp16))[name = tensor("op_9438_cast_fp16")]; tensor var_9439_to_fp16 = const()[name = tensor("op_9439_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_739_cast_fp16 = mul(x = var_9438_cast_fp16, y = var_9439_to_fp16)[name = tensor("aw_chunk_739_cast_fp16")]; tensor var_9442_equation_0 = const()[name = tensor("op_9442_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9442_cast_fp16 = einsum(equation = var_9442_equation_0, values = (var_9376_cast_fp16, var_9354_cast_fp16))[name = tensor("op_9442_cast_fp16")]; tensor var_9443_to_fp16 = const()[name = tensor("op_9443_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_741_cast_fp16 = mul(x = var_9442_cast_fp16, y = var_9443_to_fp16)[name = tensor("aw_chunk_741_cast_fp16")]; tensor var_9446_equation_0 = const()[name = tensor("op_9446_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9446_cast_fp16 = einsum(equation = var_9446_equation_0, values = (var_9376_cast_fp16, var_9355_cast_fp16))[name = tensor("op_9446_cast_fp16")]; tensor var_9447_to_fp16 = const()[name = tensor("op_9447_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_743_cast_fp16 = mul(x = var_9446_cast_fp16, y = var_9447_to_fp16)[name = tensor("aw_chunk_743_cast_fp16")]; tensor var_9450_equation_0 = const()[name = tensor("op_9450_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9450_cast_fp16 = einsum(equation = var_9450_equation_0, values = (var_9380_cast_fp16, var_9356_cast_fp16))[name = tensor("op_9450_cast_fp16")]; tensor var_9451_to_fp16 = const()[name = tensor("op_9451_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_745_cast_fp16 = mul(x = var_9450_cast_fp16, y = var_9451_to_fp16)[name = tensor("aw_chunk_745_cast_fp16")]; tensor var_9454_equation_0 = const()[name = tensor("op_9454_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9454_cast_fp16 = einsum(equation = var_9454_equation_0, values = (var_9380_cast_fp16, var_9357_cast_fp16))[name = tensor("op_9454_cast_fp16")]; tensor var_9455_to_fp16 = const()[name = tensor("op_9455_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_747_cast_fp16 = mul(x = var_9454_cast_fp16, y = var_9455_to_fp16)[name = tensor("aw_chunk_747_cast_fp16")]; tensor var_9458_equation_0 = const()[name = tensor("op_9458_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9458_cast_fp16 = einsum(equation = var_9458_equation_0, values = (var_9384_cast_fp16, var_9358_cast_fp16))[name = tensor("op_9458_cast_fp16")]; tensor var_9459_to_fp16 = const()[name = tensor("op_9459_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_749_cast_fp16 = mul(x = var_9458_cast_fp16, y = var_9459_to_fp16)[name = tensor("aw_chunk_749_cast_fp16")]; tensor var_9462_equation_0 = const()[name = tensor("op_9462_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9462_cast_fp16 = einsum(equation = var_9462_equation_0, values = (var_9384_cast_fp16, var_9359_cast_fp16))[name = tensor("op_9462_cast_fp16")]; tensor var_9463_to_fp16 = const()[name = tensor("op_9463_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_751_cast_fp16 = mul(x = var_9462_cast_fp16, y = var_9463_to_fp16)[name = tensor("aw_chunk_751_cast_fp16")]; tensor var_9466_equation_0 = const()[name = tensor("op_9466_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9466_cast_fp16 = einsum(equation = var_9466_equation_0, values = (var_9388_cast_fp16, var_9360_cast_fp16))[name = tensor("op_9466_cast_fp16")]; tensor var_9467_to_fp16 = const()[name = tensor("op_9467_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_753_cast_fp16 = mul(x = var_9466_cast_fp16, y = var_9467_to_fp16)[name = tensor("aw_chunk_753_cast_fp16")]; tensor var_9470_equation_0 = const()[name = tensor("op_9470_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9470_cast_fp16 = einsum(equation = var_9470_equation_0, values = (var_9388_cast_fp16, var_9361_cast_fp16))[name = tensor("op_9470_cast_fp16")]; tensor var_9471_to_fp16 = const()[name = tensor("op_9471_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_755_cast_fp16 = mul(x = var_9470_cast_fp16, y = var_9471_to_fp16)[name = tensor("aw_chunk_755_cast_fp16")]; tensor var_9474_equation_0 = const()[name = tensor("op_9474_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9474_cast_fp16 = einsum(equation = var_9474_equation_0, values = (var_9392_cast_fp16, var_9362_cast_fp16))[name = tensor("op_9474_cast_fp16")]; tensor var_9475_to_fp16 = const()[name = tensor("op_9475_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_757_cast_fp16 = mul(x = var_9474_cast_fp16, y = var_9475_to_fp16)[name = tensor("aw_chunk_757_cast_fp16")]; tensor var_9478_equation_0 = const()[name = tensor("op_9478_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9478_cast_fp16 = einsum(equation = var_9478_equation_0, values = (var_9392_cast_fp16, var_9363_cast_fp16))[name = tensor("op_9478_cast_fp16")]; tensor var_9479_to_fp16 = const()[name = tensor("op_9479_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_759_cast_fp16 = mul(x = var_9478_cast_fp16, y = var_9479_to_fp16)[name = tensor("aw_chunk_759_cast_fp16")]; tensor var_9482_equation_0 = const()[name = tensor("op_9482_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9482_cast_fp16 = einsum(equation = var_9482_equation_0, values = (var_9396_cast_fp16, var_9364_cast_fp16))[name = tensor("op_9482_cast_fp16")]; tensor var_9483_to_fp16 = const()[name = tensor("op_9483_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_761_cast_fp16 = mul(x = var_9482_cast_fp16, y = var_9483_to_fp16)[name = tensor("aw_chunk_761_cast_fp16")]; tensor var_9486_equation_0 = const()[name = tensor("op_9486_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9486_cast_fp16 = einsum(equation = var_9486_equation_0, values = (var_9396_cast_fp16, var_9365_cast_fp16))[name = tensor("op_9486_cast_fp16")]; tensor var_9487_to_fp16 = const()[name = tensor("op_9487_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_763_cast_fp16 = mul(x = var_9486_cast_fp16, y = var_9487_to_fp16)[name = tensor("aw_chunk_763_cast_fp16")]; tensor var_9490_equation_0 = const()[name = tensor("op_9490_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9490_cast_fp16 = einsum(equation = var_9490_equation_0, values = (var_9400_cast_fp16, var_9366_cast_fp16))[name = tensor("op_9490_cast_fp16")]; tensor var_9491_to_fp16 = const()[name = tensor("op_9491_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_765_cast_fp16 = mul(x = var_9490_cast_fp16, y = var_9491_to_fp16)[name = tensor("aw_chunk_765_cast_fp16")]; tensor var_9494_equation_0 = const()[name = tensor("op_9494_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9494_cast_fp16 = einsum(equation = var_9494_equation_0, values = (var_9400_cast_fp16, var_9367_cast_fp16))[name = tensor("op_9494_cast_fp16")]; tensor var_9495_to_fp16 = const()[name = tensor("op_9495_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_767_cast_fp16 = mul(x = var_9494_cast_fp16, y = var_9495_to_fp16)[name = tensor("aw_chunk_767_cast_fp16")]; tensor var_9497_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_737_cast_fp16)[name = tensor("op_9497_cast_fp16")]; tensor var_9498_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_739_cast_fp16)[name = tensor("op_9498_cast_fp16")]; tensor var_9499_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_741_cast_fp16)[name = tensor("op_9499_cast_fp16")]; tensor var_9500_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_743_cast_fp16)[name = tensor("op_9500_cast_fp16")]; tensor var_9501_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_745_cast_fp16)[name = tensor("op_9501_cast_fp16")]; tensor var_9502_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_747_cast_fp16)[name = tensor("op_9502_cast_fp16")]; tensor var_9503_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_749_cast_fp16)[name = tensor("op_9503_cast_fp16")]; tensor var_9504_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_751_cast_fp16)[name = tensor("op_9504_cast_fp16")]; tensor var_9505_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_753_cast_fp16)[name = tensor("op_9505_cast_fp16")]; tensor var_9506_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_755_cast_fp16)[name = tensor("op_9506_cast_fp16")]; tensor var_9507_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_757_cast_fp16)[name = tensor("op_9507_cast_fp16")]; tensor var_9508_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_759_cast_fp16)[name = tensor("op_9508_cast_fp16")]; tensor var_9509_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_761_cast_fp16)[name = tensor("op_9509_cast_fp16")]; tensor var_9510_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_763_cast_fp16)[name = tensor("op_9510_cast_fp16")]; tensor var_9511_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_765_cast_fp16)[name = tensor("op_9511_cast_fp16")]; tensor var_9512_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_767_cast_fp16)[name = tensor("op_9512_cast_fp16")]; tensor var_9514_equation_0 = const()[name = tensor("op_9514_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9514_cast_fp16 = einsum(equation = var_9514_equation_0, values = (var_9402_cast_fp16, var_9497_cast_fp16))[name = tensor("op_9514_cast_fp16")]; tensor var_9516_equation_0 = const()[name = tensor("op_9516_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9516_cast_fp16 = einsum(equation = var_9516_equation_0, values = (var_9402_cast_fp16, var_9498_cast_fp16))[name = tensor("op_9516_cast_fp16")]; tensor var_9518_equation_0 = const()[name = tensor("op_9518_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9518_cast_fp16 = einsum(equation = var_9518_equation_0, values = (var_9406_cast_fp16, var_9499_cast_fp16))[name = tensor("op_9518_cast_fp16")]; tensor var_9520_equation_0 = const()[name = tensor("op_9520_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9520_cast_fp16 = einsum(equation = var_9520_equation_0, values = (var_9406_cast_fp16, var_9500_cast_fp16))[name = tensor("op_9520_cast_fp16")]; tensor var_9522_equation_0 = const()[name = tensor("op_9522_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9522_cast_fp16 = einsum(equation = var_9522_equation_0, values = (var_9410_cast_fp16, var_9501_cast_fp16))[name = tensor("op_9522_cast_fp16")]; tensor var_9524_equation_0 = const()[name = tensor("op_9524_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9524_cast_fp16 = einsum(equation = var_9524_equation_0, values = (var_9410_cast_fp16, var_9502_cast_fp16))[name = tensor("op_9524_cast_fp16")]; tensor var_9526_equation_0 = const()[name = tensor("op_9526_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9526_cast_fp16 = einsum(equation = var_9526_equation_0, values = (var_9414_cast_fp16, var_9503_cast_fp16))[name = tensor("op_9526_cast_fp16")]; tensor var_9528_equation_0 = const()[name = tensor("op_9528_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9528_cast_fp16 = einsum(equation = var_9528_equation_0, values = (var_9414_cast_fp16, var_9504_cast_fp16))[name = tensor("op_9528_cast_fp16")]; tensor var_9530_equation_0 = const()[name = tensor("op_9530_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9530_cast_fp16 = einsum(equation = var_9530_equation_0, values = (var_9418_cast_fp16, var_9505_cast_fp16))[name = tensor("op_9530_cast_fp16")]; tensor var_9532_equation_0 = const()[name = tensor("op_9532_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9532_cast_fp16 = einsum(equation = var_9532_equation_0, values = (var_9418_cast_fp16, var_9506_cast_fp16))[name = tensor("op_9532_cast_fp16")]; tensor var_9534_equation_0 = const()[name = tensor("op_9534_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9534_cast_fp16 = einsum(equation = var_9534_equation_0, values = (var_9422_cast_fp16, var_9507_cast_fp16))[name = tensor("op_9534_cast_fp16")]; tensor var_9536_equation_0 = const()[name = tensor("op_9536_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9536_cast_fp16 = einsum(equation = var_9536_equation_0, values = (var_9422_cast_fp16, var_9508_cast_fp16))[name = tensor("op_9536_cast_fp16")]; tensor var_9538_equation_0 = const()[name = tensor("op_9538_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9538_cast_fp16 = einsum(equation = var_9538_equation_0, values = (var_9426_cast_fp16, var_9509_cast_fp16))[name = tensor("op_9538_cast_fp16")]; tensor var_9540_equation_0 = const()[name = tensor("op_9540_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9540_cast_fp16 = einsum(equation = var_9540_equation_0, values = (var_9426_cast_fp16, var_9510_cast_fp16))[name = tensor("op_9540_cast_fp16")]; tensor var_9542_equation_0 = const()[name = tensor("op_9542_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9542_cast_fp16 = einsum(equation = var_9542_equation_0, values = (var_9430_cast_fp16, var_9511_cast_fp16))[name = tensor("op_9542_cast_fp16")]; tensor var_9544_equation_0 = const()[name = tensor("op_9544_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9544_cast_fp16 = einsum(equation = var_9544_equation_0, values = (var_9430_cast_fp16, var_9512_cast_fp16))[name = tensor("op_9544_cast_fp16")]; tensor var_9546_interleave_0 = const()[name = tensor("op_9546_interleave_0"), val = tensor(false)]; tensor var_9546_cast_fp16 = concat(axis = var_8160, interleave = var_9546_interleave_0, values = (var_9514_cast_fp16, var_9516_cast_fp16))[name = tensor("op_9546_cast_fp16")]; tensor var_9548_interleave_0 = const()[name = tensor("op_9548_interleave_0"), val = tensor(false)]; tensor var_9548_cast_fp16 = concat(axis = var_8160, interleave = var_9548_interleave_0, values = (var_9518_cast_fp16, var_9520_cast_fp16))[name = tensor("op_9548_cast_fp16")]; tensor var_9550_interleave_0 = const()[name = tensor("op_9550_interleave_0"), val = tensor(false)]; tensor var_9550_cast_fp16 = concat(axis = var_8160, interleave = var_9550_interleave_0, values = (var_9522_cast_fp16, var_9524_cast_fp16))[name = tensor("op_9550_cast_fp16")]; tensor var_9552_interleave_0 = const()[name = tensor("op_9552_interleave_0"), val = tensor(false)]; tensor var_9552_cast_fp16 = concat(axis = var_8160, interleave = var_9552_interleave_0, values = (var_9526_cast_fp16, var_9528_cast_fp16))[name = tensor("op_9552_cast_fp16")]; tensor var_9554_interleave_0 = const()[name = tensor("op_9554_interleave_0"), val = tensor(false)]; tensor var_9554_cast_fp16 = concat(axis = var_8160, interleave = var_9554_interleave_0, values = (var_9530_cast_fp16, var_9532_cast_fp16))[name = tensor("op_9554_cast_fp16")]; tensor var_9556_interleave_0 = const()[name = tensor("op_9556_interleave_0"), val = tensor(false)]; tensor var_9556_cast_fp16 = concat(axis = var_8160, interleave = var_9556_interleave_0, values = (var_9534_cast_fp16, var_9536_cast_fp16))[name = tensor("op_9556_cast_fp16")]; tensor var_9558_interleave_0 = const()[name = tensor("op_9558_interleave_0"), val = tensor(false)]; tensor var_9558_cast_fp16 = concat(axis = var_8160, interleave = var_9558_interleave_0, values = (var_9538_cast_fp16, var_9540_cast_fp16))[name = tensor("op_9558_cast_fp16")]; tensor var_9560_interleave_0 = const()[name = tensor("op_9560_interleave_0"), val = tensor(false)]; tensor var_9560_cast_fp16 = concat(axis = var_8160, interleave = var_9560_interleave_0, values = (var_9542_cast_fp16, var_9544_cast_fp16))[name = tensor("op_9560_cast_fp16")]; tensor input_411_interleave_0 = const()[name = tensor("input_411_interleave_0"), val = tensor(false)]; tensor input_411_cast_fp16 = concat(axis = var_8182, interleave = input_411_interleave_0, values = (var_9546_cast_fp16, var_9548_cast_fp16, var_9550_cast_fp16, var_9552_cast_fp16, var_9554_cast_fp16, var_9556_cast_fp16, var_9558_cast_fp16, var_9560_cast_fp16))[name = tensor("input_411_cast_fp16")]; tensor var_9570_pad_type_0 = const()[name = tensor("op_9570_pad_type_0"), val = tensor("valid")]; tensor var_9570_strides_0 = const()[name = tensor("op_9570_strides_0"), val = tensor([1, 1])]; tensor var_9570_pad_0 = const()[name = tensor("op_9570_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_9570_dilations_0 = const()[name = tensor("op_9570_dilations_0"), val = tensor([1, 1])]; tensor var_9570_groups_0 = const()[name = tensor("op_9570_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(735360576)))]; tensor up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(736179840)))]; tensor var_9570_cast_fp16 = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_9570_dilations_0, groups = var_9570_groups_0, pad = var_9570_pad_0, pad_type = var_9570_pad_type_0, strides = var_9570_strides_0, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_411_cast_fp16)[name = tensor("op_9570_cast_fp16")]; tensor inputs_71_cast_fp16 = add(x = var_9570_cast_fp16, y = inputs_69_cast_fp16)[name = tensor("inputs_71_cast_fp16")]; tensor input_413_axes_0 = const()[name = tensor("input_413_axes_0"), val = tensor([1])]; tensor input_413_gamma_0_to_fp16 = const()[name = tensor("input_413_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(736181184)))]; tensor input_413_beta_0_to_fp16 = const()[name = tensor("input_413_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(736182528)))]; tensor var_9580_to_fp16 = const()[name = tensor("op_9580_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_413_cast_fp16 = layer_norm(axes = input_413_axes_0, beta = input_413_beta_0_to_fp16, epsilon = var_9580_to_fp16, gamma = input_413_gamma_0_to_fp16, x = inputs_71_cast_fp16)[name = tensor("input_413_cast_fp16")]; tensor var_9600_pad_type_0 = const()[name = tensor("op_9600_pad_type_0"), val = tensor("valid")]; tensor var_9600_strides_0 = const()[name = tensor("op_9600_strides_0"), val = tensor([1, 1])]; tensor var_9600_pad_0 = const()[name = tensor("op_9600_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_9600_dilations_0 = const()[name = tensor("op_9600_dilations_0"), val = tensor([1, 1])]; tensor var_9600_groups_0 = const()[name = tensor("op_9600_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(736183872)))]; tensor up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(742737536)))]; tensor var_9600_cast_fp16 = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_9600_dilations_0, groups = var_9600_groups_0, pad = var_9600_pad_0, pad_type = var_9600_pad_type_0, strides = var_9600_strides_0, weight = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_413_cast_fp16)[name = tensor("op_9600_cast_fp16")]; tensor var_9601_split_sizes_0 = const()[name = tensor("op_9601_split_sizes_0"), val = tensor([2560, 2560])]; tensor var_9601_axis_0 = const()[name = tensor("op_9601_axis_0"), val = tensor(1)]; tensor var_9601_cast_fp16_0, tensor var_9601_cast_fp16_1 = split(axis = var_9601_axis_0, split_sizes = var_9601_split_sizes_0, x = var_9600_cast_fp16)[name = tensor("op_9601_cast_fp16")]; tensor var_9603_mode_0 = const()[name = tensor("op_9603_mode_0"), val = tensor("EXACT")]; tensor var_9603_cast_fp16 = gelu(mode = var_9603_mode_0, x = var_9601_cast_fp16_1)[name = tensor("op_9603_cast_fp16")]; tensor input_415_cast_fp16 = mul(x = var_9601_cast_fp16_0, y = var_9603_cast_fp16)[name = tensor("input_415_cast_fp16")]; tensor var_9611_pad_type_0 = const()[name = tensor("op_9611_pad_type_0"), val = tensor("valid")]; tensor var_9611_strides_0 = const()[name = tensor("op_9611_strides_0"), val = tensor([1, 1])]; tensor var_9611_pad_0 = const()[name = tensor("op_9611_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_9611_dilations_0 = const()[name = tensor("op_9611_dilations_0"), val = tensor([1, 1])]; tensor var_9611_groups_0 = const()[name = tensor("op_9611_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(742747840)))]; tensor up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(746024704)))]; tensor var_9611_cast_fp16 = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_9611_dilations_0, groups = var_9611_groups_0, pad = var_9611_pad_0, pad_type = var_9611_pad_type_0, strides = var_9611_strides_0, weight = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_415_cast_fp16)[name = tensor("op_9611_cast_fp16")]; tensor hidden_states_253_cast_fp16 = add(x = var_9611_cast_fp16, y = inputs_71_cast_fp16)[name = tensor("hidden_states_253_cast_fp16")]; tensor var_9613 = const()[name = tensor("op_9613"), val = tensor([2, 640, 32, 32])]; tensor input_417_cast_fp16 = reshape(shape = var_9613, x = hidden_states_253_cast_fp16)[name = tensor("input_417_cast_fp16")]; tensor hidden_states_255_pad_type_0 = const()[name = tensor("hidden_states_255_pad_type_0"), val = tensor("valid")]; tensor hidden_states_255_strides_0 = const()[name = tensor("hidden_states_255_strides_0"), val = tensor([1, 1])]; tensor hidden_states_255_pad_0 = const()[name = tensor("hidden_states_255_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_255_dilations_0 = const()[name = tensor("hidden_states_255_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_255_groups_0 = const()[name = tensor("hidden_states_255_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_1_proj_out_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(746026048)))]; tensor up_blocks_2_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(746845312)))]; tensor hidden_states_255_cast_fp16 = conv(bias = up_blocks_2_attentions_1_proj_out_bias_to_fp16, dilations = hidden_states_255_dilations_0, groups = hidden_states_255_groups_0, pad = hidden_states_255_pad_0, pad_type = hidden_states_255_pad_type_0, strides = hidden_states_255_strides_0, weight = up_blocks_2_attentions_1_proj_out_weight_to_fp16, x = input_417_cast_fp16)[name = tensor("hidden_states_255_cast_fp16")]; tensor hidden_states_257_cast_fp16 = add(x = hidden_states_255_cast_fp16, y = hidden_states_243_cast_fp16)[name = tensor("hidden_states_257_cast_fp16")]; tensor input_419_interleave_0 = const()[name = tensor("input_419_interleave_0"), val = tensor(false)]; tensor cast_7 = cast(dtype = cast_7_dtype_0, x = input_63_cast_fp16)[name = tensor("cast_9")]; tensor input_419_cast_fp16 = concat(axis = var_8182, interleave = input_419_interleave_0, values = (hidden_states_257_cast_fp16, cast_7))[name = tensor("input_419_cast_fp16")]; tensor reshape_192_shape_0 = const()[name = tensor("reshape_192_shape_0"), val = tensor([2, 32, 30, 32, 32])]; tensor reshape_192_cast_fp16 = reshape(shape = reshape_192_shape_0, x = input_419_cast_fp16)[name = tensor("reshape_192_cast_fp16")]; tensor reduce_mean_144_axes_0 = const()[name = tensor("reduce_mean_144_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_144_keep_dims_0 = const()[name = tensor("reduce_mean_144_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_144_cast_fp16 = reduce_mean(axes = reduce_mean_144_axes_0, keep_dims = reduce_mean_144_keep_dims_0, x = reshape_192_cast_fp16)[name = tensor("reduce_mean_144_cast_fp16")]; tensor sub_96_cast_fp16 = sub(x = reshape_192_cast_fp16, y = reduce_mean_144_cast_fp16)[name = tensor("sub_96_cast_fp16")]; tensor square_48_cast_fp16 = square(x = sub_96_cast_fp16)[name = tensor("square_48_cast_fp16")]; tensor reduce_mean_146_axes_0 = const()[name = tensor("reduce_mean_146_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_146_keep_dims_0 = const()[name = tensor("reduce_mean_146_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_146_cast_fp16 = reduce_mean(axes = reduce_mean_146_axes_0, keep_dims = reduce_mean_146_keep_dims_0, x = square_48_cast_fp16)[name = tensor("reduce_mean_146_cast_fp16")]; tensor add_96_y_0_to_fp16 = const()[name = tensor("add_96_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_96_cast_fp16 = add(x = reduce_mean_146_cast_fp16, y = add_96_y_0_to_fp16)[name = tensor("add_96_cast_fp16")]; tensor sqrt_48_cast_fp16 = sqrt(x = add_96_cast_fp16)[name = tensor("sqrt_48_cast_fp16")]; tensor real_div_48_cast_fp16 = real_div(x = sub_96_cast_fp16, y = sqrt_48_cast_fp16)[name = tensor("real_div_48_cast_fp16")]; tensor reshape_193_shape_0 = const()[name = tensor("reshape_193_shape_0"), val = tensor([2, 960, 32, 32])]; tensor reshape_193_cast_fp16 = reshape(shape = reshape_193_shape_0, x = real_div_48_cast_fp16)[name = tensor("reshape_193_cast_fp16")]; tensor add_97_mean_0_to_fp16 = const()[name = tensor("add_97_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(746846656)))]; tensor add_97_variance_0_to_fp16 = const()[name = tensor("add_97_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(746848640)))]; tensor add_97_gamma_0_to_fp16 = const()[name = tensor("add_97_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(746850624)))]; tensor add_97_beta_0_to_fp16 = const()[name = tensor("add_97_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(746852608)))]; tensor add_97_epsilon_0_to_fp16 = const()[name = tensor("add_97_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_97_cast_fp16 = batch_norm(beta = add_97_beta_0_to_fp16, epsilon = add_97_epsilon_0_to_fp16, gamma = add_97_gamma_0_to_fp16, mean = add_97_mean_0_to_fp16, variance = add_97_variance_0_to_fp16, x = reshape_193_cast_fp16)[name = tensor("add_97_cast_fp16")]; tensor input_423_cast_fp16 = silu(x = add_97_cast_fp16)[name = tensor("input_423_cast_fp16")]; tensor hidden_states_259_pad_type_0 = const()[name = tensor("hidden_states_259_pad_type_0"), val = tensor("custom")]; tensor hidden_states_259_pad_0 = const()[name = tensor("hidden_states_259_pad_0"), val = tensor([1, 1, 1, 1])]; tensor hidden_states_259_strides_0 = const()[name = tensor("hidden_states_259_strides_0"), val = tensor([1, 1])]; tensor hidden_states_259_dilations_0 = const()[name = tensor("hidden_states_259_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_259_groups_0 = const()[name = tensor("hidden_states_259_groups_0"), val = tensor(1)]; tensor up_blocks_2_resnets_2_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(746854592)))]; tensor up_blocks_2_resnets_2_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(757913856)))]; tensor hidden_states_259_cast_fp16 = conv(bias = up_blocks_2_resnets_2_conv1_bias_to_fp16, dilations = hidden_states_259_dilations_0, groups = hidden_states_259_groups_0, pad = hidden_states_259_pad_0, pad_type = hidden_states_259_pad_type_0, strides = hidden_states_259_strides_0, weight = up_blocks_2_resnets_2_conv1_weight_to_fp16, x = input_423_cast_fp16)[name = tensor("hidden_states_259_cast_fp16")]; tensor temb_37_pad_type_0 = const()[name = tensor("temb_37_pad_type_0"), val = tensor("valid")]; tensor temb_37_strides_0 = const()[name = tensor("temb_37_strides_0"), val = tensor([1, 1])]; tensor temb_37_pad_0 = const()[name = tensor("temb_37_pad_0"), val = tensor([0, 0, 0, 0])]; tensor temb_37_dilations_0 = const()[name = tensor("temb_37_dilations_0"), val = tensor([1, 1])]; tensor temb_37_groups_0 = const()[name = tensor("temb_37_groups_0"), val = tensor(1)]; tensor up_blocks_2_resnets_2_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(757915200)))]; tensor up_blocks_2_resnets_2_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(759553664)))]; tensor temb_37_cast_fp16 = conv(bias = up_blocks_2_resnets_2_time_emb_proj_bias_to_fp16, dilations = temb_37_dilations_0, groups = temb_37_groups_0, pad = temb_37_pad_0, pad_type = temb_37_pad_type_0, strides = temb_37_strides_0, weight = up_blocks_2_resnets_2_time_emb_proj_weight_to_fp16, x = cast_9)[name = tensor("temb_37_cast_fp16")]; tensor input_427_cast_fp16 = add(x = hidden_states_259_cast_fp16, y = temb_37_cast_fp16)[name = tensor("input_427_cast_fp16")]; tensor reshape_196_shape_0 = const()[name = tensor("reshape_196_shape_0"), val = tensor([2, 32, 20, 32, 32])]; tensor reshape_196_cast_fp16 = reshape(shape = reshape_196_shape_0, x = input_427_cast_fp16)[name = tensor("reshape_196_cast_fp16")]; tensor reduce_mean_147_axes_0 = const()[name = tensor("reduce_mean_147_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_147_keep_dims_0 = const()[name = tensor("reduce_mean_147_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_147_cast_fp16 = reduce_mean(axes = reduce_mean_147_axes_0, keep_dims = reduce_mean_147_keep_dims_0, x = reshape_196_cast_fp16)[name = tensor("reduce_mean_147_cast_fp16")]; tensor sub_98_cast_fp16 = sub(x = reshape_196_cast_fp16, y = reduce_mean_147_cast_fp16)[name = tensor("sub_98_cast_fp16")]; tensor square_49_cast_fp16 = square(x = sub_98_cast_fp16)[name = tensor("square_49_cast_fp16")]; tensor reduce_mean_149_axes_0 = const()[name = tensor("reduce_mean_149_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_149_keep_dims_0 = const()[name = tensor("reduce_mean_149_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_149_cast_fp16 = reduce_mean(axes = reduce_mean_149_axes_0, keep_dims = reduce_mean_149_keep_dims_0, x = square_49_cast_fp16)[name = tensor("reduce_mean_149_cast_fp16")]; tensor add_98_y_0_to_fp16 = const()[name = tensor("add_98_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_98_cast_fp16 = add(x = reduce_mean_149_cast_fp16, y = add_98_y_0_to_fp16)[name = tensor("add_98_cast_fp16")]; tensor sqrt_49_cast_fp16 = sqrt(x = add_98_cast_fp16)[name = tensor("sqrt_49_cast_fp16")]; tensor real_div_49_cast_fp16 = real_div(x = sub_98_cast_fp16, y = sqrt_49_cast_fp16)[name = tensor("real_div_49_cast_fp16")]; tensor reshape_197_shape_0 = const()[name = tensor("reshape_197_shape_0"), val = tensor([2, 640, 32, 32])]; tensor reshape_197_cast_fp16 = reshape(shape = reshape_197_shape_0, x = real_div_49_cast_fp16)[name = tensor("reshape_197_cast_fp16")]; tensor add_99_gamma_0_to_fp16 = const()[name = tensor("add_99_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(759555008)))]; tensor add_99_beta_0_to_fp16 = const()[name = tensor("add_99_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(759556352)))]; tensor add_99_epsilon_0_to_fp16 = const()[name = tensor("add_99_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_99_cast_fp16 = batch_norm(beta = add_99_beta_0_to_fp16, epsilon = add_99_epsilon_0_to_fp16, gamma = add_99_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_197_cast_fp16)[name = tensor("add_99_cast_fp16")]; tensor input_431_cast_fp16 = silu(x = add_99_cast_fp16)[name = tensor("input_431_cast_fp16")]; tensor hidden_states_261_pad_type_0 = const()[name = tensor("hidden_states_261_pad_type_0"), val = tensor("custom")]; tensor hidden_states_261_pad_0 = const()[name = tensor("hidden_states_261_pad_0"), val = tensor([1, 1, 1, 1])]; tensor hidden_states_261_strides_0 = const()[name = tensor("hidden_states_261_strides_0"), val = tensor([1, 1])]; tensor hidden_states_261_dilations_0 = const()[name = tensor("hidden_states_261_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_261_groups_0 = const()[name = tensor("hidden_states_261_groups_0"), val = tensor(1)]; tensor up_blocks_2_resnets_2_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(759557696)))]; tensor up_blocks_2_resnets_2_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(766930560)))]; tensor hidden_states_261_cast_fp16 = conv(bias = up_blocks_2_resnets_2_conv2_bias_to_fp16, dilations = hidden_states_261_dilations_0, groups = hidden_states_261_groups_0, pad = hidden_states_261_pad_0, pad_type = hidden_states_261_pad_type_0, strides = hidden_states_261_strides_0, weight = up_blocks_2_resnets_2_conv2_weight_to_fp16, x = input_431_cast_fp16)[name = tensor("hidden_states_261_cast_fp16")]; tensor x_21_pad_type_0 = const()[name = tensor("x_21_pad_type_0"), val = tensor("valid")]; tensor x_21_strides_0 = const()[name = tensor("x_21_strides_0"), val = tensor([1, 1])]; tensor x_21_pad_0 = const()[name = tensor("x_21_pad_0"), val = tensor([0, 0, 0, 0])]; tensor x_21_dilations_0 = const()[name = tensor("x_21_dilations_0"), val = tensor([1, 1])]; tensor x_21_groups_0 = const()[name = tensor("x_21_groups_0"), val = tensor(1)]; tensor up_blocks_2_resnets_2_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(766931904)))]; tensor up_blocks_2_resnets_2_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(768160768)))]; tensor x_21_cast_fp16 = conv(bias = up_blocks_2_resnets_2_conv_shortcut_bias_to_fp16, dilations = x_21_dilations_0, groups = x_21_groups_0, pad = x_21_pad_0, pad_type = x_21_pad_type_0, strides = x_21_strides_0, weight = up_blocks_2_resnets_2_conv_shortcut_weight_to_fp16, x = input_419_cast_fp16)[name = tensor("x_21_cast_fp16")]; tensor hidden_states_263_cast_fp16 = add(x = x_21_cast_fp16, y = hidden_states_261_cast_fp16)[name = tensor("hidden_states_263_cast_fp16")]; tensor reshape_200_shape_0 = const()[name = tensor("reshape_200_shape_0"), val = tensor([2, 32, 20, 32, 32])]; tensor reshape_200_cast_fp16 = reshape(shape = reshape_200_shape_0, x = hidden_states_263_cast_fp16)[name = tensor("reshape_200_cast_fp16")]; tensor reduce_mean_150_axes_0 = const()[name = tensor("reduce_mean_150_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_150_keep_dims_0 = const()[name = tensor("reduce_mean_150_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_150_cast_fp16 = reduce_mean(axes = reduce_mean_150_axes_0, keep_dims = reduce_mean_150_keep_dims_0, x = reshape_200_cast_fp16)[name = tensor("reduce_mean_150_cast_fp16")]; tensor sub_100_cast_fp16 = sub(x = reshape_200_cast_fp16, y = reduce_mean_150_cast_fp16)[name = tensor("sub_100_cast_fp16")]; tensor square_50_cast_fp16 = square(x = sub_100_cast_fp16)[name = tensor("square_50_cast_fp16")]; tensor reduce_mean_152_axes_0 = const()[name = tensor("reduce_mean_152_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_152_keep_dims_0 = const()[name = tensor("reduce_mean_152_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_152_cast_fp16 = reduce_mean(axes = reduce_mean_152_axes_0, keep_dims = reduce_mean_152_keep_dims_0, x = square_50_cast_fp16)[name = tensor("reduce_mean_152_cast_fp16")]; tensor add_100_y_0_to_fp16 = const()[name = tensor("add_100_y_0_to_fp16"), val = tensor(0x1.1p-20)]; tensor add_100_cast_fp16 = add(x = reduce_mean_152_cast_fp16, y = add_100_y_0_to_fp16)[name = tensor("add_100_cast_fp16")]; tensor sqrt_50_cast_fp16 = sqrt(x = add_100_cast_fp16)[name = tensor("sqrt_50_cast_fp16")]; tensor real_div_50_cast_fp16 = real_div(x = sub_100_cast_fp16, y = sqrt_50_cast_fp16)[name = tensor("real_div_50_cast_fp16")]; tensor reshape_201_shape_0 = const()[name = tensor("reshape_201_shape_0"), val = tensor([2, 640, 32, 32])]; tensor reshape_201_cast_fp16 = reshape(shape = reshape_201_shape_0, x = real_div_50_cast_fp16)[name = tensor("reshape_201_cast_fp16")]; tensor add_101_gamma_0_to_fp16 = const()[name = tensor("add_101_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(768162112)))]; tensor add_101_beta_0_to_fp16 = const()[name = tensor("add_101_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(768163456)))]; tensor add_101_epsilon_0_to_fp16 = const()[name = tensor("add_101_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_101_cast_fp16 = batch_norm(beta = add_101_beta_0_to_fp16, epsilon = add_101_epsilon_0_to_fp16, gamma = add_101_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_201_cast_fp16)[name = tensor("add_101_cast_fp16")]; tensor hidden_states_265_pad_type_0 = const()[name = tensor("hidden_states_265_pad_type_0"), val = tensor("valid")]; tensor hidden_states_265_strides_0 = const()[name = tensor("hidden_states_265_strides_0"), val = tensor([1, 1])]; tensor hidden_states_265_pad_0 = const()[name = tensor("hidden_states_265_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_265_dilations_0 = const()[name = tensor("hidden_states_265_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_265_groups_0 = const()[name = tensor("hidden_states_265_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_2_proj_in_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(768164800)))]; tensor up_blocks_2_attentions_2_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(768984064)))]; tensor hidden_states_265_cast_fp16 = conv(bias = up_blocks_2_attentions_2_proj_in_bias_to_fp16, dilations = hidden_states_265_dilations_0, groups = hidden_states_265_groups_0, pad = hidden_states_265_pad_0, pad_type = hidden_states_265_pad_type_0, strides = hidden_states_265_strides_0, weight = up_blocks_2_attentions_2_proj_in_weight_to_fp16, x = add_101_cast_fp16)[name = tensor("hidden_states_265_cast_fp16")]; tensor var_9693 = const()[name = tensor("op_9693"), val = tensor([2, 640, 1, 1024])]; tensor inputs_73_cast_fp16 = reshape(shape = var_9693, x = hidden_states_265_cast_fp16)[name = tensor("inputs_73_cast_fp16")]; tensor hidden_states_267_axes_0 = const()[name = tensor("hidden_states_267_axes_0"), val = tensor([1])]; tensor hidden_states_267_gamma_0_to_fp16 = const()[name = tensor("hidden_states_267_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(768985408)))]; tensor hidden_states_267_beta_0_to_fp16 = const()[name = tensor("hidden_states_267_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(768986752)))]; tensor var_9709_to_fp16 = const()[name = tensor("op_9709_to_fp16"), val = tensor(0x1.5p-17)]; tensor hidden_states_267_cast_fp16 = layer_norm(axes = hidden_states_267_axes_0, beta = hidden_states_267_beta_0_to_fp16, epsilon = var_9709_to_fp16, gamma = hidden_states_267_gamma_0_to_fp16, x = inputs_73_cast_fp16)[name = tensor("hidden_states_267_cast_fp16")]; tensor q_49_pad_type_0 = const()[name = tensor("q_49_pad_type_0"), val = tensor("valid")]; tensor q_49_strides_0 = const()[name = tensor("q_49_strides_0"), val = tensor([1, 1])]; tensor q_49_pad_0 = const()[name = tensor("q_49_pad_0"), val = tensor([0, 0, 0, 0])]; tensor q_49_dilations_0 = const()[name = tensor("q_49_dilations_0"), val = tensor([1, 1])]; tensor q_49_groups_0 = const()[name = tensor("q_49_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(768988096)))]; tensor q_49_cast_fp16 = conv(dilations = q_49_dilations_0, groups = q_49_groups_0, pad = q_49_pad_0, pad_type = q_49_pad_type_0, strides = q_49_strides_0, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_267_cast_fp16)[name = tensor("q_49_cast_fp16")]; tensor k_97_pad_type_0 = const()[name = tensor("k_97_pad_type_0"), val = tensor("valid")]; tensor k_97_strides_0 = const()[name = tensor("k_97_strides_0"), val = tensor([1, 1])]; tensor k_97_pad_0 = const()[name = tensor("k_97_pad_0"), val = tensor([0, 0, 0, 0])]; tensor k_97_dilations_0 = const()[name = tensor("k_97_dilations_0"), val = tensor([1, 1])]; tensor k_97_groups_0 = const()[name = tensor("k_97_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(769807360)))]; tensor k_97_cast_fp16 = conv(dilations = k_97_dilations_0, groups = k_97_groups_0, pad = k_97_pad_0, pad_type = k_97_pad_type_0, strides = k_97_strides_0, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_267_cast_fp16)[name = tensor("k_97_cast_fp16")]; tensor v_49_pad_type_0 = const()[name = tensor("v_49_pad_type_0"), val = tensor("valid")]; tensor v_49_strides_0 = const()[name = tensor("v_49_strides_0"), val = tensor([1, 1])]; tensor v_49_pad_0 = const()[name = tensor("v_49_pad_0"), val = tensor([0, 0, 0, 0])]; tensor v_49_dilations_0 = const()[name = tensor("v_49_dilations_0"), val = tensor([1, 1])]; tensor v_49_groups_0 = const()[name = tensor("v_49_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(770626624)))]; tensor v_49_cast_fp16 = conv(dilations = v_49_dilations_0, groups = v_49_groups_0, pad = v_49_pad_0, pad_type = v_49_pad_type_0, strides = v_49_strides_0, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_267_cast_fp16)[name = tensor("v_49_cast_fp16")]; tensor var_9742_begin_0 = const()[name = tensor("op_9742_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9742_end_0 = const()[name = tensor("op_9742_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_9742_end_mask_0 = const()[name = tensor("op_9742_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9742_cast_fp16 = slice_by_index(begin = var_9742_begin_0, end = var_9742_end_0, end_mask = var_9742_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_9742_cast_fp16")]; tensor var_9746_begin_0 = const()[name = tensor("op_9746_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_9746_end_0 = const()[name = tensor("op_9746_end_0"), val = tensor([2, 160, 1, 1024])]; tensor var_9746_end_mask_0 = const()[name = tensor("op_9746_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9746_cast_fp16 = slice_by_index(begin = var_9746_begin_0, end = var_9746_end_0, end_mask = var_9746_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_9746_cast_fp16")]; tensor var_9750_begin_0 = const()[name = tensor("op_9750_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_9750_end_0 = const()[name = tensor("op_9750_end_0"), val = tensor([2, 240, 1, 1024])]; tensor var_9750_end_mask_0 = const()[name = tensor("op_9750_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9750_cast_fp16 = slice_by_index(begin = var_9750_begin_0, end = var_9750_end_0, end_mask = var_9750_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_9750_cast_fp16")]; tensor var_9754_begin_0 = const()[name = tensor("op_9754_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_9754_end_0 = const()[name = tensor("op_9754_end_0"), val = tensor([2, 320, 1, 1024])]; tensor var_9754_end_mask_0 = const()[name = tensor("op_9754_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9754_cast_fp16 = slice_by_index(begin = var_9754_begin_0, end = var_9754_end_0, end_mask = var_9754_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_9754_cast_fp16")]; tensor var_9758_begin_0 = const()[name = tensor("op_9758_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_9758_end_0 = const()[name = tensor("op_9758_end_0"), val = tensor([2, 400, 1, 1024])]; tensor var_9758_end_mask_0 = const()[name = tensor("op_9758_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9758_cast_fp16 = slice_by_index(begin = var_9758_begin_0, end = var_9758_end_0, end_mask = var_9758_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_9758_cast_fp16")]; tensor var_9762_begin_0 = const()[name = tensor("op_9762_begin_0"), val = tensor([0, 400, 0, 0])]; tensor var_9762_end_0 = const()[name = tensor("op_9762_end_0"), val = tensor([2, 480, 1, 1024])]; tensor var_9762_end_mask_0 = const()[name = tensor("op_9762_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9762_cast_fp16 = slice_by_index(begin = var_9762_begin_0, end = var_9762_end_0, end_mask = var_9762_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_9762_cast_fp16")]; tensor var_9766_begin_0 = const()[name = tensor("op_9766_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_9766_end_0 = const()[name = tensor("op_9766_end_0"), val = tensor([2, 560, 1, 1024])]; tensor var_9766_end_mask_0 = const()[name = tensor("op_9766_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9766_cast_fp16 = slice_by_index(begin = var_9766_begin_0, end = var_9766_end_0, end_mask = var_9766_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_9766_cast_fp16")]; tensor var_9770_begin_0 = const()[name = tensor("op_9770_begin_0"), val = tensor([0, 560, 0, 0])]; tensor var_9770_end_0 = const()[name = tensor("op_9770_end_0"), val = tensor([2, 1, 1, 1024])]; tensor var_9770_end_mask_0 = const()[name = tensor("op_9770_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9770_cast_fp16 = slice_by_index(begin = var_9770_begin_0, end = var_9770_end_0, end_mask = var_9770_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_9770_cast_fp16")]; tensor var_9773_begin_0 = const()[name = tensor("op_9773_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9773_end_0 = const()[name = tensor("op_9773_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_9773_end_mask_0 = const()[name = tensor("op_9773_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9773_cast_fp16 = slice_by_index(begin = var_9773_begin_0, end = var_9773_end_0, end_mask = var_9773_end_mask_0, x = var_9742_cast_fp16)[name = tensor("op_9773_cast_fp16")]; tensor var_9774_begin_0 = const()[name = tensor("op_9774_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9774_end_0 = const()[name = tensor("op_9774_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_9774_end_mask_0 = const()[name = tensor("op_9774_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9774_cast_fp16 = slice_by_index(begin = var_9774_begin_0, end = var_9774_end_0, end_mask = var_9774_end_mask_0, x = var_9742_cast_fp16)[name = tensor("op_9774_cast_fp16")]; tensor var_9775_begin_0 = const()[name = tensor("op_9775_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9775_end_0 = const()[name = tensor("op_9775_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_9775_end_mask_0 = const()[name = tensor("op_9775_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9775_cast_fp16 = slice_by_index(begin = var_9775_begin_0, end = var_9775_end_0, end_mask = var_9775_end_mask_0, x = var_9746_cast_fp16)[name = tensor("op_9775_cast_fp16")]; tensor var_9776_begin_0 = const()[name = tensor("op_9776_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9776_end_0 = const()[name = tensor("op_9776_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_9776_end_mask_0 = const()[name = tensor("op_9776_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9776_cast_fp16 = slice_by_index(begin = var_9776_begin_0, end = var_9776_end_0, end_mask = var_9776_end_mask_0, x = var_9746_cast_fp16)[name = tensor("op_9776_cast_fp16")]; tensor var_9777_begin_0 = const()[name = tensor("op_9777_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9777_end_0 = const()[name = tensor("op_9777_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_9777_end_mask_0 = const()[name = tensor("op_9777_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9777_cast_fp16 = slice_by_index(begin = var_9777_begin_0, end = var_9777_end_0, end_mask = var_9777_end_mask_0, x = var_9750_cast_fp16)[name = tensor("op_9777_cast_fp16")]; tensor var_9778_begin_0 = const()[name = tensor("op_9778_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9778_end_0 = const()[name = tensor("op_9778_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_9778_end_mask_0 = const()[name = tensor("op_9778_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9778_cast_fp16 = slice_by_index(begin = var_9778_begin_0, end = var_9778_end_0, end_mask = var_9778_end_mask_0, x = var_9750_cast_fp16)[name = tensor("op_9778_cast_fp16")]; tensor var_9779_begin_0 = const()[name = tensor("op_9779_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9779_end_0 = const()[name = tensor("op_9779_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_9779_end_mask_0 = const()[name = tensor("op_9779_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9779_cast_fp16 = slice_by_index(begin = var_9779_begin_0, end = var_9779_end_0, end_mask = var_9779_end_mask_0, x = var_9754_cast_fp16)[name = tensor("op_9779_cast_fp16")]; tensor var_9780_begin_0 = const()[name = tensor("op_9780_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9780_end_0 = const()[name = tensor("op_9780_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_9780_end_mask_0 = const()[name = tensor("op_9780_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9780_cast_fp16 = slice_by_index(begin = var_9780_begin_0, end = var_9780_end_0, end_mask = var_9780_end_mask_0, x = var_9754_cast_fp16)[name = tensor("op_9780_cast_fp16")]; tensor var_9781_begin_0 = const()[name = tensor("op_9781_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9781_end_0 = const()[name = tensor("op_9781_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_9781_end_mask_0 = const()[name = tensor("op_9781_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9781_cast_fp16 = slice_by_index(begin = var_9781_begin_0, end = var_9781_end_0, end_mask = var_9781_end_mask_0, x = var_9758_cast_fp16)[name = tensor("op_9781_cast_fp16")]; tensor var_9782_begin_0 = const()[name = tensor("op_9782_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9782_end_0 = const()[name = tensor("op_9782_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_9782_end_mask_0 = const()[name = tensor("op_9782_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9782_cast_fp16 = slice_by_index(begin = var_9782_begin_0, end = var_9782_end_0, end_mask = var_9782_end_mask_0, x = var_9758_cast_fp16)[name = tensor("op_9782_cast_fp16")]; tensor var_9783_begin_0 = const()[name = tensor("op_9783_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9783_end_0 = const()[name = tensor("op_9783_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_9783_end_mask_0 = const()[name = tensor("op_9783_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9783_cast_fp16 = slice_by_index(begin = var_9783_begin_0, end = var_9783_end_0, end_mask = var_9783_end_mask_0, x = var_9762_cast_fp16)[name = tensor("op_9783_cast_fp16")]; tensor var_9784_begin_0 = const()[name = tensor("op_9784_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9784_end_0 = const()[name = tensor("op_9784_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_9784_end_mask_0 = const()[name = tensor("op_9784_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9784_cast_fp16 = slice_by_index(begin = var_9784_begin_0, end = var_9784_end_0, end_mask = var_9784_end_mask_0, x = var_9762_cast_fp16)[name = tensor("op_9784_cast_fp16")]; tensor var_9785_begin_0 = const()[name = tensor("op_9785_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9785_end_0 = const()[name = tensor("op_9785_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_9785_end_mask_0 = const()[name = tensor("op_9785_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9785_cast_fp16 = slice_by_index(begin = var_9785_begin_0, end = var_9785_end_0, end_mask = var_9785_end_mask_0, x = var_9766_cast_fp16)[name = tensor("op_9785_cast_fp16")]; tensor var_9786_begin_0 = const()[name = tensor("op_9786_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9786_end_0 = const()[name = tensor("op_9786_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_9786_end_mask_0 = const()[name = tensor("op_9786_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9786_cast_fp16 = slice_by_index(begin = var_9786_begin_0, end = var_9786_end_0, end_mask = var_9786_end_mask_0, x = var_9766_cast_fp16)[name = tensor("op_9786_cast_fp16")]; tensor var_9787_begin_0 = const()[name = tensor("op_9787_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9787_end_0 = const()[name = tensor("op_9787_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_9787_end_mask_0 = const()[name = tensor("op_9787_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9787_cast_fp16 = slice_by_index(begin = var_9787_begin_0, end = var_9787_end_0, end_mask = var_9787_end_mask_0, x = var_9770_cast_fp16)[name = tensor("op_9787_cast_fp16")]; tensor var_9788_begin_0 = const()[name = tensor("op_9788_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9788_end_0 = const()[name = tensor("op_9788_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_9788_end_mask_0 = const()[name = tensor("op_9788_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9788_cast_fp16 = slice_by_index(begin = var_9788_begin_0, end = var_9788_end_0, end_mask = var_9788_end_mask_0, x = var_9770_cast_fp16)[name = tensor("op_9788_cast_fp16")]; tensor k_99_perm_0 = const()[name = tensor("k_99_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_9793_begin_0 = const()[name = tensor("op_9793_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9793_end_0 = const()[name = tensor("op_9793_end_0"), val = tensor([2, 1024, 1, 80])]; tensor var_9793_end_mask_0 = const()[name = tensor("op_9793_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_99_cast_fp16 = transpose(perm = k_99_perm_0, x = k_97_cast_fp16)[name = tensor("transpose_7")]; tensor var_9793_cast_fp16 = slice_by_index(begin = var_9793_begin_0, end = var_9793_end_0, end_mask = var_9793_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_9793_cast_fp16")]; tensor var_9797_begin_0 = const()[name = tensor("op_9797_begin_0"), val = tensor([0, 0, 0, 80])]; tensor var_9797_end_0 = const()[name = tensor("op_9797_end_0"), val = tensor([2, 1024, 1, 160])]; tensor var_9797_end_mask_0 = const()[name = tensor("op_9797_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9797_cast_fp16 = slice_by_index(begin = var_9797_begin_0, end = var_9797_end_0, end_mask = var_9797_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_9797_cast_fp16")]; tensor var_9801_begin_0 = const()[name = tensor("op_9801_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_9801_end_0 = const()[name = tensor("op_9801_end_0"), val = tensor([2, 1024, 1, 240])]; tensor var_9801_end_mask_0 = const()[name = tensor("op_9801_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9801_cast_fp16 = slice_by_index(begin = var_9801_begin_0, end = var_9801_end_0, end_mask = var_9801_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_9801_cast_fp16")]; tensor var_9805_begin_0 = const()[name = tensor("op_9805_begin_0"), val = tensor([0, 0, 0, 240])]; tensor var_9805_end_0 = const()[name = tensor("op_9805_end_0"), val = tensor([2, 1024, 1, 320])]; tensor var_9805_end_mask_0 = const()[name = tensor("op_9805_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9805_cast_fp16 = slice_by_index(begin = var_9805_begin_0, end = var_9805_end_0, end_mask = var_9805_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_9805_cast_fp16")]; tensor var_9809_begin_0 = const()[name = tensor("op_9809_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_9809_end_0 = const()[name = tensor("op_9809_end_0"), val = tensor([2, 1024, 1, 400])]; tensor var_9809_end_mask_0 = const()[name = tensor("op_9809_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9809_cast_fp16 = slice_by_index(begin = var_9809_begin_0, end = var_9809_end_0, end_mask = var_9809_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_9809_cast_fp16")]; tensor var_9813_begin_0 = const()[name = tensor("op_9813_begin_0"), val = tensor([0, 0, 0, 400])]; tensor var_9813_end_0 = const()[name = tensor("op_9813_end_0"), val = tensor([2, 1024, 1, 480])]; tensor var_9813_end_mask_0 = const()[name = tensor("op_9813_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9813_cast_fp16 = slice_by_index(begin = var_9813_begin_0, end = var_9813_end_0, end_mask = var_9813_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_9813_cast_fp16")]; tensor var_9817_begin_0 = const()[name = tensor("op_9817_begin_0"), val = tensor([0, 0, 0, 480])]; tensor var_9817_end_0 = const()[name = tensor("op_9817_end_0"), val = tensor([2, 1024, 1, 560])]; tensor var_9817_end_mask_0 = const()[name = tensor("op_9817_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9817_cast_fp16 = slice_by_index(begin = var_9817_begin_0, end = var_9817_end_0, end_mask = var_9817_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_9817_cast_fp16")]; tensor var_9821_begin_0 = const()[name = tensor("op_9821_begin_0"), val = tensor([0, 0, 0, 560])]; tensor var_9821_end_0 = const()[name = tensor("op_9821_end_0"), val = tensor([2, 1024, 1, 1])]; tensor var_9821_end_mask_0 = const()[name = tensor("op_9821_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9821_cast_fp16 = slice_by_index(begin = var_9821_begin_0, end = var_9821_end_0, end_mask = var_9821_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_9821_cast_fp16")]; tensor var_9823_begin_0 = const()[name = tensor("op_9823_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9823_end_0 = const()[name = tensor("op_9823_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_9823_end_mask_0 = const()[name = tensor("op_9823_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9823_cast_fp16 = slice_by_index(begin = var_9823_begin_0, end = var_9823_end_0, end_mask = var_9823_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_9823_cast_fp16")]; tensor var_9827_begin_0 = const()[name = tensor("op_9827_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_9827_end_0 = const()[name = tensor("op_9827_end_0"), val = tensor([2, 160, 1, 1024])]; tensor var_9827_end_mask_0 = const()[name = tensor("op_9827_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9827_cast_fp16 = slice_by_index(begin = var_9827_begin_0, end = var_9827_end_0, end_mask = var_9827_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_9827_cast_fp16")]; tensor var_9831_begin_0 = const()[name = tensor("op_9831_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_9831_end_0 = const()[name = tensor("op_9831_end_0"), val = tensor([2, 240, 1, 1024])]; tensor var_9831_end_mask_0 = const()[name = tensor("op_9831_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9831_cast_fp16 = slice_by_index(begin = var_9831_begin_0, end = var_9831_end_0, end_mask = var_9831_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_9831_cast_fp16")]; tensor var_9835_begin_0 = const()[name = tensor("op_9835_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_9835_end_0 = const()[name = tensor("op_9835_end_0"), val = tensor([2, 320, 1, 1024])]; tensor var_9835_end_mask_0 = const()[name = tensor("op_9835_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9835_cast_fp16 = slice_by_index(begin = var_9835_begin_0, end = var_9835_end_0, end_mask = var_9835_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_9835_cast_fp16")]; tensor var_9839_begin_0 = const()[name = tensor("op_9839_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_9839_end_0 = const()[name = tensor("op_9839_end_0"), val = tensor([2, 400, 1, 1024])]; tensor var_9839_end_mask_0 = const()[name = tensor("op_9839_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9839_cast_fp16 = slice_by_index(begin = var_9839_begin_0, end = var_9839_end_0, end_mask = var_9839_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_9839_cast_fp16")]; tensor var_9843_begin_0 = const()[name = tensor("op_9843_begin_0"), val = tensor([0, 400, 0, 0])]; tensor var_9843_end_0 = const()[name = tensor("op_9843_end_0"), val = tensor([2, 480, 1, 1024])]; tensor var_9843_end_mask_0 = const()[name = tensor("op_9843_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9843_cast_fp16 = slice_by_index(begin = var_9843_begin_0, end = var_9843_end_0, end_mask = var_9843_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_9843_cast_fp16")]; tensor var_9847_begin_0 = const()[name = tensor("op_9847_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_9847_end_0 = const()[name = tensor("op_9847_end_0"), val = tensor([2, 560, 1, 1024])]; tensor var_9847_end_mask_0 = const()[name = tensor("op_9847_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9847_cast_fp16 = slice_by_index(begin = var_9847_begin_0, end = var_9847_end_0, end_mask = var_9847_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_9847_cast_fp16")]; tensor var_9851_begin_0 = const()[name = tensor("op_9851_begin_0"), val = tensor([0, 560, 0, 0])]; tensor var_9851_end_0 = const()[name = tensor("op_9851_end_0"), val = tensor([2, 1, 1, 1024])]; tensor var_9851_end_mask_0 = const()[name = tensor("op_9851_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_9851_cast_fp16 = slice_by_index(begin = var_9851_begin_0, end = var_9851_end_0, end_mask = var_9851_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_9851_cast_fp16")]; tensor var_9855_equation_0 = const()[name = tensor("op_9855_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9855_cast_fp16 = einsum(equation = var_9855_equation_0, values = (var_9793_cast_fp16, var_9773_cast_fp16))[name = tensor("op_9855_cast_fp16")]; tensor var_9856_to_fp16 = const()[name = tensor("op_9856_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_769_cast_fp16 = mul(x = var_9855_cast_fp16, y = var_9856_to_fp16)[name = tensor("aw_chunk_769_cast_fp16")]; tensor var_9859_equation_0 = const()[name = tensor("op_9859_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9859_cast_fp16 = einsum(equation = var_9859_equation_0, values = (var_9793_cast_fp16, var_9774_cast_fp16))[name = tensor("op_9859_cast_fp16")]; tensor var_9860_to_fp16 = const()[name = tensor("op_9860_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_771_cast_fp16 = mul(x = var_9859_cast_fp16, y = var_9860_to_fp16)[name = tensor("aw_chunk_771_cast_fp16")]; tensor var_9863_equation_0 = const()[name = tensor("op_9863_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9863_cast_fp16 = einsum(equation = var_9863_equation_0, values = (var_9797_cast_fp16, var_9775_cast_fp16))[name = tensor("op_9863_cast_fp16")]; tensor var_9864_to_fp16 = const()[name = tensor("op_9864_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_773_cast_fp16 = mul(x = var_9863_cast_fp16, y = var_9864_to_fp16)[name = tensor("aw_chunk_773_cast_fp16")]; tensor var_9867_equation_0 = const()[name = tensor("op_9867_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9867_cast_fp16 = einsum(equation = var_9867_equation_0, values = (var_9797_cast_fp16, var_9776_cast_fp16))[name = tensor("op_9867_cast_fp16")]; tensor var_9868_to_fp16 = const()[name = tensor("op_9868_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_775_cast_fp16 = mul(x = var_9867_cast_fp16, y = var_9868_to_fp16)[name = tensor("aw_chunk_775_cast_fp16")]; tensor var_9871_equation_0 = const()[name = tensor("op_9871_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9871_cast_fp16 = einsum(equation = var_9871_equation_0, values = (var_9801_cast_fp16, var_9777_cast_fp16))[name = tensor("op_9871_cast_fp16")]; tensor var_9872_to_fp16 = const()[name = tensor("op_9872_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_777_cast_fp16 = mul(x = var_9871_cast_fp16, y = var_9872_to_fp16)[name = tensor("aw_chunk_777_cast_fp16")]; tensor var_9875_equation_0 = const()[name = tensor("op_9875_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9875_cast_fp16 = einsum(equation = var_9875_equation_0, values = (var_9801_cast_fp16, var_9778_cast_fp16))[name = tensor("op_9875_cast_fp16")]; tensor var_9876_to_fp16 = const()[name = tensor("op_9876_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_779_cast_fp16 = mul(x = var_9875_cast_fp16, y = var_9876_to_fp16)[name = tensor("aw_chunk_779_cast_fp16")]; tensor var_9879_equation_0 = const()[name = tensor("op_9879_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9879_cast_fp16 = einsum(equation = var_9879_equation_0, values = (var_9805_cast_fp16, var_9779_cast_fp16))[name = tensor("op_9879_cast_fp16")]; tensor var_9880_to_fp16 = const()[name = tensor("op_9880_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_781_cast_fp16 = mul(x = var_9879_cast_fp16, y = var_9880_to_fp16)[name = tensor("aw_chunk_781_cast_fp16")]; tensor var_9883_equation_0 = const()[name = tensor("op_9883_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9883_cast_fp16 = einsum(equation = var_9883_equation_0, values = (var_9805_cast_fp16, var_9780_cast_fp16))[name = tensor("op_9883_cast_fp16")]; tensor var_9884_to_fp16 = const()[name = tensor("op_9884_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_783_cast_fp16 = mul(x = var_9883_cast_fp16, y = var_9884_to_fp16)[name = tensor("aw_chunk_783_cast_fp16")]; tensor var_9887_equation_0 = const()[name = tensor("op_9887_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9887_cast_fp16 = einsum(equation = var_9887_equation_0, values = (var_9809_cast_fp16, var_9781_cast_fp16))[name = tensor("op_9887_cast_fp16")]; tensor var_9888_to_fp16 = const()[name = tensor("op_9888_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_785_cast_fp16 = mul(x = var_9887_cast_fp16, y = var_9888_to_fp16)[name = tensor("aw_chunk_785_cast_fp16")]; tensor var_9891_equation_0 = const()[name = tensor("op_9891_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9891_cast_fp16 = einsum(equation = var_9891_equation_0, values = (var_9809_cast_fp16, var_9782_cast_fp16))[name = tensor("op_9891_cast_fp16")]; tensor var_9892_to_fp16 = const()[name = tensor("op_9892_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_787_cast_fp16 = mul(x = var_9891_cast_fp16, y = var_9892_to_fp16)[name = tensor("aw_chunk_787_cast_fp16")]; tensor var_9895_equation_0 = const()[name = tensor("op_9895_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9895_cast_fp16 = einsum(equation = var_9895_equation_0, values = (var_9813_cast_fp16, var_9783_cast_fp16))[name = tensor("op_9895_cast_fp16")]; tensor var_9896_to_fp16 = const()[name = tensor("op_9896_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_789_cast_fp16 = mul(x = var_9895_cast_fp16, y = var_9896_to_fp16)[name = tensor("aw_chunk_789_cast_fp16")]; tensor var_9899_equation_0 = const()[name = tensor("op_9899_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9899_cast_fp16 = einsum(equation = var_9899_equation_0, values = (var_9813_cast_fp16, var_9784_cast_fp16))[name = tensor("op_9899_cast_fp16")]; tensor var_9900_to_fp16 = const()[name = tensor("op_9900_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_791_cast_fp16 = mul(x = var_9899_cast_fp16, y = var_9900_to_fp16)[name = tensor("aw_chunk_791_cast_fp16")]; tensor var_9903_equation_0 = const()[name = tensor("op_9903_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9903_cast_fp16 = einsum(equation = var_9903_equation_0, values = (var_9817_cast_fp16, var_9785_cast_fp16))[name = tensor("op_9903_cast_fp16")]; tensor var_9904_to_fp16 = const()[name = tensor("op_9904_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_793_cast_fp16 = mul(x = var_9903_cast_fp16, y = var_9904_to_fp16)[name = tensor("aw_chunk_793_cast_fp16")]; tensor var_9907_equation_0 = const()[name = tensor("op_9907_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9907_cast_fp16 = einsum(equation = var_9907_equation_0, values = (var_9817_cast_fp16, var_9786_cast_fp16))[name = tensor("op_9907_cast_fp16")]; tensor var_9908_to_fp16 = const()[name = tensor("op_9908_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_795_cast_fp16 = mul(x = var_9907_cast_fp16, y = var_9908_to_fp16)[name = tensor("aw_chunk_795_cast_fp16")]; tensor var_9911_equation_0 = const()[name = tensor("op_9911_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9911_cast_fp16 = einsum(equation = var_9911_equation_0, values = (var_9821_cast_fp16, var_9787_cast_fp16))[name = tensor("op_9911_cast_fp16")]; tensor var_9912_to_fp16 = const()[name = tensor("op_9912_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_797_cast_fp16 = mul(x = var_9911_cast_fp16, y = var_9912_to_fp16)[name = tensor("aw_chunk_797_cast_fp16")]; tensor var_9915_equation_0 = const()[name = tensor("op_9915_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9915_cast_fp16 = einsum(equation = var_9915_equation_0, values = (var_9821_cast_fp16, var_9788_cast_fp16))[name = tensor("op_9915_cast_fp16")]; tensor var_9916_to_fp16 = const()[name = tensor("op_9916_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_799_cast_fp16 = mul(x = var_9915_cast_fp16, y = var_9916_to_fp16)[name = tensor("aw_chunk_799_cast_fp16")]; tensor var_9918_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_769_cast_fp16)[name = tensor("op_9918_cast_fp16")]; tensor var_9919_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_771_cast_fp16)[name = tensor("op_9919_cast_fp16")]; tensor var_9920_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_773_cast_fp16)[name = tensor("op_9920_cast_fp16")]; tensor var_9921_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_775_cast_fp16)[name = tensor("op_9921_cast_fp16")]; tensor var_9922_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_777_cast_fp16)[name = tensor("op_9922_cast_fp16")]; tensor var_9923_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_779_cast_fp16)[name = tensor("op_9923_cast_fp16")]; tensor var_9924_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_781_cast_fp16)[name = tensor("op_9924_cast_fp16")]; tensor var_9925_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_783_cast_fp16)[name = tensor("op_9925_cast_fp16")]; tensor var_9926_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_785_cast_fp16)[name = tensor("op_9926_cast_fp16")]; tensor var_9927_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_787_cast_fp16)[name = tensor("op_9927_cast_fp16")]; tensor var_9928_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_789_cast_fp16)[name = tensor("op_9928_cast_fp16")]; tensor var_9929_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_791_cast_fp16)[name = tensor("op_9929_cast_fp16")]; tensor var_9930_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_793_cast_fp16)[name = tensor("op_9930_cast_fp16")]; tensor var_9931_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_795_cast_fp16)[name = tensor("op_9931_cast_fp16")]; tensor var_9932_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_797_cast_fp16)[name = tensor("op_9932_cast_fp16")]; tensor var_9933_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_799_cast_fp16)[name = tensor("op_9933_cast_fp16")]; tensor var_9935_equation_0 = const()[name = tensor("op_9935_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9935_cast_fp16 = einsum(equation = var_9935_equation_0, values = (var_9823_cast_fp16, var_9918_cast_fp16))[name = tensor("op_9935_cast_fp16")]; tensor var_9937_equation_0 = const()[name = tensor("op_9937_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9937_cast_fp16 = einsum(equation = var_9937_equation_0, values = (var_9823_cast_fp16, var_9919_cast_fp16))[name = tensor("op_9937_cast_fp16")]; tensor var_9939_equation_0 = const()[name = tensor("op_9939_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9939_cast_fp16 = einsum(equation = var_9939_equation_0, values = (var_9827_cast_fp16, var_9920_cast_fp16))[name = tensor("op_9939_cast_fp16")]; tensor var_9941_equation_0 = const()[name = tensor("op_9941_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9941_cast_fp16 = einsum(equation = var_9941_equation_0, values = (var_9827_cast_fp16, var_9921_cast_fp16))[name = tensor("op_9941_cast_fp16")]; tensor var_9943_equation_0 = const()[name = tensor("op_9943_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9943_cast_fp16 = einsum(equation = var_9943_equation_0, values = (var_9831_cast_fp16, var_9922_cast_fp16))[name = tensor("op_9943_cast_fp16")]; tensor var_9945_equation_0 = const()[name = tensor("op_9945_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9945_cast_fp16 = einsum(equation = var_9945_equation_0, values = (var_9831_cast_fp16, var_9923_cast_fp16))[name = tensor("op_9945_cast_fp16")]; tensor var_9947_equation_0 = const()[name = tensor("op_9947_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9947_cast_fp16 = einsum(equation = var_9947_equation_0, values = (var_9835_cast_fp16, var_9924_cast_fp16))[name = tensor("op_9947_cast_fp16")]; tensor var_9949_equation_0 = const()[name = tensor("op_9949_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9949_cast_fp16 = einsum(equation = var_9949_equation_0, values = (var_9835_cast_fp16, var_9925_cast_fp16))[name = tensor("op_9949_cast_fp16")]; tensor var_9951_equation_0 = const()[name = tensor("op_9951_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9951_cast_fp16 = einsum(equation = var_9951_equation_0, values = (var_9839_cast_fp16, var_9926_cast_fp16))[name = tensor("op_9951_cast_fp16")]; tensor var_9953_equation_0 = const()[name = tensor("op_9953_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9953_cast_fp16 = einsum(equation = var_9953_equation_0, values = (var_9839_cast_fp16, var_9927_cast_fp16))[name = tensor("op_9953_cast_fp16")]; tensor var_9955_equation_0 = const()[name = tensor("op_9955_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9955_cast_fp16 = einsum(equation = var_9955_equation_0, values = (var_9843_cast_fp16, var_9928_cast_fp16))[name = tensor("op_9955_cast_fp16")]; tensor var_9957_equation_0 = const()[name = tensor("op_9957_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9957_cast_fp16 = einsum(equation = var_9957_equation_0, values = (var_9843_cast_fp16, var_9929_cast_fp16))[name = tensor("op_9957_cast_fp16")]; tensor var_9959_equation_0 = const()[name = tensor("op_9959_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9959_cast_fp16 = einsum(equation = var_9959_equation_0, values = (var_9847_cast_fp16, var_9930_cast_fp16))[name = tensor("op_9959_cast_fp16")]; tensor var_9961_equation_0 = const()[name = tensor("op_9961_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9961_cast_fp16 = einsum(equation = var_9961_equation_0, values = (var_9847_cast_fp16, var_9931_cast_fp16))[name = tensor("op_9961_cast_fp16")]; tensor var_9963_equation_0 = const()[name = tensor("op_9963_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9963_cast_fp16 = einsum(equation = var_9963_equation_0, values = (var_9851_cast_fp16, var_9932_cast_fp16))[name = tensor("op_9963_cast_fp16")]; tensor var_9965_equation_0 = const()[name = tensor("op_9965_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9965_cast_fp16 = einsum(equation = var_9965_equation_0, values = (var_9851_cast_fp16, var_9933_cast_fp16))[name = tensor("op_9965_cast_fp16")]; tensor var_9967_interleave_0 = const()[name = tensor("op_9967_interleave_0"), val = tensor(false)]; tensor var_9967_cast_fp16 = concat(axis = var_8160, interleave = var_9967_interleave_0, values = (var_9935_cast_fp16, var_9937_cast_fp16))[name = tensor("op_9967_cast_fp16")]; tensor var_9969_interleave_0 = const()[name = tensor("op_9969_interleave_0"), val = tensor(false)]; tensor var_9969_cast_fp16 = concat(axis = var_8160, interleave = var_9969_interleave_0, values = (var_9939_cast_fp16, var_9941_cast_fp16))[name = tensor("op_9969_cast_fp16")]; tensor var_9971_interleave_0 = const()[name = tensor("op_9971_interleave_0"), val = tensor(false)]; tensor var_9971_cast_fp16 = concat(axis = var_8160, interleave = var_9971_interleave_0, values = (var_9943_cast_fp16, var_9945_cast_fp16))[name = tensor("op_9971_cast_fp16")]; tensor var_9973_interleave_0 = const()[name = tensor("op_9973_interleave_0"), val = tensor(false)]; tensor var_9973_cast_fp16 = concat(axis = var_8160, interleave = var_9973_interleave_0, values = (var_9947_cast_fp16, var_9949_cast_fp16))[name = tensor("op_9973_cast_fp16")]; tensor var_9975_interleave_0 = const()[name = tensor("op_9975_interleave_0"), val = tensor(false)]; tensor var_9975_cast_fp16 = concat(axis = var_8160, interleave = var_9975_interleave_0, values = (var_9951_cast_fp16, var_9953_cast_fp16))[name = tensor("op_9975_cast_fp16")]; tensor var_9977_interleave_0 = const()[name = tensor("op_9977_interleave_0"), val = tensor(false)]; tensor var_9977_cast_fp16 = concat(axis = var_8160, interleave = var_9977_interleave_0, values = (var_9955_cast_fp16, var_9957_cast_fp16))[name = tensor("op_9977_cast_fp16")]; tensor var_9979_interleave_0 = const()[name = tensor("op_9979_interleave_0"), val = tensor(false)]; tensor var_9979_cast_fp16 = concat(axis = var_8160, interleave = var_9979_interleave_0, values = (var_9959_cast_fp16, var_9961_cast_fp16))[name = tensor("op_9979_cast_fp16")]; tensor var_9981_interleave_0 = const()[name = tensor("op_9981_interleave_0"), val = tensor(false)]; tensor var_9981_cast_fp16 = concat(axis = var_8160, interleave = var_9981_interleave_0, values = (var_9963_cast_fp16, var_9965_cast_fp16))[name = tensor("op_9981_cast_fp16")]; tensor input_435_interleave_0 = const()[name = tensor("input_435_interleave_0"), val = tensor(false)]; tensor input_435_cast_fp16 = concat(axis = var_8182, interleave = input_435_interleave_0, values = (var_9967_cast_fp16, var_9969_cast_fp16, var_9971_cast_fp16, var_9973_cast_fp16, var_9975_cast_fp16, var_9977_cast_fp16, var_9979_cast_fp16, var_9981_cast_fp16))[name = tensor("input_435_cast_fp16")]; tensor var_9991_pad_type_0 = const()[name = tensor("op_9991_pad_type_0"), val = tensor("valid")]; tensor var_9991_strides_0 = const()[name = tensor("op_9991_strides_0"), val = tensor([1, 1])]; tensor var_9991_pad_0 = const()[name = tensor("op_9991_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_9991_dilations_0 = const()[name = tensor("op_9991_dilations_0"), val = tensor([1, 1])]; tensor var_9991_groups_0 = const()[name = tensor("op_9991_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(771445888)))]; tensor up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(772265152)))]; tensor var_9991_cast_fp16 = conv(bias = up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_9991_dilations_0, groups = var_9991_groups_0, pad = var_9991_pad_0, pad_type = var_9991_pad_type_0, strides = var_9991_strides_0, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_435_cast_fp16)[name = tensor("op_9991_cast_fp16")]; tensor inputs_75_cast_fp16 = add(x = var_9991_cast_fp16, y = inputs_73_cast_fp16)[name = tensor("inputs_75_cast_fp16")]; tensor hidden_states_269_axes_0 = const()[name = tensor("hidden_states_269_axes_0"), val = tensor([1])]; tensor hidden_states_269_gamma_0_to_fp16 = const()[name = tensor("hidden_states_269_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(772266496)))]; tensor hidden_states_269_beta_0_to_fp16 = const()[name = tensor("hidden_states_269_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(772267840)))]; tensor var_10001_to_fp16 = const()[name = tensor("op_10001_to_fp16"), val = tensor(0x1.5p-17)]; tensor hidden_states_269_cast_fp16 = layer_norm(axes = hidden_states_269_axes_0, beta = hidden_states_269_beta_0_to_fp16, epsilon = var_10001_to_fp16, gamma = hidden_states_269_gamma_0_to_fp16, x = inputs_75_cast_fp16)[name = tensor("hidden_states_269_cast_fp16")]; tensor q_51_pad_type_0 = const()[name = tensor("q_51_pad_type_0"), val = tensor("valid")]; tensor q_51_strides_0 = const()[name = tensor("q_51_strides_0"), val = tensor([1, 1])]; tensor q_51_pad_0 = const()[name = tensor("q_51_pad_0"), val = tensor([0, 0, 0, 0])]; tensor q_51_dilations_0 = const()[name = tensor("q_51_dilations_0"), val = tensor([1, 1])]; tensor q_51_groups_0 = const()[name = tensor("q_51_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(772269184)))]; tensor q_51_cast_fp16 = conv(dilations = q_51_dilations_0, groups = q_51_groups_0, pad = q_51_pad_0, pad_type = q_51_pad_type_0, strides = q_51_strides_0, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_269_cast_fp16)[name = tensor("q_51_cast_fp16")]; tensor k_101_pad_type_0 = const()[name = tensor("k_101_pad_type_0"), val = tensor("valid")]; tensor k_101_strides_0 = const()[name = tensor("k_101_strides_0"), val = tensor([1, 1])]; tensor k_101_pad_0 = const()[name = tensor("k_101_pad_0"), val = tensor([0, 0, 0, 0])]; tensor k_101_dilations_0 = const()[name = tensor("k_101_dilations_0"), val = tensor([1, 1])]; tensor k_101_groups_0 = const()[name = tensor("k_101_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(773088448)))]; tensor k_101_cast_fp16 = conv(dilations = k_101_dilations_0, groups = k_101_groups_0, pad = k_101_pad_0, pad_type = k_101_pad_type_0, strides = k_101_strides_0, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_101_cast_fp16")]; tensor v_51_pad_type_0 = const()[name = tensor("v_51_pad_type_0"), val = tensor("valid")]; tensor v_51_strides_0 = const()[name = tensor("v_51_strides_0"), val = tensor([1, 1])]; tensor v_51_pad_0 = const()[name = tensor("v_51_pad_0"), val = tensor([0, 0, 0, 0])]; tensor v_51_dilations_0 = const()[name = tensor("v_51_dilations_0"), val = tensor([1, 1])]; tensor v_51_groups_0 = const()[name = tensor("v_51_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(774071552)))]; tensor v_51_cast_fp16 = conv(dilations = v_51_dilations_0, groups = v_51_groups_0, pad = v_51_pad_0, pad_type = v_51_pad_type_0, strides = v_51_strides_0, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_51_cast_fp16")]; tensor var_10034_begin_0 = const()[name = tensor("op_10034_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10034_end_0 = const()[name = tensor("op_10034_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_10034_end_mask_0 = const()[name = tensor("op_10034_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10034_cast_fp16 = slice_by_index(begin = var_10034_begin_0, end = var_10034_end_0, end_mask = var_10034_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10034_cast_fp16")]; tensor var_10038_begin_0 = const()[name = tensor("op_10038_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_10038_end_0 = const()[name = tensor("op_10038_end_0"), val = tensor([2, 160, 1, 1024])]; tensor var_10038_end_mask_0 = const()[name = tensor("op_10038_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10038_cast_fp16 = slice_by_index(begin = var_10038_begin_0, end = var_10038_end_0, end_mask = var_10038_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10038_cast_fp16")]; tensor var_10042_begin_0 = const()[name = tensor("op_10042_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_10042_end_0 = const()[name = tensor("op_10042_end_0"), val = tensor([2, 240, 1, 1024])]; tensor var_10042_end_mask_0 = const()[name = tensor("op_10042_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10042_cast_fp16 = slice_by_index(begin = var_10042_begin_0, end = var_10042_end_0, end_mask = var_10042_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10042_cast_fp16")]; tensor var_10046_begin_0 = const()[name = tensor("op_10046_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_10046_end_0 = const()[name = tensor("op_10046_end_0"), val = tensor([2, 320, 1, 1024])]; tensor var_10046_end_mask_0 = const()[name = tensor("op_10046_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10046_cast_fp16 = slice_by_index(begin = var_10046_begin_0, end = var_10046_end_0, end_mask = var_10046_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10046_cast_fp16")]; tensor var_10050_begin_0 = const()[name = tensor("op_10050_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_10050_end_0 = const()[name = tensor("op_10050_end_0"), val = tensor([2, 400, 1, 1024])]; tensor var_10050_end_mask_0 = const()[name = tensor("op_10050_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10050_cast_fp16 = slice_by_index(begin = var_10050_begin_0, end = var_10050_end_0, end_mask = var_10050_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10050_cast_fp16")]; tensor var_10054_begin_0 = const()[name = tensor("op_10054_begin_0"), val = tensor([0, 400, 0, 0])]; tensor var_10054_end_0 = const()[name = tensor("op_10054_end_0"), val = tensor([2, 480, 1, 1024])]; tensor var_10054_end_mask_0 = const()[name = tensor("op_10054_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10054_cast_fp16 = slice_by_index(begin = var_10054_begin_0, end = var_10054_end_0, end_mask = var_10054_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10054_cast_fp16")]; tensor var_10058_begin_0 = const()[name = tensor("op_10058_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_10058_end_0 = const()[name = tensor("op_10058_end_0"), val = tensor([2, 560, 1, 1024])]; tensor var_10058_end_mask_0 = const()[name = tensor("op_10058_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10058_cast_fp16 = slice_by_index(begin = var_10058_begin_0, end = var_10058_end_0, end_mask = var_10058_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10058_cast_fp16")]; tensor var_10062_begin_0 = const()[name = tensor("op_10062_begin_0"), val = tensor([0, 560, 0, 0])]; tensor var_10062_end_0 = const()[name = tensor("op_10062_end_0"), val = tensor([2, 1, 1, 1024])]; tensor var_10062_end_mask_0 = const()[name = tensor("op_10062_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_10062_cast_fp16 = slice_by_index(begin = var_10062_begin_0, end = var_10062_end_0, end_mask = var_10062_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10062_cast_fp16")]; tensor var_10065_begin_0 = const()[name = tensor("op_10065_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10065_end_0 = const()[name = tensor("op_10065_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_10065_end_mask_0 = const()[name = tensor("op_10065_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10065_cast_fp16 = slice_by_index(begin = var_10065_begin_0, end = var_10065_end_0, end_mask = var_10065_end_mask_0, x = var_10034_cast_fp16)[name = tensor("op_10065_cast_fp16")]; tensor var_10066_begin_0 = const()[name = tensor("op_10066_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_10066_end_0 = const()[name = tensor("op_10066_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_10066_end_mask_0 = const()[name = tensor("op_10066_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_10066_cast_fp16 = slice_by_index(begin = var_10066_begin_0, end = var_10066_end_0, end_mask = var_10066_end_mask_0, x = var_10034_cast_fp16)[name = tensor("op_10066_cast_fp16")]; tensor var_10067_begin_0 = const()[name = tensor("op_10067_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10067_end_0 = const()[name = tensor("op_10067_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_10067_end_mask_0 = const()[name = tensor("op_10067_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10067_cast_fp16 = slice_by_index(begin = var_10067_begin_0, end = var_10067_end_0, end_mask = var_10067_end_mask_0, x = var_10038_cast_fp16)[name = tensor("op_10067_cast_fp16")]; tensor var_10068_begin_0 = const()[name = tensor("op_10068_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_10068_end_0 = const()[name = tensor("op_10068_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_10068_end_mask_0 = const()[name = tensor("op_10068_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_10068_cast_fp16 = slice_by_index(begin = var_10068_begin_0, end = var_10068_end_0, end_mask = var_10068_end_mask_0, x = var_10038_cast_fp16)[name = tensor("op_10068_cast_fp16")]; tensor var_10069_begin_0 = const()[name = tensor("op_10069_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10069_end_0 = const()[name = tensor("op_10069_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_10069_end_mask_0 = const()[name = tensor("op_10069_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10069_cast_fp16 = slice_by_index(begin = var_10069_begin_0, end = var_10069_end_0, end_mask = var_10069_end_mask_0, x = var_10042_cast_fp16)[name = tensor("op_10069_cast_fp16")]; tensor var_10070_begin_0 = const()[name = tensor("op_10070_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_10070_end_0 = const()[name = tensor("op_10070_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_10070_end_mask_0 = const()[name = tensor("op_10070_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_10070_cast_fp16 = slice_by_index(begin = var_10070_begin_0, end = var_10070_end_0, end_mask = var_10070_end_mask_0, x = var_10042_cast_fp16)[name = tensor("op_10070_cast_fp16")]; tensor var_10071_begin_0 = const()[name = tensor("op_10071_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10071_end_0 = const()[name = tensor("op_10071_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_10071_end_mask_0 = const()[name = tensor("op_10071_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10071_cast_fp16 = slice_by_index(begin = var_10071_begin_0, end = var_10071_end_0, end_mask = var_10071_end_mask_0, x = var_10046_cast_fp16)[name = tensor("op_10071_cast_fp16")]; tensor var_10072_begin_0 = const()[name = tensor("op_10072_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_10072_end_0 = const()[name = tensor("op_10072_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_10072_end_mask_0 = const()[name = tensor("op_10072_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_10072_cast_fp16 = slice_by_index(begin = var_10072_begin_0, end = var_10072_end_0, end_mask = var_10072_end_mask_0, x = var_10046_cast_fp16)[name = tensor("op_10072_cast_fp16")]; tensor var_10073_begin_0 = const()[name = tensor("op_10073_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10073_end_0 = const()[name = tensor("op_10073_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_10073_end_mask_0 = const()[name = tensor("op_10073_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10073_cast_fp16 = slice_by_index(begin = var_10073_begin_0, end = var_10073_end_0, end_mask = var_10073_end_mask_0, x = var_10050_cast_fp16)[name = tensor("op_10073_cast_fp16")]; tensor var_10074_begin_0 = const()[name = tensor("op_10074_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_10074_end_0 = const()[name = tensor("op_10074_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_10074_end_mask_0 = const()[name = tensor("op_10074_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_10074_cast_fp16 = slice_by_index(begin = var_10074_begin_0, end = var_10074_end_0, end_mask = var_10074_end_mask_0, x = var_10050_cast_fp16)[name = tensor("op_10074_cast_fp16")]; tensor var_10075_begin_0 = const()[name = tensor("op_10075_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10075_end_0 = const()[name = tensor("op_10075_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_10075_end_mask_0 = const()[name = tensor("op_10075_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10075_cast_fp16 = slice_by_index(begin = var_10075_begin_0, end = var_10075_end_0, end_mask = var_10075_end_mask_0, x = var_10054_cast_fp16)[name = tensor("op_10075_cast_fp16")]; tensor var_10076_begin_0 = const()[name = tensor("op_10076_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_10076_end_0 = const()[name = tensor("op_10076_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_10076_end_mask_0 = const()[name = tensor("op_10076_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_10076_cast_fp16 = slice_by_index(begin = var_10076_begin_0, end = var_10076_end_0, end_mask = var_10076_end_mask_0, x = var_10054_cast_fp16)[name = tensor("op_10076_cast_fp16")]; tensor var_10077_begin_0 = const()[name = tensor("op_10077_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10077_end_0 = const()[name = tensor("op_10077_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_10077_end_mask_0 = const()[name = tensor("op_10077_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10077_cast_fp16 = slice_by_index(begin = var_10077_begin_0, end = var_10077_end_0, end_mask = var_10077_end_mask_0, x = var_10058_cast_fp16)[name = tensor("op_10077_cast_fp16")]; tensor var_10078_begin_0 = const()[name = tensor("op_10078_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_10078_end_0 = const()[name = tensor("op_10078_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_10078_end_mask_0 = const()[name = tensor("op_10078_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_10078_cast_fp16 = slice_by_index(begin = var_10078_begin_0, end = var_10078_end_0, end_mask = var_10078_end_mask_0, x = var_10058_cast_fp16)[name = tensor("op_10078_cast_fp16")]; tensor var_10079_begin_0 = const()[name = tensor("op_10079_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10079_end_0 = const()[name = tensor("op_10079_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_10079_end_mask_0 = const()[name = tensor("op_10079_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10079_cast_fp16 = slice_by_index(begin = var_10079_begin_0, end = var_10079_end_0, end_mask = var_10079_end_mask_0, x = var_10062_cast_fp16)[name = tensor("op_10079_cast_fp16")]; tensor var_10080_begin_0 = const()[name = tensor("op_10080_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_10080_end_0 = const()[name = tensor("op_10080_end_0"), val = tensor([2, 80, 1, 1])]; tensor var_10080_end_mask_0 = const()[name = tensor("op_10080_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_10080_cast_fp16 = slice_by_index(begin = var_10080_begin_0, end = var_10080_end_0, end_mask = var_10080_end_mask_0, x = var_10062_cast_fp16)[name = tensor("op_10080_cast_fp16")]; tensor k_103_perm_0 = const()[name = tensor("k_103_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_10085_begin_0 = const()[name = tensor("op_10085_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10085_end_0 = const()[name = tensor("op_10085_end_0"), val = tensor([2, 77, 1, 80])]; tensor var_10085_end_mask_0 = const()[name = tensor("op_10085_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_103_cast_fp16 = transpose(perm = k_103_perm_0, x = k_101_cast_fp16)[name = tensor("transpose_6")]; tensor var_10085_cast_fp16 = slice_by_index(begin = var_10085_begin_0, end = var_10085_end_0, end_mask = var_10085_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10085_cast_fp16")]; tensor var_10089_begin_0 = const()[name = tensor("op_10089_begin_0"), val = tensor([0, 0, 0, 80])]; tensor var_10089_end_0 = const()[name = tensor("op_10089_end_0"), val = tensor([2, 77, 1, 160])]; tensor var_10089_end_mask_0 = const()[name = tensor("op_10089_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10089_cast_fp16 = slice_by_index(begin = var_10089_begin_0, end = var_10089_end_0, end_mask = var_10089_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10089_cast_fp16")]; tensor var_10093_begin_0 = const()[name = tensor("op_10093_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_10093_end_0 = const()[name = tensor("op_10093_end_0"), val = tensor([2, 77, 1, 240])]; tensor var_10093_end_mask_0 = const()[name = tensor("op_10093_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10093_cast_fp16 = slice_by_index(begin = var_10093_begin_0, end = var_10093_end_0, end_mask = var_10093_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10093_cast_fp16")]; tensor var_10097_begin_0 = const()[name = tensor("op_10097_begin_0"), val = tensor([0, 0, 0, 240])]; tensor var_10097_end_0 = const()[name = tensor("op_10097_end_0"), val = tensor([2, 77, 1, 320])]; tensor var_10097_end_mask_0 = const()[name = tensor("op_10097_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10097_cast_fp16 = slice_by_index(begin = var_10097_begin_0, end = var_10097_end_0, end_mask = var_10097_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10097_cast_fp16")]; tensor var_10101_begin_0 = const()[name = tensor("op_10101_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_10101_end_0 = const()[name = tensor("op_10101_end_0"), val = tensor([2, 77, 1, 400])]; tensor var_10101_end_mask_0 = const()[name = tensor("op_10101_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10101_cast_fp16 = slice_by_index(begin = var_10101_begin_0, end = var_10101_end_0, end_mask = var_10101_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10101_cast_fp16")]; tensor var_10105_begin_0 = const()[name = tensor("op_10105_begin_0"), val = tensor([0, 0, 0, 400])]; tensor var_10105_end_0 = const()[name = tensor("op_10105_end_0"), val = tensor([2, 77, 1, 480])]; tensor var_10105_end_mask_0 = const()[name = tensor("op_10105_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10105_cast_fp16 = slice_by_index(begin = var_10105_begin_0, end = var_10105_end_0, end_mask = var_10105_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10105_cast_fp16")]; tensor var_10109_begin_0 = const()[name = tensor("op_10109_begin_0"), val = tensor([0, 0, 0, 480])]; tensor var_10109_end_0 = const()[name = tensor("op_10109_end_0"), val = tensor([2, 77, 1, 560])]; tensor var_10109_end_mask_0 = const()[name = tensor("op_10109_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10109_cast_fp16 = slice_by_index(begin = var_10109_begin_0, end = var_10109_end_0, end_mask = var_10109_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10109_cast_fp16")]; tensor var_10113_begin_0 = const()[name = tensor("op_10113_begin_0"), val = tensor([0, 0, 0, 560])]; tensor var_10113_end_0 = const()[name = tensor("op_10113_end_0"), val = tensor([2, 77, 1, 1])]; tensor var_10113_end_mask_0 = const()[name = tensor("op_10113_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_10113_cast_fp16 = slice_by_index(begin = var_10113_begin_0, end = var_10113_end_0, end_mask = var_10113_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10113_cast_fp16")]; tensor var_10115_begin_0 = const()[name = tensor("op_10115_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10115_end_0 = const()[name = tensor("op_10115_end_0"), val = tensor([2, 80, 1, 77])]; tensor var_10115_end_mask_0 = const()[name = tensor("op_10115_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10115_cast_fp16 = slice_by_index(begin = var_10115_begin_0, end = var_10115_end_0, end_mask = var_10115_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10115_cast_fp16")]; tensor var_10119_begin_0 = const()[name = tensor("op_10119_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_10119_end_0 = const()[name = tensor("op_10119_end_0"), val = tensor([2, 160, 1, 77])]; tensor var_10119_end_mask_0 = const()[name = tensor("op_10119_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10119_cast_fp16 = slice_by_index(begin = var_10119_begin_0, end = var_10119_end_0, end_mask = var_10119_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10119_cast_fp16")]; tensor var_10123_begin_0 = const()[name = tensor("op_10123_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_10123_end_0 = const()[name = tensor("op_10123_end_0"), val = tensor([2, 240, 1, 77])]; tensor var_10123_end_mask_0 = const()[name = tensor("op_10123_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10123_cast_fp16 = slice_by_index(begin = var_10123_begin_0, end = var_10123_end_0, end_mask = var_10123_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10123_cast_fp16")]; tensor var_10127_begin_0 = const()[name = tensor("op_10127_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_10127_end_0 = const()[name = tensor("op_10127_end_0"), val = tensor([2, 320, 1, 77])]; tensor var_10127_end_mask_0 = const()[name = tensor("op_10127_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10127_cast_fp16 = slice_by_index(begin = var_10127_begin_0, end = var_10127_end_0, end_mask = var_10127_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10127_cast_fp16")]; tensor var_10131_begin_0 = const()[name = tensor("op_10131_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_10131_end_0 = const()[name = tensor("op_10131_end_0"), val = tensor([2, 400, 1, 77])]; tensor var_10131_end_mask_0 = const()[name = tensor("op_10131_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10131_cast_fp16 = slice_by_index(begin = var_10131_begin_0, end = var_10131_end_0, end_mask = var_10131_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10131_cast_fp16")]; tensor var_10135_begin_0 = const()[name = tensor("op_10135_begin_0"), val = tensor([0, 400, 0, 0])]; tensor var_10135_end_0 = const()[name = tensor("op_10135_end_0"), val = tensor([2, 480, 1, 77])]; tensor var_10135_end_mask_0 = const()[name = tensor("op_10135_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10135_cast_fp16 = slice_by_index(begin = var_10135_begin_0, end = var_10135_end_0, end_mask = var_10135_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10135_cast_fp16")]; tensor var_10139_begin_0 = const()[name = tensor("op_10139_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_10139_end_0 = const()[name = tensor("op_10139_end_0"), val = tensor([2, 560, 1, 77])]; tensor var_10139_end_mask_0 = const()[name = tensor("op_10139_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10139_cast_fp16 = slice_by_index(begin = var_10139_begin_0, end = var_10139_end_0, end_mask = var_10139_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10139_cast_fp16")]; tensor var_10143_begin_0 = const()[name = tensor("op_10143_begin_0"), val = tensor([0, 560, 0, 0])]; tensor var_10143_end_0 = const()[name = tensor("op_10143_end_0"), val = tensor([2, 1, 1, 77])]; tensor var_10143_end_mask_0 = const()[name = tensor("op_10143_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_10143_cast_fp16 = slice_by_index(begin = var_10143_begin_0, end = var_10143_end_0, end_mask = var_10143_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10143_cast_fp16")]; tensor var_10147_equation_0 = const()[name = tensor("op_10147_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10147_cast_fp16 = einsum(equation = var_10147_equation_0, values = (var_10085_cast_fp16, var_10065_cast_fp16))[name = tensor("op_10147_cast_fp16")]; tensor var_10148_to_fp16 = const()[name = tensor("op_10148_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_801_cast_fp16 = mul(x = var_10147_cast_fp16, y = var_10148_to_fp16)[name = tensor("aw_chunk_801_cast_fp16")]; tensor var_10151_equation_0 = const()[name = tensor("op_10151_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10151_cast_fp16 = einsum(equation = var_10151_equation_0, values = (var_10085_cast_fp16, var_10066_cast_fp16))[name = tensor("op_10151_cast_fp16")]; tensor var_10152_to_fp16 = const()[name = tensor("op_10152_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_803_cast_fp16 = mul(x = var_10151_cast_fp16, y = var_10152_to_fp16)[name = tensor("aw_chunk_803_cast_fp16")]; tensor var_10155_equation_0 = const()[name = tensor("op_10155_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10155_cast_fp16 = einsum(equation = var_10155_equation_0, values = (var_10089_cast_fp16, var_10067_cast_fp16))[name = tensor("op_10155_cast_fp16")]; tensor var_10156_to_fp16 = const()[name = tensor("op_10156_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_805_cast_fp16 = mul(x = var_10155_cast_fp16, y = var_10156_to_fp16)[name = tensor("aw_chunk_805_cast_fp16")]; tensor var_10159_equation_0 = const()[name = tensor("op_10159_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10159_cast_fp16 = einsum(equation = var_10159_equation_0, values = (var_10089_cast_fp16, var_10068_cast_fp16))[name = tensor("op_10159_cast_fp16")]; tensor var_10160_to_fp16 = const()[name = tensor("op_10160_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_807_cast_fp16 = mul(x = var_10159_cast_fp16, y = var_10160_to_fp16)[name = tensor("aw_chunk_807_cast_fp16")]; tensor var_10163_equation_0 = const()[name = tensor("op_10163_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10163_cast_fp16 = einsum(equation = var_10163_equation_0, values = (var_10093_cast_fp16, var_10069_cast_fp16))[name = tensor("op_10163_cast_fp16")]; tensor var_10164_to_fp16 = const()[name = tensor("op_10164_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_809_cast_fp16 = mul(x = var_10163_cast_fp16, y = var_10164_to_fp16)[name = tensor("aw_chunk_809_cast_fp16")]; tensor var_10167_equation_0 = const()[name = tensor("op_10167_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10167_cast_fp16 = einsum(equation = var_10167_equation_0, values = (var_10093_cast_fp16, var_10070_cast_fp16))[name = tensor("op_10167_cast_fp16")]; tensor var_10168_to_fp16 = const()[name = tensor("op_10168_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_811_cast_fp16 = mul(x = var_10167_cast_fp16, y = var_10168_to_fp16)[name = tensor("aw_chunk_811_cast_fp16")]; tensor var_10171_equation_0 = const()[name = tensor("op_10171_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10171_cast_fp16 = einsum(equation = var_10171_equation_0, values = (var_10097_cast_fp16, var_10071_cast_fp16))[name = tensor("op_10171_cast_fp16")]; tensor var_10172_to_fp16 = const()[name = tensor("op_10172_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_813_cast_fp16 = mul(x = var_10171_cast_fp16, y = var_10172_to_fp16)[name = tensor("aw_chunk_813_cast_fp16")]; tensor var_10175_equation_0 = const()[name = tensor("op_10175_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10175_cast_fp16 = einsum(equation = var_10175_equation_0, values = (var_10097_cast_fp16, var_10072_cast_fp16))[name = tensor("op_10175_cast_fp16")]; tensor var_10176_to_fp16 = const()[name = tensor("op_10176_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_815_cast_fp16 = mul(x = var_10175_cast_fp16, y = var_10176_to_fp16)[name = tensor("aw_chunk_815_cast_fp16")]; tensor var_10179_equation_0 = const()[name = tensor("op_10179_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10179_cast_fp16 = einsum(equation = var_10179_equation_0, values = (var_10101_cast_fp16, var_10073_cast_fp16))[name = tensor("op_10179_cast_fp16")]; tensor var_10180_to_fp16 = const()[name = tensor("op_10180_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_817_cast_fp16 = mul(x = var_10179_cast_fp16, y = var_10180_to_fp16)[name = tensor("aw_chunk_817_cast_fp16")]; tensor var_10183_equation_0 = const()[name = tensor("op_10183_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10183_cast_fp16 = einsum(equation = var_10183_equation_0, values = (var_10101_cast_fp16, var_10074_cast_fp16))[name = tensor("op_10183_cast_fp16")]; tensor var_10184_to_fp16 = const()[name = tensor("op_10184_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_819_cast_fp16 = mul(x = var_10183_cast_fp16, y = var_10184_to_fp16)[name = tensor("aw_chunk_819_cast_fp16")]; tensor var_10187_equation_0 = const()[name = tensor("op_10187_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10187_cast_fp16 = einsum(equation = var_10187_equation_0, values = (var_10105_cast_fp16, var_10075_cast_fp16))[name = tensor("op_10187_cast_fp16")]; tensor var_10188_to_fp16 = const()[name = tensor("op_10188_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_821_cast_fp16 = mul(x = var_10187_cast_fp16, y = var_10188_to_fp16)[name = tensor("aw_chunk_821_cast_fp16")]; tensor var_10191_equation_0 = const()[name = tensor("op_10191_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10191_cast_fp16 = einsum(equation = var_10191_equation_0, values = (var_10105_cast_fp16, var_10076_cast_fp16))[name = tensor("op_10191_cast_fp16")]; tensor var_10192_to_fp16 = const()[name = tensor("op_10192_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_823_cast_fp16 = mul(x = var_10191_cast_fp16, y = var_10192_to_fp16)[name = tensor("aw_chunk_823_cast_fp16")]; tensor var_10195_equation_0 = const()[name = tensor("op_10195_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10195_cast_fp16 = einsum(equation = var_10195_equation_0, values = (var_10109_cast_fp16, var_10077_cast_fp16))[name = tensor("op_10195_cast_fp16")]; tensor var_10196_to_fp16 = const()[name = tensor("op_10196_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_825_cast_fp16 = mul(x = var_10195_cast_fp16, y = var_10196_to_fp16)[name = tensor("aw_chunk_825_cast_fp16")]; tensor var_10199_equation_0 = const()[name = tensor("op_10199_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10199_cast_fp16 = einsum(equation = var_10199_equation_0, values = (var_10109_cast_fp16, var_10078_cast_fp16))[name = tensor("op_10199_cast_fp16")]; tensor var_10200_to_fp16 = const()[name = tensor("op_10200_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_827_cast_fp16 = mul(x = var_10199_cast_fp16, y = var_10200_to_fp16)[name = tensor("aw_chunk_827_cast_fp16")]; tensor var_10203_equation_0 = const()[name = tensor("op_10203_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10203_cast_fp16 = einsum(equation = var_10203_equation_0, values = (var_10113_cast_fp16, var_10079_cast_fp16))[name = tensor("op_10203_cast_fp16")]; tensor var_10204_to_fp16 = const()[name = tensor("op_10204_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_829_cast_fp16 = mul(x = var_10203_cast_fp16, y = var_10204_to_fp16)[name = tensor("aw_chunk_829_cast_fp16")]; tensor var_10207_equation_0 = const()[name = tensor("op_10207_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10207_cast_fp16 = einsum(equation = var_10207_equation_0, values = (var_10113_cast_fp16, var_10080_cast_fp16))[name = tensor("op_10207_cast_fp16")]; tensor var_10208_to_fp16 = const()[name = tensor("op_10208_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_831_cast_fp16 = mul(x = var_10207_cast_fp16, y = var_10208_to_fp16)[name = tensor("aw_chunk_831_cast_fp16")]; tensor var_10210_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_801_cast_fp16)[name = tensor("op_10210_cast_fp16")]; tensor var_10211_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_803_cast_fp16)[name = tensor("op_10211_cast_fp16")]; tensor var_10212_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_805_cast_fp16)[name = tensor("op_10212_cast_fp16")]; tensor var_10213_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_807_cast_fp16)[name = tensor("op_10213_cast_fp16")]; tensor var_10214_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_809_cast_fp16)[name = tensor("op_10214_cast_fp16")]; tensor var_10215_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_811_cast_fp16)[name = tensor("op_10215_cast_fp16")]; tensor var_10216_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_813_cast_fp16)[name = tensor("op_10216_cast_fp16")]; tensor var_10217_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_815_cast_fp16)[name = tensor("op_10217_cast_fp16")]; tensor var_10218_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_817_cast_fp16)[name = tensor("op_10218_cast_fp16")]; tensor var_10219_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_819_cast_fp16)[name = tensor("op_10219_cast_fp16")]; tensor var_10220_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_821_cast_fp16)[name = tensor("op_10220_cast_fp16")]; tensor var_10221_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_823_cast_fp16)[name = tensor("op_10221_cast_fp16")]; tensor var_10222_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_825_cast_fp16)[name = tensor("op_10222_cast_fp16")]; tensor var_10223_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_827_cast_fp16)[name = tensor("op_10223_cast_fp16")]; tensor var_10224_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_829_cast_fp16)[name = tensor("op_10224_cast_fp16")]; tensor var_10225_cast_fp16 = softmax(axis = var_8182, x = aw_chunk_831_cast_fp16)[name = tensor("op_10225_cast_fp16")]; tensor var_10227_equation_0 = const()[name = tensor("op_10227_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10227_cast_fp16 = einsum(equation = var_10227_equation_0, values = (var_10115_cast_fp16, var_10210_cast_fp16))[name = tensor("op_10227_cast_fp16")]; tensor var_10229_equation_0 = const()[name = tensor("op_10229_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10229_cast_fp16 = einsum(equation = var_10229_equation_0, values = (var_10115_cast_fp16, var_10211_cast_fp16))[name = tensor("op_10229_cast_fp16")]; tensor var_10231_equation_0 = const()[name = tensor("op_10231_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10231_cast_fp16 = einsum(equation = var_10231_equation_0, values = (var_10119_cast_fp16, var_10212_cast_fp16))[name = tensor("op_10231_cast_fp16")]; tensor var_10233_equation_0 = const()[name = tensor("op_10233_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10233_cast_fp16 = einsum(equation = var_10233_equation_0, values = (var_10119_cast_fp16, var_10213_cast_fp16))[name = tensor("op_10233_cast_fp16")]; tensor var_10235_equation_0 = const()[name = tensor("op_10235_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10235_cast_fp16 = einsum(equation = var_10235_equation_0, values = (var_10123_cast_fp16, var_10214_cast_fp16))[name = tensor("op_10235_cast_fp16")]; tensor var_10237_equation_0 = const()[name = tensor("op_10237_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10237_cast_fp16 = einsum(equation = var_10237_equation_0, values = (var_10123_cast_fp16, var_10215_cast_fp16))[name = tensor("op_10237_cast_fp16")]; tensor var_10239_equation_0 = const()[name = tensor("op_10239_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10239_cast_fp16 = einsum(equation = var_10239_equation_0, values = (var_10127_cast_fp16, var_10216_cast_fp16))[name = tensor("op_10239_cast_fp16")]; tensor var_10241_equation_0 = const()[name = tensor("op_10241_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10241_cast_fp16 = einsum(equation = var_10241_equation_0, values = (var_10127_cast_fp16, var_10217_cast_fp16))[name = tensor("op_10241_cast_fp16")]; tensor var_10243_equation_0 = const()[name = tensor("op_10243_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10243_cast_fp16 = einsum(equation = var_10243_equation_0, values = (var_10131_cast_fp16, var_10218_cast_fp16))[name = tensor("op_10243_cast_fp16")]; tensor var_10245_equation_0 = const()[name = tensor("op_10245_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10245_cast_fp16 = einsum(equation = var_10245_equation_0, values = (var_10131_cast_fp16, var_10219_cast_fp16))[name = tensor("op_10245_cast_fp16")]; tensor var_10247_equation_0 = const()[name = tensor("op_10247_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10247_cast_fp16 = einsum(equation = var_10247_equation_0, values = (var_10135_cast_fp16, var_10220_cast_fp16))[name = tensor("op_10247_cast_fp16")]; tensor var_10249_equation_0 = const()[name = tensor("op_10249_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10249_cast_fp16 = einsum(equation = var_10249_equation_0, values = (var_10135_cast_fp16, var_10221_cast_fp16))[name = tensor("op_10249_cast_fp16")]; tensor var_10251_equation_0 = const()[name = tensor("op_10251_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10251_cast_fp16 = einsum(equation = var_10251_equation_0, values = (var_10139_cast_fp16, var_10222_cast_fp16))[name = tensor("op_10251_cast_fp16")]; tensor var_10253_equation_0 = const()[name = tensor("op_10253_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10253_cast_fp16 = einsum(equation = var_10253_equation_0, values = (var_10139_cast_fp16, var_10223_cast_fp16))[name = tensor("op_10253_cast_fp16")]; tensor var_10255_equation_0 = const()[name = tensor("op_10255_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10255_cast_fp16 = einsum(equation = var_10255_equation_0, values = (var_10143_cast_fp16, var_10224_cast_fp16))[name = tensor("op_10255_cast_fp16")]; tensor var_10257_equation_0 = const()[name = tensor("op_10257_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10257_cast_fp16 = einsum(equation = var_10257_equation_0, values = (var_10143_cast_fp16, var_10225_cast_fp16))[name = tensor("op_10257_cast_fp16")]; tensor var_10259_interleave_0 = const()[name = tensor("op_10259_interleave_0"), val = tensor(false)]; tensor var_10259_cast_fp16 = concat(axis = var_8160, interleave = var_10259_interleave_0, values = (var_10227_cast_fp16, var_10229_cast_fp16))[name = tensor("op_10259_cast_fp16")]; tensor var_10261_interleave_0 = const()[name = tensor("op_10261_interleave_0"), val = tensor(false)]; tensor var_10261_cast_fp16 = concat(axis = var_8160, interleave = var_10261_interleave_0, values = (var_10231_cast_fp16, var_10233_cast_fp16))[name = tensor("op_10261_cast_fp16")]; tensor var_10263_interleave_0 = const()[name = tensor("op_10263_interleave_0"), val = tensor(false)]; tensor var_10263_cast_fp16 = concat(axis = var_8160, interleave = var_10263_interleave_0, values = (var_10235_cast_fp16, var_10237_cast_fp16))[name = tensor("op_10263_cast_fp16")]; tensor var_10265_interleave_0 = const()[name = tensor("op_10265_interleave_0"), val = tensor(false)]; tensor var_10265_cast_fp16 = concat(axis = var_8160, interleave = var_10265_interleave_0, values = (var_10239_cast_fp16, var_10241_cast_fp16))[name = tensor("op_10265_cast_fp16")]; tensor var_10267_interleave_0 = const()[name = tensor("op_10267_interleave_0"), val = tensor(false)]; tensor var_10267_cast_fp16 = concat(axis = var_8160, interleave = var_10267_interleave_0, values = (var_10243_cast_fp16, var_10245_cast_fp16))[name = tensor("op_10267_cast_fp16")]; tensor var_10269_interleave_0 = const()[name = tensor("op_10269_interleave_0"), val = tensor(false)]; tensor var_10269_cast_fp16 = concat(axis = var_8160, interleave = var_10269_interleave_0, values = (var_10247_cast_fp16, var_10249_cast_fp16))[name = tensor("op_10269_cast_fp16")]; tensor var_10271_interleave_0 = const()[name = tensor("op_10271_interleave_0"), val = tensor(false)]; tensor var_10271_cast_fp16 = concat(axis = var_8160, interleave = var_10271_interleave_0, values = (var_10251_cast_fp16, var_10253_cast_fp16))[name = tensor("op_10271_cast_fp16")]; tensor var_10273_interleave_0 = const()[name = tensor("op_10273_interleave_0"), val = tensor(false)]; tensor var_10273_cast_fp16 = concat(axis = var_8160, interleave = var_10273_interleave_0, values = (var_10255_cast_fp16, var_10257_cast_fp16))[name = tensor("op_10273_cast_fp16")]; tensor input_437_interleave_0 = const()[name = tensor("input_437_interleave_0"), val = tensor(false)]; tensor input_437_cast_fp16 = concat(axis = var_8182, interleave = input_437_interleave_0, values = (var_10259_cast_fp16, var_10261_cast_fp16, var_10263_cast_fp16, var_10265_cast_fp16, var_10267_cast_fp16, var_10269_cast_fp16, var_10271_cast_fp16, var_10273_cast_fp16))[name = tensor("input_437_cast_fp16")]; tensor var_10283_pad_type_0 = const()[name = tensor("op_10283_pad_type_0"), val = tensor("valid")]; tensor var_10283_strides_0 = const()[name = tensor("op_10283_strides_0"), val = tensor([1, 1])]; tensor var_10283_pad_0 = const()[name = tensor("op_10283_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_10283_dilations_0 = const()[name = tensor("op_10283_dilations_0"), val = tensor([1, 1])]; tensor var_10283_groups_0 = const()[name = tensor("op_10283_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775054656)))]; tensor up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775873920)))]; tensor var_10283_cast_fp16 = conv(bias = up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_10283_dilations_0, groups = var_10283_groups_0, pad = var_10283_pad_0, pad_type = var_10283_pad_type_0, strides = var_10283_strides_0, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_437_cast_fp16)[name = tensor("op_10283_cast_fp16")]; tensor inputs_77_cast_fp16 = add(x = var_10283_cast_fp16, y = inputs_75_cast_fp16)[name = tensor("inputs_77_cast_fp16")]; tensor input_439_axes_0 = const()[name = tensor("input_439_axes_0"), val = tensor([1])]; tensor input_439_gamma_0_to_fp16 = const()[name = tensor("input_439_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775875264)))]; tensor input_439_beta_0_to_fp16 = const()[name = tensor("input_439_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775876608)))]; tensor var_10293_to_fp16 = const()[name = tensor("op_10293_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_439_cast_fp16 = layer_norm(axes = input_439_axes_0, beta = input_439_beta_0_to_fp16, epsilon = var_10293_to_fp16, gamma = input_439_gamma_0_to_fp16, x = inputs_77_cast_fp16)[name = tensor("input_439_cast_fp16")]; tensor var_10313_pad_type_0 = const()[name = tensor("op_10313_pad_type_0"), val = tensor("valid")]; tensor var_10313_strides_0 = const()[name = tensor("op_10313_strides_0"), val = tensor([1, 1])]; tensor var_10313_pad_0 = const()[name = tensor("op_10313_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_10313_dilations_0 = const()[name = tensor("op_10313_dilations_0"), val = tensor([1, 1])]; tensor var_10313_groups_0 = const()[name = tensor("op_10313_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775877952)))]; tensor up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(782431616)))]; tensor var_10313_cast_fp16 = conv(bias = up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_10313_dilations_0, groups = var_10313_groups_0, pad = var_10313_pad_0, pad_type = var_10313_pad_type_0, strides = var_10313_strides_0, weight = up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_439_cast_fp16)[name = tensor("op_10313_cast_fp16")]; tensor var_10314_split_sizes_0 = const()[name = tensor("op_10314_split_sizes_0"), val = tensor([2560, 2560])]; tensor var_10314_axis_0 = const()[name = tensor("op_10314_axis_0"), val = tensor(1)]; tensor var_10314_cast_fp16_0, tensor var_10314_cast_fp16_1 = split(axis = var_10314_axis_0, split_sizes = var_10314_split_sizes_0, x = var_10313_cast_fp16)[name = tensor("op_10314_cast_fp16")]; tensor var_10316_mode_0 = const()[name = tensor("op_10316_mode_0"), val = tensor("EXACT")]; tensor var_10316_cast_fp16 = gelu(mode = var_10316_mode_0, x = var_10314_cast_fp16_1)[name = tensor("op_10316_cast_fp16")]; tensor input_441_cast_fp16 = mul(x = var_10314_cast_fp16_0, y = var_10316_cast_fp16)[name = tensor("input_441_cast_fp16")]; tensor var_10324_pad_type_0 = const()[name = tensor("op_10324_pad_type_0"), val = tensor("valid")]; tensor var_10324_strides_0 = const()[name = tensor("op_10324_strides_0"), val = tensor([1, 1])]; tensor var_10324_pad_0 = const()[name = tensor("op_10324_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_10324_dilations_0 = const()[name = tensor("op_10324_dilations_0"), val = tensor([1, 1])]; tensor var_10324_groups_0 = const()[name = tensor("op_10324_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(782441920)))]; tensor up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(785718784)))]; tensor var_10324_cast_fp16 = conv(bias = up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_10324_dilations_0, groups = var_10324_groups_0, pad = var_10324_pad_0, pad_type = var_10324_pad_type_0, strides = var_10324_strides_0, weight = up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_441_cast_fp16)[name = tensor("op_10324_cast_fp16")]; tensor hidden_states_273_cast_fp16 = add(x = var_10324_cast_fp16, y = inputs_77_cast_fp16)[name = tensor("hidden_states_273_cast_fp16")]; tensor var_10326 = const()[name = tensor("op_10326"), val = tensor([2, 640, 32, 32])]; tensor input_443_cast_fp16 = reshape(shape = var_10326, x = hidden_states_273_cast_fp16)[name = tensor("input_443_cast_fp16")]; tensor hidden_states_275_pad_type_0 = const()[name = tensor("hidden_states_275_pad_type_0"), val = tensor("valid")]; tensor hidden_states_275_strides_0 = const()[name = tensor("hidden_states_275_strides_0"), val = tensor([1, 1])]; tensor hidden_states_275_pad_0 = const()[name = tensor("hidden_states_275_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_275_dilations_0 = const()[name = tensor("hidden_states_275_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_275_groups_0 = const()[name = tensor("hidden_states_275_groups_0"), val = tensor(1)]; tensor up_blocks_2_attentions_2_proj_out_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(785720128)))]; tensor up_blocks_2_attentions_2_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(786539392)))]; tensor hidden_states_275_cast_fp16 = conv(bias = up_blocks_2_attentions_2_proj_out_bias_to_fp16, dilations = hidden_states_275_dilations_0, groups = hidden_states_275_groups_0, pad = hidden_states_275_pad_0, pad_type = hidden_states_275_pad_type_0, strides = hidden_states_275_strides_0, weight = up_blocks_2_attentions_2_proj_out_weight_to_fp16, x = input_443_cast_fp16)[name = tensor("hidden_states_275_cast_fp16")]; tensor input_445_cast_fp16 = add(x = hidden_states_275_cast_fp16, y = hidden_states_263_cast_fp16)[name = tensor("input_445_cast_fp16")]; tensor input_447_scale_factor_height_0 = const()[name = tensor("input_447_scale_factor_height_0"), val = tensor(0x1p+1)]; tensor input_447_scale_factor_width_0 = const()[name = tensor("input_447_scale_factor_width_0"), val = tensor(0x1p+1)]; tensor input_447_cast_fp16 = upsample_nearest_neighbor(scale_factor_height = input_447_scale_factor_height_0, scale_factor_width = input_447_scale_factor_width_0, x = input_445_cast_fp16)[name = tensor("input_447_cast_fp16")]; tensor hidden_states_277_pad_type_0 = const()[name = tensor("hidden_states_277_pad_type_0"), val = tensor("custom")]; tensor hidden_states_277_pad_0 = const()[name = tensor("hidden_states_277_pad_0"), val = tensor([1, 1, 1, 1])]; tensor hidden_states_277_strides_0 = const()[name = tensor("hidden_states_277_strides_0"), val = tensor([1, 1])]; tensor hidden_states_277_dilations_0 = const()[name = tensor("hidden_states_277_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_277_groups_0 = const()[name = tensor("hidden_states_277_groups_0"), val = tensor(1)]; tensor up_blocks_2_upsamplers_0_conv_weight_to_fp16 = const()[name = tensor("up_blocks_2_upsamplers_0_conv_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(786540736)))]; tensor up_blocks_2_upsamplers_0_conv_bias_to_fp16 = const()[name = tensor("up_blocks_2_upsamplers_0_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(793913600)))]; tensor hidden_states_277_cast_fp16 = conv(bias = up_blocks_2_upsamplers_0_conv_bias_to_fp16, dilations = hidden_states_277_dilations_0, groups = hidden_states_277_groups_0, pad = hidden_states_277_pad_0, pad_type = hidden_states_277_pad_type_0, strides = hidden_states_277_strides_0, weight = up_blocks_2_upsamplers_0_conv_weight_to_fp16, x = input_447_cast_fp16)[name = tensor("hidden_states_277_cast_fp16")]; tensor var_10347 = const()[name = tensor("op_10347"), val = tensor(3)]; tensor var_10375 = const()[name = tensor("op_10375"), val = tensor(1)]; tensor input_449_interleave_0 = const()[name = tensor("input_449_interleave_0"), val = tensor(false)]; tensor cast_8 = cast(dtype = cast_8_dtype_0, x = input_61_cast_fp16)[name = tensor("cast_10")]; tensor input_449_cast_fp16 = concat(axis = var_10375, interleave = input_449_interleave_0, values = (hidden_states_277_cast_fp16, cast_8))[name = tensor("input_449_cast_fp16")]; tensor reshape_204_shape_0 = const()[name = tensor("reshape_204_shape_0"), val = tensor([2, 32, 30, 64, 64])]; tensor reshape_204_cast_fp16 = reshape(shape = reshape_204_shape_0, x = input_449_cast_fp16)[name = tensor("reshape_204_cast_fp16")]; tensor reduce_mean_153_axes_0 = const()[name = tensor("reduce_mean_153_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_153_keep_dims_0 = const()[name = tensor("reduce_mean_153_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_153_cast_fp16 = reduce_mean(axes = reduce_mean_153_axes_0, keep_dims = reduce_mean_153_keep_dims_0, x = reshape_204_cast_fp16)[name = tensor("reduce_mean_153_cast_fp16")]; tensor sub_102_cast_fp16 = sub(x = reshape_204_cast_fp16, y = reduce_mean_153_cast_fp16)[name = tensor("sub_102_cast_fp16")]; tensor square_51_cast_fp16 = square(x = sub_102_cast_fp16)[name = tensor("square_51_cast_fp16")]; tensor reduce_mean_155_axes_0 = const()[name = tensor("reduce_mean_155_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_155_keep_dims_0 = const()[name = tensor("reduce_mean_155_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_155_cast_fp16 = reduce_mean(axes = reduce_mean_155_axes_0, keep_dims = reduce_mean_155_keep_dims_0, x = square_51_cast_fp16)[name = tensor("reduce_mean_155_cast_fp16")]; tensor add_102_y_0_to_fp16 = const()[name = tensor("add_102_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_102_cast_fp16 = add(x = reduce_mean_155_cast_fp16, y = add_102_y_0_to_fp16)[name = tensor("add_102_cast_fp16")]; tensor sqrt_51_cast_fp16 = sqrt(x = add_102_cast_fp16)[name = tensor("sqrt_51_cast_fp16")]; tensor real_div_51_cast_fp16 = real_div(x = sub_102_cast_fp16, y = sqrt_51_cast_fp16)[name = tensor("real_div_51_cast_fp16")]; tensor reshape_205_shape_0 = const()[name = tensor("reshape_205_shape_0"), val = tensor([2, 960, 64, 64])]; tensor reshape_205_cast_fp16 = reshape(shape = reshape_205_shape_0, x = real_div_51_cast_fp16)[name = tensor("reshape_205_cast_fp16")]; tensor add_103_gamma_0_to_fp16 = const()[name = tensor("add_103_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(793914944)))]; tensor add_103_beta_0_to_fp16 = const()[name = tensor("add_103_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(793916928)))]; tensor add_103_epsilon_0_to_fp16 = const()[name = tensor("add_103_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_103_cast_fp16 = batch_norm(beta = add_103_beta_0_to_fp16, epsilon = add_103_epsilon_0_to_fp16, gamma = add_103_gamma_0_to_fp16, mean = add_97_mean_0_to_fp16, variance = add_97_variance_0_to_fp16, x = reshape_205_cast_fp16)[name = tensor("add_103_cast_fp16")]; tensor input_453_cast_fp16 = silu(x = add_103_cast_fp16)[name = tensor("input_453_cast_fp16")]; tensor hidden_states_279_pad_type_0 = const()[name = tensor("hidden_states_279_pad_type_0"), val = tensor("custom")]; tensor hidden_states_279_pad_0 = const()[name = tensor("hidden_states_279_pad_0"), val = tensor([1, 1, 1, 1])]; tensor hidden_states_279_strides_0 = const()[name = tensor("hidden_states_279_strides_0"), val = tensor([1, 1])]; tensor hidden_states_279_dilations_0 = const()[name = tensor("hidden_states_279_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_279_groups_0 = const()[name = tensor("hidden_states_279_groups_0"), val = tensor(1)]; tensor up_blocks_3_resnets_0_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_3_resnets_0_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(793918912)))]; tensor up_blocks_3_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(799448576)))]; tensor hidden_states_279_cast_fp16 = conv(bias = up_blocks_3_resnets_0_conv1_bias_to_fp16, dilations = hidden_states_279_dilations_0, groups = hidden_states_279_groups_0, pad = hidden_states_279_pad_0, pad_type = hidden_states_279_pad_type_0, strides = hidden_states_279_strides_0, weight = up_blocks_3_resnets_0_conv1_weight_to_fp16, x = input_453_cast_fp16)[name = tensor("hidden_states_279_cast_fp16")]; tensor temb_39_pad_type_0 = const()[name = tensor("temb_39_pad_type_0"), val = tensor("valid")]; tensor temb_39_strides_0 = const()[name = tensor("temb_39_strides_0"), val = tensor([1, 1])]; tensor temb_39_pad_0 = const()[name = tensor("temb_39_pad_0"), val = tensor([0, 0, 0, 0])]; tensor temb_39_dilations_0 = const()[name = tensor("temb_39_dilations_0"), val = tensor([1, 1])]; tensor temb_39_groups_0 = const()[name = tensor("temb_39_groups_0"), val = tensor(1)]; tensor up_blocks_3_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_3_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(799449280)))]; tensor up_blocks_3_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(800268544)))]; tensor temb_39_cast_fp16 = conv(bias = up_blocks_3_resnets_0_time_emb_proj_bias_to_fp16, dilations = temb_39_dilations_0, groups = temb_39_groups_0, pad = temb_39_pad_0, pad_type = temb_39_pad_type_0, strides = temb_39_strides_0, weight = up_blocks_3_resnets_0_time_emb_proj_weight_to_fp16, x = cast_9)[name = tensor("temb_39_cast_fp16")]; tensor input_457_cast_fp16 = add(x = hidden_states_279_cast_fp16, y = temb_39_cast_fp16)[name = tensor("input_457_cast_fp16")]; tensor reshape_208_shape_0 = const()[name = tensor("reshape_208_shape_0"), val = tensor([2, 32, 10, 64, 64])]; tensor reshape_208_cast_fp16 = reshape(shape = reshape_208_shape_0, x = input_457_cast_fp16)[name = tensor("reshape_208_cast_fp16")]; tensor reduce_mean_156_axes_0 = const()[name = tensor("reduce_mean_156_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_156_keep_dims_0 = const()[name = tensor("reduce_mean_156_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_156_cast_fp16 = reduce_mean(axes = reduce_mean_156_axes_0, keep_dims = reduce_mean_156_keep_dims_0, x = reshape_208_cast_fp16)[name = tensor("reduce_mean_156_cast_fp16")]; tensor sub_104_cast_fp16 = sub(x = reshape_208_cast_fp16, y = reduce_mean_156_cast_fp16)[name = tensor("sub_104_cast_fp16")]; tensor square_52_cast_fp16 = square(x = sub_104_cast_fp16)[name = tensor("square_52_cast_fp16")]; tensor reduce_mean_158_axes_0 = const()[name = tensor("reduce_mean_158_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_158_keep_dims_0 = const()[name = tensor("reduce_mean_158_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_158_cast_fp16 = reduce_mean(axes = reduce_mean_158_axes_0, keep_dims = reduce_mean_158_keep_dims_0, x = square_52_cast_fp16)[name = tensor("reduce_mean_158_cast_fp16")]; tensor add_104_y_0_to_fp16 = const()[name = tensor("add_104_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_104_cast_fp16 = add(x = reduce_mean_158_cast_fp16, y = add_104_y_0_to_fp16)[name = tensor("add_104_cast_fp16")]; tensor sqrt_52_cast_fp16 = sqrt(x = add_104_cast_fp16)[name = tensor("sqrt_52_cast_fp16")]; tensor real_div_52_cast_fp16 = real_div(x = sub_104_cast_fp16, y = sqrt_52_cast_fp16)[name = tensor("real_div_52_cast_fp16")]; tensor reshape_209_shape_0 = const()[name = tensor("reshape_209_shape_0"), val = tensor([2, 320, 64, 64])]; tensor reshape_209_cast_fp16 = reshape(shape = reshape_209_shape_0, x = real_div_52_cast_fp16)[name = tensor("reshape_209_cast_fp16")]; tensor add_105_gamma_0_to_fp16 = const()[name = tensor("add_105_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(800269248)))]; tensor add_105_beta_0_to_fp16 = const()[name = tensor("add_105_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(800269952)))]; tensor add_105_epsilon_0_to_fp16 = const()[name = tensor("add_105_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_105_cast_fp16 = batch_norm(beta = add_105_beta_0_to_fp16, epsilon = add_105_epsilon_0_to_fp16, gamma = add_105_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_209_cast_fp16)[name = tensor("add_105_cast_fp16")]; tensor input_461_cast_fp16 = silu(x = add_105_cast_fp16)[name = tensor("input_461_cast_fp16")]; tensor hidden_states_281_pad_type_0 = const()[name = tensor("hidden_states_281_pad_type_0"), val = tensor("custom")]; tensor hidden_states_281_pad_0 = const()[name = tensor("hidden_states_281_pad_0"), val = tensor([1, 1, 1, 1])]; tensor hidden_states_281_strides_0 = const()[name = tensor("hidden_states_281_strides_0"), val = tensor([1, 1])]; tensor hidden_states_281_dilations_0 = const()[name = tensor("hidden_states_281_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_281_groups_0 = const()[name = tensor("hidden_states_281_groups_0"), val = tensor(1)]; tensor up_blocks_3_resnets_0_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_3_resnets_0_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(800270656)))]; tensor up_blocks_3_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(802113920)))]; tensor hidden_states_281_cast_fp16 = conv(bias = up_blocks_3_resnets_0_conv2_bias_to_fp16, dilations = hidden_states_281_dilations_0, groups = hidden_states_281_groups_0, pad = hidden_states_281_pad_0, pad_type = hidden_states_281_pad_type_0, strides = hidden_states_281_strides_0, weight = up_blocks_3_resnets_0_conv2_weight_to_fp16, x = input_461_cast_fp16)[name = tensor("hidden_states_281_cast_fp16")]; tensor x_23_pad_type_0 = const()[name = tensor("x_23_pad_type_0"), val = tensor("valid")]; tensor x_23_strides_0 = const()[name = tensor("x_23_strides_0"), val = tensor([1, 1])]; tensor x_23_pad_0 = const()[name = tensor("x_23_pad_0"), val = tensor([0, 0, 0, 0])]; tensor x_23_dilations_0 = const()[name = tensor("x_23_dilations_0"), val = tensor([1, 1])]; tensor x_23_groups_0 = const()[name = tensor("x_23_groups_0"), val = tensor(1)]; tensor up_blocks_3_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_3_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(802114624)))]; tensor up_blocks_3_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(802729088)))]; tensor x_23_cast_fp16 = conv(bias = up_blocks_3_resnets_0_conv_shortcut_bias_to_fp16, dilations = x_23_dilations_0, groups = x_23_groups_0, pad = x_23_pad_0, pad_type = x_23_pad_type_0, strides = x_23_strides_0, weight = up_blocks_3_resnets_0_conv_shortcut_weight_to_fp16, x = input_449_cast_fp16)[name = tensor("x_23_cast_fp16")]; tensor hidden_states_283_cast_fp16 = add(x = x_23_cast_fp16, y = hidden_states_281_cast_fp16)[name = tensor("hidden_states_283_cast_fp16")]; tensor reshape_212_shape_0 = const()[name = tensor("reshape_212_shape_0"), val = tensor([2, 32, 10, 64, 64])]; tensor reshape_212_cast_fp16 = reshape(shape = reshape_212_shape_0, x = hidden_states_283_cast_fp16)[name = tensor("reshape_212_cast_fp16")]; tensor reduce_mean_159_axes_0 = const()[name = tensor("reduce_mean_159_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_159_keep_dims_0 = const()[name = tensor("reduce_mean_159_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_159_cast_fp16 = reduce_mean(axes = reduce_mean_159_axes_0, keep_dims = reduce_mean_159_keep_dims_0, x = reshape_212_cast_fp16)[name = tensor("reduce_mean_159_cast_fp16")]; tensor sub_106_cast_fp16 = sub(x = reshape_212_cast_fp16, y = reduce_mean_159_cast_fp16)[name = tensor("sub_106_cast_fp16")]; tensor square_53_cast_fp16 = square(x = sub_106_cast_fp16)[name = tensor("square_53_cast_fp16")]; tensor reduce_mean_161_axes_0 = const()[name = tensor("reduce_mean_161_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_161_keep_dims_0 = const()[name = tensor("reduce_mean_161_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_161_cast_fp16 = reduce_mean(axes = reduce_mean_161_axes_0, keep_dims = reduce_mean_161_keep_dims_0, x = square_53_cast_fp16)[name = tensor("reduce_mean_161_cast_fp16")]; tensor add_106_y_0_to_fp16 = const()[name = tensor("add_106_y_0_to_fp16"), val = tensor(0x1.1p-20)]; tensor add_106_cast_fp16 = add(x = reduce_mean_161_cast_fp16, y = add_106_y_0_to_fp16)[name = tensor("add_106_cast_fp16")]; tensor sqrt_53_cast_fp16 = sqrt(x = add_106_cast_fp16)[name = tensor("sqrt_53_cast_fp16")]; tensor real_div_53_cast_fp16 = real_div(x = sub_106_cast_fp16, y = sqrt_53_cast_fp16)[name = tensor("real_div_53_cast_fp16")]; tensor reshape_213_shape_0 = const()[name = tensor("reshape_213_shape_0"), val = tensor([2, 320, 64, 64])]; tensor reshape_213_cast_fp16 = reshape(shape = reshape_213_shape_0, x = real_div_53_cast_fp16)[name = tensor("reshape_213_cast_fp16")]; tensor add_107_gamma_0_to_fp16 = const()[name = tensor("add_107_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(802729792)))]; tensor add_107_beta_0_to_fp16 = const()[name = tensor("add_107_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(802730496)))]; tensor add_107_epsilon_0_to_fp16 = const()[name = tensor("add_107_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_107_cast_fp16 = batch_norm(beta = add_107_beta_0_to_fp16, epsilon = add_107_epsilon_0_to_fp16, gamma = add_107_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_213_cast_fp16)[name = tensor("add_107_cast_fp16")]; tensor hidden_states_285_pad_type_0 = const()[name = tensor("hidden_states_285_pad_type_0"), val = tensor("valid")]; tensor hidden_states_285_strides_0 = const()[name = tensor("hidden_states_285_strides_0"), val = tensor([1, 1])]; tensor hidden_states_285_pad_0 = const()[name = tensor("hidden_states_285_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_285_dilations_0 = const()[name = tensor("hidden_states_285_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_285_groups_0 = const()[name = tensor("hidden_states_285_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(802731200)))]; tensor up_blocks_3_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(802936064)))]; tensor hidden_states_285_cast_fp16 = conv(bias = up_blocks_3_attentions_0_proj_in_bias_to_fp16, dilations = hidden_states_285_dilations_0, groups = hidden_states_285_groups_0, pad = hidden_states_285_pad_0, pad_type = hidden_states_285_pad_type_0, strides = hidden_states_285_strides_0, weight = up_blocks_3_attentions_0_proj_in_weight_to_fp16, x = add_107_cast_fp16)[name = tensor("hidden_states_285_cast_fp16")]; tensor var_10458 = const()[name = tensor("op_10458"), val = tensor([2, 320, 1, 4096])]; tensor inputs_79_cast_fp16 = reshape(shape = var_10458, x = hidden_states_285_cast_fp16)[name = tensor("inputs_79_cast_fp16")]; tensor hidden_states_287_axes_0 = const()[name = tensor("hidden_states_287_axes_0"), val = tensor([1])]; tensor hidden_states_287_gamma_0_to_fp16 = const()[name = tensor("hidden_states_287_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(802936768)))]; tensor hidden_states_287_beta_0_to_fp16 = const()[name = tensor("hidden_states_287_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(802937472)))]; tensor var_10474_to_fp16 = const()[name = tensor("op_10474_to_fp16"), val = tensor(0x1.5p-17)]; tensor hidden_states_287_cast_fp16 = layer_norm(axes = hidden_states_287_axes_0, beta = hidden_states_287_beta_0_to_fp16, epsilon = var_10474_to_fp16, gamma = hidden_states_287_gamma_0_to_fp16, x = inputs_79_cast_fp16)[name = tensor("hidden_states_287_cast_fp16")]; tensor q_53_pad_type_0 = const()[name = tensor("q_53_pad_type_0"), val = tensor("valid")]; tensor q_53_strides_0 = const()[name = tensor("q_53_strides_0"), val = tensor([1, 1])]; tensor q_53_pad_0 = const()[name = tensor("q_53_pad_0"), val = tensor([0, 0, 0, 0])]; tensor q_53_dilations_0 = const()[name = tensor("q_53_dilations_0"), val = tensor([1, 1])]; tensor q_53_groups_0 = const()[name = tensor("q_53_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(802938176)))]; tensor q_53_cast_fp16 = conv(dilations = q_53_dilations_0, groups = q_53_groups_0, pad = q_53_pad_0, pad_type = q_53_pad_type_0, strides = q_53_strides_0, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_287_cast_fp16)[name = tensor("q_53_cast_fp16")]; tensor k_105_pad_type_0 = const()[name = tensor("k_105_pad_type_0"), val = tensor("valid")]; tensor k_105_strides_0 = const()[name = tensor("k_105_strides_0"), val = tensor([1, 1])]; tensor k_105_pad_0 = const()[name = tensor("k_105_pad_0"), val = tensor([0, 0, 0, 0])]; tensor k_105_dilations_0 = const()[name = tensor("k_105_dilations_0"), val = tensor([1, 1])]; tensor k_105_groups_0 = const()[name = tensor("k_105_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(803143040)))]; tensor k_105_cast_fp16 = conv(dilations = k_105_dilations_0, groups = k_105_groups_0, pad = k_105_pad_0, pad_type = k_105_pad_type_0, strides = k_105_strides_0, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_287_cast_fp16)[name = tensor("k_105_cast_fp16")]; tensor v_53_pad_type_0 = const()[name = tensor("v_53_pad_type_0"), val = tensor("valid")]; tensor v_53_strides_0 = const()[name = tensor("v_53_strides_0"), val = tensor([1, 1])]; tensor v_53_pad_0 = const()[name = tensor("v_53_pad_0"), val = tensor([0, 0, 0, 0])]; tensor v_53_dilations_0 = const()[name = tensor("v_53_dilations_0"), val = tensor([1, 1])]; tensor v_53_groups_0 = const()[name = tensor("v_53_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(803347904)))]; tensor v_53_cast_fp16 = conv(dilations = v_53_dilations_0, groups = v_53_groups_0, pad = v_53_pad_0, pad_type = v_53_pad_type_0, strides = v_53_strides_0, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_287_cast_fp16)[name = tensor("v_53_cast_fp16")]; tensor var_10507_begin_0 = const()[name = tensor("op_10507_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10507_end_0 = const()[name = tensor("op_10507_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_10507_end_mask_0 = const()[name = tensor("op_10507_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10507_cast_fp16 = slice_by_index(begin = var_10507_begin_0, end = var_10507_end_0, end_mask = var_10507_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10507_cast_fp16")]; tensor var_10511_begin_0 = const()[name = tensor("op_10511_begin_0"), val = tensor([0, 40, 0, 0])]; tensor var_10511_end_0 = const()[name = tensor("op_10511_end_0"), val = tensor([2, 80, 1, 4096])]; tensor var_10511_end_mask_0 = const()[name = tensor("op_10511_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10511_cast_fp16 = slice_by_index(begin = var_10511_begin_0, end = var_10511_end_0, end_mask = var_10511_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10511_cast_fp16")]; tensor var_10515_begin_0 = const()[name = tensor("op_10515_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_10515_end_0 = const()[name = tensor("op_10515_end_0"), val = tensor([2, 120, 1, 4096])]; tensor var_10515_end_mask_0 = const()[name = tensor("op_10515_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10515_cast_fp16 = slice_by_index(begin = var_10515_begin_0, end = var_10515_end_0, end_mask = var_10515_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10515_cast_fp16")]; tensor var_10519_begin_0 = const()[name = tensor("op_10519_begin_0"), val = tensor([0, 120, 0, 0])]; tensor var_10519_end_0 = const()[name = tensor("op_10519_end_0"), val = tensor([2, 160, 1, 4096])]; tensor var_10519_end_mask_0 = const()[name = tensor("op_10519_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10519_cast_fp16 = slice_by_index(begin = var_10519_begin_0, end = var_10519_end_0, end_mask = var_10519_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10519_cast_fp16")]; tensor var_10523_begin_0 = const()[name = tensor("op_10523_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_10523_end_0 = const()[name = tensor("op_10523_end_0"), val = tensor([2, 200, 1, 4096])]; tensor var_10523_end_mask_0 = const()[name = tensor("op_10523_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10523_cast_fp16 = slice_by_index(begin = var_10523_begin_0, end = var_10523_end_0, end_mask = var_10523_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10523_cast_fp16")]; tensor var_10527_begin_0 = const()[name = tensor("op_10527_begin_0"), val = tensor([0, 200, 0, 0])]; tensor var_10527_end_0 = const()[name = tensor("op_10527_end_0"), val = tensor([2, 240, 1, 4096])]; tensor var_10527_end_mask_0 = const()[name = tensor("op_10527_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10527_cast_fp16 = slice_by_index(begin = var_10527_begin_0, end = var_10527_end_0, end_mask = var_10527_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10527_cast_fp16")]; tensor var_10531_begin_0 = const()[name = tensor("op_10531_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_10531_end_0 = const()[name = tensor("op_10531_end_0"), val = tensor([2, 280, 1, 4096])]; tensor var_10531_end_mask_0 = const()[name = tensor("op_10531_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10531_cast_fp16 = slice_by_index(begin = var_10531_begin_0, end = var_10531_end_0, end_mask = var_10531_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10531_cast_fp16")]; tensor var_10535_begin_0 = const()[name = tensor("op_10535_begin_0"), val = tensor([0, 280, 0, 0])]; tensor var_10535_end_0 = const()[name = tensor("op_10535_end_0"), val = tensor([2, 1, 1, 4096])]; tensor var_10535_end_mask_0 = const()[name = tensor("op_10535_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_10535_cast_fp16 = slice_by_index(begin = var_10535_begin_0, end = var_10535_end_0, end_mask = var_10535_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10535_cast_fp16")]; tensor var_10538_begin_0 = const()[name = tensor("op_10538_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10538_end_0 = const()[name = tensor("op_10538_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_10538_end_mask_0 = const()[name = tensor("op_10538_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10538_cast_fp16 = slice_by_index(begin = var_10538_begin_0, end = var_10538_end_0, end_mask = var_10538_end_mask_0, x = var_10507_cast_fp16)[name = tensor("op_10538_cast_fp16")]; tensor var_10539_begin_0 = const()[name = tensor("op_10539_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_10539_end_0 = const()[name = tensor("op_10539_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_10539_end_mask_0 = const()[name = tensor("op_10539_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10539_cast_fp16 = slice_by_index(begin = var_10539_begin_0, end = var_10539_end_0, end_mask = var_10539_end_mask_0, x = var_10507_cast_fp16)[name = tensor("op_10539_cast_fp16")]; tensor var_10540_begin_0 = const()[name = tensor("op_10540_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_10540_end_0 = const()[name = tensor("op_10540_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_10540_end_mask_0 = const()[name = tensor("op_10540_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10540_cast_fp16 = slice_by_index(begin = var_10540_begin_0, end = var_10540_end_0, end_mask = var_10540_end_mask_0, x = var_10507_cast_fp16)[name = tensor("op_10540_cast_fp16")]; tensor var_10541_begin_0 = const()[name = tensor("op_10541_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_10541_end_0 = const()[name = tensor("op_10541_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_10541_end_mask_0 = const()[name = tensor("op_10541_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10541_cast_fp16 = slice_by_index(begin = var_10541_begin_0, end = var_10541_end_0, end_mask = var_10541_end_mask_0, x = var_10507_cast_fp16)[name = tensor("op_10541_cast_fp16")]; tensor var_10542_begin_0 = const()[name = tensor("op_10542_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_10542_end_0 = const()[name = tensor("op_10542_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_10542_end_mask_0 = const()[name = tensor("op_10542_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10542_cast_fp16 = slice_by_index(begin = var_10542_begin_0, end = var_10542_end_0, end_mask = var_10542_end_mask_0, x = var_10507_cast_fp16)[name = tensor("op_10542_cast_fp16")]; tensor var_10543_begin_0 = const()[name = tensor("op_10543_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_10543_end_0 = const()[name = tensor("op_10543_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_10543_end_mask_0 = const()[name = tensor("op_10543_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10543_cast_fp16 = slice_by_index(begin = var_10543_begin_0, end = var_10543_end_0, end_mask = var_10543_end_mask_0, x = var_10507_cast_fp16)[name = tensor("op_10543_cast_fp16")]; tensor var_10544_begin_0 = const()[name = tensor("op_10544_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_10544_end_0 = const()[name = tensor("op_10544_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_10544_end_mask_0 = const()[name = tensor("op_10544_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10544_cast_fp16 = slice_by_index(begin = var_10544_begin_0, end = var_10544_end_0, end_mask = var_10544_end_mask_0, x = var_10507_cast_fp16)[name = tensor("op_10544_cast_fp16")]; tensor var_10545_begin_0 = const()[name = tensor("op_10545_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_10545_end_0 = const()[name = tensor("op_10545_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_10545_end_mask_0 = const()[name = tensor("op_10545_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_10545_cast_fp16 = slice_by_index(begin = var_10545_begin_0, end = var_10545_end_0, end_mask = var_10545_end_mask_0, x = var_10507_cast_fp16)[name = tensor("op_10545_cast_fp16")]; tensor var_10546_begin_0 = const()[name = tensor("op_10546_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10546_end_0 = const()[name = tensor("op_10546_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_10546_end_mask_0 = const()[name = tensor("op_10546_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10546_cast_fp16 = slice_by_index(begin = var_10546_begin_0, end = var_10546_end_0, end_mask = var_10546_end_mask_0, x = var_10511_cast_fp16)[name = tensor("op_10546_cast_fp16")]; tensor var_10547_begin_0 = const()[name = tensor("op_10547_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_10547_end_0 = const()[name = tensor("op_10547_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_10547_end_mask_0 = const()[name = tensor("op_10547_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10547_cast_fp16 = slice_by_index(begin = var_10547_begin_0, end = var_10547_end_0, end_mask = var_10547_end_mask_0, x = var_10511_cast_fp16)[name = tensor("op_10547_cast_fp16")]; tensor var_10548_begin_0 = const()[name = tensor("op_10548_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_10548_end_0 = const()[name = tensor("op_10548_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_10548_end_mask_0 = const()[name = tensor("op_10548_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10548_cast_fp16 = slice_by_index(begin = var_10548_begin_0, end = var_10548_end_0, end_mask = var_10548_end_mask_0, x = var_10511_cast_fp16)[name = tensor("op_10548_cast_fp16")]; tensor var_10549_begin_0 = const()[name = tensor("op_10549_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_10549_end_0 = const()[name = tensor("op_10549_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_10549_end_mask_0 = const()[name = tensor("op_10549_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10549_cast_fp16 = slice_by_index(begin = var_10549_begin_0, end = var_10549_end_0, end_mask = var_10549_end_mask_0, x = var_10511_cast_fp16)[name = tensor("op_10549_cast_fp16")]; tensor var_10550_begin_0 = const()[name = tensor("op_10550_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_10550_end_0 = const()[name = tensor("op_10550_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_10550_end_mask_0 = const()[name = tensor("op_10550_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10550_cast_fp16 = slice_by_index(begin = var_10550_begin_0, end = var_10550_end_0, end_mask = var_10550_end_mask_0, x = var_10511_cast_fp16)[name = tensor("op_10550_cast_fp16")]; tensor var_10551_begin_0 = const()[name = tensor("op_10551_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_10551_end_0 = const()[name = tensor("op_10551_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_10551_end_mask_0 = const()[name = tensor("op_10551_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10551_cast_fp16 = slice_by_index(begin = var_10551_begin_0, end = var_10551_end_0, end_mask = var_10551_end_mask_0, x = var_10511_cast_fp16)[name = tensor("op_10551_cast_fp16")]; tensor var_10552_begin_0 = const()[name = tensor("op_10552_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_10552_end_0 = const()[name = tensor("op_10552_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_10552_end_mask_0 = const()[name = tensor("op_10552_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10552_cast_fp16 = slice_by_index(begin = var_10552_begin_0, end = var_10552_end_0, end_mask = var_10552_end_mask_0, x = var_10511_cast_fp16)[name = tensor("op_10552_cast_fp16")]; tensor var_10553_begin_0 = const()[name = tensor("op_10553_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_10553_end_0 = const()[name = tensor("op_10553_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_10553_end_mask_0 = const()[name = tensor("op_10553_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_10553_cast_fp16 = slice_by_index(begin = var_10553_begin_0, end = var_10553_end_0, end_mask = var_10553_end_mask_0, x = var_10511_cast_fp16)[name = tensor("op_10553_cast_fp16")]; tensor var_10554_begin_0 = const()[name = tensor("op_10554_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10554_end_0 = const()[name = tensor("op_10554_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_10554_end_mask_0 = const()[name = tensor("op_10554_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10554_cast_fp16 = slice_by_index(begin = var_10554_begin_0, end = var_10554_end_0, end_mask = var_10554_end_mask_0, x = var_10515_cast_fp16)[name = tensor("op_10554_cast_fp16")]; tensor var_10555_begin_0 = const()[name = tensor("op_10555_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_10555_end_0 = const()[name = tensor("op_10555_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_10555_end_mask_0 = const()[name = tensor("op_10555_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10555_cast_fp16 = slice_by_index(begin = var_10555_begin_0, end = var_10555_end_0, end_mask = var_10555_end_mask_0, x = var_10515_cast_fp16)[name = tensor("op_10555_cast_fp16")]; tensor var_10556_begin_0 = const()[name = tensor("op_10556_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_10556_end_0 = const()[name = tensor("op_10556_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_10556_end_mask_0 = const()[name = tensor("op_10556_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10556_cast_fp16 = slice_by_index(begin = var_10556_begin_0, end = var_10556_end_0, end_mask = var_10556_end_mask_0, x = var_10515_cast_fp16)[name = tensor("op_10556_cast_fp16")]; tensor var_10557_begin_0 = const()[name = tensor("op_10557_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_10557_end_0 = const()[name = tensor("op_10557_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_10557_end_mask_0 = const()[name = tensor("op_10557_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10557_cast_fp16 = slice_by_index(begin = var_10557_begin_0, end = var_10557_end_0, end_mask = var_10557_end_mask_0, x = var_10515_cast_fp16)[name = tensor("op_10557_cast_fp16")]; tensor var_10558_begin_0 = const()[name = tensor("op_10558_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_10558_end_0 = const()[name = tensor("op_10558_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_10558_end_mask_0 = const()[name = tensor("op_10558_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10558_cast_fp16 = slice_by_index(begin = var_10558_begin_0, end = var_10558_end_0, end_mask = var_10558_end_mask_0, x = var_10515_cast_fp16)[name = tensor("op_10558_cast_fp16")]; tensor var_10559_begin_0 = const()[name = tensor("op_10559_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_10559_end_0 = const()[name = tensor("op_10559_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_10559_end_mask_0 = const()[name = tensor("op_10559_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10559_cast_fp16 = slice_by_index(begin = var_10559_begin_0, end = var_10559_end_0, end_mask = var_10559_end_mask_0, x = var_10515_cast_fp16)[name = tensor("op_10559_cast_fp16")]; tensor var_10560_begin_0 = const()[name = tensor("op_10560_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_10560_end_0 = const()[name = tensor("op_10560_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_10560_end_mask_0 = const()[name = tensor("op_10560_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10560_cast_fp16 = slice_by_index(begin = var_10560_begin_0, end = var_10560_end_0, end_mask = var_10560_end_mask_0, x = var_10515_cast_fp16)[name = tensor("op_10560_cast_fp16")]; tensor var_10561_begin_0 = const()[name = tensor("op_10561_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_10561_end_0 = const()[name = tensor("op_10561_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_10561_end_mask_0 = const()[name = tensor("op_10561_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_10561_cast_fp16 = slice_by_index(begin = var_10561_begin_0, end = var_10561_end_0, end_mask = var_10561_end_mask_0, x = var_10515_cast_fp16)[name = tensor("op_10561_cast_fp16")]; tensor var_10562_begin_0 = const()[name = tensor("op_10562_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10562_end_0 = const()[name = tensor("op_10562_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_10562_end_mask_0 = const()[name = tensor("op_10562_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10562_cast_fp16 = slice_by_index(begin = var_10562_begin_0, end = var_10562_end_0, end_mask = var_10562_end_mask_0, x = var_10519_cast_fp16)[name = tensor("op_10562_cast_fp16")]; tensor var_10563_begin_0 = const()[name = tensor("op_10563_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_10563_end_0 = const()[name = tensor("op_10563_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_10563_end_mask_0 = const()[name = tensor("op_10563_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10563_cast_fp16 = slice_by_index(begin = var_10563_begin_0, end = var_10563_end_0, end_mask = var_10563_end_mask_0, x = var_10519_cast_fp16)[name = tensor("op_10563_cast_fp16")]; tensor var_10564_begin_0 = const()[name = tensor("op_10564_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_10564_end_0 = const()[name = tensor("op_10564_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_10564_end_mask_0 = const()[name = tensor("op_10564_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10564_cast_fp16 = slice_by_index(begin = var_10564_begin_0, end = var_10564_end_0, end_mask = var_10564_end_mask_0, x = var_10519_cast_fp16)[name = tensor("op_10564_cast_fp16")]; tensor var_10565_begin_0 = const()[name = tensor("op_10565_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_10565_end_0 = const()[name = tensor("op_10565_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_10565_end_mask_0 = const()[name = tensor("op_10565_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10565_cast_fp16 = slice_by_index(begin = var_10565_begin_0, end = var_10565_end_0, end_mask = var_10565_end_mask_0, x = var_10519_cast_fp16)[name = tensor("op_10565_cast_fp16")]; tensor var_10566_begin_0 = const()[name = tensor("op_10566_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_10566_end_0 = const()[name = tensor("op_10566_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_10566_end_mask_0 = const()[name = tensor("op_10566_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10566_cast_fp16 = slice_by_index(begin = var_10566_begin_0, end = var_10566_end_0, end_mask = var_10566_end_mask_0, x = var_10519_cast_fp16)[name = tensor("op_10566_cast_fp16")]; tensor var_10567_begin_0 = const()[name = tensor("op_10567_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_10567_end_0 = const()[name = tensor("op_10567_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_10567_end_mask_0 = const()[name = tensor("op_10567_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10567_cast_fp16 = slice_by_index(begin = var_10567_begin_0, end = var_10567_end_0, end_mask = var_10567_end_mask_0, x = var_10519_cast_fp16)[name = tensor("op_10567_cast_fp16")]; tensor var_10568_begin_0 = const()[name = tensor("op_10568_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_10568_end_0 = const()[name = tensor("op_10568_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_10568_end_mask_0 = const()[name = tensor("op_10568_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10568_cast_fp16 = slice_by_index(begin = var_10568_begin_0, end = var_10568_end_0, end_mask = var_10568_end_mask_0, x = var_10519_cast_fp16)[name = tensor("op_10568_cast_fp16")]; tensor var_10569_begin_0 = const()[name = tensor("op_10569_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_10569_end_0 = const()[name = tensor("op_10569_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_10569_end_mask_0 = const()[name = tensor("op_10569_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_10569_cast_fp16 = slice_by_index(begin = var_10569_begin_0, end = var_10569_end_0, end_mask = var_10569_end_mask_0, x = var_10519_cast_fp16)[name = tensor("op_10569_cast_fp16")]; tensor var_10570_begin_0 = const()[name = tensor("op_10570_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10570_end_0 = const()[name = tensor("op_10570_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_10570_end_mask_0 = const()[name = tensor("op_10570_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10570_cast_fp16 = slice_by_index(begin = var_10570_begin_0, end = var_10570_end_0, end_mask = var_10570_end_mask_0, x = var_10523_cast_fp16)[name = tensor("op_10570_cast_fp16")]; tensor var_10571_begin_0 = const()[name = tensor("op_10571_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_10571_end_0 = const()[name = tensor("op_10571_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_10571_end_mask_0 = const()[name = tensor("op_10571_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10571_cast_fp16 = slice_by_index(begin = var_10571_begin_0, end = var_10571_end_0, end_mask = var_10571_end_mask_0, x = var_10523_cast_fp16)[name = tensor("op_10571_cast_fp16")]; tensor var_10572_begin_0 = const()[name = tensor("op_10572_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_10572_end_0 = const()[name = tensor("op_10572_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_10572_end_mask_0 = const()[name = tensor("op_10572_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10572_cast_fp16 = slice_by_index(begin = var_10572_begin_0, end = var_10572_end_0, end_mask = var_10572_end_mask_0, x = var_10523_cast_fp16)[name = tensor("op_10572_cast_fp16")]; tensor var_10573_begin_0 = const()[name = tensor("op_10573_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_10573_end_0 = const()[name = tensor("op_10573_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_10573_end_mask_0 = const()[name = tensor("op_10573_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10573_cast_fp16 = slice_by_index(begin = var_10573_begin_0, end = var_10573_end_0, end_mask = var_10573_end_mask_0, x = var_10523_cast_fp16)[name = tensor("op_10573_cast_fp16")]; tensor var_10574_begin_0 = const()[name = tensor("op_10574_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_10574_end_0 = const()[name = tensor("op_10574_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_10574_end_mask_0 = const()[name = tensor("op_10574_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10574_cast_fp16 = slice_by_index(begin = var_10574_begin_0, end = var_10574_end_0, end_mask = var_10574_end_mask_0, x = var_10523_cast_fp16)[name = tensor("op_10574_cast_fp16")]; tensor var_10575_begin_0 = const()[name = tensor("op_10575_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_10575_end_0 = const()[name = tensor("op_10575_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_10575_end_mask_0 = const()[name = tensor("op_10575_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10575_cast_fp16 = slice_by_index(begin = var_10575_begin_0, end = var_10575_end_0, end_mask = var_10575_end_mask_0, x = var_10523_cast_fp16)[name = tensor("op_10575_cast_fp16")]; tensor var_10576_begin_0 = const()[name = tensor("op_10576_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_10576_end_0 = const()[name = tensor("op_10576_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_10576_end_mask_0 = const()[name = tensor("op_10576_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10576_cast_fp16 = slice_by_index(begin = var_10576_begin_0, end = var_10576_end_0, end_mask = var_10576_end_mask_0, x = var_10523_cast_fp16)[name = tensor("op_10576_cast_fp16")]; tensor var_10577_begin_0 = const()[name = tensor("op_10577_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_10577_end_0 = const()[name = tensor("op_10577_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_10577_end_mask_0 = const()[name = tensor("op_10577_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_10577_cast_fp16 = slice_by_index(begin = var_10577_begin_0, end = var_10577_end_0, end_mask = var_10577_end_mask_0, x = var_10523_cast_fp16)[name = tensor("op_10577_cast_fp16")]; tensor var_10578_begin_0 = const()[name = tensor("op_10578_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10578_end_0 = const()[name = tensor("op_10578_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_10578_end_mask_0 = const()[name = tensor("op_10578_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10578_cast_fp16 = slice_by_index(begin = var_10578_begin_0, end = var_10578_end_0, end_mask = var_10578_end_mask_0, x = var_10527_cast_fp16)[name = tensor("op_10578_cast_fp16")]; tensor var_10579_begin_0 = const()[name = tensor("op_10579_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_10579_end_0 = const()[name = tensor("op_10579_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_10579_end_mask_0 = const()[name = tensor("op_10579_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10579_cast_fp16 = slice_by_index(begin = var_10579_begin_0, end = var_10579_end_0, end_mask = var_10579_end_mask_0, x = var_10527_cast_fp16)[name = tensor("op_10579_cast_fp16")]; tensor var_10580_begin_0 = const()[name = tensor("op_10580_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_10580_end_0 = const()[name = tensor("op_10580_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_10580_end_mask_0 = const()[name = tensor("op_10580_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10580_cast_fp16 = slice_by_index(begin = var_10580_begin_0, end = var_10580_end_0, end_mask = var_10580_end_mask_0, x = var_10527_cast_fp16)[name = tensor("op_10580_cast_fp16")]; tensor var_10581_begin_0 = const()[name = tensor("op_10581_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_10581_end_0 = const()[name = tensor("op_10581_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_10581_end_mask_0 = const()[name = tensor("op_10581_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10581_cast_fp16 = slice_by_index(begin = var_10581_begin_0, end = var_10581_end_0, end_mask = var_10581_end_mask_0, x = var_10527_cast_fp16)[name = tensor("op_10581_cast_fp16")]; tensor var_10582_begin_0 = const()[name = tensor("op_10582_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_10582_end_0 = const()[name = tensor("op_10582_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_10582_end_mask_0 = const()[name = tensor("op_10582_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10582_cast_fp16 = slice_by_index(begin = var_10582_begin_0, end = var_10582_end_0, end_mask = var_10582_end_mask_0, x = var_10527_cast_fp16)[name = tensor("op_10582_cast_fp16")]; tensor var_10583_begin_0 = const()[name = tensor("op_10583_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_10583_end_0 = const()[name = tensor("op_10583_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_10583_end_mask_0 = const()[name = tensor("op_10583_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10583_cast_fp16 = slice_by_index(begin = var_10583_begin_0, end = var_10583_end_0, end_mask = var_10583_end_mask_0, x = var_10527_cast_fp16)[name = tensor("op_10583_cast_fp16")]; tensor var_10584_begin_0 = const()[name = tensor("op_10584_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_10584_end_0 = const()[name = tensor("op_10584_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_10584_end_mask_0 = const()[name = tensor("op_10584_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10584_cast_fp16 = slice_by_index(begin = var_10584_begin_0, end = var_10584_end_0, end_mask = var_10584_end_mask_0, x = var_10527_cast_fp16)[name = tensor("op_10584_cast_fp16")]; tensor var_10585_begin_0 = const()[name = tensor("op_10585_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_10585_end_0 = const()[name = tensor("op_10585_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_10585_end_mask_0 = const()[name = tensor("op_10585_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_10585_cast_fp16 = slice_by_index(begin = var_10585_begin_0, end = var_10585_end_0, end_mask = var_10585_end_mask_0, x = var_10527_cast_fp16)[name = tensor("op_10585_cast_fp16")]; tensor var_10586_begin_0 = const()[name = tensor("op_10586_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10586_end_0 = const()[name = tensor("op_10586_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_10586_end_mask_0 = const()[name = tensor("op_10586_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10586_cast_fp16 = slice_by_index(begin = var_10586_begin_0, end = var_10586_end_0, end_mask = var_10586_end_mask_0, x = var_10531_cast_fp16)[name = tensor("op_10586_cast_fp16")]; tensor var_10587_begin_0 = const()[name = tensor("op_10587_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_10587_end_0 = const()[name = tensor("op_10587_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_10587_end_mask_0 = const()[name = tensor("op_10587_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10587_cast_fp16 = slice_by_index(begin = var_10587_begin_0, end = var_10587_end_0, end_mask = var_10587_end_mask_0, x = var_10531_cast_fp16)[name = tensor("op_10587_cast_fp16")]; tensor var_10588_begin_0 = const()[name = tensor("op_10588_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_10588_end_0 = const()[name = tensor("op_10588_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_10588_end_mask_0 = const()[name = tensor("op_10588_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10588_cast_fp16 = slice_by_index(begin = var_10588_begin_0, end = var_10588_end_0, end_mask = var_10588_end_mask_0, x = var_10531_cast_fp16)[name = tensor("op_10588_cast_fp16")]; tensor var_10589_begin_0 = const()[name = tensor("op_10589_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_10589_end_0 = const()[name = tensor("op_10589_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_10589_end_mask_0 = const()[name = tensor("op_10589_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10589_cast_fp16 = slice_by_index(begin = var_10589_begin_0, end = var_10589_end_0, end_mask = var_10589_end_mask_0, x = var_10531_cast_fp16)[name = tensor("op_10589_cast_fp16")]; tensor var_10590_begin_0 = const()[name = tensor("op_10590_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_10590_end_0 = const()[name = tensor("op_10590_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_10590_end_mask_0 = const()[name = tensor("op_10590_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10590_cast_fp16 = slice_by_index(begin = var_10590_begin_0, end = var_10590_end_0, end_mask = var_10590_end_mask_0, x = var_10531_cast_fp16)[name = tensor("op_10590_cast_fp16")]; tensor var_10591_begin_0 = const()[name = tensor("op_10591_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_10591_end_0 = const()[name = tensor("op_10591_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_10591_end_mask_0 = const()[name = tensor("op_10591_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10591_cast_fp16 = slice_by_index(begin = var_10591_begin_0, end = var_10591_end_0, end_mask = var_10591_end_mask_0, x = var_10531_cast_fp16)[name = tensor("op_10591_cast_fp16")]; tensor var_10592_begin_0 = const()[name = tensor("op_10592_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_10592_end_0 = const()[name = tensor("op_10592_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_10592_end_mask_0 = const()[name = tensor("op_10592_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10592_cast_fp16 = slice_by_index(begin = var_10592_begin_0, end = var_10592_end_0, end_mask = var_10592_end_mask_0, x = var_10531_cast_fp16)[name = tensor("op_10592_cast_fp16")]; tensor var_10593_begin_0 = const()[name = tensor("op_10593_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_10593_end_0 = const()[name = tensor("op_10593_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_10593_end_mask_0 = const()[name = tensor("op_10593_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_10593_cast_fp16 = slice_by_index(begin = var_10593_begin_0, end = var_10593_end_0, end_mask = var_10593_end_mask_0, x = var_10531_cast_fp16)[name = tensor("op_10593_cast_fp16")]; tensor var_10594_begin_0 = const()[name = tensor("op_10594_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10594_end_0 = const()[name = tensor("op_10594_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_10594_end_mask_0 = const()[name = tensor("op_10594_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10594_cast_fp16 = slice_by_index(begin = var_10594_begin_0, end = var_10594_end_0, end_mask = var_10594_end_mask_0, x = var_10535_cast_fp16)[name = tensor("op_10594_cast_fp16")]; tensor var_10595_begin_0 = const()[name = tensor("op_10595_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_10595_end_0 = const()[name = tensor("op_10595_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_10595_end_mask_0 = const()[name = tensor("op_10595_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10595_cast_fp16 = slice_by_index(begin = var_10595_begin_0, end = var_10595_end_0, end_mask = var_10595_end_mask_0, x = var_10535_cast_fp16)[name = tensor("op_10595_cast_fp16")]; tensor var_10596_begin_0 = const()[name = tensor("op_10596_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_10596_end_0 = const()[name = tensor("op_10596_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_10596_end_mask_0 = const()[name = tensor("op_10596_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10596_cast_fp16 = slice_by_index(begin = var_10596_begin_0, end = var_10596_end_0, end_mask = var_10596_end_mask_0, x = var_10535_cast_fp16)[name = tensor("op_10596_cast_fp16")]; tensor var_10597_begin_0 = const()[name = tensor("op_10597_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_10597_end_0 = const()[name = tensor("op_10597_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_10597_end_mask_0 = const()[name = tensor("op_10597_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10597_cast_fp16 = slice_by_index(begin = var_10597_begin_0, end = var_10597_end_0, end_mask = var_10597_end_mask_0, x = var_10535_cast_fp16)[name = tensor("op_10597_cast_fp16")]; tensor var_10598_begin_0 = const()[name = tensor("op_10598_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_10598_end_0 = const()[name = tensor("op_10598_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_10598_end_mask_0 = const()[name = tensor("op_10598_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10598_cast_fp16 = slice_by_index(begin = var_10598_begin_0, end = var_10598_end_0, end_mask = var_10598_end_mask_0, x = var_10535_cast_fp16)[name = tensor("op_10598_cast_fp16")]; tensor var_10599_begin_0 = const()[name = tensor("op_10599_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_10599_end_0 = const()[name = tensor("op_10599_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_10599_end_mask_0 = const()[name = tensor("op_10599_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10599_cast_fp16 = slice_by_index(begin = var_10599_begin_0, end = var_10599_end_0, end_mask = var_10599_end_mask_0, x = var_10535_cast_fp16)[name = tensor("op_10599_cast_fp16")]; tensor var_10600_begin_0 = const()[name = tensor("op_10600_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_10600_end_0 = const()[name = tensor("op_10600_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_10600_end_mask_0 = const()[name = tensor("op_10600_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10600_cast_fp16 = slice_by_index(begin = var_10600_begin_0, end = var_10600_end_0, end_mask = var_10600_end_mask_0, x = var_10535_cast_fp16)[name = tensor("op_10600_cast_fp16")]; tensor var_10601_begin_0 = const()[name = tensor("op_10601_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_10601_end_0 = const()[name = tensor("op_10601_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_10601_end_mask_0 = const()[name = tensor("op_10601_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_10601_cast_fp16 = slice_by_index(begin = var_10601_begin_0, end = var_10601_end_0, end_mask = var_10601_end_mask_0, x = var_10535_cast_fp16)[name = tensor("op_10601_cast_fp16")]; tensor k_107_perm_0 = const()[name = tensor("k_107_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_10606_begin_0 = const()[name = tensor("op_10606_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10606_end_0 = const()[name = tensor("op_10606_end_0"), val = tensor([2, 4096, 1, 40])]; tensor var_10606_end_mask_0 = const()[name = tensor("op_10606_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_107_cast_fp16 = transpose(perm = k_107_perm_0, x = k_105_cast_fp16)[name = tensor("transpose_5")]; tensor var_10606_cast_fp16 = slice_by_index(begin = var_10606_begin_0, end = var_10606_end_0, end_mask = var_10606_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_10606_cast_fp16")]; tensor var_10610_begin_0 = const()[name = tensor("op_10610_begin_0"), val = tensor([0, 0, 0, 40])]; tensor var_10610_end_0 = const()[name = tensor("op_10610_end_0"), val = tensor([2, 4096, 1, 80])]; tensor var_10610_end_mask_0 = const()[name = tensor("op_10610_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10610_cast_fp16 = slice_by_index(begin = var_10610_begin_0, end = var_10610_end_0, end_mask = var_10610_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_10610_cast_fp16")]; tensor var_10614_begin_0 = const()[name = tensor("op_10614_begin_0"), val = tensor([0, 0, 0, 80])]; tensor var_10614_end_0 = const()[name = tensor("op_10614_end_0"), val = tensor([2, 4096, 1, 120])]; tensor var_10614_end_mask_0 = const()[name = tensor("op_10614_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10614_cast_fp16 = slice_by_index(begin = var_10614_begin_0, end = var_10614_end_0, end_mask = var_10614_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_10614_cast_fp16")]; tensor var_10618_begin_0 = const()[name = tensor("op_10618_begin_0"), val = tensor([0, 0, 0, 120])]; tensor var_10618_end_0 = const()[name = tensor("op_10618_end_0"), val = tensor([2, 4096, 1, 160])]; tensor var_10618_end_mask_0 = const()[name = tensor("op_10618_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10618_cast_fp16 = slice_by_index(begin = var_10618_begin_0, end = var_10618_end_0, end_mask = var_10618_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_10618_cast_fp16")]; tensor var_10622_begin_0 = const()[name = tensor("op_10622_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_10622_end_0 = const()[name = tensor("op_10622_end_0"), val = tensor([2, 4096, 1, 200])]; tensor var_10622_end_mask_0 = const()[name = tensor("op_10622_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10622_cast_fp16 = slice_by_index(begin = var_10622_begin_0, end = var_10622_end_0, end_mask = var_10622_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_10622_cast_fp16")]; tensor var_10626_begin_0 = const()[name = tensor("op_10626_begin_0"), val = tensor([0, 0, 0, 200])]; tensor var_10626_end_0 = const()[name = tensor("op_10626_end_0"), val = tensor([2, 4096, 1, 240])]; tensor var_10626_end_mask_0 = const()[name = tensor("op_10626_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10626_cast_fp16 = slice_by_index(begin = var_10626_begin_0, end = var_10626_end_0, end_mask = var_10626_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_10626_cast_fp16")]; tensor var_10630_begin_0 = const()[name = tensor("op_10630_begin_0"), val = tensor([0, 0, 0, 240])]; tensor var_10630_end_0 = const()[name = tensor("op_10630_end_0"), val = tensor([2, 4096, 1, 280])]; tensor var_10630_end_mask_0 = const()[name = tensor("op_10630_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10630_cast_fp16 = slice_by_index(begin = var_10630_begin_0, end = var_10630_end_0, end_mask = var_10630_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_10630_cast_fp16")]; tensor var_10634_begin_0 = const()[name = tensor("op_10634_begin_0"), val = tensor([0, 0, 0, 280])]; tensor var_10634_end_0 = const()[name = tensor("op_10634_end_0"), val = tensor([2, 4096, 1, 1])]; tensor var_10634_end_mask_0 = const()[name = tensor("op_10634_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_10634_cast_fp16 = slice_by_index(begin = var_10634_begin_0, end = var_10634_end_0, end_mask = var_10634_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_10634_cast_fp16")]; tensor var_10636_begin_0 = const()[name = tensor("op_10636_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10636_end_0 = const()[name = tensor("op_10636_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_10636_end_mask_0 = const()[name = tensor("op_10636_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10636_cast_fp16 = slice_by_index(begin = var_10636_begin_0, end = var_10636_end_0, end_mask = var_10636_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_10636_cast_fp16")]; tensor var_10640_begin_0 = const()[name = tensor("op_10640_begin_0"), val = tensor([0, 40, 0, 0])]; tensor var_10640_end_0 = const()[name = tensor("op_10640_end_0"), val = tensor([2, 80, 1, 4096])]; tensor var_10640_end_mask_0 = const()[name = tensor("op_10640_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10640_cast_fp16 = slice_by_index(begin = var_10640_begin_0, end = var_10640_end_0, end_mask = var_10640_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_10640_cast_fp16")]; tensor var_10644_begin_0 = const()[name = tensor("op_10644_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_10644_end_0 = const()[name = tensor("op_10644_end_0"), val = tensor([2, 120, 1, 4096])]; tensor var_10644_end_mask_0 = const()[name = tensor("op_10644_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10644_cast_fp16 = slice_by_index(begin = var_10644_begin_0, end = var_10644_end_0, end_mask = var_10644_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_10644_cast_fp16")]; tensor var_10648_begin_0 = const()[name = tensor("op_10648_begin_0"), val = tensor([0, 120, 0, 0])]; tensor var_10648_end_0 = const()[name = tensor("op_10648_end_0"), val = tensor([2, 160, 1, 4096])]; tensor var_10648_end_mask_0 = const()[name = tensor("op_10648_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10648_cast_fp16 = slice_by_index(begin = var_10648_begin_0, end = var_10648_end_0, end_mask = var_10648_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_10648_cast_fp16")]; tensor var_10652_begin_0 = const()[name = tensor("op_10652_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_10652_end_0 = const()[name = tensor("op_10652_end_0"), val = tensor([2, 200, 1, 4096])]; tensor var_10652_end_mask_0 = const()[name = tensor("op_10652_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10652_cast_fp16 = slice_by_index(begin = var_10652_begin_0, end = var_10652_end_0, end_mask = var_10652_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_10652_cast_fp16")]; tensor var_10656_begin_0 = const()[name = tensor("op_10656_begin_0"), val = tensor([0, 200, 0, 0])]; tensor var_10656_end_0 = const()[name = tensor("op_10656_end_0"), val = tensor([2, 240, 1, 4096])]; tensor var_10656_end_mask_0 = const()[name = tensor("op_10656_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10656_cast_fp16 = slice_by_index(begin = var_10656_begin_0, end = var_10656_end_0, end_mask = var_10656_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_10656_cast_fp16")]; tensor var_10660_begin_0 = const()[name = tensor("op_10660_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_10660_end_0 = const()[name = tensor("op_10660_end_0"), val = tensor([2, 280, 1, 4096])]; tensor var_10660_end_mask_0 = const()[name = tensor("op_10660_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10660_cast_fp16 = slice_by_index(begin = var_10660_begin_0, end = var_10660_end_0, end_mask = var_10660_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_10660_cast_fp16")]; tensor var_10664_begin_0 = const()[name = tensor("op_10664_begin_0"), val = tensor([0, 280, 0, 0])]; tensor var_10664_end_0 = const()[name = tensor("op_10664_end_0"), val = tensor([2, 1, 1, 4096])]; tensor var_10664_end_mask_0 = const()[name = tensor("op_10664_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_10664_cast_fp16 = slice_by_index(begin = var_10664_begin_0, end = var_10664_end_0, end_mask = var_10664_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_10664_cast_fp16")]; tensor var_10668_equation_0 = const()[name = tensor("op_10668_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10668_cast_fp16 = einsum(equation = var_10668_equation_0, values = (var_10606_cast_fp16, var_10538_cast_fp16))[name = tensor("op_10668_cast_fp16")]; tensor var_10669_to_fp16 = const()[name = tensor("op_10669_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_833_cast_fp16 = mul(x = var_10668_cast_fp16, y = var_10669_to_fp16)[name = tensor("aw_chunk_833_cast_fp16")]; tensor var_10672_equation_0 = const()[name = tensor("op_10672_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10672_cast_fp16 = einsum(equation = var_10672_equation_0, values = (var_10606_cast_fp16, var_10539_cast_fp16))[name = tensor("op_10672_cast_fp16")]; tensor var_10673_to_fp16 = const()[name = tensor("op_10673_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_835_cast_fp16 = mul(x = var_10672_cast_fp16, y = var_10673_to_fp16)[name = tensor("aw_chunk_835_cast_fp16")]; tensor var_10676_equation_0 = const()[name = tensor("op_10676_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10676_cast_fp16 = einsum(equation = var_10676_equation_0, values = (var_10606_cast_fp16, var_10540_cast_fp16))[name = tensor("op_10676_cast_fp16")]; tensor var_10677_to_fp16 = const()[name = tensor("op_10677_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_837_cast_fp16 = mul(x = var_10676_cast_fp16, y = var_10677_to_fp16)[name = tensor("aw_chunk_837_cast_fp16")]; tensor var_10680_equation_0 = const()[name = tensor("op_10680_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10680_cast_fp16 = einsum(equation = var_10680_equation_0, values = (var_10606_cast_fp16, var_10541_cast_fp16))[name = tensor("op_10680_cast_fp16")]; tensor var_10681_to_fp16 = const()[name = tensor("op_10681_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_839_cast_fp16 = mul(x = var_10680_cast_fp16, y = var_10681_to_fp16)[name = tensor("aw_chunk_839_cast_fp16")]; tensor var_10684_equation_0 = const()[name = tensor("op_10684_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10684_cast_fp16 = einsum(equation = var_10684_equation_0, values = (var_10606_cast_fp16, var_10542_cast_fp16))[name = tensor("op_10684_cast_fp16")]; tensor var_10685_to_fp16 = const()[name = tensor("op_10685_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_841_cast_fp16 = mul(x = var_10684_cast_fp16, y = var_10685_to_fp16)[name = tensor("aw_chunk_841_cast_fp16")]; tensor var_10688_equation_0 = const()[name = tensor("op_10688_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10688_cast_fp16 = einsum(equation = var_10688_equation_0, values = (var_10606_cast_fp16, var_10543_cast_fp16))[name = tensor("op_10688_cast_fp16")]; tensor var_10689_to_fp16 = const()[name = tensor("op_10689_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_843_cast_fp16 = mul(x = var_10688_cast_fp16, y = var_10689_to_fp16)[name = tensor("aw_chunk_843_cast_fp16")]; tensor var_10692_equation_0 = const()[name = tensor("op_10692_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10692_cast_fp16 = einsum(equation = var_10692_equation_0, values = (var_10606_cast_fp16, var_10544_cast_fp16))[name = tensor("op_10692_cast_fp16")]; tensor var_10693_to_fp16 = const()[name = tensor("op_10693_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_845_cast_fp16 = mul(x = var_10692_cast_fp16, y = var_10693_to_fp16)[name = tensor("aw_chunk_845_cast_fp16")]; tensor var_10696_equation_0 = const()[name = tensor("op_10696_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10696_cast_fp16 = einsum(equation = var_10696_equation_0, values = (var_10606_cast_fp16, var_10545_cast_fp16))[name = tensor("op_10696_cast_fp16")]; tensor var_10697_to_fp16 = const()[name = tensor("op_10697_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_847_cast_fp16 = mul(x = var_10696_cast_fp16, y = var_10697_to_fp16)[name = tensor("aw_chunk_847_cast_fp16")]; tensor var_10700_equation_0 = const()[name = tensor("op_10700_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10700_cast_fp16 = einsum(equation = var_10700_equation_0, values = (var_10610_cast_fp16, var_10546_cast_fp16))[name = tensor("op_10700_cast_fp16")]; tensor var_10701_to_fp16 = const()[name = tensor("op_10701_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_849_cast_fp16 = mul(x = var_10700_cast_fp16, y = var_10701_to_fp16)[name = tensor("aw_chunk_849_cast_fp16")]; tensor var_10704_equation_0 = const()[name = tensor("op_10704_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10704_cast_fp16 = einsum(equation = var_10704_equation_0, values = (var_10610_cast_fp16, var_10547_cast_fp16))[name = tensor("op_10704_cast_fp16")]; tensor var_10705_to_fp16 = const()[name = tensor("op_10705_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_851_cast_fp16 = mul(x = var_10704_cast_fp16, y = var_10705_to_fp16)[name = tensor("aw_chunk_851_cast_fp16")]; tensor var_10708_equation_0 = const()[name = tensor("op_10708_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10708_cast_fp16 = einsum(equation = var_10708_equation_0, values = (var_10610_cast_fp16, var_10548_cast_fp16))[name = tensor("op_10708_cast_fp16")]; tensor var_10709_to_fp16 = const()[name = tensor("op_10709_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_853_cast_fp16 = mul(x = var_10708_cast_fp16, y = var_10709_to_fp16)[name = tensor("aw_chunk_853_cast_fp16")]; tensor var_10712_equation_0 = const()[name = tensor("op_10712_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10712_cast_fp16 = einsum(equation = var_10712_equation_0, values = (var_10610_cast_fp16, var_10549_cast_fp16))[name = tensor("op_10712_cast_fp16")]; tensor var_10713_to_fp16 = const()[name = tensor("op_10713_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_855_cast_fp16 = mul(x = var_10712_cast_fp16, y = var_10713_to_fp16)[name = tensor("aw_chunk_855_cast_fp16")]; tensor var_10716_equation_0 = const()[name = tensor("op_10716_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10716_cast_fp16 = einsum(equation = var_10716_equation_0, values = (var_10610_cast_fp16, var_10550_cast_fp16))[name = tensor("op_10716_cast_fp16")]; tensor var_10717_to_fp16 = const()[name = tensor("op_10717_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_857_cast_fp16 = mul(x = var_10716_cast_fp16, y = var_10717_to_fp16)[name = tensor("aw_chunk_857_cast_fp16")]; tensor var_10720_equation_0 = const()[name = tensor("op_10720_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10720_cast_fp16 = einsum(equation = var_10720_equation_0, values = (var_10610_cast_fp16, var_10551_cast_fp16))[name = tensor("op_10720_cast_fp16")]; tensor var_10721_to_fp16 = const()[name = tensor("op_10721_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_859_cast_fp16 = mul(x = var_10720_cast_fp16, y = var_10721_to_fp16)[name = tensor("aw_chunk_859_cast_fp16")]; tensor var_10724_equation_0 = const()[name = tensor("op_10724_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10724_cast_fp16 = einsum(equation = var_10724_equation_0, values = (var_10610_cast_fp16, var_10552_cast_fp16))[name = tensor("op_10724_cast_fp16")]; tensor var_10725_to_fp16 = const()[name = tensor("op_10725_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_861_cast_fp16 = mul(x = var_10724_cast_fp16, y = var_10725_to_fp16)[name = tensor("aw_chunk_861_cast_fp16")]; tensor var_10728_equation_0 = const()[name = tensor("op_10728_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10728_cast_fp16 = einsum(equation = var_10728_equation_0, values = (var_10610_cast_fp16, var_10553_cast_fp16))[name = tensor("op_10728_cast_fp16")]; tensor var_10729_to_fp16 = const()[name = tensor("op_10729_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_863_cast_fp16 = mul(x = var_10728_cast_fp16, y = var_10729_to_fp16)[name = tensor("aw_chunk_863_cast_fp16")]; tensor var_10732_equation_0 = const()[name = tensor("op_10732_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10732_cast_fp16 = einsum(equation = var_10732_equation_0, values = (var_10614_cast_fp16, var_10554_cast_fp16))[name = tensor("op_10732_cast_fp16")]; tensor var_10733_to_fp16 = const()[name = tensor("op_10733_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_865_cast_fp16 = mul(x = var_10732_cast_fp16, y = var_10733_to_fp16)[name = tensor("aw_chunk_865_cast_fp16")]; tensor var_10736_equation_0 = const()[name = tensor("op_10736_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10736_cast_fp16 = einsum(equation = var_10736_equation_0, values = (var_10614_cast_fp16, var_10555_cast_fp16))[name = tensor("op_10736_cast_fp16")]; tensor var_10737_to_fp16 = const()[name = tensor("op_10737_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_867_cast_fp16 = mul(x = var_10736_cast_fp16, y = var_10737_to_fp16)[name = tensor("aw_chunk_867_cast_fp16")]; tensor var_10740_equation_0 = const()[name = tensor("op_10740_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10740_cast_fp16 = einsum(equation = var_10740_equation_0, values = (var_10614_cast_fp16, var_10556_cast_fp16))[name = tensor("op_10740_cast_fp16")]; tensor var_10741_to_fp16 = const()[name = tensor("op_10741_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_869_cast_fp16 = mul(x = var_10740_cast_fp16, y = var_10741_to_fp16)[name = tensor("aw_chunk_869_cast_fp16")]; tensor var_10744_equation_0 = const()[name = tensor("op_10744_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10744_cast_fp16 = einsum(equation = var_10744_equation_0, values = (var_10614_cast_fp16, var_10557_cast_fp16))[name = tensor("op_10744_cast_fp16")]; tensor var_10745_to_fp16 = const()[name = tensor("op_10745_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_871_cast_fp16 = mul(x = var_10744_cast_fp16, y = var_10745_to_fp16)[name = tensor("aw_chunk_871_cast_fp16")]; tensor var_10748_equation_0 = const()[name = tensor("op_10748_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10748_cast_fp16 = einsum(equation = var_10748_equation_0, values = (var_10614_cast_fp16, var_10558_cast_fp16))[name = tensor("op_10748_cast_fp16")]; tensor var_10749_to_fp16 = const()[name = tensor("op_10749_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_873_cast_fp16 = mul(x = var_10748_cast_fp16, y = var_10749_to_fp16)[name = tensor("aw_chunk_873_cast_fp16")]; tensor var_10752_equation_0 = const()[name = tensor("op_10752_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10752_cast_fp16 = einsum(equation = var_10752_equation_0, values = (var_10614_cast_fp16, var_10559_cast_fp16))[name = tensor("op_10752_cast_fp16")]; tensor var_10753_to_fp16 = const()[name = tensor("op_10753_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_875_cast_fp16 = mul(x = var_10752_cast_fp16, y = var_10753_to_fp16)[name = tensor("aw_chunk_875_cast_fp16")]; tensor var_10756_equation_0 = const()[name = tensor("op_10756_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10756_cast_fp16 = einsum(equation = var_10756_equation_0, values = (var_10614_cast_fp16, var_10560_cast_fp16))[name = tensor("op_10756_cast_fp16")]; tensor var_10757_to_fp16 = const()[name = tensor("op_10757_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_877_cast_fp16 = mul(x = var_10756_cast_fp16, y = var_10757_to_fp16)[name = tensor("aw_chunk_877_cast_fp16")]; tensor var_10760_equation_0 = const()[name = tensor("op_10760_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10760_cast_fp16 = einsum(equation = var_10760_equation_0, values = (var_10614_cast_fp16, var_10561_cast_fp16))[name = tensor("op_10760_cast_fp16")]; tensor var_10761_to_fp16 = const()[name = tensor("op_10761_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_879_cast_fp16 = mul(x = var_10760_cast_fp16, y = var_10761_to_fp16)[name = tensor("aw_chunk_879_cast_fp16")]; tensor var_10764_equation_0 = const()[name = tensor("op_10764_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10764_cast_fp16 = einsum(equation = var_10764_equation_0, values = (var_10618_cast_fp16, var_10562_cast_fp16))[name = tensor("op_10764_cast_fp16")]; tensor var_10765_to_fp16 = const()[name = tensor("op_10765_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_881_cast_fp16 = mul(x = var_10764_cast_fp16, y = var_10765_to_fp16)[name = tensor("aw_chunk_881_cast_fp16")]; tensor var_10768_equation_0 = const()[name = tensor("op_10768_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10768_cast_fp16 = einsum(equation = var_10768_equation_0, values = (var_10618_cast_fp16, var_10563_cast_fp16))[name = tensor("op_10768_cast_fp16")]; tensor var_10769_to_fp16 = const()[name = tensor("op_10769_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_883_cast_fp16 = mul(x = var_10768_cast_fp16, y = var_10769_to_fp16)[name = tensor("aw_chunk_883_cast_fp16")]; tensor var_10772_equation_0 = const()[name = tensor("op_10772_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10772_cast_fp16 = einsum(equation = var_10772_equation_0, values = (var_10618_cast_fp16, var_10564_cast_fp16))[name = tensor("op_10772_cast_fp16")]; tensor var_10773_to_fp16 = const()[name = tensor("op_10773_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_885_cast_fp16 = mul(x = var_10772_cast_fp16, y = var_10773_to_fp16)[name = tensor("aw_chunk_885_cast_fp16")]; tensor var_10776_equation_0 = const()[name = tensor("op_10776_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10776_cast_fp16 = einsum(equation = var_10776_equation_0, values = (var_10618_cast_fp16, var_10565_cast_fp16))[name = tensor("op_10776_cast_fp16")]; tensor var_10777_to_fp16 = const()[name = tensor("op_10777_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_887_cast_fp16 = mul(x = var_10776_cast_fp16, y = var_10777_to_fp16)[name = tensor("aw_chunk_887_cast_fp16")]; tensor var_10780_equation_0 = const()[name = tensor("op_10780_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10780_cast_fp16 = einsum(equation = var_10780_equation_0, values = (var_10618_cast_fp16, var_10566_cast_fp16))[name = tensor("op_10780_cast_fp16")]; tensor var_10781_to_fp16 = const()[name = tensor("op_10781_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_889_cast_fp16 = mul(x = var_10780_cast_fp16, y = var_10781_to_fp16)[name = tensor("aw_chunk_889_cast_fp16")]; tensor var_10784_equation_0 = const()[name = tensor("op_10784_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10784_cast_fp16 = einsum(equation = var_10784_equation_0, values = (var_10618_cast_fp16, var_10567_cast_fp16))[name = tensor("op_10784_cast_fp16")]; tensor var_10785_to_fp16 = const()[name = tensor("op_10785_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_891_cast_fp16 = mul(x = var_10784_cast_fp16, y = var_10785_to_fp16)[name = tensor("aw_chunk_891_cast_fp16")]; tensor var_10788_equation_0 = const()[name = tensor("op_10788_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10788_cast_fp16 = einsum(equation = var_10788_equation_0, values = (var_10618_cast_fp16, var_10568_cast_fp16))[name = tensor("op_10788_cast_fp16")]; tensor var_10789_to_fp16 = const()[name = tensor("op_10789_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_893_cast_fp16 = mul(x = var_10788_cast_fp16, y = var_10789_to_fp16)[name = tensor("aw_chunk_893_cast_fp16")]; tensor var_10792_equation_0 = const()[name = tensor("op_10792_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10792_cast_fp16 = einsum(equation = var_10792_equation_0, values = (var_10618_cast_fp16, var_10569_cast_fp16))[name = tensor("op_10792_cast_fp16")]; tensor var_10793_to_fp16 = const()[name = tensor("op_10793_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_895_cast_fp16 = mul(x = var_10792_cast_fp16, y = var_10793_to_fp16)[name = tensor("aw_chunk_895_cast_fp16")]; tensor var_10796_equation_0 = const()[name = tensor("op_10796_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10796_cast_fp16 = einsum(equation = var_10796_equation_0, values = (var_10622_cast_fp16, var_10570_cast_fp16))[name = tensor("op_10796_cast_fp16")]; tensor var_10797_to_fp16 = const()[name = tensor("op_10797_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_897_cast_fp16 = mul(x = var_10796_cast_fp16, y = var_10797_to_fp16)[name = tensor("aw_chunk_897_cast_fp16")]; tensor var_10800_equation_0 = const()[name = tensor("op_10800_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10800_cast_fp16 = einsum(equation = var_10800_equation_0, values = (var_10622_cast_fp16, var_10571_cast_fp16))[name = tensor("op_10800_cast_fp16")]; tensor var_10801_to_fp16 = const()[name = tensor("op_10801_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_899_cast_fp16 = mul(x = var_10800_cast_fp16, y = var_10801_to_fp16)[name = tensor("aw_chunk_899_cast_fp16")]; tensor var_10804_equation_0 = const()[name = tensor("op_10804_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10804_cast_fp16 = einsum(equation = var_10804_equation_0, values = (var_10622_cast_fp16, var_10572_cast_fp16))[name = tensor("op_10804_cast_fp16")]; tensor var_10805_to_fp16 = const()[name = tensor("op_10805_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_901_cast_fp16 = mul(x = var_10804_cast_fp16, y = var_10805_to_fp16)[name = tensor("aw_chunk_901_cast_fp16")]; tensor var_10808_equation_0 = const()[name = tensor("op_10808_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10808_cast_fp16 = einsum(equation = var_10808_equation_0, values = (var_10622_cast_fp16, var_10573_cast_fp16))[name = tensor("op_10808_cast_fp16")]; tensor var_10809_to_fp16 = const()[name = tensor("op_10809_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_903_cast_fp16 = mul(x = var_10808_cast_fp16, y = var_10809_to_fp16)[name = tensor("aw_chunk_903_cast_fp16")]; tensor var_10812_equation_0 = const()[name = tensor("op_10812_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10812_cast_fp16 = einsum(equation = var_10812_equation_0, values = (var_10622_cast_fp16, var_10574_cast_fp16))[name = tensor("op_10812_cast_fp16")]; tensor var_10813_to_fp16 = const()[name = tensor("op_10813_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_905_cast_fp16 = mul(x = var_10812_cast_fp16, y = var_10813_to_fp16)[name = tensor("aw_chunk_905_cast_fp16")]; tensor var_10816_equation_0 = const()[name = tensor("op_10816_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10816_cast_fp16 = einsum(equation = var_10816_equation_0, values = (var_10622_cast_fp16, var_10575_cast_fp16))[name = tensor("op_10816_cast_fp16")]; tensor var_10817_to_fp16 = const()[name = tensor("op_10817_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_907_cast_fp16 = mul(x = var_10816_cast_fp16, y = var_10817_to_fp16)[name = tensor("aw_chunk_907_cast_fp16")]; tensor var_10820_equation_0 = const()[name = tensor("op_10820_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10820_cast_fp16 = einsum(equation = var_10820_equation_0, values = (var_10622_cast_fp16, var_10576_cast_fp16))[name = tensor("op_10820_cast_fp16")]; tensor var_10821_to_fp16 = const()[name = tensor("op_10821_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_909_cast_fp16 = mul(x = var_10820_cast_fp16, y = var_10821_to_fp16)[name = tensor("aw_chunk_909_cast_fp16")]; tensor var_10824_equation_0 = const()[name = tensor("op_10824_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10824_cast_fp16 = einsum(equation = var_10824_equation_0, values = (var_10622_cast_fp16, var_10577_cast_fp16))[name = tensor("op_10824_cast_fp16")]; tensor var_10825_to_fp16 = const()[name = tensor("op_10825_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_911_cast_fp16 = mul(x = var_10824_cast_fp16, y = var_10825_to_fp16)[name = tensor("aw_chunk_911_cast_fp16")]; tensor var_10828_equation_0 = const()[name = tensor("op_10828_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10828_cast_fp16 = einsum(equation = var_10828_equation_0, values = (var_10626_cast_fp16, var_10578_cast_fp16))[name = tensor("op_10828_cast_fp16")]; tensor var_10829_to_fp16 = const()[name = tensor("op_10829_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_913_cast_fp16 = mul(x = var_10828_cast_fp16, y = var_10829_to_fp16)[name = tensor("aw_chunk_913_cast_fp16")]; tensor var_10832_equation_0 = const()[name = tensor("op_10832_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10832_cast_fp16 = einsum(equation = var_10832_equation_0, values = (var_10626_cast_fp16, var_10579_cast_fp16))[name = tensor("op_10832_cast_fp16")]; tensor var_10833_to_fp16 = const()[name = tensor("op_10833_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_915_cast_fp16 = mul(x = var_10832_cast_fp16, y = var_10833_to_fp16)[name = tensor("aw_chunk_915_cast_fp16")]; tensor var_10836_equation_0 = const()[name = tensor("op_10836_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10836_cast_fp16 = einsum(equation = var_10836_equation_0, values = (var_10626_cast_fp16, var_10580_cast_fp16))[name = tensor("op_10836_cast_fp16")]; tensor var_10837_to_fp16 = const()[name = tensor("op_10837_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_917_cast_fp16 = mul(x = var_10836_cast_fp16, y = var_10837_to_fp16)[name = tensor("aw_chunk_917_cast_fp16")]; tensor var_10840_equation_0 = const()[name = tensor("op_10840_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10840_cast_fp16 = einsum(equation = var_10840_equation_0, values = (var_10626_cast_fp16, var_10581_cast_fp16))[name = tensor("op_10840_cast_fp16")]; tensor var_10841_to_fp16 = const()[name = tensor("op_10841_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_919_cast_fp16 = mul(x = var_10840_cast_fp16, y = var_10841_to_fp16)[name = tensor("aw_chunk_919_cast_fp16")]; tensor var_10844_equation_0 = const()[name = tensor("op_10844_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10844_cast_fp16 = einsum(equation = var_10844_equation_0, values = (var_10626_cast_fp16, var_10582_cast_fp16))[name = tensor("op_10844_cast_fp16")]; tensor var_10845_to_fp16 = const()[name = tensor("op_10845_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_921_cast_fp16 = mul(x = var_10844_cast_fp16, y = var_10845_to_fp16)[name = tensor("aw_chunk_921_cast_fp16")]; tensor var_10848_equation_0 = const()[name = tensor("op_10848_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10848_cast_fp16 = einsum(equation = var_10848_equation_0, values = (var_10626_cast_fp16, var_10583_cast_fp16))[name = tensor("op_10848_cast_fp16")]; tensor var_10849_to_fp16 = const()[name = tensor("op_10849_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_923_cast_fp16 = mul(x = var_10848_cast_fp16, y = var_10849_to_fp16)[name = tensor("aw_chunk_923_cast_fp16")]; tensor var_10852_equation_0 = const()[name = tensor("op_10852_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10852_cast_fp16 = einsum(equation = var_10852_equation_0, values = (var_10626_cast_fp16, var_10584_cast_fp16))[name = tensor("op_10852_cast_fp16")]; tensor var_10853_to_fp16 = const()[name = tensor("op_10853_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_925_cast_fp16 = mul(x = var_10852_cast_fp16, y = var_10853_to_fp16)[name = tensor("aw_chunk_925_cast_fp16")]; tensor var_10856_equation_0 = const()[name = tensor("op_10856_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10856_cast_fp16 = einsum(equation = var_10856_equation_0, values = (var_10626_cast_fp16, var_10585_cast_fp16))[name = tensor("op_10856_cast_fp16")]; tensor var_10857_to_fp16 = const()[name = tensor("op_10857_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_927_cast_fp16 = mul(x = var_10856_cast_fp16, y = var_10857_to_fp16)[name = tensor("aw_chunk_927_cast_fp16")]; tensor var_10860_equation_0 = const()[name = tensor("op_10860_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10860_cast_fp16 = einsum(equation = var_10860_equation_0, values = (var_10630_cast_fp16, var_10586_cast_fp16))[name = tensor("op_10860_cast_fp16")]; tensor var_10861_to_fp16 = const()[name = tensor("op_10861_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_929_cast_fp16 = mul(x = var_10860_cast_fp16, y = var_10861_to_fp16)[name = tensor("aw_chunk_929_cast_fp16")]; tensor var_10864_equation_0 = const()[name = tensor("op_10864_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10864_cast_fp16 = einsum(equation = var_10864_equation_0, values = (var_10630_cast_fp16, var_10587_cast_fp16))[name = tensor("op_10864_cast_fp16")]; tensor var_10865_to_fp16 = const()[name = tensor("op_10865_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_931_cast_fp16 = mul(x = var_10864_cast_fp16, y = var_10865_to_fp16)[name = tensor("aw_chunk_931_cast_fp16")]; tensor var_10868_equation_0 = const()[name = tensor("op_10868_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10868_cast_fp16 = einsum(equation = var_10868_equation_0, values = (var_10630_cast_fp16, var_10588_cast_fp16))[name = tensor("op_10868_cast_fp16")]; tensor var_10869_to_fp16 = const()[name = tensor("op_10869_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_933_cast_fp16 = mul(x = var_10868_cast_fp16, y = var_10869_to_fp16)[name = tensor("aw_chunk_933_cast_fp16")]; tensor var_10872_equation_0 = const()[name = tensor("op_10872_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10872_cast_fp16 = einsum(equation = var_10872_equation_0, values = (var_10630_cast_fp16, var_10589_cast_fp16))[name = tensor("op_10872_cast_fp16")]; tensor var_10873_to_fp16 = const()[name = tensor("op_10873_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_935_cast_fp16 = mul(x = var_10872_cast_fp16, y = var_10873_to_fp16)[name = tensor("aw_chunk_935_cast_fp16")]; tensor var_10876_equation_0 = const()[name = tensor("op_10876_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10876_cast_fp16 = einsum(equation = var_10876_equation_0, values = (var_10630_cast_fp16, var_10590_cast_fp16))[name = tensor("op_10876_cast_fp16")]; tensor var_10877_to_fp16 = const()[name = tensor("op_10877_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_937_cast_fp16 = mul(x = var_10876_cast_fp16, y = var_10877_to_fp16)[name = tensor("aw_chunk_937_cast_fp16")]; tensor var_10880_equation_0 = const()[name = tensor("op_10880_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10880_cast_fp16 = einsum(equation = var_10880_equation_0, values = (var_10630_cast_fp16, var_10591_cast_fp16))[name = tensor("op_10880_cast_fp16")]; tensor var_10881_to_fp16 = const()[name = tensor("op_10881_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_939_cast_fp16 = mul(x = var_10880_cast_fp16, y = var_10881_to_fp16)[name = tensor("aw_chunk_939_cast_fp16")]; tensor var_10884_equation_0 = const()[name = tensor("op_10884_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10884_cast_fp16 = einsum(equation = var_10884_equation_0, values = (var_10630_cast_fp16, var_10592_cast_fp16))[name = tensor("op_10884_cast_fp16")]; tensor var_10885_to_fp16 = const()[name = tensor("op_10885_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_941_cast_fp16 = mul(x = var_10884_cast_fp16, y = var_10885_to_fp16)[name = tensor("aw_chunk_941_cast_fp16")]; tensor var_10888_equation_0 = const()[name = tensor("op_10888_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10888_cast_fp16 = einsum(equation = var_10888_equation_0, values = (var_10630_cast_fp16, var_10593_cast_fp16))[name = tensor("op_10888_cast_fp16")]; tensor var_10889_to_fp16 = const()[name = tensor("op_10889_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_943_cast_fp16 = mul(x = var_10888_cast_fp16, y = var_10889_to_fp16)[name = tensor("aw_chunk_943_cast_fp16")]; tensor var_10892_equation_0 = const()[name = tensor("op_10892_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10892_cast_fp16 = einsum(equation = var_10892_equation_0, values = (var_10634_cast_fp16, var_10594_cast_fp16))[name = tensor("op_10892_cast_fp16")]; tensor var_10893_to_fp16 = const()[name = tensor("op_10893_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_945_cast_fp16 = mul(x = var_10892_cast_fp16, y = var_10893_to_fp16)[name = tensor("aw_chunk_945_cast_fp16")]; tensor var_10896_equation_0 = const()[name = tensor("op_10896_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10896_cast_fp16 = einsum(equation = var_10896_equation_0, values = (var_10634_cast_fp16, var_10595_cast_fp16))[name = tensor("op_10896_cast_fp16")]; tensor var_10897_to_fp16 = const()[name = tensor("op_10897_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_947_cast_fp16 = mul(x = var_10896_cast_fp16, y = var_10897_to_fp16)[name = tensor("aw_chunk_947_cast_fp16")]; tensor var_10900_equation_0 = const()[name = tensor("op_10900_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10900_cast_fp16 = einsum(equation = var_10900_equation_0, values = (var_10634_cast_fp16, var_10596_cast_fp16))[name = tensor("op_10900_cast_fp16")]; tensor var_10901_to_fp16 = const()[name = tensor("op_10901_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_949_cast_fp16 = mul(x = var_10900_cast_fp16, y = var_10901_to_fp16)[name = tensor("aw_chunk_949_cast_fp16")]; tensor var_10904_equation_0 = const()[name = tensor("op_10904_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10904_cast_fp16 = einsum(equation = var_10904_equation_0, values = (var_10634_cast_fp16, var_10597_cast_fp16))[name = tensor("op_10904_cast_fp16")]; tensor var_10905_to_fp16 = const()[name = tensor("op_10905_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_951_cast_fp16 = mul(x = var_10904_cast_fp16, y = var_10905_to_fp16)[name = tensor("aw_chunk_951_cast_fp16")]; tensor var_10908_equation_0 = const()[name = tensor("op_10908_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10908_cast_fp16 = einsum(equation = var_10908_equation_0, values = (var_10634_cast_fp16, var_10598_cast_fp16))[name = tensor("op_10908_cast_fp16")]; tensor var_10909_to_fp16 = const()[name = tensor("op_10909_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_953_cast_fp16 = mul(x = var_10908_cast_fp16, y = var_10909_to_fp16)[name = tensor("aw_chunk_953_cast_fp16")]; tensor var_10912_equation_0 = const()[name = tensor("op_10912_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10912_cast_fp16 = einsum(equation = var_10912_equation_0, values = (var_10634_cast_fp16, var_10599_cast_fp16))[name = tensor("op_10912_cast_fp16")]; tensor var_10913_to_fp16 = const()[name = tensor("op_10913_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_955_cast_fp16 = mul(x = var_10912_cast_fp16, y = var_10913_to_fp16)[name = tensor("aw_chunk_955_cast_fp16")]; tensor var_10916_equation_0 = const()[name = tensor("op_10916_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10916_cast_fp16 = einsum(equation = var_10916_equation_0, values = (var_10634_cast_fp16, var_10600_cast_fp16))[name = tensor("op_10916_cast_fp16")]; tensor var_10917_to_fp16 = const()[name = tensor("op_10917_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_957_cast_fp16 = mul(x = var_10916_cast_fp16, y = var_10917_to_fp16)[name = tensor("aw_chunk_957_cast_fp16")]; tensor var_10920_equation_0 = const()[name = tensor("op_10920_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10920_cast_fp16 = einsum(equation = var_10920_equation_0, values = (var_10634_cast_fp16, var_10601_cast_fp16))[name = tensor("op_10920_cast_fp16")]; tensor var_10921_to_fp16 = const()[name = tensor("op_10921_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_959_cast_fp16 = mul(x = var_10920_cast_fp16, y = var_10921_to_fp16)[name = tensor("aw_chunk_959_cast_fp16")]; tensor var_10923_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_833_cast_fp16)[name = tensor("op_10923_cast_fp16")]; tensor var_10924_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_835_cast_fp16)[name = tensor("op_10924_cast_fp16")]; tensor var_10925_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_837_cast_fp16)[name = tensor("op_10925_cast_fp16")]; tensor var_10926_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_839_cast_fp16)[name = tensor("op_10926_cast_fp16")]; tensor var_10927_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_841_cast_fp16)[name = tensor("op_10927_cast_fp16")]; tensor var_10928_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_843_cast_fp16)[name = tensor("op_10928_cast_fp16")]; tensor var_10929_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_845_cast_fp16)[name = tensor("op_10929_cast_fp16")]; tensor var_10930_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_847_cast_fp16)[name = tensor("op_10930_cast_fp16")]; tensor var_10931_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_849_cast_fp16)[name = tensor("op_10931_cast_fp16")]; tensor var_10932_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_851_cast_fp16)[name = tensor("op_10932_cast_fp16")]; tensor var_10933_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_853_cast_fp16)[name = tensor("op_10933_cast_fp16")]; tensor var_10934_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_855_cast_fp16)[name = tensor("op_10934_cast_fp16")]; tensor var_10935_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_857_cast_fp16)[name = tensor("op_10935_cast_fp16")]; tensor var_10936_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_859_cast_fp16)[name = tensor("op_10936_cast_fp16")]; tensor var_10937_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_861_cast_fp16)[name = tensor("op_10937_cast_fp16")]; tensor var_10938_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_863_cast_fp16)[name = tensor("op_10938_cast_fp16")]; tensor var_10939_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_865_cast_fp16)[name = tensor("op_10939_cast_fp16")]; tensor var_10940_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_867_cast_fp16)[name = tensor("op_10940_cast_fp16")]; tensor var_10941_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_869_cast_fp16)[name = tensor("op_10941_cast_fp16")]; tensor var_10942_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_871_cast_fp16)[name = tensor("op_10942_cast_fp16")]; tensor var_10943_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_873_cast_fp16)[name = tensor("op_10943_cast_fp16")]; tensor var_10944_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_875_cast_fp16)[name = tensor("op_10944_cast_fp16")]; tensor var_10945_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_877_cast_fp16)[name = tensor("op_10945_cast_fp16")]; tensor var_10946_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_879_cast_fp16)[name = tensor("op_10946_cast_fp16")]; tensor var_10947_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_881_cast_fp16)[name = tensor("op_10947_cast_fp16")]; tensor var_10948_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_883_cast_fp16)[name = tensor("op_10948_cast_fp16")]; tensor var_10949_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_885_cast_fp16)[name = tensor("op_10949_cast_fp16")]; tensor var_10950_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_887_cast_fp16)[name = tensor("op_10950_cast_fp16")]; tensor var_10951_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_889_cast_fp16)[name = tensor("op_10951_cast_fp16")]; tensor var_10952_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_891_cast_fp16)[name = tensor("op_10952_cast_fp16")]; tensor var_10953_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_893_cast_fp16)[name = tensor("op_10953_cast_fp16")]; tensor var_10954_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_895_cast_fp16)[name = tensor("op_10954_cast_fp16")]; tensor var_10955_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_897_cast_fp16)[name = tensor("op_10955_cast_fp16")]; tensor var_10956_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_899_cast_fp16)[name = tensor("op_10956_cast_fp16")]; tensor var_10957_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_901_cast_fp16)[name = tensor("op_10957_cast_fp16")]; tensor var_10958_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_903_cast_fp16)[name = tensor("op_10958_cast_fp16")]; tensor var_10959_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_905_cast_fp16)[name = tensor("op_10959_cast_fp16")]; tensor var_10960_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_907_cast_fp16)[name = tensor("op_10960_cast_fp16")]; tensor var_10961_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_909_cast_fp16)[name = tensor("op_10961_cast_fp16")]; tensor var_10962_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_911_cast_fp16)[name = tensor("op_10962_cast_fp16")]; tensor var_10963_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_913_cast_fp16)[name = tensor("op_10963_cast_fp16")]; tensor var_10964_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_915_cast_fp16)[name = tensor("op_10964_cast_fp16")]; tensor var_10965_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_917_cast_fp16)[name = tensor("op_10965_cast_fp16")]; tensor var_10966_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_919_cast_fp16)[name = tensor("op_10966_cast_fp16")]; tensor var_10967_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_921_cast_fp16)[name = tensor("op_10967_cast_fp16")]; tensor var_10968_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_923_cast_fp16)[name = tensor("op_10968_cast_fp16")]; tensor var_10969_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_925_cast_fp16)[name = tensor("op_10969_cast_fp16")]; tensor var_10970_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_927_cast_fp16)[name = tensor("op_10970_cast_fp16")]; tensor var_10971_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_929_cast_fp16)[name = tensor("op_10971_cast_fp16")]; tensor var_10972_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_931_cast_fp16)[name = tensor("op_10972_cast_fp16")]; tensor var_10973_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_933_cast_fp16)[name = tensor("op_10973_cast_fp16")]; tensor var_10974_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_935_cast_fp16)[name = tensor("op_10974_cast_fp16")]; tensor var_10975_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_937_cast_fp16)[name = tensor("op_10975_cast_fp16")]; tensor var_10976_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_939_cast_fp16)[name = tensor("op_10976_cast_fp16")]; tensor var_10977_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_941_cast_fp16)[name = tensor("op_10977_cast_fp16")]; tensor var_10978_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_943_cast_fp16)[name = tensor("op_10978_cast_fp16")]; tensor var_10979_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_945_cast_fp16)[name = tensor("op_10979_cast_fp16")]; tensor var_10980_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_947_cast_fp16)[name = tensor("op_10980_cast_fp16")]; tensor var_10981_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_949_cast_fp16)[name = tensor("op_10981_cast_fp16")]; tensor var_10982_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_951_cast_fp16)[name = tensor("op_10982_cast_fp16")]; tensor var_10983_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_953_cast_fp16)[name = tensor("op_10983_cast_fp16")]; tensor var_10984_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_955_cast_fp16)[name = tensor("op_10984_cast_fp16")]; tensor var_10985_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_957_cast_fp16)[name = tensor("op_10985_cast_fp16")]; tensor var_10986_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_959_cast_fp16)[name = tensor("op_10986_cast_fp16")]; tensor var_10988_equation_0 = const()[name = tensor("op_10988_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10988_cast_fp16 = einsum(equation = var_10988_equation_0, values = (var_10636_cast_fp16, var_10923_cast_fp16))[name = tensor("op_10988_cast_fp16")]; tensor var_10990_equation_0 = const()[name = tensor("op_10990_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10990_cast_fp16 = einsum(equation = var_10990_equation_0, values = (var_10636_cast_fp16, var_10924_cast_fp16))[name = tensor("op_10990_cast_fp16")]; tensor var_10992_equation_0 = const()[name = tensor("op_10992_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10992_cast_fp16 = einsum(equation = var_10992_equation_0, values = (var_10636_cast_fp16, var_10925_cast_fp16))[name = tensor("op_10992_cast_fp16")]; tensor var_10994_equation_0 = const()[name = tensor("op_10994_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10994_cast_fp16 = einsum(equation = var_10994_equation_0, values = (var_10636_cast_fp16, var_10926_cast_fp16))[name = tensor("op_10994_cast_fp16")]; tensor var_10996_equation_0 = const()[name = tensor("op_10996_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10996_cast_fp16 = einsum(equation = var_10996_equation_0, values = (var_10636_cast_fp16, var_10927_cast_fp16))[name = tensor("op_10996_cast_fp16")]; tensor var_10998_equation_0 = const()[name = tensor("op_10998_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10998_cast_fp16 = einsum(equation = var_10998_equation_0, values = (var_10636_cast_fp16, var_10928_cast_fp16))[name = tensor("op_10998_cast_fp16")]; tensor var_11000_equation_0 = const()[name = tensor("op_11000_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11000_cast_fp16 = einsum(equation = var_11000_equation_0, values = (var_10636_cast_fp16, var_10929_cast_fp16))[name = tensor("op_11000_cast_fp16")]; tensor var_11002_equation_0 = const()[name = tensor("op_11002_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11002_cast_fp16 = einsum(equation = var_11002_equation_0, values = (var_10636_cast_fp16, var_10930_cast_fp16))[name = tensor("op_11002_cast_fp16")]; tensor var_11004_equation_0 = const()[name = tensor("op_11004_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11004_cast_fp16 = einsum(equation = var_11004_equation_0, values = (var_10640_cast_fp16, var_10931_cast_fp16))[name = tensor("op_11004_cast_fp16")]; tensor var_11006_equation_0 = const()[name = tensor("op_11006_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11006_cast_fp16 = einsum(equation = var_11006_equation_0, values = (var_10640_cast_fp16, var_10932_cast_fp16))[name = tensor("op_11006_cast_fp16")]; tensor var_11008_equation_0 = const()[name = tensor("op_11008_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11008_cast_fp16 = einsum(equation = var_11008_equation_0, values = (var_10640_cast_fp16, var_10933_cast_fp16))[name = tensor("op_11008_cast_fp16")]; tensor var_11010_equation_0 = const()[name = tensor("op_11010_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11010_cast_fp16 = einsum(equation = var_11010_equation_0, values = (var_10640_cast_fp16, var_10934_cast_fp16))[name = tensor("op_11010_cast_fp16")]; tensor var_11012_equation_0 = const()[name = tensor("op_11012_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11012_cast_fp16 = einsum(equation = var_11012_equation_0, values = (var_10640_cast_fp16, var_10935_cast_fp16))[name = tensor("op_11012_cast_fp16")]; tensor var_11014_equation_0 = const()[name = tensor("op_11014_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11014_cast_fp16 = einsum(equation = var_11014_equation_0, values = (var_10640_cast_fp16, var_10936_cast_fp16))[name = tensor("op_11014_cast_fp16")]; tensor var_11016_equation_0 = const()[name = tensor("op_11016_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11016_cast_fp16 = einsum(equation = var_11016_equation_0, values = (var_10640_cast_fp16, var_10937_cast_fp16))[name = tensor("op_11016_cast_fp16")]; tensor var_11018_equation_0 = const()[name = tensor("op_11018_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11018_cast_fp16 = einsum(equation = var_11018_equation_0, values = (var_10640_cast_fp16, var_10938_cast_fp16))[name = tensor("op_11018_cast_fp16")]; tensor var_11020_equation_0 = const()[name = tensor("op_11020_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11020_cast_fp16 = einsum(equation = var_11020_equation_0, values = (var_10644_cast_fp16, var_10939_cast_fp16))[name = tensor("op_11020_cast_fp16")]; tensor var_11022_equation_0 = const()[name = tensor("op_11022_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11022_cast_fp16 = einsum(equation = var_11022_equation_0, values = (var_10644_cast_fp16, var_10940_cast_fp16))[name = tensor("op_11022_cast_fp16")]; tensor var_11024_equation_0 = const()[name = tensor("op_11024_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11024_cast_fp16 = einsum(equation = var_11024_equation_0, values = (var_10644_cast_fp16, var_10941_cast_fp16))[name = tensor("op_11024_cast_fp16")]; tensor var_11026_equation_0 = const()[name = tensor("op_11026_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11026_cast_fp16 = einsum(equation = var_11026_equation_0, values = (var_10644_cast_fp16, var_10942_cast_fp16))[name = tensor("op_11026_cast_fp16")]; tensor var_11028_equation_0 = const()[name = tensor("op_11028_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11028_cast_fp16 = einsum(equation = var_11028_equation_0, values = (var_10644_cast_fp16, var_10943_cast_fp16))[name = tensor("op_11028_cast_fp16")]; tensor var_11030_equation_0 = const()[name = tensor("op_11030_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11030_cast_fp16 = einsum(equation = var_11030_equation_0, values = (var_10644_cast_fp16, var_10944_cast_fp16))[name = tensor("op_11030_cast_fp16")]; tensor var_11032_equation_0 = const()[name = tensor("op_11032_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11032_cast_fp16 = einsum(equation = var_11032_equation_0, values = (var_10644_cast_fp16, var_10945_cast_fp16))[name = tensor("op_11032_cast_fp16")]; tensor var_11034_equation_0 = const()[name = tensor("op_11034_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11034_cast_fp16 = einsum(equation = var_11034_equation_0, values = (var_10644_cast_fp16, var_10946_cast_fp16))[name = tensor("op_11034_cast_fp16")]; tensor var_11036_equation_0 = const()[name = tensor("op_11036_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11036_cast_fp16 = einsum(equation = var_11036_equation_0, values = (var_10648_cast_fp16, var_10947_cast_fp16))[name = tensor("op_11036_cast_fp16")]; tensor var_11038_equation_0 = const()[name = tensor("op_11038_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11038_cast_fp16 = einsum(equation = var_11038_equation_0, values = (var_10648_cast_fp16, var_10948_cast_fp16))[name = tensor("op_11038_cast_fp16")]; tensor var_11040_equation_0 = const()[name = tensor("op_11040_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11040_cast_fp16 = einsum(equation = var_11040_equation_0, values = (var_10648_cast_fp16, var_10949_cast_fp16))[name = tensor("op_11040_cast_fp16")]; tensor var_11042_equation_0 = const()[name = tensor("op_11042_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11042_cast_fp16 = einsum(equation = var_11042_equation_0, values = (var_10648_cast_fp16, var_10950_cast_fp16))[name = tensor("op_11042_cast_fp16")]; tensor var_11044_equation_0 = const()[name = tensor("op_11044_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11044_cast_fp16 = einsum(equation = var_11044_equation_0, values = (var_10648_cast_fp16, var_10951_cast_fp16))[name = tensor("op_11044_cast_fp16")]; tensor var_11046_equation_0 = const()[name = tensor("op_11046_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11046_cast_fp16 = einsum(equation = var_11046_equation_0, values = (var_10648_cast_fp16, var_10952_cast_fp16))[name = tensor("op_11046_cast_fp16")]; tensor var_11048_equation_0 = const()[name = tensor("op_11048_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11048_cast_fp16 = einsum(equation = var_11048_equation_0, values = (var_10648_cast_fp16, var_10953_cast_fp16))[name = tensor("op_11048_cast_fp16")]; tensor var_11050_equation_0 = const()[name = tensor("op_11050_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11050_cast_fp16 = einsum(equation = var_11050_equation_0, values = (var_10648_cast_fp16, var_10954_cast_fp16))[name = tensor("op_11050_cast_fp16")]; tensor var_11052_equation_0 = const()[name = tensor("op_11052_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11052_cast_fp16 = einsum(equation = var_11052_equation_0, values = (var_10652_cast_fp16, var_10955_cast_fp16))[name = tensor("op_11052_cast_fp16")]; tensor var_11054_equation_0 = const()[name = tensor("op_11054_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11054_cast_fp16 = einsum(equation = var_11054_equation_0, values = (var_10652_cast_fp16, var_10956_cast_fp16))[name = tensor("op_11054_cast_fp16")]; tensor var_11056_equation_0 = const()[name = tensor("op_11056_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11056_cast_fp16 = einsum(equation = var_11056_equation_0, values = (var_10652_cast_fp16, var_10957_cast_fp16))[name = tensor("op_11056_cast_fp16")]; tensor var_11058_equation_0 = const()[name = tensor("op_11058_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11058_cast_fp16 = einsum(equation = var_11058_equation_0, values = (var_10652_cast_fp16, var_10958_cast_fp16))[name = tensor("op_11058_cast_fp16")]; tensor var_11060_equation_0 = const()[name = tensor("op_11060_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11060_cast_fp16 = einsum(equation = var_11060_equation_0, values = (var_10652_cast_fp16, var_10959_cast_fp16))[name = tensor("op_11060_cast_fp16")]; tensor var_11062_equation_0 = const()[name = tensor("op_11062_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11062_cast_fp16 = einsum(equation = var_11062_equation_0, values = (var_10652_cast_fp16, var_10960_cast_fp16))[name = tensor("op_11062_cast_fp16")]; tensor var_11064_equation_0 = const()[name = tensor("op_11064_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11064_cast_fp16 = einsum(equation = var_11064_equation_0, values = (var_10652_cast_fp16, var_10961_cast_fp16))[name = tensor("op_11064_cast_fp16")]; tensor var_11066_equation_0 = const()[name = tensor("op_11066_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11066_cast_fp16 = einsum(equation = var_11066_equation_0, values = (var_10652_cast_fp16, var_10962_cast_fp16))[name = tensor("op_11066_cast_fp16")]; tensor var_11068_equation_0 = const()[name = tensor("op_11068_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11068_cast_fp16 = einsum(equation = var_11068_equation_0, values = (var_10656_cast_fp16, var_10963_cast_fp16))[name = tensor("op_11068_cast_fp16")]; tensor var_11070_equation_0 = const()[name = tensor("op_11070_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11070_cast_fp16 = einsum(equation = var_11070_equation_0, values = (var_10656_cast_fp16, var_10964_cast_fp16))[name = tensor("op_11070_cast_fp16")]; tensor var_11072_equation_0 = const()[name = tensor("op_11072_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11072_cast_fp16 = einsum(equation = var_11072_equation_0, values = (var_10656_cast_fp16, var_10965_cast_fp16))[name = tensor("op_11072_cast_fp16")]; tensor var_11074_equation_0 = const()[name = tensor("op_11074_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11074_cast_fp16 = einsum(equation = var_11074_equation_0, values = (var_10656_cast_fp16, var_10966_cast_fp16))[name = tensor("op_11074_cast_fp16")]; tensor var_11076_equation_0 = const()[name = tensor("op_11076_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11076_cast_fp16 = einsum(equation = var_11076_equation_0, values = (var_10656_cast_fp16, var_10967_cast_fp16))[name = tensor("op_11076_cast_fp16")]; tensor var_11078_equation_0 = const()[name = tensor("op_11078_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11078_cast_fp16 = einsum(equation = var_11078_equation_0, values = (var_10656_cast_fp16, var_10968_cast_fp16))[name = tensor("op_11078_cast_fp16")]; tensor var_11080_equation_0 = const()[name = tensor("op_11080_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11080_cast_fp16 = einsum(equation = var_11080_equation_0, values = (var_10656_cast_fp16, var_10969_cast_fp16))[name = tensor("op_11080_cast_fp16")]; tensor var_11082_equation_0 = const()[name = tensor("op_11082_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11082_cast_fp16 = einsum(equation = var_11082_equation_0, values = (var_10656_cast_fp16, var_10970_cast_fp16))[name = tensor("op_11082_cast_fp16")]; tensor var_11084_equation_0 = const()[name = tensor("op_11084_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11084_cast_fp16 = einsum(equation = var_11084_equation_0, values = (var_10660_cast_fp16, var_10971_cast_fp16))[name = tensor("op_11084_cast_fp16")]; tensor var_11086_equation_0 = const()[name = tensor("op_11086_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11086_cast_fp16 = einsum(equation = var_11086_equation_0, values = (var_10660_cast_fp16, var_10972_cast_fp16))[name = tensor("op_11086_cast_fp16")]; tensor var_11088_equation_0 = const()[name = tensor("op_11088_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11088_cast_fp16 = einsum(equation = var_11088_equation_0, values = (var_10660_cast_fp16, var_10973_cast_fp16))[name = tensor("op_11088_cast_fp16")]; tensor var_11090_equation_0 = const()[name = tensor("op_11090_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11090_cast_fp16 = einsum(equation = var_11090_equation_0, values = (var_10660_cast_fp16, var_10974_cast_fp16))[name = tensor("op_11090_cast_fp16")]; tensor var_11092_equation_0 = const()[name = tensor("op_11092_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11092_cast_fp16 = einsum(equation = var_11092_equation_0, values = (var_10660_cast_fp16, var_10975_cast_fp16))[name = tensor("op_11092_cast_fp16")]; tensor var_11094_equation_0 = const()[name = tensor("op_11094_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11094_cast_fp16 = einsum(equation = var_11094_equation_0, values = (var_10660_cast_fp16, var_10976_cast_fp16))[name = tensor("op_11094_cast_fp16")]; tensor var_11096_equation_0 = const()[name = tensor("op_11096_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11096_cast_fp16 = einsum(equation = var_11096_equation_0, values = (var_10660_cast_fp16, var_10977_cast_fp16))[name = tensor("op_11096_cast_fp16")]; tensor var_11098_equation_0 = const()[name = tensor("op_11098_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11098_cast_fp16 = einsum(equation = var_11098_equation_0, values = (var_10660_cast_fp16, var_10978_cast_fp16))[name = tensor("op_11098_cast_fp16")]; tensor var_11100_equation_0 = const()[name = tensor("op_11100_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11100_cast_fp16 = einsum(equation = var_11100_equation_0, values = (var_10664_cast_fp16, var_10979_cast_fp16))[name = tensor("op_11100_cast_fp16")]; tensor var_11102_equation_0 = const()[name = tensor("op_11102_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11102_cast_fp16 = einsum(equation = var_11102_equation_0, values = (var_10664_cast_fp16, var_10980_cast_fp16))[name = tensor("op_11102_cast_fp16")]; tensor var_11104_equation_0 = const()[name = tensor("op_11104_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11104_cast_fp16 = einsum(equation = var_11104_equation_0, values = (var_10664_cast_fp16, var_10981_cast_fp16))[name = tensor("op_11104_cast_fp16")]; tensor var_11106_equation_0 = const()[name = tensor("op_11106_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11106_cast_fp16 = einsum(equation = var_11106_equation_0, values = (var_10664_cast_fp16, var_10982_cast_fp16))[name = tensor("op_11106_cast_fp16")]; tensor var_11108_equation_0 = const()[name = tensor("op_11108_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11108_cast_fp16 = einsum(equation = var_11108_equation_0, values = (var_10664_cast_fp16, var_10983_cast_fp16))[name = tensor("op_11108_cast_fp16")]; tensor var_11110_equation_0 = const()[name = tensor("op_11110_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11110_cast_fp16 = einsum(equation = var_11110_equation_0, values = (var_10664_cast_fp16, var_10984_cast_fp16))[name = tensor("op_11110_cast_fp16")]; tensor var_11112_equation_0 = const()[name = tensor("op_11112_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11112_cast_fp16 = einsum(equation = var_11112_equation_0, values = (var_10664_cast_fp16, var_10985_cast_fp16))[name = tensor("op_11112_cast_fp16")]; tensor var_11114_equation_0 = const()[name = tensor("op_11114_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11114_cast_fp16 = einsum(equation = var_11114_equation_0, values = (var_10664_cast_fp16, var_10986_cast_fp16))[name = tensor("op_11114_cast_fp16")]; tensor var_11116_interleave_0 = const()[name = tensor("op_11116_interleave_0"), val = tensor(false)]; tensor var_11116_cast_fp16 = concat(axis = var_10347, interleave = var_11116_interleave_0, values = (var_10988_cast_fp16, var_10990_cast_fp16, var_10992_cast_fp16, var_10994_cast_fp16, var_10996_cast_fp16, var_10998_cast_fp16, var_11000_cast_fp16, var_11002_cast_fp16))[name = tensor("op_11116_cast_fp16")]; tensor var_11118_interleave_0 = const()[name = tensor("op_11118_interleave_0"), val = tensor(false)]; tensor var_11118_cast_fp16 = concat(axis = var_10347, interleave = var_11118_interleave_0, values = (var_11004_cast_fp16, var_11006_cast_fp16, var_11008_cast_fp16, var_11010_cast_fp16, var_11012_cast_fp16, var_11014_cast_fp16, var_11016_cast_fp16, var_11018_cast_fp16))[name = tensor("op_11118_cast_fp16")]; tensor var_11120_interleave_0 = const()[name = tensor("op_11120_interleave_0"), val = tensor(false)]; tensor var_11120_cast_fp16 = concat(axis = var_10347, interleave = var_11120_interleave_0, values = (var_11020_cast_fp16, var_11022_cast_fp16, var_11024_cast_fp16, var_11026_cast_fp16, var_11028_cast_fp16, var_11030_cast_fp16, var_11032_cast_fp16, var_11034_cast_fp16))[name = tensor("op_11120_cast_fp16")]; tensor var_11122_interleave_0 = const()[name = tensor("op_11122_interleave_0"), val = tensor(false)]; tensor var_11122_cast_fp16 = concat(axis = var_10347, interleave = var_11122_interleave_0, values = (var_11036_cast_fp16, var_11038_cast_fp16, var_11040_cast_fp16, var_11042_cast_fp16, var_11044_cast_fp16, var_11046_cast_fp16, var_11048_cast_fp16, var_11050_cast_fp16))[name = tensor("op_11122_cast_fp16")]; tensor var_11124_interleave_0 = const()[name = tensor("op_11124_interleave_0"), val = tensor(false)]; tensor var_11124_cast_fp16 = concat(axis = var_10347, interleave = var_11124_interleave_0, values = (var_11052_cast_fp16, var_11054_cast_fp16, var_11056_cast_fp16, var_11058_cast_fp16, var_11060_cast_fp16, var_11062_cast_fp16, var_11064_cast_fp16, var_11066_cast_fp16))[name = tensor("op_11124_cast_fp16")]; tensor var_11126_interleave_0 = const()[name = tensor("op_11126_interleave_0"), val = tensor(false)]; tensor var_11126_cast_fp16 = concat(axis = var_10347, interleave = var_11126_interleave_0, values = (var_11068_cast_fp16, var_11070_cast_fp16, var_11072_cast_fp16, var_11074_cast_fp16, var_11076_cast_fp16, var_11078_cast_fp16, var_11080_cast_fp16, var_11082_cast_fp16))[name = tensor("op_11126_cast_fp16")]; tensor var_11128_interleave_0 = const()[name = tensor("op_11128_interleave_0"), val = tensor(false)]; tensor var_11128_cast_fp16 = concat(axis = var_10347, interleave = var_11128_interleave_0, values = (var_11084_cast_fp16, var_11086_cast_fp16, var_11088_cast_fp16, var_11090_cast_fp16, var_11092_cast_fp16, var_11094_cast_fp16, var_11096_cast_fp16, var_11098_cast_fp16))[name = tensor("op_11128_cast_fp16")]; tensor var_11130_interleave_0 = const()[name = tensor("op_11130_interleave_0"), val = tensor(false)]; tensor var_11130_cast_fp16 = concat(axis = var_10347, interleave = var_11130_interleave_0, values = (var_11100_cast_fp16, var_11102_cast_fp16, var_11104_cast_fp16, var_11106_cast_fp16, var_11108_cast_fp16, var_11110_cast_fp16, var_11112_cast_fp16, var_11114_cast_fp16))[name = tensor("op_11130_cast_fp16")]; tensor input_465_interleave_0 = const()[name = tensor("input_465_interleave_0"), val = tensor(false)]; tensor input_465_cast_fp16 = concat(axis = var_10375, interleave = input_465_interleave_0, values = (var_11116_cast_fp16, var_11118_cast_fp16, var_11120_cast_fp16, var_11122_cast_fp16, var_11124_cast_fp16, var_11126_cast_fp16, var_11128_cast_fp16, var_11130_cast_fp16))[name = tensor("input_465_cast_fp16")]; tensor var_11140_pad_type_0 = const()[name = tensor("op_11140_pad_type_0"), val = tensor("valid")]; tensor var_11140_strides_0 = const()[name = tensor("op_11140_strides_0"), val = tensor([1, 1])]; tensor var_11140_pad_0 = const()[name = tensor("op_11140_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_11140_dilations_0 = const()[name = tensor("op_11140_dilations_0"), val = tensor([1, 1])]; tensor var_11140_groups_0 = const()[name = tensor("op_11140_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(803552768)))]; tensor up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(803757632)))]; tensor var_11140_cast_fp16 = conv(bias = up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_11140_dilations_0, groups = var_11140_groups_0, pad = var_11140_pad_0, pad_type = var_11140_pad_type_0, strides = var_11140_strides_0, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_465_cast_fp16)[name = tensor("op_11140_cast_fp16")]; tensor inputs_81_cast_fp16 = add(x = var_11140_cast_fp16, y = inputs_79_cast_fp16)[name = tensor("inputs_81_cast_fp16")]; tensor hidden_states_289_axes_0 = const()[name = tensor("hidden_states_289_axes_0"), val = tensor([1])]; tensor hidden_states_289_gamma_0_to_fp16 = const()[name = tensor("hidden_states_289_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(803758336)))]; tensor hidden_states_289_beta_0_to_fp16 = const()[name = tensor("hidden_states_289_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(803759040)))]; tensor var_11150_to_fp16 = const()[name = tensor("op_11150_to_fp16"), val = tensor(0x1.5p-17)]; tensor hidden_states_289_cast_fp16 = layer_norm(axes = hidden_states_289_axes_0, beta = hidden_states_289_beta_0_to_fp16, epsilon = var_11150_to_fp16, gamma = hidden_states_289_gamma_0_to_fp16, x = inputs_81_cast_fp16)[name = tensor("hidden_states_289_cast_fp16")]; tensor q_55_pad_type_0 = const()[name = tensor("q_55_pad_type_0"), val = tensor("valid")]; tensor q_55_strides_0 = const()[name = tensor("q_55_strides_0"), val = tensor([1, 1])]; tensor q_55_pad_0 = const()[name = tensor("q_55_pad_0"), val = tensor([0, 0, 0, 0])]; tensor q_55_dilations_0 = const()[name = tensor("q_55_dilations_0"), val = tensor([1, 1])]; tensor q_55_groups_0 = const()[name = tensor("q_55_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(803759744)))]; tensor q_55_cast_fp16 = conv(dilations = q_55_dilations_0, groups = q_55_groups_0, pad = q_55_pad_0, pad_type = q_55_pad_type_0, strides = q_55_strides_0, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_289_cast_fp16)[name = tensor("q_55_cast_fp16")]; tensor k_109_pad_type_0 = const()[name = tensor("k_109_pad_type_0"), val = tensor("valid")]; tensor k_109_strides_0 = const()[name = tensor("k_109_strides_0"), val = tensor([1, 1])]; tensor k_109_pad_0 = const()[name = tensor("k_109_pad_0"), val = tensor([0, 0, 0, 0])]; tensor k_109_dilations_0 = const()[name = tensor("k_109_dilations_0"), val = tensor([1, 1])]; tensor k_109_groups_0 = const()[name = tensor("k_109_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(803964608)))]; tensor k_109_cast_fp16 = conv(dilations = k_109_dilations_0, groups = k_109_groups_0, pad = k_109_pad_0, pad_type = k_109_pad_type_0, strides = k_109_strides_0, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_109_cast_fp16")]; tensor v_55_pad_type_0 = const()[name = tensor("v_55_pad_type_0"), val = tensor("valid")]; tensor v_55_strides_0 = const()[name = tensor("v_55_strides_0"), val = tensor([1, 1])]; tensor v_55_pad_0 = const()[name = tensor("v_55_pad_0"), val = tensor([0, 0, 0, 0])]; tensor v_55_dilations_0 = const()[name = tensor("v_55_dilations_0"), val = tensor([1, 1])]; tensor v_55_groups_0 = const()[name = tensor("v_55_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(804456192)))]; tensor v_55_cast_fp16 = conv(dilations = v_55_dilations_0, groups = v_55_groups_0, pad = v_55_pad_0, pad_type = v_55_pad_type_0, strides = v_55_strides_0, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_55_cast_fp16")]; tensor var_11183_begin_0 = const()[name = tensor("op_11183_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11183_end_0 = const()[name = tensor("op_11183_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_11183_end_mask_0 = const()[name = tensor("op_11183_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11183_cast_fp16 = slice_by_index(begin = var_11183_begin_0, end = var_11183_end_0, end_mask = var_11183_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11183_cast_fp16")]; tensor var_11187_begin_0 = const()[name = tensor("op_11187_begin_0"), val = tensor([0, 40, 0, 0])]; tensor var_11187_end_0 = const()[name = tensor("op_11187_end_0"), val = tensor([2, 80, 1, 4096])]; tensor var_11187_end_mask_0 = const()[name = tensor("op_11187_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11187_cast_fp16 = slice_by_index(begin = var_11187_begin_0, end = var_11187_end_0, end_mask = var_11187_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11187_cast_fp16")]; tensor var_11191_begin_0 = const()[name = tensor("op_11191_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_11191_end_0 = const()[name = tensor("op_11191_end_0"), val = tensor([2, 120, 1, 4096])]; tensor var_11191_end_mask_0 = const()[name = tensor("op_11191_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11191_cast_fp16 = slice_by_index(begin = var_11191_begin_0, end = var_11191_end_0, end_mask = var_11191_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11191_cast_fp16")]; tensor var_11195_begin_0 = const()[name = tensor("op_11195_begin_0"), val = tensor([0, 120, 0, 0])]; tensor var_11195_end_0 = const()[name = tensor("op_11195_end_0"), val = tensor([2, 160, 1, 4096])]; tensor var_11195_end_mask_0 = const()[name = tensor("op_11195_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11195_cast_fp16 = slice_by_index(begin = var_11195_begin_0, end = var_11195_end_0, end_mask = var_11195_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11195_cast_fp16")]; tensor var_11199_begin_0 = const()[name = tensor("op_11199_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_11199_end_0 = const()[name = tensor("op_11199_end_0"), val = tensor([2, 200, 1, 4096])]; tensor var_11199_end_mask_0 = const()[name = tensor("op_11199_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11199_cast_fp16 = slice_by_index(begin = var_11199_begin_0, end = var_11199_end_0, end_mask = var_11199_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11199_cast_fp16")]; tensor var_11203_begin_0 = const()[name = tensor("op_11203_begin_0"), val = tensor([0, 200, 0, 0])]; tensor var_11203_end_0 = const()[name = tensor("op_11203_end_0"), val = tensor([2, 240, 1, 4096])]; tensor var_11203_end_mask_0 = const()[name = tensor("op_11203_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11203_cast_fp16 = slice_by_index(begin = var_11203_begin_0, end = var_11203_end_0, end_mask = var_11203_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11203_cast_fp16")]; tensor var_11207_begin_0 = const()[name = tensor("op_11207_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_11207_end_0 = const()[name = tensor("op_11207_end_0"), val = tensor([2, 280, 1, 4096])]; tensor var_11207_end_mask_0 = const()[name = tensor("op_11207_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11207_cast_fp16 = slice_by_index(begin = var_11207_begin_0, end = var_11207_end_0, end_mask = var_11207_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11207_cast_fp16")]; tensor var_11211_begin_0 = const()[name = tensor("op_11211_begin_0"), val = tensor([0, 280, 0, 0])]; tensor var_11211_end_0 = const()[name = tensor("op_11211_end_0"), val = tensor([2, 1, 1, 4096])]; tensor var_11211_end_mask_0 = const()[name = tensor("op_11211_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_11211_cast_fp16 = slice_by_index(begin = var_11211_begin_0, end = var_11211_end_0, end_mask = var_11211_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11211_cast_fp16")]; tensor var_11214_begin_0 = const()[name = tensor("op_11214_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11214_end_0 = const()[name = tensor("op_11214_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_11214_end_mask_0 = const()[name = tensor("op_11214_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11214_cast_fp16 = slice_by_index(begin = var_11214_begin_0, end = var_11214_end_0, end_mask = var_11214_end_mask_0, x = var_11183_cast_fp16)[name = tensor("op_11214_cast_fp16")]; tensor var_11215_begin_0 = const()[name = tensor("op_11215_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_11215_end_0 = const()[name = tensor("op_11215_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_11215_end_mask_0 = const()[name = tensor("op_11215_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11215_cast_fp16 = slice_by_index(begin = var_11215_begin_0, end = var_11215_end_0, end_mask = var_11215_end_mask_0, x = var_11183_cast_fp16)[name = tensor("op_11215_cast_fp16")]; tensor var_11216_begin_0 = const()[name = tensor("op_11216_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_11216_end_0 = const()[name = tensor("op_11216_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_11216_end_mask_0 = const()[name = tensor("op_11216_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11216_cast_fp16 = slice_by_index(begin = var_11216_begin_0, end = var_11216_end_0, end_mask = var_11216_end_mask_0, x = var_11183_cast_fp16)[name = tensor("op_11216_cast_fp16")]; tensor var_11217_begin_0 = const()[name = tensor("op_11217_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_11217_end_0 = const()[name = tensor("op_11217_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_11217_end_mask_0 = const()[name = tensor("op_11217_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11217_cast_fp16 = slice_by_index(begin = var_11217_begin_0, end = var_11217_end_0, end_mask = var_11217_end_mask_0, x = var_11183_cast_fp16)[name = tensor("op_11217_cast_fp16")]; tensor var_11218_begin_0 = const()[name = tensor("op_11218_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_11218_end_0 = const()[name = tensor("op_11218_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_11218_end_mask_0 = const()[name = tensor("op_11218_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11218_cast_fp16 = slice_by_index(begin = var_11218_begin_0, end = var_11218_end_0, end_mask = var_11218_end_mask_0, x = var_11183_cast_fp16)[name = tensor("op_11218_cast_fp16")]; tensor var_11219_begin_0 = const()[name = tensor("op_11219_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_11219_end_0 = const()[name = tensor("op_11219_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_11219_end_mask_0 = const()[name = tensor("op_11219_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11219_cast_fp16 = slice_by_index(begin = var_11219_begin_0, end = var_11219_end_0, end_mask = var_11219_end_mask_0, x = var_11183_cast_fp16)[name = tensor("op_11219_cast_fp16")]; tensor var_11220_begin_0 = const()[name = tensor("op_11220_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_11220_end_0 = const()[name = tensor("op_11220_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_11220_end_mask_0 = const()[name = tensor("op_11220_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11220_cast_fp16 = slice_by_index(begin = var_11220_begin_0, end = var_11220_end_0, end_mask = var_11220_end_mask_0, x = var_11183_cast_fp16)[name = tensor("op_11220_cast_fp16")]; tensor var_11221_begin_0 = const()[name = tensor("op_11221_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_11221_end_0 = const()[name = tensor("op_11221_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_11221_end_mask_0 = const()[name = tensor("op_11221_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_11221_cast_fp16 = slice_by_index(begin = var_11221_begin_0, end = var_11221_end_0, end_mask = var_11221_end_mask_0, x = var_11183_cast_fp16)[name = tensor("op_11221_cast_fp16")]; tensor var_11222_begin_0 = const()[name = tensor("op_11222_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11222_end_0 = const()[name = tensor("op_11222_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_11222_end_mask_0 = const()[name = tensor("op_11222_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11222_cast_fp16 = slice_by_index(begin = var_11222_begin_0, end = var_11222_end_0, end_mask = var_11222_end_mask_0, x = var_11187_cast_fp16)[name = tensor("op_11222_cast_fp16")]; tensor var_11223_begin_0 = const()[name = tensor("op_11223_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_11223_end_0 = const()[name = tensor("op_11223_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_11223_end_mask_0 = const()[name = tensor("op_11223_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11223_cast_fp16 = slice_by_index(begin = var_11223_begin_0, end = var_11223_end_0, end_mask = var_11223_end_mask_0, x = var_11187_cast_fp16)[name = tensor("op_11223_cast_fp16")]; tensor var_11224_begin_0 = const()[name = tensor("op_11224_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_11224_end_0 = const()[name = tensor("op_11224_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_11224_end_mask_0 = const()[name = tensor("op_11224_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11224_cast_fp16 = slice_by_index(begin = var_11224_begin_0, end = var_11224_end_0, end_mask = var_11224_end_mask_0, x = var_11187_cast_fp16)[name = tensor("op_11224_cast_fp16")]; tensor var_11225_begin_0 = const()[name = tensor("op_11225_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_11225_end_0 = const()[name = tensor("op_11225_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_11225_end_mask_0 = const()[name = tensor("op_11225_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11225_cast_fp16 = slice_by_index(begin = var_11225_begin_0, end = var_11225_end_0, end_mask = var_11225_end_mask_0, x = var_11187_cast_fp16)[name = tensor("op_11225_cast_fp16")]; tensor var_11226_begin_0 = const()[name = tensor("op_11226_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_11226_end_0 = const()[name = tensor("op_11226_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_11226_end_mask_0 = const()[name = tensor("op_11226_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11226_cast_fp16 = slice_by_index(begin = var_11226_begin_0, end = var_11226_end_0, end_mask = var_11226_end_mask_0, x = var_11187_cast_fp16)[name = tensor("op_11226_cast_fp16")]; tensor var_11227_begin_0 = const()[name = tensor("op_11227_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_11227_end_0 = const()[name = tensor("op_11227_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_11227_end_mask_0 = const()[name = tensor("op_11227_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11227_cast_fp16 = slice_by_index(begin = var_11227_begin_0, end = var_11227_end_0, end_mask = var_11227_end_mask_0, x = var_11187_cast_fp16)[name = tensor("op_11227_cast_fp16")]; tensor var_11228_begin_0 = const()[name = tensor("op_11228_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_11228_end_0 = const()[name = tensor("op_11228_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_11228_end_mask_0 = const()[name = tensor("op_11228_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11228_cast_fp16 = slice_by_index(begin = var_11228_begin_0, end = var_11228_end_0, end_mask = var_11228_end_mask_0, x = var_11187_cast_fp16)[name = tensor("op_11228_cast_fp16")]; tensor var_11229_begin_0 = const()[name = tensor("op_11229_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_11229_end_0 = const()[name = tensor("op_11229_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_11229_end_mask_0 = const()[name = tensor("op_11229_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_11229_cast_fp16 = slice_by_index(begin = var_11229_begin_0, end = var_11229_end_0, end_mask = var_11229_end_mask_0, x = var_11187_cast_fp16)[name = tensor("op_11229_cast_fp16")]; tensor var_11230_begin_0 = const()[name = tensor("op_11230_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11230_end_0 = const()[name = tensor("op_11230_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_11230_end_mask_0 = const()[name = tensor("op_11230_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11230_cast_fp16 = slice_by_index(begin = var_11230_begin_0, end = var_11230_end_0, end_mask = var_11230_end_mask_0, x = var_11191_cast_fp16)[name = tensor("op_11230_cast_fp16")]; tensor var_11231_begin_0 = const()[name = tensor("op_11231_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_11231_end_0 = const()[name = tensor("op_11231_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_11231_end_mask_0 = const()[name = tensor("op_11231_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11231_cast_fp16 = slice_by_index(begin = var_11231_begin_0, end = var_11231_end_0, end_mask = var_11231_end_mask_0, x = var_11191_cast_fp16)[name = tensor("op_11231_cast_fp16")]; tensor var_11232_begin_0 = const()[name = tensor("op_11232_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_11232_end_0 = const()[name = tensor("op_11232_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_11232_end_mask_0 = const()[name = tensor("op_11232_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11232_cast_fp16 = slice_by_index(begin = var_11232_begin_0, end = var_11232_end_0, end_mask = var_11232_end_mask_0, x = var_11191_cast_fp16)[name = tensor("op_11232_cast_fp16")]; tensor var_11233_begin_0 = const()[name = tensor("op_11233_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_11233_end_0 = const()[name = tensor("op_11233_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_11233_end_mask_0 = const()[name = tensor("op_11233_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11233_cast_fp16 = slice_by_index(begin = var_11233_begin_0, end = var_11233_end_0, end_mask = var_11233_end_mask_0, x = var_11191_cast_fp16)[name = tensor("op_11233_cast_fp16")]; tensor var_11234_begin_0 = const()[name = tensor("op_11234_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_11234_end_0 = const()[name = tensor("op_11234_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_11234_end_mask_0 = const()[name = tensor("op_11234_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11234_cast_fp16 = slice_by_index(begin = var_11234_begin_0, end = var_11234_end_0, end_mask = var_11234_end_mask_0, x = var_11191_cast_fp16)[name = tensor("op_11234_cast_fp16")]; tensor var_11235_begin_0 = const()[name = tensor("op_11235_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_11235_end_0 = const()[name = tensor("op_11235_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_11235_end_mask_0 = const()[name = tensor("op_11235_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11235_cast_fp16 = slice_by_index(begin = var_11235_begin_0, end = var_11235_end_0, end_mask = var_11235_end_mask_0, x = var_11191_cast_fp16)[name = tensor("op_11235_cast_fp16")]; tensor var_11236_begin_0 = const()[name = tensor("op_11236_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_11236_end_0 = const()[name = tensor("op_11236_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_11236_end_mask_0 = const()[name = tensor("op_11236_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11236_cast_fp16 = slice_by_index(begin = var_11236_begin_0, end = var_11236_end_0, end_mask = var_11236_end_mask_0, x = var_11191_cast_fp16)[name = tensor("op_11236_cast_fp16")]; tensor var_11237_begin_0 = const()[name = tensor("op_11237_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_11237_end_0 = const()[name = tensor("op_11237_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_11237_end_mask_0 = const()[name = tensor("op_11237_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_11237_cast_fp16 = slice_by_index(begin = var_11237_begin_0, end = var_11237_end_0, end_mask = var_11237_end_mask_0, x = var_11191_cast_fp16)[name = tensor("op_11237_cast_fp16")]; tensor var_11238_begin_0 = const()[name = tensor("op_11238_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11238_end_0 = const()[name = tensor("op_11238_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_11238_end_mask_0 = const()[name = tensor("op_11238_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11238_cast_fp16 = slice_by_index(begin = var_11238_begin_0, end = var_11238_end_0, end_mask = var_11238_end_mask_0, x = var_11195_cast_fp16)[name = tensor("op_11238_cast_fp16")]; tensor var_11239_begin_0 = const()[name = tensor("op_11239_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_11239_end_0 = const()[name = tensor("op_11239_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_11239_end_mask_0 = const()[name = tensor("op_11239_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11239_cast_fp16 = slice_by_index(begin = var_11239_begin_0, end = var_11239_end_0, end_mask = var_11239_end_mask_0, x = var_11195_cast_fp16)[name = tensor("op_11239_cast_fp16")]; tensor var_11240_begin_0 = const()[name = tensor("op_11240_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_11240_end_0 = const()[name = tensor("op_11240_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_11240_end_mask_0 = const()[name = tensor("op_11240_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11240_cast_fp16 = slice_by_index(begin = var_11240_begin_0, end = var_11240_end_0, end_mask = var_11240_end_mask_0, x = var_11195_cast_fp16)[name = tensor("op_11240_cast_fp16")]; tensor var_11241_begin_0 = const()[name = tensor("op_11241_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_11241_end_0 = const()[name = tensor("op_11241_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_11241_end_mask_0 = const()[name = tensor("op_11241_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11241_cast_fp16 = slice_by_index(begin = var_11241_begin_0, end = var_11241_end_0, end_mask = var_11241_end_mask_0, x = var_11195_cast_fp16)[name = tensor("op_11241_cast_fp16")]; tensor var_11242_begin_0 = const()[name = tensor("op_11242_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_11242_end_0 = const()[name = tensor("op_11242_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_11242_end_mask_0 = const()[name = tensor("op_11242_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11242_cast_fp16 = slice_by_index(begin = var_11242_begin_0, end = var_11242_end_0, end_mask = var_11242_end_mask_0, x = var_11195_cast_fp16)[name = tensor("op_11242_cast_fp16")]; tensor var_11243_begin_0 = const()[name = tensor("op_11243_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_11243_end_0 = const()[name = tensor("op_11243_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_11243_end_mask_0 = const()[name = tensor("op_11243_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11243_cast_fp16 = slice_by_index(begin = var_11243_begin_0, end = var_11243_end_0, end_mask = var_11243_end_mask_0, x = var_11195_cast_fp16)[name = tensor("op_11243_cast_fp16")]; tensor var_11244_begin_0 = const()[name = tensor("op_11244_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_11244_end_0 = const()[name = tensor("op_11244_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_11244_end_mask_0 = const()[name = tensor("op_11244_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11244_cast_fp16 = slice_by_index(begin = var_11244_begin_0, end = var_11244_end_0, end_mask = var_11244_end_mask_0, x = var_11195_cast_fp16)[name = tensor("op_11244_cast_fp16")]; tensor var_11245_begin_0 = const()[name = tensor("op_11245_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_11245_end_0 = const()[name = tensor("op_11245_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_11245_end_mask_0 = const()[name = tensor("op_11245_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_11245_cast_fp16 = slice_by_index(begin = var_11245_begin_0, end = var_11245_end_0, end_mask = var_11245_end_mask_0, x = var_11195_cast_fp16)[name = tensor("op_11245_cast_fp16")]; tensor var_11246_begin_0 = const()[name = tensor("op_11246_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11246_end_0 = const()[name = tensor("op_11246_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_11246_end_mask_0 = const()[name = tensor("op_11246_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11246_cast_fp16 = slice_by_index(begin = var_11246_begin_0, end = var_11246_end_0, end_mask = var_11246_end_mask_0, x = var_11199_cast_fp16)[name = tensor("op_11246_cast_fp16")]; tensor var_11247_begin_0 = const()[name = tensor("op_11247_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_11247_end_0 = const()[name = tensor("op_11247_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_11247_end_mask_0 = const()[name = tensor("op_11247_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11247_cast_fp16 = slice_by_index(begin = var_11247_begin_0, end = var_11247_end_0, end_mask = var_11247_end_mask_0, x = var_11199_cast_fp16)[name = tensor("op_11247_cast_fp16")]; tensor var_11248_begin_0 = const()[name = tensor("op_11248_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_11248_end_0 = const()[name = tensor("op_11248_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_11248_end_mask_0 = const()[name = tensor("op_11248_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11248_cast_fp16 = slice_by_index(begin = var_11248_begin_0, end = var_11248_end_0, end_mask = var_11248_end_mask_0, x = var_11199_cast_fp16)[name = tensor("op_11248_cast_fp16")]; tensor var_11249_begin_0 = const()[name = tensor("op_11249_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_11249_end_0 = const()[name = tensor("op_11249_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_11249_end_mask_0 = const()[name = tensor("op_11249_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11249_cast_fp16 = slice_by_index(begin = var_11249_begin_0, end = var_11249_end_0, end_mask = var_11249_end_mask_0, x = var_11199_cast_fp16)[name = tensor("op_11249_cast_fp16")]; tensor var_11250_begin_0 = const()[name = tensor("op_11250_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_11250_end_0 = const()[name = tensor("op_11250_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_11250_end_mask_0 = const()[name = tensor("op_11250_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11250_cast_fp16 = slice_by_index(begin = var_11250_begin_0, end = var_11250_end_0, end_mask = var_11250_end_mask_0, x = var_11199_cast_fp16)[name = tensor("op_11250_cast_fp16")]; tensor var_11251_begin_0 = const()[name = tensor("op_11251_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_11251_end_0 = const()[name = tensor("op_11251_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_11251_end_mask_0 = const()[name = tensor("op_11251_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11251_cast_fp16 = slice_by_index(begin = var_11251_begin_0, end = var_11251_end_0, end_mask = var_11251_end_mask_0, x = var_11199_cast_fp16)[name = tensor("op_11251_cast_fp16")]; tensor var_11252_begin_0 = const()[name = tensor("op_11252_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_11252_end_0 = const()[name = tensor("op_11252_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_11252_end_mask_0 = const()[name = tensor("op_11252_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11252_cast_fp16 = slice_by_index(begin = var_11252_begin_0, end = var_11252_end_0, end_mask = var_11252_end_mask_0, x = var_11199_cast_fp16)[name = tensor("op_11252_cast_fp16")]; tensor var_11253_begin_0 = const()[name = tensor("op_11253_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_11253_end_0 = const()[name = tensor("op_11253_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_11253_end_mask_0 = const()[name = tensor("op_11253_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_11253_cast_fp16 = slice_by_index(begin = var_11253_begin_0, end = var_11253_end_0, end_mask = var_11253_end_mask_0, x = var_11199_cast_fp16)[name = tensor("op_11253_cast_fp16")]; tensor var_11254_begin_0 = const()[name = tensor("op_11254_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11254_end_0 = const()[name = tensor("op_11254_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_11254_end_mask_0 = const()[name = tensor("op_11254_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11254_cast_fp16 = slice_by_index(begin = var_11254_begin_0, end = var_11254_end_0, end_mask = var_11254_end_mask_0, x = var_11203_cast_fp16)[name = tensor("op_11254_cast_fp16")]; tensor var_11255_begin_0 = const()[name = tensor("op_11255_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_11255_end_0 = const()[name = tensor("op_11255_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_11255_end_mask_0 = const()[name = tensor("op_11255_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11255_cast_fp16 = slice_by_index(begin = var_11255_begin_0, end = var_11255_end_0, end_mask = var_11255_end_mask_0, x = var_11203_cast_fp16)[name = tensor("op_11255_cast_fp16")]; tensor var_11256_begin_0 = const()[name = tensor("op_11256_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_11256_end_0 = const()[name = tensor("op_11256_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_11256_end_mask_0 = const()[name = tensor("op_11256_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11256_cast_fp16 = slice_by_index(begin = var_11256_begin_0, end = var_11256_end_0, end_mask = var_11256_end_mask_0, x = var_11203_cast_fp16)[name = tensor("op_11256_cast_fp16")]; tensor var_11257_begin_0 = const()[name = tensor("op_11257_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_11257_end_0 = const()[name = tensor("op_11257_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_11257_end_mask_0 = const()[name = tensor("op_11257_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11257_cast_fp16 = slice_by_index(begin = var_11257_begin_0, end = var_11257_end_0, end_mask = var_11257_end_mask_0, x = var_11203_cast_fp16)[name = tensor("op_11257_cast_fp16")]; tensor var_11258_begin_0 = const()[name = tensor("op_11258_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_11258_end_0 = const()[name = tensor("op_11258_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_11258_end_mask_0 = const()[name = tensor("op_11258_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11258_cast_fp16 = slice_by_index(begin = var_11258_begin_0, end = var_11258_end_0, end_mask = var_11258_end_mask_0, x = var_11203_cast_fp16)[name = tensor("op_11258_cast_fp16")]; tensor var_11259_begin_0 = const()[name = tensor("op_11259_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_11259_end_0 = const()[name = tensor("op_11259_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_11259_end_mask_0 = const()[name = tensor("op_11259_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11259_cast_fp16 = slice_by_index(begin = var_11259_begin_0, end = var_11259_end_0, end_mask = var_11259_end_mask_0, x = var_11203_cast_fp16)[name = tensor("op_11259_cast_fp16")]; tensor var_11260_begin_0 = const()[name = tensor("op_11260_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_11260_end_0 = const()[name = tensor("op_11260_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_11260_end_mask_0 = const()[name = tensor("op_11260_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11260_cast_fp16 = slice_by_index(begin = var_11260_begin_0, end = var_11260_end_0, end_mask = var_11260_end_mask_0, x = var_11203_cast_fp16)[name = tensor("op_11260_cast_fp16")]; tensor var_11261_begin_0 = const()[name = tensor("op_11261_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_11261_end_0 = const()[name = tensor("op_11261_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_11261_end_mask_0 = const()[name = tensor("op_11261_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_11261_cast_fp16 = slice_by_index(begin = var_11261_begin_0, end = var_11261_end_0, end_mask = var_11261_end_mask_0, x = var_11203_cast_fp16)[name = tensor("op_11261_cast_fp16")]; tensor var_11262_begin_0 = const()[name = tensor("op_11262_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11262_end_0 = const()[name = tensor("op_11262_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_11262_end_mask_0 = const()[name = tensor("op_11262_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11262_cast_fp16 = slice_by_index(begin = var_11262_begin_0, end = var_11262_end_0, end_mask = var_11262_end_mask_0, x = var_11207_cast_fp16)[name = tensor("op_11262_cast_fp16")]; tensor var_11263_begin_0 = const()[name = tensor("op_11263_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_11263_end_0 = const()[name = tensor("op_11263_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_11263_end_mask_0 = const()[name = tensor("op_11263_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11263_cast_fp16 = slice_by_index(begin = var_11263_begin_0, end = var_11263_end_0, end_mask = var_11263_end_mask_0, x = var_11207_cast_fp16)[name = tensor("op_11263_cast_fp16")]; tensor var_11264_begin_0 = const()[name = tensor("op_11264_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_11264_end_0 = const()[name = tensor("op_11264_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_11264_end_mask_0 = const()[name = tensor("op_11264_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11264_cast_fp16 = slice_by_index(begin = var_11264_begin_0, end = var_11264_end_0, end_mask = var_11264_end_mask_0, x = var_11207_cast_fp16)[name = tensor("op_11264_cast_fp16")]; tensor var_11265_begin_0 = const()[name = tensor("op_11265_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_11265_end_0 = const()[name = tensor("op_11265_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_11265_end_mask_0 = const()[name = tensor("op_11265_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11265_cast_fp16 = slice_by_index(begin = var_11265_begin_0, end = var_11265_end_0, end_mask = var_11265_end_mask_0, x = var_11207_cast_fp16)[name = tensor("op_11265_cast_fp16")]; tensor var_11266_begin_0 = const()[name = tensor("op_11266_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_11266_end_0 = const()[name = tensor("op_11266_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_11266_end_mask_0 = const()[name = tensor("op_11266_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11266_cast_fp16 = slice_by_index(begin = var_11266_begin_0, end = var_11266_end_0, end_mask = var_11266_end_mask_0, x = var_11207_cast_fp16)[name = tensor("op_11266_cast_fp16")]; tensor var_11267_begin_0 = const()[name = tensor("op_11267_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_11267_end_0 = const()[name = tensor("op_11267_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_11267_end_mask_0 = const()[name = tensor("op_11267_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11267_cast_fp16 = slice_by_index(begin = var_11267_begin_0, end = var_11267_end_0, end_mask = var_11267_end_mask_0, x = var_11207_cast_fp16)[name = tensor("op_11267_cast_fp16")]; tensor var_11268_begin_0 = const()[name = tensor("op_11268_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_11268_end_0 = const()[name = tensor("op_11268_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_11268_end_mask_0 = const()[name = tensor("op_11268_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11268_cast_fp16 = slice_by_index(begin = var_11268_begin_0, end = var_11268_end_0, end_mask = var_11268_end_mask_0, x = var_11207_cast_fp16)[name = tensor("op_11268_cast_fp16")]; tensor var_11269_begin_0 = const()[name = tensor("op_11269_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_11269_end_0 = const()[name = tensor("op_11269_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_11269_end_mask_0 = const()[name = tensor("op_11269_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_11269_cast_fp16 = slice_by_index(begin = var_11269_begin_0, end = var_11269_end_0, end_mask = var_11269_end_mask_0, x = var_11207_cast_fp16)[name = tensor("op_11269_cast_fp16")]; tensor var_11270_begin_0 = const()[name = tensor("op_11270_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11270_end_0 = const()[name = tensor("op_11270_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_11270_end_mask_0 = const()[name = tensor("op_11270_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11270_cast_fp16 = slice_by_index(begin = var_11270_begin_0, end = var_11270_end_0, end_mask = var_11270_end_mask_0, x = var_11211_cast_fp16)[name = tensor("op_11270_cast_fp16")]; tensor var_11271_begin_0 = const()[name = tensor("op_11271_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_11271_end_0 = const()[name = tensor("op_11271_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_11271_end_mask_0 = const()[name = tensor("op_11271_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11271_cast_fp16 = slice_by_index(begin = var_11271_begin_0, end = var_11271_end_0, end_mask = var_11271_end_mask_0, x = var_11211_cast_fp16)[name = tensor("op_11271_cast_fp16")]; tensor var_11272_begin_0 = const()[name = tensor("op_11272_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_11272_end_0 = const()[name = tensor("op_11272_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_11272_end_mask_0 = const()[name = tensor("op_11272_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11272_cast_fp16 = slice_by_index(begin = var_11272_begin_0, end = var_11272_end_0, end_mask = var_11272_end_mask_0, x = var_11211_cast_fp16)[name = tensor("op_11272_cast_fp16")]; tensor var_11273_begin_0 = const()[name = tensor("op_11273_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_11273_end_0 = const()[name = tensor("op_11273_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_11273_end_mask_0 = const()[name = tensor("op_11273_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11273_cast_fp16 = slice_by_index(begin = var_11273_begin_0, end = var_11273_end_0, end_mask = var_11273_end_mask_0, x = var_11211_cast_fp16)[name = tensor("op_11273_cast_fp16")]; tensor var_11274_begin_0 = const()[name = tensor("op_11274_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_11274_end_0 = const()[name = tensor("op_11274_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_11274_end_mask_0 = const()[name = tensor("op_11274_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11274_cast_fp16 = slice_by_index(begin = var_11274_begin_0, end = var_11274_end_0, end_mask = var_11274_end_mask_0, x = var_11211_cast_fp16)[name = tensor("op_11274_cast_fp16")]; tensor var_11275_begin_0 = const()[name = tensor("op_11275_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_11275_end_0 = const()[name = tensor("op_11275_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_11275_end_mask_0 = const()[name = tensor("op_11275_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11275_cast_fp16 = slice_by_index(begin = var_11275_begin_0, end = var_11275_end_0, end_mask = var_11275_end_mask_0, x = var_11211_cast_fp16)[name = tensor("op_11275_cast_fp16")]; tensor var_11276_begin_0 = const()[name = tensor("op_11276_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_11276_end_0 = const()[name = tensor("op_11276_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_11276_end_mask_0 = const()[name = tensor("op_11276_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11276_cast_fp16 = slice_by_index(begin = var_11276_begin_0, end = var_11276_end_0, end_mask = var_11276_end_mask_0, x = var_11211_cast_fp16)[name = tensor("op_11276_cast_fp16")]; tensor var_11277_begin_0 = const()[name = tensor("op_11277_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_11277_end_0 = const()[name = tensor("op_11277_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_11277_end_mask_0 = const()[name = tensor("op_11277_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_11277_cast_fp16 = slice_by_index(begin = var_11277_begin_0, end = var_11277_end_0, end_mask = var_11277_end_mask_0, x = var_11211_cast_fp16)[name = tensor("op_11277_cast_fp16")]; tensor k_111_perm_0 = const()[name = tensor("k_111_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_11282_begin_0 = const()[name = tensor("op_11282_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11282_end_0 = const()[name = tensor("op_11282_end_0"), val = tensor([2, 77, 1, 40])]; tensor var_11282_end_mask_0 = const()[name = tensor("op_11282_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_111_cast_fp16 = transpose(perm = k_111_perm_0, x = k_109_cast_fp16)[name = tensor("transpose_4")]; tensor var_11282_cast_fp16 = slice_by_index(begin = var_11282_begin_0, end = var_11282_end_0, end_mask = var_11282_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11282_cast_fp16")]; tensor var_11286_begin_0 = const()[name = tensor("op_11286_begin_0"), val = tensor([0, 0, 0, 40])]; tensor var_11286_end_0 = const()[name = tensor("op_11286_end_0"), val = tensor([2, 77, 1, 80])]; tensor var_11286_end_mask_0 = const()[name = tensor("op_11286_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11286_cast_fp16 = slice_by_index(begin = var_11286_begin_0, end = var_11286_end_0, end_mask = var_11286_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11286_cast_fp16")]; tensor var_11290_begin_0 = const()[name = tensor("op_11290_begin_0"), val = tensor([0, 0, 0, 80])]; tensor var_11290_end_0 = const()[name = tensor("op_11290_end_0"), val = tensor([2, 77, 1, 120])]; tensor var_11290_end_mask_0 = const()[name = tensor("op_11290_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11290_cast_fp16 = slice_by_index(begin = var_11290_begin_0, end = var_11290_end_0, end_mask = var_11290_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11290_cast_fp16")]; tensor var_11294_begin_0 = const()[name = tensor("op_11294_begin_0"), val = tensor([0, 0, 0, 120])]; tensor var_11294_end_0 = const()[name = tensor("op_11294_end_0"), val = tensor([2, 77, 1, 160])]; tensor var_11294_end_mask_0 = const()[name = tensor("op_11294_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11294_cast_fp16 = slice_by_index(begin = var_11294_begin_0, end = var_11294_end_0, end_mask = var_11294_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11294_cast_fp16")]; tensor var_11298_begin_0 = const()[name = tensor("op_11298_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_11298_end_0 = const()[name = tensor("op_11298_end_0"), val = tensor([2, 77, 1, 200])]; tensor var_11298_end_mask_0 = const()[name = tensor("op_11298_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11298_cast_fp16 = slice_by_index(begin = var_11298_begin_0, end = var_11298_end_0, end_mask = var_11298_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11298_cast_fp16")]; tensor var_11302_begin_0 = const()[name = tensor("op_11302_begin_0"), val = tensor([0, 0, 0, 200])]; tensor var_11302_end_0 = const()[name = tensor("op_11302_end_0"), val = tensor([2, 77, 1, 240])]; tensor var_11302_end_mask_0 = const()[name = tensor("op_11302_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11302_cast_fp16 = slice_by_index(begin = var_11302_begin_0, end = var_11302_end_0, end_mask = var_11302_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11302_cast_fp16")]; tensor var_11306_begin_0 = const()[name = tensor("op_11306_begin_0"), val = tensor([0, 0, 0, 240])]; tensor var_11306_end_0 = const()[name = tensor("op_11306_end_0"), val = tensor([2, 77, 1, 280])]; tensor var_11306_end_mask_0 = const()[name = tensor("op_11306_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11306_cast_fp16 = slice_by_index(begin = var_11306_begin_0, end = var_11306_end_0, end_mask = var_11306_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11306_cast_fp16")]; tensor var_11310_begin_0 = const()[name = tensor("op_11310_begin_0"), val = tensor([0, 0, 0, 280])]; tensor var_11310_end_0 = const()[name = tensor("op_11310_end_0"), val = tensor([2, 77, 1, 1])]; tensor var_11310_end_mask_0 = const()[name = tensor("op_11310_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_11310_cast_fp16 = slice_by_index(begin = var_11310_begin_0, end = var_11310_end_0, end_mask = var_11310_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11310_cast_fp16")]; tensor var_11312_begin_0 = const()[name = tensor("op_11312_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11312_end_0 = const()[name = tensor("op_11312_end_0"), val = tensor([2, 40, 1, 77])]; tensor var_11312_end_mask_0 = const()[name = tensor("op_11312_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11312_cast_fp16 = slice_by_index(begin = var_11312_begin_0, end = var_11312_end_0, end_mask = var_11312_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11312_cast_fp16")]; tensor var_11316_begin_0 = const()[name = tensor("op_11316_begin_0"), val = tensor([0, 40, 0, 0])]; tensor var_11316_end_0 = const()[name = tensor("op_11316_end_0"), val = tensor([2, 80, 1, 77])]; tensor var_11316_end_mask_0 = const()[name = tensor("op_11316_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11316_cast_fp16 = slice_by_index(begin = var_11316_begin_0, end = var_11316_end_0, end_mask = var_11316_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11316_cast_fp16")]; tensor var_11320_begin_0 = const()[name = tensor("op_11320_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_11320_end_0 = const()[name = tensor("op_11320_end_0"), val = tensor([2, 120, 1, 77])]; tensor var_11320_end_mask_0 = const()[name = tensor("op_11320_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11320_cast_fp16 = slice_by_index(begin = var_11320_begin_0, end = var_11320_end_0, end_mask = var_11320_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11320_cast_fp16")]; tensor var_11324_begin_0 = const()[name = tensor("op_11324_begin_0"), val = tensor([0, 120, 0, 0])]; tensor var_11324_end_0 = const()[name = tensor("op_11324_end_0"), val = tensor([2, 160, 1, 77])]; tensor var_11324_end_mask_0 = const()[name = tensor("op_11324_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11324_cast_fp16 = slice_by_index(begin = var_11324_begin_0, end = var_11324_end_0, end_mask = var_11324_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11324_cast_fp16")]; tensor var_11328_begin_0 = const()[name = tensor("op_11328_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_11328_end_0 = const()[name = tensor("op_11328_end_0"), val = tensor([2, 200, 1, 77])]; tensor var_11328_end_mask_0 = const()[name = tensor("op_11328_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11328_cast_fp16 = slice_by_index(begin = var_11328_begin_0, end = var_11328_end_0, end_mask = var_11328_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11328_cast_fp16")]; tensor var_11332_begin_0 = const()[name = tensor("op_11332_begin_0"), val = tensor([0, 200, 0, 0])]; tensor var_11332_end_0 = const()[name = tensor("op_11332_end_0"), val = tensor([2, 240, 1, 77])]; tensor var_11332_end_mask_0 = const()[name = tensor("op_11332_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11332_cast_fp16 = slice_by_index(begin = var_11332_begin_0, end = var_11332_end_0, end_mask = var_11332_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11332_cast_fp16")]; tensor var_11336_begin_0 = const()[name = tensor("op_11336_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_11336_end_0 = const()[name = tensor("op_11336_end_0"), val = tensor([2, 280, 1, 77])]; tensor var_11336_end_mask_0 = const()[name = tensor("op_11336_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11336_cast_fp16 = slice_by_index(begin = var_11336_begin_0, end = var_11336_end_0, end_mask = var_11336_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11336_cast_fp16")]; tensor var_11340_begin_0 = const()[name = tensor("op_11340_begin_0"), val = tensor([0, 280, 0, 0])]; tensor var_11340_end_0 = const()[name = tensor("op_11340_end_0"), val = tensor([2, 1, 1, 77])]; tensor var_11340_end_mask_0 = const()[name = tensor("op_11340_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_11340_cast_fp16 = slice_by_index(begin = var_11340_begin_0, end = var_11340_end_0, end_mask = var_11340_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11340_cast_fp16")]; tensor var_11344_equation_0 = const()[name = tensor("op_11344_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11344_cast_fp16 = einsum(equation = var_11344_equation_0, values = (var_11282_cast_fp16, var_11214_cast_fp16))[name = tensor("op_11344_cast_fp16")]; tensor var_11345_to_fp16 = const()[name = tensor("op_11345_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_961_cast_fp16 = mul(x = var_11344_cast_fp16, y = var_11345_to_fp16)[name = tensor("aw_chunk_961_cast_fp16")]; tensor var_11348_equation_0 = const()[name = tensor("op_11348_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11348_cast_fp16 = einsum(equation = var_11348_equation_0, values = (var_11282_cast_fp16, var_11215_cast_fp16))[name = tensor("op_11348_cast_fp16")]; tensor var_11349_to_fp16 = const()[name = tensor("op_11349_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_963_cast_fp16 = mul(x = var_11348_cast_fp16, y = var_11349_to_fp16)[name = tensor("aw_chunk_963_cast_fp16")]; tensor var_11352_equation_0 = const()[name = tensor("op_11352_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11352_cast_fp16 = einsum(equation = var_11352_equation_0, values = (var_11282_cast_fp16, var_11216_cast_fp16))[name = tensor("op_11352_cast_fp16")]; tensor var_11353_to_fp16 = const()[name = tensor("op_11353_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_965_cast_fp16 = mul(x = var_11352_cast_fp16, y = var_11353_to_fp16)[name = tensor("aw_chunk_965_cast_fp16")]; tensor var_11356_equation_0 = const()[name = tensor("op_11356_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11356_cast_fp16 = einsum(equation = var_11356_equation_0, values = (var_11282_cast_fp16, var_11217_cast_fp16))[name = tensor("op_11356_cast_fp16")]; tensor var_11357_to_fp16 = const()[name = tensor("op_11357_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_967_cast_fp16 = mul(x = var_11356_cast_fp16, y = var_11357_to_fp16)[name = tensor("aw_chunk_967_cast_fp16")]; tensor var_11360_equation_0 = const()[name = tensor("op_11360_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11360_cast_fp16 = einsum(equation = var_11360_equation_0, values = (var_11282_cast_fp16, var_11218_cast_fp16))[name = tensor("op_11360_cast_fp16")]; tensor var_11361_to_fp16 = const()[name = tensor("op_11361_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_969_cast_fp16 = mul(x = var_11360_cast_fp16, y = var_11361_to_fp16)[name = tensor("aw_chunk_969_cast_fp16")]; tensor var_11364_equation_0 = const()[name = tensor("op_11364_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11364_cast_fp16 = einsum(equation = var_11364_equation_0, values = (var_11282_cast_fp16, var_11219_cast_fp16))[name = tensor("op_11364_cast_fp16")]; tensor var_11365_to_fp16 = const()[name = tensor("op_11365_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_971_cast_fp16 = mul(x = var_11364_cast_fp16, y = var_11365_to_fp16)[name = tensor("aw_chunk_971_cast_fp16")]; tensor var_11368_equation_0 = const()[name = tensor("op_11368_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11368_cast_fp16 = einsum(equation = var_11368_equation_0, values = (var_11282_cast_fp16, var_11220_cast_fp16))[name = tensor("op_11368_cast_fp16")]; tensor var_11369_to_fp16 = const()[name = tensor("op_11369_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_973_cast_fp16 = mul(x = var_11368_cast_fp16, y = var_11369_to_fp16)[name = tensor("aw_chunk_973_cast_fp16")]; tensor var_11372_equation_0 = const()[name = tensor("op_11372_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11372_cast_fp16 = einsum(equation = var_11372_equation_0, values = (var_11282_cast_fp16, var_11221_cast_fp16))[name = tensor("op_11372_cast_fp16")]; tensor var_11373_to_fp16 = const()[name = tensor("op_11373_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_975_cast_fp16 = mul(x = var_11372_cast_fp16, y = var_11373_to_fp16)[name = tensor("aw_chunk_975_cast_fp16")]; tensor var_11376_equation_0 = const()[name = tensor("op_11376_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11376_cast_fp16 = einsum(equation = var_11376_equation_0, values = (var_11286_cast_fp16, var_11222_cast_fp16))[name = tensor("op_11376_cast_fp16")]; tensor var_11377_to_fp16 = const()[name = tensor("op_11377_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_977_cast_fp16 = mul(x = var_11376_cast_fp16, y = var_11377_to_fp16)[name = tensor("aw_chunk_977_cast_fp16")]; tensor var_11380_equation_0 = const()[name = tensor("op_11380_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11380_cast_fp16 = einsum(equation = var_11380_equation_0, values = (var_11286_cast_fp16, var_11223_cast_fp16))[name = tensor("op_11380_cast_fp16")]; tensor var_11381_to_fp16 = const()[name = tensor("op_11381_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_979_cast_fp16 = mul(x = var_11380_cast_fp16, y = var_11381_to_fp16)[name = tensor("aw_chunk_979_cast_fp16")]; tensor var_11384_equation_0 = const()[name = tensor("op_11384_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11384_cast_fp16 = einsum(equation = var_11384_equation_0, values = (var_11286_cast_fp16, var_11224_cast_fp16))[name = tensor("op_11384_cast_fp16")]; tensor var_11385_to_fp16 = const()[name = tensor("op_11385_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_981_cast_fp16 = mul(x = var_11384_cast_fp16, y = var_11385_to_fp16)[name = tensor("aw_chunk_981_cast_fp16")]; tensor var_11388_equation_0 = const()[name = tensor("op_11388_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11388_cast_fp16 = einsum(equation = var_11388_equation_0, values = (var_11286_cast_fp16, var_11225_cast_fp16))[name = tensor("op_11388_cast_fp16")]; tensor var_11389_to_fp16 = const()[name = tensor("op_11389_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_983_cast_fp16 = mul(x = var_11388_cast_fp16, y = var_11389_to_fp16)[name = tensor("aw_chunk_983_cast_fp16")]; tensor var_11392_equation_0 = const()[name = tensor("op_11392_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11392_cast_fp16 = einsum(equation = var_11392_equation_0, values = (var_11286_cast_fp16, var_11226_cast_fp16))[name = tensor("op_11392_cast_fp16")]; tensor var_11393_to_fp16 = const()[name = tensor("op_11393_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_985_cast_fp16 = mul(x = var_11392_cast_fp16, y = var_11393_to_fp16)[name = tensor("aw_chunk_985_cast_fp16")]; tensor var_11396_equation_0 = const()[name = tensor("op_11396_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11396_cast_fp16 = einsum(equation = var_11396_equation_0, values = (var_11286_cast_fp16, var_11227_cast_fp16))[name = tensor("op_11396_cast_fp16")]; tensor var_11397_to_fp16 = const()[name = tensor("op_11397_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_987_cast_fp16 = mul(x = var_11396_cast_fp16, y = var_11397_to_fp16)[name = tensor("aw_chunk_987_cast_fp16")]; tensor var_11400_equation_0 = const()[name = tensor("op_11400_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11400_cast_fp16 = einsum(equation = var_11400_equation_0, values = (var_11286_cast_fp16, var_11228_cast_fp16))[name = tensor("op_11400_cast_fp16")]; tensor var_11401_to_fp16 = const()[name = tensor("op_11401_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_989_cast_fp16 = mul(x = var_11400_cast_fp16, y = var_11401_to_fp16)[name = tensor("aw_chunk_989_cast_fp16")]; tensor var_11404_equation_0 = const()[name = tensor("op_11404_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11404_cast_fp16 = einsum(equation = var_11404_equation_0, values = (var_11286_cast_fp16, var_11229_cast_fp16))[name = tensor("op_11404_cast_fp16")]; tensor var_11405_to_fp16 = const()[name = tensor("op_11405_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_991_cast_fp16 = mul(x = var_11404_cast_fp16, y = var_11405_to_fp16)[name = tensor("aw_chunk_991_cast_fp16")]; tensor var_11408_equation_0 = const()[name = tensor("op_11408_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11408_cast_fp16 = einsum(equation = var_11408_equation_0, values = (var_11290_cast_fp16, var_11230_cast_fp16))[name = tensor("op_11408_cast_fp16")]; tensor var_11409_to_fp16 = const()[name = tensor("op_11409_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_993_cast_fp16 = mul(x = var_11408_cast_fp16, y = var_11409_to_fp16)[name = tensor("aw_chunk_993_cast_fp16")]; tensor var_11412_equation_0 = const()[name = tensor("op_11412_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11412_cast_fp16 = einsum(equation = var_11412_equation_0, values = (var_11290_cast_fp16, var_11231_cast_fp16))[name = tensor("op_11412_cast_fp16")]; tensor var_11413_to_fp16 = const()[name = tensor("op_11413_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_995_cast_fp16 = mul(x = var_11412_cast_fp16, y = var_11413_to_fp16)[name = tensor("aw_chunk_995_cast_fp16")]; tensor var_11416_equation_0 = const()[name = tensor("op_11416_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11416_cast_fp16 = einsum(equation = var_11416_equation_0, values = (var_11290_cast_fp16, var_11232_cast_fp16))[name = tensor("op_11416_cast_fp16")]; tensor var_11417_to_fp16 = const()[name = tensor("op_11417_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_997_cast_fp16 = mul(x = var_11416_cast_fp16, y = var_11417_to_fp16)[name = tensor("aw_chunk_997_cast_fp16")]; tensor var_11420_equation_0 = const()[name = tensor("op_11420_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11420_cast_fp16 = einsum(equation = var_11420_equation_0, values = (var_11290_cast_fp16, var_11233_cast_fp16))[name = tensor("op_11420_cast_fp16")]; tensor var_11421_to_fp16 = const()[name = tensor("op_11421_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_999_cast_fp16 = mul(x = var_11420_cast_fp16, y = var_11421_to_fp16)[name = tensor("aw_chunk_999_cast_fp16")]; tensor var_11424_equation_0 = const()[name = tensor("op_11424_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11424_cast_fp16 = einsum(equation = var_11424_equation_0, values = (var_11290_cast_fp16, var_11234_cast_fp16))[name = tensor("op_11424_cast_fp16")]; tensor var_11425_to_fp16 = const()[name = tensor("op_11425_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1001_cast_fp16 = mul(x = var_11424_cast_fp16, y = var_11425_to_fp16)[name = tensor("aw_chunk_1001_cast_fp16")]; tensor var_11428_equation_0 = const()[name = tensor("op_11428_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11428_cast_fp16 = einsum(equation = var_11428_equation_0, values = (var_11290_cast_fp16, var_11235_cast_fp16))[name = tensor("op_11428_cast_fp16")]; tensor var_11429_to_fp16 = const()[name = tensor("op_11429_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1003_cast_fp16 = mul(x = var_11428_cast_fp16, y = var_11429_to_fp16)[name = tensor("aw_chunk_1003_cast_fp16")]; tensor var_11432_equation_0 = const()[name = tensor("op_11432_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11432_cast_fp16 = einsum(equation = var_11432_equation_0, values = (var_11290_cast_fp16, var_11236_cast_fp16))[name = tensor("op_11432_cast_fp16")]; tensor var_11433_to_fp16 = const()[name = tensor("op_11433_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1005_cast_fp16 = mul(x = var_11432_cast_fp16, y = var_11433_to_fp16)[name = tensor("aw_chunk_1005_cast_fp16")]; tensor var_11436_equation_0 = const()[name = tensor("op_11436_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11436_cast_fp16 = einsum(equation = var_11436_equation_0, values = (var_11290_cast_fp16, var_11237_cast_fp16))[name = tensor("op_11436_cast_fp16")]; tensor var_11437_to_fp16 = const()[name = tensor("op_11437_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1007_cast_fp16 = mul(x = var_11436_cast_fp16, y = var_11437_to_fp16)[name = tensor("aw_chunk_1007_cast_fp16")]; tensor var_11440_equation_0 = const()[name = tensor("op_11440_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11440_cast_fp16 = einsum(equation = var_11440_equation_0, values = (var_11294_cast_fp16, var_11238_cast_fp16))[name = tensor("op_11440_cast_fp16")]; tensor var_11441_to_fp16 = const()[name = tensor("op_11441_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1009_cast_fp16 = mul(x = var_11440_cast_fp16, y = var_11441_to_fp16)[name = tensor("aw_chunk_1009_cast_fp16")]; tensor var_11444_equation_0 = const()[name = tensor("op_11444_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11444_cast_fp16 = einsum(equation = var_11444_equation_0, values = (var_11294_cast_fp16, var_11239_cast_fp16))[name = tensor("op_11444_cast_fp16")]; tensor var_11445_to_fp16 = const()[name = tensor("op_11445_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1011_cast_fp16 = mul(x = var_11444_cast_fp16, y = var_11445_to_fp16)[name = tensor("aw_chunk_1011_cast_fp16")]; tensor var_11448_equation_0 = const()[name = tensor("op_11448_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11448_cast_fp16 = einsum(equation = var_11448_equation_0, values = (var_11294_cast_fp16, var_11240_cast_fp16))[name = tensor("op_11448_cast_fp16")]; tensor var_11449_to_fp16 = const()[name = tensor("op_11449_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1013_cast_fp16 = mul(x = var_11448_cast_fp16, y = var_11449_to_fp16)[name = tensor("aw_chunk_1013_cast_fp16")]; tensor var_11452_equation_0 = const()[name = tensor("op_11452_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11452_cast_fp16 = einsum(equation = var_11452_equation_0, values = (var_11294_cast_fp16, var_11241_cast_fp16))[name = tensor("op_11452_cast_fp16")]; tensor var_11453_to_fp16 = const()[name = tensor("op_11453_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1015_cast_fp16 = mul(x = var_11452_cast_fp16, y = var_11453_to_fp16)[name = tensor("aw_chunk_1015_cast_fp16")]; tensor var_11456_equation_0 = const()[name = tensor("op_11456_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11456_cast_fp16 = einsum(equation = var_11456_equation_0, values = (var_11294_cast_fp16, var_11242_cast_fp16))[name = tensor("op_11456_cast_fp16")]; tensor var_11457_to_fp16 = const()[name = tensor("op_11457_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1017_cast_fp16 = mul(x = var_11456_cast_fp16, y = var_11457_to_fp16)[name = tensor("aw_chunk_1017_cast_fp16")]; tensor var_11460_equation_0 = const()[name = tensor("op_11460_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11460_cast_fp16 = einsum(equation = var_11460_equation_0, values = (var_11294_cast_fp16, var_11243_cast_fp16))[name = tensor("op_11460_cast_fp16")]; tensor var_11461_to_fp16 = const()[name = tensor("op_11461_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1019_cast_fp16 = mul(x = var_11460_cast_fp16, y = var_11461_to_fp16)[name = tensor("aw_chunk_1019_cast_fp16")]; tensor var_11464_equation_0 = const()[name = tensor("op_11464_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11464_cast_fp16 = einsum(equation = var_11464_equation_0, values = (var_11294_cast_fp16, var_11244_cast_fp16))[name = tensor("op_11464_cast_fp16")]; tensor var_11465_to_fp16 = const()[name = tensor("op_11465_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1021_cast_fp16 = mul(x = var_11464_cast_fp16, y = var_11465_to_fp16)[name = tensor("aw_chunk_1021_cast_fp16")]; tensor var_11468_equation_0 = const()[name = tensor("op_11468_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11468_cast_fp16 = einsum(equation = var_11468_equation_0, values = (var_11294_cast_fp16, var_11245_cast_fp16))[name = tensor("op_11468_cast_fp16")]; tensor var_11469_to_fp16 = const()[name = tensor("op_11469_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1023_cast_fp16 = mul(x = var_11468_cast_fp16, y = var_11469_to_fp16)[name = tensor("aw_chunk_1023_cast_fp16")]; tensor var_11472_equation_0 = const()[name = tensor("op_11472_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11472_cast_fp16 = einsum(equation = var_11472_equation_0, values = (var_11298_cast_fp16, var_11246_cast_fp16))[name = tensor("op_11472_cast_fp16")]; tensor var_11473_to_fp16 = const()[name = tensor("op_11473_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1025_cast_fp16 = mul(x = var_11472_cast_fp16, y = var_11473_to_fp16)[name = tensor("aw_chunk_1025_cast_fp16")]; tensor var_11476_equation_0 = const()[name = tensor("op_11476_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11476_cast_fp16 = einsum(equation = var_11476_equation_0, values = (var_11298_cast_fp16, var_11247_cast_fp16))[name = tensor("op_11476_cast_fp16")]; tensor var_11477_to_fp16 = const()[name = tensor("op_11477_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1027_cast_fp16 = mul(x = var_11476_cast_fp16, y = var_11477_to_fp16)[name = tensor("aw_chunk_1027_cast_fp16")]; tensor var_11480_equation_0 = const()[name = tensor("op_11480_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11480_cast_fp16 = einsum(equation = var_11480_equation_0, values = (var_11298_cast_fp16, var_11248_cast_fp16))[name = tensor("op_11480_cast_fp16")]; tensor var_11481_to_fp16 = const()[name = tensor("op_11481_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1029_cast_fp16 = mul(x = var_11480_cast_fp16, y = var_11481_to_fp16)[name = tensor("aw_chunk_1029_cast_fp16")]; tensor var_11484_equation_0 = const()[name = tensor("op_11484_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11484_cast_fp16 = einsum(equation = var_11484_equation_0, values = (var_11298_cast_fp16, var_11249_cast_fp16))[name = tensor("op_11484_cast_fp16")]; tensor var_11485_to_fp16 = const()[name = tensor("op_11485_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1031_cast_fp16 = mul(x = var_11484_cast_fp16, y = var_11485_to_fp16)[name = tensor("aw_chunk_1031_cast_fp16")]; tensor var_11488_equation_0 = const()[name = tensor("op_11488_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11488_cast_fp16 = einsum(equation = var_11488_equation_0, values = (var_11298_cast_fp16, var_11250_cast_fp16))[name = tensor("op_11488_cast_fp16")]; tensor var_11489_to_fp16 = const()[name = tensor("op_11489_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1033_cast_fp16 = mul(x = var_11488_cast_fp16, y = var_11489_to_fp16)[name = tensor("aw_chunk_1033_cast_fp16")]; tensor var_11492_equation_0 = const()[name = tensor("op_11492_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11492_cast_fp16 = einsum(equation = var_11492_equation_0, values = (var_11298_cast_fp16, var_11251_cast_fp16))[name = tensor("op_11492_cast_fp16")]; tensor var_11493_to_fp16 = const()[name = tensor("op_11493_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1035_cast_fp16 = mul(x = var_11492_cast_fp16, y = var_11493_to_fp16)[name = tensor("aw_chunk_1035_cast_fp16")]; tensor var_11496_equation_0 = const()[name = tensor("op_11496_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11496_cast_fp16 = einsum(equation = var_11496_equation_0, values = (var_11298_cast_fp16, var_11252_cast_fp16))[name = tensor("op_11496_cast_fp16")]; tensor var_11497_to_fp16 = const()[name = tensor("op_11497_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1037_cast_fp16 = mul(x = var_11496_cast_fp16, y = var_11497_to_fp16)[name = tensor("aw_chunk_1037_cast_fp16")]; tensor var_11500_equation_0 = const()[name = tensor("op_11500_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11500_cast_fp16 = einsum(equation = var_11500_equation_0, values = (var_11298_cast_fp16, var_11253_cast_fp16))[name = tensor("op_11500_cast_fp16")]; tensor var_11501_to_fp16 = const()[name = tensor("op_11501_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1039_cast_fp16 = mul(x = var_11500_cast_fp16, y = var_11501_to_fp16)[name = tensor("aw_chunk_1039_cast_fp16")]; tensor var_11504_equation_0 = const()[name = tensor("op_11504_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11504_cast_fp16 = einsum(equation = var_11504_equation_0, values = (var_11302_cast_fp16, var_11254_cast_fp16))[name = tensor("op_11504_cast_fp16")]; tensor var_11505_to_fp16 = const()[name = tensor("op_11505_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1041_cast_fp16 = mul(x = var_11504_cast_fp16, y = var_11505_to_fp16)[name = tensor("aw_chunk_1041_cast_fp16")]; tensor var_11508_equation_0 = const()[name = tensor("op_11508_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11508_cast_fp16 = einsum(equation = var_11508_equation_0, values = (var_11302_cast_fp16, var_11255_cast_fp16))[name = tensor("op_11508_cast_fp16")]; tensor var_11509_to_fp16 = const()[name = tensor("op_11509_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1043_cast_fp16 = mul(x = var_11508_cast_fp16, y = var_11509_to_fp16)[name = tensor("aw_chunk_1043_cast_fp16")]; tensor var_11512_equation_0 = const()[name = tensor("op_11512_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11512_cast_fp16 = einsum(equation = var_11512_equation_0, values = (var_11302_cast_fp16, var_11256_cast_fp16))[name = tensor("op_11512_cast_fp16")]; tensor var_11513_to_fp16 = const()[name = tensor("op_11513_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1045_cast_fp16 = mul(x = var_11512_cast_fp16, y = var_11513_to_fp16)[name = tensor("aw_chunk_1045_cast_fp16")]; tensor var_11516_equation_0 = const()[name = tensor("op_11516_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11516_cast_fp16 = einsum(equation = var_11516_equation_0, values = (var_11302_cast_fp16, var_11257_cast_fp16))[name = tensor("op_11516_cast_fp16")]; tensor var_11517_to_fp16 = const()[name = tensor("op_11517_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1047_cast_fp16 = mul(x = var_11516_cast_fp16, y = var_11517_to_fp16)[name = tensor("aw_chunk_1047_cast_fp16")]; tensor var_11520_equation_0 = const()[name = tensor("op_11520_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11520_cast_fp16 = einsum(equation = var_11520_equation_0, values = (var_11302_cast_fp16, var_11258_cast_fp16))[name = tensor("op_11520_cast_fp16")]; tensor var_11521_to_fp16 = const()[name = tensor("op_11521_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1049_cast_fp16 = mul(x = var_11520_cast_fp16, y = var_11521_to_fp16)[name = tensor("aw_chunk_1049_cast_fp16")]; tensor var_11524_equation_0 = const()[name = tensor("op_11524_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11524_cast_fp16 = einsum(equation = var_11524_equation_0, values = (var_11302_cast_fp16, var_11259_cast_fp16))[name = tensor("op_11524_cast_fp16")]; tensor var_11525_to_fp16 = const()[name = tensor("op_11525_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1051_cast_fp16 = mul(x = var_11524_cast_fp16, y = var_11525_to_fp16)[name = tensor("aw_chunk_1051_cast_fp16")]; tensor var_11528_equation_0 = const()[name = tensor("op_11528_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11528_cast_fp16 = einsum(equation = var_11528_equation_0, values = (var_11302_cast_fp16, var_11260_cast_fp16))[name = tensor("op_11528_cast_fp16")]; tensor var_11529_to_fp16 = const()[name = tensor("op_11529_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1053_cast_fp16 = mul(x = var_11528_cast_fp16, y = var_11529_to_fp16)[name = tensor("aw_chunk_1053_cast_fp16")]; tensor var_11532_equation_0 = const()[name = tensor("op_11532_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11532_cast_fp16 = einsum(equation = var_11532_equation_0, values = (var_11302_cast_fp16, var_11261_cast_fp16))[name = tensor("op_11532_cast_fp16")]; tensor var_11533_to_fp16 = const()[name = tensor("op_11533_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1055_cast_fp16 = mul(x = var_11532_cast_fp16, y = var_11533_to_fp16)[name = tensor("aw_chunk_1055_cast_fp16")]; tensor var_11536_equation_0 = const()[name = tensor("op_11536_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11536_cast_fp16 = einsum(equation = var_11536_equation_0, values = (var_11306_cast_fp16, var_11262_cast_fp16))[name = tensor("op_11536_cast_fp16")]; tensor var_11537_to_fp16 = const()[name = tensor("op_11537_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1057_cast_fp16 = mul(x = var_11536_cast_fp16, y = var_11537_to_fp16)[name = tensor("aw_chunk_1057_cast_fp16")]; tensor var_11540_equation_0 = const()[name = tensor("op_11540_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11540_cast_fp16 = einsum(equation = var_11540_equation_0, values = (var_11306_cast_fp16, var_11263_cast_fp16))[name = tensor("op_11540_cast_fp16")]; tensor var_11541_to_fp16 = const()[name = tensor("op_11541_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1059_cast_fp16 = mul(x = var_11540_cast_fp16, y = var_11541_to_fp16)[name = tensor("aw_chunk_1059_cast_fp16")]; tensor var_11544_equation_0 = const()[name = tensor("op_11544_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11544_cast_fp16 = einsum(equation = var_11544_equation_0, values = (var_11306_cast_fp16, var_11264_cast_fp16))[name = tensor("op_11544_cast_fp16")]; tensor var_11545_to_fp16 = const()[name = tensor("op_11545_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1061_cast_fp16 = mul(x = var_11544_cast_fp16, y = var_11545_to_fp16)[name = tensor("aw_chunk_1061_cast_fp16")]; tensor var_11548_equation_0 = const()[name = tensor("op_11548_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11548_cast_fp16 = einsum(equation = var_11548_equation_0, values = (var_11306_cast_fp16, var_11265_cast_fp16))[name = tensor("op_11548_cast_fp16")]; tensor var_11549_to_fp16 = const()[name = tensor("op_11549_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1063_cast_fp16 = mul(x = var_11548_cast_fp16, y = var_11549_to_fp16)[name = tensor("aw_chunk_1063_cast_fp16")]; tensor var_11552_equation_0 = const()[name = tensor("op_11552_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11552_cast_fp16 = einsum(equation = var_11552_equation_0, values = (var_11306_cast_fp16, var_11266_cast_fp16))[name = tensor("op_11552_cast_fp16")]; tensor var_11553_to_fp16 = const()[name = tensor("op_11553_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1065_cast_fp16 = mul(x = var_11552_cast_fp16, y = var_11553_to_fp16)[name = tensor("aw_chunk_1065_cast_fp16")]; tensor var_11556_equation_0 = const()[name = tensor("op_11556_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11556_cast_fp16 = einsum(equation = var_11556_equation_0, values = (var_11306_cast_fp16, var_11267_cast_fp16))[name = tensor("op_11556_cast_fp16")]; tensor var_11557_to_fp16 = const()[name = tensor("op_11557_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1067_cast_fp16 = mul(x = var_11556_cast_fp16, y = var_11557_to_fp16)[name = tensor("aw_chunk_1067_cast_fp16")]; tensor var_11560_equation_0 = const()[name = tensor("op_11560_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11560_cast_fp16 = einsum(equation = var_11560_equation_0, values = (var_11306_cast_fp16, var_11268_cast_fp16))[name = tensor("op_11560_cast_fp16")]; tensor var_11561_to_fp16 = const()[name = tensor("op_11561_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1069_cast_fp16 = mul(x = var_11560_cast_fp16, y = var_11561_to_fp16)[name = tensor("aw_chunk_1069_cast_fp16")]; tensor var_11564_equation_0 = const()[name = tensor("op_11564_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11564_cast_fp16 = einsum(equation = var_11564_equation_0, values = (var_11306_cast_fp16, var_11269_cast_fp16))[name = tensor("op_11564_cast_fp16")]; tensor var_11565_to_fp16 = const()[name = tensor("op_11565_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1071_cast_fp16 = mul(x = var_11564_cast_fp16, y = var_11565_to_fp16)[name = tensor("aw_chunk_1071_cast_fp16")]; tensor var_11568_equation_0 = const()[name = tensor("op_11568_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11568_cast_fp16 = einsum(equation = var_11568_equation_0, values = (var_11310_cast_fp16, var_11270_cast_fp16))[name = tensor("op_11568_cast_fp16")]; tensor var_11569_to_fp16 = const()[name = tensor("op_11569_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1073_cast_fp16 = mul(x = var_11568_cast_fp16, y = var_11569_to_fp16)[name = tensor("aw_chunk_1073_cast_fp16")]; tensor var_11572_equation_0 = const()[name = tensor("op_11572_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11572_cast_fp16 = einsum(equation = var_11572_equation_0, values = (var_11310_cast_fp16, var_11271_cast_fp16))[name = tensor("op_11572_cast_fp16")]; tensor var_11573_to_fp16 = const()[name = tensor("op_11573_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1075_cast_fp16 = mul(x = var_11572_cast_fp16, y = var_11573_to_fp16)[name = tensor("aw_chunk_1075_cast_fp16")]; tensor var_11576_equation_0 = const()[name = tensor("op_11576_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11576_cast_fp16 = einsum(equation = var_11576_equation_0, values = (var_11310_cast_fp16, var_11272_cast_fp16))[name = tensor("op_11576_cast_fp16")]; tensor var_11577_to_fp16 = const()[name = tensor("op_11577_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1077_cast_fp16 = mul(x = var_11576_cast_fp16, y = var_11577_to_fp16)[name = tensor("aw_chunk_1077_cast_fp16")]; tensor var_11580_equation_0 = const()[name = tensor("op_11580_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11580_cast_fp16 = einsum(equation = var_11580_equation_0, values = (var_11310_cast_fp16, var_11273_cast_fp16))[name = tensor("op_11580_cast_fp16")]; tensor var_11581_to_fp16 = const()[name = tensor("op_11581_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1079_cast_fp16 = mul(x = var_11580_cast_fp16, y = var_11581_to_fp16)[name = tensor("aw_chunk_1079_cast_fp16")]; tensor var_11584_equation_0 = const()[name = tensor("op_11584_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11584_cast_fp16 = einsum(equation = var_11584_equation_0, values = (var_11310_cast_fp16, var_11274_cast_fp16))[name = tensor("op_11584_cast_fp16")]; tensor var_11585_to_fp16 = const()[name = tensor("op_11585_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1081_cast_fp16 = mul(x = var_11584_cast_fp16, y = var_11585_to_fp16)[name = tensor("aw_chunk_1081_cast_fp16")]; tensor var_11588_equation_0 = const()[name = tensor("op_11588_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11588_cast_fp16 = einsum(equation = var_11588_equation_0, values = (var_11310_cast_fp16, var_11275_cast_fp16))[name = tensor("op_11588_cast_fp16")]; tensor var_11589_to_fp16 = const()[name = tensor("op_11589_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1083_cast_fp16 = mul(x = var_11588_cast_fp16, y = var_11589_to_fp16)[name = tensor("aw_chunk_1083_cast_fp16")]; tensor var_11592_equation_0 = const()[name = tensor("op_11592_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11592_cast_fp16 = einsum(equation = var_11592_equation_0, values = (var_11310_cast_fp16, var_11276_cast_fp16))[name = tensor("op_11592_cast_fp16")]; tensor var_11593_to_fp16 = const()[name = tensor("op_11593_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1085_cast_fp16 = mul(x = var_11592_cast_fp16, y = var_11593_to_fp16)[name = tensor("aw_chunk_1085_cast_fp16")]; tensor var_11596_equation_0 = const()[name = tensor("op_11596_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11596_cast_fp16 = einsum(equation = var_11596_equation_0, values = (var_11310_cast_fp16, var_11277_cast_fp16))[name = tensor("op_11596_cast_fp16")]; tensor var_11597_to_fp16 = const()[name = tensor("op_11597_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1087_cast_fp16 = mul(x = var_11596_cast_fp16, y = var_11597_to_fp16)[name = tensor("aw_chunk_1087_cast_fp16")]; tensor var_11599_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_961_cast_fp16)[name = tensor("op_11599_cast_fp16")]; tensor var_11600_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_963_cast_fp16)[name = tensor("op_11600_cast_fp16")]; tensor var_11601_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_965_cast_fp16)[name = tensor("op_11601_cast_fp16")]; tensor var_11602_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_967_cast_fp16)[name = tensor("op_11602_cast_fp16")]; tensor var_11603_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_969_cast_fp16)[name = tensor("op_11603_cast_fp16")]; tensor var_11604_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_971_cast_fp16)[name = tensor("op_11604_cast_fp16")]; tensor var_11605_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_973_cast_fp16)[name = tensor("op_11605_cast_fp16")]; tensor var_11606_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_975_cast_fp16)[name = tensor("op_11606_cast_fp16")]; tensor var_11607_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_977_cast_fp16)[name = tensor("op_11607_cast_fp16")]; tensor var_11608_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_979_cast_fp16)[name = tensor("op_11608_cast_fp16")]; tensor var_11609_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_981_cast_fp16)[name = tensor("op_11609_cast_fp16")]; tensor var_11610_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_983_cast_fp16)[name = tensor("op_11610_cast_fp16")]; tensor var_11611_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_985_cast_fp16)[name = tensor("op_11611_cast_fp16")]; tensor var_11612_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_987_cast_fp16)[name = tensor("op_11612_cast_fp16")]; tensor var_11613_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_989_cast_fp16)[name = tensor("op_11613_cast_fp16")]; tensor var_11614_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_991_cast_fp16)[name = tensor("op_11614_cast_fp16")]; tensor var_11615_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_993_cast_fp16)[name = tensor("op_11615_cast_fp16")]; tensor var_11616_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_995_cast_fp16)[name = tensor("op_11616_cast_fp16")]; tensor var_11617_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_997_cast_fp16)[name = tensor("op_11617_cast_fp16")]; tensor var_11618_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_999_cast_fp16)[name = tensor("op_11618_cast_fp16")]; tensor var_11619_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1001_cast_fp16)[name = tensor("op_11619_cast_fp16")]; tensor var_11620_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1003_cast_fp16)[name = tensor("op_11620_cast_fp16")]; tensor var_11621_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1005_cast_fp16)[name = tensor("op_11621_cast_fp16")]; tensor var_11622_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1007_cast_fp16)[name = tensor("op_11622_cast_fp16")]; tensor var_11623_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1009_cast_fp16)[name = tensor("op_11623_cast_fp16")]; tensor var_11624_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1011_cast_fp16)[name = tensor("op_11624_cast_fp16")]; tensor var_11625_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1013_cast_fp16)[name = tensor("op_11625_cast_fp16")]; tensor var_11626_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1015_cast_fp16)[name = tensor("op_11626_cast_fp16")]; tensor var_11627_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1017_cast_fp16)[name = tensor("op_11627_cast_fp16")]; tensor var_11628_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1019_cast_fp16)[name = tensor("op_11628_cast_fp16")]; tensor var_11629_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1021_cast_fp16)[name = tensor("op_11629_cast_fp16")]; tensor var_11630_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1023_cast_fp16)[name = tensor("op_11630_cast_fp16")]; tensor var_11631_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1025_cast_fp16)[name = tensor("op_11631_cast_fp16")]; tensor var_11632_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1027_cast_fp16)[name = tensor("op_11632_cast_fp16")]; tensor var_11633_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1029_cast_fp16)[name = tensor("op_11633_cast_fp16")]; tensor var_11634_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1031_cast_fp16)[name = tensor("op_11634_cast_fp16")]; tensor var_11635_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1033_cast_fp16)[name = tensor("op_11635_cast_fp16")]; tensor var_11636_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1035_cast_fp16)[name = tensor("op_11636_cast_fp16")]; tensor var_11637_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1037_cast_fp16)[name = tensor("op_11637_cast_fp16")]; tensor var_11638_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1039_cast_fp16)[name = tensor("op_11638_cast_fp16")]; tensor var_11639_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1041_cast_fp16)[name = tensor("op_11639_cast_fp16")]; tensor var_11640_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1043_cast_fp16)[name = tensor("op_11640_cast_fp16")]; tensor var_11641_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1045_cast_fp16)[name = tensor("op_11641_cast_fp16")]; tensor var_11642_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1047_cast_fp16)[name = tensor("op_11642_cast_fp16")]; tensor var_11643_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1049_cast_fp16)[name = tensor("op_11643_cast_fp16")]; tensor var_11644_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1051_cast_fp16)[name = tensor("op_11644_cast_fp16")]; tensor var_11645_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1053_cast_fp16)[name = tensor("op_11645_cast_fp16")]; tensor var_11646_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1055_cast_fp16)[name = tensor("op_11646_cast_fp16")]; tensor var_11647_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1057_cast_fp16)[name = tensor("op_11647_cast_fp16")]; tensor var_11648_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1059_cast_fp16)[name = tensor("op_11648_cast_fp16")]; tensor var_11649_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1061_cast_fp16)[name = tensor("op_11649_cast_fp16")]; tensor var_11650_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1063_cast_fp16)[name = tensor("op_11650_cast_fp16")]; tensor var_11651_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1065_cast_fp16)[name = tensor("op_11651_cast_fp16")]; tensor var_11652_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1067_cast_fp16)[name = tensor("op_11652_cast_fp16")]; tensor var_11653_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1069_cast_fp16)[name = tensor("op_11653_cast_fp16")]; tensor var_11654_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1071_cast_fp16)[name = tensor("op_11654_cast_fp16")]; tensor var_11655_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1073_cast_fp16)[name = tensor("op_11655_cast_fp16")]; tensor var_11656_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1075_cast_fp16)[name = tensor("op_11656_cast_fp16")]; tensor var_11657_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1077_cast_fp16)[name = tensor("op_11657_cast_fp16")]; tensor var_11658_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1079_cast_fp16)[name = tensor("op_11658_cast_fp16")]; tensor var_11659_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1081_cast_fp16)[name = tensor("op_11659_cast_fp16")]; tensor var_11660_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1083_cast_fp16)[name = tensor("op_11660_cast_fp16")]; tensor var_11661_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1085_cast_fp16)[name = tensor("op_11661_cast_fp16")]; tensor var_11662_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1087_cast_fp16)[name = tensor("op_11662_cast_fp16")]; tensor var_11664_equation_0 = const()[name = tensor("op_11664_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11664_cast_fp16 = einsum(equation = var_11664_equation_0, values = (var_11312_cast_fp16, var_11599_cast_fp16))[name = tensor("op_11664_cast_fp16")]; tensor var_11666_equation_0 = const()[name = tensor("op_11666_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11666_cast_fp16 = einsum(equation = var_11666_equation_0, values = (var_11312_cast_fp16, var_11600_cast_fp16))[name = tensor("op_11666_cast_fp16")]; tensor var_11668_equation_0 = const()[name = tensor("op_11668_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11668_cast_fp16 = einsum(equation = var_11668_equation_0, values = (var_11312_cast_fp16, var_11601_cast_fp16))[name = tensor("op_11668_cast_fp16")]; tensor var_11670_equation_0 = const()[name = tensor("op_11670_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11670_cast_fp16 = einsum(equation = var_11670_equation_0, values = (var_11312_cast_fp16, var_11602_cast_fp16))[name = tensor("op_11670_cast_fp16")]; tensor var_11672_equation_0 = const()[name = tensor("op_11672_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11672_cast_fp16 = einsum(equation = var_11672_equation_0, values = (var_11312_cast_fp16, var_11603_cast_fp16))[name = tensor("op_11672_cast_fp16")]; tensor var_11674_equation_0 = const()[name = tensor("op_11674_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11674_cast_fp16 = einsum(equation = var_11674_equation_0, values = (var_11312_cast_fp16, var_11604_cast_fp16))[name = tensor("op_11674_cast_fp16")]; tensor var_11676_equation_0 = const()[name = tensor("op_11676_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11676_cast_fp16 = einsum(equation = var_11676_equation_0, values = (var_11312_cast_fp16, var_11605_cast_fp16))[name = tensor("op_11676_cast_fp16")]; tensor var_11678_equation_0 = const()[name = tensor("op_11678_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11678_cast_fp16 = einsum(equation = var_11678_equation_0, values = (var_11312_cast_fp16, var_11606_cast_fp16))[name = tensor("op_11678_cast_fp16")]; tensor var_11680_equation_0 = const()[name = tensor("op_11680_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11680_cast_fp16 = einsum(equation = var_11680_equation_0, values = (var_11316_cast_fp16, var_11607_cast_fp16))[name = tensor("op_11680_cast_fp16")]; tensor var_11682_equation_0 = const()[name = tensor("op_11682_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11682_cast_fp16 = einsum(equation = var_11682_equation_0, values = (var_11316_cast_fp16, var_11608_cast_fp16))[name = tensor("op_11682_cast_fp16")]; tensor var_11684_equation_0 = const()[name = tensor("op_11684_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11684_cast_fp16 = einsum(equation = var_11684_equation_0, values = (var_11316_cast_fp16, var_11609_cast_fp16))[name = tensor("op_11684_cast_fp16")]; tensor var_11686_equation_0 = const()[name = tensor("op_11686_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11686_cast_fp16 = einsum(equation = var_11686_equation_0, values = (var_11316_cast_fp16, var_11610_cast_fp16))[name = tensor("op_11686_cast_fp16")]; tensor var_11688_equation_0 = const()[name = tensor("op_11688_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11688_cast_fp16 = einsum(equation = var_11688_equation_0, values = (var_11316_cast_fp16, var_11611_cast_fp16))[name = tensor("op_11688_cast_fp16")]; tensor var_11690_equation_0 = const()[name = tensor("op_11690_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11690_cast_fp16 = einsum(equation = var_11690_equation_0, values = (var_11316_cast_fp16, var_11612_cast_fp16))[name = tensor("op_11690_cast_fp16")]; tensor var_11692_equation_0 = const()[name = tensor("op_11692_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11692_cast_fp16 = einsum(equation = var_11692_equation_0, values = (var_11316_cast_fp16, var_11613_cast_fp16))[name = tensor("op_11692_cast_fp16")]; tensor var_11694_equation_0 = const()[name = tensor("op_11694_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11694_cast_fp16 = einsum(equation = var_11694_equation_0, values = (var_11316_cast_fp16, var_11614_cast_fp16))[name = tensor("op_11694_cast_fp16")]; tensor var_11696_equation_0 = const()[name = tensor("op_11696_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11696_cast_fp16 = einsum(equation = var_11696_equation_0, values = (var_11320_cast_fp16, var_11615_cast_fp16))[name = tensor("op_11696_cast_fp16")]; tensor var_11698_equation_0 = const()[name = tensor("op_11698_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11698_cast_fp16 = einsum(equation = var_11698_equation_0, values = (var_11320_cast_fp16, var_11616_cast_fp16))[name = tensor("op_11698_cast_fp16")]; tensor var_11700_equation_0 = const()[name = tensor("op_11700_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11700_cast_fp16 = einsum(equation = var_11700_equation_0, values = (var_11320_cast_fp16, var_11617_cast_fp16))[name = tensor("op_11700_cast_fp16")]; tensor var_11702_equation_0 = const()[name = tensor("op_11702_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11702_cast_fp16 = einsum(equation = var_11702_equation_0, values = (var_11320_cast_fp16, var_11618_cast_fp16))[name = tensor("op_11702_cast_fp16")]; tensor var_11704_equation_0 = const()[name = tensor("op_11704_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11704_cast_fp16 = einsum(equation = var_11704_equation_0, values = (var_11320_cast_fp16, var_11619_cast_fp16))[name = tensor("op_11704_cast_fp16")]; tensor var_11706_equation_0 = const()[name = tensor("op_11706_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11706_cast_fp16 = einsum(equation = var_11706_equation_0, values = (var_11320_cast_fp16, var_11620_cast_fp16))[name = tensor("op_11706_cast_fp16")]; tensor var_11708_equation_0 = const()[name = tensor("op_11708_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11708_cast_fp16 = einsum(equation = var_11708_equation_0, values = (var_11320_cast_fp16, var_11621_cast_fp16))[name = tensor("op_11708_cast_fp16")]; tensor var_11710_equation_0 = const()[name = tensor("op_11710_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11710_cast_fp16 = einsum(equation = var_11710_equation_0, values = (var_11320_cast_fp16, var_11622_cast_fp16))[name = tensor("op_11710_cast_fp16")]; tensor var_11712_equation_0 = const()[name = tensor("op_11712_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11712_cast_fp16 = einsum(equation = var_11712_equation_0, values = (var_11324_cast_fp16, var_11623_cast_fp16))[name = tensor("op_11712_cast_fp16")]; tensor var_11714_equation_0 = const()[name = tensor("op_11714_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11714_cast_fp16 = einsum(equation = var_11714_equation_0, values = (var_11324_cast_fp16, var_11624_cast_fp16))[name = tensor("op_11714_cast_fp16")]; tensor var_11716_equation_0 = const()[name = tensor("op_11716_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11716_cast_fp16 = einsum(equation = var_11716_equation_0, values = (var_11324_cast_fp16, var_11625_cast_fp16))[name = tensor("op_11716_cast_fp16")]; tensor var_11718_equation_0 = const()[name = tensor("op_11718_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11718_cast_fp16 = einsum(equation = var_11718_equation_0, values = (var_11324_cast_fp16, var_11626_cast_fp16))[name = tensor("op_11718_cast_fp16")]; tensor var_11720_equation_0 = const()[name = tensor("op_11720_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11720_cast_fp16 = einsum(equation = var_11720_equation_0, values = (var_11324_cast_fp16, var_11627_cast_fp16))[name = tensor("op_11720_cast_fp16")]; tensor var_11722_equation_0 = const()[name = tensor("op_11722_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11722_cast_fp16 = einsum(equation = var_11722_equation_0, values = (var_11324_cast_fp16, var_11628_cast_fp16))[name = tensor("op_11722_cast_fp16")]; tensor var_11724_equation_0 = const()[name = tensor("op_11724_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11724_cast_fp16 = einsum(equation = var_11724_equation_0, values = (var_11324_cast_fp16, var_11629_cast_fp16))[name = tensor("op_11724_cast_fp16")]; tensor var_11726_equation_0 = const()[name = tensor("op_11726_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11726_cast_fp16 = einsum(equation = var_11726_equation_0, values = (var_11324_cast_fp16, var_11630_cast_fp16))[name = tensor("op_11726_cast_fp16")]; tensor var_11728_equation_0 = const()[name = tensor("op_11728_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11728_cast_fp16 = einsum(equation = var_11728_equation_0, values = (var_11328_cast_fp16, var_11631_cast_fp16))[name = tensor("op_11728_cast_fp16")]; tensor var_11730_equation_0 = const()[name = tensor("op_11730_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11730_cast_fp16 = einsum(equation = var_11730_equation_0, values = (var_11328_cast_fp16, var_11632_cast_fp16))[name = tensor("op_11730_cast_fp16")]; tensor var_11732_equation_0 = const()[name = tensor("op_11732_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11732_cast_fp16 = einsum(equation = var_11732_equation_0, values = (var_11328_cast_fp16, var_11633_cast_fp16))[name = tensor("op_11732_cast_fp16")]; tensor var_11734_equation_0 = const()[name = tensor("op_11734_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11734_cast_fp16 = einsum(equation = var_11734_equation_0, values = (var_11328_cast_fp16, var_11634_cast_fp16))[name = tensor("op_11734_cast_fp16")]; tensor var_11736_equation_0 = const()[name = tensor("op_11736_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11736_cast_fp16 = einsum(equation = var_11736_equation_0, values = (var_11328_cast_fp16, var_11635_cast_fp16))[name = tensor("op_11736_cast_fp16")]; tensor var_11738_equation_0 = const()[name = tensor("op_11738_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11738_cast_fp16 = einsum(equation = var_11738_equation_0, values = (var_11328_cast_fp16, var_11636_cast_fp16))[name = tensor("op_11738_cast_fp16")]; tensor var_11740_equation_0 = const()[name = tensor("op_11740_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11740_cast_fp16 = einsum(equation = var_11740_equation_0, values = (var_11328_cast_fp16, var_11637_cast_fp16))[name = tensor("op_11740_cast_fp16")]; tensor var_11742_equation_0 = const()[name = tensor("op_11742_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11742_cast_fp16 = einsum(equation = var_11742_equation_0, values = (var_11328_cast_fp16, var_11638_cast_fp16))[name = tensor("op_11742_cast_fp16")]; tensor var_11744_equation_0 = const()[name = tensor("op_11744_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11744_cast_fp16 = einsum(equation = var_11744_equation_0, values = (var_11332_cast_fp16, var_11639_cast_fp16))[name = tensor("op_11744_cast_fp16")]; tensor var_11746_equation_0 = const()[name = tensor("op_11746_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11746_cast_fp16 = einsum(equation = var_11746_equation_0, values = (var_11332_cast_fp16, var_11640_cast_fp16))[name = tensor("op_11746_cast_fp16")]; tensor var_11748_equation_0 = const()[name = tensor("op_11748_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11748_cast_fp16 = einsum(equation = var_11748_equation_0, values = (var_11332_cast_fp16, var_11641_cast_fp16))[name = tensor("op_11748_cast_fp16")]; tensor var_11750_equation_0 = const()[name = tensor("op_11750_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11750_cast_fp16 = einsum(equation = var_11750_equation_0, values = (var_11332_cast_fp16, var_11642_cast_fp16))[name = tensor("op_11750_cast_fp16")]; tensor var_11752_equation_0 = const()[name = tensor("op_11752_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11752_cast_fp16 = einsum(equation = var_11752_equation_0, values = (var_11332_cast_fp16, var_11643_cast_fp16))[name = tensor("op_11752_cast_fp16")]; tensor var_11754_equation_0 = const()[name = tensor("op_11754_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11754_cast_fp16 = einsum(equation = var_11754_equation_0, values = (var_11332_cast_fp16, var_11644_cast_fp16))[name = tensor("op_11754_cast_fp16")]; tensor var_11756_equation_0 = const()[name = tensor("op_11756_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11756_cast_fp16 = einsum(equation = var_11756_equation_0, values = (var_11332_cast_fp16, var_11645_cast_fp16))[name = tensor("op_11756_cast_fp16")]; tensor var_11758_equation_0 = const()[name = tensor("op_11758_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11758_cast_fp16 = einsum(equation = var_11758_equation_0, values = (var_11332_cast_fp16, var_11646_cast_fp16))[name = tensor("op_11758_cast_fp16")]; tensor var_11760_equation_0 = const()[name = tensor("op_11760_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11760_cast_fp16 = einsum(equation = var_11760_equation_0, values = (var_11336_cast_fp16, var_11647_cast_fp16))[name = tensor("op_11760_cast_fp16")]; tensor var_11762_equation_0 = const()[name = tensor("op_11762_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11762_cast_fp16 = einsum(equation = var_11762_equation_0, values = (var_11336_cast_fp16, var_11648_cast_fp16))[name = tensor("op_11762_cast_fp16")]; tensor var_11764_equation_0 = const()[name = tensor("op_11764_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11764_cast_fp16 = einsum(equation = var_11764_equation_0, values = (var_11336_cast_fp16, var_11649_cast_fp16))[name = tensor("op_11764_cast_fp16")]; tensor var_11766_equation_0 = const()[name = tensor("op_11766_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11766_cast_fp16 = einsum(equation = var_11766_equation_0, values = (var_11336_cast_fp16, var_11650_cast_fp16))[name = tensor("op_11766_cast_fp16")]; tensor var_11768_equation_0 = const()[name = tensor("op_11768_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11768_cast_fp16 = einsum(equation = var_11768_equation_0, values = (var_11336_cast_fp16, var_11651_cast_fp16))[name = tensor("op_11768_cast_fp16")]; tensor var_11770_equation_0 = const()[name = tensor("op_11770_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11770_cast_fp16 = einsum(equation = var_11770_equation_0, values = (var_11336_cast_fp16, var_11652_cast_fp16))[name = tensor("op_11770_cast_fp16")]; tensor var_11772_equation_0 = const()[name = tensor("op_11772_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11772_cast_fp16 = einsum(equation = var_11772_equation_0, values = (var_11336_cast_fp16, var_11653_cast_fp16))[name = tensor("op_11772_cast_fp16")]; tensor var_11774_equation_0 = const()[name = tensor("op_11774_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11774_cast_fp16 = einsum(equation = var_11774_equation_0, values = (var_11336_cast_fp16, var_11654_cast_fp16))[name = tensor("op_11774_cast_fp16")]; tensor var_11776_equation_0 = const()[name = tensor("op_11776_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11776_cast_fp16 = einsum(equation = var_11776_equation_0, values = (var_11340_cast_fp16, var_11655_cast_fp16))[name = tensor("op_11776_cast_fp16")]; tensor var_11778_equation_0 = const()[name = tensor("op_11778_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11778_cast_fp16 = einsum(equation = var_11778_equation_0, values = (var_11340_cast_fp16, var_11656_cast_fp16))[name = tensor("op_11778_cast_fp16")]; tensor var_11780_equation_0 = const()[name = tensor("op_11780_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11780_cast_fp16 = einsum(equation = var_11780_equation_0, values = (var_11340_cast_fp16, var_11657_cast_fp16))[name = tensor("op_11780_cast_fp16")]; tensor var_11782_equation_0 = const()[name = tensor("op_11782_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11782_cast_fp16 = einsum(equation = var_11782_equation_0, values = (var_11340_cast_fp16, var_11658_cast_fp16))[name = tensor("op_11782_cast_fp16")]; tensor var_11784_equation_0 = const()[name = tensor("op_11784_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11784_cast_fp16 = einsum(equation = var_11784_equation_0, values = (var_11340_cast_fp16, var_11659_cast_fp16))[name = tensor("op_11784_cast_fp16")]; tensor var_11786_equation_0 = const()[name = tensor("op_11786_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11786_cast_fp16 = einsum(equation = var_11786_equation_0, values = (var_11340_cast_fp16, var_11660_cast_fp16))[name = tensor("op_11786_cast_fp16")]; tensor var_11788_equation_0 = const()[name = tensor("op_11788_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11788_cast_fp16 = einsum(equation = var_11788_equation_0, values = (var_11340_cast_fp16, var_11661_cast_fp16))[name = tensor("op_11788_cast_fp16")]; tensor var_11790_equation_0 = const()[name = tensor("op_11790_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11790_cast_fp16 = einsum(equation = var_11790_equation_0, values = (var_11340_cast_fp16, var_11662_cast_fp16))[name = tensor("op_11790_cast_fp16")]; tensor var_11792_interleave_0 = const()[name = tensor("op_11792_interleave_0"), val = tensor(false)]; tensor var_11792_cast_fp16 = concat(axis = var_10347, interleave = var_11792_interleave_0, values = (var_11664_cast_fp16, var_11666_cast_fp16, var_11668_cast_fp16, var_11670_cast_fp16, var_11672_cast_fp16, var_11674_cast_fp16, var_11676_cast_fp16, var_11678_cast_fp16))[name = tensor("op_11792_cast_fp16")]; tensor var_11794_interleave_0 = const()[name = tensor("op_11794_interleave_0"), val = tensor(false)]; tensor var_11794_cast_fp16 = concat(axis = var_10347, interleave = var_11794_interleave_0, values = (var_11680_cast_fp16, var_11682_cast_fp16, var_11684_cast_fp16, var_11686_cast_fp16, var_11688_cast_fp16, var_11690_cast_fp16, var_11692_cast_fp16, var_11694_cast_fp16))[name = tensor("op_11794_cast_fp16")]; tensor var_11796_interleave_0 = const()[name = tensor("op_11796_interleave_0"), val = tensor(false)]; tensor var_11796_cast_fp16 = concat(axis = var_10347, interleave = var_11796_interleave_0, values = (var_11696_cast_fp16, var_11698_cast_fp16, var_11700_cast_fp16, var_11702_cast_fp16, var_11704_cast_fp16, var_11706_cast_fp16, var_11708_cast_fp16, var_11710_cast_fp16))[name = tensor("op_11796_cast_fp16")]; tensor var_11798_interleave_0 = const()[name = tensor("op_11798_interleave_0"), val = tensor(false)]; tensor var_11798_cast_fp16 = concat(axis = var_10347, interleave = var_11798_interleave_0, values = (var_11712_cast_fp16, var_11714_cast_fp16, var_11716_cast_fp16, var_11718_cast_fp16, var_11720_cast_fp16, var_11722_cast_fp16, var_11724_cast_fp16, var_11726_cast_fp16))[name = tensor("op_11798_cast_fp16")]; tensor var_11800_interleave_0 = const()[name = tensor("op_11800_interleave_0"), val = tensor(false)]; tensor var_11800_cast_fp16 = concat(axis = var_10347, interleave = var_11800_interleave_0, values = (var_11728_cast_fp16, var_11730_cast_fp16, var_11732_cast_fp16, var_11734_cast_fp16, var_11736_cast_fp16, var_11738_cast_fp16, var_11740_cast_fp16, var_11742_cast_fp16))[name = tensor("op_11800_cast_fp16")]; tensor var_11802_interleave_0 = const()[name = tensor("op_11802_interleave_0"), val = tensor(false)]; tensor var_11802_cast_fp16 = concat(axis = var_10347, interleave = var_11802_interleave_0, values = (var_11744_cast_fp16, var_11746_cast_fp16, var_11748_cast_fp16, var_11750_cast_fp16, var_11752_cast_fp16, var_11754_cast_fp16, var_11756_cast_fp16, var_11758_cast_fp16))[name = tensor("op_11802_cast_fp16")]; tensor var_11804_interleave_0 = const()[name = tensor("op_11804_interleave_0"), val = tensor(false)]; tensor var_11804_cast_fp16 = concat(axis = var_10347, interleave = var_11804_interleave_0, values = (var_11760_cast_fp16, var_11762_cast_fp16, var_11764_cast_fp16, var_11766_cast_fp16, var_11768_cast_fp16, var_11770_cast_fp16, var_11772_cast_fp16, var_11774_cast_fp16))[name = tensor("op_11804_cast_fp16")]; tensor var_11806_interleave_0 = const()[name = tensor("op_11806_interleave_0"), val = tensor(false)]; tensor var_11806_cast_fp16 = concat(axis = var_10347, interleave = var_11806_interleave_0, values = (var_11776_cast_fp16, var_11778_cast_fp16, var_11780_cast_fp16, var_11782_cast_fp16, var_11784_cast_fp16, var_11786_cast_fp16, var_11788_cast_fp16, var_11790_cast_fp16))[name = tensor("op_11806_cast_fp16")]; tensor input_467_interleave_0 = const()[name = tensor("input_467_interleave_0"), val = tensor(false)]; tensor input_467_cast_fp16 = concat(axis = var_10375, interleave = input_467_interleave_0, values = (var_11792_cast_fp16, var_11794_cast_fp16, var_11796_cast_fp16, var_11798_cast_fp16, var_11800_cast_fp16, var_11802_cast_fp16, var_11804_cast_fp16, var_11806_cast_fp16))[name = tensor("input_467_cast_fp16")]; tensor var_11816_pad_type_0 = const()[name = tensor("op_11816_pad_type_0"), val = tensor("valid")]; tensor var_11816_strides_0 = const()[name = tensor("op_11816_strides_0"), val = tensor([1, 1])]; tensor var_11816_pad_0 = const()[name = tensor("op_11816_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_11816_dilations_0 = const()[name = tensor("op_11816_dilations_0"), val = tensor([1, 1])]; tensor var_11816_groups_0 = const()[name = tensor("op_11816_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(804947776)))]; tensor up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(805152640)))]; tensor var_11816_cast_fp16 = conv(bias = up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_11816_dilations_0, groups = var_11816_groups_0, pad = var_11816_pad_0, pad_type = var_11816_pad_type_0, strides = var_11816_strides_0, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_467_cast_fp16)[name = tensor("op_11816_cast_fp16")]; tensor inputs_83_cast_fp16 = add(x = var_11816_cast_fp16, y = inputs_81_cast_fp16)[name = tensor("inputs_83_cast_fp16")]; tensor input_469_axes_0 = const()[name = tensor("input_469_axes_0"), val = tensor([1])]; tensor input_469_gamma_0_to_fp16 = const()[name = tensor("input_469_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(805153344)))]; tensor input_469_beta_0_to_fp16 = const()[name = tensor("input_469_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(805154048)))]; tensor var_11826_to_fp16 = const()[name = tensor("op_11826_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_469_cast_fp16 = layer_norm(axes = input_469_axes_0, beta = input_469_beta_0_to_fp16, epsilon = var_11826_to_fp16, gamma = input_469_gamma_0_to_fp16, x = inputs_83_cast_fp16)[name = tensor("input_469_cast_fp16")]; tensor var_11846_pad_type_0 = const()[name = tensor("op_11846_pad_type_0"), val = tensor("valid")]; tensor var_11846_strides_0 = const()[name = tensor("op_11846_strides_0"), val = tensor([1, 1])]; tensor var_11846_pad_0 = const()[name = tensor("op_11846_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_11846_dilations_0 = const()[name = tensor("op_11846_dilations_0"), val = tensor([1, 1])]; tensor var_11846_groups_0 = const()[name = tensor("op_11846_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(805154752)))]; tensor up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(806793216)))]; tensor var_11846_cast_fp16 = conv(bias = up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_11846_dilations_0, groups = var_11846_groups_0, pad = var_11846_pad_0, pad_type = var_11846_pad_type_0, strides = var_11846_strides_0, weight = up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_469_cast_fp16)[name = tensor("op_11846_cast_fp16")]; tensor var_11847_split_sizes_0 = const()[name = tensor("op_11847_split_sizes_0"), val = tensor([1280, 1280])]; tensor var_11847_axis_0 = const()[name = tensor("op_11847_axis_0"), val = tensor(1)]; tensor var_11847_cast_fp16_0, tensor var_11847_cast_fp16_1 = split(axis = var_11847_axis_0, split_sizes = var_11847_split_sizes_0, x = var_11846_cast_fp16)[name = tensor("op_11847_cast_fp16")]; tensor var_11849_mode_0 = const()[name = tensor("op_11849_mode_0"), val = tensor("EXACT")]; tensor var_11849_cast_fp16 = gelu(mode = var_11849_mode_0, x = var_11847_cast_fp16_1)[name = tensor("op_11849_cast_fp16")]; tensor input_471_cast_fp16 = mul(x = var_11847_cast_fp16_0, y = var_11849_cast_fp16)[name = tensor("input_471_cast_fp16")]; tensor var_11857_pad_type_0 = const()[name = tensor("op_11857_pad_type_0"), val = tensor("valid")]; tensor var_11857_strides_0 = const()[name = tensor("op_11857_strides_0"), val = tensor([1, 1])]; tensor var_11857_pad_0 = const()[name = tensor("op_11857_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_11857_dilations_0 = const()[name = tensor("op_11857_dilations_0"), val = tensor([1, 1])]; tensor var_11857_groups_0 = const()[name = tensor("op_11857_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(806798400)))]; tensor up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(807617664)))]; tensor var_11857_cast_fp16 = conv(bias = up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_11857_dilations_0, groups = var_11857_groups_0, pad = var_11857_pad_0, pad_type = var_11857_pad_type_0, strides = var_11857_strides_0, weight = up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_471_cast_fp16)[name = tensor("op_11857_cast_fp16")]; tensor hidden_states_293_cast_fp16 = add(x = var_11857_cast_fp16, y = inputs_83_cast_fp16)[name = tensor("hidden_states_293_cast_fp16")]; tensor var_11859 = const()[name = tensor("op_11859"), val = tensor([2, 320, 64, 64])]; tensor input_473_cast_fp16 = reshape(shape = var_11859, x = hidden_states_293_cast_fp16)[name = tensor("input_473_cast_fp16")]; tensor hidden_states_295_pad_type_0 = const()[name = tensor("hidden_states_295_pad_type_0"), val = tensor("valid")]; tensor hidden_states_295_strides_0 = const()[name = tensor("hidden_states_295_strides_0"), val = tensor([1, 1])]; tensor hidden_states_295_pad_0 = const()[name = tensor("hidden_states_295_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_295_dilations_0 = const()[name = tensor("hidden_states_295_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_295_groups_0 = const()[name = tensor("hidden_states_295_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(807618368)))]; tensor up_blocks_3_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(807823232)))]; tensor hidden_states_295_cast_fp16 = conv(bias = up_blocks_3_attentions_0_proj_out_bias_to_fp16, dilations = hidden_states_295_dilations_0, groups = hidden_states_295_groups_0, pad = hidden_states_295_pad_0, pad_type = hidden_states_295_pad_type_0, strides = hidden_states_295_strides_0, weight = up_blocks_3_attentions_0_proj_out_weight_to_fp16, x = input_473_cast_fp16)[name = tensor("hidden_states_295_cast_fp16")]; tensor hidden_states_297_cast_fp16 = add(x = hidden_states_295_cast_fp16, y = hidden_states_283_cast_fp16)[name = tensor("hidden_states_297_cast_fp16")]; tensor input_475_interleave_0 = const()[name = tensor("input_475_interleave_0"), val = tensor(false)]; tensor cast_12 = cast(dtype = cast_12_dtype_0, x = input_35_cast_fp16)[name = tensor("cast_11")]; tensor input_475_cast_fp16 = concat(axis = var_10375, interleave = input_475_interleave_0, values = (hidden_states_297_cast_fp16, cast_12))[name = tensor("input_475_cast_fp16")]; tensor reshape_216_shape_0 = const()[name = tensor("reshape_216_shape_0"), val = tensor([2, 32, 20, 64, 64])]; tensor reshape_216_cast_fp16 = reshape(shape = reshape_216_shape_0, x = input_475_cast_fp16)[name = tensor("reshape_216_cast_fp16")]; tensor reduce_mean_162_axes_0 = const()[name = tensor("reduce_mean_162_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_162_keep_dims_0 = const()[name = tensor("reduce_mean_162_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_162_cast_fp16 = reduce_mean(axes = reduce_mean_162_axes_0, keep_dims = reduce_mean_162_keep_dims_0, x = reshape_216_cast_fp16)[name = tensor("reduce_mean_162_cast_fp16")]; tensor sub_108_cast_fp16 = sub(x = reshape_216_cast_fp16, y = reduce_mean_162_cast_fp16)[name = tensor("sub_108_cast_fp16")]; tensor square_54_cast_fp16 = square(x = sub_108_cast_fp16)[name = tensor("square_54_cast_fp16")]; tensor reduce_mean_164_axes_0 = const()[name = tensor("reduce_mean_164_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_164_keep_dims_0 = const()[name = tensor("reduce_mean_164_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_164_cast_fp16 = reduce_mean(axes = reduce_mean_164_axes_0, keep_dims = reduce_mean_164_keep_dims_0, x = square_54_cast_fp16)[name = tensor("reduce_mean_164_cast_fp16")]; tensor add_108_y_0_to_fp16 = const()[name = tensor("add_108_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_108_cast_fp16 = add(x = reduce_mean_164_cast_fp16, y = add_108_y_0_to_fp16)[name = tensor("add_108_cast_fp16")]; tensor sqrt_54_cast_fp16 = sqrt(x = add_108_cast_fp16)[name = tensor("sqrt_54_cast_fp16")]; tensor real_div_54_cast_fp16 = real_div(x = sub_108_cast_fp16, y = sqrt_54_cast_fp16)[name = tensor("real_div_54_cast_fp16")]; tensor reshape_217_shape_0 = const()[name = tensor("reshape_217_shape_0"), val = tensor([2, 640, 64, 64])]; tensor reshape_217_cast_fp16 = reshape(shape = reshape_217_shape_0, x = real_div_54_cast_fp16)[name = tensor("reshape_217_cast_fp16")]; tensor add_109_gamma_0_to_fp16 = const()[name = tensor("add_109_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(807823936)))]; tensor add_109_beta_0_to_fp16 = const()[name = tensor("add_109_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(807825280)))]; tensor add_109_epsilon_0_to_fp16 = const()[name = tensor("add_109_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_109_cast_fp16 = batch_norm(beta = add_109_beta_0_to_fp16, epsilon = add_109_epsilon_0_to_fp16, gamma = add_109_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_217_cast_fp16)[name = tensor("add_109_cast_fp16")]; tensor input_479_cast_fp16 = silu(x = add_109_cast_fp16)[name = tensor("input_479_cast_fp16")]; tensor hidden_states_299_pad_type_0 = const()[name = tensor("hidden_states_299_pad_type_0"), val = tensor("custom")]; tensor hidden_states_299_pad_0 = const()[name = tensor("hidden_states_299_pad_0"), val = tensor([1, 1, 1, 1])]; tensor hidden_states_299_strides_0 = const()[name = tensor("hidden_states_299_strides_0"), val = tensor([1, 1])]; tensor hidden_states_299_dilations_0 = const()[name = tensor("hidden_states_299_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_299_groups_0 = const()[name = tensor("hidden_states_299_groups_0"), val = tensor(1)]; tensor up_blocks_3_resnets_1_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_3_resnets_1_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(807826624)))]; tensor up_blocks_3_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(811513088)))]; tensor hidden_states_299_cast_fp16 = conv(bias = up_blocks_3_resnets_1_conv1_bias_to_fp16, dilations = hidden_states_299_dilations_0, groups = hidden_states_299_groups_0, pad = hidden_states_299_pad_0, pad_type = hidden_states_299_pad_type_0, strides = hidden_states_299_strides_0, weight = up_blocks_3_resnets_1_conv1_weight_to_fp16, x = input_479_cast_fp16)[name = tensor("hidden_states_299_cast_fp16")]; tensor temb_41_pad_type_0 = const()[name = tensor("temb_41_pad_type_0"), val = tensor("valid")]; tensor temb_41_strides_0 = const()[name = tensor("temb_41_strides_0"), val = tensor([1, 1])]; tensor temb_41_pad_0 = const()[name = tensor("temb_41_pad_0"), val = tensor([0, 0, 0, 0])]; tensor temb_41_dilations_0 = const()[name = tensor("temb_41_dilations_0"), val = tensor([1, 1])]; tensor temb_41_groups_0 = const()[name = tensor("temb_41_groups_0"), val = tensor(1)]; tensor up_blocks_3_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_3_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(811513792)))]; tensor up_blocks_3_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(812333056)))]; tensor temb_41_cast_fp16 = conv(bias = up_blocks_3_resnets_1_time_emb_proj_bias_to_fp16, dilations = temb_41_dilations_0, groups = temb_41_groups_0, pad = temb_41_pad_0, pad_type = temb_41_pad_type_0, strides = temb_41_strides_0, weight = up_blocks_3_resnets_1_time_emb_proj_weight_to_fp16, x = cast_9)[name = tensor("temb_41_cast_fp16")]; tensor input_483_cast_fp16 = add(x = hidden_states_299_cast_fp16, y = temb_41_cast_fp16)[name = tensor("input_483_cast_fp16")]; tensor reshape_220_shape_0 = const()[name = tensor("reshape_220_shape_0"), val = tensor([2, 32, 10, 64, 64])]; tensor reshape_220_cast_fp16 = reshape(shape = reshape_220_shape_0, x = input_483_cast_fp16)[name = tensor("reshape_220_cast_fp16")]; tensor reduce_mean_165_axes_0 = const()[name = tensor("reduce_mean_165_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_165_keep_dims_0 = const()[name = tensor("reduce_mean_165_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_165_cast_fp16 = reduce_mean(axes = reduce_mean_165_axes_0, keep_dims = reduce_mean_165_keep_dims_0, x = reshape_220_cast_fp16)[name = tensor("reduce_mean_165_cast_fp16")]; tensor sub_110_cast_fp16 = sub(x = reshape_220_cast_fp16, y = reduce_mean_165_cast_fp16)[name = tensor("sub_110_cast_fp16")]; tensor square_55_cast_fp16 = square(x = sub_110_cast_fp16)[name = tensor("square_55_cast_fp16")]; tensor reduce_mean_167_axes_0 = const()[name = tensor("reduce_mean_167_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_167_keep_dims_0 = const()[name = tensor("reduce_mean_167_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_167_cast_fp16 = reduce_mean(axes = reduce_mean_167_axes_0, keep_dims = reduce_mean_167_keep_dims_0, x = square_55_cast_fp16)[name = tensor("reduce_mean_167_cast_fp16")]; tensor add_110_y_0_to_fp16 = const()[name = tensor("add_110_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_110_cast_fp16 = add(x = reduce_mean_167_cast_fp16, y = add_110_y_0_to_fp16)[name = tensor("add_110_cast_fp16")]; tensor sqrt_55_cast_fp16 = sqrt(x = add_110_cast_fp16)[name = tensor("sqrt_55_cast_fp16")]; tensor real_div_55_cast_fp16 = real_div(x = sub_110_cast_fp16, y = sqrt_55_cast_fp16)[name = tensor("real_div_55_cast_fp16")]; tensor reshape_221_shape_0 = const()[name = tensor("reshape_221_shape_0"), val = tensor([2, 320, 64, 64])]; tensor reshape_221_cast_fp16 = reshape(shape = reshape_221_shape_0, x = real_div_55_cast_fp16)[name = tensor("reshape_221_cast_fp16")]; tensor add_111_gamma_0_to_fp16 = const()[name = tensor("add_111_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(812333760)))]; tensor add_111_beta_0_to_fp16 = const()[name = tensor("add_111_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(812334464)))]; tensor add_111_epsilon_0_to_fp16 = const()[name = tensor("add_111_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_111_cast_fp16 = batch_norm(beta = add_111_beta_0_to_fp16, epsilon = add_111_epsilon_0_to_fp16, gamma = add_111_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_221_cast_fp16)[name = tensor("add_111_cast_fp16")]; tensor input_487_cast_fp16 = silu(x = add_111_cast_fp16)[name = tensor("input_487_cast_fp16")]; tensor hidden_states_301_pad_type_0 = const()[name = tensor("hidden_states_301_pad_type_0"), val = tensor("custom")]; tensor hidden_states_301_pad_0 = const()[name = tensor("hidden_states_301_pad_0"), val = tensor([1, 1, 1, 1])]; tensor hidden_states_301_strides_0 = const()[name = tensor("hidden_states_301_strides_0"), val = tensor([1, 1])]; tensor hidden_states_301_dilations_0 = const()[name = tensor("hidden_states_301_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_301_groups_0 = const()[name = tensor("hidden_states_301_groups_0"), val = tensor(1)]; tensor up_blocks_3_resnets_1_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_3_resnets_1_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(812335168)))]; tensor up_blocks_3_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814178432)))]; tensor hidden_states_301_cast_fp16 = conv(bias = up_blocks_3_resnets_1_conv2_bias_to_fp16, dilations = hidden_states_301_dilations_0, groups = hidden_states_301_groups_0, pad = hidden_states_301_pad_0, pad_type = hidden_states_301_pad_type_0, strides = hidden_states_301_strides_0, weight = up_blocks_3_resnets_1_conv2_weight_to_fp16, x = input_487_cast_fp16)[name = tensor("hidden_states_301_cast_fp16")]; tensor x_25_pad_type_0 = const()[name = tensor("x_25_pad_type_0"), val = tensor("valid")]; tensor x_25_strides_0 = const()[name = tensor("x_25_strides_0"), val = tensor([1, 1])]; tensor x_25_pad_0 = const()[name = tensor("x_25_pad_0"), val = tensor([0, 0, 0, 0])]; tensor x_25_dilations_0 = const()[name = tensor("x_25_dilations_0"), val = tensor([1, 1])]; tensor x_25_groups_0 = const()[name = tensor("x_25_groups_0"), val = tensor(1)]; tensor up_blocks_3_resnets_1_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_3_resnets_1_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814179136)))]; tensor up_blocks_3_resnets_1_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_1_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814588800)))]; tensor x_25_cast_fp16 = conv(bias = up_blocks_3_resnets_1_conv_shortcut_bias_to_fp16, dilations = x_25_dilations_0, groups = x_25_groups_0, pad = x_25_pad_0, pad_type = x_25_pad_type_0, strides = x_25_strides_0, weight = up_blocks_3_resnets_1_conv_shortcut_weight_to_fp16, x = input_475_cast_fp16)[name = tensor("x_25_cast_fp16")]; tensor hidden_states_303_cast_fp16 = add(x = x_25_cast_fp16, y = hidden_states_301_cast_fp16)[name = tensor("hidden_states_303_cast_fp16")]; tensor reshape_224_shape_0 = const()[name = tensor("reshape_224_shape_0"), val = tensor([2, 32, 10, 64, 64])]; tensor reshape_224_cast_fp16 = reshape(shape = reshape_224_shape_0, x = hidden_states_303_cast_fp16)[name = tensor("reshape_224_cast_fp16")]; tensor reduce_mean_168_axes_0 = const()[name = tensor("reduce_mean_168_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_168_keep_dims_0 = const()[name = tensor("reduce_mean_168_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_168_cast_fp16 = reduce_mean(axes = reduce_mean_168_axes_0, keep_dims = reduce_mean_168_keep_dims_0, x = reshape_224_cast_fp16)[name = tensor("reduce_mean_168_cast_fp16")]; tensor sub_112_cast_fp16 = sub(x = reshape_224_cast_fp16, y = reduce_mean_168_cast_fp16)[name = tensor("sub_112_cast_fp16")]; tensor square_56_cast_fp16 = square(x = sub_112_cast_fp16)[name = tensor("square_56_cast_fp16")]; tensor reduce_mean_170_axes_0 = const()[name = tensor("reduce_mean_170_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_170_keep_dims_0 = const()[name = tensor("reduce_mean_170_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_170_cast_fp16 = reduce_mean(axes = reduce_mean_170_axes_0, keep_dims = reduce_mean_170_keep_dims_0, x = square_56_cast_fp16)[name = tensor("reduce_mean_170_cast_fp16")]; tensor add_112_y_0_to_fp16 = const()[name = tensor("add_112_y_0_to_fp16"), val = tensor(0x1.1p-20)]; tensor add_112_cast_fp16 = add(x = reduce_mean_170_cast_fp16, y = add_112_y_0_to_fp16)[name = tensor("add_112_cast_fp16")]; tensor sqrt_56_cast_fp16 = sqrt(x = add_112_cast_fp16)[name = tensor("sqrt_56_cast_fp16")]; tensor real_div_56_cast_fp16 = real_div(x = sub_112_cast_fp16, y = sqrt_56_cast_fp16)[name = tensor("real_div_56_cast_fp16")]; tensor reshape_225_shape_0 = const()[name = tensor("reshape_225_shape_0"), val = tensor([2, 320, 64, 64])]; tensor reshape_225_cast_fp16 = reshape(shape = reshape_225_shape_0, x = real_div_56_cast_fp16)[name = tensor("reshape_225_cast_fp16")]; tensor add_113_gamma_0_to_fp16 = const()[name = tensor("add_113_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814589504)))]; tensor add_113_beta_0_to_fp16 = const()[name = tensor("add_113_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814590208)))]; tensor add_113_epsilon_0_to_fp16 = const()[name = tensor("add_113_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_113_cast_fp16 = batch_norm(beta = add_113_beta_0_to_fp16, epsilon = add_113_epsilon_0_to_fp16, gamma = add_113_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_225_cast_fp16)[name = tensor("add_113_cast_fp16")]; tensor hidden_states_305_pad_type_0 = const()[name = tensor("hidden_states_305_pad_type_0"), val = tensor("valid")]; tensor hidden_states_305_strides_0 = const()[name = tensor("hidden_states_305_strides_0"), val = tensor([1, 1])]; tensor hidden_states_305_pad_0 = const()[name = tensor("hidden_states_305_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_305_dilations_0 = const()[name = tensor("hidden_states_305_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_305_groups_0 = const()[name = tensor("hidden_states_305_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_1_proj_in_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814590912)))]; tensor up_blocks_3_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814795776)))]; tensor hidden_states_305_cast_fp16 = conv(bias = up_blocks_3_attentions_1_proj_in_bias_to_fp16, dilations = hidden_states_305_dilations_0, groups = hidden_states_305_groups_0, pad = hidden_states_305_pad_0, pad_type = hidden_states_305_pad_type_0, strides = hidden_states_305_strides_0, weight = up_blocks_3_attentions_1_proj_in_weight_to_fp16, x = add_113_cast_fp16)[name = tensor("hidden_states_305_cast_fp16")]; tensor var_11939 = const()[name = tensor("op_11939"), val = tensor([2, 320, 1, 4096])]; tensor inputs_85_cast_fp16 = reshape(shape = var_11939, x = hidden_states_305_cast_fp16)[name = tensor("inputs_85_cast_fp16")]; tensor hidden_states_307_axes_0 = const()[name = tensor("hidden_states_307_axes_0"), val = tensor([1])]; tensor hidden_states_307_gamma_0_to_fp16 = const()[name = tensor("hidden_states_307_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814796480)))]; tensor hidden_states_307_beta_0_to_fp16 = const()[name = tensor("hidden_states_307_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814797184)))]; tensor var_11955_to_fp16 = const()[name = tensor("op_11955_to_fp16"), val = tensor(0x1.5p-17)]; tensor hidden_states_307_cast_fp16 = layer_norm(axes = hidden_states_307_axes_0, beta = hidden_states_307_beta_0_to_fp16, epsilon = var_11955_to_fp16, gamma = hidden_states_307_gamma_0_to_fp16, x = inputs_85_cast_fp16)[name = tensor("hidden_states_307_cast_fp16")]; tensor q_57_pad_type_0 = const()[name = tensor("q_57_pad_type_0"), val = tensor("valid")]; tensor q_57_strides_0 = const()[name = tensor("q_57_strides_0"), val = tensor([1, 1])]; tensor q_57_pad_0 = const()[name = tensor("q_57_pad_0"), val = tensor([0, 0, 0, 0])]; tensor q_57_dilations_0 = const()[name = tensor("q_57_dilations_0"), val = tensor([1, 1])]; tensor q_57_groups_0 = const()[name = tensor("q_57_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814797888)))]; tensor q_57_cast_fp16 = conv(dilations = q_57_dilations_0, groups = q_57_groups_0, pad = q_57_pad_0, pad_type = q_57_pad_type_0, strides = q_57_strides_0, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_307_cast_fp16)[name = tensor("q_57_cast_fp16")]; tensor k_113_pad_type_0 = const()[name = tensor("k_113_pad_type_0"), val = tensor("valid")]; tensor k_113_strides_0 = const()[name = tensor("k_113_strides_0"), val = tensor([1, 1])]; tensor k_113_pad_0 = const()[name = tensor("k_113_pad_0"), val = tensor([0, 0, 0, 0])]; tensor k_113_dilations_0 = const()[name = tensor("k_113_dilations_0"), val = tensor([1, 1])]; tensor k_113_groups_0 = const()[name = tensor("k_113_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(815002752)))]; tensor k_113_cast_fp16 = conv(dilations = k_113_dilations_0, groups = k_113_groups_0, pad = k_113_pad_0, pad_type = k_113_pad_type_0, strides = k_113_strides_0, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_307_cast_fp16)[name = tensor("k_113_cast_fp16")]; tensor v_57_pad_type_0 = const()[name = tensor("v_57_pad_type_0"), val = tensor("valid")]; tensor v_57_strides_0 = const()[name = tensor("v_57_strides_0"), val = tensor([1, 1])]; tensor v_57_pad_0 = const()[name = tensor("v_57_pad_0"), val = tensor([0, 0, 0, 0])]; tensor v_57_dilations_0 = const()[name = tensor("v_57_dilations_0"), val = tensor([1, 1])]; tensor v_57_groups_0 = const()[name = tensor("v_57_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(815207616)))]; tensor v_57_cast_fp16 = conv(dilations = v_57_dilations_0, groups = v_57_groups_0, pad = v_57_pad_0, pad_type = v_57_pad_type_0, strides = v_57_strides_0, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_307_cast_fp16)[name = tensor("v_57_cast_fp16")]; tensor var_11988_begin_0 = const()[name = tensor("op_11988_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11988_end_0 = const()[name = tensor("op_11988_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_11988_end_mask_0 = const()[name = tensor("op_11988_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11988_cast_fp16 = slice_by_index(begin = var_11988_begin_0, end = var_11988_end_0, end_mask = var_11988_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_11988_cast_fp16")]; tensor var_11992_begin_0 = const()[name = tensor("op_11992_begin_0"), val = tensor([0, 40, 0, 0])]; tensor var_11992_end_0 = const()[name = tensor("op_11992_end_0"), val = tensor([2, 80, 1, 4096])]; tensor var_11992_end_mask_0 = const()[name = tensor("op_11992_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11992_cast_fp16 = slice_by_index(begin = var_11992_begin_0, end = var_11992_end_0, end_mask = var_11992_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_11992_cast_fp16")]; tensor var_11996_begin_0 = const()[name = tensor("op_11996_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_11996_end_0 = const()[name = tensor("op_11996_end_0"), val = tensor([2, 120, 1, 4096])]; tensor var_11996_end_mask_0 = const()[name = tensor("op_11996_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11996_cast_fp16 = slice_by_index(begin = var_11996_begin_0, end = var_11996_end_0, end_mask = var_11996_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_11996_cast_fp16")]; tensor var_12000_begin_0 = const()[name = tensor("op_12000_begin_0"), val = tensor([0, 120, 0, 0])]; tensor var_12000_end_0 = const()[name = tensor("op_12000_end_0"), val = tensor([2, 160, 1, 4096])]; tensor var_12000_end_mask_0 = const()[name = tensor("op_12000_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_12000_cast_fp16 = slice_by_index(begin = var_12000_begin_0, end = var_12000_end_0, end_mask = var_12000_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12000_cast_fp16")]; tensor var_12004_begin_0 = const()[name = tensor("op_12004_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_12004_end_0 = const()[name = tensor("op_12004_end_0"), val = tensor([2, 200, 1, 4096])]; tensor var_12004_end_mask_0 = const()[name = tensor("op_12004_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_12004_cast_fp16 = slice_by_index(begin = var_12004_begin_0, end = var_12004_end_0, end_mask = var_12004_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12004_cast_fp16")]; tensor var_12008_begin_0 = const()[name = tensor("op_12008_begin_0"), val = tensor([0, 200, 0, 0])]; tensor var_12008_end_0 = const()[name = tensor("op_12008_end_0"), val = tensor([2, 240, 1, 4096])]; tensor var_12008_end_mask_0 = const()[name = tensor("op_12008_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_12008_cast_fp16 = slice_by_index(begin = var_12008_begin_0, end = var_12008_end_0, end_mask = var_12008_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12008_cast_fp16")]; tensor var_12012_begin_0 = const()[name = tensor("op_12012_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_12012_end_0 = const()[name = tensor("op_12012_end_0"), val = tensor([2, 280, 1, 4096])]; tensor var_12012_end_mask_0 = const()[name = tensor("op_12012_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_12012_cast_fp16 = slice_by_index(begin = var_12012_begin_0, end = var_12012_end_0, end_mask = var_12012_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12012_cast_fp16")]; tensor var_12016_begin_0 = const()[name = tensor("op_12016_begin_0"), val = tensor([0, 280, 0, 0])]; tensor var_12016_end_0 = const()[name = tensor("op_12016_end_0"), val = tensor([2, 1, 1, 4096])]; tensor var_12016_end_mask_0 = const()[name = tensor("op_12016_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_12016_cast_fp16 = slice_by_index(begin = var_12016_begin_0, end = var_12016_end_0, end_mask = var_12016_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12016_cast_fp16")]; tensor var_12019_begin_0 = const()[name = tensor("op_12019_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_12019_end_0 = const()[name = tensor("op_12019_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_12019_end_mask_0 = const()[name = tensor("op_12019_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12019_cast_fp16 = slice_by_index(begin = var_12019_begin_0, end = var_12019_end_0, end_mask = var_12019_end_mask_0, x = var_11988_cast_fp16)[name = tensor("op_12019_cast_fp16")]; tensor var_12020_begin_0 = const()[name = tensor("op_12020_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_12020_end_0 = const()[name = tensor("op_12020_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_12020_end_mask_0 = const()[name = tensor("op_12020_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12020_cast_fp16 = slice_by_index(begin = var_12020_begin_0, end = var_12020_end_0, end_mask = var_12020_end_mask_0, x = var_11988_cast_fp16)[name = tensor("op_12020_cast_fp16")]; tensor var_12021_begin_0 = const()[name = tensor("op_12021_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_12021_end_0 = const()[name = tensor("op_12021_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_12021_end_mask_0 = const()[name = tensor("op_12021_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12021_cast_fp16 = slice_by_index(begin = var_12021_begin_0, end = var_12021_end_0, end_mask = var_12021_end_mask_0, x = var_11988_cast_fp16)[name = tensor("op_12021_cast_fp16")]; tensor var_12022_begin_0 = const()[name = tensor("op_12022_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_12022_end_0 = const()[name = tensor("op_12022_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_12022_end_mask_0 = const()[name = tensor("op_12022_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12022_cast_fp16 = slice_by_index(begin = var_12022_begin_0, end = var_12022_end_0, end_mask = var_12022_end_mask_0, x = var_11988_cast_fp16)[name = tensor("op_12022_cast_fp16")]; tensor var_12023_begin_0 = const()[name = tensor("op_12023_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_12023_end_0 = const()[name = tensor("op_12023_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_12023_end_mask_0 = const()[name = tensor("op_12023_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12023_cast_fp16 = slice_by_index(begin = var_12023_begin_0, end = var_12023_end_0, end_mask = var_12023_end_mask_0, x = var_11988_cast_fp16)[name = tensor("op_12023_cast_fp16")]; tensor var_12024_begin_0 = const()[name = tensor("op_12024_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_12024_end_0 = const()[name = tensor("op_12024_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_12024_end_mask_0 = const()[name = tensor("op_12024_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12024_cast_fp16 = slice_by_index(begin = var_12024_begin_0, end = var_12024_end_0, end_mask = var_12024_end_mask_0, x = var_11988_cast_fp16)[name = tensor("op_12024_cast_fp16")]; tensor var_12025_begin_0 = const()[name = tensor("op_12025_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_12025_end_0 = const()[name = tensor("op_12025_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_12025_end_mask_0 = const()[name = tensor("op_12025_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12025_cast_fp16 = slice_by_index(begin = var_12025_begin_0, end = var_12025_end_0, end_mask = var_12025_end_mask_0, x = var_11988_cast_fp16)[name = tensor("op_12025_cast_fp16")]; tensor var_12026_begin_0 = const()[name = tensor("op_12026_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_12026_end_0 = const()[name = tensor("op_12026_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_12026_end_mask_0 = const()[name = tensor("op_12026_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_12026_cast_fp16 = slice_by_index(begin = var_12026_begin_0, end = var_12026_end_0, end_mask = var_12026_end_mask_0, x = var_11988_cast_fp16)[name = tensor("op_12026_cast_fp16")]; tensor var_12027_begin_0 = const()[name = tensor("op_12027_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_12027_end_0 = const()[name = tensor("op_12027_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_12027_end_mask_0 = const()[name = tensor("op_12027_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12027_cast_fp16 = slice_by_index(begin = var_12027_begin_0, end = var_12027_end_0, end_mask = var_12027_end_mask_0, x = var_11992_cast_fp16)[name = tensor("op_12027_cast_fp16")]; tensor var_12028_begin_0 = const()[name = tensor("op_12028_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_12028_end_0 = const()[name = tensor("op_12028_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_12028_end_mask_0 = const()[name = tensor("op_12028_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12028_cast_fp16 = slice_by_index(begin = var_12028_begin_0, end = var_12028_end_0, end_mask = var_12028_end_mask_0, x = var_11992_cast_fp16)[name = tensor("op_12028_cast_fp16")]; tensor var_12029_begin_0 = const()[name = tensor("op_12029_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_12029_end_0 = const()[name = tensor("op_12029_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_12029_end_mask_0 = const()[name = tensor("op_12029_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12029_cast_fp16 = slice_by_index(begin = var_12029_begin_0, end = var_12029_end_0, end_mask = var_12029_end_mask_0, x = var_11992_cast_fp16)[name = tensor("op_12029_cast_fp16")]; tensor var_12030_begin_0 = const()[name = tensor("op_12030_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_12030_end_0 = const()[name = tensor("op_12030_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_12030_end_mask_0 = const()[name = tensor("op_12030_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12030_cast_fp16 = slice_by_index(begin = var_12030_begin_0, end = var_12030_end_0, end_mask = var_12030_end_mask_0, x = var_11992_cast_fp16)[name = tensor("op_12030_cast_fp16")]; tensor var_12031_begin_0 = const()[name = tensor("op_12031_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_12031_end_0 = const()[name = tensor("op_12031_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_12031_end_mask_0 = const()[name = tensor("op_12031_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12031_cast_fp16 = slice_by_index(begin = var_12031_begin_0, end = var_12031_end_0, end_mask = var_12031_end_mask_0, x = var_11992_cast_fp16)[name = tensor("op_12031_cast_fp16")]; tensor var_12032_begin_0 = const()[name = tensor("op_12032_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_12032_end_0 = const()[name = tensor("op_12032_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_12032_end_mask_0 = const()[name = tensor("op_12032_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12032_cast_fp16 = slice_by_index(begin = var_12032_begin_0, end = var_12032_end_0, end_mask = var_12032_end_mask_0, x = var_11992_cast_fp16)[name = tensor("op_12032_cast_fp16")]; tensor var_12033_begin_0 = const()[name = tensor("op_12033_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_12033_end_0 = const()[name = tensor("op_12033_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_12033_end_mask_0 = const()[name = tensor("op_12033_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12033_cast_fp16 = slice_by_index(begin = var_12033_begin_0, end = var_12033_end_0, end_mask = var_12033_end_mask_0, x = var_11992_cast_fp16)[name = tensor("op_12033_cast_fp16")]; tensor var_12034_begin_0 = const()[name = tensor("op_12034_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_12034_end_0 = const()[name = tensor("op_12034_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_12034_end_mask_0 = const()[name = tensor("op_12034_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_12034_cast_fp16 = slice_by_index(begin = var_12034_begin_0, end = var_12034_end_0, end_mask = var_12034_end_mask_0, x = var_11992_cast_fp16)[name = tensor("op_12034_cast_fp16")]; tensor var_12035_begin_0 = const()[name = tensor("op_12035_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_12035_end_0 = const()[name = tensor("op_12035_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_12035_end_mask_0 = const()[name = tensor("op_12035_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12035_cast_fp16 = slice_by_index(begin = var_12035_begin_0, end = var_12035_end_0, end_mask = var_12035_end_mask_0, x = var_11996_cast_fp16)[name = tensor("op_12035_cast_fp16")]; tensor var_12036_begin_0 = const()[name = tensor("op_12036_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_12036_end_0 = const()[name = tensor("op_12036_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_12036_end_mask_0 = const()[name = tensor("op_12036_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12036_cast_fp16 = slice_by_index(begin = var_12036_begin_0, end = var_12036_end_0, end_mask = var_12036_end_mask_0, x = var_11996_cast_fp16)[name = tensor("op_12036_cast_fp16")]; tensor var_12037_begin_0 = const()[name = tensor("op_12037_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_12037_end_0 = const()[name = tensor("op_12037_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_12037_end_mask_0 = const()[name = tensor("op_12037_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12037_cast_fp16 = slice_by_index(begin = var_12037_begin_0, end = var_12037_end_0, end_mask = var_12037_end_mask_0, x = var_11996_cast_fp16)[name = tensor("op_12037_cast_fp16")]; tensor var_12038_begin_0 = const()[name = tensor("op_12038_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_12038_end_0 = const()[name = tensor("op_12038_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_12038_end_mask_0 = const()[name = tensor("op_12038_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12038_cast_fp16 = slice_by_index(begin = var_12038_begin_0, end = var_12038_end_0, end_mask = var_12038_end_mask_0, x = var_11996_cast_fp16)[name = tensor("op_12038_cast_fp16")]; tensor var_12039_begin_0 = const()[name = tensor("op_12039_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_12039_end_0 = const()[name = tensor("op_12039_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_12039_end_mask_0 = const()[name = tensor("op_12039_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12039_cast_fp16 = slice_by_index(begin = var_12039_begin_0, end = var_12039_end_0, end_mask = var_12039_end_mask_0, x = var_11996_cast_fp16)[name = tensor("op_12039_cast_fp16")]; tensor var_12040_begin_0 = const()[name = tensor("op_12040_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_12040_end_0 = const()[name = tensor("op_12040_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_12040_end_mask_0 = const()[name = tensor("op_12040_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12040_cast_fp16 = slice_by_index(begin = var_12040_begin_0, end = var_12040_end_0, end_mask = var_12040_end_mask_0, x = var_11996_cast_fp16)[name = tensor("op_12040_cast_fp16")]; tensor var_12041_begin_0 = const()[name = tensor("op_12041_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_12041_end_0 = const()[name = tensor("op_12041_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_12041_end_mask_0 = const()[name = tensor("op_12041_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12041_cast_fp16 = slice_by_index(begin = var_12041_begin_0, end = var_12041_end_0, end_mask = var_12041_end_mask_0, x = var_11996_cast_fp16)[name = tensor("op_12041_cast_fp16")]; tensor var_12042_begin_0 = const()[name = tensor("op_12042_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_12042_end_0 = const()[name = tensor("op_12042_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_12042_end_mask_0 = const()[name = tensor("op_12042_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_12042_cast_fp16 = slice_by_index(begin = var_12042_begin_0, end = var_12042_end_0, end_mask = var_12042_end_mask_0, x = var_11996_cast_fp16)[name = tensor("op_12042_cast_fp16")]; tensor var_12043_begin_0 = const()[name = tensor("op_12043_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_12043_end_0 = const()[name = tensor("op_12043_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_12043_end_mask_0 = const()[name = tensor("op_12043_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12043_cast_fp16 = slice_by_index(begin = var_12043_begin_0, end = var_12043_end_0, end_mask = var_12043_end_mask_0, x = var_12000_cast_fp16)[name = tensor("op_12043_cast_fp16")]; tensor var_12044_begin_0 = const()[name = tensor("op_12044_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_12044_end_0 = const()[name = tensor("op_12044_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_12044_end_mask_0 = const()[name = tensor("op_12044_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12044_cast_fp16 = slice_by_index(begin = var_12044_begin_0, end = var_12044_end_0, end_mask = var_12044_end_mask_0, x = var_12000_cast_fp16)[name = tensor("op_12044_cast_fp16")]; tensor var_12045_begin_0 = const()[name = tensor("op_12045_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_12045_end_0 = const()[name = tensor("op_12045_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_12045_end_mask_0 = const()[name = tensor("op_12045_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12045_cast_fp16 = slice_by_index(begin = var_12045_begin_0, end = var_12045_end_0, end_mask = var_12045_end_mask_0, x = var_12000_cast_fp16)[name = tensor("op_12045_cast_fp16")]; tensor var_12046_begin_0 = const()[name = tensor("op_12046_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_12046_end_0 = const()[name = tensor("op_12046_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_12046_end_mask_0 = const()[name = tensor("op_12046_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12046_cast_fp16 = slice_by_index(begin = var_12046_begin_0, end = var_12046_end_0, end_mask = var_12046_end_mask_0, x = var_12000_cast_fp16)[name = tensor("op_12046_cast_fp16")]; tensor var_12047_begin_0 = const()[name = tensor("op_12047_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_12047_end_0 = const()[name = tensor("op_12047_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_12047_end_mask_0 = const()[name = tensor("op_12047_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12047_cast_fp16 = slice_by_index(begin = var_12047_begin_0, end = var_12047_end_0, end_mask = var_12047_end_mask_0, x = var_12000_cast_fp16)[name = tensor("op_12047_cast_fp16")]; tensor var_12048_begin_0 = const()[name = tensor("op_12048_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_12048_end_0 = const()[name = tensor("op_12048_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_12048_end_mask_0 = const()[name = tensor("op_12048_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12048_cast_fp16 = slice_by_index(begin = var_12048_begin_0, end = var_12048_end_0, end_mask = var_12048_end_mask_0, x = var_12000_cast_fp16)[name = tensor("op_12048_cast_fp16")]; tensor var_12049_begin_0 = const()[name = tensor("op_12049_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_12049_end_0 = const()[name = tensor("op_12049_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_12049_end_mask_0 = const()[name = tensor("op_12049_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12049_cast_fp16 = slice_by_index(begin = var_12049_begin_0, end = var_12049_end_0, end_mask = var_12049_end_mask_0, x = var_12000_cast_fp16)[name = tensor("op_12049_cast_fp16")]; tensor var_12050_begin_0 = const()[name = tensor("op_12050_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_12050_end_0 = const()[name = tensor("op_12050_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_12050_end_mask_0 = const()[name = tensor("op_12050_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_12050_cast_fp16 = slice_by_index(begin = var_12050_begin_0, end = var_12050_end_0, end_mask = var_12050_end_mask_0, x = var_12000_cast_fp16)[name = tensor("op_12050_cast_fp16")]; tensor var_12051_begin_0 = const()[name = tensor("op_12051_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_12051_end_0 = const()[name = tensor("op_12051_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_12051_end_mask_0 = const()[name = tensor("op_12051_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12051_cast_fp16 = slice_by_index(begin = var_12051_begin_0, end = var_12051_end_0, end_mask = var_12051_end_mask_0, x = var_12004_cast_fp16)[name = tensor("op_12051_cast_fp16")]; tensor var_12052_begin_0 = const()[name = tensor("op_12052_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_12052_end_0 = const()[name = tensor("op_12052_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_12052_end_mask_0 = const()[name = tensor("op_12052_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12052_cast_fp16 = slice_by_index(begin = var_12052_begin_0, end = var_12052_end_0, end_mask = var_12052_end_mask_0, x = var_12004_cast_fp16)[name = tensor("op_12052_cast_fp16")]; tensor var_12053_begin_0 = const()[name = tensor("op_12053_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_12053_end_0 = const()[name = tensor("op_12053_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_12053_end_mask_0 = const()[name = tensor("op_12053_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12053_cast_fp16 = slice_by_index(begin = var_12053_begin_0, end = var_12053_end_0, end_mask = var_12053_end_mask_0, x = var_12004_cast_fp16)[name = tensor("op_12053_cast_fp16")]; tensor var_12054_begin_0 = const()[name = tensor("op_12054_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_12054_end_0 = const()[name = tensor("op_12054_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_12054_end_mask_0 = const()[name = tensor("op_12054_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12054_cast_fp16 = slice_by_index(begin = var_12054_begin_0, end = var_12054_end_0, end_mask = var_12054_end_mask_0, x = var_12004_cast_fp16)[name = tensor("op_12054_cast_fp16")]; tensor var_12055_begin_0 = const()[name = tensor("op_12055_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_12055_end_0 = const()[name = tensor("op_12055_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_12055_end_mask_0 = const()[name = tensor("op_12055_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12055_cast_fp16 = slice_by_index(begin = var_12055_begin_0, end = var_12055_end_0, end_mask = var_12055_end_mask_0, x = var_12004_cast_fp16)[name = tensor("op_12055_cast_fp16")]; tensor var_12056_begin_0 = const()[name = tensor("op_12056_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_12056_end_0 = const()[name = tensor("op_12056_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_12056_end_mask_0 = const()[name = tensor("op_12056_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12056_cast_fp16 = slice_by_index(begin = var_12056_begin_0, end = var_12056_end_0, end_mask = var_12056_end_mask_0, x = var_12004_cast_fp16)[name = tensor("op_12056_cast_fp16")]; tensor var_12057_begin_0 = const()[name = tensor("op_12057_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_12057_end_0 = const()[name = tensor("op_12057_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_12057_end_mask_0 = const()[name = tensor("op_12057_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12057_cast_fp16 = slice_by_index(begin = var_12057_begin_0, end = var_12057_end_0, end_mask = var_12057_end_mask_0, x = var_12004_cast_fp16)[name = tensor("op_12057_cast_fp16")]; tensor var_12058_begin_0 = const()[name = tensor("op_12058_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_12058_end_0 = const()[name = tensor("op_12058_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_12058_end_mask_0 = const()[name = tensor("op_12058_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_12058_cast_fp16 = slice_by_index(begin = var_12058_begin_0, end = var_12058_end_0, end_mask = var_12058_end_mask_0, x = var_12004_cast_fp16)[name = tensor("op_12058_cast_fp16")]; tensor var_12059_begin_0 = const()[name = tensor("op_12059_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_12059_end_0 = const()[name = tensor("op_12059_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_12059_end_mask_0 = const()[name = tensor("op_12059_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12059_cast_fp16 = slice_by_index(begin = var_12059_begin_0, end = var_12059_end_0, end_mask = var_12059_end_mask_0, x = var_12008_cast_fp16)[name = tensor("op_12059_cast_fp16")]; tensor var_12060_begin_0 = const()[name = tensor("op_12060_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_12060_end_0 = const()[name = tensor("op_12060_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_12060_end_mask_0 = const()[name = tensor("op_12060_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12060_cast_fp16 = slice_by_index(begin = var_12060_begin_0, end = var_12060_end_0, end_mask = var_12060_end_mask_0, x = var_12008_cast_fp16)[name = tensor("op_12060_cast_fp16")]; tensor var_12061_begin_0 = const()[name = tensor("op_12061_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_12061_end_0 = const()[name = tensor("op_12061_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_12061_end_mask_0 = const()[name = tensor("op_12061_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12061_cast_fp16 = slice_by_index(begin = var_12061_begin_0, end = var_12061_end_0, end_mask = var_12061_end_mask_0, x = var_12008_cast_fp16)[name = tensor("op_12061_cast_fp16")]; tensor var_12062_begin_0 = const()[name = tensor("op_12062_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_12062_end_0 = const()[name = tensor("op_12062_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_12062_end_mask_0 = const()[name = tensor("op_12062_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12062_cast_fp16 = slice_by_index(begin = var_12062_begin_0, end = var_12062_end_0, end_mask = var_12062_end_mask_0, x = var_12008_cast_fp16)[name = tensor("op_12062_cast_fp16")]; tensor var_12063_begin_0 = const()[name = tensor("op_12063_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_12063_end_0 = const()[name = tensor("op_12063_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_12063_end_mask_0 = const()[name = tensor("op_12063_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12063_cast_fp16 = slice_by_index(begin = var_12063_begin_0, end = var_12063_end_0, end_mask = var_12063_end_mask_0, x = var_12008_cast_fp16)[name = tensor("op_12063_cast_fp16")]; tensor var_12064_begin_0 = const()[name = tensor("op_12064_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_12064_end_0 = const()[name = tensor("op_12064_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_12064_end_mask_0 = const()[name = tensor("op_12064_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12064_cast_fp16 = slice_by_index(begin = var_12064_begin_0, end = var_12064_end_0, end_mask = var_12064_end_mask_0, x = var_12008_cast_fp16)[name = tensor("op_12064_cast_fp16")]; tensor var_12065_begin_0 = const()[name = tensor("op_12065_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_12065_end_0 = const()[name = tensor("op_12065_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_12065_end_mask_0 = const()[name = tensor("op_12065_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12065_cast_fp16 = slice_by_index(begin = var_12065_begin_0, end = var_12065_end_0, end_mask = var_12065_end_mask_0, x = var_12008_cast_fp16)[name = tensor("op_12065_cast_fp16")]; tensor var_12066_begin_0 = const()[name = tensor("op_12066_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_12066_end_0 = const()[name = tensor("op_12066_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_12066_end_mask_0 = const()[name = tensor("op_12066_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_12066_cast_fp16 = slice_by_index(begin = var_12066_begin_0, end = var_12066_end_0, end_mask = var_12066_end_mask_0, x = var_12008_cast_fp16)[name = tensor("op_12066_cast_fp16")]; tensor var_12067_begin_0 = const()[name = tensor("op_12067_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_12067_end_0 = const()[name = tensor("op_12067_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_12067_end_mask_0 = const()[name = tensor("op_12067_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12067_cast_fp16 = slice_by_index(begin = var_12067_begin_0, end = var_12067_end_0, end_mask = var_12067_end_mask_0, x = var_12012_cast_fp16)[name = tensor("op_12067_cast_fp16")]; tensor var_12068_begin_0 = const()[name = tensor("op_12068_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_12068_end_0 = const()[name = tensor("op_12068_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_12068_end_mask_0 = const()[name = tensor("op_12068_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12068_cast_fp16 = slice_by_index(begin = var_12068_begin_0, end = var_12068_end_0, end_mask = var_12068_end_mask_0, x = var_12012_cast_fp16)[name = tensor("op_12068_cast_fp16")]; tensor var_12069_begin_0 = const()[name = tensor("op_12069_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_12069_end_0 = const()[name = tensor("op_12069_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_12069_end_mask_0 = const()[name = tensor("op_12069_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12069_cast_fp16 = slice_by_index(begin = var_12069_begin_0, end = var_12069_end_0, end_mask = var_12069_end_mask_0, x = var_12012_cast_fp16)[name = tensor("op_12069_cast_fp16")]; tensor var_12070_begin_0 = const()[name = tensor("op_12070_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_12070_end_0 = const()[name = tensor("op_12070_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_12070_end_mask_0 = const()[name = tensor("op_12070_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12070_cast_fp16 = slice_by_index(begin = var_12070_begin_0, end = var_12070_end_0, end_mask = var_12070_end_mask_0, x = var_12012_cast_fp16)[name = tensor("op_12070_cast_fp16")]; tensor var_12071_begin_0 = const()[name = tensor("op_12071_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_12071_end_0 = const()[name = tensor("op_12071_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_12071_end_mask_0 = const()[name = tensor("op_12071_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12071_cast_fp16 = slice_by_index(begin = var_12071_begin_0, end = var_12071_end_0, end_mask = var_12071_end_mask_0, x = var_12012_cast_fp16)[name = tensor("op_12071_cast_fp16")]; tensor var_12072_begin_0 = const()[name = tensor("op_12072_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_12072_end_0 = const()[name = tensor("op_12072_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_12072_end_mask_0 = const()[name = tensor("op_12072_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12072_cast_fp16 = slice_by_index(begin = var_12072_begin_0, end = var_12072_end_0, end_mask = var_12072_end_mask_0, x = var_12012_cast_fp16)[name = tensor("op_12072_cast_fp16")]; tensor var_12073_begin_0 = const()[name = tensor("op_12073_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_12073_end_0 = const()[name = tensor("op_12073_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_12073_end_mask_0 = const()[name = tensor("op_12073_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12073_cast_fp16 = slice_by_index(begin = var_12073_begin_0, end = var_12073_end_0, end_mask = var_12073_end_mask_0, x = var_12012_cast_fp16)[name = tensor("op_12073_cast_fp16")]; tensor var_12074_begin_0 = const()[name = tensor("op_12074_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_12074_end_0 = const()[name = tensor("op_12074_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_12074_end_mask_0 = const()[name = tensor("op_12074_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_12074_cast_fp16 = slice_by_index(begin = var_12074_begin_0, end = var_12074_end_0, end_mask = var_12074_end_mask_0, x = var_12012_cast_fp16)[name = tensor("op_12074_cast_fp16")]; tensor var_12075_begin_0 = const()[name = tensor("op_12075_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_12075_end_0 = const()[name = tensor("op_12075_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_12075_end_mask_0 = const()[name = tensor("op_12075_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12075_cast_fp16 = slice_by_index(begin = var_12075_begin_0, end = var_12075_end_0, end_mask = var_12075_end_mask_0, x = var_12016_cast_fp16)[name = tensor("op_12075_cast_fp16")]; tensor var_12076_begin_0 = const()[name = tensor("op_12076_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_12076_end_0 = const()[name = tensor("op_12076_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_12076_end_mask_0 = const()[name = tensor("op_12076_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12076_cast_fp16 = slice_by_index(begin = var_12076_begin_0, end = var_12076_end_0, end_mask = var_12076_end_mask_0, x = var_12016_cast_fp16)[name = tensor("op_12076_cast_fp16")]; tensor var_12077_begin_0 = const()[name = tensor("op_12077_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_12077_end_0 = const()[name = tensor("op_12077_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_12077_end_mask_0 = const()[name = tensor("op_12077_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12077_cast_fp16 = slice_by_index(begin = var_12077_begin_0, end = var_12077_end_0, end_mask = var_12077_end_mask_0, x = var_12016_cast_fp16)[name = tensor("op_12077_cast_fp16")]; tensor var_12078_begin_0 = const()[name = tensor("op_12078_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_12078_end_0 = const()[name = tensor("op_12078_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_12078_end_mask_0 = const()[name = tensor("op_12078_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12078_cast_fp16 = slice_by_index(begin = var_12078_begin_0, end = var_12078_end_0, end_mask = var_12078_end_mask_0, x = var_12016_cast_fp16)[name = tensor("op_12078_cast_fp16")]; tensor var_12079_begin_0 = const()[name = tensor("op_12079_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_12079_end_0 = const()[name = tensor("op_12079_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_12079_end_mask_0 = const()[name = tensor("op_12079_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12079_cast_fp16 = slice_by_index(begin = var_12079_begin_0, end = var_12079_end_0, end_mask = var_12079_end_mask_0, x = var_12016_cast_fp16)[name = tensor("op_12079_cast_fp16")]; tensor var_12080_begin_0 = const()[name = tensor("op_12080_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_12080_end_0 = const()[name = tensor("op_12080_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_12080_end_mask_0 = const()[name = tensor("op_12080_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12080_cast_fp16 = slice_by_index(begin = var_12080_begin_0, end = var_12080_end_0, end_mask = var_12080_end_mask_0, x = var_12016_cast_fp16)[name = tensor("op_12080_cast_fp16")]; tensor var_12081_begin_0 = const()[name = tensor("op_12081_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_12081_end_0 = const()[name = tensor("op_12081_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_12081_end_mask_0 = const()[name = tensor("op_12081_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12081_cast_fp16 = slice_by_index(begin = var_12081_begin_0, end = var_12081_end_0, end_mask = var_12081_end_mask_0, x = var_12016_cast_fp16)[name = tensor("op_12081_cast_fp16")]; tensor var_12082_begin_0 = const()[name = tensor("op_12082_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_12082_end_0 = const()[name = tensor("op_12082_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_12082_end_mask_0 = const()[name = tensor("op_12082_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_12082_cast_fp16 = slice_by_index(begin = var_12082_begin_0, end = var_12082_end_0, end_mask = var_12082_end_mask_0, x = var_12016_cast_fp16)[name = tensor("op_12082_cast_fp16")]; tensor k_115_perm_0 = const()[name = tensor("k_115_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_12087_begin_0 = const()[name = tensor("op_12087_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_12087_end_0 = const()[name = tensor("op_12087_end_0"), val = tensor([2, 4096, 1, 40])]; tensor var_12087_end_mask_0 = const()[name = tensor("op_12087_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_115_cast_fp16 = transpose(perm = k_115_perm_0, x = k_113_cast_fp16)[name = tensor("transpose_3")]; tensor var_12087_cast_fp16 = slice_by_index(begin = var_12087_begin_0, end = var_12087_end_0, end_mask = var_12087_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12087_cast_fp16")]; tensor var_12091_begin_0 = const()[name = tensor("op_12091_begin_0"), val = tensor([0, 0, 0, 40])]; tensor var_12091_end_0 = const()[name = tensor("op_12091_end_0"), val = tensor([2, 4096, 1, 80])]; tensor var_12091_end_mask_0 = const()[name = tensor("op_12091_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12091_cast_fp16 = slice_by_index(begin = var_12091_begin_0, end = var_12091_end_0, end_mask = var_12091_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12091_cast_fp16")]; tensor var_12095_begin_0 = const()[name = tensor("op_12095_begin_0"), val = tensor([0, 0, 0, 80])]; tensor var_12095_end_0 = const()[name = tensor("op_12095_end_0"), val = tensor([2, 4096, 1, 120])]; tensor var_12095_end_mask_0 = const()[name = tensor("op_12095_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12095_cast_fp16 = slice_by_index(begin = var_12095_begin_0, end = var_12095_end_0, end_mask = var_12095_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12095_cast_fp16")]; tensor var_12099_begin_0 = const()[name = tensor("op_12099_begin_0"), val = tensor([0, 0, 0, 120])]; tensor var_12099_end_0 = const()[name = tensor("op_12099_end_0"), val = tensor([2, 4096, 1, 160])]; tensor var_12099_end_mask_0 = const()[name = tensor("op_12099_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12099_cast_fp16 = slice_by_index(begin = var_12099_begin_0, end = var_12099_end_0, end_mask = var_12099_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12099_cast_fp16")]; tensor var_12103_begin_0 = const()[name = tensor("op_12103_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_12103_end_0 = const()[name = tensor("op_12103_end_0"), val = tensor([2, 4096, 1, 200])]; tensor var_12103_end_mask_0 = const()[name = tensor("op_12103_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12103_cast_fp16 = slice_by_index(begin = var_12103_begin_0, end = var_12103_end_0, end_mask = var_12103_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12103_cast_fp16")]; tensor var_12107_begin_0 = const()[name = tensor("op_12107_begin_0"), val = tensor([0, 0, 0, 200])]; tensor var_12107_end_0 = const()[name = tensor("op_12107_end_0"), val = tensor([2, 4096, 1, 240])]; tensor var_12107_end_mask_0 = const()[name = tensor("op_12107_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12107_cast_fp16 = slice_by_index(begin = var_12107_begin_0, end = var_12107_end_0, end_mask = var_12107_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12107_cast_fp16")]; tensor var_12111_begin_0 = const()[name = tensor("op_12111_begin_0"), val = tensor([0, 0, 0, 240])]; tensor var_12111_end_0 = const()[name = tensor("op_12111_end_0"), val = tensor([2, 4096, 1, 280])]; tensor var_12111_end_mask_0 = const()[name = tensor("op_12111_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12111_cast_fp16 = slice_by_index(begin = var_12111_begin_0, end = var_12111_end_0, end_mask = var_12111_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12111_cast_fp16")]; tensor var_12115_begin_0 = const()[name = tensor("op_12115_begin_0"), val = tensor([0, 0, 0, 280])]; tensor var_12115_end_0 = const()[name = tensor("op_12115_end_0"), val = tensor([2, 4096, 1, 1])]; tensor var_12115_end_mask_0 = const()[name = tensor("op_12115_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_12115_cast_fp16 = slice_by_index(begin = var_12115_begin_0, end = var_12115_end_0, end_mask = var_12115_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12115_cast_fp16")]; tensor var_12117_begin_0 = const()[name = tensor("op_12117_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_12117_end_0 = const()[name = tensor("op_12117_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_12117_end_mask_0 = const()[name = tensor("op_12117_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_12117_cast_fp16 = slice_by_index(begin = var_12117_begin_0, end = var_12117_end_0, end_mask = var_12117_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12117_cast_fp16")]; tensor var_12121_begin_0 = const()[name = tensor("op_12121_begin_0"), val = tensor([0, 40, 0, 0])]; tensor var_12121_end_0 = const()[name = tensor("op_12121_end_0"), val = tensor([2, 80, 1, 4096])]; tensor var_12121_end_mask_0 = const()[name = tensor("op_12121_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_12121_cast_fp16 = slice_by_index(begin = var_12121_begin_0, end = var_12121_end_0, end_mask = var_12121_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12121_cast_fp16")]; tensor var_12125_begin_0 = const()[name = tensor("op_12125_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_12125_end_0 = const()[name = tensor("op_12125_end_0"), val = tensor([2, 120, 1, 4096])]; tensor var_12125_end_mask_0 = const()[name = tensor("op_12125_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_12125_cast_fp16 = slice_by_index(begin = var_12125_begin_0, end = var_12125_end_0, end_mask = var_12125_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12125_cast_fp16")]; tensor var_12129_begin_0 = const()[name = tensor("op_12129_begin_0"), val = tensor([0, 120, 0, 0])]; tensor var_12129_end_0 = const()[name = tensor("op_12129_end_0"), val = tensor([2, 160, 1, 4096])]; tensor var_12129_end_mask_0 = const()[name = tensor("op_12129_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_12129_cast_fp16 = slice_by_index(begin = var_12129_begin_0, end = var_12129_end_0, end_mask = var_12129_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12129_cast_fp16")]; tensor var_12133_begin_0 = const()[name = tensor("op_12133_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_12133_end_0 = const()[name = tensor("op_12133_end_0"), val = tensor([2, 200, 1, 4096])]; tensor var_12133_end_mask_0 = const()[name = tensor("op_12133_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_12133_cast_fp16 = slice_by_index(begin = var_12133_begin_0, end = var_12133_end_0, end_mask = var_12133_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12133_cast_fp16")]; tensor var_12137_begin_0 = const()[name = tensor("op_12137_begin_0"), val = tensor([0, 200, 0, 0])]; tensor var_12137_end_0 = const()[name = tensor("op_12137_end_0"), val = tensor([2, 240, 1, 4096])]; tensor var_12137_end_mask_0 = const()[name = tensor("op_12137_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_12137_cast_fp16 = slice_by_index(begin = var_12137_begin_0, end = var_12137_end_0, end_mask = var_12137_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12137_cast_fp16")]; tensor var_12141_begin_0 = const()[name = tensor("op_12141_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_12141_end_0 = const()[name = tensor("op_12141_end_0"), val = tensor([2, 280, 1, 4096])]; tensor var_12141_end_mask_0 = const()[name = tensor("op_12141_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_12141_cast_fp16 = slice_by_index(begin = var_12141_begin_0, end = var_12141_end_0, end_mask = var_12141_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12141_cast_fp16")]; tensor var_12145_begin_0 = const()[name = tensor("op_12145_begin_0"), val = tensor([0, 280, 0, 0])]; tensor var_12145_end_0 = const()[name = tensor("op_12145_end_0"), val = tensor([2, 1, 1, 4096])]; tensor var_12145_end_mask_0 = const()[name = tensor("op_12145_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_12145_cast_fp16 = slice_by_index(begin = var_12145_begin_0, end = var_12145_end_0, end_mask = var_12145_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12145_cast_fp16")]; tensor var_12149_equation_0 = const()[name = tensor("op_12149_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12149_cast_fp16 = einsum(equation = var_12149_equation_0, values = (var_12087_cast_fp16, var_12019_cast_fp16))[name = tensor("op_12149_cast_fp16")]; tensor var_12150_to_fp16 = const()[name = tensor("op_12150_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1089_cast_fp16 = mul(x = var_12149_cast_fp16, y = var_12150_to_fp16)[name = tensor("aw_chunk_1089_cast_fp16")]; tensor var_12153_equation_0 = const()[name = tensor("op_12153_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12153_cast_fp16 = einsum(equation = var_12153_equation_0, values = (var_12087_cast_fp16, var_12020_cast_fp16))[name = tensor("op_12153_cast_fp16")]; tensor var_12154_to_fp16 = const()[name = tensor("op_12154_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1091_cast_fp16 = mul(x = var_12153_cast_fp16, y = var_12154_to_fp16)[name = tensor("aw_chunk_1091_cast_fp16")]; tensor var_12157_equation_0 = const()[name = tensor("op_12157_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12157_cast_fp16 = einsum(equation = var_12157_equation_0, values = (var_12087_cast_fp16, var_12021_cast_fp16))[name = tensor("op_12157_cast_fp16")]; tensor var_12158_to_fp16 = const()[name = tensor("op_12158_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1093_cast_fp16 = mul(x = var_12157_cast_fp16, y = var_12158_to_fp16)[name = tensor("aw_chunk_1093_cast_fp16")]; tensor var_12161_equation_0 = const()[name = tensor("op_12161_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12161_cast_fp16 = einsum(equation = var_12161_equation_0, values = (var_12087_cast_fp16, var_12022_cast_fp16))[name = tensor("op_12161_cast_fp16")]; tensor var_12162_to_fp16 = const()[name = tensor("op_12162_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1095_cast_fp16 = mul(x = var_12161_cast_fp16, y = var_12162_to_fp16)[name = tensor("aw_chunk_1095_cast_fp16")]; tensor var_12165_equation_0 = const()[name = tensor("op_12165_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12165_cast_fp16 = einsum(equation = var_12165_equation_0, values = (var_12087_cast_fp16, var_12023_cast_fp16))[name = tensor("op_12165_cast_fp16")]; tensor var_12166_to_fp16 = const()[name = tensor("op_12166_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1097_cast_fp16 = mul(x = var_12165_cast_fp16, y = var_12166_to_fp16)[name = tensor("aw_chunk_1097_cast_fp16")]; tensor var_12169_equation_0 = const()[name = tensor("op_12169_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12169_cast_fp16 = einsum(equation = var_12169_equation_0, values = (var_12087_cast_fp16, var_12024_cast_fp16))[name = tensor("op_12169_cast_fp16")]; tensor var_12170_to_fp16 = const()[name = tensor("op_12170_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1099_cast_fp16 = mul(x = var_12169_cast_fp16, y = var_12170_to_fp16)[name = tensor("aw_chunk_1099_cast_fp16")]; tensor var_12173_equation_0 = const()[name = tensor("op_12173_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12173_cast_fp16 = einsum(equation = var_12173_equation_0, values = (var_12087_cast_fp16, var_12025_cast_fp16))[name = tensor("op_12173_cast_fp16")]; tensor var_12174_to_fp16 = const()[name = tensor("op_12174_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1101_cast_fp16 = mul(x = var_12173_cast_fp16, y = var_12174_to_fp16)[name = tensor("aw_chunk_1101_cast_fp16")]; tensor var_12177_equation_0 = const()[name = tensor("op_12177_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12177_cast_fp16 = einsum(equation = var_12177_equation_0, values = (var_12087_cast_fp16, var_12026_cast_fp16))[name = tensor("op_12177_cast_fp16")]; tensor var_12178_to_fp16 = const()[name = tensor("op_12178_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1103_cast_fp16 = mul(x = var_12177_cast_fp16, y = var_12178_to_fp16)[name = tensor("aw_chunk_1103_cast_fp16")]; tensor var_12181_equation_0 = const()[name = tensor("op_12181_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12181_cast_fp16 = einsum(equation = var_12181_equation_0, values = (var_12091_cast_fp16, var_12027_cast_fp16))[name = tensor("op_12181_cast_fp16")]; tensor var_12182_to_fp16 = const()[name = tensor("op_12182_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1105_cast_fp16 = mul(x = var_12181_cast_fp16, y = var_12182_to_fp16)[name = tensor("aw_chunk_1105_cast_fp16")]; tensor var_12185_equation_0 = const()[name = tensor("op_12185_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12185_cast_fp16 = einsum(equation = var_12185_equation_0, values = (var_12091_cast_fp16, var_12028_cast_fp16))[name = tensor("op_12185_cast_fp16")]; tensor var_12186_to_fp16 = const()[name = tensor("op_12186_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1107_cast_fp16 = mul(x = var_12185_cast_fp16, y = var_12186_to_fp16)[name = tensor("aw_chunk_1107_cast_fp16")]; tensor var_12189_equation_0 = const()[name = tensor("op_12189_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12189_cast_fp16 = einsum(equation = var_12189_equation_0, values = (var_12091_cast_fp16, var_12029_cast_fp16))[name = tensor("op_12189_cast_fp16")]; tensor var_12190_to_fp16 = const()[name = tensor("op_12190_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1109_cast_fp16 = mul(x = var_12189_cast_fp16, y = var_12190_to_fp16)[name = tensor("aw_chunk_1109_cast_fp16")]; tensor var_12193_equation_0 = const()[name = tensor("op_12193_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12193_cast_fp16 = einsum(equation = var_12193_equation_0, values = (var_12091_cast_fp16, var_12030_cast_fp16))[name = tensor("op_12193_cast_fp16")]; tensor var_12194_to_fp16 = const()[name = tensor("op_12194_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1111_cast_fp16 = mul(x = var_12193_cast_fp16, y = var_12194_to_fp16)[name = tensor("aw_chunk_1111_cast_fp16")]; tensor var_12197_equation_0 = const()[name = tensor("op_12197_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12197_cast_fp16 = einsum(equation = var_12197_equation_0, values = (var_12091_cast_fp16, var_12031_cast_fp16))[name = tensor("op_12197_cast_fp16")]; tensor var_12198_to_fp16 = const()[name = tensor("op_12198_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1113_cast_fp16 = mul(x = var_12197_cast_fp16, y = var_12198_to_fp16)[name = tensor("aw_chunk_1113_cast_fp16")]; tensor var_12201_equation_0 = const()[name = tensor("op_12201_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12201_cast_fp16 = einsum(equation = var_12201_equation_0, values = (var_12091_cast_fp16, var_12032_cast_fp16))[name = tensor("op_12201_cast_fp16")]; tensor var_12202_to_fp16 = const()[name = tensor("op_12202_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1115_cast_fp16 = mul(x = var_12201_cast_fp16, y = var_12202_to_fp16)[name = tensor("aw_chunk_1115_cast_fp16")]; tensor var_12205_equation_0 = const()[name = tensor("op_12205_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12205_cast_fp16 = einsum(equation = var_12205_equation_0, values = (var_12091_cast_fp16, var_12033_cast_fp16))[name = tensor("op_12205_cast_fp16")]; tensor var_12206_to_fp16 = const()[name = tensor("op_12206_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1117_cast_fp16 = mul(x = var_12205_cast_fp16, y = var_12206_to_fp16)[name = tensor("aw_chunk_1117_cast_fp16")]; tensor var_12209_equation_0 = const()[name = tensor("op_12209_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12209_cast_fp16 = einsum(equation = var_12209_equation_0, values = (var_12091_cast_fp16, var_12034_cast_fp16))[name = tensor("op_12209_cast_fp16")]; tensor var_12210_to_fp16 = const()[name = tensor("op_12210_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1119_cast_fp16 = mul(x = var_12209_cast_fp16, y = var_12210_to_fp16)[name = tensor("aw_chunk_1119_cast_fp16")]; tensor var_12213_equation_0 = const()[name = tensor("op_12213_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12213_cast_fp16 = einsum(equation = var_12213_equation_0, values = (var_12095_cast_fp16, var_12035_cast_fp16))[name = tensor("op_12213_cast_fp16")]; tensor var_12214_to_fp16 = const()[name = tensor("op_12214_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1121_cast_fp16 = mul(x = var_12213_cast_fp16, y = var_12214_to_fp16)[name = tensor("aw_chunk_1121_cast_fp16")]; tensor var_12217_equation_0 = const()[name = tensor("op_12217_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12217_cast_fp16 = einsum(equation = var_12217_equation_0, values = (var_12095_cast_fp16, var_12036_cast_fp16))[name = tensor("op_12217_cast_fp16")]; tensor var_12218_to_fp16 = const()[name = tensor("op_12218_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1123_cast_fp16 = mul(x = var_12217_cast_fp16, y = var_12218_to_fp16)[name = tensor("aw_chunk_1123_cast_fp16")]; tensor var_12221_equation_0 = const()[name = tensor("op_12221_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12221_cast_fp16 = einsum(equation = var_12221_equation_0, values = (var_12095_cast_fp16, var_12037_cast_fp16))[name = tensor("op_12221_cast_fp16")]; tensor var_12222_to_fp16 = const()[name = tensor("op_12222_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1125_cast_fp16 = mul(x = var_12221_cast_fp16, y = var_12222_to_fp16)[name = tensor("aw_chunk_1125_cast_fp16")]; tensor var_12225_equation_0 = const()[name = tensor("op_12225_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12225_cast_fp16 = einsum(equation = var_12225_equation_0, values = (var_12095_cast_fp16, var_12038_cast_fp16))[name = tensor("op_12225_cast_fp16")]; tensor var_12226_to_fp16 = const()[name = tensor("op_12226_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1127_cast_fp16 = mul(x = var_12225_cast_fp16, y = var_12226_to_fp16)[name = tensor("aw_chunk_1127_cast_fp16")]; tensor var_12229_equation_0 = const()[name = tensor("op_12229_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12229_cast_fp16 = einsum(equation = var_12229_equation_0, values = (var_12095_cast_fp16, var_12039_cast_fp16))[name = tensor("op_12229_cast_fp16")]; tensor var_12230_to_fp16 = const()[name = tensor("op_12230_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1129_cast_fp16 = mul(x = var_12229_cast_fp16, y = var_12230_to_fp16)[name = tensor("aw_chunk_1129_cast_fp16")]; tensor var_12233_equation_0 = const()[name = tensor("op_12233_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12233_cast_fp16 = einsum(equation = var_12233_equation_0, values = (var_12095_cast_fp16, var_12040_cast_fp16))[name = tensor("op_12233_cast_fp16")]; tensor var_12234_to_fp16 = const()[name = tensor("op_12234_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1131_cast_fp16 = mul(x = var_12233_cast_fp16, y = var_12234_to_fp16)[name = tensor("aw_chunk_1131_cast_fp16")]; tensor var_12237_equation_0 = const()[name = tensor("op_12237_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12237_cast_fp16 = einsum(equation = var_12237_equation_0, values = (var_12095_cast_fp16, var_12041_cast_fp16))[name = tensor("op_12237_cast_fp16")]; tensor var_12238_to_fp16 = const()[name = tensor("op_12238_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1133_cast_fp16 = mul(x = var_12237_cast_fp16, y = var_12238_to_fp16)[name = tensor("aw_chunk_1133_cast_fp16")]; tensor var_12241_equation_0 = const()[name = tensor("op_12241_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12241_cast_fp16 = einsum(equation = var_12241_equation_0, values = (var_12095_cast_fp16, var_12042_cast_fp16))[name = tensor("op_12241_cast_fp16")]; tensor var_12242_to_fp16 = const()[name = tensor("op_12242_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1135_cast_fp16 = mul(x = var_12241_cast_fp16, y = var_12242_to_fp16)[name = tensor("aw_chunk_1135_cast_fp16")]; tensor var_12245_equation_0 = const()[name = tensor("op_12245_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12245_cast_fp16 = einsum(equation = var_12245_equation_0, values = (var_12099_cast_fp16, var_12043_cast_fp16))[name = tensor("op_12245_cast_fp16")]; tensor var_12246_to_fp16 = const()[name = tensor("op_12246_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1137_cast_fp16 = mul(x = var_12245_cast_fp16, y = var_12246_to_fp16)[name = tensor("aw_chunk_1137_cast_fp16")]; tensor var_12249_equation_0 = const()[name = tensor("op_12249_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12249_cast_fp16 = einsum(equation = var_12249_equation_0, values = (var_12099_cast_fp16, var_12044_cast_fp16))[name = tensor("op_12249_cast_fp16")]; tensor var_12250_to_fp16 = const()[name = tensor("op_12250_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1139_cast_fp16 = mul(x = var_12249_cast_fp16, y = var_12250_to_fp16)[name = tensor("aw_chunk_1139_cast_fp16")]; tensor var_12253_equation_0 = const()[name = tensor("op_12253_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12253_cast_fp16 = einsum(equation = var_12253_equation_0, values = (var_12099_cast_fp16, var_12045_cast_fp16))[name = tensor("op_12253_cast_fp16")]; tensor var_12254_to_fp16 = const()[name = tensor("op_12254_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1141_cast_fp16 = mul(x = var_12253_cast_fp16, y = var_12254_to_fp16)[name = tensor("aw_chunk_1141_cast_fp16")]; tensor var_12257_equation_0 = const()[name = tensor("op_12257_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12257_cast_fp16 = einsum(equation = var_12257_equation_0, values = (var_12099_cast_fp16, var_12046_cast_fp16))[name = tensor("op_12257_cast_fp16")]; tensor var_12258_to_fp16 = const()[name = tensor("op_12258_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1143_cast_fp16 = mul(x = var_12257_cast_fp16, y = var_12258_to_fp16)[name = tensor("aw_chunk_1143_cast_fp16")]; tensor var_12261_equation_0 = const()[name = tensor("op_12261_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12261_cast_fp16 = einsum(equation = var_12261_equation_0, values = (var_12099_cast_fp16, var_12047_cast_fp16))[name = tensor("op_12261_cast_fp16")]; tensor var_12262_to_fp16 = const()[name = tensor("op_12262_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1145_cast_fp16 = mul(x = var_12261_cast_fp16, y = var_12262_to_fp16)[name = tensor("aw_chunk_1145_cast_fp16")]; tensor var_12265_equation_0 = const()[name = tensor("op_12265_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12265_cast_fp16 = einsum(equation = var_12265_equation_0, values = (var_12099_cast_fp16, var_12048_cast_fp16))[name = tensor("op_12265_cast_fp16")]; tensor var_12266_to_fp16 = const()[name = tensor("op_12266_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1147_cast_fp16 = mul(x = var_12265_cast_fp16, y = var_12266_to_fp16)[name = tensor("aw_chunk_1147_cast_fp16")]; tensor var_12269_equation_0 = const()[name = tensor("op_12269_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12269_cast_fp16 = einsum(equation = var_12269_equation_0, values = (var_12099_cast_fp16, var_12049_cast_fp16))[name = tensor("op_12269_cast_fp16")]; tensor var_12270_to_fp16 = const()[name = tensor("op_12270_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1149_cast_fp16 = mul(x = var_12269_cast_fp16, y = var_12270_to_fp16)[name = tensor("aw_chunk_1149_cast_fp16")]; tensor var_12273_equation_0 = const()[name = tensor("op_12273_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12273_cast_fp16 = einsum(equation = var_12273_equation_0, values = (var_12099_cast_fp16, var_12050_cast_fp16))[name = tensor("op_12273_cast_fp16")]; tensor var_12274_to_fp16 = const()[name = tensor("op_12274_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1151_cast_fp16 = mul(x = var_12273_cast_fp16, y = var_12274_to_fp16)[name = tensor("aw_chunk_1151_cast_fp16")]; tensor var_12277_equation_0 = const()[name = tensor("op_12277_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12277_cast_fp16 = einsum(equation = var_12277_equation_0, values = (var_12103_cast_fp16, var_12051_cast_fp16))[name = tensor("op_12277_cast_fp16")]; tensor var_12278_to_fp16 = const()[name = tensor("op_12278_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1153_cast_fp16 = mul(x = var_12277_cast_fp16, y = var_12278_to_fp16)[name = tensor("aw_chunk_1153_cast_fp16")]; tensor var_12281_equation_0 = const()[name = tensor("op_12281_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12281_cast_fp16 = einsum(equation = var_12281_equation_0, values = (var_12103_cast_fp16, var_12052_cast_fp16))[name = tensor("op_12281_cast_fp16")]; tensor var_12282_to_fp16 = const()[name = tensor("op_12282_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1155_cast_fp16 = mul(x = var_12281_cast_fp16, y = var_12282_to_fp16)[name = tensor("aw_chunk_1155_cast_fp16")]; tensor var_12285_equation_0 = const()[name = tensor("op_12285_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12285_cast_fp16 = einsum(equation = var_12285_equation_0, values = (var_12103_cast_fp16, var_12053_cast_fp16))[name = tensor("op_12285_cast_fp16")]; tensor var_12286_to_fp16 = const()[name = tensor("op_12286_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1157_cast_fp16 = mul(x = var_12285_cast_fp16, y = var_12286_to_fp16)[name = tensor("aw_chunk_1157_cast_fp16")]; tensor var_12289_equation_0 = const()[name = tensor("op_12289_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12289_cast_fp16 = einsum(equation = var_12289_equation_0, values = (var_12103_cast_fp16, var_12054_cast_fp16))[name = tensor("op_12289_cast_fp16")]; tensor var_12290_to_fp16 = const()[name = tensor("op_12290_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1159_cast_fp16 = mul(x = var_12289_cast_fp16, y = var_12290_to_fp16)[name = tensor("aw_chunk_1159_cast_fp16")]; tensor var_12293_equation_0 = const()[name = tensor("op_12293_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12293_cast_fp16 = einsum(equation = var_12293_equation_0, values = (var_12103_cast_fp16, var_12055_cast_fp16))[name = tensor("op_12293_cast_fp16")]; tensor var_12294_to_fp16 = const()[name = tensor("op_12294_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1161_cast_fp16 = mul(x = var_12293_cast_fp16, y = var_12294_to_fp16)[name = tensor("aw_chunk_1161_cast_fp16")]; tensor var_12297_equation_0 = const()[name = tensor("op_12297_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12297_cast_fp16 = einsum(equation = var_12297_equation_0, values = (var_12103_cast_fp16, var_12056_cast_fp16))[name = tensor("op_12297_cast_fp16")]; tensor var_12298_to_fp16 = const()[name = tensor("op_12298_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1163_cast_fp16 = mul(x = var_12297_cast_fp16, y = var_12298_to_fp16)[name = tensor("aw_chunk_1163_cast_fp16")]; tensor var_12301_equation_0 = const()[name = tensor("op_12301_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12301_cast_fp16 = einsum(equation = var_12301_equation_0, values = (var_12103_cast_fp16, var_12057_cast_fp16))[name = tensor("op_12301_cast_fp16")]; tensor var_12302_to_fp16 = const()[name = tensor("op_12302_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1165_cast_fp16 = mul(x = var_12301_cast_fp16, y = var_12302_to_fp16)[name = tensor("aw_chunk_1165_cast_fp16")]; tensor var_12305_equation_0 = const()[name = tensor("op_12305_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12305_cast_fp16 = einsum(equation = var_12305_equation_0, values = (var_12103_cast_fp16, var_12058_cast_fp16))[name = tensor("op_12305_cast_fp16")]; tensor var_12306_to_fp16 = const()[name = tensor("op_12306_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1167_cast_fp16 = mul(x = var_12305_cast_fp16, y = var_12306_to_fp16)[name = tensor("aw_chunk_1167_cast_fp16")]; tensor var_12309_equation_0 = const()[name = tensor("op_12309_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12309_cast_fp16 = einsum(equation = var_12309_equation_0, values = (var_12107_cast_fp16, var_12059_cast_fp16))[name = tensor("op_12309_cast_fp16")]; tensor var_12310_to_fp16 = const()[name = tensor("op_12310_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1169_cast_fp16 = mul(x = var_12309_cast_fp16, y = var_12310_to_fp16)[name = tensor("aw_chunk_1169_cast_fp16")]; tensor var_12313_equation_0 = const()[name = tensor("op_12313_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12313_cast_fp16 = einsum(equation = var_12313_equation_0, values = (var_12107_cast_fp16, var_12060_cast_fp16))[name = tensor("op_12313_cast_fp16")]; tensor var_12314_to_fp16 = const()[name = tensor("op_12314_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1171_cast_fp16 = mul(x = var_12313_cast_fp16, y = var_12314_to_fp16)[name = tensor("aw_chunk_1171_cast_fp16")]; tensor var_12317_equation_0 = const()[name = tensor("op_12317_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12317_cast_fp16 = einsum(equation = var_12317_equation_0, values = (var_12107_cast_fp16, var_12061_cast_fp16))[name = tensor("op_12317_cast_fp16")]; tensor var_12318_to_fp16 = const()[name = tensor("op_12318_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1173_cast_fp16 = mul(x = var_12317_cast_fp16, y = var_12318_to_fp16)[name = tensor("aw_chunk_1173_cast_fp16")]; tensor var_12321_equation_0 = const()[name = tensor("op_12321_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12321_cast_fp16 = einsum(equation = var_12321_equation_0, values = (var_12107_cast_fp16, var_12062_cast_fp16))[name = tensor("op_12321_cast_fp16")]; tensor var_12322_to_fp16 = const()[name = tensor("op_12322_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1175_cast_fp16 = mul(x = var_12321_cast_fp16, y = var_12322_to_fp16)[name = tensor("aw_chunk_1175_cast_fp16")]; tensor var_12325_equation_0 = const()[name = tensor("op_12325_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12325_cast_fp16 = einsum(equation = var_12325_equation_0, values = (var_12107_cast_fp16, var_12063_cast_fp16))[name = tensor("op_12325_cast_fp16")]; tensor var_12326_to_fp16 = const()[name = tensor("op_12326_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1177_cast_fp16 = mul(x = var_12325_cast_fp16, y = var_12326_to_fp16)[name = tensor("aw_chunk_1177_cast_fp16")]; tensor var_12329_equation_0 = const()[name = tensor("op_12329_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12329_cast_fp16 = einsum(equation = var_12329_equation_0, values = (var_12107_cast_fp16, var_12064_cast_fp16))[name = tensor("op_12329_cast_fp16")]; tensor var_12330_to_fp16 = const()[name = tensor("op_12330_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1179_cast_fp16 = mul(x = var_12329_cast_fp16, y = var_12330_to_fp16)[name = tensor("aw_chunk_1179_cast_fp16")]; tensor var_12333_equation_0 = const()[name = tensor("op_12333_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12333_cast_fp16 = einsum(equation = var_12333_equation_0, values = (var_12107_cast_fp16, var_12065_cast_fp16))[name = tensor("op_12333_cast_fp16")]; tensor var_12334_to_fp16 = const()[name = tensor("op_12334_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1181_cast_fp16 = mul(x = var_12333_cast_fp16, y = var_12334_to_fp16)[name = tensor("aw_chunk_1181_cast_fp16")]; tensor var_12337_equation_0 = const()[name = tensor("op_12337_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12337_cast_fp16 = einsum(equation = var_12337_equation_0, values = (var_12107_cast_fp16, var_12066_cast_fp16))[name = tensor("op_12337_cast_fp16")]; tensor var_12338_to_fp16 = const()[name = tensor("op_12338_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1183_cast_fp16 = mul(x = var_12337_cast_fp16, y = var_12338_to_fp16)[name = tensor("aw_chunk_1183_cast_fp16")]; tensor var_12341_equation_0 = const()[name = tensor("op_12341_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12341_cast_fp16 = einsum(equation = var_12341_equation_0, values = (var_12111_cast_fp16, var_12067_cast_fp16))[name = tensor("op_12341_cast_fp16")]; tensor var_12342_to_fp16 = const()[name = tensor("op_12342_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1185_cast_fp16 = mul(x = var_12341_cast_fp16, y = var_12342_to_fp16)[name = tensor("aw_chunk_1185_cast_fp16")]; tensor var_12345_equation_0 = const()[name = tensor("op_12345_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12345_cast_fp16 = einsum(equation = var_12345_equation_0, values = (var_12111_cast_fp16, var_12068_cast_fp16))[name = tensor("op_12345_cast_fp16")]; tensor var_12346_to_fp16 = const()[name = tensor("op_12346_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1187_cast_fp16 = mul(x = var_12345_cast_fp16, y = var_12346_to_fp16)[name = tensor("aw_chunk_1187_cast_fp16")]; tensor var_12349_equation_0 = const()[name = tensor("op_12349_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12349_cast_fp16 = einsum(equation = var_12349_equation_0, values = (var_12111_cast_fp16, var_12069_cast_fp16))[name = tensor("op_12349_cast_fp16")]; tensor var_12350_to_fp16 = const()[name = tensor("op_12350_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1189_cast_fp16 = mul(x = var_12349_cast_fp16, y = var_12350_to_fp16)[name = tensor("aw_chunk_1189_cast_fp16")]; tensor var_12353_equation_0 = const()[name = tensor("op_12353_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12353_cast_fp16 = einsum(equation = var_12353_equation_0, values = (var_12111_cast_fp16, var_12070_cast_fp16))[name = tensor("op_12353_cast_fp16")]; tensor var_12354_to_fp16 = const()[name = tensor("op_12354_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1191_cast_fp16 = mul(x = var_12353_cast_fp16, y = var_12354_to_fp16)[name = tensor("aw_chunk_1191_cast_fp16")]; tensor var_12357_equation_0 = const()[name = tensor("op_12357_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12357_cast_fp16 = einsum(equation = var_12357_equation_0, values = (var_12111_cast_fp16, var_12071_cast_fp16))[name = tensor("op_12357_cast_fp16")]; tensor var_12358_to_fp16 = const()[name = tensor("op_12358_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1193_cast_fp16 = mul(x = var_12357_cast_fp16, y = var_12358_to_fp16)[name = tensor("aw_chunk_1193_cast_fp16")]; tensor var_12361_equation_0 = const()[name = tensor("op_12361_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12361_cast_fp16 = einsum(equation = var_12361_equation_0, values = (var_12111_cast_fp16, var_12072_cast_fp16))[name = tensor("op_12361_cast_fp16")]; tensor var_12362_to_fp16 = const()[name = tensor("op_12362_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1195_cast_fp16 = mul(x = var_12361_cast_fp16, y = var_12362_to_fp16)[name = tensor("aw_chunk_1195_cast_fp16")]; tensor var_12365_equation_0 = const()[name = tensor("op_12365_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12365_cast_fp16 = einsum(equation = var_12365_equation_0, values = (var_12111_cast_fp16, var_12073_cast_fp16))[name = tensor("op_12365_cast_fp16")]; tensor var_12366_to_fp16 = const()[name = tensor("op_12366_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1197_cast_fp16 = mul(x = var_12365_cast_fp16, y = var_12366_to_fp16)[name = tensor("aw_chunk_1197_cast_fp16")]; tensor var_12369_equation_0 = const()[name = tensor("op_12369_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12369_cast_fp16 = einsum(equation = var_12369_equation_0, values = (var_12111_cast_fp16, var_12074_cast_fp16))[name = tensor("op_12369_cast_fp16")]; tensor var_12370_to_fp16 = const()[name = tensor("op_12370_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1199_cast_fp16 = mul(x = var_12369_cast_fp16, y = var_12370_to_fp16)[name = tensor("aw_chunk_1199_cast_fp16")]; tensor var_12373_equation_0 = const()[name = tensor("op_12373_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12373_cast_fp16 = einsum(equation = var_12373_equation_0, values = (var_12115_cast_fp16, var_12075_cast_fp16))[name = tensor("op_12373_cast_fp16")]; tensor var_12374_to_fp16 = const()[name = tensor("op_12374_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1201_cast_fp16 = mul(x = var_12373_cast_fp16, y = var_12374_to_fp16)[name = tensor("aw_chunk_1201_cast_fp16")]; tensor var_12377_equation_0 = const()[name = tensor("op_12377_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12377_cast_fp16 = einsum(equation = var_12377_equation_0, values = (var_12115_cast_fp16, var_12076_cast_fp16))[name = tensor("op_12377_cast_fp16")]; tensor var_12378_to_fp16 = const()[name = tensor("op_12378_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1203_cast_fp16 = mul(x = var_12377_cast_fp16, y = var_12378_to_fp16)[name = tensor("aw_chunk_1203_cast_fp16")]; tensor var_12381_equation_0 = const()[name = tensor("op_12381_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12381_cast_fp16 = einsum(equation = var_12381_equation_0, values = (var_12115_cast_fp16, var_12077_cast_fp16))[name = tensor("op_12381_cast_fp16")]; tensor var_12382_to_fp16 = const()[name = tensor("op_12382_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1205_cast_fp16 = mul(x = var_12381_cast_fp16, y = var_12382_to_fp16)[name = tensor("aw_chunk_1205_cast_fp16")]; tensor var_12385_equation_0 = const()[name = tensor("op_12385_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12385_cast_fp16 = einsum(equation = var_12385_equation_0, values = (var_12115_cast_fp16, var_12078_cast_fp16))[name = tensor("op_12385_cast_fp16")]; tensor var_12386_to_fp16 = const()[name = tensor("op_12386_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1207_cast_fp16 = mul(x = var_12385_cast_fp16, y = var_12386_to_fp16)[name = tensor("aw_chunk_1207_cast_fp16")]; tensor var_12389_equation_0 = const()[name = tensor("op_12389_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12389_cast_fp16 = einsum(equation = var_12389_equation_0, values = (var_12115_cast_fp16, var_12079_cast_fp16))[name = tensor("op_12389_cast_fp16")]; tensor var_12390_to_fp16 = const()[name = tensor("op_12390_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1209_cast_fp16 = mul(x = var_12389_cast_fp16, y = var_12390_to_fp16)[name = tensor("aw_chunk_1209_cast_fp16")]; tensor var_12393_equation_0 = const()[name = tensor("op_12393_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12393_cast_fp16 = einsum(equation = var_12393_equation_0, values = (var_12115_cast_fp16, var_12080_cast_fp16))[name = tensor("op_12393_cast_fp16")]; tensor var_12394_to_fp16 = const()[name = tensor("op_12394_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1211_cast_fp16 = mul(x = var_12393_cast_fp16, y = var_12394_to_fp16)[name = tensor("aw_chunk_1211_cast_fp16")]; tensor var_12397_equation_0 = const()[name = tensor("op_12397_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12397_cast_fp16 = einsum(equation = var_12397_equation_0, values = (var_12115_cast_fp16, var_12081_cast_fp16))[name = tensor("op_12397_cast_fp16")]; tensor var_12398_to_fp16 = const()[name = tensor("op_12398_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1213_cast_fp16 = mul(x = var_12397_cast_fp16, y = var_12398_to_fp16)[name = tensor("aw_chunk_1213_cast_fp16")]; tensor var_12401_equation_0 = const()[name = tensor("op_12401_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12401_cast_fp16 = einsum(equation = var_12401_equation_0, values = (var_12115_cast_fp16, var_12082_cast_fp16))[name = tensor("op_12401_cast_fp16")]; tensor var_12402_to_fp16 = const()[name = tensor("op_12402_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1215_cast_fp16 = mul(x = var_12401_cast_fp16, y = var_12402_to_fp16)[name = tensor("aw_chunk_1215_cast_fp16")]; tensor var_12404_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1089_cast_fp16)[name = tensor("op_12404_cast_fp16")]; tensor var_12405_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1091_cast_fp16)[name = tensor("op_12405_cast_fp16")]; tensor var_12406_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1093_cast_fp16)[name = tensor("op_12406_cast_fp16")]; tensor var_12407_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1095_cast_fp16)[name = tensor("op_12407_cast_fp16")]; tensor var_12408_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1097_cast_fp16)[name = tensor("op_12408_cast_fp16")]; tensor var_12409_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1099_cast_fp16)[name = tensor("op_12409_cast_fp16")]; tensor var_12410_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1101_cast_fp16)[name = tensor("op_12410_cast_fp16")]; tensor var_12411_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1103_cast_fp16)[name = tensor("op_12411_cast_fp16")]; tensor var_12412_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1105_cast_fp16)[name = tensor("op_12412_cast_fp16")]; tensor var_12413_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1107_cast_fp16)[name = tensor("op_12413_cast_fp16")]; tensor var_12414_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1109_cast_fp16)[name = tensor("op_12414_cast_fp16")]; tensor var_12415_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1111_cast_fp16)[name = tensor("op_12415_cast_fp16")]; tensor var_12416_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1113_cast_fp16)[name = tensor("op_12416_cast_fp16")]; tensor var_12417_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1115_cast_fp16)[name = tensor("op_12417_cast_fp16")]; tensor var_12418_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1117_cast_fp16)[name = tensor("op_12418_cast_fp16")]; tensor var_12419_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1119_cast_fp16)[name = tensor("op_12419_cast_fp16")]; tensor var_12420_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1121_cast_fp16)[name = tensor("op_12420_cast_fp16")]; tensor var_12421_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1123_cast_fp16)[name = tensor("op_12421_cast_fp16")]; tensor var_12422_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1125_cast_fp16)[name = tensor("op_12422_cast_fp16")]; tensor var_12423_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1127_cast_fp16)[name = tensor("op_12423_cast_fp16")]; tensor var_12424_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1129_cast_fp16)[name = tensor("op_12424_cast_fp16")]; tensor var_12425_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1131_cast_fp16)[name = tensor("op_12425_cast_fp16")]; tensor var_12426_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1133_cast_fp16)[name = tensor("op_12426_cast_fp16")]; tensor var_12427_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1135_cast_fp16)[name = tensor("op_12427_cast_fp16")]; tensor var_12428_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1137_cast_fp16)[name = tensor("op_12428_cast_fp16")]; tensor var_12429_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1139_cast_fp16)[name = tensor("op_12429_cast_fp16")]; tensor var_12430_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1141_cast_fp16)[name = tensor("op_12430_cast_fp16")]; tensor var_12431_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1143_cast_fp16)[name = tensor("op_12431_cast_fp16")]; tensor var_12432_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1145_cast_fp16)[name = tensor("op_12432_cast_fp16")]; tensor var_12433_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1147_cast_fp16)[name = tensor("op_12433_cast_fp16")]; tensor var_12434_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1149_cast_fp16)[name = tensor("op_12434_cast_fp16")]; tensor var_12435_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1151_cast_fp16)[name = tensor("op_12435_cast_fp16")]; tensor var_12436_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1153_cast_fp16)[name = tensor("op_12436_cast_fp16")]; tensor var_12437_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1155_cast_fp16)[name = tensor("op_12437_cast_fp16")]; tensor var_12438_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1157_cast_fp16)[name = tensor("op_12438_cast_fp16")]; tensor var_12439_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1159_cast_fp16)[name = tensor("op_12439_cast_fp16")]; tensor var_12440_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1161_cast_fp16)[name = tensor("op_12440_cast_fp16")]; tensor var_12441_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1163_cast_fp16)[name = tensor("op_12441_cast_fp16")]; tensor var_12442_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1165_cast_fp16)[name = tensor("op_12442_cast_fp16")]; tensor var_12443_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1167_cast_fp16)[name = tensor("op_12443_cast_fp16")]; tensor var_12444_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1169_cast_fp16)[name = tensor("op_12444_cast_fp16")]; tensor var_12445_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1171_cast_fp16)[name = tensor("op_12445_cast_fp16")]; tensor var_12446_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1173_cast_fp16)[name = tensor("op_12446_cast_fp16")]; tensor var_12447_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1175_cast_fp16)[name = tensor("op_12447_cast_fp16")]; tensor var_12448_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1177_cast_fp16)[name = tensor("op_12448_cast_fp16")]; tensor var_12449_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1179_cast_fp16)[name = tensor("op_12449_cast_fp16")]; tensor var_12450_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1181_cast_fp16)[name = tensor("op_12450_cast_fp16")]; tensor var_12451_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1183_cast_fp16)[name = tensor("op_12451_cast_fp16")]; tensor var_12452_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1185_cast_fp16)[name = tensor("op_12452_cast_fp16")]; tensor var_12453_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1187_cast_fp16)[name = tensor("op_12453_cast_fp16")]; tensor var_12454_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1189_cast_fp16)[name = tensor("op_12454_cast_fp16")]; tensor var_12455_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1191_cast_fp16)[name = tensor("op_12455_cast_fp16")]; tensor var_12456_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1193_cast_fp16)[name = tensor("op_12456_cast_fp16")]; tensor var_12457_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1195_cast_fp16)[name = tensor("op_12457_cast_fp16")]; tensor var_12458_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1197_cast_fp16)[name = tensor("op_12458_cast_fp16")]; tensor var_12459_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1199_cast_fp16)[name = tensor("op_12459_cast_fp16")]; tensor var_12460_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1201_cast_fp16)[name = tensor("op_12460_cast_fp16")]; tensor var_12461_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1203_cast_fp16)[name = tensor("op_12461_cast_fp16")]; tensor var_12462_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1205_cast_fp16)[name = tensor("op_12462_cast_fp16")]; tensor var_12463_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1207_cast_fp16)[name = tensor("op_12463_cast_fp16")]; tensor var_12464_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1209_cast_fp16)[name = tensor("op_12464_cast_fp16")]; tensor var_12465_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1211_cast_fp16)[name = tensor("op_12465_cast_fp16")]; tensor var_12466_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1213_cast_fp16)[name = tensor("op_12466_cast_fp16")]; tensor var_12467_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1215_cast_fp16)[name = tensor("op_12467_cast_fp16")]; tensor var_12469_equation_0 = const()[name = tensor("op_12469_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12469_cast_fp16 = einsum(equation = var_12469_equation_0, values = (var_12117_cast_fp16, var_12404_cast_fp16))[name = tensor("op_12469_cast_fp16")]; tensor var_12471_equation_0 = const()[name = tensor("op_12471_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12471_cast_fp16 = einsum(equation = var_12471_equation_0, values = (var_12117_cast_fp16, var_12405_cast_fp16))[name = tensor("op_12471_cast_fp16")]; tensor var_12473_equation_0 = const()[name = tensor("op_12473_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12473_cast_fp16 = einsum(equation = var_12473_equation_0, values = (var_12117_cast_fp16, var_12406_cast_fp16))[name = tensor("op_12473_cast_fp16")]; tensor var_12475_equation_0 = const()[name = tensor("op_12475_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12475_cast_fp16 = einsum(equation = var_12475_equation_0, values = (var_12117_cast_fp16, var_12407_cast_fp16))[name = tensor("op_12475_cast_fp16")]; tensor var_12477_equation_0 = const()[name = tensor("op_12477_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12477_cast_fp16 = einsum(equation = var_12477_equation_0, values = (var_12117_cast_fp16, var_12408_cast_fp16))[name = tensor("op_12477_cast_fp16")]; tensor var_12479_equation_0 = const()[name = tensor("op_12479_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12479_cast_fp16 = einsum(equation = var_12479_equation_0, values = (var_12117_cast_fp16, var_12409_cast_fp16))[name = tensor("op_12479_cast_fp16")]; tensor var_12481_equation_0 = const()[name = tensor("op_12481_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12481_cast_fp16 = einsum(equation = var_12481_equation_0, values = (var_12117_cast_fp16, var_12410_cast_fp16))[name = tensor("op_12481_cast_fp16")]; tensor var_12483_equation_0 = const()[name = tensor("op_12483_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12483_cast_fp16 = einsum(equation = var_12483_equation_0, values = (var_12117_cast_fp16, var_12411_cast_fp16))[name = tensor("op_12483_cast_fp16")]; tensor var_12485_equation_0 = const()[name = tensor("op_12485_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12485_cast_fp16 = einsum(equation = var_12485_equation_0, values = (var_12121_cast_fp16, var_12412_cast_fp16))[name = tensor("op_12485_cast_fp16")]; tensor var_12487_equation_0 = const()[name = tensor("op_12487_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12487_cast_fp16 = einsum(equation = var_12487_equation_0, values = (var_12121_cast_fp16, var_12413_cast_fp16))[name = tensor("op_12487_cast_fp16")]; tensor var_12489_equation_0 = const()[name = tensor("op_12489_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12489_cast_fp16 = einsum(equation = var_12489_equation_0, values = (var_12121_cast_fp16, var_12414_cast_fp16))[name = tensor("op_12489_cast_fp16")]; tensor var_12491_equation_0 = const()[name = tensor("op_12491_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12491_cast_fp16 = einsum(equation = var_12491_equation_0, values = (var_12121_cast_fp16, var_12415_cast_fp16))[name = tensor("op_12491_cast_fp16")]; tensor var_12493_equation_0 = const()[name = tensor("op_12493_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12493_cast_fp16 = einsum(equation = var_12493_equation_0, values = (var_12121_cast_fp16, var_12416_cast_fp16))[name = tensor("op_12493_cast_fp16")]; tensor var_12495_equation_0 = const()[name = tensor("op_12495_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12495_cast_fp16 = einsum(equation = var_12495_equation_0, values = (var_12121_cast_fp16, var_12417_cast_fp16))[name = tensor("op_12495_cast_fp16")]; tensor var_12497_equation_0 = const()[name = tensor("op_12497_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12497_cast_fp16 = einsum(equation = var_12497_equation_0, values = (var_12121_cast_fp16, var_12418_cast_fp16))[name = tensor("op_12497_cast_fp16")]; tensor var_12499_equation_0 = const()[name = tensor("op_12499_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12499_cast_fp16 = einsum(equation = var_12499_equation_0, values = (var_12121_cast_fp16, var_12419_cast_fp16))[name = tensor("op_12499_cast_fp16")]; tensor var_12501_equation_0 = const()[name = tensor("op_12501_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12501_cast_fp16 = einsum(equation = var_12501_equation_0, values = (var_12125_cast_fp16, var_12420_cast_fp16))[name = tensor("op_12501_cast_fp16")]; tensor var_12503_equation_0 = const()[name = tensor("op_12503_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12503_cast_fp16 = einsum(equation = var_12503_equation_0, values = (var_12125_cast_fp16, var_12421_cast_fp16))[name = tensor("op_12503_cast_fp16")]; tensor var_12505_equation_0 = const()[name = tensor("op_12505_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12505_cast_fp16 = einsum(equation = var_12505_equation_0, values = (var_12125_cast_fp16, var_12422_cast_fp16))[name = tensor("op_12505_cast_fp16")]; tensor var_12507_equation_0 = const()[name = tensor("op_12507_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12507_cast_fp16 = einsum(equation = var_12507_equation_0, values = (var_12125_cast_fp16, var_12423_cast_fp16))[name = tensor("op_12507_cast_fp16")]; tensor var_12509_equation_0 = const()[name = tensor("op_12509_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12509_cast_fp16 = einsum(equation = var_12509_equation_0, values = (var_12125_cast_fp16, var_12424_cast_fp16))[name = tensor("op_12509_cast_fp16")]; tensor var_12511_equation_0 = const()[name = tensor("op_12511_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12511_cast_fp16 = einsum(equation = var_12511_equation_0, values = (var_12125_cast_fp16, var_12425_cast_fp16))[name = tensor("op_12511_cast_fp16")]; tensor var_12513_equation_0 = const()[name = tensor("op_12513_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12513_cast_fp16 = einsum(equation = var_12513_equation_0, values = (var_12125_cast_fp16, var_12426_cast_fp16))[name = tensor("op_12513_cast_fp16")]; tensor var_12515_equation_0 = const()[name = tensor("op_12515_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12515_cast_fp16 = einsum(equation = var_12515_equation_0, values = (var_12125_cast_fp16, var_12427_cast_fp16))[name = tensor("op_12515_cast_fp16")]; tensor var_12517_equation_0 = const()[name = tensor("op_12517_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12517_cast_fp16 = einsum(equation = var_12517_equation_0, values = (var_12129_cast_fp16, var_12428_cast_fp16))[name = tensor("op_12517_cast_fp16")]; tensor var_12519_equation_0 = const()[name = tensor("op_12519_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12519_cast_fp16 = einsum(equation = var_12519_equation_0, values = (var_12129_cast_fp16, var_12429_cast_fp16))[name = tensor("op_12519_cast_fp16")]; tensor var_12521_equation_0 = const()[name = tensor("op_12521_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12521_cast_fp16 = einsum(equation = var_12521_equation_0, values = (var_12129_cast_fp16, var_12430_cast_fp16))[name = tensor("op_12521_cast_fp16")]; tensor var_12523_equation_0 = const()[name = tensor("op_12523_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12523_cast_fp16 = einsum(equation = var_12523_equation_0, values = (var_12129_cast_fp16, var_12431_cast_fp16))[name = tensor("op_12523_cast_fp16")]; tensor var_12525_equation_0 = const()[name = tensor("op_12525_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12525_cast_fp16 = einsum(equation = var_12525_equation_0, values = (var_12129_cast_fp16, var_12432_cast_fp16))[name = tensor("op_12525_cast_fp16")]; tensor var_12527_equation_0 = const()[name = tensor("op_12527_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12527_cast_fp16 = einsum(equation = var_12527_equation_0, values = (var_12129_cast_fp16, var_12433_cast_fp16))[name = tensor("op_12527_cast_fp16")]; tensor var_12529_equation_0 = const()[name = tensor("op_12529_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12529_cast_fp16 = einsum(equation = var_12529_equation_0, values = (var_12129_cast_fp16, var_12434_cast_fp16))[name = tensor("op_12529_cast_fp16")]; tensor var_12531_equation_0 = const()[name = tensor("op_12531_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12531_cast_fp16 = einsum(equation = var_12531_equation_0, values = (var_12129_cast_fp16, var_12435_cast_fp16))[name = tensor("op_12531_cast_fp16")]; tensor var_12533_equation_0 = const()[name = tensor("op_12533_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12533_cast_fp16 = einsum(equation = var_12533_equation_0, values = (var_12133_cast_fp16, var_12436_cast_fp16))[name = tensor("op_12533_cast_fp16")]; tensor var_12535_equation_0 = const()[name = tensor("op_12535_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12535_cast_fp16 = einsum(equation = var_12535_equation_0, values = (var_12133_cast_fp16, var_12437_cast_fp16))[name = tensor("op_12535_cast_fp16")]; tensor var_12537_equation_0 = const()[name = tensor("op_12537_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12537_cast_fp16 = einsum(equation = var_12537_equation_0, values = (var_12133_cast_fp16, var_12438_cast_fp16))[name = tensor("op_12537_cast_fp16")]; tensor var_12539_equation_0 = const()[name = tensor("op_12539_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12539_cast_fp16 = einsum(equation = var_12539_equation_0, values = (var_12133_cast_fp16, var_12439_cast_fp16))[name = tensor("op_12539_cast_fp16")]; tensor var_12541_equation_0 = const()[name = tensor("op_12541_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12541_cast_fp16 = einsum(equation = var_12541_equation_0, values = (var_12133_cast_fp16, var_12440_cast_fp16))[name = tensor("op_12541_cast_fp16")]; tensor var_12543_equation_0 = const()[name = tensor("op_12543_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12543_cast_fp16 = einsum(equation = var_12543_equation_0, values = (var_12133_cast_fp16, var_12441_cast_fp16))[name = tensor("op_12543_cast_fp16")]; tensor var_12545_equation_0 = const()[name = tensor("op_12545_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12545_cast_fp16 = einsum(equation = var_12545_equation_0, values = (var_12133_cast_fp16, var_12442_cast_fp16))[name = tensor("op_12545_cast_fp16")]; tensor var_12547_equation_0 = const()[name = tensor("op_12547_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12547_cast_fp16 = einsum(equation = var_12547_equation_0, values = (var_12133_cast_fp16, var_12443_cast_fp16))[name = tensor("op_12547_cast_fp16")]; tensor var_12549_equation_0 = const()[name = tensor("op_12549_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12549_cast_fp16 = einsum(equation = var_12549_equation_0, values = (var_12137_cast_fp16, var_12444_cast_fp16))[name = tensor("op_12549_cast_fp16")]; tensor var_12551_equation_0 = const()[name = tensor("op_12551_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12551_cast_fp16 = einsum(equation = var_12551_equation_0, values = (var_12137_cast_fp16, var_12445_cast_fp16))[name = tensor("op_12551_cast_fp16")]; tensor var_12553_equation_0 = const()[name = tensor("op_12553_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12553_cast_fp16 = einsum(equation = var_12553_equation_0, values = (var_12137_cast_fp16, var_12446_cast_fp16))[name = tensor("op_12553_cast_fp16")]; tensor var_12555_equation_0 = const()[name = tensor("op_12555_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12555_cast_fp16 = einsum(equation = var_12555_equation_0, values = (var_12137_cast_fp16, var_12447_cast_fp16))[name = tensor("op_12555_cast_fp16")]; tensor var_12557_equation_0 = const()[name = tensor("op_12557_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12557_cast_fp16 = einsum(equation = var_12557_equation_0, values = (var_12137_cast_fp16, var_12448_cast_fp16))[name = tensor("op_12557_cast_fp16")]; tensor var_12559_equation_0 = const()[name = tensor("op_12559_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12559_cast_fp16 = einsum(equation = var_12559_equation_0, values = (var_12137_cast_fp16, var_12449_cast_fp16))[name = tensor("op_12559_cast_fp16")]; tensor var_12561_equation_0 = const()[name = tensor("op_12561_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12561_cast_fp16 = einsum(equation = var_12561_equation_0, values = (var_12137_cast_fp16, var_12450_cast_fp16))[name = tensor("op_12561_cast_fp16")]; tensor var_12563_equation_0 = const()[name = tensor("op_12563_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12563_cast_fp16 = einsum(equation = var_12563_equation_0, values = (var_12137_cast_fp16, var_12451_cast_fp16))[name = tensor("op_12563_cast_fp16")]; tensor var_12565_equation_0 = const()[name = tensor("op_12565_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12565_cast_fp16 = einsum(equation = var_12565_equation_0, values = (var_12141_cast_fp16, var_12452_cast_fp16))[name = tensor("op_12565_cast_fp16")]; tensor var_12567_equation_0 = const()[name = tensor("op_12567_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12567_cast_fp16 = einsum(equation = var_12567_equation_0, values = (var_12141_cast_fp16, var_12453_cast_fp16))[name = tensor("op_12567_cast_fp16")]; tensor var_12569_equation_0 = const()[name = tensor("op_12569_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12569_cast_fp16 = einsum(equation = var_12569_equation_0, values = (var_12141_cast_fp16, var_12454_cast_fp16))[name = tensor("op_12569_cast_fp16")]; tensor var_12571_equation_0 = const()[name = tensor("op_12571_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12571_cast_fp16 = einsum(equation = var_12571_equation_0, values = (var_12141_cast_fp16, var_12455_cast_fp16))[name = tensor("op_12571_cast_fp16")]; tensor var_12573_equation_0 = const()[name = tensor("op_12573_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12573_cast_fp16 = einsum(equation = var_12573_equation_0, values = (var_12141_cast_fp16, var_12456_cast_fp16))[name = tensor("op_12573_cast_fp16")]; tensor var_12575_equation_0 = const()[name = tensor("op_12575_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12575_cast_fp16 = einsum(equation = var_12575_equation_0, values = (var_12141_cast_fp16, var_12457_cast_fp16))[name = tensor("op_12575_cast_fp16")]; tensor var_12577_equation_0 = const()[name = tensor("op_12577_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12577_cast_fp16 = einsum(equation = var_12577_equation_0, values = (var_12141_cast_fp16, var_12458_cast_fp16))[name = tensor("op_12577_cast_fp16")]; tensor var_12579_equation_0 = const()[name = tensor("op_12579_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12579_cast_fp16 = einsum(equation = var_12579_equation_0, values = (var_12141_cast_fp16, var_12459_cast_fp16))[name = tensor("op_12579_cast_fp16")]; tensor var_12581_equation_0 = const()[name = tensor("op_12581_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12581_cast_fp16 = einsum(equation = var_12581_equation_0, values = (var_12145_cast_fp16, var_12460_cast_fp16))[name = tensor("op_12581_cast_fp16")]; tensor var_12583_equation_0 = const()[name = tensor("op_12583_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12583_cast_fp16 = einsum(equation = var_12583_equation_0, values = (var_12145_cast_fp16, var_12461_cast_fp16))[name = tensor("op_12583_cast_fp16")]; tensor var_12585_equation_0 = const()[name = tensor("op_12585_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12585_cast_fp16 = einsum(equation = var_12585_equation_0, values = (var_12145_cast_fp16, var_12462_cast_fp16))[name = tensor("op_12585_cast_fp16")]; tensor var_12587_equation_0 = const()[name = tensor("op_12587_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12587_cast_fp16 = einsum(equation = var_12587_equation_0, values = (var_12145_cast_fp16, var_12463_cast_fp16))[name = tensor("op_12587_cast_fp16")]; tensor var_12589_equation_0 = const()[name = tensor("op_12589_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12589_cast_fp16 = einsum(equation = var_12589_equation_0, values = (var_12145_cast_fp16, var_12464_cast_fp16))[name = tensor("op_12589_cast_fp16")]; tensor var_12591_equation_0 = const()[name = tensor("op_12591_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12591_cast_fp16 = einsum(equation = var_12591_equation_0, values = (var_12145_cast_fp16, var_12465_cast_fp16))[name = tensor("op_12591_cast_fp16")]; tensor var_12593_equation_0 = const()[name = tensor("op_12593_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12593_cast_fp16 = einsum(equation = var_12593_equation_0, values = (var_12145_cast_fp16, var_12466_cast_fp16))[name = tensor("op_12593_cast_fp16")]; tensor var_12595_equation_0 = const()[name = tensor("op_12595_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_12595_cast_fp16 = einsum(equation = var_12595_equation_0, values = (var_12145_cast_fp16, var_12467_cast_fp16))[name = tensor("op_12595_cast_fp16")]; tensor var_12597_interleave_0 = const()[name = tensor("op_12597_interleave_0"), val = tensor(false)]; tensor var_12597_cast_fp16 = concat(axis = var_10347, interleave = var_12597_interleave_0, values = (var_12469_cast_fp16, var_12471_cast_fp16, var_12473_cast_fp16, var_12475_cast_fp16, var_12477_cast_fp16, var_12479_cast_fp16, var_12481_cast_fp16, var_12483_cast_fp16))[name = tensor("op_12597_cast_fp16")]; tensor var_12599_interleave_0 = const()[name = tensor("op_12599_interleave_0"), val = tensor(false)]; tensor var_12599_cast_fp16 = concat(axis = var_10347, interleave = var_12599_interleave_0, values = (var_12485_cast_fp16, var_12487_cast_fp16, var_12489_cast_fp16, var_12491_cast_fp16, var_12493_cast_fp16, var_12495_cast_fp16, var_12497_cast_fp16, var_12499_cast_fp16))[name = tensor("op_12599_cast_fp16")]; tensor var_12601_interleave_0 = const()[name = tensor("op_12601_interleave_0"), val = tensor(false)]; tensor var_12601_cast_fp16 = concat(axis = var_10347, interleave = var_12601_interleave_0, values = (var_12501_cast_fp16, var_12503_cast_fp16, var_12505_cast_fp16, var_12507_cast_fp16, var_12509_cast_fp16, var_12511_cast_fp16, var_12513_cast_fp16, var_12515_cast_fp16))[name = tensor("op_12601_cast_fp16")]; tensor var_12603_interleave_0 = const()[name = tensor("op_12603_interleave_0"), val = tensor(false)]; tensor var_12603_cast_fp16 = concat(axis = var_10347, interleave = var_12603_interleave_0, values = (var_12517_cast_fp16, var_12519_cast_fp16, var_12521_cast_fp16, var_12523_cast_fp16, var_12525_cast_fp16, var_12527_cast_fp16, var_12529_cast_fp16, var_12531_cast_fp16))[name = tensor("op_12603_cast_fp16")]; tensor var_12605_interleave_0 = const()[name = tensor("op_12605_interleave_0"), val = tensor(false)]; tensor var_12605_cast_fp16 = concat(axis = var_10347, interleave = var_12605_interleave_0, values = (var_12533_cast_fp16, var_12535_cast_fp16, var_12537_cast_fp16, var_12539_cast_fp16, var_12541_cast_fp16, var_12543_cast_fp16, var_12545_cast_fp16, var_12547_cast_fp16))[name = tensor("op_12605_cast_fp16")]; tensor var_12607_interleave_0 = const()[name = tensor("op_12607_interleave_0"), val = tensor(false)]; tensor var_12607_cast_fp16 = concat(axis = var_10347, interleave = var_12607_interleave_0, values = (var_12549_cast_fp16, var_12551_cast_fp16, var_12553_cast_fp16, var_12555_cast_fp16, var_12557_cast_fp16, var_12559_cast_fp16, var_12561_cast_fp16, var_12563_cast_fp16))[name = tensor("op_12607_cast_fp16")]; tensor var_12609_interleave_0 = const()[name = tensor("op_12609_interleave_0"), val = tensor(false)]; tensor var_12609_cast_fp16 = concat(axis = var_10347, interleave = var_12609_interleave_0, values = (var_12565_cast_fp16, var_12567_cast_fp16, var_12569_cast_fp16, var_12571_cast_fp16, var_12573_cast_fp16, var_12575_cast_fp16, var_12577_cast_fp16, var_12579_cast_fp16))[name = tensor("op_12609_cast_fp16")]; tensor var_12611_interleave_0 = const()[name = tensor("op_12611_interleave_0"), val = tensor(false)]; tensor var_12611_cast_fp16 = concat(axis = var_10347, interleave = var_12611_interleave_0, values = (var_12581_cast_fp16, var_12583_cast_fp16, var_12585_cast_fp16, var_12587_cast_fp16, var_12589_cast_fp16, var_12591_cast_fp16, var_12593_cast_fp16, var_12595_cast_fp16))[name = tensor("op_12611_cast_fp16")]; tensor input_491_interleave_0 = const()[name = tensor("input_491_interleave_0"), val = tensor(false)]; tensor input_491_cast_fp16 = concat(axis = var_10375, interleave = input_491_interleave_0, values = (var_12597_cast_fp16, var_12599_cast_fp16, var_12601_cast_fp16, var_12603_cast_fp16, var_12605_cast_fp16, var_12607_cast_fp16, var_12609_cast_fp16, var_12611_cast_fp16))[name = tensor("input_491_cast_fp16")]; tensor var_12621_pad_type_0 = const()[name = tensor("op_12621_pad_type_0"), val = tensor("valid")]; tensor var_12621_strides_0 = const()[name = tensor("op_12621_strides_0"), val = tensor([1, 1])]; tensor var_12621_pad_0 = const()[name = tensor("op_12621_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_12621_dilations_0 = const()[name = tensor("op_12621_dilations_0"), val = tensor([1, 1])]; tensor var_12621_groups_0 = const()[name = tensor("op_12621_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(815412480)))]; tensor up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(815617344)))]; tensor var_12621_cast_fp16 = conv(bias = up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_12621_dilations_0, groups = var_12621_groups_0, pad = var_12621_pad_0, pad_type = var_12621_pad_type_0, strides = var_12621_strides_0, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_491_cast_fp16)[name = tensor("op_12621_cast_fp16")]; tensor inputs_87_cast_fp16 = add(x = var_12621_cast_fp16, y = inputs_85_cast_fp16)[name = tensor("inputs_87_cast_fp16")]; tensor hidden_states_309_axes_0 = const()[name = tensor("hidden_states_309_axes_0"), val = tensor([1])]; tensor hidden_states_309_gamma_0_to_fp16 = const()[name = tensor("hidden_states_309_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(815618048)))]; tensor hidden_states_309_beta_0_to_fp16 = const()[name = tensor("hidden_states_309_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(815618752)))]; tensor var_12631_to_fp16 = const()[name = tensor("op_12631_to_fp16"), val = tensor(0x1.5p-17)]; tensor hidden_states_309_cast_fp16 = layer_norm(axes = hidden_states_309_axes_0, beta = hidden_states_309_beta_0_to_fp16, epsilon = var_12631_to_fp16, gamma = hidden_states_309_gamma_0_to_fp16, x = inputs_87_cast_fp16)[name = tensor("hidden_states_309_cast_fp16")]; tensor q_59_pad_type_0 = const()[name = tensor("q_59_pad_type_0"), val = tensor("valid")]; tensor q_59_strides_0 = const()[name = tensor("q_59_strides_0"), val = tensor([1, 1])]; tensor q_59_pad_0 = const()[name = tensor("q_59_pad_0"), val = tensor([0, 0, 0, 0])]; tensor q_59_dilations_0 = const()[name = tensor("q_59_dilations_0"), val = tensor([1, 1])]; tensor q_59_groups_0 = const()[name = tensor("q_59_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(815619456)))]; tensor q_59_cast_fp16 = conv(dilations = q_59_dilations_0, groups = q_59_groups_0, pad = q_59_pad_0, pad_type = q_59_pad_type_0, strides = q_59_strides_0, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_309_cast_fp16)[name = tensor("q_59_cast_fp16")]; tensor k_117_pad_type_0 = const()[name = tensor("k_117_pad_type_0"), val = tensor("valid")]; tensor k_117_strides_0 = const()[name = tensor("k_117_strides_0"), val = tensor([1, 1])]; tensor k_117_pad_0 = const()[name = tensor("k_117_pad_0"), val = tensor([0, 0, 0, 0])]; tensor k_117_dilations_0 = const()[name = tensor("k_117_dilations_0"), val = tensor([1, 1])]; tensor k_117_groups_0 = const()[name = tensor("k_117_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(815824320)))]; tensor k_117_cast_fp16 = conv(dilations = k_117_dilations_0, groups = k_117_groups_0, pad = k_117_pad_0, pad_type = k_117_pad_type_0, strides = k_117_strides_0, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_117_cast_fp16")]; tensor v_59_pad_type_0 = const()[name = tensor("v_59_pad_type_0"), val = tensor("valid")]; tensor v_59_strides_0 = const()[name = tensor("v_59_strides_0"), val = tensor([1, 1])]; tensor v_59_pad_0 = const()[name = tensor("v_59_pad_0"), val = tensor([0, 0, 0, 0])]; tensor v_59_dilations_0 = const()[name = tensor("v_59_dilations_0"), val = tensor([1, 1])]; tensor v_59_groups_0 = const()[name = tensor("v_59_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(816315904)))]; tensor v_59_cast_fp16 = conv(dilations = v_59_dilations_0, groups = v_59_groups_0, pad = v_59_pad_0, pad_type = v_59_pad_type_0, strides = v_59_strides_0, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_59_cast_fp16")]; tensor var_12664_begin_0 = const()[name = tensor("op_12664_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_12664_end_0 = const()[name = tensor("op_12664_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_12664_end_mask_0 = const()[name = tensor("op_12664_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_12664_cast_fp16 = slice_by_index(begin = var_12664_begin_0, end = var_12664_end_0, end_mask = var_12664_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12664_cast_fp16")]; tensor var_12668_begin_0 = const()[name = tensor("op_12668_begin_0"), val = tensor([0, 40, 0, 0])]; tensor var_12668_end_0 = const()[name = tensor("op_12668_end_0"), val = tensor([2, 80, 1, 4096])]; tensor var_12668_end_mask_0 = const()[name = tensor("op_12668_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_12668_cast_fp16 = slice_by_index(begin = var_12668_begin_0, end = var_12668_end_0, end_mask = var_12668_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12668_cast_fp16")]; tensor var_12672_begin_0 = const()[name = tensor("op_12672_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_12672_end_0 = const()[name = tensor("op_12672_end_0"), val = tensor([2, 120, 1, 4096])]; tensor var_12672_end_mask_0 = const()[name = tensor("op_12672_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_12672_cast_fp16 = slice_by_index(begin = var_12672_begin_0, end = var_12672_end_0, end_mask = var_12672_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12672_cast_fp16")]; tensor var_12676_begin_0 = const()[name = tensor("op_12676_begin_0"), val = tensor([0, 120, 0, 0])]; tensor var_12676_end_0 = const()[name = tensor("op_12676_end_0"), val = tensor([2, 160, 1, 4096])]; tensor var_12676_end_mask_0 = const()[name = tensor("op_12676_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_12676_cast_fp16 = slice_by_index(begin = var_12676_begin_0, end = var_12676_end_0, end_mask = var_12676_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12676_cast_fp16")]; tensor var_12680_begin_0 = const()[name = tensor("op_12680_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_12680_end_0 = const()[name = tensor("op_12680_end_0"), val = tensor([2, 200, 1, 4096])]; tensor var_12680_end_mask_0 = const()[name = tensor("op_12680_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_12680_cast_fp16 = slice_by_index(begin = var_12680_begin_0, end = var_12680_end_0, end_mask = var_12680_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12680_cast_fp16")]; tensor var_12684_begin_0 = const()[name = tensor("op_12684_begin_0"), val = tensor([0, 200, 0, 0])]; tensor var_12684_end_0 = const()[name = tensor("op_12684_end_0"), val = tensor([2, 240, 1, 4096])]; tensor var_12684_end_mask_0 = const()[name = tensor("op_12684_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_12684_cast_fp16 = slice_by_index(begin = var_12684_begin_0, end = var_12684_end_0, end_mask = var_12684_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12684_cast_fp16")]; tensor var_12688_begin_0 = const()[name = tensor("op_12688_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_12688_end_0 = const()[name = tensor("op_12688_end_0"), val = tensor([2, 280, 1, 4096])]; tensor var_12688_end_mask_0 = const()[name = tensor("op_12688_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_12688_cast_fp16 = slice_by_index(begin = var_12688_begin_0, end = var_12688_end_0, end_mask = var_12688_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12688_cast_fp16")]; tensor var_12692_begin_0 = const()[name = tensor("op_12692_begin_0"), val = tensor([0, 280, 0, 0])]; tensor var_12692_end_0 = const()[name = tensor("op_12692_end_0"), val = tensor([2, 1, 1, 4096])]; tensor var_12692_end_mask_0 = const()[name = tensor("op_12692_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_12692_cast_fp16 = slice_by_index(begin = var_12692_begin_0, end = var_12692_end_0, end_mask = var_12692_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12692_cast_fp16")]; tensor var_12695_begin_0 = const()[name = tensor("op_12695_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_12695_end_0 = const()[name = tensor("op_12695_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_12695_end_mask_0 = const()[name = tensor("op_12695_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12695_cast_fp16 = slice_by_index(begin = var_12695_begin_0, end = var_12695_end_0, end_mask = var_12695_end_mask_0, x = var_12664_cast_fp16)[name = tensor("op_12695_cast_fp16")]; tensor var_12696_begin_0 = const()[name = tensor("op_12696_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_12696_end_0 = const()[name = tensor("op_12696_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_12696_end_mask_0 = const()[name = tensor("op_12696_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12696_cast_fp16 = slice_by_index(begin = var_12696_begin_0, end = var_12696_end_0, end_mask = var_12696_end_mask_0, x = var_12664_cast_fp16)[name = tensor("op_12696_cast_fp16")]; tensor var_12697_begin_0 = const()[name = tensor("op_12697_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_12697_end_0 = const()[name = tensor("op_12697_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_12697_end_mask_0 = const()[name = tensor("op_12697_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12697_cast_fp16 = slice_by_index(begin = var_12697_begin_0, end = var_12697_end_0, end_mask = var_12697_end_mask_0, x = var_12664_cast_fp16)[name = tensor("op_12697_cast_fp16")]; tensor var_12698_begin_0 = const()[name = tensor("op_12698_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_12698_end_0 = const()[name = tensor("op_12698_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_12698_end_mask_0 = const()[name = tensor("op_12698_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12698_cast_fp16 = slice_by_index(begin = var_12698_begin_0, end = var_12698_end_0, end_mask = var_12698_end_mask_0, x = var_12664_cast_fp16)[name = tensor("op_12698_cast_fp16")]; tensor var_12699_begin_0 = const()[name = tensor("op_12699_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_12699_end_0 = const()[name = tensor("op_12699_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_12699_end_mask_0 = const()[name = tensor("op_12699_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12699_cast_fp16 = slice_by_index(begin = var_12699_begin_0, end = var_12699_end_0, end_mask = var_12699_end_mask_0, x = var_12664_cast_fp16)[name = tensor("op_12699_cast_fp16")]; tensor var_12700_begin_0 = const()[name = tensor("op_12700_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_12700_end_0 = const()[name = tensor("op_12700_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_12700_end_mask_0 = const()[name = tensor("op_12700_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12700_cast_fp16 = slice_by_index(begin = var_12700_begin_0, end = var_12700_end_0, end_mask = var_12700_end_mask_0, x = var_12664_cast_fp16)[name = tensor("op_12700_cast_fp16")]; tensor var_12701_begin_0 = const()[name = tensor("op_12701_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_12701_end_0 = const()[name = tensor("op_12701_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_12701_end_mask_0 = const()[name = tensor("op_12701_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12701_cast_fp16 = slice_by_index(begin = var_12701_begin_0, end = var_12701_end_0, end_mask = var_12701_end_mask_0, x = var_12664_cast_fp16)[name = tensor("op_12701_cast_fp16")]; tensor var_12702_begin_0 = const()[name = tensor("op_12702_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_12702_end_0 = const()[name = tensor("op_12702_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_12702_end_mask_0 = const()[name = tensor("op_12702_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_12702_cast_fp16 = slice_by_index(begin = var_12702_begin_0, end = var_12702_end_0, end_mask = var_12702_end_mask_0, x = var_12664_cast_fp16)[name = tensor("op_12702_cast_fp16")]; tensor var_12703_begin_0 = const()[name = tensor("op_12703_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_12703_end_0 = const()[name = tensor("op_12703_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_12703_end_mask_0 = const()[name = tensor("op_12703_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12703_cast_fp16 = slice_by_index(begin = var_12703_begin_0, end = var_12703_end_0, end_mask = var_12703_end_mask_0, x = var_12668_cast_fp16)[name = tensor("op_12703_cast_fp16")]; tensor var_12704_begin_0 = const()[name = tensor("op_12704_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_12704_end_0 = const()[name = tensor("op_12704_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_12704_end_mask_0 = const()[name = tensor("op_12704_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12704_cast_fp16 = slice_by_index(begin = var_12704_begin_0, end = var_12704_end_0, end_mask = var_12704_end_mask_0, x = var_12668_cast_fp16)[name = tensor("op_12704_cast_fp16")]; tensor var_12705_begin_0 = const()[name = tensor("op_12705_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_12705_end_0 = const()[name = tensor("op_12705_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_12705_end_mask_0 = const()[name = tensor("op_12705_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12705_cast_fp16 = slice_by_index(begin = var_12705_begin_0, end = var_12705_end_0, end_mask = var_12705_end_mask_0, x = var_12668_cast_fp16)[name = tensor("op_12705_cast_fp16")]; tensor var_12706_begin_0 = const()[name = tensor("op_12706_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_12706_end_0 = const()[name = tensor("op_12706_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_12706_end_mask_0 = const()[name = tensor("op_12706_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12706_cast_fp16 = slice_by_index(begin = var_12706_begin_0, end = var_12706_end_0, end_mask = var_12706_end_mask_0, x = var_12668_cast_fp16)[name = tensor("op_12706_cast_fp16")]; tensor var_12707_begin_0 = const()[name = tensor("op_12707_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_12707_end_0 = const()[name = tensor("op_12707_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_12707_end_mask_0 = const()[name = tensor("op_12707_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12707_cast_fp16 = slice_by_index(begin = var_12707_begin_0, end = var_12707_end_0, end_mask = var_12707_end_mask_0, x = var_12668_cast_fp16)[name = tensor("op_12707_cast_fp16")]; tensor var_12708_begin_0 = const()[name = tensor("op_12708_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_12708_end_0 = const()[name = tensor("op_12708_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_12708_end_mask_0 = const()[name = tensor("op_12708_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12708_cast_fp16 = slice_by_index(begin = var_12708_begin_0, end = var_12708_end_0, end_mask = var_12708_end_mask_0, x = var_12668_cast_fp16)[name = tensor("op_12708_cast_fp16")]; tensor var_12709_begin_0 = const()[name = tensor("op_12709_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_12709_end_0 = const()[name = tensor("op_12709_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_12709_end_mask_0 = const()[name = tensor("op_12709_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12709_cast_fp16 = slice_by_index(begin = var_12709_begin_0, end = var_12709_end_0, end_mask = var_12709_end_mask_0, x = var_12668_cast_fp16)[name = tensor("op_12709_cast_fp16")]; tensor var_12710_begin_0 = const()[name = tensor("op_12710_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_12710_end_0 = const()[name = tensor("op_12710_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_12710_end_mask_0 = const()[name = tensor("op_12710_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_12710_cast_fp16 = slice_by_index(begin = var_12710_begin_0, end = var_12710_end_0, end_mask = var_12710_end_mask_0, x = var_12668_cast_fp16)[name = tensor("op_12710_cast_fp16")]; tensor var_12711_begin_0 = const()[name = tensor("op_12711_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_12711_end_0 = const()[name = tensor("op_12711_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_12711_end_mask_0 = const()[name = tensor("op_12711_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12711_cast_fp16 = slice_by_index(begin = var_12711_begin_0, end = var_12711_end_0, end_mask = var_12711_end_mask_0, x = var_12672_cast_fp16)[name = tensor("op_12711_cast_fp16")]; tensor var_12712_begin_0 = const()[name = tensor("op_12712_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_12712_end_0 = const()[name = tensor("op_12712_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_12712_end_mask_0 = const()[name = tensor("op_12712_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12712_cast_fp16 = slice_by_index(begin = var_12712_begin_0, end = var_12712_end_0, end_mask = var_12712_end_mask_0, x = var_12672_cast_fp16)[name = tensor("op_12712_cast_fp16")]; tensor var_12713_begin_0 = const()[name = tensor("op_12713_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_12713_end_0 = const()[name = tensor("op_12713_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_12713_end_mask_0 = const()[name = tensor("op_12713_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12713_cast_fp16 = slice_by_index(begin = var_12713_begin_0, end = var_12713_end_0, end_mask = var_12713_end_mask_0, x = var_12672_cast_fp16)[name = tensor("op_12713_cast_fp16")]; tensor var_12714_begin_0 = const()[name = tensor("op_12714_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_12714_end_0 = const()[name = tensor("op_12714_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_12714_end_mask_0 = const()[name = tensor("op_12714_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12714_cast_fp16 = slice_by_index(begin = var_12714_begin_0, end = var_12714_end_0, end_mask = var_12714_end_mask_0, x = var_12672_cast_fp16)[name = tensor("op_12714_cast_fp16")]; tensor var_12715_begin_0 = const()[name = tensor("op_12715_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_12715_end_0 = const()[name = tensor("op_12715_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_12715_end_mask_0 = const()[name = tensor("op_12715_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12715_cast_fp16 = slice_by_index(begin = var_12715_begin_0, end = var_12715_end_0, end_mask = var_12715_end_mask_0, x = var_12672_cast_fp16)[name = tensor("op_12715_cast_fp16")]; tensor var_12716_begin_0 = const()[name = tensor("op_12716_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_12716_end_0 = const()[name = tensor("op_12716_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_12716_end_mask_0 = const()[name = tensor("op_12716_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12716_cast_fp16 = slice_by_index(begin = var_12716_begin_0, end = var_12716_end_0, end_mask = var_12716_end_mask_0, x = var_12672_cast_fp16)[name = tensor("op_12716_cast_fp16")]; tensor var_12717_begin_0 = const()[name = tensor("op_12717_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_12717_end_0 = const()[name = tensor("op_12717_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_12717_end_mask_0 = const()[name = tensor("op_12717_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12717_cast_fp16 = slice_by_index(begin = var_12717_begin_0, end = var_12717_end_0, end_mask = var_12717_end_mask_0, x = var_12672_cast_fp16)[name = tensor("op_12717_cast_fp16")]; tensor var_12718_begin_0 = const()[name = tensor("op_12718_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_12718_end_0 = const()[name = tensor("op_12718_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_12718_end_mask_0 = const()[name = tensor("op_12718_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_12718_cast_fp16 = slice_by_index(begin = var_12718_begin_0, end = var_12718_end_0, end_mask = var_12718_end_mask_0, x = var_12672_cast_fp16)[name = tensor("op_12718_cast_fp16")]; tensor var_12719_begin_0 = const()[name = tensor("op_12719_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_12719_end_0 = const()[name = tensor("op_12719_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_12719_end_mask_0 = const()[name = tensor("op_12719_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12719_cast_fp16 = slice_by_index(begin = var_12719_begin_0, end = var_12719_end_0, end_mask = var_12719_end_mask_0, x = var_12676_cast_fp16)[name = tensor("op_12719_cast_fp16")]; tensor var_12720_begin_0 = const()[name = tensor("op_12720_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_12720_end_0 = const()[name = tensor("op_12720_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_12720_end_mask_0 = const()[name = tensor("op_12720_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12720_cast_fp16 = slice_by_index(begin = var_12720_begin_0, end = var_12720_end_0, end_mask = var_12720_end_mask_0, x = var_12676_cast_fp16)[name = tensor("op_12720_cast_fp16")]; tensor var_12721_begin_0 = const()[name = tensor("op_12721_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_12721_end_0 = const()[name = tensor("op_12721_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_12721_end_mask_0 = const()[name = tensor("op_12721_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12721_cast_fp16 = slice_by_index(begin = var_12721_begin_0, end = var_12721_end_0, end_mask = var_12721_end_mask_0, x = var_12676_cast_fp16)[name = tensor("op_12721_cast_fp16")]; tensor var_12722_begin_0 = const()[name = tensor("op_12722_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_12722_end_0 = const()[name = tensor("op_12722_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_12722_end_mask_0 = const()[name = tensor("op_12722_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12722_cast_fp16 = slice_by_index(begin = var_12722_begin_0, end = var_12722_end_0, end_mask = var_12722_end_mask_0, x = var_12676_cast_fp16)[name = tensor("op_12722_cast_fp16")]; tensor var_12723_begin_0 = const()[name = tensor("op_12723_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_12723_end_0 = const()[name = tensor("op_12723_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_12723_end_mask_0 = const()[name = tensor("op_12723_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12723_cast_fp16 = slice_by_index(begin = var_12723_begin_0, end = var_12723_end_0, end_mask = var_12723_end_mask_0, x = var_12676_cast_fp16)[name = tensor("op_12723_cast_fp16")]; tensor var_12724_begin_0 = const()[name = tensor("op_12724_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_12724_end_0 = const()[name = tensor("op_12724_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_12724_end_mask_0 = const()[name = tensor("op_12724_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12724_cast_fp16 = slice_by_index(begin = var_12724_begin_0, end = var_12724_end_0, end_mask = var_12724_end_mask_0, x = var_12676_cast_fp16)[name = tensor("op_12724_cast_fp16")]; tensor var_12725_begin_0 = const()[name = tensor("op_12725_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_12725_end_0 = const()[name = tensor("op_12725_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_12725_end_mask_0 = const()[name = tensor("op_12725_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12725_cast_fp16 = slice_by_index(begin = var_12725_begin_0, end = var_12725_end_0, end_mask = var_12725_end_mask_0, x = var_12676_cast_fp16)[name = tensor("op_12725_cast_fp16")]; tensor var_12726_begin_0 = const()[name = tensor("op_12726_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_12726_end_0 = const()[name = tensor("op_12726_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_12726_end_mask_0 = const()[name = tensor("op_12726_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_12726_cast_fp16 = slice_by_index(begin = var_12726_begin_0, end = var_12726_end_0, end_mask = var_12726_end_mask_0, x = var_12676_cast_fp16)[name = tensor("op_12726_cast_fp16")]; tensor var_12727_begin_0 = const()[name = tensor("op_12727_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_12727_end_0 = const()[name = tensor("op_12727_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_12727_end_mask_0 = const()[name = tensor("op_12727_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12727_cast_fp16 = slice_by_index(begin = var_12727_begin_0, end = var_12727_end_0, end_mask = var_12727_end_mask_0, x = var_12680_cast_fp16)[name = tensor("op_12727_cast_fp16")]; tensor var_12728_begin_0 = const()[name = tensor("op_12728_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_12728_end_0 = const()[name = tensor("op_12728_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_12728_end_mask_0 = const()[name = tensor("op_12728_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12728_cast_fp16 = slice_by_index(begin = var_12728_begin_0, end = var_12728_end_0, end_mask = var_12728_end_mask_0, x = var_12680_cast_fp16)[name = tensor("op_12728_cast_fp16")]; tensor var_12729_begin_0 = const()[name = tensor("op_12729_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_12729_end_0 = const()[name = tensor("op_12729_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_12729_end_mask_0 = const()[name = tensor("op_12729_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12729_cast_fp16 = slice_by_index(begin = var_12729_begin_0, end = var_12729_end_0, end_mask = var_12729_end_mask_0, x = var_12680_cast_fp16)[name = tensor("op_12729_cast_fp16")]; tensor var_12730_begin_0 = const()[name = tensor("op_12730_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_12730_end_0 = const()[name = tensor("op_12730_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_12730_end_mask_0 = const()[name = tensor("op_12730_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12730_cast_fp16 = slice_by_index(begin = var_12730_begin_0, end = var_12730_end_0, end_mask = var_12730_end_mask_0, x = var_12680_cast_fp16)[name = tensor("op_12730_cast_fp16")]; tensor var_12731_begin_0 = const()[name = tensor("op_12731_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_12731_end_0 = const()[name = tensor("op_12731_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_12731_end_mask_0 = const()[name = tensor("op_12731_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12731_cast_fp16 = slice_by_index(begin = var_12731_begin_0, end = var_12731_end_0, end_mask = var_12731_end_mask_0, x = var_12680_cast_fp16)[name = tensor("op_12731_cast_fp16")]; tensor var_12732_begin_0 = const()[name = tensor("op_12732_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_12732_end_0 = const()[name = tensor("op_12732_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_12732_end_mask_0 = const()[name = tensor("op_12732_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12732_cast_fp16 = slice_by_index(begin = var_12732_begin_0, end = var_12732_end_0, end_mask = var_12732_end_mask_0, x = var_12680_cast_fp16)[name = tensor("op_12732_cast_fp16")]; tensor var_12733_begin_0 = const()[name = tensor("op_12733_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_12733_end_0 = const()[name = tensor("op_12733_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_12733_end_mask_0 = const()[name = tensor("op_12733_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12733_cast_fp16 = slice_by_index(begin = var_12733_begin_0, end = var_12733_end_0, end_mask = var_12733_end_mask_0, x = var_12680_cast_fp16)[name = tensor("op_12733_cast_fp16")]; tensor var_12734_begin_0 = const()[name = tensor("op_12734_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_12734_end_0 = const()[name = tensor("op_12734_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_12734_end_mask_0 = const()[name = tensor("op_12734_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_12734_cast_fp16 = slice_by_index(begin = var_12734_begin_0, end = var_12734_end_0, end_mask = var_12734_end_mask_0, x = var_12680_cast_fp16)[name = tensor("op_12734_cast_fp16")]; tensor var_12735_begin_0 = const()[name = tensor("op_12735_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_12735_end_0 = const()[name = tensor("op_12735_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_12735_end_mask_0 = const()[name = tensor("op_12735_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12735_cast_fp16 = slice_by_index(begin = var_12735_begin_0, end = var_12735_end_0, end_mask = var_12735_end_mask_0, x = var_12684_cast_fp16)[name = tensor("op_12735_cast_fp16")]; tensor var_12736_begin_0 = const()[name = tensor("op_12736_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_12736_end_0 = const()[name = tensor("op_12736_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_12736_end_mask_0 = const()[name = tensor("op_12736_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12736_cast_fp16 = slice_by_index(begin = var_12736_begin_0, end = var_12736_end_0, end_mask = var_12736_end_mask_0, x = var_12684_cast_fp16)[name = tensor("op_12736_cast_fp16")]; tensor var_12737_begin_0 = const()[name = tensor("op_12737_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_12737_end_0 = const()[name = tensor("op_12737_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_12737_end_mask_0 = const()[name = tensor("op_12737_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12737_cast_fp16 = slice_by_index(begin = var_12737_begin_0, end = var_12737_end_0, end_mask = var_12737_end_mask_0, x = var_12684_cast_fp16)[name = tensor("op_12737_cast_fp16")]; tensor var_12738_begin_0 = const()[name = tensor("op_12738_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_12738_end_0 = const()[name = tensor("op_12738_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_12738_end_mask_0 = const()[name = tensor("op_12738_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12738_cast_fp16 = slice_by_index(begin = var_12738_begin_0, end = var_12738_end_0, end_mask = var_12738_end_mask_0, x = var_12684_cast_fp16)[name = tensor("op_12738_cast_fp16")]; tensor var_12739_begin_0 = const()[name = tensor("op_12739_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_12739_end_0 = const()[name = tensor("op_12739_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_12739_end_mask_0 = const()[name = tensor("op_12739_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12739_cast_fp16 = slice_by_index(begin = var_12739_begin_0, end = var_12739_end_0, end_mask = var_12739_end_mask_0, x = var_12684_cast_fp16)[name = tensor("op_12739_cast_fp16")]; tensor var_12740_begin_0 = const()[name = tensor("op_12740_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_12740_end_0 = const()[name = tensor("op_12740_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_12740_end_mask_0 = const()[name = tensor("op_12740_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12740_cast_fp16 = slice_by_index(begin = var_12740_begin_0, end = var_12740_end_0, end_mask = var_12740_end_mask_0, x = var_12684_cast_fp16)[name = tensor("op_12740_cast_fp16")]; tensor var_12741_begin_0 = const()[name = tensor("op_12741_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_12741_end_0 = const()[name = tensor("op_12741_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_12741_end_mask_0 = const()[name = tensor("op_12741_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12741_cast_fp16 = slice_by_index(begin = var_12741_begin_0, end = var_12741_end_0, end_mask = var_12741_end_mask_0, x = var_12684_cast_fp16)[name = tensor("op_12741_cast_fp16")]; tensor var_12742_begin_0 = const()[name = tensor("op_12742_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_12742_end_0 = const()[name = tensor("op_12742_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_12742_end_mask_0 = const()[name = tensor("op_12742_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_12742_cast_fp16 = slice_by_index(begin = var_12742_begin_0, end = var_12742_end_0, end_mask = var_12742_end_mask_0, x = var_12684_cast_fp16)[name = tensor("op_12742_cast_fp16")]; tensor var_12743_begin_0 = const()[name = tensor("op_12743_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_12743_end_0 = const()[name = tensor("op_12743_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_12743_end_mask_0 = const()[name = tensor("op_12743_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12743_cast_fp16 = slice_by_index(begin = var_12743_begin_0, end = var_12743_end_0, end_mask = var_12743_end_mask_0, x = var_12688_cast_fp16)[name = tensor("op_12743_cast_fp16")]; tensor var_12744_begin_0 = const()[name = tensor("op_12744_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_12744_end_0 = const()[name = tensor("op_12744_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_12744_end_mask_0 = const()[name = tensor("op_12744_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12744_cast_fp16 = slice_by_index(begin = var_12744_begin_0, end = var_12744_end_0, end_mask = var_12744_end_mask_0, x = var_12688_cast_fp16)[name = tensor("op_12744_cast_fp16")]; tensor var_12745_begin_0 = const()[name = tensor("op_12745_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_12745_end_0 = const()[name = tensor("op_12745_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_12745_end_mask_0 = const()[name = tensor("op_12745_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12745_cast_fp16 = slice_by_index(begin = var_12745_begin_0, end = var_12745_end_0, end_mask = var_12745_end_mask_0, x = var_12688_cast_fp16)[name = tensor("op_12745_cast_fp16")]; tensor var_12746_begin_0 = const()[name = tensor("op_12746_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_12746_end_0 = const()[name = tensor("op_12746_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_12746_end_mask_0 = const()[name = tensor("op_12746_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12746_cast_fp16 = slice_by_index(begin = var_12746_begin_0, end = var_12746_end_0, end_mask = var_12746_end_mask_0, x = var_12688_cast_fp16)[name = tensor("op_12746_cast_fp16")]; tensor var_12747_begin_0 = const()[name = tensor("op_12747_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_12747_end_0 = const()[name = tensor("op_12747_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_12747_end_mask_0 = const()[name = tensor("op_12747_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12747_cast_fp16 = slice_by_index(begin = var_12747_begin_0, end = var_12747_end_0, end_mask = var_12747_end_mask_0, x = var_12688_cast_fp16)[name = tensor("op_12747_cast_fp16")]; tensor var_12748_begin_0 = const()[name = tensor("op_12748_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_12748_end_0 = const()[name = tensor("op_12748_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_12748_end_mask_0 = const()[name = tensor("op_12748_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12748_cast_fp16 = slice_by_index(begin = var_12748_begin_0, end = var_12748_end_0, end_mask = var_12748_end_mask_0, x = var_12688_cast_fp16)[name = tensor("op_12748_cast_fp16")]; tensor var_12749_begin_0 = const()[name = tensor("op_12749_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_12749_end_0 = const()[name = tensor("op_12749_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_12749_end_mask_0 = const()[name = tensor("op_12749_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12749_cast_fp16 = slice_by_index(begin = var_12749_begin_0, end = var_12749_end_0, end_mask = var_12749_end_mask_0, x = var_12688_cast_fp16)[name = tensor("op_12749_cast_fp16")]; tensor var_12750_begin_0 = const()[name = tensor("op_12750_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_12750_end_0 = const()[name = tensor("op_12750_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_12750_end_mask_0 = const()[name = tensor("op_12750_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_12750_cast_fp16 = slice_by_index(begin = var_12750_begin_0, end = var_12750_end_0, end_mask = var_12750_end_mask_0, x = var_12688_cast_fp16)[name = tensor("op_12750_cast_fp16")]; tensor var_12751_begin_0 = const()[name = tensor("op_12751_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_12751_end_0 = const()[name = tensor("op_12751_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_12751_end_mask_0 = const()[name = tensor("op_12751_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12751_cast_fp16 = slice_by_index(begin = var_12751_begin_0, end = var_12751_end_0, end_mask = var_12751_end_mask_0, x = var_12692_cast_fp16)[name = tensor("op_12751_cast_fp16")]; tensor var_12752_begin_0 = const()[name = tensor("op_12752_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_12752_end_0 = const()[name = tensor("op_12752_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_12752_end_mask_0 = const()[name = tensor("op_12752_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12752_cast_fp16 = slice_by_index(begin = var_12752_begin_0, end = var_12752_end_0, end_mask = var_12752_end_mask_0, x = var_12692_cast_fp16)[name = tensor("op_12752_cast_fp16")]; tensor var_12753_begin_0 = const()[name = tensor("op_12753_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_12753_end_0 = const()[name = tensor("op_12753_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_12753_end_mask_0 = const()[name = tensor("op_12753_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12753_cast_fp16 = slice_by_index(begin = var_12753_begin_0, end = var_12753_end_0, end_mask = var_12753_end_mask_0, x = var_12692_cast_fp16)[name = tensor("op_12753_cast_fp16")]; tensor var_12754_begin_0 = const()[name = tensor("op_12754_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_12754_end_0 = const()[name = tensor("op_12754_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_12754_end_mask_0 = const()[name = tensor("op_12754_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12754_cast_fp16 = slice_by_index(begin = var_12754_begin_0, end = var_12754_end_0, end_mask = var_12754_end_mask_0, x = var_12692_cast_fp16)[name = tensor("op_12754_cast_fp16")]; tensor var_12755_begin_0 = const()[name = tensor("op_12755_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_12755_end_0 = const()[name = tensor("op_12755_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_12755_end_mask_0 = const()[name = tensor("op_12755_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12755_cast_fp16 = slice_by_index(begin = var_12755_begin_0, end = var_12755_end_0, end_mask = var_12755_end_mask_0, x = var_12692_cast_fp16)[name = tensor("op_12755_cast_fp16")]; tensor var_12756_begin_0 = const()[name = tensor("op_12756_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_12756_end_0 = const()[name = tensor("op_12756_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_12756_end_mask_0 = const()[name = tensor("op_12756_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12756_cast_fp16 = slice_by_index(begin = var_12756_begin_0, end = var_12756_end_0, end_mask = var_12756_end_mask_0, x = var_12692_cast_fp16)[name = tensor("op_12756_cast_fp16")]; tensor var_12757_begin_0 = const()[name = tensor("op_12757_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_12757_end_0 = const()[name = tensor("op_12757_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_12757_end_mask_0 = const()[name = tensor("op_12757_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12757_cast_fp16 = slice_by_index(begin = var_12757_begin_0, end = var_12757_end_0, end_mask = var_12757_end_mask_0, x = var_12692_cast_fp16)[name = tensor("op_12757_cast_fp16")]; tensor var_12758_begin_0 = const()[name = tensor("op_12758_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_12758_end_0 = const()[name = tensor("op_12758_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_12758_end_mask_0 = const()[name = tensor("op_12758_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_12758_cast_fp16 = slice_by_index(begin = var_12758_begin_0, end = var_12758_end_0, end_mask = var_12758_end_mask_0, x = var_12692_cast_fp16)[name = tensor("op_12758_cast_fp16")]; tensor k_119_perm_0 = const()[name = tensor("k_119_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_12763_begin_0 = const()[name = tensor("op_12763_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_12763_end_0 = const()[name = tensor("op_12763_end_0"), val = tensor([2, 77, 1, 40])]; tensor var_12763_end_mask_0 = const()[name = tensor("op_12763_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_119_cast_fp16 = transpose(perm = k_119_perm_0, x = k_117_cast_fp16)[name = tensor("transpose_2")]; tensor var_12763_cast_fp16 = slice_by_index(begin = var_12763_begin_0, end = var_12763_end_0, end_mask = var_12763_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12763_cast_fp16")]; tensor var_12767_begin_0 = const()[name = tensor("op_12767_begin_0"), val = tensor([0, 0, 0, 40])]; tensor var_12767_end_0 = const()[name = tensor("op_12767_end_0"), val = tensor([2, 77, 1, 80])]; tensor var_12767_end_mask_0 = const()[name = tensor("op_12767_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12767_cast_fp16 = slice_by_index(begin = var_12767_begin_0, end = var_12767_end_0, end_mask = var_12767_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12767_cast_fp16")]; tensor var_12771_begin_0 = const()[name = tensor("op_12771_begin_0"), val = tensor([0, 0, 0, 80])]; tensor var_12771_end_0 = const()[name = tensor("op_12771_end_0"), val = tensor([2, 77, 1, 120])]; tensor var_12771_end_mask_0 = const()[name = tensor("op_12771_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12771_cast_fp16 = slice_by_index(begin = var_12771_begin_0, end = var_12771_end_0, end_mask = var_12771_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12771_cast_fp16")]; tensor var_12775_begin_0 = const()[name = tensor("op_12775_begin_0"), val = tensor([0, 0, 0, 120])]; tensor var_12775_end_0 = const()[name = tensor("op_12775_end_0"), val = tensor([2, 77, 1, 160])]; tensor var_12775_end_mask_0 = const()[name = tensor("op_12775_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12775_cast_fp16 = slice_by_index(begin = var_12775_begin_0, end = var_12775_end_0, end_mask = var_12775_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12775_cast_fp16")]; tensor var_12779_begin_0 = const()[name = tensor("op_12779_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_12779_end_0 = const()[name = tensor("op_12779_end_0"), val = tensor([2, 77, 1, 200])]; tensor var_12779_end_mask_0 = const()[name = tensor("op_12779_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12779_cast_fp16 = slice_by_index(begin = var_12779_begin_0, end = var_12779_end_0, end_mask = var_12779_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12779_cast_fp16")]; tensor var_12783_begin_0 = const()[name = tensor("op_12783_begin_0"), val = tensor([0, 0, 0, 200])]; tensor var_12783_end_0 = const()[name = tensor("op_12783_end_0"), val = tensor([2, 77, 1, 240])]; tensor var_12783_end_mask_0 = const()[name = tensor("op_12783_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12783_cast_fp16 = slice_by_index(begin = var_12783_begin_0, end = var_12783_end_0, end_mask = var_12783_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12783_cast_fp16")]; tensor var_12787_begin_0 = const()[name = tensor("op_12787_begin_0"), val = tensor([0, 0, 0, 240])]; tensor var_12787_end_0 = const()[name = tensor("op_12787_end_0"), val = tensor([2, 77, 1, 280])]; tensor var_12787_end_mask_0 = const()[name = tensor("op_12787_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_12787_cast_fp16 = slice_by_index(begin = var_12787_begin_0, end = var_12787_end_0, end_mask = var_12787_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12787_cast_fp16")]; tensor var_12791_begin_0 = const()[name = tensor("op_12791_begin_0"), val = tensor([0, 0, 0, 280])]; tensor var_12791_end_0 = const()[name = tensor("op_12791_end_0"), val = tensor([2, 77, 1, 1])]; tensor var_12791_end_mask_0 = const()[name = tensor("op_12791_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_12791_cast_fp16 = slice_by_index(begin = var_12791_begin_0, end = var_12791_end_0, end_mask = var_12791_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12791_cast_fp16")]; tensor var_12793_begin_0 = const()[name = tensor("op_12793_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_12793_end_0 = const()[name = tensor("op_12793_end_0"), val = tensor([2, 40, 1, 77])]; tensor var_12793_end_mask_0 = const()[name = tensor("op_12793_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_12793_cast_fp16 = slice_by_index(begin = var_12793_begin_0, end = var_12793_end_0, end_mask = var_12793_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12793_cast_fp16")]; tensor var_12797_begin_0 = const()[name = tensor("op_12797_begin_0"), val = tensor([0, 40, 0, 0])]; tensor var_12797_end_0 = const()[name = tensor("op_12797_end_0"), val = tensor([2, 80, 1, 77])]; tensor var_12797_end_mask_0 = const()[name = tensor("op_12797_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_12797_cast_fp16 = slice_by_index(begin = var_12797_begin_0, end = var_12797_end_0, end_mask = var_12797_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12797_cast_fp16")]; tensor var_12801_begin_0 = const()[name = tensor("op_12801_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_12801_end_0 = const()[name = tensor("op_12801_end_0"), val = tensor([2, 120, 1, 77])]; tensor var_12801_end_mask_0 = const()[name = tensor("op_12801_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_12801_cast_fp16 = slice_by_index(begin = var_12801_begin_0, end = var_12801_end_0, end_mask = var_12801_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12801_cast_fp16")]; tensor var_12805_begin_0 = const()[name = tensor("op_12805_begin_0"), val = tensor([0, 120, 0, 0])]; tensor var_12805_end_0 = const()[name = tensor("op_12805_end_0"), val = tensor([2, 160, 1, 77])]; tensor var_12805_end_mask_0 = const()[name = tensor("op_12805_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_12805_cast_fp16 = slice_by_index(begin = var_12805_begin_0, end = var_12805_end_0, end_mask = var_12805_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12805_cast_fp16")]; tensor var_12809_begin_0 = const()[name = tensor("op_12809_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_12809_end_0 = const()[name = tensor("op_12809_end_0"), val = tensor([2, 200, 1, 77])]; tensor var_12809_end_mask_0 = const()[name = tensor("op_12809_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_12809_cast_fp16 = slice_by_index(begin = var_12809_begin_0, end = var_12809_end_0, end_mask = var_12809_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12809_cast_fp16")]; tensor var_12813_begin_0 = const()[name = tensor("op_12813_begin_0"), val = tensor([0, 200, 0, 0])]; tensor var_12813_end_0 = const()[name = tensor("op_12813_end_0"), val = tensor([2, 240, 1, 77])]; tensor var_12813_end_mask_0 = const()[name = tensor("op_12813_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_12813_cast_fp16 = slice_by_index(begin = var_12813_begin_0, end = var_12813_end_0, end_mask = var_12813_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12813_cast_fp16")]; tensor var_12817_begin_0 = const()[name = tensor("op_12817_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_12817_end_0 = const()[name = tensor("op_12817_end_0"), val = tensor([2, 280, 1, 77])]; tensor var_12817_end_mask_0 = const()[name = tensor("op_12817_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_12817_cast_fp16 = slice_by_index(begin = var_12817_begin_0, end = var_12817_end_0, end_mask = var_12817_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12817_cast_fp16")]; tensor var_12821_begin_0 = const()[name = tensor("op_12821_begin_0"), val = tensor([0, 280, 0, 0])]; tensor var_12821_end_0 = const()[name = tensor("op_12821_end_0"), val = tensor([2, 1, 1, 77])]; tensor var_12821_end_mask_0 = const()[name = tensor("op_12821_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_12821_cast_fp16 = slice_by_index(begin = var_12821_begin_0, end = var_12821_end_0, end_mask = var_12821_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12821_cast_fp16")]; tensor var_12825_equation_0 = const()[name = tensor("op_12825_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12825_cast_fp16 = einsum(equation = var_12825_equation_0, values = (var_12763_cast_fp16, var_12695_cast_fp16))[name = tensor("op_12825_cast_fp16")]; tensor var_12826_to_fp16 = const()[name = tensor("op_12826_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1217_cast_fp16 = mul(x = var_12825_cast_fp16, y = var_12826_to_fp16)[name = tensor("aw_chunk_1217_cast_fp16")]; tensor var_12829_equation_0 = const()[name = tensor("op_12829_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12829_cast_fp16 = einsum(equation = var_12829_equation_0, values = (var_12763_cast_fp16, var_12696_cast_fp16))[name = tensor("op_12829_cast_fp16")]; tensor var_12830_to_fp16 = const()[name = tensor("op_12830_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1219_cast_fp16 = mul(x = var_12829_cast_fp16, y = var_12830_to_fp16)[name = tensor("aw_chunk_1219_cast_fp16")]; tensor var_12833_equation_0 = const()[name = tensor("op_12833_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12833_cast_fp16 = einsum(equation = var_12833_equation_0, values = (var_12763_cast_fp16, var_12697_cast_fp16))[name = tensor("op_12833_cast_fp16")]; tensor var_12834_to_fp16 = const()[name = tensor("op_12834_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1221_cast_fp16 = mul(x = var_12833_cast_fp16, y = var_12834_to_fp16)[name = tensor("aw_chunk_1221_cast_fp16")]; tensor var_12837_equation_0 = const()[name = tensor("op_12837_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12837_cast_fp16 = einsum(equation = var_12837_equation_0, values = (var_12763_cast_fp16, var_12698_cast_fp16))[name = tensor("op_12837_cast_fp16")]; tensor var_12838_to_fp16 = const()[name = tensor("op_12838_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1223_cast_fp16 = mul(x = var_12837_cast_fp16, y = var_12838_to_fp16)[name = tensor("aw_chunk_1223_cast_fp16")]; tensor var_12841_equation_0 = const()[name = tensor("op_12841_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12841_cast_fp16 = einsum(equation = var_12841_equation_0, values = (var_12763_cast_fp16, var_12699_cast_fp16))[name = tensor("op_12841_cast_fp16")]; tensor var_12842_to_fp16 = const()[name = tensor("op_12842_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1225_cast_fp16 = mul(x = var_12841_cast_fp16, y = var_12842_to_fp16)[name = tensor("aw_chunk_1225_cast_fp16")]; tensor var_12845_equation_0 = const()[name = tensor("op_12845_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12845_cast_fp16 = einsum(equation = var_12845_equation_0, values = (var_12763_cast_fp16, var_12700_cast_fp16))[name = tensor("op_12845_cast_fp16")]; tensor var_12846_to_fp16 = const()[name = tensor("op_12846_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1227_cast_fp16 = mul(x = var_12845_cast_fp16, y = var_12846_to_fp16)[name = tensor("aw_chunk_1227_cast_fp16")]; tensor var_12849_equation_0 = const()[name = tensor("op_12849_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12849_cast_fp16 = einsum(equation = var_12849_equation_0, values = (var_12763_cast_fp16, var_12701_cast_fp16))[name = tensor("op_12849_cast_fp16")]; tensor var_12850_to_fp16 = const()[name = tensor("op_12850_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1229_cast_fp16 = mul(x = var_12849_cast_fp16, y = var_12850_to_fp16)[name = tensor("aw_chunk_1229_cast_fp16")]; tensor var_12853_equation_0 = const()[name = tensor("op_12853_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12853_cast_fp16 = einsum(equation = var_12853_equation_0, values = (var_12763_cast_fp16, var_12702_cast_fp16))[name = tensor("op_12853_cast_fp16")]; tensor var_12854_to_fp16 = const()[name = tensor("op_12854_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1231_cast_fp16 = mul(x = var_12853_cast_fp16, y = var_12854_to_fp16)[name = tensor("aw_chunk_1231_cast_fp16")]; tensor var_12857_equation_0 = const()[name = tensor("op_12857_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12857_cast_fp16 = einsum(equation = var_12857_equation_0, values = (var_12767_cast_fp16, var_12703_cast_fp16))[name = tensor("op_12857_cast_fp16")]; tensor var_12858_to_fp16 = const()[name = tensor("op_12858_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1233_cast_fp16 = mul(x = var_12857_cast_fp16, y = var_12858_to_fp16)[name = tensor("aw_chunk_1233_cast_fp16")]; tensor var_12861_equation_0 = const()[name = tensor("op_12861_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12861_cast_fp16 = einsum(equation = var_12861_equation_0, values = (var_12767_cast_fp16, var_12704_cast_fp16))[name = tensor("op_12861_cast_fp16")]; tensor var_12862_to_fp16 = const()[name = tensor("op_12862_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1235_cast_fp16 = mul(x = var_12861_cast_fp16, y = var_12862_to_fp16)[name = tensor("aw_chunk_1235_cast_fp16")]; tensor var_12865_equation_0 = const()[name = tensor("op_12865_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12865_cast_fp16 = einsum(equation = var_12865_equation_0, values = (var_12767_cast_fp16, var_12705_cast_fp16))[name = tensor("op_12865_cast_fp16")]; tensor var_12866_to_fp16 = const()[name = tensor("op_12866_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1237_cast_fp16 = mul(x = var_12865_cast_fp16, y = var_12866_to_fp16)[name = tensor("aw_chunk_1237_cast_fp16")]; tensor var_12869_equation_0 = const()[name = tensor("op_12869_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12869_cast_fp16 = einsum(equation = var_12869_equation_0, values = (var_12767_cast_fp16, var_12706_cast_fp16))[name = tensor("op_12869_cast_fp16")]; tensor var_12870_to_fp16 = const()[name = tensor("op_12870_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1239_cast_fp16 = mul(x = var_12869_cast_fp16, y = var_12870_to_fp16)[name = tensor("aw_chunk_1239_cast_fp16")]; tensor var_12873_equation_0 = const()[name = tensor("op_12873_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12873_cast_fp16 = einsum(equation = var_12873_equation_0, values = (var_12767_cast_fp16, var_12707_cast_fp16))[name = tensor("op_12873_cast_fp16")]; tensor var_12874_to_fp16 = const()[name = tensor("op_12874_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1241_cast_fp16 = mul(x = var_12873_cast_fp16, y = var_12874_to_fp16)[name = tensor("aw_chunk_1241_cast_fp16")]; tensor var_12877_equation_0 = const()[name = tensor("op_12877_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12877_cast_fp16 = einsum(equation = var_12877_equation_0, values = (var_12767_cast_fp16, var_12708_cast_fp16))[name = tensor("op_12877_cast_fp16")]; tensor var_12878_to_fp16 = const()[name = tensor("op_12878_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1243_cast_fp16 = mul(x = var_12877_cast_fp16, y = var_12878_to_fp16)[name = tensor("aw_chunk_1243_cast_fp16")]; tensor var_12881_equation_0 = const()[name = tensor("op_12881_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12881_cast_fp16 = einsum(equation = var_12881_equation_0, values = (var_12767_cast_fp16, var_12709_cast_fp16))[name = tensor("op_12881_cast_fp16")]; tensor var_12882_to_fp16 = const()[name = tensor("op_12882_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1245_cast_fp16 = mul(x = var_12881_cast_fp16, y = var_12882_to_fp16)[name = tensor("aw_chunk_1245_cast_fp16")]; tensor var_12885_equation_0 = const()[name = tensor("op_12885_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12885_cast_fp16 = einsum(equation = var_12885_equation_0, values = (var_12767_cast_fp16, var_12710_cast_fp16))[name = tensor("op_12885_cast_fp16")]; tensor var_12886_to_fp16 = const()[name = tensor("op_12886_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1247_cast_fp16 = mul(x = var_12885_cast_fp16, y = var_12886_to_fp16)[name = tensor("aw_chunk_1247_cast_fp16")]; tensor var_12889_equation_0 = const()[name = tensor("op_12889_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12889_cast_fp16 = einsum(equation = var_12889_equation_0, values = (var_12771_cast_fp16, var_12711_cast_fp16))[name = tensor("op_12889_cast_fp16")]; tensor var_12890_to_fp16 = const()[name = tensor("op_12890_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1249_cast_fp16 = mul(x = var_12889_cast_fp16, y = var_12890_to_fp16)[name = tensor("aw_chunk_1249_cast_fp16")]; tensor var_12893_equation_0 = const()[name = tensor("op_12893_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12893_cast_fp16 = einsum(equation = var_12893_equation_0, values = (var_12771_cast_fp16, var_12712_cast_fp16))[name = tensor("op_12893_cast_fp16")]; tensor var_12894_to_fp16 = const()[name = tensor("op_12894_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1251_cast_fp16 = mul(x = var_12893_cast_fp16, y = var_12894_to_fp16)[name = tensor("aw_chunk_1251_cast_fp16")]; tensor var_12897_equation_0 = const()[name = tensor("op_12897_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12897_cast_fp16 = einsum(equation = var_12897_equation_0, values = (var_12771_cast_fp16, var_12713_cast_fp16))[name = tensor("op_12897_cast_fp16")]; tensor var_12898_to_fp16 = const()[name = tensor("op_12898_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1253_cast_fp16 = mul(x = var_12897_cast_fp16, y = var_12898_to_fp16)[name = tensor("aw_chunk_1253_cast_fp16")]; tensor var_12901_equation_0 = const()[name = tensor("op_12901_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12901_cast_fp16 = einsum(equation = var_12901_equation_0, values = (var_12771_cast_fp16, var_12714_cast_fp16))[name = tensor("op_12901_cast_fp16")]; tensor var_12902_to_fp16 = const()[name = tensor("op_12902_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1255_cast_fp16 = mul(x = var_12901_cast_fp16, y = var_12902_to_fp16)[name = tensor("aw_chunk_1255_cast_fp16")]; tensor var_12905_equation_0 = const()[name = tensor("op_12905_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12905_cast_fp16 = einsum(equation = var_12905_equation_0, values = (var_12771_cast_fp16, var_12715_cast_fp16))[name = tensor("op_12905_cast_fp16")]; tensor var_12906_to_fp16 = const()[name = tensor("op_12906_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1257_cast_fp16 = mul(x = var_12905_cast_fp16, y = var_12906_to_fp16)[name = tensor("aw_chunk_1257_cast_fp16")]; tensor var_12909_equation_0 = const()[name = tensor("op_12909_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12909_cast_fp16 = einsum(equation = var_12909_equation_0, values = (var_12771_cast_fp16, var_12716_cast_fp16))[name = tensor("op_12909_cast_fp16")]; tensor var_12910_to_fp16 = const()[name = tensor("op_12910_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1259_cast_fp16 = mul(x = var_12909_cast_fp16, y = var_12910_to_fp16)[name = tensor("aw_chunk_1259_cast_fp16")]; tensor var_12913_equation_0 = const()[name = tensor("op_12913_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12913_cast_fp16 = einsum(equation = var_12913_equation_0, values = (var_12771_cast_fp16, var_12717_cast_fp16))[name = tensor("op_12913_cast_fp16")]; tensor var_12914_to_fp16 = const()[name = tensor("op_12914_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1261_cast_fp16 = mul(x = var_12913_cast_fp16, y = var_12914_to_fp16)[name = tensor("aw_chunk_1261_cast_fp16")]; tensor var_12917_equation_0 = const()[name = tensor("op_12917_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12917_cast_fp16 = einsum(equation = var_12917_equation_0, values = (var_12771_cast_fp16, var_12718_cast_fp16))[name = tensor("op_12917_cast_fp16")]; tensor var_12918_to_fp16 = const()[name = tensor("op_12918_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1263_cast_fp16 = mul(x = var_12917_cast_fp16, y = var_12918_to_fp16)[name = tensor("aw_chunk_1263_cast_fp16")]; tensor var_12921_equation_0 = const()[name = tensor("op_12921_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12921_cast_fp16 = einsum(equation = var_12921_equation_0, values = (var_12775_cast_fp16, var_12719_cast_fp16))[name = tensor("op_12921_cast_fp16")]; tensor var_12922_to_fp16 = const()[name = tensor("op_12922_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1265_cast_fp16 = mul(x = var_12921_cast_fp16, y = var_12922_to_fp16)[name = tensor("aw_chunk_1265_cast_fp16")]; tensor var_12925_equation_0 = const()[name = tensor("op_12925_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12925_cast_fp16 = einsum(equation = var_12925_equation_0, values = (var_12775_cast_fp16, var_12720_cast_fp16))[name = tensor("op_12925_cast_fp16")]; tensor var_12926_to_fp16 = const()[name = tensor("op_12926_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1267_cast_fp16 = mul(x = var_12925_cast_fp16, y = var_12926_to_fp16)[name = tensor("aw_chunk_1267_cast_fp16")]; tensor var_12929_equation_0 = const()[name = tensor("op_12929_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12929_cast_fp16 = einsum(equation = var_12929_equation_0, values = (var_12775_cast_fp16, var_12721_cast_fp16))[name = tensor("op_12929_cast_fp16")]; tensor var_12930_to_fp16 = const()[name = tensor("op_12930_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1269_cast_fp16 = mul(x = var_12929_cast_fp16, y = var_12930_to_fp16)[name = tensor("aw_chunk_1269_cast_fp16")]; tensor var_12933_equation_0 = const()[name = tensor("op_12933_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12933_cast_fp16 = einsum(equation = var_12933_equation_0, values = (var_12775_cast_fp16, var_12722_cast_fp16))[name = tensor("op_12933_cast_fp16")]; tensor var_12934_to_fp16 = const()[name = tensor("op_12934_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1271_cast_fp16 = mul(x = var_12933_cast_fp16, y = var_12934_to_fp16)[name = tensor("aw_chunk_1271_cast_fp16")]; tensor var_12937_equation_0 = const()[name = tensor("op_12937_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12937_cast_fp16 = einsum(equation = var_12937_equation_0, values = (var_12775_cast_fp16, var_12723_cast_fp16))[name = tensor("op_12937_cast_fp16")]; tensor var_12938_to_fp16 = const()[name = tensor("op_12938_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1273_cast_fp16 = mul(x = var_12937_cast_fp16, y = var_12938_to_fp16)[name = tensor("aw_chunk_1273_cast_fp16")]; tensor var_12941_equation_0 = const()[name = tensor("op_12941_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12941_cast_fp16 = einsum(equation = var_12941_equation_0, values = (var_12775_cast_fp16, var_12724_cast_fp16))[name = tensor("op_12941_cast_fp16")]; tensor var_12942_to_fp16 = const()[name = tensor("op_12942_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1275_cast_fp16 = mul(x = var_12941_cast_fp16, y = var_12942_to_fp16)[name = tensor("aw_chunk_1275_cast_fp16")]; tensor var_12945_equation_0 = const()[name = tensor("op_12945_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12945_cast_fp16 = einsum(equation = var_12945_equation_0, values = (var_12775_cast_fp16, var_12725_cast_fp16))[name = tensor("op_12945_cast_fp16")]; tensor var_12946_to_fp16 = const()[name = tensor("op_12946_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1277_cast_fp16 = mul(x = var_12945_cast_fp16, y = var_12946_to_fp16)[name = tensor("aw_chunk_1277_cast_fp16")]; tensor var_12949_equation_0 = const()[name = tensor("op_12949_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12949_cast_fp16 = einsum(equation = var_12949_equation_0, values = (var_12775_cast_fp16, var_12726_cast_fp16))[name = tensor("op_12949_cast_fp16")]; tensor var_12950_to_fp16 = const()[name = tensor("op_12950_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1279_cast_fp16 = mul(x = var_12949_cast_fp16, y = var_12950_to_fp16)[name = tensor("aw_chunk_1279_cast_fp16")]; tensor var_12953_equation_0 = const()[name = tensor("op_12953_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12953_cast_fp16 = einsum(equation = var_12953_equation_0, values = (var_12779_cast_fp16, var_12727_cast_fp16))[name = tensor("op_12953_cast_fp16")]; tensor var_12954_to_fp16 = const()[name = tensor("op_12954_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1281_cast_fp16 = mul(x = var_12953_cast_fp16, y = var_12954_to_fp16)[name = tensor("aw_chunk_1281_cast_fp16")]; tensor var_12957_equation_0 = const()[name = tensor("op_12957_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12957_cast_fp16 = einsum(equation = var_12957_equation_0, values = (var_12779_cast_fp16, var_12728_cast_fp16))[name = tensor("op_12957_cast_fp16")]; tensor var_12958_to_fp16 = const()[name = tensor("op_12958_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1283_cast_fp16 = mul(x = var_12957_cast_fp16, y = var_12958_to_fp16)[name = tensor("aw_chunk_1283_cast_fp16")]; tensor var_12961_equation_0 = const()[name = tensor("op_12961_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12961_cast_fp16 = einsum(equation = var_12961_equation_0, values = (var_12779_cast_fp16, var_12729_cast_fp16))[name = tensor("op_12961_cast_fp16")]; tensor var_12962_to_fp16 = const()[name = tensor("op_12962_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1285_cast_fp16 = mul(x = var_12961_cast_fp16, y = var_12962_to_fp16)[name = tensor("aw_chunk_1285_cast_fp16")]; tensor var_12965_equation_0 = const()[name = tensor("op_12965_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12965_cast_fp16 = einsum(equation = var_12965_equation_0, values = (var_12779_cast_fp16, var_12730_cast_fp16))[name = tensor("op_12965_cast_fp16")]; tensor var_12966_to_fp16 = const()[name = tensor("op_12966_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1287_cast_fp16 = mul(x = var_12965_cast_fp16, y = var_12966_to_fp16)[name = tensor("aw_chunk_1287_cast_fp16")]; tensor var_12969_equation_0 = const()[name = tensor("op_12969_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12969_cast_fp16 = einsum(equation = var_12969_equation_0, values = (var_12779_cast_fp16, var_12731_cast_fp16))[name = tensor("op_12969_cast_fp16")]; tensor var_12970_to_fp16 = const()[name = tensor("op_12970_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1289_cast_fp16 = mul(x = var_12969_cast_fp16, y = var_12970_to_fp16)[name = tensor("aw_chunk_1289_cast_fp16")]; tensor var_12973_equation_0 = const()[name = tensor("op_12973_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12973_cast_fp16 = einsum(equation = var_12973_equation_0, values = (var_12779_cast_fp16, var_12732_cast_fp16))[name = tensor("op_12973_cast_fp16")]; tensor var_12974_to_fp16 = const()[name = tensor("op_12974_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1291_cast_fp16 = mul(x = var_12973_cast_fp16, y = var_12974_to_fp16)[name = tensor("aw_chunk_1291_cast_fp16")]; tensor var_12977_equation_0 = const()[name = tensor("op_12977_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12977_cast_fp16 = einsum(equation = var_12977_equation_0, values = (var_12779_cast_fp16, var_12733_cast_fp16))[name = tensor("op_12977_cast_fp16")]; tensor var_12978_to_fp16 = const()[name = tensor("op_12978_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1293_cast_fp16 = mul(x = var_12977_cast_fp16, y = var_12978_to_fp16)[name = tensor("aw_chunk_1293_cast_fp16")]; tensor var_12981_equation_0 = const()[name = tensor("op_12981_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12981_cast_fp16 = einsum(equation = var_12981_equation_0, values = (var_12779_cast_fp16, var_12734_cast_fp16))[name = tensor("op_12981_cast_fp16")]; tensor var_12982_to_fp16 = const()[name = tensor("op_12982_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1295_cast_fp16 = mul(x = var_12981_cast_fp16, y = var_12982_to_fp16)[name = tensor("aw_chunk_1295_cast_fp16")]; tensor var_12985_equation_0 = const()[name = tensor("op_12985_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12985_cast_fp16 = einsum(equation = var_12985_equation_0, values = (var_12783_cast_fp16, var_12735_cast_fp16))[name = tensor("op_12985_cast_fp16")]; tensor var_12986_to_fp16 = const()[name = tensor("op_12986_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1297_cast_fp16 = mul(x = var_12985_cast_fp16, y = var_12986_to_fp16)[name = tensor("aw_chunk_1297_cast_fp16")]; tensor var_12989_equation_0 = const()[name = tensor("op_12989_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12989_cast_fp16 = einsum(equation = var_12989_equation_0, values = (var_12783_cast_fp16, var_12736_cast_fp16))[name = tensor("op_12989_cast_fp16")]; tensor var_12990_to_fp16 = const()[name = tensor("op_12990_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1299_cast_fp16 = mul(x = var_12989_cast_fp16, y = var_12990_to_fp16)[name = tensor("aw_chunk_1299_cast_fp16")]; tensor var_12993_equation_0 = const()[name = tensor("op_12993_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12993_cast_fp16 = einsum(equation = var_12993_equation_0, values = (var_12783_cast_fp16, var_12737_cast_fp16))[name = tensor("op_12993_cast_fp16")]; tensor var_12994_to_fp16 = const()[name = tensor("op_12994_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1301_cast_fp16 = mul(x = var_12993_cast_fp16, y = var_12994_to_fp16)[name = tensor("aw_chunk_1301_cast_fp16")]; tensor var_12997_equation_0 = const()[name = tensor("op_12997_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_12997_cast_fp16 = einsum(equation = var_12997_equation_0, values = (var_12783_cast_fp16, var_12738_cast_fp16))[name = tensor("op_12997_cast_fp16")]; tensor var_12998_to_fp16 = const()[name = tensor("op_12998_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1303_cast_fp16 = mul(x = var_12997_cast_fp16, y = var_12998_to_fp16)[name = tensor("aw_chunk_1303_cast_fp16")]; tensor var_13001_equation_0 = const()[name = tensor("op_13001_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13001_cast_fp16 = einsum(equation = var_13001_equation_0, values = (var_12783_cast_fp16, var_12739_cast_fp16))[name = tensor("op_13001_cast_fp16")]; tensor var_13002_to_fp16 = const()[name = tensor("op_13002_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1305_cast_fp16 = mul(x = var_13001_cast_fp16, y = var_13002_to_fp16)[name = tensor("aw_chunk_1305_cast_fp16")]; tensor var_13005_equation_0 = const()[name = tensor("op_13005_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13005_cast_fp16 = einsum(equation = var_13005_equation_0, values = (var_12783_cast_fp16, var_12740_cast_fp16))[name = tensor("op_13005_cast_fp16")]; tensor var_13006_to_fp16 = const()[name = tensor("op_13006_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1307_cast_fp16 = mul(x = var_13005_cast_fp16, y = var_13006_to_fp16)[name = tensor("aw_chunk_1307_cast_fp16")]; tensor var_13009_equation_0 = const()[name = tensor("op_13009_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13009_cast_fp16 = einsum(equation = var_13009_equation_0, values = (var_12783_cast_fp16, var_12741_cast_fp16))[name = tensor("op_13009_cast_fp16")]; tensor var_13010_to_fp16 = const()[name = tensor("op_13010_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1309_cast_fp16 = mul(x = var_13009_cast_fp16, y = var_13010_to_fp16)[name = tensor("aw_chunk_1309_cast_fp16")]; tensor var_13013_equation_0 = const()[name = tensor("op_13013_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13013_cast_fp16 = einsum(equation = var_13013_equation_0, values = (var_12783_cast_fp16, var_12742_cast_fp16))[name = tensor("op_13013_cast_fp16")]; tensor var_13014_to_fp16 = const()[name = tensor("op_13014_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1311_cast_fp16 = mul(x = var_13013_cast_fp16, y = var_13014_to_fp16)[name = tensor("aw_chunk_1311_cast_fp16")]; tensor var_13017_equation_0 = const()[name = tensor("op_13017_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13017_cast_fp16 = einsum(equation = var_13017_equation_0, values = (var_12787_cast_fp16, var_12743_cast_fp16))[name = tensor("op_13017_cast_fp16")]; tensor var_13018_to_fp16 = const()[name = tensor("op_13018_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1313_cast_fp16 = mul(x = var_13017_cast_fp16, y = var_13018_to_fp16)[name = tensor("aw_chunk_1313_cast_fp16")]; tensor var_13021_equation_0 = const()[name = tensor("op_13021_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13021_cast_fp16 = einsum(equation = var_13021_equation_0, values = (var_12787_cast_fp16, var_12744_cast_fp16))[name = tensor("op_13021_cast_fp16")]; tensor var_13022_to_fp16 = const()[name = tensor("op_13022_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1315_cast_fp16 = mul(x = var_13021_cast_fp16, y = var_13022_to_fp16)[name = tensor("aw_chunk_1315_cast_fp16")]; tensor var_13025_equation_0 = const()[name = tensor("op_13025_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13025_cast_fp16 = einsum(equation = var_13025_equation_0, values = (var_12787_cast_fp16, var_12745_cast_fp16))[name = tensor("op_13025_cast_fp16")]; tensor var_13026_to_fp16 = const()[name = tensor("op_13026_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1317_cast_fp16 = mul(x = var_13025_cast_fp16, y = var_13026_to_fp16)[name = tensor("aw_chunk_1317_cast_fp16")]; tensor var_13029_equation_0 = const()[name = tensor("op_13029_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13029_cast_fp16 = einsum(equation = var_13029_equation_0, values = (var_12787_cast_fp16, var_12746_cast_fp16))[name = tensor("op_13029_cast_fp16")]; tensor var_13030_to_fp16 = const()[name = tensor("op_13030_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1319_cast_fp16 = mul(x = var_13029_cast_fp16, y = var_13030_to_fp16)[name = tensor("aw_chunk_1319_cast_fp16")]; tensor var_13033_equation_0 = const()[name = tensor("op_13033_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13033_cast_fp16 = einsum(equation = var_13033_equation_0, values = (var_12787_cast_fp16, var_12747_cast_fp16))[name = tensor("op_13033_cast_fp16")]; tensor var_13034_to_fp16 = const()[name = tensor("op_13034_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1321_cast_fp16 = mul(x = var_13033_cast_fp16, y = var_13034_to_fp16)[name = tensor("aw_chunk_1321_cast_fp16")]; tensor var_13037_equation_0 = const()[name = tensor("op_13037_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13037_cast_fp16 = einsum(equation = var_13037_equation_0, values = (var_12787_cast_fp16, var_12748_cast_fp16))[name = tensor("op_13037_cast_fp16")]; tensor var_13038_to_fp16 = const()[name = tensor("op_13038_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1323_cast_fp16 = mul(x = var_13037_cast_fp16, y = var_13038_to_fp16)[name = tensor("aw_chunk_1323_cast_fp16")]; tensor var_13041_equation_0 = const()[name = tensor("op_13041_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13041_cast_fp16 = einsum(equation = var_13041_equation_0, values = (var_12787_cast_fp16, var_12749_cast_fp16))[name = tensor("op_13041_cast_fp16")]; tensor var_13042_to_fp16 = const()[name = tensor("op_13042_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1325_cast_fp16 = mul(x = var_13041_cast_fp16, y = var_13042_to_fp16)[name = tensor("aw_chunk_1325_cast_fp16")]; tensor var_13045_equation_0 = const()[name = tensor("op_13045_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13045_cast_fp16 = einsum(equation = var_13045_equation_0, values = (var_12787_cast_fp16, var_12750_cast_fp16))[name = tensor("op_13045_cast_fp16")]; tensor var_13046_to_fp16 = const()[name = tensor("op_13046_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1327_cast_fp16 = mul(x = var_13045_cast_fp16, y = var_13046_to_fp16)[name = tensor("aw_chunk_1327_cast_fp16")]; tensor var_13049_equation_0 = const()[name = tensor("op_13049_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13049_cast_fp16 = einsum(equation = var_13049_equation_0, values = (var_12791_cast_fp16, var_12751_cast_fp16))[name = tensor("op_13049_cast_fp16")]; tensor var_13050_to_fp16 = const()[name = tensor("op_13050_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1329_cast_fp16 = mul(x = var_13049_cast_fp16, y = var_13050_to_fp16)[name = tensor("aw_chunk_1329_cast_fp16")]; tensor var_13053_equation_0 = const()[name = tensor("op_13053_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13053_cast_fp16 = einsum(equation = var_13053_equation_0, values = (var_12791_cast_fp16, var_12752_cast_fp16))[name = tensor("op_13053_cast_fp16")]; tensor var_13054_to_fp16 = const()[name = tensor("op_13054_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1331_cast_fp16 = mul(x = var_13053_cast_fp16, y = var_13054_to_fp16)[name = tensor("aw_chunk_1331_cast_fp16")]; tensor var_13057_equation_0 = const()[name = tensor("op_13057_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13057_cast_fp16 = einsum(equation = var_13057_equation_0, values = (var_12791_cast_fp16, var_12753_cast_fp16))[name = tensor("op_13057_cast_fp16")]; tensor var_13058_to_fp16 = const()[name = tensor("op_13058_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1333_cast_fp16 = mul(x = var_13057_cast_fp16, y = var_13058_to_fp16)[name = tensor("aw_chunk_1333_cast_fp16")]; tensor var_13061_equation_0 = const()[name = tensor("op_13061_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13061_cast_fp16 = einsum(equation = var_13061_equation_0, values = (var_12791_cast_fp16, var_12754_cast_fp16))[name = tensor("op_13061_cast_fp16")]; tensor var_13062_to_fp16 = const()[name = tensor("op_13062_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1335_cast_fp16 = mul(x = var_13061_cast_fp16, y = var_13062_to_fp16)[name = tensor("aw_chunk_1335_cast_fp16")]; tensor var_13065_equation_0 = const()[name = tensor("op_13065_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13065_cast_fp16 = einsum(equation = var_13065_equation_0, values = (var_12791_cast_fp16, var_12755_cast_fp16))[name = tensor("op_13065_cast_fp16")]; tensor var_13066_to_fp16 = const()[name = tensor("op_13066_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1337_cast_fp16 = mul(x = var_13065_cast_fp16, y = var_13066_to_fp16)[name = tensor("aw_chunk_1337_cast_fp16")]; tensor var_13069_equation_0 = const()[name = tensor("op_13069_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13069_cast_fp16 = einsum(equation = var_13069_equation_0, values = (var_12791_cast_fp16, var_12756_cast_fp16))[name = tensor("op_13069_cast_fp16")]; tensor var_13070_to_fp16 = const()[name = tensor("op_13070_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1339_cast_fp16 = mul(x = var_13069_cast_fp16, y = var_13070_to_fp16)[name = tensor("aw_chunk_1339_cast_fp16")]; tensor var_13073_equation_0 = const()[name = tensor("op_13073_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13073_cast_fp16 = einsum(equation = var_13073_equation_0, values = (var_12791_cast_fp16, var_12757_cast_fp16))[name = tensor("op_13073_cast_fp16")]; tensor var_13074_to_fp16 = const()[name = tensor("op_13074_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1341_cast_fp16 = mul(x = var_13073_cast_fp16, y = var_13074_to_fp16)[name = tensor("aw_chunk_1341_cast_fp16")]; tensor var_13077_equation_0 = const()[name = tensor("op_13077_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13077_cast_fp16 = einsum(equation = var_13077_equation_0, values = (var_12791_cast_fp16, var_12758_cast_fp16))[name = tensor("op_13077_cast_fp16")]; tensor var_13078_to_fp16 = const()[name = tensor("op_13078_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1343_cast_fp16 = mul(x = var_13077_cast_fp16, y = var_13078_to_fp16)[name = tensor("aw_chunk_1343_cast_fp16")]; tensor var_13080_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1217_cast_fp16)[name = tensor("op_13080_cast_fp16")]; tensor var_13081_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1219_cast_fp16)[name = tensor("op_13081_cast_fp16")]; tensor var_13082_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1221_cast_fp16)[name = tensor("op_13082_cast_fp16")]; tensor var_13083_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1223_cast_fp16)[name = tensor("op_13083_cast_fp16")]; tensor var_13084_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1225_cast_fp16)[name = tensor("op_13084_cast_fp16")]; tensor var_13085_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1227_cast_fp16)[name = tensor("op_13085_cast_fp16")]; tensor var_13086_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1229_cast_fp16)[name = tensor("op_13086_cast_fp16")]; tensor var_13087_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1231_cast_fp16)[name = tensor("op_13087_cast_fp16")]; tensor var_13088_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1233_cast_fp16)[name = tensor("op_13088_cast_fp16")]; tensor var_13089_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1235_cast_fp16)[name = tensor("op_13089_cast_fp16")]; tensor var_13090_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1237_cast_fp16)[name = tensor("op_13090_cast_fp16")]; tensor var_13091_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1239_cast_fp16)[name = tensor("op_13091_cast_fp16")]; tensor var_13092_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1241_cast_fp16)[name = tensor("op_13092_cast_fp16")]; tensor var_13093_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1243_cast_fp16)[name = tensor("op_13093_cast_fp16")]; tensor var_13094_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1245_cast_fp16)[name = tensor("op_13094_cast_fp16")]; tensor var_13095_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1247_cast_fp16)[name = tensor("op_13095_cast_fp16")]; tensor var_13096_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1249_cast_fp16)[name = tensor("op_13096_cast_fp16")]; tensor var_13097_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1251_cast_fp16)[name = tensor("op_13097_cast_fp16")]; tensor var_13098_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1253_cast_fp16)[name = tensor("op_13098_cast_fp16")]; tensor var_13099_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1255_cast_fp16)[name = tensor("op_13099_cast_fp16")]; tensor var_13100_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1257_cast_fp16)[name = tensor("op_13100_cast_fp16")]; tensor var_13101_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1259_cast_fp16)[name = tensor("op_13101_cast_fp16")]; tensor var_13102_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1261_cast_fp16)[name = tensor("op_13102_cast_fp16")]; tensor var_13103_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1263_cast_fp16)[name = tensor("op_13103_cast_fp16")]; tensor var_13104_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1265_cast_fp16)[name = tensor("op_13104_cast_fp16")]; tensor var_13105_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1267_cast_fp16)[name = tensor("op_13105_cast_fp16")]; tensor var_13106_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1269_cast_fp16)[name = tensor("op_13106_cast_fp16")]; tensor var_13107_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1271_cast_fp16)[name = tensor("op_13107_cast_fp16")]; tensor var_13108_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1273_cast_fp16)[name = tensor("op_13108_cast_fp16")]; tensor var_13109_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1275_cast_fp16)[name = tensor("op_13109_cast_fp16")]; tensor var_13110_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1277_cast_fp16)[name = tensor("op_13110_cast_fp16")]; tensor var_13111_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1279_cast_fp16)[name = tensor("op_13111_cast_fp16")]; tensor var_13112_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1281_cast_fp16)[name = tensor("op_13112_cast_fp16")]; tensor var_13113_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1283_cast_fp16)[name = tensor("op_13113_cast_fp16")]; tensor var_13114_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1285_cast_fp16)[name = tensor("op_13114_cast_fp16")]; tensor var_13115_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1287_cast_fp16)[name = tensor("op_13115_cast_fp16")]; tensor var_13116_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1289_cast_fp16)[name = tensor("op_13116_cast_fp16")]; tensor var_13117_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1291_cast_fp16)[name = tensor("op_13117_cast_fp16")]; tensor var_13118_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1293_cast_fp16)[name = tensor("op_13118_cast_fp16")]; tensor var_13119_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1295_cast_fp16)[name = tensor("op_13119_cast_fp16")]; tensor var_13120_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1297_cast_fp16)[name = tensor("op_13120_cast_fp16")]; tensor var_13121_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1299_cast_fp16)[name = tensor("op_13121_cast_fp16")]; tensor var_13122_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1301_cast_fp16)[name = tensor("op_13122_cast_fp16")]; tensor var_13123_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1303_cast_fp16)[name = tensor("op_13123_cast_fp16")]; tensor var_13124_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1305_cast_fp16)[name = tensor("op_13124_cast_fp16")]; tensor var_13125_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1307_cast_fp16)[name = tensor("op_13125_cast_fp16")]; tensor var_13126_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1309_cast_fp16)[name = tensor("op_13126_cast_fp16")]; tensor var_13127_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1311_cast_fp16)[name = tensor("op_13127_cast_fp16")]; tensor var_13128_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1313_cast_fp16)[name = tensor("op_13128_cast_fp16")]; tensor var_13129_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1315_cast_fp16)[name = tensor("op_13129_cast_fp16")]; tensor var_13130_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1317_cast_fp16)[name = tensor("op_13130_cast_fp16")]; tensor var_13131_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1319_cast_fp16)[name = tensor("op_13131_cast_fp16")]; tensor var_13132_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1321_cast_fp16)[name = tensor("op_13132_cast_fp16")]; tensor var_13133_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1323_cast_fp16)[name = tensor("op_13133_cast_fp16")]; tensor var_13134_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1325_cast_fp16)[name = tensor("op_13134_cast_fp16")]; tensor var_13135_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1327_cast_fp16)[name = tensor("op_13135_cast_fp16")]; tensor var_13136_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1329_cast_fp16)[name = tensor("op_13136_cast_fp16")]; tensor var_13137_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1331_cast_fp16)[name = tensor("op_13137_cast_fp16")]; tensor var_13138_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1333_cast_fp16)[name = tensor("op_13138_cast_fp16")]; tensor var_13139_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1335_cast_fp16)[name = tensor("op_13139_cast_fp16")]; tensor var_13140_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1337_cast_fp16)[name = tensor("op_13140_cast_fp16")]; tensor var_13141_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1339_cast_fp16)[name = tensor("op_13141_cast_fp16")]; tensor var_13142_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1341_cast_fp16)[name = tensor("op_13142_cast_fp16")]; tensor var_13143_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1343_cast_fp16)[name = tensor("op_13143_cast_fp16")]; tensor var_13145_equation_0 = const()[name = tensor("op_13145_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13145_cast_fp16 = einsum(equation = var_13145_equation_0, values = (var_12793_cast_fp16, var_13080_cast_fp16))[name = tensor("op_13145_cast_fp16")]; tensor var_13147_equation_0 = const()[name = tensor("op_13147_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13147_cast_fp16 = einsum(equation = var_13147_equation_0, values = (var_12793_cast_fp16, var_13081_cast_fp16))[name = tensor("op_13147_cast_fp16")]; tensor var_13149_equation_0 = const()[name = tensor("op_13149_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13149_cast_fp16 = einsum(equation = var_13149_equation_0, values = (var_12793_cast_fp16, var_13082_cast_fp16))[name = tensor("op_13149_cast_fp16")]; tensor var_13151_equation_0 = const()[name = tensor("op_13151_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13151_cast_fp16 = einsum(equation = var_13151_equation_0, values = (var_12793_cast_fp16, var_13083_cast_fp16))[name = tensor("op_13151_cast_fp16")]; tensor var_13153_equation_0 = const()[name = tensor("op_13153_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13153_cast_fp16 = einsum(equation = var_13153_equation_0, values = (var_12793_cast_fp16, var_13084_cast_fp16))[name = tensor("op_13153_cast_fp16")]; tensor var_13155_equation_0 = const()[name = tensor("op_13155_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13155_cast_fp16 = einsum(equation = var_13155_equation_0, values = (var_12793_cast_fp16, var_13085_cast_fp16))[name = tensor("op_13155_cast_fp16")]; tensor var_13157_equation_0 = const()[name = tensor("op_13157_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13157_cast_fp16 = einsum(equation = var_13157_equation_0, values = (var_12793_cast_fp16, var_13086_cast_fp16))[name = tensor("op_13157_cast_fp16")]; tensor var_13159_equation_0 = const()[name = tensor("op_13159_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13159_cast_fp16 = einsum(equation = var_13159_equation_0, values = (var_12793_cast_fp16, var_13087_cast_fp16))[name = tensor("op_13159_cast_fp16")]; tensor var_13161_equation_0 = const()[name = tensor("op_13161_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13161_cast_fp16 = einsum(equation = var_13161_equation_0, values = (var_12797_cast_fp16, var_13088_cast_fp16))[name = tensor("op_13161_cast_fp16")]; tensor var_13163_equation_0 = const()[name = tensor("op_13163_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13163_cast_fp16 = einsum(equation = var_13163_equation_0, values = (var_12797_cast_fp16, var_13089_cast_fp16))[name = tensor("op_13163_cast_fp16")]; tensor var_13165_equation_0 = const()[name = tensor("op_13165_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13165_cast_fp16 = einsum(equation = var_13165_equation_0, values = (var_12797_cast_fp16, var_13090_cast_fp16))[name = tensor("op_13165_cast_fp16")]; tensor var_13167_equation_0 = const()[name = tensor("op_13167_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13167_cast_fp16 = einsum(equation = var_13167_equation_0, values = (var_12797_cast_fp16, var_13091_cast_fp16))[name = tensor("op_13167_cast_fp16")]; tensor var_13169_equation_0 = const()[name = tensor("op_13169_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13169_cast_fp16 = einsum(equation = var_13169_equation_0, values = (var_12797_cast_fp16, var_13092_cast_fp16))[name = tensor("op_13169_cast_fp16")]; tensor var_13171_equation_0 = const()[name = tensor("op_13171_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13171_cast_fp16 = einsum(equation = var_13171_equation_0, values = (var_12797_cast_fp16, var_13093_cast_fp16))[name = tensor("op_13171_cast_fp16")]; tensor var_13173_equation_0 = const()[name = tensor("op_13173_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13173_cast_fp16 = einsum(equation = var_13173_equation_0, values = (var_12797_cast_fp16, var_13094_cast_fp16))[name = tensor("op_13173_cast_fp16")]; tensor var_13175_equation_0 = const()[name = tensor("op_13175_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13175_cast_fp16 = einsum(equation = var_13175_equation_0, values = (var_12797_cast_fp16, var_13095_cast_fp16))[name = tensor("op_13175_cast_fp16")]; tensor var_13177_equation_0 = const()[name = tensor("op_13177_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13177_cast_fp16 = einsum(equation = var_13177_equation_0, values = (var_12801_cast_fp16, var_13096_cast_fp16))[name = tensor("op_13177_cast_fp16")]; tensor var_13179_equation_0 = const()[name = tensor("op_13179_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13179_cast_fp16 = einsum(equation = var_13179_equation_0, values = (var_12801_cast_fp16, var_13097_cast_fp16))[name = tensor("op_13179_cast_fp16")]; tensor var_13181_equation_0 = const()[name = tensor("op_13181_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13181_cast_fp16 = einsum(equation = var_13181_equation_0, values = (var_12801_cast_fp16, var_13098_cast_fp16))[name = tensor("op_13181_cast_fp16")]; tensor var_13183_equation_0 = const()[name = tensor("op_13183_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13183_cast_fp16 = einsum(equation = var_13183_equation_0, values = (var_12801_cast_fp16, var_13099_cast_fp16))[name = tensor("op_13183_cast_fp16")]; tensor var_13185_equation_0 = const()[name = tensor("op_13185_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13185_cast_fp16 = einsum(equation = var_13185_equation_0, values = (var_12801_cast_fp16, var_13100_cast_fp16))[name = tensor("op_13185_cast_fp16")]; tensor var_13187_equation_0 = const()[name = tensor("op_13187_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13187_cast_fp16 = einsum(equation = var_13187_equation_0, values = (var_12801_cast_fp16, var_13101_cast_fp16))[name = tensor("op_13187_cast_fp16")]; tensor var_13189_equation_0 = const()[name = tensor("op_13189_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13189_cast_fp16 = einsum(equation = var_13189_equation_0, values = (var_12801_cast_fp16, var_13102_cast_fp16))[name = tensor("op_13189_cast_fp16")]; tensor var_13191_equation_0 = const()[name = tensor("op_13191_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13191_cast_fp16 = einsum(equation = var_13191_equation_0, values = (var_12801_cast_fp16, var_13103_cast_fp16))[name = tensor("op_13191_cast_fp16")]; tensor var_13193_equation_0 = const()[name = tensor("op_13193_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13193_cast_fp16 = einsum(equation = var_13193_equation_0, values = (var_12805_cast_fp16, var_13104_cast_fp16))[name = tensor("op_13193_cast_fp16")]; tensor var_13195_equation_0 = const()[name = tensor("op_13195_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13195_cast_fp16 = einsum(equation = var_13195_equation_0, values = (var_12805_cast_fp16, var_13105_cast_fp16))[name = tensor("op_13195_cast_fp16")]; tensor var_13197_equation_0 = const()[name = tensor("op_13197_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13197_cast_fp16 = einsum(equation = var_13197_equation_0, values = (var_12805_cast_fp16, var_13106_cast_fp16))[name = tensor("op_13197_cast_fp16")]; tensor var_13199_equation_0 = const()[name = tensor("op_13199_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13199_cast_fp16 = einsum(equation = var_13199_equation_0, values = (var_12805_cast_fp16, var_13107_cast_fp16))[name = tensor("op_13199_cast_fp16")]; tensor var_13201_equation_0 = const()[name = tensor("op_13201_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13201_cast_fp16 = einsum(equation = var_13201_equation_0, values = (var_12805_cast_fp16, var_13108_cast_fp16))[name = tensor("op_13201_cast_fp16")]; tensor var_13203_equation_0 = const()[name = tensor("op_13203_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13203_cast_fp16 = einsum(equation = var_13203_equation_0, values = (var_12805_cast_fp16, var_13109_cast_fp16))[name = tensor("op_13203_cast_fp16")]; tensor var_13205_equation_0 = const()[name = tensor("op_13205_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13205_cast_fp16 = einsum(equation = var_13205_equation_0, values = (var_12805_cast_fp16, var_13110_cast_fp16))[name = tensor("op_13205_cast_fp16")]; tensor var_13207_equation_0 = const()[name = tensor("op_13207_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13207_cast_fp16 = einsum(equation = var_13207_equation_0, values = (var_12805_cast_fp16, var_13111_cast_fp16))[name = tensor("op_13207_cast_fp16")]; tensor var_13209_equation_0 = const()[name = tensor("op_13209_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13209_cast_fp16 = einsum(equation = var_13209_equation_0, values = (var_12809_cast_fp16, var_13112_cast_fp16))[name = tensor("op_13209_cast_fp16")]; tensor var_13211_equation_0 = const()[name = tensor("op_13211_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13211_cast_fp16 = einsum(equation = var_13211_equation_0, values = (var_12809_cast_fp16, var_13113_cast_fp16))[name = tensor("op_13211_cast_fp16")]; tensor var_13213_equation_0 = const()[name = tensor("op_13213_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13213_cast_fp16 = einsum(equation = var_13213_equation_0, values = (var_12809_cast_fp16, var_13114_cast_fp16))[name = tensor("op_13213_cast_fp16")]; tensor var_13215_equation_0 = const()[name = tensor("op_13215_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13215_cast_fp16 = einsum(equation = var_13215_equation_0, values = (var_12809_cast_fp16, var_13115_cast_fp16))[name = tensor("op_13215_cast_fp16")]; tensor var_13217_equation_0 = const()[name = tensor("op_13217_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13217_cast_fp16 = einsum(equation = var_13217_equation_0, values = (var_12809_cast_fp16, var_13116_cast_fp16))[name = tensor("op_13217_cast_fp16")]; tensor var_13219_equation_0 = const()[name = tensor("op_13219_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13219_cast_fp16 = einsum(equation = var_13219_equation_0, values = (var_12809_cast_fp16, var_13117_cast_fp16))[name = tensor("op_13219_cast_fp16")]; tensor var_13221_equation_0 = const()[name = tensor("op_13221_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13221_cast_fp16 = einsum(equation = var_13221_equation_0, values = (var_12809_cast_fp16, var_13118_cast_fp16))[name = tensor("op_13221_cast_fp16")]; tensor var_13223_equation_0 = const()[name = tensor("op_13223_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13223_cast_fp16 = einsum(equation = var_13223_equation_0, values = (var_12809_cast_fp16, var_13119_cast_fp16))[name = tensor("op_13223_cast_fp16")]; tensor var_13225_equation_0 = const()[name = tensor("op_13225_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13225_cast_fp16 = einsum(equation = var_13225_equation_0, values = (var_12813_cast_fp16, var_13120_cast_fp16))[name = tensor("op_13225_cast_fp16")]; tensor var_13227_equation_0 = const()[name = tensor("op_13227_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13227_cast_fp16 = einsum(equation = var_13227_equation_0, values = (var_12813_cast_fp16, var_13121_cast_fp16))[name = tensor("op_13227_cast_fp16")]; tensor var_13229_equation_0 = const()[name = tensor("op_13229_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13229_cast_fp16 = einsum(equation = var_13229_equation_0, values = (var_12813_cast_fp16, var_13122_cast_fp16))[name = tensor("op_13229_cast_fp16")]; tensor var_13231_equation_0 = const()[name = tensor("op_13231_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13231_cast_fp16 = einsum(equation = var_13231_equation_0, values = (var_12813_cast_fp16, var_13123_cast_fp16))[name = tensor("op_13231_cast_fp16")]; tensor var_13233_equation_0 = const()[name = tensor("op_13233_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13233_cast_fp16 = einsum(equation = var_13233_equation_0, values = (var_12813_cast_fp16, var_13124_cast_fp16))[name = tensor("op_13233_cast_fp16")]; tensor var_13235_equation_0 = const()[name = tensor("op_13235_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13235_cast_fp16 = einsum(equation = var_13235_equation_0, values = (var_12813_cast_fp16, var_13125_cast_fp16))[name = tensor("op_13235_cast_fp16")]; tensor var_13237_equation_0 = const()[name = tensor("op_13237_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13237_cast_fp16 = einsum(equation = var_13237_equation_0, values = (var_12813_cast_fp16, var_13126_cast_fp16))[name = tensor("op_13237_cast_fp16")]; tensor var_13239_equation_0 = const()[name = tensor("op_13239_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13239_cast_fp16 = einsum(equation = var_13239_equation_0, values = (var_12813_cast_fp16, var_13127_cast_fp16))[name = tensor("op_13239_cast_fp16")]; tensor var_13241_equation_0 = const()[name = tensor("op_13241_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13241_cast_fp16 = einsum(equation = var_13241_equation_0, values = (var_12817_cast_fp16, var_13128_cast_fp16))[name = tensor("op_13241_cast_fp16")]; tensor var_13243_equation_0 = const()[name = tensor("op_13243_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13243_cast_fp16 = einsum(equation = var_13243_equation_0, values = (var_12817_cast_fp16, var_13129_cast_fp16))[name = tensor("op_13243_cast_fp16")]; tensor var_13245_equation_0 = const()[name = tensor("op_13245_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13245_cast_fp16 = einsum(equation = var_13245_equation_0, values = (var_12817_cast_fp16, var_13130_cast_fp16))[name = tensor("op_13245_cast_fp16")]; tensor var_13247_equation_0 = const()[name = tensor("op_13247_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13247_cast_fp16 = einsum(equation = var_13247_equation_0, values = (var_12817_cast_fp16, var_13131_cast_fp16))[name = tensor("op_13247_cast_fp16")]; tensor var_13249_equation_0 = const()[name = tensor("op_13249_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13249_cast_fp16 = einsum(equation = var_13249_equation_0, values = (var_12817_cast_fp16, var_13132_cast_fp16))[name = tensor("op_13249_cast_fp16")]; tensor var_13251_equation_0 = const()[name = tensor("op_13251_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13251_cast_fp16 = einsum(equation = var_13251_equation_0, values = (var_12817_cast_fp16, var_13133_cast_fp16))[name = tensor("op_13251_cast_fp16")]; tensor var_13253_equation_0 = const()[name = tensor("op_13253_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13253_cast_fp16 = einsum(equation = var_13253_equation_0, values = (var_12817_cast_fp16, var_13134_cast_fp16))[name = tensor("op_13253_cast_fp16")]; tensor var_13255_equation_0 = const()[name = tensor("op_13255_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13255_cast_fp16 = einsum(equation = var_13255_equation_0, values = (var_12817_cast_fp16, var_13135_cast_fp16))[name = tensor("op_13255_cast_fp16")]; tensor var_13257_equation_0 = const()[name = tensor("op_13257_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13257_cast_fp16 = einsum(equation = var_13257_equation_0, values = (var_12821_cast_fp16, var_13136_cast_fp16))[name = tensor("op_13257_cast_fp16")]; tensor var_13259_equation_0 = const()[name = tensor("op_13259_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13259_cast_fp16 = einsum(equation = var_13259_equation_0, values = (var_12821_cast_fp16, var_13137_cast_fp16))[name = tensor("op_13259_cast_fp16")]; tensor var_13261_equation_0 = const()[name = tensor("op_13261_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13261_cast_fp16 = einsum(equation = var_13261_equation_0, values = (var_12821_cast_fp16, var_13138_cast_fp16))[name = tensor("op_13261_cast_fp16")]; tensor var_13263_equation_0 = const()[name = tensor("op_13263_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13263_cast_fp16 = einsum(equation = var_13263_equation_0, values = (var_12821_cast_fp16, var_13139_cast_fp16))[name = tensor("op_13263_cast_fp16")]; tensor var_13265_equation_0 = const()[name = tensor("op_13265_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13265_cast_fp16 = einsum(equation = var_13265_equation_0, values = (var_12821_cast_fp16, var_13140_cast_fp16))[name = tensor("op_13265_cast_fp16")]; tensor var_13267_equation_0 = const()[name = tensor("op_13267_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13267_cast_fp16 = einsum(equation = var_13267_equation_0, values = (var_12821_cast_fp16, var_13141_cast_fp16))[name = tensor("op_13267_cast_fp16")]; tensor var_13269_equation_0 = const()[name = tensor("op_13269_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13269_cast_fp16 = einsum(equation = var_13269_equation_0, values = (var_12821_cast_fp16, var_13142_cast_fp16))[name = tensor("op_13269_cast_fp16")]; tensor var_13271_equation_0 = const()[name = tensor("op_13271_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13271_cast_fp16 = einsum(equation = var_13271_equation_0, values = (var_12821_cast_fp16, var_13143_cast_fp16))[name = tensor("op_13271_cast_fp16")]; tensor var_13273_interleave_0 = const()[name = tensor("op_13273_interleave_0"), val = tensor(false)]; tensor var_13273_cast_fp16 = concat(axis = var_10347, interleave = var_13273_interleave_0, values = (var_13145_cast_fp16, var_13147_cast_fp16, var_13149_cast_fp16, var_13151_cast_fp16, var_13153_cast_fp16, var_13155_cast_fp16, var_13157_cast_fp16, var_13159_cast_fp16))[name = tensor("op_13273_cast_fp16")]; tensor var_13275_interleave_0 = const()[name = tensor("op_13275_interleave_0"), val = tensor(false)]; tensor var_13275_cast_fp16 = concat(axis = var_10347, interleave = var_13275_interleave_0, values = (var_13161_cast_fp16, var_13163_cast_fp16, var_13165_cast_fp16, var_13167_cast_fp16, var_13169_cast_fp16, var_13171_cast_fp16, var_13173_cast_fp16, var_13175_cast_fp16))[name = tensor("op_13275_cast_fp16")]; tensor var_13277_interleave_0 = const()[name = tensor("op_13277_interleave_0"), val = tensor(false)]; tensor var_13277_cast_fp16 = concat(axis = var_10347, interleave = var_13277_interleave_0, values = (var_13177_cast_fp16, var_13179_cast_fp16, var_13181_cast_fp16, var_13183_cast_fp16, var_13185_cast_fp16, var_13187_cast_fp16, var_13189_cast_fp16, var_13191_cast_fp16))[name = tensor("op_13277_cast_fp16")]; tensor var_13279_interleave_0 = const()[name = tensor("op_13279_interleave_0"), val = tensor(false)]; tensor var_13279_cast_fp16 = concat(axis = var_10347, interleave = var_13279_interleave_0, values = (var_13193_cast_fp16, var_13195_cast_fp16, var_13197_cast_fp16, var_13199_cast_fp16, var_13201_cast_fp16, var_13203_cast_fp16, var_13205_cast_fp16, var_13207_cast_fp16))[name = tensor("op_13279_cast_fp16")]; tensor var_13281_interleave_0 = const()[name = tensor("op_13281_interleave_0"), val = tensor(false)]; tensor var_13281_cast_fp16 = concat(axis = var_10347, interleave = var_13281_interleave_0, values = (var_13209_cast_fp16, var_13211_cast_fp16, var_13213_cast_fp16, var_13215_cast_fp16, var_13217_cast_fp16, var_13219_cast_fp16, var_13221_cast_fp16, var_13223_cast_fp16))[name = tensor("op_13281_cast_fp16")]; tensor var_13283_interleave_0 = const()[name = tensor("op_13283_interleave_0"), val = tensor(false)]; tensor var_13283_cast_fp16 = concat(axis = var_10347, interleave = var_13283_interleave_0, values = (var_13225_cast_fp16, var_13227_cast_fp16, var_13229_cast_fp16, var_13231_cast_fp16, var_13233_cast_fp16, var_13235_cast_fp16, var_13237_cast_fp16, var_13239_cast_fp16))[name = tensor("op_13283_cast_fp16")]; tensor var_13285_interleave_0 = const()[name = tensor("op_13285_interleave_0"), val = tensor(false)]; tensor var_13285_cast_fp16 = concat(axis = var_10347, interleave = var_13285_interleave_0, values = (var_13241_cast_fp16, var_13243_cast_fp16, var_13245_cast_fp16, var_13247_cast_fp16, var_13249_cast_fp16, var_13251_cast_fp16, var_13253_cast_fp16, var_13255_cast_fp16))[name = tensor("op_13285_cast_fp16")]; tensor var_13287_interleave_0 = const()[name = tensor("op_13287_interleave_0"), val = tensor(false)]; tensor var_13287_cast_fp16 = concat(axis = var_10347, interleave = var_13287_interleave_0, values = (var_13257_cast_fp16, var_13259_cast_fp16, var_13261_cast_fp16, var_13263_cast_fp16, var_13265_cast_fp16, var_13267_cast_fp16, var_13269_cast_fp16, var_13271_cast_fp16))[name = tensor("op_13287_cast_fp16")]; tensor input_493_interleave_0 = const()[name = tensor("input_493_interleave_0"), val = tensor(false)]; tensor input_493_cast_fp16 = concat(axis = var_10375, interleave = input_493_interleave_0, values = (var_13273_cast_fp16, var_13275_cast_fp16, var_13277_cast_fp16, var_13279_cast_fp16, var_13281_cast_fp16, var_13283_cast_fp16, var_13285_cast_fp16, var_13287_cast_fp16))[name = tensor("input_493_cast_fp16")]; tensor var_13297_pad_type_0 = const()[name = tensor("op_13297_pad_type_0"), val = tensor("valid")]; tensor var_13297_strides_0 = const()[name = tensor("op_13297_strides_0"), val = tensor([1, 1])]; tensor var_13297_pad_0 = const()[name = tensor("op_13297_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_13297_dilations_0 = const()[name = tensor("op_13297_dilations_0"), val = tensor([1, 1])]; tensor var_13297_groups_0 = const()[name = tensor("op_13297_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(816807488)))]; tensor up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(817012352)))]; tensor var_13297_cast_fp16 = conv(bias = up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_13297_dilations_0, groups = var_13297_groups_0, pad = var_13297_pad_0, pad_type = var_13297_pad_type_0, strides = var_13297_strides_0, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_493_cast_fp16)[name = tensor("op_13297_cast_fp16")]; tensor inputs_89_cast_fp16 = add(x = var_13297_cast_fp16, y = inputs_87_cast_fp16)[name = tensor("inputs_89_cast_fp16")]; tensor input_495_axes_0 = const()[name = tensor("input_495_axes_0"), val = tensor([1])]; tensor input_495_gamma_0_to_fp16 = const()[name = tensor("input_495_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(817013056)))]; tensor input_495_beta_0_to_fp16 = const()[name = tensor("input_495_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(817013760)))]; tensor var_13307_to_fp16 = const()[name = tensor("op_13307_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_495_cast_fp16 = layer_norm(axes = input_495_axes_0, beta = input_495_beta_0_to_fp16, epsilon = var_13307_to_fp16, gamma = input_495_gamma_0_to_fp16, x = inputs_89_cast_fp16)[name = tensor("input_495_cast_fp16")]; tensor var_13327_pad_type_0 = const()[name = tensor("op_13327_pad_type_0"), val = tensor("valid")]; tensor var_13327_strides_0 = const()[name = tensor("op_13327_strides_0"), val = tensor([1, 1])]; tensor var_13327_pad_0 = const()[name = tensor("op_13327_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_13327_dilations_0 = const()[name = tensor("op_13327_dilations_0"), val = tensor([1, 1])]; tensor var_13327_groups_0 = const()[name = tensor("op_13327_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(817014464)))]; tensor up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(818652928)))]; tensor var_13327_cast_fp16 = conv(bias = up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_13327_dilations_0, groups = var_13327_groups_0, pad = var_13327_pad_0, pad_type = var_13327_pad_type_0, strides = var_13327_strides_0, weight = up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_495_cast_fp16)[name = tensor("op_13327_cast_fp16")]; tensor var_13328_split_sizes_0 = const()[name = tensor("op_13328_split_sizes_0"), val = tensor([1280, 1280])]; tensor var_13328_axis_0 = const()[name = tensor("op_13328_axis_0"), val = tensor(1)]; tensor var_13328_cast_fp16_0, tensor var_13328_cast_fp16_1 = split(axis = var_13328_axis_0, split_sizes = var_13328_split_sizes_0, x = var_13327_cast_fp16)[name = tensor("op_13328_cast_fp16")]; tensor var_13330_mode_0 = const()[name = tensor("op_13330_mode_0"), val = tensor("EXACT")]; tensor var_13330_cast_fp16 = gelu(mode = var_13330_mode_0, x = var_13328_cast_fp16_1)[name = tensor("op_13330_cast_fp16")]; tensor input_497_cast_fp16 = mul(x = var_13328_cast_fp16_0, y = var_13330_cast_fp16)[name = tensor("input_497_cast_fp16")]; tensor var_13338_pad_type_0 = const()[name = tensor("op_13338_pad_type_0"), val = tensor("valid")]; tensor var_13338_strides_0 = const()[name = tensor("op_13338_strides_0"), val = tensor([1, 1])]; tensor var_13338_pad_0 = const()[name = tensor("op_13338_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_13338_dilations_0 = const()[name = tensor("op_13338_dilations_0"), val = tensor([1, 1])]; tensor var_13338_groups_0 = const()[name = tensor("op_13338_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(818658112)))]; tensor up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(819477376)))]; tensor var_13338_cast_fp16 = conv(bias = up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_13338_dilations_0, groups = var_13338_groups_0, pad = var_13338_pad_0, pad_type = var_13338_pad_type_0, strides = var_13338_strides_0, weight = up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_497_cast_fp16)[name = tensor("op_13338_cast_fp16")]; tensor hidden_states_313_cast_fp16 = add(x = var_13338_cast_fp16, y = inputs_89_cast_fp16)[name = tensor("hidden_states_313_cast_fp16")]; tensor var_13340 = const()[name = tensor("op_13340"), val = tensor([2, 320, 64, 64])]; tensor input_499_cast_fp16 = reshape(shape = var_13340, x = hidden_states_313_cast_fp16)[name = tensor("input_499_cast_fp16")]; tensor hidden_states_315_pad_type_0 = const()[name = tensor("hidden_states_315_pad_type_0"), val = tensor("valid")]; tensor hidden_states_315_strides_0 = const()[name = tensor("hidden_states_315_strides_0"), val = tensor([1, 1])]; tensor hidden_states_315_pad_0 = const()[name = tensor("hidden_states_315_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_315_dilations_0 = const()[name = tensor("hidden_states_315_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_315_groups_0 = const()[name = tensor("hidden_states_315_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_1_proj_out_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(819478080)))]; tensor up_blocks_3_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(819682944)))]; tensor hidden_states_315_cast_fp16 = conv(bias = up_blocks_3_attentions_1_proj_out_bias_to_fp16, dilations = hidden_states_315_dilations_0, groups = hidden_states_315_groups_0, pad = hidden_states_315_pad_0, pad_type = hidden_states_315_pad_type_0, strides = hidden_states_315_strides_0, weight = up_blocks_3_attentions_1_proj_out_weight_to_fp16, x = input_499_cast_fp16)[name = tensor("hidden_states_315_cast_fp16")]; tensor hidden_states_317_cast_fp16 = add(x = hidden_states_315_cast_fp16, y = hidden_states_303_cast_fp16)[name = tensor("hidden_states_317_cast_fp16")]; tensor input_501_interleave_0 = const()[name = tensor("input_501_interleave_0"), val = tensor(false)]; tensor cast_2 = cast(dtype = cast_2_dtype_0, x = input_7_cast_fp16)[name = tensor("cast_13")]; tensor input_501_cast_fp16 = concat(axis = var_10375, interleave = input_501_interleave_0, values = (hidden_states_317_cast_fp16, cast_2))[name = tensor("input_501_cast_fp16")]; tensor reshape_228_shape_0 = const()[name = tensor("reshape_228_shape_0"), val = tensor([2, 32, 20, 64, 64])]; tensor reshape_228_cast_fp16 = reshape(shape = reshape_228_shape_0, x = input_501_cast_fp16)[name = tensor("reshape_228_cast_fp16")]; tensor reduce_mean_171_axes_0 = const()[name = tensor("reduce_mean_171_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_171_keep_dims_0 = const()[name = tensor("reduce_mean_171_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_171_cast_fp16 = reduce_mean(axes = reduce_mean_171_axes_0, keep_dims = reduce_mean_171_keep_dims_0, x = reshape_228_cast_fp16)[name = tensor("reduce_mean_171_cast_fp16")]; tensor sub_114_cast_fp16 = sub(x = reshape_228_cast_fp16, y = reduce_mean_171_cast_fp16)[name = tensor("sub_114_cast_fp16")]; tensor square_57_cast_fp16 = square(x = sub_114_cast_fp16)[name = tensor("square_57_cast_fp16")]; tensor reduce_mean_173_axes_0 = const()[name = tensor("reduce_mean_173_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_173_keep_dims_0 = const()[name = tensor("reduce_mean_173_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_173_cast_fp16 = reduce_mean(axes = reduce_mean_173_axes_0, keep_dims = reduce_mean_173_keep_dims_0, x = square_57_cast_fp16)[name = tensor("reduce_mean_173_cast_fp16")]; tensor add_114_y_0_to_fp16 = const()[name = tensor("add_114_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_114_cast_fp16 = add(x = reduce_mean_173_cast_fp16, y = add_114_y_0_to_fp16)[name = tensor("add_114_cast_fp16")]; tensor sqrt_57_cast_fp16 = sqrt(x = add_114_cast_fp16)[name = tensor("sqrt_57_cast_fp16")]; tensor real_div_57_cast_fp16 = real_div(x = sub_114_cast_fp16, y = sqrt_57_cast_fp16)[name = tensor("real_div_57_cast_fp16")]; tensor reshape_229_shape_0 = const()[name = tensor("reshape_229_shape_0"), val = tensor([2, 640, 64, 64])]; tensor reshape_229_cast_fp16 = reshape(shape = reshape_229_shape_0, x = real_div_57_cast_fp16)[name = tensor("reshape_229_cast_fp16")]; tensor add_115_gamma_0_to_fp16 = const()[name = tensor("add_115_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(819683648)))]; tensor add_115_beta_0_to_fp16 = const()[name = tensor("add_115_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(819684992)))]; tensor add_115_epsilon_0_to_fp16 = const()[name = tensor("add_115_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_115_cast_fp16 = batch_norm(beta = add_115_beta_0_to_fp16, epsilon = add_115_epsilon_0_to_fp16, gamma = add_115_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_229_cast_fp16)[name = tensor("add_115_cast_fp16")]; tensor input_505_cast_fp16 = silu(x = add_115_cast_fp16)[name = tensor("input_505_cast_fp16")]; tensor hidden_states_319_pad_type_0 = const()[name = tensor("hidden_states_319_pad_type_0"), val = tensor("custom")]; tensor hidden_states_319_pad_0 = const()[name = tensor("hidden_states_319_pad_0"), val = tensor([1, 1, 1, 1])]; tensor hidden_states_319_strides_0 = const()[name = tensor("hidden_states_319_strides_0"), val = tensor([1, 1])]; tensor hidden_states_319_dilations_0 = const()[name = tensor("hidden_states_319_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_319_groups_0 = const()[name = tensor("hidden_states_319_groups_0"), val = tensor(1)]; tensor up_blocks_3_resnets_2_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_3_resnets_2_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(819686336)))]; tensor up_blocks_3_resnets_2_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_2_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(823372800)))]; tensor hidden_states_319_cast_fp16 = conv(bias = up_blocks_3_resnets_2_conv1_bias_to_fp16, dilations = hidden_states_319_dilations_0, groups = hidden_states_319_groups_0, pad = hidden_states_319_pad_0, pad_type = hidden_states_319_pad_type_0, strides = hidden_states_319_strides_0, weight = up_blocks_3_resnets_2_conv1_weight_to_fp16, x = input_505_cast_fp16)[name = tensor("hidden_states_319_cast_fp16")]; tensor temb_pad_type_0 = const()[name = tensor("temb_pad_type_0"), val = tensor("valid")]; tensor temb_strides_0 = const()[name = tensor("temb_strides_0"), val = tensor([1, 1])]; tensor temb_pad_0 = const()[name = tensor("temb_pad_0"), val = tensor([0, 0, 0, 0])]; tensor temb_dilations_0 = const()[name = tensor("temb_dilations_0"), val = tensor([1, 1])]; tensor temb_groups_0 = const()[name = tensor("temb_groups_0"), val = tensor(1)]; tensor up_blocks_3_resnets_2_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_3_resnets_2_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(823373504)))]; tensor up_blocks_3_resnets_2_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_2_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(824192768)))]; tensor temb_cast_fp16 = conv(bias = up_blocks_3_resnets_2_time_emb_proj_bias_to_fp16, dilations = temb_dilations_0, groups = temb_groups_0, pad = temb_pad_0, pad_type = temb_pad_type_0, strides = temb_strides_0, weight = up_blocks_3_resnets_2_time_emb_proj_weight_to_fp16, x = cast_9)[name = tensor("temb_cast_fp16")]; tensor input_509_cast_fp16 = add(x = hidden_states_319_cast_fp16, y = temb_cast_fp16)[name = tensor("input_509_cast_fp16")]; tensor reshape_232_shape_0 = const()[name = tensor("reshape_232_shape_0"), val = tensor([2, 32, 10, 64, 64])]; tensor reshape_232_cast_fp16 = reshape(shape = reshape_232_shape_0, x = input_509_cast_fp16)[name = tensor("reshape_232_cast_fp16")]; tensor reduce_mean_174_axes_0 = const()[name = tensor("reduce_mean_174_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_174_keep_dims_0 = const()[name = tensor("reduce_mean_174_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_174_cast_fp16 = reduce_mean(axes = reduce_mean_174_axes_0, keep_dims = reduce_mean_174_keep_dims_0, x = reshape_232_cast_fp16)[name = tensor("reduce_mean_174_cast_fp16")]; tensor sub_116_cast_fp16 = sub(x = reshape_232_cast_fp16, y = reduce_mean_174_cast_fp16)[name = tensor("sub_116_cast_fp16")]; tensor square_58_cast_fp16 = square(x = sub_116_cast_fp16)[name = tensor("square_58_cast_fp16")]; tensor reduce_mean_176_axes_0 = const()[name = tensor("reduce_mean_176_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_176_keep_dims_0 = const()[name = tensor("reduce_mean_176_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_176_cast_fp16 = reduce_mean(axes = reduce_mean_176_axes_0, keep_dims = reduce_mean_176_keep_dims_0, x = square_58_cast_fp16)[name = tensor("reduce_mean_176_cast_fp16")]; tensor add_116_y_0_to_fp16 = const()[name = tensor("add_116_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_116_cast_fp16 = add(x = reduce_mean_176_cast_fp16, y = add_116_y_0_to_fp16)[name = tensor("add_116_cast_fp16")]; tensor sqrt_58_cast_fp16 = sqrt(x = add_116_cast_fp16)[name = tensor("sqrt_58_cast_fp16")]; tensor real_div_58_cast_fp16 = real_div(x = sub_116_cast_fp16, y = sqrt_58_cast_fp16)[name = tensor("real_div_58_cast_fp16")]; tensor reshape_233_shape_0 = const()[name = tensor("reshape_233_shape_0"), val = tensor([2, 320, 64, 64])]; tensor reshape_233_cast_fp16 = reshape(shape = reshape_233_shape_0, x = real_div_58_cast_fp16)[name = tensor("reshape_233_cast_fp16")]; tensor add_117_gamma_0_to_fp16 = const()[name = tensor("add_117_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(824193472)))]; tensor add_117_beta_0_to_fp16 = const()[name = tensor("add_117_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(824194176)))]; tensor add_117_epsilon_0_to_fp16 = const()[name = tensor("add_117_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_117_cast_fp16 = batch_norm(beta = add_117_beta_0_to_fp16, epsilon = add_117_epsilon_0_to_fp16, gamma = add_117_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_233_cast_fp16)[name = tensor("add_117_cast_fp16")]; tensor input_513_cast_fp16 = silu(x = add_117_cast_fp16)[name = tensor("input_513_cast_fp16")]; tensor hidden_states_321_pad_type_0 = const()[name = tensor("hidden_states_321_pad_type_0"), val = tensor("custom")]; tensor hidden_states_321_pad_0 = const()[name = tensor("hidden_states_321_pad_0"), val = tensor([1, 1, 1, 1])]; tensor hidden_states_321_strides_0 = const()[name = tensor("hidden_states_321_strides_0"), val = tensor([1, 1])]; tensor hidden_states_321_dilations_0 = const()[name = tensor("hidden_states_321_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_321_groups_0 = const()[name = tensor("hidden_states_321_groups_0"), val = tensor(1)]; tensor up_blocks_3_resnets_2_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_3_resnets_2_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(824194880)))]; tensor up_blocks_3_resnets_2_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_2_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(826038144)))]; tensor hidden_states_321_cast_fp16 = conv(bias = up_blocks_3_resnets_2_conv2_bias_to_fp16, dilations = hidden_states_321_dilations_0, groups = hidden_states_321_groups_0, pad = hidden_states_321_pad_0, pad_type = hidden_states_321_pad_type_0, strides = hidden_states_321_strides_0, weight = up_blocks_3_resnets_2_conv2_weight_to_fp16, x = input_513_cast_fp16)[name = tensor("hidden_states_321_cast_fp16")]; tensor x_pad_type_0 = const()[name = tensor("x_pad_type_0"), val = tensor("valid")]; tensor x_strides_0 = const()[name = tensor("x_strides_0"), val = tensor([1, 1])]; tensor x_pad_0 = const()[name = tensor("x_pad_0"), val = tensor([0, 0, 0, 0])]; tensor x_dilations_0 = const()[name = tensor("x_dilations_0"), val = tensor([1, 1])]; tensor x_groups_0 = const()[name = tensor("x_groups_0"), val = tensor(1)]; tensor up_blocks_3_resnets_2_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_3_resnets_2_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(826038848)))]; tensor up_blocks_3_resnets_2_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_2_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(826448512)))]; tensor x_cast_fp16 = conv(bias = up_blocks_3_resnets_2_conv_shortcut_bias_to_fp16, dilations = x_dilations_0, groups = x_groups_0, pad = x_pad_0, pad_type = x_pad_type_0, strides = x_strides_0, weight = up_blocks_3_resnets_2_conv_shortcut_weight_to_fp16, x = input_501_cast_fp16)[name = tensor("x_cast_fp16")]; tensor hidden_states_323_cast_fp16 = add(x = x_cast_fp16, y = hidden_states_321_cast_fp16)[name = tensor("hidden_states_323_cast_fp16")]; tensor reshape_236_shape_0 = const()[name = tensor("reshape_236_shape_0"), val = tensor([2, 32, 10, 64, 64])]; tensor reshape_236_cast_fp16 = reshape(shape = reshape_236_shape_0, x = hidden_states_323_cast_fp16)[name = tensor("reshape_236_cast_fp16")]; tensor reduce_mean_177_axes_0 = const()[name = tensor("reduce_mean_177_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_177_keep_dims_0 = const()[name = tensor("reduce_mean_177_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_177_cast_fp16 = reduce_mean(axes = reduce_mean_177_axes_0, keep_dims = reduce_mean_177_keep_dims_0, x = reshape_236_cast_fp16)[name = tensor("reduce_mean_177_cast_fp16")]; tensor sub_118_cast_fp16 = sub(x = reshape_236_cast_fp16, y = reduce_mean_177_cast_fp16)[name = tensor("sub_118_cast_fp16")]; tensor square_59_cast_fp16 = square(x = sub_118_cast_fp16)[name = tensor("square_59_cast_fp16")]; tensor reduce_mean_179_axes_0 = const()[name = tensor("reduce_mean_179_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_179_keep_dims_0 = const()[name = tensor("reduce_mean_179_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_179_cast_fp16 = reduce_mean(axes = reduce_mean_179_axes_0, keep_dims = reduce_mean_179_keep_dims_0, x = square_59_cast_fp16)[name = tensor("reduce_mean_179_cast_fp16")]; tensor add_118_y_0_to_fp16 = const()[name = tensor("add_118_y_0_to_fp16"), val = tensor(0x1.1p-20)]; tensor add_118_cast_fp16 = add(x = reduce_mean_179_cast_fp16, y = add_118_y_0_to_fp16)[name = tensor("add_118_cast_fp16")]; tensor sqrt_59_cast_fp16 = sqrt(x = add_118_cast_fp16)[name = tensor("sqrt_59_cast_fp16")]; tensor real_div_59_cast_fp16 = real_div(x = sub_118_cast_fp16, y = sqrt_59_cast_fp16)[name = tensor("real_div_59_cast_fp16")]; tensor reshape_237_shape_0 = const()[name = tensor("reshape_237_shape_0"), val = tensor([2, 320, 64, 64])]; tensor reshape_237_cast_fp16 = reshape(shape = reshape_237_shape_0, x = real_div_59_cast_fp16)[name = tensor("reshape_237_cast_fp16")]; tensor add_119_gamma_0_to_fp16 = const()[name = tensor("add_119_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(826449216)))]; tensor add_119_beta_0_to_fp16 = const()[name = tensor("add_119_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(826449920)))]; tensor add_119_epsilon_0_to_fp16 = const()[name = tensor("add_119_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_119_cast_fp16 = batch_norm(beta = add_119_beta_0_to_fp16, epsilon = add_119_epsilon_0_to_fp16, gamma = add_119_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_237_cast_fp16)[name = tensor("add_119_cast_fp16")]; tensor hidden_states_325_pad_type_0 = const()[name = tensor("hidden_states_325_pad_type_0"), val = tensor("valid")]; tensor hidden_states_325_strides_0 = const()[name = tensor("hidden_states_325_strides_0"), val = tensor([1, 1])]; tensor hidden_states_325_pad_0 = const()[name = tensor("hidden_states_325_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_325_dilations_0 = const()[name = tensor("hidden_states_325_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_325_groups_0 = const()[name = tensor("hidden_states_325_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_2_proj_in_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(826450624)))]; tensor up_blocks_3_attentions_2_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(826655488)))]; tensor hidden_states_325_cast_fp16 = conv(bias = up_blocks_3_attentions_2_proj_in_bias_to_fp16, dilations = hidden_states_325_dilations_0, groups = hidden_states_325_groups_0, pad = hidden_states_325_pad_0, pad_type = hidden_states_325_pad_type_0, strides = hidden_states_325_strides_0, weight = up_blocks_3_attentions_2_proj_in_weight_to_fp16, x = add_119_cast_fp16)[name = tensor("hidden_states_325_cast_fp16")]; tensor var_13420 = const()[name = tensor("op_13420"), val = tensor([2, 320, 1, 4096])]; tensor inputs_91_cast_fp16 = reshape(shape = var_13420, x = hidden_states_325_cast_fp16)[name = tensor("inputs_91_cast_fp16")]; tensor hidden_states_327_axes_0 = const()[name = tensor("hidden_states_327_axes_0"), val = tensor([1])]; tensor hidden_states_327_gamma_0_to_fp16 = const()[name = tensor("hidden_states_327_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(826656192)))]; tensor hidden_states_327_beta_0_to_fp16 = const()[name = tensor("hidden_states_327_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(826656896)))]; tensor var_13436_to_fp16 = const()[name = tensor("op_13436_to_fp16"), val = tensor(0x1.5p-17)]; tensor hidden_states_327_cast_fp16 = layer_norm(axes = hidden_states_327_axes_0, beta = hidden_states_327_beta_0_to_fp16, epsilon = var_13436_to_fp16, gamma = hidden_states_327_gamma_0_to_fp16, x = inputs_91_cast_fp16)[name = tensor("hidden_states_327_cast_fp16")]; tensor q_61_pad_type_0 = const()[name = tensor("q_61_pad_type_0"), val = tensor("valid")]; tensor q_61_strides_0 = const()[name = tensor("q_61_strides_0"), val = tensor([1, 1])]; tensor q_61_pad_0 = const()[name = tensor("q_61_pad_0"), val = tensor([0, 0, 0, 0])]; tensor q_61_dilations_0 = const()[name = tensor("q_61_dilations_0"), val = tensor([1, 1])]; tensor q_61_groups_0 = const()[name = tensor("q_61_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(826657600)))]; tensor q_61_cast_fp16 = conv(dilations = q_61_dilations_0, groups = q_61_groups_0, pad = q_61_pad_0, pad_type = q_61_pad_type_0, strides = q_61_strides_0, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_327_cast_fp16)[name = tensor("q_61_cast_fp16")]; tensor k_121_pad_type_0 = const()[name = tensor("k_121_pad_type_0"), val = tensor("valid")]; tensor k_121_strides_0 = const()[name = tensor("k_121_strides_0"), val = tensor([1, 1])]; tensor k_121_pad_0 = const()[name = tensor("k_121_pad_0"), val = tensor([0, 0, 0, 0])]; tensor k_121_dilations_0 = const()[name = tensor("k_121_dilations_0"), val = tensor([1, 1])]; tensor k_121_groups_0 = const()[name = tensor("k_121_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(826862464)))]; tensor k_121_cast_fp16 = conv(dilations = k_121_dilations_0, groups = k_121_groups_0, pad = k_121_pad_0, pad_type = k_121_pad_type_0, strides = k_121_strides_0, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_327_cast_fp16)[name = tensor("k_121_cast_fp16")]; tensor v_61_pad_type_0 = const()[name = tensor("v_61_pad_type_0"), val = tensor("valid")]; tensor v_61_strides_0 = const()[name = tensor("v_61_strides_0"), val = tensor([1, 1])]; tensor v_61_pad_0 = const()[name = tensor("v_61_pad_0"), val = tensor([0, 0, 0, 0])]; tensor v_61_dilations_0 = const()[name = tensor("v_61_dilations_0"), val = tensor([1, 1])]; tensor v_61_groups_0 = const()[name = tensor("v_61_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(827067328)))]; tensor v_61_cast_fp16 = conv(dilations = v_61_dilations_0, groups = v_61_groups_0, pad = v_61_pad_0, pad_type = v_61_pad_type_0, strides = v_61_strides_0, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_327_cast_fp16)[name = tensor("v_61_cast_fp16")]; tensor var_13469_begin_0 = const()[name = tensor("op_13469_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_13469_end_0 = const()[name = tensor("op_13469_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_13469_end_mask_0 = const()[name = tensor("op_13469_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_13469_cast_fp16 = slice_by_index(begin = var_13469_begin_0, end = var_13469_end_0, end_mask = var_13469_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_13469_cast_fp16")]; tensor var_13473_begin_0 = const()[name = tensor("op_13473_begin_0"), val = tensor([0, 40, 0, 0])]; tensor var_13473_end_0 = const()[name = tensor("op_13473_end_0"), val = tensor([2, 80, 1, 4096])]; tensor var_13473_end_mask_0 = const()[name = tensor("op_13473_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_13473_cast_fp16 = slice_by_index(begin = var_13473_begin_0, end = var_13473_end_0, end_mask = var_13473_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_13473_cast_fp16")]; tensor var_13477_begin_0 = const()[name = tensor("op_13477_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_13477_end_0 = const()[name = tensor("op_13477_end_0"), val = tensor([2, 120, 1, 4096])]; tensor var_13477_end_mask_0 = const()[name = tensor("op_13477_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_13477_cast_fp16 = slice_by_index(begin = var_13477_begin_0, end = var_13477_end_0, end_mask = var_13477_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_13477_cast_fp16")]; tensor var_13481_begin_0 = const()[name = tensor("op_13481_begin_0"), val = tensor([0, 120, 0, 0])]; tensor var_13481_end_0 = const()[name = tensor("op_13481_end_0"), val = tensor([2, 160, 1, 4096])]; tensor var_13481_end_mask_0 = const()[name = tensor("op_13481_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_13481_cast_fp16 = slice_by_index(begin = var_13481_begin_0, end = var_13481_end_0, end_mask = var_13481_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_13481_cast_fp16")]; tensor var_13485_begin_0 = const()[name = tensor("op_13485_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_13485_end_0 = const()[name = tensor("op_13485_end_0"), val = tensor([2, 200, 1, 4096])]; tensor var_13485_end_mask_0 = const()[name = tensor("op_13485_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_13485_cast_fp16 = slice_by_index(begin = var_13485_begin_0, end = var_13485_end_0, end_mask = var_13485_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_13485_cast_fp16")]; tensor var_13489_begin_0 = const()[name = tensor("op_13489_begin_0"), val = tensor([0, 200, 0, 0])]; tensor var_13489_end_0 = const()[name = tensor("op_13489_end_0"), val = tensor([2, 240, 1, 4096])]; tensor var_13489_end_mask_0 = const()[name = tensor("op_13489_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_13489_cast_fp16 = slice_by_index(begin = var_13489_begin_0, end = var_13489_end_0, end_mask = var_13489_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_13489_cast_fp16")]; tensor var_13493_begin_0 = const()[name = tensor("op_13493_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_13493_end_0 = const()[name = tensor("op_13493_end_0"), val = tensor([2, 280, 1, 4096])]; tensor var_13493_end_mask_0 = const()[name = tensor("op_13493_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_13493_cast_fp16 = slice_by_index(begin = var_13493_begin_0, end = var_13493_end_0, end_mask = var_13493_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_13493_cast_fp16")]; tensor var_13497_begin_0 = const()[name = tensor("op_13497_begin_0"), val = tensor([0, 280, 0, 0])]; tensor var_13497_end_0 = const()[name = tensor("op_13497_end_0"), val = tensor([2, 1, 1, 4096])]; tensor var_13497_end_mask_0 = const()[name = tensor("op_13497_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_13497_cast_fp16 = slice_by_index(begin = var_13497_begin_0, end = var_13497_end_0, end_mask = var_13497_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_13497_cast_fp16")]; tensor var_13500_begin_0 = const()[name = tensor("op_13500_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_13500_end_0 = const()[name = tensor("op_13500_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_13500_end_mask_0 = const()[name = tensor("op_13500_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13500_cast_fp16 = slice_by_index(begin = var_13500_begin_0, end = var_13500_end_0, end_mask = var_13500_end_mask_0, x = var_13469_cast_fp16)[name = tensor("op_13500_cast_fp16")]; tensor var_13501_begin_0 = const()[name = tensor("op_13501_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_13501_end_0 = const()[name = tensor("op_13501_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_13501_end_mask_0 = const()[name = tensor("op_13501_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13501_cast_fp16 = slice_by_index(begin = var_13501_begin_0, end = var_13501_end_0, end_mask = var_13501_end_mask_0, x = var_13469_cast_fp16)[name = tensor("op_13501_cast_fp16")]; tensor var_13502_begin_0 = const()[name = tensor("op_13502_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_13502_end_0 = const()[name = tensor("op_13502_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_13502_end_mask_0 = const()[name = tensor("op_13502_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13502_cast_fp16 = slice_by_index(begin = var_13502_begin_0, end = var_13502_end_0, end_mask = var_13502_end_mask_0, x = var_13469_cast_fp16)[name = tensor("op_13502_cast_fp16")]; tensor var_13503_begin_0 = const()[name = tensor("op_13503_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_13503_end_0 = const()[name = tensor("op_13503_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_13503_end_mask_0 = const()[name = tensor("op_13503_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13503_cast_fp16 = slice_by_index(begin = var_13503_begin_0, end = var_13503_end_0, end_mask = var_13503_end_mask_0, x = var_13469_cast_fp16)[name = tensor("op_13503_cast_fp16")]; tensor var_13504_begin_0 = const()[name = tensor("op_13504_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_13504_end_0 = const()[name = tensor("op_13504_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_13504_end_mask_0 = const()[name = tensor("op_13504_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13504_cast_fp16 = slice_by_index(begin = var_13504_begin_0, end = var_13504_end_0, end_mask = var_13504_end_mask_0, x = var_13469_cast_fp16)[name = tensor("op_13504_cast_fp16")]; tensor var_13505_begin_0 = const()[name = tensor("op_13505_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_13505_end_0 = const()[name = tensor("op_13505_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_13505_end_mask_0 = const()[name = tensor("op_13505_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13505_cast_fp16 = slice_by_index(begin = var_13505_begin_0, end = var_13505_end_0, end_mask = var_13505_end_mask_0, x = var_13469_cast_fp16)[name = tensor("op_13505_cast_fp16")]; tensor var_13506_begin_0 = const()[name = tensor("op_13506_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_13506_end_0 = const()[name = tensor("op_13506_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_13506_end_mask_0 = const()[name = tensor("op_13506_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13506_cast_fp16 = slice_by_index(begin = var_13506_begin_0, end = var_13506_end_0, end_mask = var_13506_end_mask_0, x = var_13469_cast_fp16)[name = tensor("op_13506_cast_fp16")]; tensor var_13507_begin_0 = const()[name = tensor("op_13507_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_13507_end_0 = const()[name = tensor("op_13507_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_13507_end_mask_0 = const()[name = tensor("op_13507_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_13507_cast_fp16 = slice_by_index(begin = var_13507_begin_0, end = var_13507_end_0, end_mask = var_13507_end_mask_0, x = var_13469_cast_fp16)[name = tensor("op_13507_cast_fp16")]; tensor var_13508_begin_0 = const()[name = tensor("op_13508_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_13508_end_0 = const()[name = tensor("op_13508_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_13508_end_mask_0 = const()[name = tensor("op_13508_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13508_cast_fp16 = slice_by_index(begin = var_13508_begin_0, end = var_13508_end_0, end_mask = var_13508_end_mask_0, x = var_13473_cast_fp16)[name = tensor("op_13508_cast_fp16")]; tensor var_13509_begin_0 = const()[name = tensor("op_13509_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_13509_end_0 = const()[name = tensor("op_13509_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_13509_end_mask_0 = const()[name = tensor("op_13509_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13509_cast_fp16 = slice_by_index(begin = var_13509_begin_0, end = var_13509_end_0, end_mask = var_13509_end_mask_0, x = var_13473_cast_fp16)[name = tensor("op_13509_cast_fp16")]; tensor var_13510_begin_0 = const()[name = tensor("op_13510_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_13510_end_0 = const()[name = tensor("op_13510_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_13510_end_mask_0 = const()[name = tensor("op_13510_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13510_cast_fp16 = slice_by_index(begin = var_13510_begin_0, end = var_13510_end_0, end_mask = var_13510_end_mask_0, x = var_13473_cast_fp16)[name = tensor("op_13510_cast_fp16")]; tensor var_13511_begin_0 = const()[name = tensor("op_13511_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_13511_end_0 = const()[name = tensor("op_13511_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_13511_end_mask_0 = const()[name = tensor("op_13511_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13511_cast_fp16 = slice_by_index(begin = var_13511_begin_0, end = var_13511_end_0, end_mask = var_13511_end_mask_0, x = var_13473_cast_fp16)[name = tensor("op_13511_cast_fp16")]; tensor var_13512_begin_0 = const()[name = tensor("op_13512_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_13512_end_0 = const()[name = tensor("op_13512_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_13512_end_mask_0 = const()[name = tensor("op_13512_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13512_cast_fp16 = slice_by_index(begin = var_13512_begin_0, end = var_13512_end_0, end_mask = var_13512_end_mask_0, x = var_13473_cast_fp16)[name = tensor("op_13512_cast_fp16")]; tensor var_13513_begin_0 = const()[name = tensor("op_13513_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_13513_end_0 = const()[name = tensor("op_13513_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_13513_end_mask_0 = const()[name = tensor("op_13513_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13513_cast_fp16 = slice_by_index(begin = var_13513_begin_0, end = var_13513_end_0, end_mask = var_13513_end_mask_0, x = var_13473_cast_fp16)[name = tensor("op_13513_cast_fp16")]; tensor var_13514_begin_0 = const()[name = tensor("op_13514_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_13514_end_0 = const()[name = tensor("op_13514_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_13514_end_mask_0 = const()[name = tensor("op_13514_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13514_cast_fp16 = slice_by_index(begin = var_13514_begin_0, end = var_13514_end_0, end_mask = var_13514_end_mask_0, x = var_13473_cast_fp16)[name = tensor("op_13514_cast_fp16")]; tensor var_13515_begin_0 = const()[name = tensor("op_13515_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_13515_end_0 = const()[name = tensor("op_13515_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_13515_end_mask_0 = const()[name = tensor("op_13515_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_13515_cast_fp16 = slice_by_index(begin = var_13515_begin_0, end = var_13515_end_0, end_mask = var_13515_end_mask_0, x = var_13473_cast_fp16)[name = tensor("op_13515_cast_fp16")]; tensor var_13516_begin_0 = const()[name = tensor("op_13516_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_13516_end_0 = const()[name = tensor("op_13516_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_13516_end_mask_0 = const()[name = tensor("op_13516_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13516_cast_fp16 = slice_by_index(begin = var_13516_begin_0, end = var_13516_end_0, end_mask = var_13516_end_mask_0, x = var_13477_cast_fp16)[name = tensor("op_13516_cast_fp16")]; tensor var_13517_begin_0 = const()[name = tensor("op_13517_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_13517_end_0 = const()[name = tensor("op_13517_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_13517_end_mask_0 = const()[name = tensor("op_13517_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13517_cast_fp16 = slice_by_index(begin = var_13517_begin_0, end = var_13517_end_0, end_mask = var_13517_end_mask_0, x = var_13477_cast_fp16)[name = tensor("op_13517_cast_fp16")]; tensor var_13518_begin_0 = const()[name = tensor("op_13518_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_13518_end_0 = const()[name = tensor("op_13518_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_13518_end_mask_0 = const()[name = tensor("op_13518_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13518_cast_fp16 = slice_by_index(begin = var_13518_begin_0, end = var_13518_end_0, end_mask = var_13518_end_mask_0, x = var_13477_cast_fp16)[name = tensor("op_13518_cast_fp16")]; tensor var_13519_begin_0 = const()[name = tensor("op_13519_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_13519_end_0 = const()[name = tensor("op_13519_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_13519_end_mask_0 = const()[name = tensor("op_13519_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13519_cast_fp16 = slice_by_index(begin = var_13519_begin_0, end = var_13519_end_0, end_mask = var_13519_end_mask_0, x = var_13477_cast_fp16)[name = tensor("op_13519_cast_fp16")]; tensor var_13520_begin_0 = const()[name = tensor("op_13520_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_13520_end_0 = const()[name = tensor("op_13520_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_13520_end_mask_0 = const()[name = tensor("op_13520_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13520_cast_fp16 = slice_by_index(begin = var_13520_begin_0, end = var_13520_end_0, end_mask = var_13520_end_mask_0, x = var_13477_cast_fp16)[name = tensor("op_13520_cast_fp16")]; tensor var_13521_begin_0 = const()[name = tensor("op_13521_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_13521_end_0 = const()[name = tensor("op_13521_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_13521_end_mask_0 = const()[name = tensor("op_13521_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13521_cast_fp16 = slice_by_index(begin = var_13521_begin_0, end = var_13521_end_0, end_mask = var_13521_end_mask_0, x = var_13477_cast_fp16)[name = tensor("op_13521_cast_fp16")]; tensor var_13522_begin_0 = const()[name = tensor("op_13522_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_13522_end_0 = const()[name = tensor("op_13522_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_13522_end_mask_0 = const()[name = tensor("op_13522_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13522_cast_fp16 = slice_by_index(begin = var_13522_begin_0, end = var_13522_end_0, end_mask = var_13522_end_mask_0, x = var_13477_cast_fp16)[name = tensor("op_13522_cast_fp16")]; tensor var_13523_begin_0 = const()[name = tensor("op_13523_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_13523_end_0 = const()[name = tensor("op_13523_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_13523_end_mask_0 = const()[name = tensor("op_13523_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_13523_cast_fp16 = slice_by_index(begin = var_13523_begin_0, end = var_13523_end_0, end_mask = var_13523_end_mask_0, x = var_13477_cast_fp16)[name = tensor("op_13523_cast_fp16")]; tensor var_13524_begin_0 = const()[name = tensor("op_13524_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_13524_end_0 = const()[name = tensor("op_13524_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_13524_end_mask_0 = const()[name = tensor("op_13524_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13524_cast_fp16 = slice_by_index(begin = var_13524_begin_0, end = var_13524_end_0, end_mask = var_13524_end_mask_0, x = var_13481_cast_fp16)[name = tensor("op_13524_cast_fp16")]; tensor var_13525_begin_0 = const()[name = tensor("op_13525_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_13525_end_0 = const()[name = tensor("op_13525_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_13525_end_mask_0 = const()[name = tensor("op_13525_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13525_cast_fp16 = slice_by_index(begin = var_13525_begin_0, end = var_13525_end_0, end_mask = var_13525_end_mask_0, x = var_13481_cast_fp16)[name = tensor("op_13525_cast_fp16")]; tensor var_13526_begin_0 = const()[name = tensor("op_13526_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_13526_end_0 = const()[name = tensor("op_13526_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_13526_end_mask_0 = const()[name = tensor("op_13526_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13526_cast_fp16 = slice_by_index(begin = var_13526_begin_0, end = var_13526_end_0, end_mask = var_13526_end_mask_0, x = var_13481_cast_fp16)[name = tensor("op_13526_cast_fp16")]; tensor var_13527_begin_0 = const()[name = tensor("op_13527_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_13527_end_0 = const()[name = tensor("op_13527_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_13527_end_mask_0 = const()[name = tensor("op_13527_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13527_cast_fp16 = slice_by_index(begin = var_13527_begin_0, end = var_13527_end_0, end_mask = var_13527_end_mask_0, x = var_13481_cast_fp16)[name = tensor("op_13527_cast_fp16")]; tensor var_13528_begin_0 = const()[name = tensor("op_13528_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_13528_end_0 = const()[name = tensor("op_13528_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_13528_end_mask_0 = const()[name = tensor("op_13528_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13528_cast_fp16 = slice_by_index(begin = var_13528_begin_0, end = var_13528_end_0, end_mask = var_13528_end_mask_0, x = var_13481_cast_fp16)[name = tensor("op_13528_cast_fp16")]; tensor var_13529_begin_0 = const()[name = tensor("op_13529_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_13529_end_0 = const()[name = tensor("op_13529_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_13529_end_mask_0 = const()[name = tensor("op_13529_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13529_cast_fp16 = slice_by_index(begin = var_13529_begin_0, end = var_13529_end_0, end_mask = var_13529_end_mask_0, x = var_13481_cast_fp16)[name = tensor("op_13529_cast_fp16")]; tensor var_13530_begin_0 = const()[name = tensor("op_13530_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_13530_end_0 = const()[name = tensor("op_13530_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_13530_end_mask_0 = const()[name = tensor("op_13530_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13530_cast_fp16 = slice_by_index(begin = var_13530_begin_0, end = var_13530_end_0, end_mask = var_13530_end_mask_0, x = var_13481_cast_fp16)[name = tensor("op_13530_cast_fp16")]; tensor var_13531_begin_0 = const()[name = tensor("op_13531_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_13531_end_0 = const()[name = tensor("op_13531_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_13531_end_mask_0 = const()[name = tensor("op_13531_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_13531_cast_fp16 = slice_by_index(begin = var_13531_begin_0, end = var_13531_end_0, end_mask = var_13531_end_mask_0, x = var_13481_cast_fp16)[name = tensor("op_13531_cast_fp16")]; tensor var_13532_begin_0 = const()[name = tensor("op_13532_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_13532_end_0 = const()[name = tensor("op_13532_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_13532_end_mask_0 = const()[name = tensor("op_13532_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13532_cast_fp16 = slice_by_index(begin = var_13532_begin_0, end = var_13532_end_0, end_mask = var_13532_end_mask_0, x = var_13485_cast_fp16)[name = tensor("op_13532_cast_fp16")]; tensor var_13533_begin_0 = const()[name = tensor("op_13533_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_13533_end_0 = const()[name = tensor("op_13533_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_13533_end_mask_0 = const()[name = tensor("op_13533_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13533_cast_fp16 = slice_by_index(begin = var_13533_begin_0, end = var_13533_end_0, end_mask = var_13533_end_mask_0, x = var_13485_cast_fp16)[name = tensor("op_13533_cast_fp16")]; tensor var_13534_begin_0 = const()[name = tensor("op_13534_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_13534_end_0 = const()[name = tensor("op_13534_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_13534_end_mask_0 = const()[name = tensor("op_13534_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13534_cast_fp16 = slice_by_index(begin = var_13534_begin_0, end = var_13534_end_0, end_mask = var_13534_end_mask_0, x = var_13485_cast_fp16)[name = tensor("op_13534_cast_fp16")]; tensor var_13535_begin_0 = const()[name = tensor("op_13535_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_13535_end_0 = const()[name = tensor("op_13535_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_13535_end_mask_0 = const()[name = tensor("op_13535_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13535_cast_fp16 = slice_by_index(begin = var_13535_begin_0, end = var_13535_end_0, end_mask = var_13535_end_mask_0, x = var_13485_cast_fp16)[name = tensor("op_13535_cast_fp16")]; tensor var_13536_begin_0 = const()[name = tensor("op_13536_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_13536_end_0 = const()[name = tensor("op_13536_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_13536_end_mask_0 = const()[name = tensor("op_13536_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13536_cast_fp16 = slice_by_index(begin = var_13536_begin_0, end = var_13536_end_0, end_mask = var_13536_end_mask_0, x = var_13485_cast_fp16)[name = tensor("op_13536_cast_fp16")]; tensor var_13537_begin_0 = const()[name = tensor("op_13537_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_13537_end_0 = const()[name = tensor("op_13537_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_13537_end_mask_0 = const()[name = tensor("op_13537_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13537_cast_fp16 = slice_by_index(begin = var_13537_begin_0, end = var_13537_end_0, end_mask = var_13537_end_mask_0, x = var_13485_cast_fp16)[name = tensor("op_13537_cast_fp16")]; tensor var_13538_begin_0 = const()[name = tensor("op_13538_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_13538_end_0 = const()[name = tensor("op_13538_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_13538_end_mask_0 = const()[name = tensor("op_13538_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13538_cast_fp16 = slice_by_index(begin = var_13538_begin_0, end = var_13538_end_0, end_mask = var_13538_end_mask_0, x = var_13485_cast_fp16)[name = tensor("op_13538_cast_fp16")]; tensor var_13539_begin_0 = const()[name = tensor("op_13539_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_13539_end_0 = const()[name = tensor("op_13539_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_13539_end_mask_0 = const()[name = tensor("op_13539_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_13539_cast_fp16 = slice_by_index(begin = var_13539_begin_0, end = var_13539_end_0, end_mask = var_13539_end_mask_0, x = var_13485_cast_fp16)[name = tensor("op_13539_cast_fp16")]; tensor var_13540_begin_0 = const()[name = tensor("op_13540_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_13540_end_0 = const()[name = tensor("op_13540_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_13540_end_mask_0 = const()[name = tensor("op_13540_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13540_cast_fp16 = slice_by_index(begin = var_13540_begin_0, end = var_13540_end_0, end_mask = var_13540_end_mask_0, x = var_13489_cast_fp16)[name = tensor("op_13540_cast_fp16")]; tensor var_13541_begin_0 = const()[name = tensor("op_13541_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_13541_end_0 = const()[name = tensor("op_13541_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_13541_end_mask_0 = const()[name = tensor("op_13541_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13541_cast_fp16 = slice_by_index(begin = var_13541_begin_0, end = var_13541_end_0, end_mask = var_13541_end_mask_0, x = var_13489_cast_fp16)[name = tensor("op_13541_cast_fp16")]; tensor var_13542_begin_0 = const()[name = tensor("op_13542_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_13542_end_0 = const()[name = tensor("op_13542_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_13542_end_mask_0 = const()[name = tensor("op_13542_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13542_cast_fp16 = slice_by_index(begin = var_13542_begin_0, end = var_13542_end_0, end_mask = var_13542_end_mask_0, x = var_13489_cast_fp16)[name = tensor("op_13542_cast_fp16")]; tensor var_13543_begin_0 = const()[name = tensor("op_13543_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_13543_end_0 = const()[name = tensor("op_13543_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_13543_end_mask_0 = const()[name = tensor("op_13543_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13543_cast_fp16 = slice_by_index(begin = var_13543_begin_0, end = var_13543_end_0, end_mask = var_13543_end_mask_0, x = var_13489_cast_fp16)[name = tensor("op_13543_cast_fp16")]; tensor var_13544_begin_0 = const()[name = tensor("op_13544_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_13544_end_0 = const()[name = tensor("op_13544_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_13544_end_mask_0 = const()[name = tensor("op_13544_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13544_cast_fp16 = slice_by_index(begin = var_13544_begin_0, end = var_13544_end_0, end_mask = var_13544_end_mask_0, x = var_13489_cast_fp16)[name = tensor("op_13544_cast_fp16")]; tensor var_13545_begin_0 = const()[name = tensor("op_13545_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_13545_end_0 = const()[name = tensor("op_13545_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_13545_end_mask_0 = const()[name = tensor("op_13545_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13545_cast_fp16 = slice_by_index(begin = var_13545_begin_0, end = var_13545_end_0, end_mask = var_13545_end_mask_0, x = var_13489_cast_fp16)[name = tensor("op_13545_cast_fp16")]; tensor var_13546_begin_0 = const()[name = tensor("op_13546_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_13546_end_0 = const()[name = tensor("op_13546_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_13546_end_mask_0 = const()[name = tensor("op_13546_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13546_cast_fp16 = slice_by_index(begin = var_13546_begin_0, end = var_13546_end_0, end_mask = var_13546_end_mask_0, x = var_13489_cast_fp16)[name = tensor("op_13546_cast_fp16")]; tensor var_13547_begin_0 = const()[name = tensor("op_13547_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_13547_end_0 = const()[name = tensor("op_13547_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_13547_end_mask_0 = const()[name = tensor("op_13547_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_13547_cast_fp16 = slice_by_index(begin = var_13547_begin_0, end = var_13547_end_0, end_mask = var_13547_end_mask_0, x = var_13489_cast_fp16)[name = tensor("op_13547_cast_fp16")]; tensor var_13548_begin_0 = const()[name = tensor("op_13548_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_13548_end_0 = const()[name = tensor("op_13548_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_13548_end_mask_0 = const()[name = tensor("op_13548_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13548_cast_fp16 = slice_by_index(begin = var_13548_begin_0, end = var_13548_end_0, end_mask = var_13548_end_mask_0, x = var_13493_cast_fp16)[name = tensor("op_13548_cast_fp16")]; tensor var_13549_begin_0 = const()[name = tensor("op_13549_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_13549_end_0 = const()[name = tensor("op_13549_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_13549_end_mask_0 = const()[name = tensor("op_13549_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13549_cast_fp16 = slice_by_index(begin = var_13549_begin_0, end = var_13549_end_0, end_mask = var_13549_end_mask_0, x = var_13493_cast_fp16)[name = tensor("op_13549_cast_fp16")]; tensor var_13550_begin_0 = const()[name = tensor("op_13550_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_13550_end_0 = const()[name = tensor("op_13550_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_13550_end_mask_0 = const()[name = tensor("op_13550_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13550_cast_fp16 = slice_by_index(begin = var_13550_begin_0, end = var_13550_end_0, end_mask = var_13550_end_mask_0, x = var_13493_cast_fp16)[name = tensor("op_13550_cast_fp16")]; tensor var_13551_begin_0 = const()[name = tensor("op_13551_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_13551_end_0 = const()[name = tensor("op_13551_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_13551_end_mask_0 = const()[name = tensor("op_13551_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13551_cast_fp16 = slice_by_index(begin = var_13551_begin_0, end = var_13551_end_0, end_mask = var_13551_end_mask_0, x = var_13493_cast_fp16)[name = tensor("op_13551_cast_fp16")]; tensor var_13552_begin_0 = const()[name = tensor("op_13552_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_13552_end_0 = const()[name = tensor("op_13552_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_13552_end_mask_0 = const()[name = tensor("op_13552_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13552_cast_fp16 = slice_by_index(begin = var_13552_begin_0, end = var_13552_end_0, end_mask = var_13552_end_mask_0, x = var_13493_cast_fp16)[name = tensor("op_13552_cast_fp16")]; tensor var_13553_begin_0 = const()[name = tensor("op_13553_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_13553_end_0 = const()[name = tensor("op_13553_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_13553_end_mask_0 = const()[name = tensor("op_13553_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13553_cast_fp16 = slice_by_index(begin = var_13553_begin_0, end = var_13553_end_0, end_mask = var_13553_end_mask_0, x = var_13493_cast_fp16)[name = tensor("op_13553_cast_fp16")]; tensor var_13554_begin_0 = const()[name = tensor("op_13554_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_13554_end_0 = const()[name = tensor("op_13554_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_13554_end_mask_0 = const()[name = tensor("op_13554_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13554_cast_fp16 = slice_by_index(begin = var_13554_begin_0, end = var_13554_end_0, end_mask = var_13554_end_mask_0, x = var_13493_cast_fp16)[name = tensor("op_13554_cast_fp16")]; tensor var_13555_begin_0 = const()[name = tensor("op_13555_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_13555_end_0 = const()[name = tensor("op_13555_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_13555_end_mask_0 = const()[name = tensor("op_13555_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_13555_cast_fp16 = slice_by_index(begin = var_13555_begin_0, end = var_13555_end_0, end_mask = var_13555_end_mask_0, x = var_13493_cast_fp16)[name = tensor("op_13555_cast_fp16")]; tensor var_13556_begin_0 = const()[name = tensor("op_13556_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_13556_end_0 = const()[name = tensor("op_13556_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_13556_end_mask_0 = const()[name = tensor("op_13556_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13556_cast_fp16 = slice_by_index(begin = var_13556_begin_0, end = var_13556_end_0, end_mask = var_13556_end_mask_0, x = var_13497_cast_fp16)[name = tensor("op_13556_cast_fp16")]; tensor var_13557_begin_0 = const()[name = tensor("op_13557_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_13557_end_0 = const()[name = tensor("op_13557_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_13557_end_mask_0 = const()[name = tensor("op_13557_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13557_cast_fp16 = slice_by_index(begin = var_13557_begin_0, end = var_13557_end_0, end_mask = var_13557_end_mask_0, x = var_13497_cast_fp16)[name = tensor("op_13557_cast_fp16")]; tensor var_13558_begin_0 = const()[name = tensor("op_13558_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_13558_end_0 = const()[name = tensor("op_13558_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_13558_end_mask_0 = const()[name = tensor("op_13558_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13558_cast_fp16 = slice_by_index(begin = var_13558_begin_0, end = var_13558_end_0, end_mask = var_13558_end_mask_0, x = var_13497_cast_fp16)[name = tensor("op_13558_cast_fp16")]; tensor var_13559_begin_0 = const()[name = tensor("op_13559_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_13559_end_0 = const()[name = tensor("op_13559_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_13559_end_mask_0 = const()[name = tensor("op_13559_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13559_cast_fp16 = slice_by_index(begin = var_13559_begin_0, end = var_13559_end_0, end_mask = var_13559_end_mask_0, x = var_13497_cast_fp16)[name = tensor("op_13559_cast_fp16")]; tensor var_13560_begin_0 = const()[name = tensor("op_13560_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_13560_end_0 = const()[name = tensor("op_13560_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_13560_end_mask_0 = const()[name = tensor("op_13560_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13560_cast_fp16 = slice_by_index(begin = var_13560_begin_0, end = var_13560_end_0, end_mask = var_13560_end_mask_0, x = var_13497_cast_fp16)[name = tensor("op_13560_cast_fp16")]; tensor var_13561_begin_0 = const()[name = tensor("op_13561_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_13561_end_0 = const()[name = tensor("op_13561_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_13561_end_mask_0 = const()[name = tensor("op_13561_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13561_cast_fp16 = slice_by_index(begin = var_13561_begin_0, end = var_13561_end_0, end_mask = var_13561_end_mask_0, x = var_13497_cast_fp16)[name = tensor("op_13561_cast_fp16")]; tensor var_13562_begin_0 = const()[name = tensor("op_13562_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_13562_end_0 = const()[name = tensor("op_13562_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_13562_end_mask_0 = const()[name = tensor("op_13562_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13562_cast_fp16 = slice_by_index(begin = var_13562_begin_0, end = var_13562_end_0, end_mask = var_13562_end_mask_0, x = var_13497_cast_fp16)[name = tensor("op_13562_cast_fp16")]; tensor var_13563_begin_0 = const()[name = tensor("op_13563_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_13563_end_0 = const()[name = tensor("op_13563_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_13563_end_mask_0 = const()[name = tensor("op_13563_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_13563_cast_fp16 = slice_by_index(begin = var_13563_begin_0, end = var_13563_end_0, end_mask = var_13563_end_mask_0, x = var_13497_cast_fp16)[name = tensor("op_13563_cast_fp16")]; tensor k_123_perm_0 = const()[name = tensor("k_123_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_13568_begin_0 = const()[name = tensor("op_13568_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_13568_end_0 = const()[name = tensor("op_13568_end_0"), val = tensor([2, 4096, 1, 40])]; tensor var_13568_end_mask_0 = const()[name = tensor("op_13568_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_123_cast_fp16 = transpose(perm = k_123_perm_0, x = k_121_cast_fp16)[name = tensor("transpose_1")]; tensor var_13568_cast_fp16 = slice_by_index(begin = var_13568_begin_0, end = var_13568_end_0, end_mask = var_13568_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13568_cast_fp16")]; tensor var_13572_begin_0 = const()[name = tensor("op_13572_begin_0"), val = tensor([0, 0, 0, 40])]; tensor var_13572_end_0 = const()[name = tensor("op_13572_end_0"), val = tensor([2, 4096, 1, 80])]; tensor var_13572_end_mask_0 = const()[name = tensor("op_13572_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13572_cast_fp16 = slice_by_index(begin = var_13572_begin_0, end = var_13572_end_0, end_mask = var_13572_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13572_cast_fp16")]; tensor var_13576_begin_0 = const()[name = tensor("op_13576_begin_0"), val = tensor([0, 0, 0, 80])]; tensor var_13576_end_0 = const()[name = tensor("op_13576_end_0"), val = tensor([2, 4096, 1, 120])]; tensor var_13576_end_mask_0 = const()[name = tensor("op_13576_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13576_cast_fp16 = slice_by_index(begin = var_13576_begin_0, end = var_13576_end_0, end_mask = var_13576_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13576_cast_fp16")]; tensor var_13580_begin_0 = const()[name = tensor("op_13580_begin_0"), val = tensor([0, 0, 0, 120])]; tensor var_13580_end_0 = const()[name = tensor("op_13580_end_0"), val = tensor([2, 4096, 1, 160])]; tensor var_13580_end_mask_0 = const()[name = tensor("op_13580_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13580_cast_fp16 = slice_by_index(begin = var_13580_begin_0, end = var_13580_end_0, end_mask = var_13580_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13580_cast_fp16")]; tensor var_13584_begin_0 = const()[name = tensor("op_13584_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_13584_end_0 = const()[name = tensor("op_13584_end_0"), val = tensor([2, 4096, 1, 200])]; tensor var_13584_end_mask_0 = const()[name = tensor("op_13584_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13584_cast_fp16 = slice_by_index(begin = var_13584_begin_0, end = var_13584_end_0, end_mask = var_13584_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13584_cast_fp16")]; tensor var_13588_begin_0 = const()[name = tensor("op_13588_begin_0"), val = tensor([0, 0, 0, 200])]; tensor var_13588_end_0 = const()[name = tensor("op_13588_end_0"), val = tensor([2, 4096, 1, 240])]; tensor var_13588_end_mask_0 = const()[name = tensor("op_13588_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13588_cast_fp16 = slice_by_index(begin = var_13588_begin_0, end = var_13588_end_0, end_mask = var_13588_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13588_cast_fp16")]; tensor var_13592_begin_0 = const()[name = tensor("op_13592_begin_0"), val = tensor([0, 0, 0, 240])]; tensor var_13592_end_0 = const()[name = tensor("op_13592_end_0"), val = tensor([2, 4096, 1, 280])]; tensor var_13592_end_mask_0 = const()[name = tensor("op_13592_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_13592_cast_fp16 = slice_by_index(begin = var_13592_begin_0, end = var_13592_end_0, end_mask = var_13592_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13592_cast_fp16")]; tensor var_13596_begin_0 = const()[name = tensor("op_13596_begin_0"), val = tensor([0, 0, 0, 280])]; tensor var_13596_end_0 = const()[name = tensor("op_13596_end_0"), val = tensor([2, 4096, 1, 1])]; tensor var_13596_end_mask_0 = const()[name = tensor("op_13596_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_13596_cast_fp16 = slice_by_index(begin = var_13596_begin_0, end = var_13596_end_0, end_mask = var_13596_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13596_cast_fp16")]; tensor var_13598_begin_0 = const()[name = tensor("op_13598_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_13598_end_0 = const()[name = tensor("op_13598_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_13598_end_mask_0 = const()[name = tensor("op_13598_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_13598_cast_fp16 = slice_by_index(begin = var_13598_begin_0, end = var_13598_end_0, end_mask = var_13598_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13598_cast_fp16")]; tensor var_13602_begin_0 = const()[name = tensor("op_13602_begin_0"), val = tensor([0, 40, 0, 0])]; tensor var_13602_end_0 = const()[name = tensor("op_13602_end_0"), val = tensor([2, 80, 1, 4096])]; tensor var_13602_end_mask_0 = const()[name = tensor("op_13602_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_13602_cast_fp16 = slice_by_index(begin = var_13602_begin_0, end = var_13602_end_0, end_mask = var_13602_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13602_cast_fp16")]; tensor var_13606_begin_0 = const()[name = tensor("op_13606_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_13606_end_0 = const()[name = tensor("op_13606_end_0"), val = tensor([2, 120, 1, 4096])]; tensor var_13606_end_mask_0 = const()[name = tensor("op_13606_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_13606_cast_fp16 = slice_by_index(begin = var_13606_begin_0, end = var_13606_end_0, end_mask = var_13606_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13606_cast_fp16")]; tensor var_13610_begin_0 = const()[name = tensor("op_13610_begin_0"), val = tensor([0, 120, 0, 0])]; tensor var_13610_end_0 = const()[name = tensor("op_13610_end_0"), val = tensor([2, 160, 1, 4096])]; tensor var_13610_end_mask_0 = const()[name = tensor("op_13610_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_13610_cast_fp16 = slice_by_index(begin = var_13610_begin_0, end = var_13610_end_0, end_mask = var_13610_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13610_cast_fp16")]; tensor var_13614_begin_0 = const()[name = tensor("op_13614_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_13614_end_0 = const()[name = tensor("op_13614_end_0"), val = tensor([2, 200, 1, 4096])]; tensor var_13614_end_mask_0 = const()[name = tensor("op_13614_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_13614_cast_fp16 = slice_by_index(begin = var_13614_begin_0, end = var_13614_end_0, end_mask = var_13614_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13614_cast_fp16")]; tensor var_13618_begin_0 = const()[name = tensor("op_13618_begin_0"), val = tensor([0, 200, 0, 0])]; tensor var_13618_end_0 = const()[name = tensor("op_13618_end_0"), val = tensor([2, 240, 1, 4096])]; tensor var_13618_end_mask_0 = const()[name = tensor("op_13618_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_13618_cast_fp16 = slice_by_index(begin = var_13618_begin_0, end = var_13618_end_0, end_mask = var_13618_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13618_cast_fp16")]; tensor var_13622_begin_0 = const()[name = tensor("op_13622_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_13622_end_0 = const()[name = tensor("op_13622_end_0"), val = tensor([2, 280, 1, 4096])]; tensor var_13622_end_mask_0 = const()[name = tensor("op_13622_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_13622_cast_fp16 = slice_by_index(begin = var_13622_begin_0, end = var_13622_end_0, end_mask = var_13622_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13622_cast_fp16")]; tensor var_13626_begin_0 = const()[name = tensor("op_13626_begin_0"), val = tensor([0, 280, 0, 0])]; tensor var_13626_end_0 = const()[name = tensor("op_13626_end_0"), val = tensor([2, 1, 1, 4096])]; tensor var_13626_end_mask_0 = const()[name = tensor("op_13626_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_13626_cast_fp16 = slice_by_index(begin = var_13626_begin_0, end = var_13626_end_0, end_mask = var_13626_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13626_cast_fp16")]; tensor var_13630_equation_0 = const()[name = tensor("op_13630_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13630_cast_fp16 = einsum(equation = var_13630_equation_0, values = (var_13568_cast_fp16, var_13500_cast_fp16))[name = tensor("op_13630_cast_fp16")]; tensor var_13631_to_fp16 = const()[name = tensor("op_13631_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1345_cast_fp16 = mul(x = var_13630_cast_fp16, y = var_13631_to_fp16)[name = tensor("aw_chunk_1345_cast_fp16")]; tensor var_13634_equation_0 = const()[name = tensor("op_13634_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13634_cast_fp16 = einsum(equation = var_13634_equation_0, values = (var_13568_cast_fp16, var_13501_cast_fp16))[name = tensor("op_13634_cast_fp16")]; tensor var_13635_to_fp16 = const()[name = tensor("op_13635_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1347_cast_fp16 = mul(x = var_13634_cast_fp16, y = var_13635_to_fp16)[name = tensor("aw_chunk_1347_cast_fp16")]; tensor var_13638_equation_0 = const()[name = tensor("op_13638_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13638_cast_fp16 = einsum(equation = var_13638_equation_0, values = (var_13568_cast_fp16, var_13502_cast_fp16))[name = tensor("op_13638_cast_fp16")]; tensor var_13639_to_fp16 = const()[name = tensor("op_13639_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1349_cast_fp16 = mul(x = var_13638_cast_fp16, y = var_13639_to_fp16)[name = tensor("aw_chunk_1349_cast_fp16")]; tensor var_13642_equation_0 = const()[name = tensor("op_13642_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13642_cast_fp16 = einsum(equation = var_13642_equation_0, values = (var_13568_cast_fp16, var_13503_cast_fp16))[name = tensor("op_13642_cast_fp16")]; tensor var_13643_to_fp16 = const()[name = tensor("op_13643_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1351_cast_fp16 = mul(x = var_13642_cast_fp16, y = var_13643_to_fp16)[name = tensor("aw_chunk_1351_cast_fp16")]; tensor var_13646_equation_0 = const()[name = tensor("op_13646_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13646_cast_fp16 = einsum(equation = var_13646_equation_0, values = (var_13568_cast_fp16, var_13504_cast_fp16))[name = tensor("op_13646_cast_fp16")]; tensor var_13647_to_fp16 = const()[name = tensor("op_13647_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1353_cast_fp16 = mul(x = var_13646_cast_fp16, y = var_13647_to_fp16)[name = tensor("aw_chunk_1353_cast_fp16")]; tensor var_13650_equation_0 = const()[name = tensor("op_13650_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13650_cast_fp16 = einsum(equation = var_13650_equation_0, values = (var_13568_cast_fp16, var_13505_cast_fp16))[name = tensor("op_13650_cast_fp16")]; tensor var_13651_to_fp16 = const()[name = tensor("op_13651_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1355_cast_fp16 = mul(x = var_13650_cast_fp16, y = var_13651_to_fp16)[name = tensor("aw_chunk_1355_cast_fp16")]; tensor var_13654_equation_0 = const()[name = tensor("op_13654_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13654_cast_fp16 = einsum(equation = var_13654_equation_0, values = (var_13568_cast_fp16, var_13506_cast_fp16))[name = tensor("op_13654_cast_fp16")]; tensor var_13655_to_fp16 = const()[name = tensor("op_13655_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1357_cast_fp16 = mul(x = var_13654_cast_fp16, y = var_13655_to_fp16)[name = tensor("aw_chunk_1357_cast_fp16")]; tensor var_13658_equation_0 = const()[name = tensor("op_13658_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13658_cast_fp16 = einsum(equation = var_13658_equation_0, values = (var_13568_cast_fp16, var_13507_cast_fp16))[name = tensor("op_13658_cast_fp16")]; tensor var_13659_to_fp16 = const()[name = tensor("op_13659_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1359_cast_fp16 = mul(x = var_13658_cast_fp16, y = var_13659_to_fp16)[name = tensor("aw_chunk_1359_cast_fp16")]; tensor var_13662_equation_0 = const()[name = tensor("op_13662_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13662_cast_fp16 = einsum(equation = var_13662_equation_0, values = (var_13572_cast_fp16, var_13508_cast_fp16))[name = tensor("op_13662_cast_fp16")]; tensor var_13663_to_fp16 = const()[name = tensor("op_13663_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1361_cast_fp16 = mul(x = var_13662_cast_fp16, y = var_13663_to_fp16)[name = tensor("aw_chunk_1361_cast_fp16")]; tensor var_13666_equation_0 = const()[name = tensor("op_13666_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13666_cast_fp16 = einsum(equation = var_13666_equation_0, values = (var_13572_cast_fp16, var_13509_cast_fp16))[name = tensor("op_13666_cast_fp16")]; tensor var_13667_to_fp16 = const()[name = tensor("op_13667_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1363_cast_fp16 = mul(x = var_13666_cast_fp16, y = var_13667_to_fp16)[name = tensor("aw_chunk_1363_cast_fp16")]; tensor var_13670_equation_0 = const()[name = tensor("op_13670_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13670_cast_fp16 = einsum(equation = var_13670_equation_0, values = (var_13572_cast_fp16, var_13510_cast_fp16))[name = tensor("op_13670_cast_fp16")]; tensor var_13671_to_fp16 = const()[name = tensor("op_13671_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1365_cast_fp16 = mul(x = var_13670_cast_fp16, y = var_13671_to_fp16)[name = tensor("aw_chunk_1365_cast_fp16")]; tensor var_13674_equation_0 = const()[name = tensor("op_13674_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13674_cast_fp16 = einsum(equation = var_13674_equation_0, values = (var_13572_cast_fp16, var_13511_cast_fp16))[name = tensor("op_13674_cast_fp16")]; tensor var_13675_to_fp16 = const()[name = tensor("op_13675_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1367_cast_fp16 = mul(x = var_13674_cast_fp16, y = var_13675_to_fp16)[name = tensor("aw_chunk_1367_cast_fp16")]; tensor var_13678_equation_0 = const()[name = tensor("op_13678_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13678_cast_fp16 = einsum(equation = var_13678_equation_0, values = (var_13572_cast_fp16, var_13512_cast_fp16))[name = tensor("op_13678_cast_fp16")]; tensor var_13679_to_fp16 = const()[name = tensor("op_13679_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1369_cast_fp16 = mul(x = var_13678_cast_fp16, y = var_13679_to_fp16)[name = tensor("aw_chunk_1369_cast_fp16")]; tensor var_13682_equation_0 = const()[name = tensor("op_13682_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13682_cast_fp16 = einsum(equation = var_13682_equation_0, values = (var_13572_cast_fp16, var_13513_cast_fp16))[name = tensor("op_13682_cast_fp16")]; tensor var_13683_to_fp16 = const()[name = tensor("op_13683_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1371_cast_fp16 = mul(x = var_13682_cast_fp16, y = var_13683_to_fp16)[name = tensor("aw_chunk_1371_cast_fp16")]; tensor var_13686_equation_0 = const()[name = tensor("op_13686_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13686_cast_fp16 = einsum(equation = var_13686_equation_0, values = (var_13572_cast_fp16, var_13514_cast_fp16))[name = tensor("op_13686_cast_fp16")]; tensor var_13687_to_fp16 = const()[name = tensor("op_13687_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1373_cast_fp16 = mul(x = var_13686_cast_fp16, y = var_13687_to_fp16)[name = tensor("aw_chunk_1373_cast_fp16")]; tensor var_13690_equation_0 = const()[name = tensor("op_13690_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13690_cast_fp16 = einsum(equation = var_13690_equation_0, values = (var_13572_cast_fp16, var_13515_cast_fp16))[name = tensor("op_13690_cast_fp16")]; tensor var_13691_to_fp16 = const()[name = tensor("op_13691_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1375_cast_fp16 = mul(x = var_13690_cast_fp16, y = var_13691_to_fp16)[name = tensor("aw_chunk_1375_cast_fp16")]; tensor var_13694_equation_0 = const()[name = tensor("op_13694_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13694_cast_fp16 = einsum(equation = var_13694_equation_0, values = (var_13576_cast_fp16, var_13516_cast_fp16))[name = tensor("op_13694_cast_fp16")]; tensor var_13695_to_fp16 = const()[name = tensor("op_13695_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1377_cast_fp16 = mul(x = var_13694_cast_fp16, y = var_13695_to_fp16)[name = tensor("aw_chunk_1377_cast_fp16")]; tensor var_13698_equation_0 = const()[name = tensor("op_13698_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13698_cast_fp16 = einsum(equation = var_13698_equation_0, values = (var_13576_cast_fp16, var_13517_cast_fp16))[name = tensor("op_13698_cast_fp16")]; tensor var_13699_to_fp16 = const()[name = tensor("op_13699_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1379_cast_fp16 = mul(x = var_13698_cast_fp16, y = var_13699_to_fp16)[name = tensor("aw_chunk_1379_cast_fp16")]; tensor var_13702_equation_0 = const()[name = tensor("op_13702_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13702_cast_fp16 = einsum(equation = var_13702_equation_0, values = (var_13576_cast_fp16, var_13518_cast_fp16))[name = tensor("op_13702_cast_fp16")]; tensor var_13703_to_fp16 = const()[name = tensor("op_13703_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1381_cast_fp16 = mul(x = var_13702_cast_fp16, y = var_13703_to_fp16)[name = tensor("aw_chunk_1381_cast_fp16")]; tensor var_13706_equation_0 = const()[name = tensor("op_13706_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13706_cast_fp16 = einsum(equation = var_13706_equation_0, values = (var_13576_cast_fp16, var_13519_cast_fp16))[name = tensor("op_13706_cast_fp16")]; tensor var_13707_to_fp16 = const()[name = tensor("op_13707_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1383_cast_fp16 = mul(x = var_13706_cast_fp16, y = var_13707_to_fp16)[name = tensor("aw_chunk_1383_cast_fp16")]; tensor var_13710_equation_0 = const()[name = tensor("op_13710_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13710_cast_fp16 = einsum(equation = var_13710_equation_0, values = (var_13576_cast_fp16, var_13520_cast_fp16))[name = tensor("op_13710_cast_fp16")]; tensor var_13711_to_fp16 = const()[name = tensor("op_13711_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1385_cast_fp16 = mul(x = var_13710_cast_fp16, y = var_13711_to_fp16)[name = tensor("aw_chunk_1385_cast_fp16")]; tensor var_13714_equation_0 = const()[name = tensor("op_13714_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13714_cast_fp16 = einsum(equation = var_13714_equation_0, values = (var_13576_cast_fp16, var_13521_cast_fp16))[name = tensor("op_13714_cast_fp16")]; tensor var_13715_to_fp16 = const()[name = tensor("op_13715_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1387_cast_fp16 = mul(x = var_13714_cast_fp16, y = var_13715_to_fp16)[name = tensor("aw_chunk_1387_cast_fp16")]; tensor var_13718_equation_0 = const()[name = tensor("op_13718_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13718_cast_fp16 = einsum(equation = var_13718_equation_0, values = (var_13576_cast_fp16, var_13522_cast_fp16))[name = tensor("op_13718_cast_fp16")]; tensor var_13719_to_fp16 = const()[name = tensor("op_13719_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1389_cast_fp16 = mul(x = var_13718_cast_fp16, y = var_13719_to_fp16)[name = tensor("aw_chunk_1389_cast_fp16")]; tensor var_13722_equation_0 = const()[name = tensor("op_13722_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13722_cast_fp16 = einsum(equation = var_13722_equation_0, values = (var_13576_cast_fp16, var_13523_cast_fp16))[name = tensor("op_13722_cast_fp16")]; tensor var_13723_to_fp16 = const()[name = tensor("op_13723_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1391_cast_fp16 = mul(x = var_13722_cast_fp16, y = var_13723_to_fp16)[name = tensor("aw_chunk_1391_cast_fp16")]; tensor var_13726_equation_0 = const()[name = tensor("op_13726_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13726_cast_fp16 = einsum(equation = var_13726_equation_0, values = (var_13580_cast_fp16, var_13524_cast_fp16))[name = tensor("op_13726_cast_fp16")]; tensor var_13727_to_fp16 = const()[name = tensor("op_13727_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1393_cast_fp16 = mul(x = var_13726_cast_fp16, y = var_13727_to_fp16)[name = tensor("aw_chunk_1393_cast_fp16")]; tensor var_13730_equation_0 = const()[name = tensor("op_13730_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13730_cast_fp16 = einsum(equation = var_13730_equation_0, values = (var_13580_cast_fp16, var_13525_cast_fp16))[name = tensor("op_13730_cast_fp16")]; tensor var_13731_to_fp16 = const()[name = tensor("op_13731_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1395_cast_fp16 = mul(x = var_13730_cast_fp16, y = var_13731_to_fp16)[name = tensor("aw_chunk_1395_cast_fp16")]; tensor var_13734_equation_0 = const()[name = tensor("op_13734_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13734_cast_fp16 = einsum(equation = var_13734_equation_0, values = (var_13580_cast_fp16, var_13526_cast_fp16))[name = tensor("op_13734_cast_fp16")]; tensor var_13735_to_fp16 = const()[name = tensor("op_13735_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1397_cast_fp16 = mul(x = var_13734_cast_fp16, y = var_13735_to_fp16)[name = tensor("aw_chunk_1397_cast_fp16")]; tensor var_13738_equation_0 = const()[name = tensor("op_13738_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13738_cast_fp16 = einsum(equation = var_13738_equation_0, values = (var_13580_cast_fp16, var_13527_cast_fp16))[name = tensor("op_13738_cast_fp16")]; tensor var_13739_to_fp16 = const()[name = tensor("op_13739_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1399_cast_fp16 = mul(x = var_13738_cast_fp16, y = var_13739_to_fp16)[name = tensor("aw_chunk_1399_cast_fp16")]; tensor var_13742_equation_0 = const()[name = tensor("op_13742_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13742_cast_fp16 = einsum(equation = var_13742_equation_0, values = (var_13580_cast_fp16, var_13528_cast_fp16))[name = tensor("op_13742_cast_fp16")]; tensor var_13743_to_fp16 = const()[name = tensor("op_13743_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1401_cast_fp16 = mul(x = var_13742_cast_fp16, y = var_13743_to_fp16)[name = tensor("aw_chunk_1401_cast_fp16")]; tensor var_13746_equation_0 = const()[name = tensor("op_13746_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13746_cast_fp16 = einsum(equation = var_13746_equation_0, values = (var_13580_cast_fp16, var_13529_cast_fp16))[name = tensor("op_13746_cast_fp16")]; tensor var_13747_to_fp16 = const()[name = tensor("op_13747_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1403_cast_fp16 = mul(x = var_13746_cast_fp16, y = var_13747_to_fp16)[name = tensor("aw_chunk_1403_cast_fp16")]; tensor var_13750_equation_0 = const()[name = tensor("op_13750_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13750_cast_fp16 = einsum(equation = var_13750_equation_0, values = (var_13580_cast_fp16, var_13530_cast_fp16))[name = tensor("op_13750_cast_fp16")]; tensor var_13751_to_fp16 = const()[name = tensor("op_13751_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1405_cast_fp16 = mul(x = var_13750_cast_fp16, y = var_13751_to_fp16)[name = tensor("aw_chunk_1405_cast_fp16")]; tensor var_13754_equation_0 = const()[name = tensor("op_13754_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13754_cast_fp16 = einsum(equation = var_13754_equation_0, values = (var_13580_cast_fp16, var_13531_cast_fp16))[name = tensor("op_13754_cast_fp16")]; tensor var_13755_to_fp16 = const()[name = tensor("op_13755_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1407_cast_fp16 = mul(x = var_13754_cast_fp16, y = var_13755_to_fp16)[name = tensor("aw_chunk_1407_cast_fp16")]; tensor var_13758_equation_0 = const()[name = tensor("op_13758_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13758_cast_fp16 = einsum(equation = var_13758_equation_0, values = (var_13584_cast_fp16, var_13532_cast_fp16))[name = tensor("op_13758_cast_fp16")]; tensor var_13759_to_fp16 = const()[name = tensor("op_13759_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1409_cast_fp16 = mul(x = var_13758_cast_fp16, y = var_13759_to_fp16)[name = tensor("aw_chunk_1409_cast_fp16")]; tensor var_13762_equation_0 = const()[name = tensor("op_13762_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13762_cast_fp16 = einsum(equation = var_13762_equation_0, values = (var_13584_cast_fp16, var_13533_cast_fp16))[name = tensor("op_13762_cast_fp16")]; tensor var_13763_to_fp16 = const()[name = tensor("op_13763_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1411_cast_fp16 = mul(x = var_13762_cast_fp16, y = var_13763_to_fp16)[name = tensor("aw_chunk_1411_cast_fp16")]; tensor var_13766_equation_0 = const()[name = tensor("op_13766_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13766_cast_fp16 = einsum(equation = var_13766_equation_0, values = (var_13584_cast_fp16, var_13534_cast_fp16))[name = tensor("op_13766_cast_fp16")]; tensor var_13767_to_fp16 = const()[name = tensor("op_13767_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1413_cast_fp16 = mul(x = var_13766_cast_fp16, y = var_13767_to_fp16)[name = tensor("aw_chunk_1413_cast_fp16")]; tensor var_13770_equation_0 = const()[name = tensor("op_13770_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13770_cast_fp16 = einsum(equation = var_13770_equation_0, values = (var_13584_cast_fp16, var_13535_cast_fp16))[name = tensor("op_13770_cast_fp16")]; tensor var_13771_to_fp16 = const()[name = tensor("op_13771_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1415_cast_fp16 = mul(x = var_13770_cast_fp16, y = var_13771_to_fp16)[name = tensor("aw_chunk_1415_cast_fp16")]; tensor var_13774_equation_0 = const()[name = tensor("op_13774_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13774_cast_fp16 = einsum(equation = var_13774_equation_0, values = (var_13584_cast_fp16, var_13536_cast_fp16))[name = tensor("op_13774_cast_fp16")]; tensor var_13775_to_fp16 = const()[name = tensor("op_13775_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1417_cast_fp16 = mul(x = var_13774_cast_fp16, y = var_13775_to_fp16)[name = tensor("aw_chunk_1417_cast_fp16")]; tensor var_13778_equation_0 = const()[name = tensor("op_13778_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13778_cast_fp16 = einsum(equation = var_13778_equation_0, values = (var_13584_cast_fp16, var_13537_cast_fp16))[name = tensor("op_13778_cast_fp16")]; tensor var_13779_to_fp16 = const()[name = tensor("op_13779_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1419_cast_fp16 = mul(x = var_13778_cast_fp16, y = var_13779_to_fp16)[name = tensor("aw_chunk_1419_cast_fp16")]; tensor var_13782_equation_0 = const()[name = tensor("op_13782_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13782_cast_fp16 = einsum(equation = var_13782_equation_0, values = (var_13584_cast_fp16, var_13538_cast_fp16))[name = tensor("op_13782_cast_fp16")]; tensor var_13783_to_fp16 = const()[name = tensor("op_13783_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1421_cast_fp16 = mul(x = var_13782_cast_fp16, y = var_13783_to_fp16)[name = tensor("aw_chunk_1421_cast_fp16")]; tensor var_13786_equation_0 = const()[name = tensor("op_13786_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13786_cast_fp16 = einsum(equation = var_13786_equation_0, values = (var_13584_cast_fp16, var_13539_cast_fp16))[name = tensor("op_13786_cast_fp16")]; tensor var_13787_to_fp16 = const()[name = tensor("op_13787_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1423_cast_fp16 = mul(x = var_13786_cast_fp16, y = var_13787_to_fp16)[name = tensor("aw_chunk_1423_cast_fp16")]; tensor var_13790_equation_0 = const()[name = tensor("op_13790_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13790_cast_fp16 = einsum(equation = var_13790_equation_0, values = (var_13588_cast_fp16, var_13540_cast_fp16))[name = tensor("op_13790_cast_fp16")]; tensor var_13791_to_fp16 = const()[name = tensor("op_13791_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1425_cast_fp16 = mul(x = var_13790_cast_fp16, y = var_13791_to_fp16)[name = tensor("aw_chunk_1425_cast_fp16")]; tensor var_13794_equation_0 = const()[name = tensor("op_13794_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13794_cast_fp16 = einsum(equation = var_13794_equation_0, values = (var_13588_cast_fp16, var_13541_cast_fp16))[name = tensor("op_13794_cast_fp16")]; tensor var_13795_to_fp16 = const()[name = tensor("op_13795_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1427_cast_fp16 = mul(x = var_13794_cast_fp16, y = var_13795_to_fp16)[name = tensor("aw_chunk_1427_cast_fp16")]; tensor var_13798_equation_0 = const()[name = tensor("op_13798_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13798_cast_fp16 = einsum(equation = var_13798_equation_0, values = (var_13588_cast_fp16, var_13542_cast_fp16))[name = tensor("op_13798_cast_fp16")]; tensor var_13799_to_fp16 = const()[name = tensor("op_13799_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1429_cast_fp16 = mul(x = var_13798_cast_fp16, y = var_13799_to_fp16)[name = tensor("aw_chunk_1429_cast_fp16")]; tensor var_13802_equation_0 = const()[name = tensor("op_13802_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13802_cast_fp16 = einsum(equation = var_13802_equation_0, values = (var_13588_cast_fp16, var_13543_cast_fp16))[name = tensor("op_13802_cast_fp16")]; tensor var_13803_to_fp16 = const()[name = tensor("op_13803_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1431_cast_fp16 = mul(x = var_13802_cast_fp16, y = var_13803_to_fp16)[name = tensor("aw_chunk_1431_cast_fp16")]; tensor var_13806_equation_0 = const()[name = tensor("op_13806_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13806_cast_fp16 = einsum(equation = var_13806_equation_0, values = (var_13588_cast_fp16, var_13544_cast_fp16))[name = tensor("op_13806_cast_fp16")]; tensor var_13807_to_fp16 = const()[name = tensor("op_13807_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1433_cast_fp16 = mul(x = var_13806_cast_fp16, y = var_13807_to_fp16)[name = tensor("aw_chunk_1433_cast_fp16")]; tensor var_13810_equation_0 = const()[name = tensor("op_13810_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13810_cast_fp16 = einsum(equation = var_13810_equation_0, values = (var_13588_cast_fp16, var_13545_cast_fp16))[name = tensor("op_13810_cast_fp16")]; tensor var_13811_to_fp16 = const()[name = tensor("op_13811_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1435_cast_fp16 = mul(x = var_13810_cast_fp16, y = var_13811_to_fp16)[name = tensor("aw_chunk_1435_cast_fp16")]; tensor var_13814_equation_0 = const()[name = tensor("op_13814_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13814_cast_fp16 = einsum(equation = var_13814_equation_0, values = (var_13588_cast_fp16, var_13546_cast_fp16))[name = tensor("op_13814_cast_fp16")]; tensor var_13815_to_fp16 = const()[name = tensor("op_13815_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1437_cast_fp16 = mul(x = var_13814_cast_fp16, y = var_13815_to_fp16)[name = tensor("aw_chunk_1437_cast_fp16")]; tensor var_13818_equation_0 = const()[name = tensor("op_13818_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13818_cast_fp16 = einsum(equation = var_13818_equation_0, values = (var_13588_cast_fp16, var_13547_cast_fp16))[name = tensor("op_13818_cast_fp16")]; tensor var_13819_to_fp16 = const()[name = tensor("op_13819_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1439_cast_fp16 = mul(x = var_13818_cast_fp16, y = var_13819_to_fp16)[name = tensor("aw_chunk_1439_cast_fp16")]; tensor var_13822_equation_0 = const()[name = tensor("op_13822_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13822_cast_fp16 = einsum(equation = var_13822_equation_0, values = (var_13592_cast_fp16, var_13548_cast_fp16))[name = tensor("op_13822_cast_fp16")]; tensor var_13823_to_fp16 = const()[name = tensor("op_13823_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1441_cast_fp16 = mul(x = var_13822_cast_fp16, y = var_13823_to_fp16)[name = tensor("aw_chunk_1441_cast_fp16")]; tensor var_13826_equation_0 = const()[name = tensor("op_13826_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13826_cast_fp16 = einsum(equation = var_13826_equation_0, values = (var_13592_cast_fp16, var_13549_cast_fp16))[name = tensor("op_13826_cast_fp16")]; tensor var_13827_to_fp16 = const()[name = tensor("op_13827_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1443_cast_fp16 = mul(x = var_13826_cast_fp16, y = var_13827_to_fp16)[name = tensor("aw_chunk_1443_cast_fp16")]; tensor var_13830_equation_0 = const()[name = tensor("op_13830_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13830_cast_fp16 = einsum(equation = var_13830_equation_0, values = (var_13592_cast_fp16, var_13550_cast_fp16))[name = tensor("op_13830_cast_fp16")]; tensor var_13831_to_fp16 = const()[name = tensor("op_13831_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1445_cast_fp16 = mul(x = var_13830_cast_fp16, y = var_13831_to_fp16)[name = tensor("aw_chunk_1445_cast_fp16")]; tensor var_13834_equation_0 = const()[name = tensor("op_13834_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13834_cast_fp16 = einsum(equation = var_13834_equation_0, values = (var_13592_cast_fp16, var_13551_cast_fp16))[name = tensor("op_13834_cast_fp16")]; tensor var_13835_to_fp16 = const()[name = tensor("op_13835_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1447_cast_fp16 = mul(x = var_13834_cast_fp16, y = var_13835_to_fp16)[name = tensor("aw_chunk_1447_cast_fp16")]; tensor var_13838_equation_0 = const()[name = tensor("op_13838_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13838_cast_fp16 = einsum(equation = var_13838_equation_0, values = (var_13592_cast_fp16, var_13552_cast_fp16))[name = tensor("op_13838_cast_fp16")]; tensor var_13839_to_fp16 = const()[name = tensor("op_13839_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1449_cast_fp16 = mul(x = var_13838_cast_fp16, y = var_13839_to_fp16)[name = tensor("aw_chunk_1449_cast_fp16")]; tensor var_13842_equation_0 = const()[name = tensor("op_13842_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13842_cast_fp16 = einsum(equation = var_13842_equation_0, values = (var_13592_cast_fp16, var_13553_cast_fp16))[name = tensor("op_13842_cast_fp16")]; tensor var_13843_to_fp16 = const()[name = tensor("op_13843_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1451_cast_fp16 = mul(x = var_13842_cast_fp16, y = var_13843_to_fp16)[name = tensor("aw_chunk_1451_cast_fp16")]; tensor var_13846_equation_0 = const()[name = tensor("op_13846_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13846_cast_fp16 = einsum(equation = var_13846_equation_0, values = (var_13592_cast_fp16, var_13554_cast_fp16))[name = tensor("op_13846_cast_fp16")]; tensor var_13847_to_fp16 = const()[name = tensor("op_13847_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1453_cast_fp16 = mul(x = var_13846_cast_fp16, y = var_13847_to_fp16)[name = tensor("aw_chunk_1453_cast_fp16")]; tensor var_13850_equation_0 = const()[name = tensor("op_13850_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13850_cast_fp16 = einsum(equation = var_13850_equation_0, values = (var_13592_cast_fp16, var_13555_cast_fp16))[name = tensor("op_13850_cast_fp16")]; tensor var_13851_to_fp16 = const()[name = tensor("op_13851_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1455_cast_fp16 = mul(x = var_13850_cast_fp16, y = var_13851_to_fp16)[name = tensor("aw_chunk_1455_cast_fp16")]; tensor var_13854_equation_0 = const()[name = tensor("op_13854_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13854_cast_fp16 = einsum(equation = var_13854_equation_0, values = (var_13596_cast_fp16, var_13556_cast_fp16))[name = tensor("op_13854_cast_fp16")]; tensor var_13855_to_fp16 = const()[name = tensor("op_13855_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1457_cast_fp16 = mul(x = var_13854_cast_fp16, y = var_13855_to_fp16)[name = tensor("aw_chunk_1457_cast_fp16")]; tensor var_13858_equation_0 = const()[name = tensor("op_13858_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13858_cast_fp16 = einsum(equation = var_13858_equation_0, values = (var_13596_cast_fp16, var_13557_cast_fp16))[name = tensor("op_13858_cast_fp16")]; tensor var_13859_to_fp16 = const()[name = tensor("op_13859_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1459_cast_fp16 = mul(x = var_13858_cast_fp16, y = var_13859_to_fp16)[name = tensor("aw_chunk_1459_cast_fp16")]; tensor var_13862_equation_0 = const()[name = tensor("op_13862_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13862_cast_fp16 = einsum(equation = var_13862_equation_0, values = (var_13596_cast_fp16, var_13558_cast_fp16))[name = tensor("op_13862_cast_fp16")]; tensor var_13863_to_fp16 = const()[name = tensor("op_13863_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1461_cast_fp16 = mul(x = var_13862_cast_fp16, y = var_13863_to_fp16)[name = tensor("aw_chunk_1461_cast_fp16")]; tensor var_13866_equation_0 = const()[name = tensor("op_13866_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13866_cast_fp16 = einsum(equation = var_13866_equation_0, values = (var_13596_cast_fp16, var_13559_cast_fp16))[name = tensor("op_13866_cast_fp16")]; tensor var_13867_to_fp16 = const()[name = tensor("op_13867_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1463_cast_fp16 = mul(x = var_13866_cast_fp16, y = var_13867_to_fp16)[name = tensor("aw_chunk_1463_cast_fp16")]; tensor var_13870_equation_0 = const()[name = tensor("op_13870_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13870_cast_fp16 = einsum(equation = var_13870_equation_0, values = (var_13596_cast_fp16, var_13560_cast_fp16))[name = tensor("op_13870_cast_fp16")]; tensor var_13871_to_fp16 = const()[name = tensor("op_13871_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1465_cast_fp16 = mul(x = var_13870_cast_fp16, y = var_13871_to_fp16)[name = tensor("aw_chunk_1465_cast_fp16")]; tensor var_13874_equation_0 = const()[name = tensor("op_13874_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13874_cast_fp16 = einsum(equation = var_13874_equation_0, values = (var_13596_cast_fp16, var_13561_cast_fp16))[name = tensor("op_13874_cast_fp16")]; tensor var_13875_to_fp16 = const()[name = tensor("op_13875_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1467_cast_fp16 = mul(x = var_13874_cast_fp16, y = var_13875_to_fp16)[name = tensor("aw_chunk_1467_cast_fp16")]; tensor var_13878_equation_0 = const()[name = tensor("op_13878_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13878_cast_fp16 = einsum(equation = var_13878_equation_0, values = (var_13596_cast_fp16, var_13562_cast_fp16))[name = tensor("op_13878_cast_fp16")]; tensor var_13879_to_fp16 = const()[name = tensor("op_13879_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1469_cast_fp16 = mul(x = var_13878_cast_fp16, y = var_13879_to_fp16)[name = tensor("aw_chunk_1469_cast_fp16")]; tensor var_13882_equation_0 = const()[name = tensor("op_13882_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_13882_cast_fp16 = einsum(equation = var_13882_equation_0, values = (var_13596_cast_fp16, var_13563_cast_fp16))[name = tensor("op_13882_cast_fp16")]; tensor var_13883_to_fp16 = const()[name = tensor("op_13883_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1471_cast_fp16 = mul(x = var_13882_cast_fp16, y = var_13883_to_fp16)[name = tensor("aw_chunk_1471_cast_fp16")]; tensor var_13885_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1345_cast_fp16)[name = tensor("op_13885_cast_fp16")]; tensor var_13886_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1347_cast_fp16)[name = tensor("op_13886_cast_fp16")]; tensor var_13887_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1349_cast_fp16)[name = tensor("op_13887_cast_fp16")]; tensor var_13888_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1351_cast_fp16)[name = tensor("op_13888_cast_fp16")]; tensor var_13889_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1353_cast_fp16)[name = tensor("op_13889_cast_fp16")]; tensor var_13890_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1355_cast_fp16)[name = tensor("op_13890_cast_fp16")]; tensor var_13891_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1357_cast_fp16)[name = tensor("op_13891_cast_fp16")]; tensor var_13892_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1359_cast_fp16)[name = tensor("op_13892_cast_fp16")]; tensor var_13893_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1361_cast_fp16)[name = tensor("op_13893_cast_fp16")]; tensor var_13894_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1363_cast_fp16)[name = tensor("op_13894_cast_fp16")]; tensor var_13895_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1365_cast_fp16)[name = tensor("op_13895_cast_fp16")]; tensor var_13896_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1367_cast_fp16)[name = tensor("op_13896_cast_fp16")]; tensor var_13897_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1369_cast_fp16)[name = tensor("op_13897_cast_fp16")]; tensor var_13898_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1371_cast_fp16)[name = tensor("op_13898_cast_fp16")]; tensor var_13899_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1373_cast_fp16)[name = tensor("op_13899_cast_fp16")]; tensor var_13900_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1375_cast_fp16)[name = tensor("op_13900_cast_fp16")]; tensor var_13901_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1377_cast_fp16)[name = tensor("op_13901_cast_fp16")]; tensor var_13902_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1379_cast_fp16)[name = tensor("op_13902_cast_fp16")]; tensor var_13903_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1381_cast_fp16)[name = tensor("op_13903_cast_fp16")]; tensor var_13904_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1383_cast_fp16)[name = tensor("op_13904_cast_fp16")]; tensor var_13905_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1385_cast_fp16)[name = tensor("op_13905_cast_fp16")]; tensor var_13906_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1387_cast_fp16)[name = tensor("op_13906_cast_fp16")]; tensor var_13907_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1389_cast_fp16)[name = tensor("op_13907_cast_fp16")]; tensor var_13908_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1391_cast_fp16)[name = tensor("op_13908_cast_fp16")]; tensor var_13909_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1393_cast_fp16)[name = tensor("op_13909_cast_fp16")]; tensor var_13910_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1395_cast_fp16)[name = tensor("op_13910_cast_fp16")]; tensor var_13911_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1397_cast_fp16)[name = tensor("op_13911_cast_fp16")]; tensor var_13912_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1399_cast_fp16)[name = tensor("op_13912_cast_fp16")]; tensor var_13913_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1401_cast_fp16)[name = tensor("op_13913_cast_fp16")]; tensor var_13914_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1403_cast_fp16)[name = tensor("op_13914_cast_fp16")]; tensor var_13915_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1405_cast_fp16)[name = tensor("op_13915_cast_fp16")]; tensor var_13916_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1407_cast_fp16)[name = tensor("op_13916_cast_fp16")]; tensor var_13917_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1409_cast_fp16)[name = tensor("op_13917_cast_fp16")]; tensor var_13918_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1411_cast_fp16)[name = tensor("op_13918_cast_fp16")]; tensor var_13919_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1413_cast_fp16)[name = tensor("op_13919_cast_fp16")]; tensor var_13920_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1415_cast_fp16)[name = tensor("op_13920_cast_fp16")]; tensor var_13921_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1417_cast_fp16)[name = tensor("op_13921_cast_fp16")]; tensor var_13922_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1419_cast_fp16)[name = tensor("op_13922_cast_fp16")]; tensor var_13923_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1421_cast_fp16)[name = tensor("op_13923_cast_fp16")]; tensor var_13924_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1423_cast_fp16)[name = tensor("op_13924_cast_fp16")]; tensor var_13925_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1425_cast_fp16)[name = tensor("op_13925_cast_fp16")]; tensor var_13926_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1427_cast_fp16)[name = tensor("op_13926_cast_fp16")]; tensor var_13927_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1429_cast_fp16)[name = tensor("op_13927_cast_fp16")]; tensor var_13928_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1431_cast_fp16)[name = tensor("op_13928_cast_fp16")]; tensor var_13929_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1433_cast_fp16)[name = tensor("op_13929_cast_fp16")]; tensor var_13930_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1435_cast_fp16)[name = tensor("op_13930_cast_fp16")]; tensor var_13931_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1437_cast_fp16)[name = tensor("op_13931_cast_fp16")]; tensor var_13932_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1439_cast_fp16)[name = tensor("op_13932_cast_fp16")]; tensor var_13933_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1441_cast_fp16)[name = tensor("op_13933_cast_fp16")]; tensor var_13934_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1443_cast_fp16)[name = tensor("op_13934_cast_fp16")]; tensor var_13935_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1445_cast_fp16)[name = tensor("op_13935_cast_fp16")]; tensor var_13936_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1447_cast_fp16)[name = tensor("op_13936_cast_fp16")]; tensor var_13937_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1449_cast_fp16)[name = tensor("op_13937_cast_fp16")]; tensor var_13938_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1451_cast_fp16)[name = tensor("op_13938_cast_fp16")]; tensor var_13939_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1453_cast_fp16)[name = tensor("op_13939_cast_fp16")]; tensor var_13940_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1455_cast_fp16)[name = tensor("op_13940_cast_fp16")]; tensor var_13941_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1457_cast_fp16)[name = tensor("op_13941_cast_fp16")]; tensor var_13942_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1459_cast_fp16)[name = tensor("op_13942_cast_fp16")]; tensor var_13943_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1461_cast_fp16)[name = tensor("op_13943_cast_fp16")]; tensor var_13944_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1463_cast_fp16)[name = tensor("op_13944_cast_fp16")]; tensor var_13945_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1465_cast_fp16)[name = tensor("op_13945_cast_fp16")]; tensor var_13946_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1467_cast_fp16)[name = tensor("op_13946_cast_fp16")]; tensor var_13947_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1469_cast_fp16)[name = tensor("op_13947_cast_fp16")]; tensor var_13948_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1471_cast_fp16)[name = tensor("op_13948_cast_fp16")]; tensor var_13950_equation_0 = const()[name = tensor("op_13950_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13950_cast_fp16 = einsum(equation = var_13950_equation_0, values = (var_13598_cast_fp16, var_13885_cast_fp16))[name = tensor("op_13950_cast_fp16")]; tensor var_13952_equation_0 = const()[name = tensor("op_13952_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13952_cast_fp16 = einsum(equation = var_13952_equation_0, values = (var_13598_cast_fp16, var_13886_cast_fp16))[name = tensor("op_13952_cast_fp16")]; tensor var_13954_equation_0 = const()[name = tensor("op_13954_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13954_cast_fp16 = einsum(equation = var_13954_equation_0, values = (var_13598_cast_fp16, var_13887_cast_fp16))[name = tensor("op_13954_cast_fp16")]; tensor var_13956_equation_0 = const()[name = tensor("op_13956_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13956_cast_fp16 = einsum(equation = var_13956_equation_0, values = (var_13598_cast_fp16, var_13888_cast_fp16))[name = tensor("op_13956_cast_fp16")]; tensor var_13958_equation_0 = const()[name = tensor("op_13958_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13958_cast_fp16 = einsum(equation = var_13958_equation_0, values = (var_13598_cast_fp16, var_13889_cast_fp16))[name = tensor("op_13958_cast_fp16")]; tensor var_13960_equation_0 = const()[name = tensor("op_13960_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13960_cast_fp16 = einsum(equation = var_13960_equation_0, values = (var_13598_cast_fp16, var_13890_cast_fp16))[name = tensor("op_13960_cast_fp16")]; tensor var_13962_equation_0 = const()[name = tensor("op_13962_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13962_cast_fp16 = einsum(equation = var_13962_equation_0, values = (var_13598_cast_fp16, var_13891_cast_fp16))[name = tensor("op_13962_cast_fp16")]; tensor var_13964_equation_0 = const()[name = tensor("op_13964_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13964_cast_fp16 = einsum(equation = var_13964_equation_0, values = (var_13598_cast_fp16, var_13892_cast_fp16))[name = tensor("op_13964_cast_fp16")]; tensor var_13966_equation_0 = const()[name = tensor("op_13966_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13966_cast_fp16 = einsum(equation = var_13966_equation_0, values = (var_13602_cast_fp16, var_13893_cast_fp16))[name = tensor("op_13966_cast_fp16")]; tensor var_13968_equation_0 = const()[name = tensor("op_13968_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13968_cast_fp16 = einsum(equation = var_13968_equation_0, values = (var_13602_cast_fp16, var_13894_cast_fp16))[name = tensor("op_13968_cast_fp16")]; tensor var_13970_equation_0 = const()[name = tensor("op_13970_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13970_cast_fp16 = einsum(equation = var_13970_equation_0, values = (var_13602_cast_fp16, var_13895_cast_fp16))[name = tensor("op_13970_cast_fp16")]; tensor var_13972_equation_0 = const()[name = tensor("op_13972_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13972_cast_fp16 = einsum(equation = var_13972_equation_0, values = (var_13602_cast_fp16, var_13896_cast_fp16))[name = tensor("op_13972_cast_fp16")]; tensor var_13974_equation_0 = const()[name = tensor("op_13974_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13974_cast_fp16 = einsum(equation = var_13974_equation_0, values = (var_13602_cast_fp16, var_13897_cast_fp16))[name = tensor("op_13974_cast_fp16")]; tensor var_13976_equation_0 = const()[name = tensor("op_13976_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13976_cast_fp16 = einsum(equation = var_13976_equation_0, values = (var_13602_cast_fp16, var_13898_cast_fp16))[name = tensor("op_13976_cast_fp16")]; tensor var_13978_equation_0 = const()[name = tensor("op_13978_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13978_cast_fp16 = einsum(equation = var_13978_equation_0, values = (var_13602_cast_fp16, var_13899_cast_fp16))[name = tensor("op_13978_cast_fp16")]; tensor var_13980_equation_0 = const()[name = tensor("op_13980_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13980_cast_fp16 = einsum(equation = var_13980_equation_0, values = (var_13602_cast_fp16, var_13900_cast_fp16))[name = tensor("op_13980_cast_fp16")]; tensor var_13982_equation_0 = const()[name = tensor("op_13982_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13982_cast_fp16 = einsum(equation = var_13982_equation_0, values = (var_13606_cast_fp16, var_13901_cast_fp16))[name = tensor("op_13982_cast_fp16")]; tensor var_13984_equation_0 = const()[name = tensor("op_13984_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13984_cast_fp16 = einsum(equation = var_13984_equation_0, values = (var_13606_cast_fp16, var_13902_cast_fp16))[name = tensor("op_13984_cast_fp16")]; tensor var_13986_equation_0 = const()[name = tensor("op_13986_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13986_cast_fp16 = einsum(equation = var_13986_equation_0, values = (var_13606_cast_fp16, var_13903_cast_fp16))[name = tensor("op_13986_cast_fp16")]; tensor var_13988_equation_0 = const()[name = tensor("op_13988_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13988_cast_fp16 = einsum(equation = var_13988_equation_0, values = (var_13606_cast_fp16, var_13904_cast_fp16))[name = tensor("op_13988_cast_fp16")]; tensor var_13990_equation_0 = const()[name = tensor("op_13990_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13990_cast_fp16 = einsum(equation = var_13990_equation_0, values = (var_13606_cast_fp16, var_13905_cast_fp16))[name = tensor("op_13990_cast_fp16")]; tensor var_13992_equation_0 = const()[name = tensor("op_13992_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13992_cast_fp16 = einsum(equation = var_13992_equation_0, values = (var_13606_cast_fp16, var_13906_cast_fp16))[name = tensor("op_13992_cast_fp16")]; tensor var_13994_equation_0 = const()[name = tensor("op_13994_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13994_cast_fp16 = einsum(equation = var_13994_equation_0, values = (var_13606_cast_fp16, var_13907_cast_fp16))[name = tensor("op_13994_cast_fp16")]; tensor var_13996_equation_0 = const()[name = tensor("op_13996_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13996_cast_fp16 = einsum(equation = var_13996_equation_0, values = (var_13606_cast_fp16, var_13908_cast_fp16))[name = tensor("op_13996_cast_fp16")]; tensor var_13998_equation_0 = const()[name = tensor("op_13998_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_13998_cast_fp16 = einsum(equation = var_13998_equation_0, values = (var_13610_cast_fp16, var_13909_cast_fp16))[name = tensor("op_13998_cast_fp16")]; tensor var_14000_equation_0 = const()[name = tensor("op_14000_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14000_cast_fp16 = einsum(equation = var_14000_equation_0, values = (var_13610_cast_fp16, var_13910_cast_fp16))[name = tensor("op_14000_cast_fp16")]; tensor var_14002_equation_0 = const()[name = tensor("op_14002_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14002_cast_fp16 = einsum(equation = var_14002_equation_0, values = (var_13610_cast_fp16, var_13911_cast_fp16))[name = tensor("op_14002_cast_fp16")]; tensor var_14004_equation_0 = const()[name = tensor("op_14004_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14004_cast_fp16 = einsum(equation = var_14004_equation_0, values = (var_13610_cast_fp16, var_13912_cast_fp16))[name = tensor("op_14004_cast_fp16")]; tensor var_14006_equation_0 = const()[name = tensor("op_14006_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14006_cast_fp16 = einsum(equation = var_14006_equation_0, values = (var_13610_cast_fp16, var_13913_cast_fp16))[name = tensor("op_14006_cast_fp16")]; tensor var_14008_equation_0 = const()[name = tensor("op_14008_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14008_cast_fp16 = einsum(equation = var_14008_equation_0, values = (var_13610_cast_fp16, var_13914_cast_fp16))[name = tensor("op_14008_cast_fp16")]; tensor var_14010_equation_0 = const()[name = tensor("op_14010_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14010_cast_fp16 = einsum(equation = var_14010_equation_0, values = (var_13610_cast_fp16, var_13915_cast_fp16))[name = tensor("op_14010_cast_fp16")]; tensor var_14012_equation_0 = const()[name = tensor("op_14012_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14012_cast_fp16 = einsum(equation = var_14012_equation_0, values = (var_13610_cast_fp16, var_13916_cast_fp16))[name = tensor("op_14012_cast_fp16")]; tensor var_14014_equation_0 = const()[name = tensor("op_14014_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14014_cast_fp16 = einsum(equation = var_14014_equation_0, values = (var_13614_cast_fp16, var_13917_cast_fp16))[name = tensor("op_14014_cast_fp16")]; tensor var_14016_equation_0 = const()[name = tensor("op_14016_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14016_cast_fp16 = einsum(equation = var_14016_equation_0, values = (var_13614_cast_fp16, var_13918_cast_fp16))[name = tensor("op_14016_cast_fp16")]; tensor var_14018_equation_0 = const()[name = tensor("op_14018_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14018_cast_fp16 = einsum(equation = var_14018_equation_0, values = (var_13614_cast_fp16, var_13919_cast_fp16))[name = tensor("op_14018_cast_fp16")]; tensor var_14020_equation_0 = const()[name = tensor("op_14020_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14020_cast_fp16 = einsum(equation = var_14020_equation_0, values = (var_13614_cast_fp16, var_13920_cast_fp16))[name = tensor("op_14020_cast_fp16")]; tensor var_14022_equation_0 = const()[name = tensor("op_14022_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14022_cast_fp16 = einsum(equation = var_14022_equation_0, values = (var_13614_cast_fp16, var_13921_cast_fp16))[name = tensor("op_14022_cast_fp16")]; tensor var_14024_equation_0 = const()[name = tensor("op_14024_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14024_cast_fp16 = einsum(equation = var_14024_equation_0, values = (var_13614_cast_fp16, var_13922_cast_fp16))[name = tensor("op_14024_cast_fp16")]; tensor var_14026_equation_0 = const()[name = tensor("op_14026_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14026_cast_fp16 = einsum(equation = var_14026_equation_0, values = (var_13614_cast_fp16, var_13923_cast_fp16))[name = tensor("op_14026_cast_fp16")]; tensor var_14028_equation_0 = const()[name = tensor("op_14028_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14028_cast_fp16 = einsum(equation = var_14028_equation_0, values = (var_13614_cast_fp16, var_13924_cast_fp16))[name = tensor("op_14028_cast_fp16")]; tensor var_14030_equation_0 = const()[name = tensor("op_14030_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14030_cast_fp16 = einsum(equation = var_14030_equation_0, values = (var_13618_cast_fp16, var_13925_cast_fp16))[name = tensor("op_14030_cast_fp16")]; tensor var_14032_equation_0 = const()[name = tensor("op_14032_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14032_cast_fp16 = einsum(equation = var_14032_equation_0, values = (var_13618_cast_fp16, var_13926_cast_fp16))[name = tensor("op_14032_cast_fp16")]; tensor var_14034_equation_0 = const()[name = tensor("op_14034_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14034_cast_fp16 = einsum(equation = var_14034_equation_0, values = (var_13618_cast_fp16, var_13927_cast_fp16))[name = tensor("op_14034_cast_fp16")]; tensor var_14036_equation_0 = const()[name = tensor("op_14036_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14036_cast_fp16 = einsum(equation = var_14036_equation_0, values = (var_13618_cast_fp16, var_13928_cast_fp16))[name = tensor("op_14036_cast_fp16")]; tensor var_14038_equation_0 = const()[name = tensor("op_14038_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14038_cast_fp16 = einsum(equation = var_14038_equation_0, values = (var_13618_cast_fp16, var_13929_cast_fp16))[name = tensor("op_14038_cast_fp16")]; tensor var_14040_equation_0 = const()[name = tensor("op_14040_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14040_cast_fp16 = einsum(equation = var_14040_equation_0, values = (var_13618_cast_fp16, var_13930_cast_fp16))[name = tensor("op_14040_cast_fp16")]; tensor var_14042_equation_0 = const()[name = tensor("op_14042_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14042_cast_fp16 = einsum(equation = var_14042_equation_0, values = (var_13618_cast_fp16, var_13931_cast_fp16))[name = tensor("op_14042_cast_fp16")]; tensor var_14044_equation_0 = const()[name = tensor("op_14044_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14044_cast_fp16 = einsum(equation = var_14044_equation_0, values = (var_13618_cast_fp16, var_13932_cast_fp16))[name = tensor("op_14044_cast_fp16")]; tensor var_14046_equation_0 = const()[name = tensor("op_14046_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14046_cast_fp16 = einsum(equation = var_14046_equation_0, values = (var_13622_cast_fp16, var_13933_cast_fp16))[name = tensor("op_14046_cast_fp16")]; tensor var_14048_equation_0 = const()[name = tensor("op_14048_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14048_cast_fp16 = einsum(equation = var_14048_equation_0, values = (var_13622_cast_fp16, var_13934_cast_fp16))[name = tensor("op_14048_cast_fp16")]; tensor var_14050_equation_0 = const()[name = tensor("op_14050_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14050_cast_fp16 = einsum(equation = var_14050_equation_0, values = (var_13622_cast_fp16, var_13935_cast_fp16))[name = tensor("op_14050_cast_fp16")]; tensor var_14052_equation_0 = const()[name = tensor("op_14052_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14052_cast_fp16 = einsum(equation = var_14052_equation_0, values = (var_13622_cast_fp16, var_13936_cast_fp16))[name = tensor("op_14052_cast_fp16")]; tensor var_14054_equation_0 = const()[name = tensor("op_14054_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14054_cast_fp16 = einsum(equation = var_14054_equation_0, values = (var_13622_cast_fp16, var_13937_cast_fp16))[name = tensor("op_14054_cast_fp16")]; tensor var_14056_equation_0 = const()[name = tensor("op_14056_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14056_cast_fp16 = einsum(equation = var_14056_equation_0, values = (var_13622_cast_fp16, var_13938_cast_fp16))[name = tensor("op_14056_cast_fp16")]; tensor var_14058_equation_0 = const()[name = tensor("op_14058_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14058_cast_fp16 = einsum(equation = var_14058_equation_0, values = (var_13622_cast_fp16, var_13939_cast_fp16))[name = tensor("op_14058_cast_fp16")]; tensor var_14060_equation_0 = const()[name = tensor("op_14060_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14060_cast_fp16 = einsum(equation = var_14060_equation_0, values = (var_13622_cast_fp16, var_13940_cast_fp16))[name = tensor("op_14060_cast_fp16")]; tensor var_14062_equation_0 = const()[name = tensor("op_14062_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14062_cast_fp16 = einsum(equation = var_14062_equation_0, values = (var_13626_cast_fp16, var_13941_cast_fp16))[name = tensor("op_14062_cast_fp16")]; tensor var_14064_equation_0 = const()[name = tensor("op_14064_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14064_cast_fp16 = einsum(equation = var_14064_equation_0, values = (var_13626_cast_fp16, var_13942_cast_fp16))[name = tensor("op_14064_cast_fp16")]; tensor var_14066_equation_0 = const()[name = tensor("op_14066_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14066_cast_fp16 = einsum(equation = var_14066_equation_0, values = (var_13626_cast_fp16, var_13943_cast_fp16))[name = tensor("op_14066_cast_fp16")]; tensor var_14068_equation_0 = const()[name = tensor("op_14068_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14068_cast_fp16 = einsum(equation = var_14068_equation_0, values = (var_13626_cast_fp16, var_13944_cast_fp16))[name = tensor("op_14068_cast_fp16")]; tensor var_14070_equation_0 = const()[name = tensor("op_14070_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14070_cast_fp16 = einsum(equation = var_14070_equation_0, values = (var_13626_cast_fp16, var_13945_cast_fp16))[name = tensor("op_14070_cast_fp16")]; tensor var_14072_equation_0 = const()[name = tensor("op_14072_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14072_cast_fp16 = einsum(equation = var_14072_equation_0, values = (var_13626_cast_fp16, var_13946_cast_fp16))[name = tensor("op_14072_cast_fp16")]; tensor var_14074_equation_0 = const()[name = tensor("op_14074_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14074_cast_fp16 = einsum(equation = var_14074_equation_0, values = (var_13626_cast_fp16, var_13947_cast_fp16))[name = tensor("op_14074_cast_fp16")]; tensor var_14076_equation_0 = const()[name = tensor("op_14076_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14076_cast_fp16 = einsum(equation = var_14076_equation_0, values = (var_13626_cast_fp16, var_13948_cast_fp16))[name = tensor("op_14076_cast_fp16")]; tensor var_14078_interleave_0 = const()[name = tensor("op_14078_interleave_0"), val = tensor(false)]; tensor var_14078_cast_fp16 = concat(axis = var_10347, interleave = var_14078_interleave_0, values = (var_13950_cast_fp16, var_13952_cast_fp16, var_13954_cast_fp16, var_13956_cast_fp16, var_13958_cast_fp16, var_13960_cast_fp16, var_13962_cast_fp16, var_13964_cast_fp16))[name = tensor("op_14078_cast_fp16")]; tensor var_14080_interleave_0 = const()[name = tensor("op_14080_interleave_0"), val = tensor(false)]; tensor var_14080_cast_fp16 = concat(axis = var_10347, interleave = var_14080_interleave_0, values = (var_13966_cast_fp16, var_13968_cast_fp16, var_13970_cast_fp16, var_13972_cast_fp16, var_13974_cast_fp16, var_13976_cast_fp16, var_13978_cast_fp16, var_13980_cast_fp16))[name = tensor("op_14080_cast_fp16")]; tensor var_14082_interleave_0 = const()[name = tensor("op_14082_interleave_0"), val = tensor(false)]; tensor var_14082_cast_fp16 = concat(axis = var_10347, interleave = var_14082_interleave_0, values = (var_13982_cast_fp16, var_13984_cast_fp16, var_13986_cast_fp16, var_13988_cast_fp16, var_13990_cast_fp16, var_13992_cast_fp16, var_13994_cast_fp16, var_13996_cast_fp16))[name = tensor("op_14082_cast_fp16")]; tensor var_14084_interleave_0 = const()[name = tensor("op_14084_interleave_0"), val = tensor(false)]; tensor var_14084_cast_fp16 = concat(axis = var_10347, interleave = var_14084_interleave_0, values = (var_13998_cast_fp16, var_14000_cast_fp16, var_14002_cast_fp16, var_14004_cast_fp16, var_14006_cast_fp16, var_14008_cast_fp16, var_14010_cast_fp16, var_14012_cast_fp16))[name = tensor("op_14084_cast_fp16")]; tensor var_14086_interleave_0 = const()[name = tensor("op_14086_interleave_0"), val = tensor(false)]; tensor var_14086_cast_fp16 = concat(axis = var_10347, interleave = var_14086_interleave_0, values = (var_14014_cast_fp16, var_14016_cast_fp16, var_14018_cast_fp16, var_14020_cast_fp16, var_14022_cast_fp16, var_14024_cast_fp16, var_14026_cast_fp16, var_14028_cast_fp16))[name = tensor("op_14086_cast_fp16")]; tensor var_14088_interleave_0 = const()[name = tensor("op_14088_interleave_0"), val = tensor(false)]; tensor var_14088_cast_fp16 = concat(axis = var_10347, interleave = var_14088_interleave_0, values = (var_14030_cast_fp16, var_14032_cast_fp16, var_14034_cast_fp16, var_14036_cast_fp16, var_14038_cast_fp16, var_14040_cast_fp16, var_14042_cast_fp16, var_14044_cast_fp16))[name = tensor("op_14088_cast_fp16")]; tensor var_14090_interleave_0 = const()[name = tensor("op_14090_interleave_0"), val = tensor(false)]; tensor var_14090_cast_fp16 = concat(axis = var_10347, interleave = var_14090_interleave_0, values = (var_14046_cast_fp16, var_14048_cast_fp16, var_14050_cast_fp16, var_14052_cast_fp16, var_14054_cast_fp16, var_14056_cast_fp16, var_14058_cast_fp16, var_14060_cast_fp16))[name = tensor("op_14090_cast_fp16")]; tensor var_14092_interleave_0 = const()[name = tensor("op_14092_interleave_0"), val = tensor(false)]; tensor var_14092_cast_fp16 = concat(axis = var_10347, interleave = var_14092_interleave_0, values = (var_14062_cast_fp16, var_14064_cast_fp16, var_14066_cast_fp16, var_14068_cast_fp16, var_14070_cast_fp16, var_14072_cast_fp16, var_14074_cast_fp16, var_14076_cast_fp16))[name = tensor("op_14092_cast_fp16")]; tensor input_517_interleave_0 = const()[name = tensor("input_517_interleave_0"), val = tensor(false)]; tensor input_517_cast_fp16 = concat(axis = var_10375, interleave = input_517_interleave_0, values = (var_14078_cast_fp16, var_14080_cast_fp16, var_14082_cast_fp16, var_14084_cast_fp16, var_14086_cast_fp16, var_14088_cast_fp16, var_14090_cast_fp16, var_14092_cast_fp16))[name = tensor("input_517_cast_fp16")]; tensor var_14102_pad_type_0 = const()[name = tensor("op_14102_pad_type_0"), val = tensor("valid")]; tensor var_14102_strides_0 = const()[name = tensor("op_14102_strides_0"), val = tensor([1, 1])]; tensor var_14102_pad_0 = const()[name = tensor("op_14102_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_14102_dilations_0 = const()[name = tensor("op_14102_dilations_0"), val = tensor([1, 1])]; tensor var_14102_groups_0 = const()[name = tensor("op_14102_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(827272192)))]; tensor up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(827477056)))]; tensor var_14102_cast_fp16 = conv(bias = up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_14102_dilations_0, groups = var_14102_groups_0, pad = var_14102_pad_0, pad_type = var_14102_pad_type_0, strides = var_14102_strides_0, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_517_cast_fp16)[name = tensor("op_14102_cast_fp16")]; tensor inputs_93_cast_fp16 = add(x = var_14102_cast_fp16, y = inputs_91_cast_fp16)[name = tensor("inputs_93_cast_fp16")]; tensor hidden_states_329_axes_0 = const()[name = tensor("hidden_states_329_axes_0"), val = tensor([1])]; tensor hidden_states_329_gamma_0_to_fp16 = const()[name = tensor("hidden_states_329_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(827477760)))]; tensor hidden_states_329_beta_0_to_fp16 = const()[name = tensor("hidden_states_329_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(827478464)))]; tensor var_14112_to_fp16 = const()[name = tensor("op_14112_to_fp16"), val = tensor(0x1.5p-17)]; tensor hidden_states_329_cast_fp16 = layer_norm(axes = hidden_states_329_axes_0, beta = hidden_states_329_beta_0_to_fp16, epsilon = var_14112_to_fp16, gamma = hidden_states_329_gamma_0_to_fp16, x = inputs_93_cast_fp16)[name = tensor("hidden_states_329_cast_fp16")]; tensor q_pad_type_0 = const()[name = tensor("q_pad_type_0"), val = tensor("valid")]; tensor q_strides_0 = const()[name = tensor("q_strides_0"), val = tensor([1, 1])]; tensor q_pad_0 = const()[name = tensor("q_pad_0"), val = tensor([0, 0, 0, 0])]; tensor q_dilations_0 = const()[name = tensor("q_dilations_0"), val = tensor([1, 1])]; tensor q_groups_0 = const()[name = tensor("q_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(827479168)))]; tensor q_cast_fp16 = conv(dilations = q_dilations_0, groups = q_groups_0, pad = q_pad_0, pad_type = q_pad_type_0, strides = q_strides_0, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_329_cast_fp16)[name = tensor("q_cast_fp16")]; tensor k_125_pad_type_0 = const()[name = tensor("k_125_pad_type_0"), val = tensor("valid")]; tensor k_125_strides_0 = const()[name = tensor("k_125_strides_0"), val = tensor([1, 1])]; tensor k_125_pad_0 = const()[name = tensor("k_125_pad_0"), val = tensor([0, 0, 0, 0])]; tensor k_125_dilations_0 = const()[name = tensor("k_125_dilations_0"), val = tensor([1, 1])]; tensor k_125_groups_0 = const()[name = tensor("k_125_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(827684032)))]; tensor k_125_cast_fp16 = conv(dilations = k_125_dilations_0, groups = k_125_groups_0, pad = k_125_pad_0, pad_type = k_125_pad_type_0, strides = k_125_strides_0, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_125_cast_fp16")]; tensor v_pad_type_0 = const()[name = tensor("v_pad_type_0"), val = tensor("valid")]; tensor v_strides_0 = const()[name = tensor("v_strides_0"), val = tensor([1, 1])]; tensor v_pad_0 = const()[name = tensor("v_pad_0"), val = tensor([0, 0, 0, 0])]; tensor v_dilations_0 = const()[name = tensor("v_dilations_0"), val = tensor([1, 1])]; tensor v_groups_0 = const()[name = tensor("v_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(828175616)))]; tensor v_cast_fp16 = conv(dilations = v_dilations_0, groups = v_groups_0, pad = v_pad_0, pad_type = v_pad_type_0, strides = v_strides_0, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_cast_fp16")]; tensor var_14145_begin_0 = const()[name = tensor("op_14145_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_14145_end_0 = const()[name = tensor("op_14145_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_14145_end_mask_0 = const()[name = tensor("op_14145_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_14145_cast_fp16 = slice_by_index(begin = var_14145_begin_0, end = var_14145_end_0, end_mask = var_14145_end_mask_0, x = q_cast_fp16)[name = tensor("op_14145_cast_fp16")]; tensor var_14149_begin_0 = const()[name = tensor("op_14149_begin_0"), val = tensor([0, 40, 0, 0])]; tensor var_14149_end_0 = const()[name = tensor("op_14149_end_0"), val = tensor([2, 80, 1, 4096])]; tensor var_14149_end_mask_0 = const()[name = tensor("op_14149_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_14149_cast_fp16 = slice_by_index(begin = var_14149_begin_0, end = var_14149_end_0, end_mask = var_14149_end_mask_0, x = q_cast_fp16)[name = tensor("op_14149_cast_fp16")]; tensor var_14153_begin_0 = const()[name = tensor("op_14153_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_14153_end_0 = const()[name = tensor("op_14153_end_0"), val = tensor([2, 120, 1, 4096])]; tensor var_14153_end_mask_0 = const()[name = tensor("op_14153_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_14153_cast_fp16 = slice_by_index(begin = var_14153_begin_0, end = var_14153_end_0, end_mask = var_14153_end_mask_0, x = q_cast_fp16)[name = tensor("op_14153_cast_fp16")]; tensor var_14157_begin_0 = const()[name = tensor("op_14157_begin_0"), val = tensor([0, 120, 0, 0])]; tensor var_14157_end_0 = const()[name = tensor("op_14157_end_0"), val = tensor([2, 160, 1, 4096])]; tensor var_14157_end_mask_0 = const()[name = tensor("op_14157_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_14157_cast_fp16 = slice_by_index(begin = var_14157_begin_0, end = var_14157_end_0, end_mask = var_14157_end_mask_0, x = q_cast_fp16)[name = tensor("op_14157_cast_fp16")]; tensor var_14161_begin_0 = const()[name = tensor("op_14161_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_14161_end_0 = const()[name = tensor("op_14161_end_0"), val = tensor([2, 200, 1, 4096])]; tensor var_14161_end_mask_0 = const()[name = tensor("op_14161_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_14161_cast_fp16 = slice_by_index(begin = var_14161_begin_0, end = var_14161_end_0, end_mask = var_14161_end_mask_0, x = q_cast_fp16)[name = tensor("op_14161_cast_fp16")]; tensor var_14165_begin_0 = const()[name = tensor("op_14165_begin_0"), val = tensor([0, 200, 0, 0])]; tensor var_14165_end_0 = const()[name = tensor("op_14165_end_0"), val = tensor([2, 240, 1, 4096])]; tensor var_14165_end_mask_0 = const()[name = tensor("op_14165_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_14165_cast_fp16 = slice_by_index(begin = var_14165_begin_0, end = var_14165_end_0, end_mask = var_14165_end_mask_0, x = q_cast_fp16)[name = tensor("op_14165_cast_fp16")]; tensor var_14169_begin_0 = const()[name = tensor("op_14169_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_14169_end_0 = const()[name = tensor("op_14169_end_0"), val = tensor([2, 280, 1, 4096])]; tensor var_14169_end_mask_0 = const()[name = tensor("op_14169_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_14169_cast_fp16 = slice_by_index(begin = var_14169_begin_0, end = var_14169_end_0, end_mask = var_14169_end_mask_0, x = q_cast_fp16)[name = tensor("op_14169_cast_fp16")]; tensor var_14173_begin_0 = const()[name = tensor("op_14173_begin_0"), val = tensor([0, 280, 0, 0])]; tensor var_14173_end_0 = const()[name = tensor("op_14173_end_0"), val = tensor([2, 1, 1, 4096])]; tensor var_14173_end_mask_0 = const()[name = tensor("op_14173_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_14173_cast_fp16 = slice_by_index(begin = var_14173_begin_0, end = var_14173_end_0, end_mask = var_14173_end_mask_0, x = q_cast_fp16)[name = tensor("op_14173_cast_fp16")]; tensor var_14176_begin_0 = const()[name = tensor("op_14176_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_14176_end_0 = const()[name = tensor("op_14176_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_14176_end_mask_0 = const()[name = tensor("op_14176_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14176_cast_fp16 = slice_by_index(begin = var_14176_begin_0, end = var_14176_end_0, end_mask = var_14176_end_mask_0, x = var_14145_cast_fp16)[name = tensor("op_14176_cast_fp16")]; tensor var_14177_begin_0 = const()[name = tensor("op_14177_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_14177_end_0 = const()[name = tensor("op_14177_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_14177_end_mask_0 = const()[name = tensor("op_14177_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14177_cast_fp16 = slice_by_index(begin = var_14177_begin_0, end = var_14177_end_0, end_mask = var_14177_end_mask_0, x = var_14145_cast_fp16)[name = tensor("op_14177_cast_fp16")]; tensor var_14178_begin_0 = const()[name = tensor("op_14178_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_14178_end_0 = const()[name = tensor("op_14178_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_14178_end_mask_0 = const()[name = tensor("op_14178_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14178_cast_fp16 = slice_by_index(begin = var_14178_begin_0, end = var_14178_end_0, end_mask = var_14178_end_mask_0, x = var_14145_cast_fp16)[name = tensor("op_14178_cast_fp16")]; tensor var_14179_begin_0 = const()[name = tensor("op_14179_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_14179_end_0 = const()[name = tensor("op_14179_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_14179_end_mask_0 = const()[name = tensor("op_14179_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14179_cast_fp16 = slice_by_index(begin = var_14179_begin_0, end = var_14179_end_0, end_mask = var_14179_end_mask_0, x = var_14145_cast_fp16)[name = tensor("op_14179_cast_fp16")]; tensor var_14180_begin_0 = const()[name = tensor("op_14180_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_14180_end_0 = const()[name = tensor("op_14180_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_14180_end_mask_0 = const()[name = tensor("op_14180_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14180_cast_fp16 = slice_by_index(begin = var_14180_begin_0, end = var_14180_end_0, end_mask = var_14180_end_mask_0, x = var_14145_cast_fp16)[name = tensor("op_14180_cast_fp16")]; tensor var_14181_begin_0 = const()[name = tensor("op_14181_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_14181_end_0 = const()[name = tensor("op_14181_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_14181_end_mask_0 = const()[name = tensor("op_14181_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14181_cast_fp16 = slice_by_index(begin = var_14181_begin_0, end = var_14181_end_0, end_mask = var_14181_end_mask_0, x = var_14145_cast_fp16)[name = tensor("op_14181_cast_fp16")]; tensor var_14182_begin_0 = const()[name = tensor("op_14182_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_14182_end_0 = const()[name = tensor("op_14182_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_14182_end_mask_0 = const()[name = tensor("op_14182_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14182_cast_fp16 = slice_by_index(begin = var_14182_begin_0, end = var_14182_end_0, end_mask = var_14182_end_mask_0, x = var_14145_cast_fp16)[name = tensor("op_14182_cast_fp16")]; tensor var_14183_begin_0 = const()[name = tensor("op_14183_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_14183_end_0 = const()[name = tensor("op_14183_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_14183_end_mask_0 = const()[name = tensor("op_14183_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_14183_cast_fp16 = slice_by_index(begin = var_14183_begin_0, end = var_14183_end_0, end_mask = var_14183_end_mask_0, x = var_14145_cast_fp16)[name = tensor("op_14183_cast_fp16")]; tensor var_14184_begin_0 = const()[name = tensor("op_14184_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_14184_end_0 = const()[name = tensor("op_14184_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_14184_end_mask_0 = const()[name = tensor("op_14184_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14184_cast_fp16 = slice_by_index(begin = var_14184_begin_0, end = var_14184_end_0, end_mask = var_14184_end_mask_0, x = var_14149_cast_fp16)[name = tensor("op_14184_cast_fp16")]; tensor var_14185_begin_0 = const()[name = tensor("op_14185_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_14185_end_0 = const()[name = tensor("op_14185_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_14185_end_mask_0 = const()[name = tensor("op_14185_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14185_cast_fp16 = slice_by_index(begin = var_14185_begin_0, end = var_14185_end_0, end_mask = var_14185_end_mask_0, x = var_14149_cast_fp16)[name = tensor("op_14185_cast_fp16")]; tensor var_14186_begin_0 = const()[name = tensor("op_14186_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_14186_end_0 = const()[name = tensor("op_14186_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_14186_end_mask_0 = const()[name = tensor("op_14186_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14186_cast_fp16 = slice_by_index(begin = var_14186_begin_0, end = var_14186_end_0, end_mask = var_14186_end_mask_0, x = var_14149_cast_fp16)[name = tensor("op_14186_cast_fp16")]; tensor var_14187_begin_0 = const()[name = tensor("op_14187_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_14187_end_0 = const()[name = tensor("op_14187_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_14187_end_mask_0 = const()[name = tensor("op_14187_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14187_cast_fp16 = slice_by_index(begin = var_14187_begin_0, end = var_14187_end_0, end_mask = var_14187_end_mask_0, x = var_14149_cast_fp16)[name = tensor("op_14187_cast_fp16")]; tensor var_14188_begin_0 = const()[name = tensor("op_14188_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_14188_end_0 = const()[name = tensor("op_14188_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_14188_end_mask_0 = const()[name = tensor("op_14188_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14188_cast_fp16 = slice_by_index(begin = var_14188_begin_0, end = var_14188_end_0, end_mask = var_14188_end_mask_0, x = var_14149_cast_fp16)[name = tensor("op_14188_cast_fp16")]; tensor var_14189_begin_0 = const()[name = tensor("op_14189_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_14189_end_0 = const()[name = tensor("op_14189_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_14189_end_mask_0 = const()[name = tensor("op_14189_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14189_cast_fp16 = slice_by_index(begin = var_14189_begin_0, end = var_14189_end_0, end_mask = var_14189_end_mask_0, x = var_14149_cast_fp16)[name = tensor("op_14189_cast_fp16")]; tensor var_14190_begin_0 = const()[name = tensor("op_14190_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_14190_end_0 = const()[name = tensor("op_14190_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_14190_end_mask_0 = const()[name = tensor("op_14190_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14190_cast_fp16 = slice_by_index(begin = var_14190_begin_0, end = var_14190_end_0, end_mask = var_14190_end_mask_0, x = var_14149_cast_fp16)[name = tensor("op_14190_cast_fp16")]; tensor var_14191_begin_0 = const()[name = tensor("op_14191_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_14191_end_0 = const()[name = tensor("op_14191_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_14191_end_mask_0 = const()[name = tensor("op_14191_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_14191_cast_fp16 = slice_by_index(begin = var_14191_begin_0, end = var_14191_end_0, end_mask = var_14191_end_mask_0, x = var_14149_cast_fp16)[name = tensor("op_14191_cast_fp16")]; tensor var_14192_begin_0 = const()[name = tensor("op_14192_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_14192_end_0 = const()[name = tensor("op_14192_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_14192_end_mask_0 = const()[name = tensor("op_14192_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14192_cast_fp16 = slice_by_index(begin = var_14192_begin_0, end = var_14192_end_0, end_mask = var_14192_end_mask_0, x = var_14153_cast_fp16)[name = tensor("op_14192_cast_fp16")]; tensor var_14193_begin_0 = const()[name = tensor("op_14193_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_14193_end_0 = const()[name = tensor("op_14193_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_14193_end_mask_0 = const()[name = tensor("op_14193_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14193_cast_fp16 = slice_by_index(begin = var_14193_begin_0, end = var_14193_end_0, end_mask = var_14193_end_mask_0, x = var_14153_cast_fp16)[name = tensor("op_14193_cast_fp16")]; tensor var_14194_begin_0 = const()[name = tensor("op_14194_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_14194_end_0 = const()[name = tensor("op_14194_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_14194_end_mask_0 = const()[name = tensor("op_14194_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14194_cast_fp16 = slice_by_index(begin = var_14194_begin_0, end = var_14194_end_0, end_mask = var_14194_end_mask_0, x = var_14153_cast_fp16)[name = tensor("op_14194_cast_fp16")]; tensor var_14195_begin_0 = const()[name = tensor("op_14195_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_14195_end_0 = const()[name = tensor("op_14195_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_14195_end_mask_0 = const()[name = tensor("op_14195_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14195_cast_fp16 = slice_by_index(begin = var_14195_begin_0, end = var_14195_end_0, end_mask = var_14195_end_mask_0, x = var_14153_cast_fp16)[name = tensor("op_14195_cast_fp16")]; tensor var_14196_begin_0 = const()[name = tensor("op_14196_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_14196_end_0 = const()[name = tensor("op_14196_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_14196_end_mask_0 = const()[name = tensor("op_14196_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14196_cast_fp16 = slice_by_index(begin = var_14196_begin_0, end = var_14196_end_0, end_mask = var_14196_end_mask_0, x = var_14153_cast_fp16)[name = tensor("op_14196_cast_fp16")]; tensor var_14197_begin_0 = const()[name = tensor("op_14197_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_14197_end_0 = const()[name = tensor("op_14197_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_14197_end_mask_0 = const()[name = tensor("op_14197_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14197_cast_fp16 = slice_by_index(begin = var_14197_begin_0, end = var_14197_end_0, end_mask = var_14197_end_mask_0, x = var_14153_cast_fp16)[name = tensor("op_14197_cast_fp16")]; tensor var_14198_begin_0 = const()[name = tensor("op_14198_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_14198_end_0 = const()[name = tensor("op_14198_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_14198_end_mask_0 = const()[name = tensor("op_14198_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14198_cast_fp16 = slice_by_index(begin = var_14198_begin_0, end = var_14198_end_0, end_mask = var_14198_end_mask_0, x = var_14153_cast_fp16)[name = tensor("op_14198_cast_fp16")]; tensor var_14199_begin_0 = const()[name = tensor("op_14199_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_14199_end_0 = const()[name = tensor("op_14199_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_14199_end_mask_0 = const()[name = tensor("op_14199_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_14199_cast_fp16 = slice_by_index(begin = var_14199_begin_0, end = var_14199_end_0, end_mask = var_14199_end_mask_0, x = var_14153_cast_fp16)[name = tensor("op_14199_cast_fp16")]; tensor var_14200_begin_0 = const()[name = tensor("op_14200_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_14200_end_0 = const()[name = tensor("op_14200_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_14200_end_mask_0 = const()[name = tensor("op_14200_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14200_cast_fp16 = slice_by_index(begin = var_14200_begin_0, end = var_14200_end_0, end_mask = var_14200_end_mask_0, x = var_14157_cast_fp16)[name = tensor("op_14200_cast_fp16")]; tensor var_14201_begin_0 = const()[name = tensor("op_14201_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_14201_end_0 = const()[name = tensor("op_14201_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_14201_end_mask_0 = const()[name = tensor("op_14201_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14201_cast_fp16 = slice_by_index(begin = var_14201_begin_0, end = var_14201_end_0, end_mask = var_14201_end_mask_0, x = var_14157_cast_fp16)[name = tensor("op_14201_cast_fp16")]; tensor var_14202_begin_0 = const()[name = tensor("op_14202_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_14202_end_0 = const()[name = tensor("op_14202_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_14202_end_mask_0 = const()[name = tensor("op_14202_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14202_cast_fp16 = slice_by_index(begin = var_14202_begin_0, end = var_14202_end_0, end_mask = var_14202_end_mask_0, x = var_14157_cast_fp16)[name = tensor("op_14202_cast_fp16")]; tensor var_14203_begin_0 = const()[name = tensor("op_14203_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_14203_end_0 = const()[name = tensor("op_14203_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_14203_end_mask_0 = const()[name = tensor("op_14203_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14203_cast_fp16 = slice_by_index(begin = var_14203_begin_0, end = var_14203_end_0, end_mask = var_14203_end_mask_0, x = var_14157_cast_fp16)[name = tensor("op_14203_cast_fp16")]; tensor var_14204_begin_0 = const()[name = tensor("op_14204_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_14204_end_0 = const()[name = tensor("op_14204_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_14204_end_mask_0 = const()[name = tensor("op_14204_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14204_cast_fp16 = slice_by_index(begin = var_14204_begin_0, end = var_14204_end_0, end_mask = var_14204_end_mask_0, x = var_14157_cast_fp16)[name = tensor("op_14204_cast_fp16")]; tensor var_14205_begin_0 = const()[name = tensor("op_14205_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_14205_end_0 = const()[name = tensor("op_14205_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_14205_end_mask_0 = const()[name = tensor("op_14205_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14205_cast_fp16 = slice_by_index(begin = var_14205_begin_0, end = var_14205_end_0, end_mask = var_14205_end_mask_0, x = var_14157_cast_fp16)[name = tensor("op_14205_cast_fp16")]; tensor var_14206_begin_0 = const()[name = tensor("op_14206_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_14206_end_0 = const()[name = tensor("op_14206_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_14206_end_mask_0 = const()[name = tensor("op_14206_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14206_cast_fp16 = slice_by_index(begin = var_14206_begin_0, end = var_14206_end_0, end_mask = var_14206_end_mask_0, x = var_14157_cast_fp16)[name = tensor("op_14206_cast_fp16")]; tensor var_14207_begin_0 = const()[name = tensor("op_14207_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_14207_end_0 = const()[name = tensor("op_14207_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_14207_end_mask_0 = const()[name = tensor("op_14207_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_14207_cast_fp16 = slice_by_index(begin = var_14207_begin_0, end = var_14207_end_0, end_mask = var_14207_end_mask_0, x = var_14157_cast_fp16)[name = tensor("op_14207_cast_fp16")]; tensor var_14208_begin_0 = const()[name = tensor("op_14208_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_14208_end_0 = const()[name = tensor("op_14208_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_14208_end_mask_0 = const()[name = tensor("op_14208_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14208_cast_fp16 = slice_by_index(begin = var_14208_begin_0, end = var_14208_end_0, end_mask = var_14208_end_mask_0, x = var_14161_cast_fp16)[name = tensor("op_14208_cast_fp16")]; tensor var_14209_begin_0 = const()[name = tensor("op_14209_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_14209_end_0 = const()[name = tensor("op_14209_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_14209_end_mask_0 = const()[name = tensor("op_14209_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14209_cast_fp16 = slice_by_index(begin = var_14209_begin_0, end = var_14209_end_0, end_mask = var_14209_end_mask_0, x = var_14161_cast_fp16)[name = tensor("op_14209_cast_fp16")]; tensor var_14210_begin_0 = const()[name = tensor("op_14210_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_14210_end_0 = const()[name = tensor("op_14210_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_14210_end_mask_0 = const()[name = tensor("op_14210_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14210_cast_fp16 = slice_by_index(begin = var_14210_begin_0, end = var_14210_end_0, end_mask = var_14210_end_mask_0, x = var_14161_cast_fp16)[name = tensor("op_14210_cast_fp16")]; tensor var_14211_begin_0 = const()[name = tensor("op_14211_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_14211_end_0 = const()[name = tensor("op_14211_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_14211_end_mask_0 = const()[name = tensor("op_14211_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14211_cast_fp16 = slice_by_index(begin = var_14211_begin_0, end = var_14211_end_0, end_mask = var_14211_end_mask_0, x = var_14161_cast_fp16)[name = tensor("op_14211_cast_fp16")]; tensor var_14212_begin_0 = const()[name = tensor("op_14212_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_14212_end_0 = const()[name = tensor("op_14212_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_14212_end_mask_0 = const()[name = tensor("op_14212_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14212_cast_fp16 = slice_by_index(begin = var_14212_begin_0, end = var_14212_end_0, end_mask = var_14212_end_mask_0, x = var_14161_cast_fp16)[name = tensor("op_14212_cast_fp16")]; tensor var_14213_begin_0 = const()[name = tensor("op_14213_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_14213_end_0 = const()[name = tensor("op_14213_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_14213_end_mask_0 = const()[name = tensor("op_14213_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14213_cast_fp16 = slice_by_index(begin = var_14213_begin_0, end = var_14213_end_0, end_mask = var_14213_end_mask_0, x = var_14161_cast_fp16)[name = tensor("op_14213_cast_fp16")]; tensor var_14214_begin_0 = const()[name = tensor("op_14214_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_14214_end_0 = const()[name = tensor("op_14214_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_14214_end_mask_0 = const()[name = tensor("op_14214_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14214_cast_fp16 = slice_by_index(begin = var_14214_begin_0, end = var_14214_end_0, end_mask = var_14214_end_mask_0, x = var_14161_cast_fp16)[name = tensor("op_14214_cast_fp16")]; tensor var_14215_begin_0 = const()[name = tensor("op_14215_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_14215_end_0 = const()[name = tensor("op_14215_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_14215_end_mask_0 = const()[name = tensor("op_14215_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_14215_cast_fp16 = slice_by_index(begin = var_14215_begin_0, end = var_14215_end_0, end_mask = var_14215_end_mask_0, x = var_14161_cast_fp16)[name = tensor("op_14215_cast_fp16")]; tensor var_14216_begin_0 = const()[name = tensor("op_14216_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_14216_end_0 = const()[name = tensor("op_14216_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_14216_end_mask_0 = const()[name = tensor("op_14216_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14216_cast_fp16 = slice_by_index(begin = var_14216_begin_0, end = var_14216_end_0, end_mask = var_14216_end_mask_0, x = var_14165_cast_fp16)[name = tensor("op_14216_cast_fp16")]; tensor var_14217_begin_0 = const()[name = tensor("op_14217_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_14217_end_0 = const()[name = tensor("op_14217_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_14217_end_mask_0 = const()[name = tensor("op_14217_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14217_cast_fp16 = slice_by_index(begin = var_14217_begin_0, end = var_14217_end_0, end_mask = var_14217_end_mask_0, x = var_14165_cast_fp16)[name = tensor("op_14217_cast_fp16")]; tensor var_14218_begin_0 = const()[name = tensor("op_14218_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_14218_end_0 = const()[name = tensor("op_14218_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_14218_end_mask_0 = const()[name = tensor("op_14218_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14218_cast_fp16 = slice_by_index(begin = var_14218_begin_0, end = var_14218_end_0, end_mask = var_14218_end_mask_0, x = var_14165_cast_fp16)[name = tensor("op_14218_cast_fp16")]; tensor var_14219_begin_0 = const()[name = tensor("op_14219_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_14219_end_0 = const()[name = tensor("op_14219_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_14219_end_mask_0 = const()[name = tensor("op_14219_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14219_cast_fp16 = slice_by_index(begin = var_14219_begin_0, end = var_14219_end_0, end_mask = var_14219_end_mask_0, x = var_14165_cast_fp16)[name = tensor("op_14219_cast_fp16")]; tensor var_14220_begin_0 = const()[name = tensor("op_14220_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_14220_end_0 = const()[name = tensor("op_14220_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_14220_end_mask_0 = const()[name = tensor("op_14220_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14220_cast_fp16 = slice_by_index(begin = var_14220_begin_0, end = var_14220_end_0, end_mask = var_14220_end_mask_0, x = var_14165_cast_fp16)[name = tensor("op_14220_cast_fp16")]; tensor var_14221_begin_0 = const()[name = tensor("op_14221_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_14221_end_0 = const()[name = tensor("op_14221_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_14221_end_mask_0 = const()[name = tensor("op_14221_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14221_cast_fp16 = slice_by_index(begin = var_14221_begin_0, end = var_14221_end_0, end_mask = var_14221_end_mask_0, x = var_14165_cast_fp16)[name = tensor("op_14221_cast_fp16")]; tensor var_14222_begin_0 = const()[name = tensor("op_14222_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_14222_end_0 = const()[name = tensor("op_14222_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_14222_end_mask_0 = const()[name = tensor("op_14222_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14222_cast_fp16 = slice_by_index(begin = var_14222_begin_0, end = var_14222_end_0, end_mask = var_14222_end_mask_0, x = var_14165_cast_fp16)[name = tensor("op_14222_cast_fp16")]; tensor var_14223_begin_0 = const()[name = tensor("op_14223_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_14223_end_0 = const()[name = tensor("op_14223_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_14223_end_mask_0 = const()[name = tensor("op_14223_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_14223_cast_fp16 = slice_by_index(begin = var_14223_begin_0, end = var_14223_end_0, end_mask = var_14223_end_mask_0, x = var_14165_cast_fp16)[name = tensor("op_14223_cast_fp16")]; tensor var_14224_begin_0 = const()[name = tensor("op_14224_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_14224_end_0 = const()[name = tensor("op_14224_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_14224_end_mask_0 = const()[name = tensor("op_14224_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14224_cast_fp16 = slice_by_index(begin = var_14224_begin_0, end = var_14224_end_0, end_mask = var_14224_end_mask_0, x = var_14169_cast_fp16)[name = tensor("op_14224_cast_fp16")]; tensor var_14225_begin_0 = const()[name = tensor("op_14225_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_14225_end_0 = const()[name = tensor("op_14225_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_14225_end_mask_0 = const()[name = tensor("op_14225_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14225_cast_fp16 = slice_by_index(begin = var_14225_begin_0, end = var_14225_end_0, end_mask = var_14225_end_mask_0, x = var_14169_cast_fp16)[name = tensor("op_14225_cast_fp16")]; tensor var_14226_begin_0 = const()[name = tensor("op_14226_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_14226_end_0 = const()[name = tensor("op_14226_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_14226_end_mask_0 = const()[name = tensor("op_14226_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14226_cast_fp16 = slice_by_index(begin = var_14226_begin_0, end = var_14226_end_0, end_mask = var_14226_end_mask_0, x = var_14169_cast_fp16)[name = tensor("op_14226_cast_fp16")]; tensor var_14227_begin_0 = const()[name = tensor("op_14227_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_14227_end_0 = const()[name = tensor("op_14227_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_14227_end_mask_0 = const()[name = tensor("op_14227_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14227_cast_fp16 = slice_by_index(begin = var_14227_begin_0, end = var_14227_end_0, end_mask = var_14227_end_mask_0, x = var_14169_cast_fp16)[name = tensor("op_14227_cast_fp16")]; tensor var_14228_begin_0 = const()[name = tensor("op_14228_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_14228_end_0 = const()[name = tensor("op_14228_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_14228_end_mask_0 = const()[name = tensor("op_14228_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14228_cast_fp16 = slice_by_index(begin = var_14228_begin_0, end = var_14228_end_0, end_mask = var_14228_end_mask_0, x = var_14169_cast_fp16)[name = tensor("op_14228_cast_fp16")]; tensor var_14229_begin_0 = const()[name = tensor("op_14229_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_14229_end_0 = const()[name = tensor("op_14229_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_14229_end_mask_0 = const()[name = tensor("op_14229_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14229_cast_fp16 = slice_by_index(begin = var_14229_begin_0, end = var_14229_end_0, end_mask = var_14229_end_mask_0, x = var_14169_cast_fp16)[name = tensor("op_14229_cast_fp16")]; tensor var_14230_begin_0 = const()[name = tensor("op_14230_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_14230_end_0 = const()[name = tensor("op_14230_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_14230_end_mask_0 = const()[name = tensor("op_14230_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14230_cast_fp16 = slice_by_index(begin = var_14230_begin_0, end = var_14230_end_0, end_mask = var_14230_end_mask_0, x = var_14169_cast_fp16)[name = tensor("op_14230_cast_fp16")]; tensor var_14231_begin_0 = const()[name = tensor("op_14231_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_14231_end_0 = const()[name = tensor("op_14231_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_14231_end_mask_0 = const()[name = tensor("op_14231_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_14231_cast_fp16 = slice_by_index(begin = var_14231_begin_0, end = var_14231_end_0, end_mask = var_14231_end_mask_0, x = var_14169_cast_fp16)[name = tensor("op_14231_cast_fp16")]; tensor var_14232_begin_0 = const()[name = tensor("op_14232_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_14232_end_0 = const()[name = tensor("op_14232_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_14232_end_mask_0 = const()[name = tensor("op_14232_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14232_cast_fp16 = slice_by_index(begin = var_14232_begin_0, end = var_14232_end_0, end_mask = var_14232_end_mask_0, x = var_14173_cast_fp16)[name = tensor("op_14232_cast_fp16")]; tensor var_14233_begin_0 = const()[name = tensor("op_14233_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_14233_end_0 = const()[name = tensor("op_14233_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_14233_end_mask_0 = const()[name = tensor("op_14233_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14233_cast_fp16 = slice_by_index(begin = var_14233_begin_0, end = var_14233_end_0, end_mask = var_14233_end_mask_0, x = var_14173_cast_fp16)[name = tensor("op_14233_cast_fp16")]; tensor var_14234_begin_0 = const()[name = tensor("op_14234_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_14234_end_0 = const()[name = tensor("op_14234_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_14234_end_mask_0 = const()[name = tensor("op_14234_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14234_cast_fp16 = slice_by_index(begin = var_14234_begin_0, end = var_14234_end_0, end_mask = var_14234_end_mask_0, x = var_14173_cast_fp16)[name = tensor("op_14234_cast_fp16")]; tensor var_14235_begin_0 = const()[name = tensor("op_14235_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_14235_end_0 = const()[name = tensor("op_14235_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_14235_end_mask_0 = const()[name = tensor("op_14235_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14235_cast_fp16 = slice_by_index(begin = var_14235_begin_0, end = var_14235_end_0, end_mask = var_14235_end_mask_0, x = var_14173_cast_fp16)[name = tensor("op_14235_cast_fp16")]; tensor var_14236_begin_0 = const()[name = tensor("op_14236_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_14236_end_0 = const()[name = tensor("op_14236_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_14236_end_mask_0 = const()[name = tensor("op_14236_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14236_cast_fp16 = slice_by_index(begin = var_14236_begin_0, end = var_14236_end_0, end_mask = var_14236_end_mask_0, x = var_14173_cast_fp16)[name = tensor("op_14236_cast_fp16")]; tensor var_14237_begin_0 = const()[name = tensor("op_14237_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_14237_end_0 = const()[name = tensor("op_14237_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_14237_end_mask_0 = const()[name = tensor("op_14237_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14237_cast_fp16 = slice_by_index(begin = var_14237_begin_0, end = var_14237_end_0, end_mask = var_14237_end_mask_0, x = var_14173_cast_fp16)[name = tensor("op_14237_cast_fp16")]; tensor var_14238_begin_0 = const()[name = tensor("op_14238_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_14238_end_0 = const()[name = tensor("op_14238_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_14238_end_mask_0 = const()[name = tensor("op_14238_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14238_cast_fp16 = slice_by_index(begin = var_14238_begin_0, end = var_14238_end_0, end_mask = var_14238_end_mask_0, x = var_14173_cast_fp16)[name = tensor("op_14238_cast_fp16")]; tensor var_14239_begin_0 = const()[name = tensor("op_14239_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_14239_end_0 = const()[name = tensor("op_14239_end_0"), val = tensor([2, 40, 1, 1])]; tensor var_14239_end_mask_0 = const()[name = tensor("op_14239_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_14239_cast_fp16 = slice_by_index(begin = var_14239_begin_0, end = var_14239_end_0, end_mask = var_14239_end_mask_0, x = var_14173_cast_fp16)[name = tensor("op_14239_cast_fp16")]; tensor k_perm_0 = const()[name = tensor("k_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_14244_begin_0 = const()[name = tensor("op_14244_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_14244_end_0 = const()[name = tensor("op_14244_end_0"), val = tensor([2, 77, 1, 40])]; tensor var_14244_end_mask_0 = const()[name = tensor("op_14244_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_cast_fp16 = transpose(perm = k_perm_0, x = k_125_cast_fp16)[name = tensor("transpose_0")]; tensor var_14244_cast_fp16 = slice_by_index(begin = var_14244_begin_0, end = var_14244_end_0, end_mask = var_14244_end_mask_0, x = k_cast_fp16)[name = tensor("op_14244_cast_fp16")]; tensor var_14248_begin_0 = const()[name = tensor("op_14248_begin_0"), val = tensor([0, 0, 0, 40])]; tensor var_14248_end_0 = const()[name = tensor("op_14248_end_0"), val = tensor([2, 77, 1, 80])]; tensor var_14248_end_mask_0 = const()[name = tensor("op_14248_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14248_cast_fp16 = slice_by_index(begin = var_14248_begin_0, end = var_14248_end_0, end_mask = var_14248_end_mask_0, x = k_cast_fp16)[name = tensor("op_14248_cast_fp16")]; tensor var_14252_begin_0 = const()[name = tensor("op_14252_begin_0"), val = tensor([0, 0, 0, 80])]; tensor var_14252_end_0 = const()[name = tensor("op_14252_end_0"), val = tensor([2, 77, 1, 120])]; tensor var_14252_end_mask_0 = const()[name = tensor("op_14252_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14252_cast_fp16 = slice_by_index(begin = var_14252_begin_0, end = var_14252_end_0, end_mask = var_14252_end_mask_0, x = k_cast_fp16)[name = tensor("op_14252_cast_fp16")]; tensor var_14256_begin_0 = const()[name = tensor("op_14256_begin_0"), val = tensor([0, 0, 0, 120])]; tensor var_14256_end_0 = const()[name = tensor("op_14256_end_0"), val = tensor([2, 77, 1, 160])]; tensor var_14256_end_mask_0 = const()[name = tensor("op_14256_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14256_cast_fp16 = slice_by_index(begin = var_14256_begin_0, end = var_14256_end_0, end_mask = var_14256_end_mask_0, x = k_cast_fp16)[name = tensor("op_14256_cast_fp16")]; tensor var_14260_begin_0 = const()[name = tensor("op_14260_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_14260_end_0 = const()[name = tensor("op_14260_end_0"), val = tensor([2, 77, 1, 200])]; tensor var_14260_end_mask_0 = const()[name = tensor("op_14260_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14260_cast_fp16 = slice_by_index(begin = var_14260_begin_0, end = var_14260_end_0, end_mask = var_14260_end_mask_0, x = k_cast_fp16)[name = tensor("op_14260_cast_fp16")]; tensor var_14264_begin_0 = const()[name = tensor("op_14264_begin_0"), val = tensor([0, 0, 0, 200])]; tensor var_14264_end_0 = const()[name = tensor("op_14264_end_0"), val = tensor([2, 77, 1, 240])]; tensor var_14264_end_mask_0 = const()[name = tensor("op_14264_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14264_cast_fp16 = slice_by_index(begin = var_14264_begin_0, end = var_14264_end_0, end_mask = var_14264_end_mask_0, x = k_cast_fp16)[name = tensor("op_14264_cast_fp16")]; tensor var_14268_begin_0 = const()[name = tensor("op_14268_begin_0"), val = tensor([0, 0, 0, 240])]; tensor var_14268_end_0 = const()[name = tensor("op_14268_end_0"), val = tensor([2, 77, 1, 280])]; tensor var_14268_end_mask_0 = const()[name = tensor("op_14268_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_14268_cast_fp16 = slice_by_index(begin = var_14268_begin_0, end = var_14268_end_0, end_mask = var_14268_end_mask_0, x = k_cast_fp16)[name = tensor("op_14268_cast_fp16")]; tensor var_14272_begin_0 = const()[name = tensor("op_14272_begin_0"), val = tensor([0, 0, 0, 280])]; tensor var_14272_end_0 = const()[name = tensor("op_14272_end_0"), val = tensor([2, 77, 1, 1])]; tensor var_14272_end_mask_0 = const()[name = tensor("op_14272_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_14272_cast_fp16 = slice_by_index(begin = var_14272_begin_0, end = var_14272_end_0, end_mask = var_14272_end_mask_0, x = k_cast_fp16)[name = tensor("op_14272_cast_fp16")]; tensor var_14274_begin_0 = const()[name = tensor("op_14274_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_14274_end_0 = const()[name = tensor("op_14274_end_0"), val = tensor([2, 40, 1, 77])]; tensor var_14274_end_mask_0 = const()[name = tensor("op_14274_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_14274_cast_fp16 = slice_by_index(begin = var_14274_begin_0, end = var_14274_end_0, end_mask = var_14274_end_mask_0, x = v_cast_fp16)[name = tensor("op_14274_cast_fp16")]; tensor var_14278_begin_0 = const()[name = tensor("op_14278_begin_0"), val = tensor([0, 40, 0, 0])]; tensor var_14278_end_0 = const()[name = tensor("op_14278_end_0"), val = tensor([2, 80, 1, 77])]; tensor var_14278_end_mask_0 = const()[name = tensor("op_14278_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_14278_cast_fp16 = slice_by_index(begin = var_14278_begin_0, end = var_14278_end_0, end_mask = var_14278_end_mask_0, x = v_cast_fp16)[name = tensor("op_14278_cast_fp16")]; tensor var_14282_begin_0 = const()[name = tensor("op_14282_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_14282_end_0 = const()[name = tensor("op_14282_end_0"), val = tensor([2, 120, 1, 77])]; tensor var_14282_end_mask_0 = const()[name = tensor("op_14282_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_14282_cast_fp16 = slice_by_index(begin = var_14282_begin_0, end = var_14282_end_0, end_mask = var_14282_end_mask_0, x = v_cast_fp16)[name = tensor("op_14282_cast_fp16")]; tensor var_14286_begin_0 = const()[name = tensor("op_14286_begin_0"), val = tensor([0, 120, 0, 0])]; tensor var_14286_end_0 = const()[name = tensor("op_14286_end_0"), val = tensor([2, 160, 1, 77])]; tensor var_14286_end_mask_0 = const()[name = tensor("op_14286_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_14286_cast_fp16 = slice_by_index(begin = var_14286_begin_0, end = var_14286_end_0, end_mask = var_14286_end_mask_0, x = v_cast_fp16)[name = tensor("op_14286_cast_fp16")]; tensor var_14290_begin_0 = const()[name = tensor("op_14290_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_14290_end_0 = const()[name = tensor("op_14290_end_0"), val = tensor([2, 200, 1, 77])]; tensor var_14290_end_mask_0 = const()[name = tensor("op_14290_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_14290_cast_fp16 = slice_by_index(begin = var_14290_begin_0, end = var_14290_end_0, end_mask = var_14290_end_mask_0, x = v_cast_fp16)[name = tensor("op_14290_cast_fp16")]; tensor var_14294_begin_0 = const()[name = tensor("op_14294_begin_0"), val = tensor([0, 200, 0, 0])]; tensor var_14294_end_0 = const()[name = tensor("op_14294_end_0"), val = tensor([2, 240, 1, 77])]; tensor var_14294_end_mask_0 = const()[name = tensor("op_14294_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_14294_cast_fp16 = slice_by_index(begin = var_14294_begin_0, end = var_14294_end_0, end_mask = var_14294_end_mask_0, x = v_cast_fp16)[name = tensor("op_14294_cast_fp16")]; tensor var_14298_begin_0 = const()[name = tensor("op_14298_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_14298_end_0 = const()[name = tensor("op_14298_end_0"), val = tensor([2, 280, 1, 77])]; tensor var_14298_end_mask_0 = const()[name = tensor("op_14298_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_14298_cast_fp16 = slice_by_index(begin = var_14298_begin_0, end = var_14298_end_0, end_mask = var_14298_end_mask_0, x = v_cast_fp16)[name = tensor("op_14298_cast_fp16")]; tensor var_14302_begin_0 = const()[name = tensor("op_14302_begin_0"), val = tensor([0, 280, 0, 0])]; tensor var_14302_end_0 = const()[name = tensor("op_14302_end_0"), val = tensor([2, 1, 1, 77])]; tensor var_14302_end_mask_0 = const()[name = tensor("op_14302_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_14302_cast_fp16 = slice_by_index(begin = var_14302_begin_0, end = var_14302_end_0, end_mask = var_14302_end_mask_0, x = v_cast_fp16)[name = tensor("op_14302_cast_fp16")]; tensor var_14306_equation_0 = const()[name = tensor("op_14306_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14306_cast_fp16 = einsum(equation = var_14306_equation_0, values = (var_14244_cast_fp16, var_14176_cast_fp16))[name = tensor("op_14306_cast_fp16")]; tensor var_14307_to_fp16 = const()[name = tensor("op_14307_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1473_cast_fp16 = mul(x = var_14306_cast_fp16, y = var_14307_to_fp16)[name = tensor("aw_chunk_1473_cast_fp16")]; tensor var_14310_equation_0 = const()[name = tensor("op_14310_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14310_cast_fp16 = einsum(equation = var_14310_equation_0, values = (var_14244_cast_fp16, var_14177_cast_fp16))[name = tensor("op_14310_cast_fp16")]; tensor var_14311_to_fp16 = const()[name = tensor("op_14311_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1475_cast_fp16 = mul(x = var_14310_cast_fp16, y = var_14311_to_fp16)[name = tensor("aw_chunk_1475_cast_fp16")]; tensor var_14314_equation_0 = const()[name = tensor("op_14314_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14314_cast_fp16 = einsum(equation = var_14314_equation_0, values = (var_14244_cast_fp16, var_14178_cast_fp16))[name = tensor("op_14314_cast_fp16")]; tensor var_14315_to_fp16 = const()[name = tensor("op_14315_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1477_cast_fp16 = mul(x = var_14314_cast_fp16, y = var_14315_to_fp16)[name = tensor("aw_chunk_1477_cast_fp16")]; tensor var_14318_equation_0 = const()[name = tensor("op_14318_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14318_cast_fp16 = einsum(equation = var_14318_equation_0, values = (var_14244_cast_fp16, var_14179_cast_fp16))[name = tensor("op_14318_cast_fp16")]; tensor var_14319_to_fp16 = const()[name = tensor("op_14319_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1479_cast_fp16 = mul(x = var_14318_cast_fp16, y = var_14319_to_fp16)[name = tensor("aw_chunk_1479_cast_fp16")]; tensor var_14322_equation_0 = const()[name = tensor("op_14322_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14322_cast_fp16 = einsum(equation = var_14322_equation_0, values = (var_14244_cast_fp16, var_14180_cast_fp16))[name = tensor("op_14322_cast_fp16")]; tensor var_14323_to_fp16 = const()[name = tensor("op_14323_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1481_cast_fp16 = mul(x = var_14322_cast_fp16, y = var_14323_to_fp16)[name = tensor("aw_chunk_1481_cast_fp16")]; tensor var_14326_equation_0 = const()[name = tensor("op_14326_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14326_cast_fp16 = einsum(equation = var_14326_equation_0, values = (var_14244_cast_fp16, var_14181_cast_fp16))[name = tensor("op_14326_cast_fp16")]; tensor var_14327_to_fp16 = const()[name = tensor("op_14327_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1483_cast_fp16 = mul(x = var_14326_cast_fp16, y = var_14327_to_fp16)[name = tensor("aw_chunk_1483_cast_fp16")]; tensor var_14330_equation_0 = const()[name = tensor("op_14330_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14330_cast_fp16 = einsum(equation = var_14330_equation_0, values = (var_14244_cast_fp16, var_14182_cast_fp16))[name = tensor("op_14330_cast_fp16")]; tensor var_14331_to_fp16 = const()[name = tensor("op_14331_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1485_cast_fp16 = mul(x = var_14330_cast_fp16, y = var_14331_to_fp16)[name = tensor("aw_chunk_1485_cast_fp16")]; tensor var_14334_equation_0 = const()[name = tensor("op_14334_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14334_cast_fp16 = einsum(equation = var_14334_equation_0, values = (var_14244_cast_fp16, var_14183_cast_fp16))[name = tensor("op_14334_cast_fp16")]; tensor var_14335_to_fp16 = const()[name = tensor("op_14335_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1487_cast_fp16 = mul(x = var_14334_cast_fp16, y = var_14335_to_fp16)[name = tensor("aw_chunk_1487_cast_fp16")]; tensor var_14338_equation_0 = const()[name = tensor("op_14338_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14338_cast_fp16 = einsum(equation = var_14338_equation_0, values = (var_14248_cast_fp16, var_14184_cast_fp16))[name = tensor("op_14338_cast_fp16")]; tensor var_14339_to_fp16 = const()[name = tensor("op_14339_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1489_cast_fp16 = mul(x = var_14338_cast_fp16, y = var_14339_to_fp16)[name = tensor("aw_chunk_1489_cast_fp16")]; tensor var_14342_equation_0 = const()[name = tensor("op_14342_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14342_cast_fp16 = einsum(equation = var_14342_equation_0, values = (var_14248_cast_fp16, var_14185_cast_fp16))[name = tensor("op_14342_cast_fp16")]; tensor var_14343_to_fp16 = const()[name = tensor("op_14343_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1491_cast_fp16 = mul(x = var_14342_cast_fp16, y = var_14343_to_fp16)[name = tensor("aw_chunk_1491_cast_fp16")]; tensor var_14346_equation_0 = const()[name = tensor("op_14346_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14346_cast_fp16 = einsum(equation = var_14346_equation_0, values = (var_14248_cast_fp16, var_14186_cast_fp16))[name = tensor("op_14346_cast_fp16")]; tensor var_14347_to_fp16 = const()[name = tensor("op_14347_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1493_cast_fp16 = mul(x = var_14346_cast_fp16, y = var_14347_to_fp16)[name = tensor("aw_chunk_1493_cast_fp16")]; tensor var_14350_equation_0 = const()[name = tensor("op_14350_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14350_cast_fp16 = einsum(equation = var_14350_equation_0, values = (var_14248_cast_fp16, var_14187_cast_fp16))[name = tensor("op_14350_cast_fp16")]; tensor var_14351_to_fp16 = const()[name = tensor("op_14351_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1495_cast_fp16 = mul(x = var_14350_cast_fp16, y = var_14351_to_fp16)[name = tensor("aw_chunk_1495_cast_fp16")]; tensor var_14354_equation_0 = const()[name = tensor("op_14354_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14354_cast_fp16 = einsum(equation = var_14354_equation_0, values = (var_14248_cast_fp16, var_14188_cast_fp16))[name = tensor("op_14354_cast_fp16")]; tensor var_14355_to_fp16 = const()[name = tensor("op_14355_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1497_cast_fp16 = mul(x = var_14354_cast_fp16, y = var_14355_to_fp16)[name = tensor("aw_chunk_1497_cast_fp16")]; tensor var_14358_equation_0 = const()[name = tensor("op_14358_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14358_cast_fp16 = einsum(equation = var_14358_equation_0, values = (var_14248_cast_fp16, var_14189_cast_fp16))[name = tensor("op_14358_cast_fp16")]; tensor var_14359_to_fp16 = const()[name = tensor("op_14359_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1499_cast_fp16 = mul(x = var_14358_cast_fp16, y = var_14359_to_fp16)[name = tensor("aw_chunk_1499_cast_fp16")]; tensor var_14362_equation_0 = const()[name = tensor("op_14362_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14362_cast_fp16 = einsum(equation = var_14362_equation_0, values = (var_14248_cast_fp16, var_14190_cast_fp16))[name = tensor("op_14362_cast_fp16")]; tensor var_14363_to_fp16 = const()[name = tensor("op_14363_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1501_cast_fp16 = mul(x = var_14362_cast_fp16, y = var_14363_to_fp16)[name = tensor("aw_chunk_1501_cast_fp16")]; tensor var_14366_equation_0 = const()[name = tensor("op_14366_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14366_cast_fp16 = einsum(equation = var_14366_equation_0, values = (var_14248_cast_fp16, var_14191_cast_fp16))[name = tensor("op_14366_cast_fp16")]; tensor var_14367_to_fp16 = const()[name = tensor("op_14367_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1503_cast_fp16 = mul(x = var_14366_cast_fp16, y = var_14367_to_fp16)[name = tensor("aw_chunk_1503_cast_fp16")]; tensor var_14370_equation_0 = const()[name = tensor("op_14370_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14370_cast_fp16 = einsum(equation = var_14370_equation_0, values = (var_14252_cast_fp16, var_14192_cast_fp16))[name = tensor("op_14370_cast_fp16")]; tensor var_14371_to_fp16 = const()[name = tensor("op_14371_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1505_cast_fp16 = mul(x = var_14370_cast_fp16, y = var_14371_to_fp16)[name = tensor("aw_chunk_1505_cast_fp16")]; tensor var_14374_equation_0 = const()[name = tensor("op_14374_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14374_cast_fp16 = einsum(equation = var_14374_equation_0, values = (var_14252_cast_fp16, var_14193_cast_fp16))[name = tensor("op_14374_cast_fp16")]; tensor var_14375_to_fp16 = const()[name = tensor("op_14375_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1507_cast_fp16 = mul(x = var_14374_cast_fp16, y = var_14375_to_fp16)[name = tensor("aw_chunk_1507_cast_fp16")]; tensor var_14378_equation_0 = const()[name = tensor("op_14378_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14378_cast_fp16 = einsum(equation = var_14378_equation_0, values = (var_14252_cast_fp16, var_14194_cast_fp16))[name = tensor("op_14378_cast_fp16")]; tensor var_14379_to_fp16 = const()[name = tensor("op_14379_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1509_cast_fp16 = mul(x = var_14378_cast_fp16, y = var_14379_to_fp16)[name = tensor("aw_chunk_1509_cast_fp16")]; tensor var_14382_equation_0 = const()[name = tensor("op_14382_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14382_cast_fp16 = einsum(equation = var_14382_equation_0, values = (var_14252_cast_fp16, var_14195_cast_fp16))[name = tensor("op_14382_cast_fp16")]; tensor var_14383_to_fp16 = const()[name = tensor("op_14383_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1511_cast_fp16 = mul(x = var_14382_cast_fp16, y = var_14383_to_fp16)[name = tensor("aw_chunk_1511_cast_fp16")]; tensor var_14386_equation_0 = const()[name = tensor("op_14386_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14386_cast_fp16 = einsum(equation = var_14386_equation_0, values = (var_14252_cast_fp16, var_14196_cast_fp16))[name = tensor("op_14386_cast_fp16")]; tensor var_14387_to_fp16 = const()[name = tensor("op_14387_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1513_cast_fp16 = mul(x = var_14386_cast_fp16, y = var_14387_to_fp16)[name = tensor("aw_chunk_1513_cast_fp16")]; tensor var_14390_equation_0 = const()[name = tensor("op_14390_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14390_cast_fp16 = einsum(equation = var_14390_equation_0, values = (var_14252_cast_fp16, var_14197_cast_fp16))[name = tensor("op_14390_cast_fp16")]; tensor var_14391_to_fp16 = const()[name = tensor("op_14391_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1515_cast_fp16 = mul(x = var_14390_cast_fp16, y = var_14391_to_fp16)[name = tensor("aw_chunk_1515_cast_fp16")]; tensor var_14394_equation_0 = const()[name = tensor("op_14394_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14394_cast_fp16 = einsum(equation = var_14394_equation_0, values = (var_14252_cast_fp16, var_14198_cast_fp16))[name = tensor("op_14394_cast_fp16")]; tensor var_14395_to_fp16 = const()[name = tensor("op_14395_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1517_cast_fp16 = mul(x = var_14394_cast_fp16, y = var_14395_to_fp16)[name = tensor("aw_chunk_1517_cast_fp16")]; tensor var_14398_equation_0 = const()[name = tensor("op_14398_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14398_cast_fp16 = einsum(equation = var_14398_equation_0, values = (var_14252_cast_fp16, var_14199_cast_fp16))[name = tensor("op_14398_cast_fp16")]; tensor var_14399_to_fp16 = const()[name = tensor("op_14399_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1519_cast_fp16 = mul(x = var_14398_cast_fp16, y = var_14399_to_fp16)[name = tensor("aw_chunk_1519_cast_fp16")]; tensor var_14402_equation_0 = const()[name = tensor("op_14402_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14402_cast_fp16 = einsum(equation = var_14402_equation_0, values = (var_14256_cast_fp16, var_14200_cast_fp16))[name = tensor("op_14402_cast_fp16")]; tensor var_14403_to_fp16 = const()[name = tensor("op_14403_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1521_cast_fp16 = mul(x = var_14402_cast_fp16, y = var_14403_to_fp16)[name = tensor("aw_chunk_1521_cast_fp16")]; tensor var_14406_equation_0 = const()[name = tensor("op_14406_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14406_cast_fp16 = einsum(equation = var_14406_equation_0, values = (var_14256_cast_fp16, var_14201_cast_fp16))[name = tensor("op_14406_cast_fp16")]; tensor var_14407_to_fp16 = const()[name = tensor("op_14407_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1523_cast_fp16 = mul(x = var_14406_cast_fp16, y = var_14407_to_fp16)[name = tensor("aw_chunk_1523_cast_fp16")]; tensor var_14410_equation_0 = const()[name = tensor("op_14410_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14410_cast_fp16 = einsum(equation = var_14410_equation_0, values = (var_14256_cast_fp16, var_14202_cast_fp16))[name = tensor("op_14410_cast_fp16")]; tensor var_14411_to_fp16 = const()[name = tensor("op_14411_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1525_cast_fp16 = mul(x = var_14410_cast_fp16, y = var_14411_to_fp16)[name = tensor("aw_chunk_1525_cast_fp16")]; tensor var_14414_equation_0 = const()[name = tensor("op_14414_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14414_cast_fp16 = einsum(equation = var_14414_equation_0, values = (var_14256_cast_fp16, var_14203_cast_fp16))[name = tensor("op_14414_cast_fp16")]; tensor var_14415_to_fp16 = const()[name = tensor("op_14415_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1527_cast_fp16 = mul(x = var_14414_cast_fp16, y = var_14415_to_fp16)[name = tensor("aw_chunk_1527_cast_fp16")]; tensor var_14418_equation_0 = const()[name = tensor("op_14418_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14418_cast_fp16 = einsum(equation = var_14418_equation_0, values = (var_14256_cast_fp16, var_14204_cast_fp16))[name = tensor("op_14418_cast_fp16")]; tensor var_14419_to_fp16 = const()[name = tensor("op_14419_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1529_cast_fp16 = mul(x = var_14418_cast_fp16, y = var_14419_to_fp16)[name = tensor("aw_chunk_1529_cast_fp16")]; tensor var_14422_equation_0 = const()[name = tensor("op_14422_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14422_cast_fp16 = einsum(equation = var_14422_equation_0, values = (var_14256_cast_fp16, var_14205_cast_fp16))[name = tensor("op_14422_cast_fp16")]; tensor var_14423_to_fp16 = const()[name = tensor("op_14423_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1531_cast_fp16 = mul(x = var_14422_cast_fp16, y = var_14423_to_fp16)[name = tensor("aw_chunk_1531_cast_fp16")]; tensor var_14426_equation_0 = const()[name = tensor("op_14426_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14426_cast_fp16 = einsum(equation = var_14426_equation_0, values = (var_14256_cast_fp16, var_14206_cast_fp16))[name = tensor("op_14426_cast_fp16")]; tensor var_14427_to_fp16 = const()[name = tensor("op_14427_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1533_cast_fp16 = mul(x = var_14426_cast_fp16, y = var_14427_to_fp16)[name = tensor("aw_chunk_1533_cast_fp16")]; tensor var_14430_equation_0 = const()[name = tensor("op_14430_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14430_cast_fp16 = einsum(equation = var_14430_equation_0, values = (var_14256_cast_fp16, var_14207_cast_fp16))[name = tensor("op_14430_cast_fp16")]; tensor var_14431_to_fp16 = const()[name = tensor("op_14431_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1535_cast_fp16 = mul(x = var_14430_cast_fp16, y = var_14431_to_fp16)[name = tensor("aw_chunk_1535_cast_fp16")]; tensor var_14434_equation_0 = const()[name = tensor("op_14434_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14434_cast_fp16 = einsum(equation = var_14434_equation_0, values = (var_14260_cast_fp16, var_14208_cast_fp16))[name = tensor("op_14434_cast_fp16")]; tensor var_14435_to_fp16 = const()[name = tensor("op_14435_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1537_cast_fp16 = mul(x = var_14434_cast_fp16, y = var_14435_to_fp16)[name = tensor("aw_chunk_1537_cast_fp16")]; tensor var_14438_equation_0 = const()[name = tensor("op_14438_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14438_cast_fp16 = einsum(equation = var_14438_equation_0, values = (var_14260_cast_fp16, var_14209_cast_fp16))[name = tensor("op_14438_cast_fp16")]; tensor var_14439_to_fp16 = const()[name = tensor("op_14439_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1539_cast_fp16 = mul(x = var_14438_cast_fp16, y = var_14439_to_fp16)[name = tensor("aw_chunk_1539_cast_fp16")]; tensor var_14442_equation_0 = const()[name = tensor("op_14442_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14442_cast_fp16 = einsum(equation = var_14442_equation_0, values = (var_14260_cast_fp16, var_14210_cast_fp16))[name = tensor("op_14442_cast_fp16")]; tensor var_14443_to_fp16 = const()[name = tensor("op_14443_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1541_cast_fp16 = mul(x = var_14442_cast_fp16, y = var_14443_to_fp16)[name = tensor("aw_chunk_1541_cast_fp16")]; tensor var_14446_equation_0 = const()[name = tensor("op_14446_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14446_cast_fp16 = einsum(equation = var_14446_equation_0, values = (var_14260_cast_fp16, var_14211_cast_fp16))[name = tensor("op_14446_cast_fp16")]; tensor var_14447_to_fp16 = const()[name = tensor("op_14447_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1543_cast_fp16 = mul(x = var_14446_cast_fp16, y = var_14447_to_fp16)[name = tensor("aw_chunk_1543_cast_fp16")]; tensor var_14450_equation_0 = const()[name = tensor("op_14450_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14450_cast_fp16 = einsum(equation = var_14450_equation_0, values = (var_14260_cast_fp16, var_14212_cast_fp16))[name = tensor("op_14450_cast_fp16")]; tensor var_14451_to_fp16 = const()[name = tensor("op_14451_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1545_cast_fp16 = mul(x = var_14450_cast_fp16, y = var_14451_to_fp16)[name = tensor("aw_chunk_1545_cast_fp16")]; tensor var_14454_equation_0 = const()[name = tensor("op_14454_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14454_cast_fp16 = einsum(equation = var_14454_equation_0, values = (var_14260_cast_fp16, var_14213_cast_fp16))[name = tensor("op_14454_cast_fp16")]; tensor var_14455_to_fp16 = const()[name = tensor("op_14455_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1547_cast_fp16 = mul(x = var_14454_cast_fp16, y = var_14455_to_fp16)[name = tensor("aw_chunk_1547_cast_fp16")]; tensor var_14458_equation_0 = const()[name = tensor("op_14458_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14458_cast_fp16 = einsum(equation = var_14458_equation_0, values = (var_14260_cast_fp16, var_14214_cast_fp16))[name = tensor("op_14458_cast_fp16")]; tensor var_14459_to_fp16 = const()[name = tensor("op_14459_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1549_cast_fp16 = mul(x = var_14458_cast_fp16, y = var_14459_to_fp16)[name = tensor("aw_chunk_1549_cast_fp16")]; tensor var_14462_equation_0 = const()[name = tensor("op_14462_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14462_cast_fp16 = einsum(equation = var_14462_equation_0, values = (var_14260_cast_fp16, var_14215_cast_fp16))[name = tensor("op_14462_cast_fp16")]; tensor var_14463_to_fp16 = const()[name = tensor("op_14463_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1551_cast_fp16 = mul(x = var_14462_cast_fp16, y = var_14463_to_fp16)[name = tensor("aw_chunk_1551_cast_fp16")]; tensor var_14466_equation_0 = const()[name = tensor("op_14466_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14466_cast_fp16 = einsum(equation = var_14466_equation_0, values = (var_14264_cast_fp16, var_14216_cast_fp16))[name = tensor("op_14466_cast_fp16")]; tensor var_14467_to_fp16 = const()[name = tensor("op_14467_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1553_cast_fp16 = mul(x = var_14466_cast_fp16, y = var_14467_to_fp16)[name = tensor("aw_chunk_1553_cast_fp16")]; tensor var_14470_equation_0 = const()[name = tensor("op_14470_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14470_cast_fp16 = einsum(equation = var_14470_equation_0, values = (var_14264_cast_fp16, var_14217_cast_fp16))[name = tensor("op_14470_cast_fp16")]; tensor var_14471_to_fp16 = const()[name = tensor("op_14471_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1555_cast_fp16 = mul(x = var_14470_cast_fp16, y = var_14471_to_fp16)[name = tensor("aw_chunk_1555_cast_fp16")]; tensor var_14474_equation_0 = const()[name = tensor("op_14474_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14474_cast_fp16 = einsum(equation = var_14474_equation_0, values = (var_14264_cast_fp16, var_14218_cast_fp16))[name = tensor("op_14474_cast_fp16")]; tensor var_14475_to_fp16 = const()[name = tensor("op_14475_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1557_cast_fp16 = mul(x = var_14474_cast_fp16, y = var_14475_to_fp16)[name = tensor("aw_chunk_1557_cast_fp16")]; tensor var_14478_equation_0 = const()[name = tensor("op_14478_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14478_cast_fp16 = einsum(equation = var_14478_equation_0, values = (var_14264_cast_fp16, var_14219_cast_fp16))[name = tensor("op_14478_cast_fp16")]; tensor var_14479_to_fp16 = const()[name = tensor("op_14479_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1559_cast_fp16 = mul(x = var_14478_cast_fp16, y = var_14479_to_fp16)[name = tensor("aw_chunk_1559_cast_fp16")]; tensor var_14482_equation_0 = const()[name = tensor("op_14482_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14482_cast_fp16 = einsum(equation = var_14482_equation_0, values = (var_14264_cast_fp16, var_14220_cast_fp16))[name = tensor("op_14482_cast_fp16")]; tensor var_14483_to_fp16 = const()[name = tensor("op_14483_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1561_cast_fp16 = mul(x = var_14482_cast_fp16, y = var_14483_to_fp16)[name = tensor("aw_chunk_1561_cast_fp16")]; tensor var_14486_equation_0 = const()[name = tensor("op_14486_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14486_cast_fp16 = einsum(equation = var_14486_equation_0, values = (var_14264_cast_fp16, var_14221_cast_fp16))[name = tensor("op_14486_cast_fp16")]; tensor var_14487_to_fp16 = const()[name = tensor("op_14487_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1563_cast_fp16 = mul(x = var_14486_cast_fp16, y = var_14487_to_fp16)[name = tensor("aw_chunk_1563_cast_fp16")]; tensor var_14490_equation_0 = const()[name = tensor("op_14490_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14490_cast_fp16 = einsum(equation = var_14490_equation_0, values = (var_14264_cast_fp16, var_14222_cast_fp16))[name = tensor("op_14490_cast_fp16")]; tensor var_14491_to_fp16 = const()[name = tensor("op_14491_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1565_cast_fp16 = mul(x = var_14490_cast_fp16, y = var_14491_to_fp16)[name = tensor("aw_chunk_1565_cast_fp16")]; tensor var_14494_equation_0 = const()[name = tensor("op_14494_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14494_cast_fp16 = einsum(equation = var_14494_equation_0, values = (var_14264_cast_fp16, var_14223_cast_fp16))[name = tensor("op_14494_cast_fp16")]; tensor var_14495_to_fp16 = const()[name = tensor("op_14495_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1567_cast_fp16 = mul(x = var_14494_cast_fp16, y = var_14495_to_fp16)[name = tensor("aw_chunk_1567_cast_fp16")]; tensor var_14498_equation_0 = const()[name = tensor("op_14498_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14498_cast_fp16 = einsum(equation = var_14498_equation_0, values = (var_14268_cast_fp16, var_14224_cast_fp16))[name = tensor("op_14498_cast_fp16")]; tensor var_14499_to_fp16 = const()[name = tensor("op_14499_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1569_cast_fp16 = mul(x = var_14498_cast_fp16, y = var_14499_to_fp16)[name = tensor("aw_chunk_1569_cast_fp16")]; tensor var_14502_equation_0 = const()[name = tensor("op_14502_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14502_cast_fp16 = einsum(equation = var_14502_equation_0, values = (var_14268_cast_fp16, var_14225_cast_fp16))[name = tensor("op_14502_cast_fp16")]; tensor var_14503_to_fp16 = const()[name = tensor("op_14503_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1571_cast_fp16 = mul(x = var_14502_cast_fp16, y = var_14503_to_fp16)[name = tensor("aw_chunk_1571_cast_fp16")]; tensor var_14506_equation_0 = const()[name = tensor("op_14506_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14506_cast_fp16 = einsum(equation = var_14506_equation_0, values = (var_14268_cast_fp16, var_14226_cast_fp16))[name = tensor("op_14506_cast_fp16")]; tensor var_14507_to_fp16 = const()[name = tensor("op_14507_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1573_cast_fp16 = mul(x = var_14506_cast_fp16, y = var_14507_to_fp16)[name = tensor("aw_chunk_1573_cast_fp16")]; tensor var_14510_equation_0 = const()[name = tensor("op_14510_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14510_cast_fp16 = einsum(equation = var_14510_equation_0, values = (var_14268_cast_fp16, var_14227_cast_fp16))[name = tensor("op_14510_cast_fp16")]; tensor var_14511_to_fp16 = const()[name = tensor("op_14511_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1575_cast_fp16 = mul(x = var_14510_cast_fp16, y = var_14511_to_fp16)[name = tensor("aw_chunk_1575_cast_fp16")]; tensor var_14514_equation_0 = const()[name = tensor("op_14514_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14514_cast_fp16 = einsum(equation = var_14514_equation_0, values = (var_14268_cast_fp16, var_14228_cast_fp16))[name = tensor("op_14514_cast_fp16")]; tensor var_14515_to_fp16 = const()[name = tensor("op_14515_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1577_cast_fp16 = mul(x = var_14514_cast_fp16, y = var_14515_to_fp16)[name = tensor("aw_chunk_1577_cast_fp16")]; tensor var_14518_equation_0 = const()[name = tensor("op_14518_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14518_cast_fp16 = einsum(equation = var_14518_equation_0, values = (var_14268_cast_fp16, var_14229_cast_fp16))[name = tensor("op_14518_cast_fp16")]; tensor var_14519_to_fp16 = const()[name = tensor("op_14519_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1579_cast_fp16 = mul(x = var_14518_cast_fp16, y = var_14519_to_fp16)[name = tensor("aw_chunk_1579_cast_fp16")]; tensor var_14522_equation_0 = const()[name = tensor("op_14522_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14522_cast_fp16 = einsum(equation = var_14522_equation_0, values = (var_14268_cast_fp16, var_14230_cast_fp16))[name = tensor("op_14522_cast_fp16")]; tensor var_14523_to_fp16 = const()[name = tensor("op_14523_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1581_cast_fp16 = mul(x = var_14522_cast_fp16, y = var_14523_to_fp16)[name = tensor("aw_chunk_1581_cast_fp16")]; tensor var_14526_equation_0 = const()[name = tensor("op_14526_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14526_cast_fp16 = einsum(equation = var_14526_equation_0, values = (var_14268_cast_fp16, var_14231_cast_fp16))[name = tensor("op_14526_cast_fp16")]; tensor var_14527_to_fp16 = const()[name = tensor("op_14527_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1583_cast_fp16 = mul(x = var_14526_cast_fp16, y = var_14527_to_fp16)[name = tensor("aw_chunk_1583_cast_fp16")]; tensor var_14530_equation_0 = const()[name = tensor("op_14530_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14530_cast_fp16 = einsum(equation = var_14530_equation_0, values = (var_14272_cast_fp16, var_14232_cast_fp16))[name = tensor("op_14530_cast_fp16")]; tensor var_14531_to_fp16 = const()[name = tensor("op_14531_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1585_cast_fp16 = mul(x = var_14530_cast_fp16, y = var_14531_to_fp16)[name = tensor("aw_chunk_1585_cast_fp16")]; tensor var_14534_equation_0 = const()[name = tensor("op_14534_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14534_cast_fp16 = einsum(equation = var_14534_equation_0, values = (var_14272_cast_fp16, var_14233_cast_fp16))[name = tensor("op_14534_cast_fp16")]; tensor var_14535_to_fp16 = const()[name = tensor("op_14535_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1587_cast_fp16 = mul(x = var_14534_cast_fp16, y = var_14535_to_fp16)[name = tensor("aw_chunk_1587_cast_fp16")]; tensor var_14538_equation_0 = const()[name = tensor("op_14538_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14538_cast_fp16 = einsum(equation = var_14538_equation_0, values = (var_14272_cast_fp16, var_14234_cast_fp16))[name = tensor("op_14538_cast_fp16")]; tensor var_14539_to_fp16 = const()[name = tensor("op_14539_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1589_cast_fp16 = mul(x = var_14538_cast_fp16, y = var_14539_to_fp16)[name = tensor("aw_chunk_1589_cast_fp16")]; tensor var_14542_equation_0 = const()[name = tensor("op_14542_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14542_cast_fp16 = einsum(equation = var_14542_equation_0, values = (var_14272_cast_fp16, var_14235_cast_fp16))[name = tensor("op_14542_cast_fp16")]; tensor var_14543_to_fp16 = const()[name = tensor("op_14543_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1591_cast_fp16 = mul(x = var_14542_cast_fp16, y = var_14543_to_fp16)[name = tensor("aw_chunk_1591_cast_fp16")]; tensor var_14546_equation_0 = const()[name = tensor("op_14546_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14546_cast_fp16 = einsum(equation = var_14546_equation_0, values = (var_14272_cast_fp16, var_14236_cast_fp16))[name = tensor("op_14546_cast_fp16")]; tensor var_14547_to_fp16 = const()[name = tensor("op_14547_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1593_cast_fp16 = mul(x = var_14546_cast_fp16, y = var_14547_to_fp16)[name = tensor("aw_chunk_1593_cast_fp16")]; tensor var_14550_equation_0 = const()[name = tensor("op_14550_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14550_cast_fp16 = einsum(equation = var_14550_equation_0, values = (var_14272_cast_fp16, var_14237_cast_fp16))[name = tensor("op_14550_cast_fp16")]; tensor var_14551_to_fp16 = const()[name = tensor("op_14551_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1595_cast_fp16 = mul(x = var_14550_cast_fp16, y = var_14551_to_fp16)[name = tensor("aw_chunk_1595_cast_fp16")]; tensor var_14554_equation_0 = const()[name = tensor("op_14554_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14554_cast_fp16 = einsum(equation = var_14554_equation_0, values = (var_14272_cast_fp16, var_14238_cast_fp16))[name = tensor("op_14554_cast_fp16")]; tensor var_14555_to_fp16 = const()[name = tensor("op_14555_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1597_cast_fp16 = mul(x = var_14554_cast_fp16, y = var_14555_to_fp16)[name = tensor("aw_chunk_1597_cast_fp16")]; tensor var_14558_equation_0 = const()[name = tensor("op_14558_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_14558_cast_fp16 = einsum(equation = var_14558_equation_0, values = (var_14272_cast_fp16, var_14239_cast_fp16))[name = tensor("op_14558_cast_fp16")]; tensor var_14559_to_fp16 = const()[name = tensor("op_14559_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_cast_fp16 = mul(x = var_14558_cast_fp16, y = var_14559_to_fp16)[name = tensor("aw_chunk_cast_fp16")]; tensor var_14561_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1473_cast_fp16)[name = tensor("op_14561_cast_fp16")]; tensor var_14562_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1475_cast_fp16)[name = tensor("op_14562_cast_fp16")]; tensor var_14563_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1477_cast_fp16)[name = tensor("op_14563_cast_fp16")]; tensor var_14564_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1479_cast_fp16)[name = tensor("op_14564_cast_fp16")]; tensor var_14565_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1481_cast_fp16)[name = tensor("op_14565_cast_fp16")]; tensor var_14566_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1483_cast_fp16)[name = tensor("op_14566_cast_fp16")]; tensor var_14567_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1485_cast_fp16)[name = tensor("op_14567_cast_fp16")]; tensor var_14568_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1487_cast_fp16)[name = tensor("op_14568_cast_fp16")]; tensor var_14569_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1489_cast_fp16)[name = tensor("op_14569_cast_fp16")]; tensor var_14570_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1491_cast_fp16)[name = tensor("op_14570_cast_fp16")]; tensor var_14571_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1493_cast_fp16)[name = tensor("op_14571_cast_fp16")]; tensor var_14572_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1495_cast_fp16)[name = tensor("op_14572_cast_fp16")]; tensor var_14573_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1497_cast_fp16)[name = tensor("op_14573_cast_fp16")]; tensor var_14574_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1499_cast_fp16)[name = tensor("op_14574_cast_fp16")]; tensor var_14575_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1501_cast_fp16)[name = tensor("op_14575_cast_fp16")]; tensor var_14576_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1503_cast_fp16)[name = tensor("op_14576_cast_fp16")]; tensor var_14577_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1505_cast_fp16)[name = tensor("op_14577_cast_fp16")]; tensor var_14578_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1507_cast_fp16)[name = tensor("op_14578_cast_fp16")]; tensor var_14579_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1509_cast_fp16)[name = tensor("op_14579_cast_fp16")]; tensor var_14580_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1511_cast_fp16)[name = tensor("op_14580_cast_fp16")]; tensor var_14581_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1513_cast_fp16)[name = tensor("op_14581_cast_fp16")]; tensor var_14582_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1515_cast_fp16)[name = tensor("op_14582_cast_fp16")]; tensor var_14583_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1517_cast_fp16)[name = tensor("op_14583_cast_fp16")]; tensor var_14584_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1519_cast_fp16)[name = tensor("op_14584_cast_fp16")]; tensor var_14585_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1521_cast_fp16)[name = tensor("op_14585_cast_fp16")]; tensor var_14586_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1523_cast_fp16)[name = tensor("op_14586_cast_fp16")]; tensor var_14587_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1525_cast_fp16)[name = tensor("op_14587_cast_fp16")]; tensor var_14588_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1527_cast_fp16)[name = tensor("op_14588_cast_fp16")]; tensor var_14589_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1529_cast_fp16)[name = tensor("op_14589_cast_fp16")]; tensor var_14590_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1531_cast_fp16)[name = tensor("op_14590_cast_fp16")]; tensor var_14591_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1533_cast_fp16)[name = tensor("op_14591_cast_fp16")]; tensor var_14592_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1535_cast_fp16)[name = tensor("op_14592_cast_fp16")]; tensor var_14593_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1537_cast_fp16)[name = tensor("op_14593_cast_fp16")]; tensor var_14594_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1539_cast_fp16)[name = tensor("op_14594_cast_fp16")]; tensor var_14595_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1541_cast_fp16)[name = tensor("op_14595_cast_fp16")]; tensor var_14596_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1543_cast_fp16)[name = tensor("op_14596_cast_fp16")]; tensor var_14597_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1545_cast_fp16)[name = tensor("op_14597_cast_fp16")]; tensor var_14598_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1547_cast_fp16)[name = tensor("op_14598_cast_fp16")]; tensor var_14599_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1549_cast_fp16)[name = tensor("op_14599_cast_fp16")]; tensor var_14600_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1551_cast_fp16)[name = tensor("op_14600_cast_fp16")]; tensor var_14601_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1553_cast_fp16)[name = tensor("op_14601_cast_fp16")]; tensor var_14602_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1555_cast_fp16)[name = tensor("op_14602_cast_fp16")]; tensor var_14603_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1557_cast_fp16)[name = tensor("op_14603_cast_fp16")]; tensor var_14604_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1559_cast_fp16)[name = tensor("op_14604_cast_fp16")]; tensor var_14605_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1561_cast_fp16)[name = tensor("op_14605_cast_fp16")]; tensor var_14606_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1563_cast_fp16)[name = tensor("op_14606_cast_fp16")]; tensor var_14607_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1565_cast_fp16)[name = tensor("op_14607_cast_fp16")]; tensor var_14608_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1567_cast_fp16)[name = tensor("op_14608_cast_fp16")]; tensor var_14609_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1569_cast_fp16)[name = tensor("op_14609_cast_fp16")]; tensor var_14610_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1571_cast_fp16)[name = tensor("op_14610_cast_fp16")]; tensor var_14611_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1573_cast_fp16)[name = tensor("op_14611_cast_fp16")]; tensor var_14612_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1575_cast_fp16)[name = tensor("op_14612_cast_fp16")]; tensor var_14613_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1577_cast_fp16)[name = tensor("op_14613_cast_fp16")]; tensor var_14614_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1579_cast_fp16)[name = tensor("op_14614_cast_fp16")]; tensor var_14615_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1581_cast_fp16)[name = tensor("op_14615_cast_fp16")]; tensor var_14616_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1583_cast_fp16)[name = tensor("op_14616_cast_fp16")]; tensor var_14617_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1585_cast_fp16)[name = tensor("op_14617_cast_fp16")]; tensor var_14618_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1587_cast_fp16)[name = tensor("op_14618_cast_fp16")]; tensor var_14619_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1589_cast_fp16)[name = tensor("op_14619_cast_fp16")]; tensor var_14620_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1591_cast_fp16)[name = tensor("op_14620_cast_fp16")]; tensor var_14621_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1593_cast_fp16)[name = tensor("op_14621_cast_fp16")]; tensor var_14622_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1595_cast_fp16)[name = tensor("op_14622_cast_fp16")]; tensor var_14623_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_1597_cast_fp16)[name = tensor("op_14623_cast_fp16")]; tensor var_14624_cast_fp16 = softmax(axis = var_10375, x = aw_chunk_cast_fp16)[name = tensor("op_14624_cast_fp16")]; tensor var_14626_equation_0 = const()[name = tensor("op_14626_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14626_cast_fp16 = einsum(equation = var_14626_equation_0, values = (var_14274_cast_fp16, var_14561_cast_fp16))[name = tensor("op_14626_cast_fp16")]; tensor var_14628_equation_0 = const()[name = tensor("op_14628_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14628_cast_fp16 = einsum(equation = var_14628_equation_0, values = (var_14274_cast_fp16, var_14562_cast_fp16))[name = tensor("op_14628_cast_fp16")]; tensor var_14630_equation_0 = const()[name = tensor("op_14630_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14630_cast_fp16 = einsum(equation = var_14630_equation_0, values = (var_14274_cast_fp16, var_14563_cast_fp16))[name = tensor("op_14630_cast_fp16")]; tensor var_14632_equation_0 = const()[name = tensor("op_14632_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14632_cast_fp16 = einsum(equation = var_14632_equation_0, values = (var_14274_cast_fp16, var_14564_cast_fp16))[name = tensor("op_14632_cast_fp16")]; tensor var_14634_equation_0 = const()[name = tensor("op_14634_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14634_cast_fp16 = einsum(equation = var_14634_equation_0, values = (var_14274_cast_fp16, var_14565_cast_fp16))[name = tensor("op_14634_cast_fp16")]; tensor var_14636_equation_0 = const()[name = tensor("op_14636_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14636_cast_fp16 = einsum(equation = var_14636_equation_0, values = (var_14274_cast_fp16, var_14566_cast_fp16))[name = tensor("op_14636_cast_fp16")]; tensor var_14638_equation_0 = const()[name = tensor("op_14638_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14638_cast_fp16 = einsum(equation = var_14638_equation_0, values = (var_14274_cast_fp16, var_14567_cast_fp16))[name = tensor("op_14638_cast_fp16")]; tensor var_14640_equation_0 = const()[name = tensor("op_14640_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14640_cast_fp16 = einsum(equation = var_14640_equation_0, values = (var_14274_cast_fp16, var_14568_cast_fp16))[name = tensor("op_14640_cast_fp16")]; tensor var_14642_equation_0 = const()[name = tensor("op_14642_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14642_cast_fp16 = einsum(equation = var_14642_equation_0, values = (var_14278_cast_fp16, var_14569_cast_fp16))[name = tensor("op_14642_cast_fp16")]; tensor var_14644_equation_0 = const()[name = tensor("op_14644_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14644_cast_fp16 = einsum(equation = var_14644_equation_0, values = (var_14278_cast_fp16, var_14570_cast_fp16))[name = tensor("op_14644_cast_fp16")]; tensor var_14646_equation_0 = const()[name = tensor("op_14646_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14646_cast_fp16 = einsum(equation = var_14646_equation_0, values = (var_14278_cast_fp16, var_14571_cast_fp16))[name = tensor("op_14646_cast_fp16")]; tensor var_14648_equation_0 = const()[name = tensor("op_14648_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14648_cast_fp16 = einsum(equation = var_14648_equation_0, values = (var_14278_cast_fp16, var_14572_cast_fp16))[name = tensor("op_14648_cast_fp16")]; tensor var_14650_equation_0 = const()[name = tensor("op_14650_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14650_cast_fp16 = einsum(equation = var_14650_equation_0, values = (var_14278_cast_fp16, var_14573_cast_fp16))[name = tensor("op_14650_cast_fp16")]; tensor var_14652_equation_0 = const()[name = tensor("op_14652_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14652_cast_fp16 = einsum(equation = var_14652_equation_0, values = (var_14278_cast_fp16, var_14574_cast_fp16))[name = tensor("op_14652_cast_fp16")]; tensor var_14654_equation_0 = const()[name = tensor("op_14654_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14654_cast_fp16 = einsum(equation = var_14654_equation_0, values = (var_14278_cast_fp16, var_14575_cast_fp16))[name = tensor("op_14654_cast_fp16")]; tensor var_14656_equation_0 = const()[name = tensor("op_14656_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14656_cast_fp16 = einsum(equation = var_14656_equation_0, values = (var_14278_cast_fp16, var_14576_cast_fp16))[name = tensor("op_14656_cast_fp16")]; tensor var_14658_equation_0 = const()[name = tensor("op_14658_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14658_cast_fp16 = einsum(equation = var_14658_equation_0, values = (var_14282_cast_fp16, var_14577_cast_fp16))[name = tensor("op_14658_cast_fp16")]; tensor var_14660_equation_0 = const()[name = tensor("op_14660_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14660_cast_fp16 = einsum(equation = var_14660_equation_0, values = (var_14282_cast_fp16, var_14578_cast_fp16))[name = tensor("op_14660_cast_fp16")]; tensor var_14662_equation_0 = const()[name = tensor("op_14662_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14662_cast_fp16 = einsum(equation = var_14662_equation_0, values = (var_14282_cast_fp16, var_14579_cast_fp16))[name = tensor("op_14662_cast_fp16")]; tensor var_14664_equation_0 = const()[name = tensor("op_14664_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14664_cast_fp16 = einsum(equation = var_14664_equation_0, values = (var_14282_cast_fp16, var_14580_cast_fp16))[name = tensor("op_14664_cast_fp16")]; tensor var_14666_equation_0 = const()[name = tensor("op_14666_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14666_cast_fp16 = einsum(equation = var_14666_equation_0, values = (var_14282_cast_fp16, var_14581_cast_fp16))[name = tensor("op_14666_cast_fp16")]; tensor var_14668_equation_0 = const()[name = tensor("op_14668_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14668_cast_fp16 = einsum(equation = var_14668_equation_0, values = (var_14282_cast_fp16, var_14582_cast_fp16))[name = tensor("op_14668_cast_fp16")]; tensor var_14670_equation_0 = const()[name = tensor("op_14670_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14670_cast_fp16 = einsum(equation = var_14670_equation_0, values = (var_14282_cast_fp16, var_14583_cast_fp16))[name = tensor("op_14670_cast_fp16")]; tensor var_14672_equation_0 = const()[name = tensor("op_14672_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14672_cast_fp16 = einsum(equation = var_14672_equation_0, values = (var_14282_cast_fp16, var_14584_cast_fp16))[name = tensor("op_14672_cast_fp16")]; tensor var_14674_equation_0 = const()[name = tensor("op_14674_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14674_cast_fp16 = einsum(equation = var_14674_equation_0, values = (var_14286_cast_fp16, var_14585_cast_fp16))[name = tensor("op_14674_cast_fp16")]; tensor var_14676_equation_0 = const()[name = tensor("op_14676_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14676_cast_fp16 = einsum(equation = var_14676_equation_0, values = (var_14286_cast_fp16, var_14586_cast_fp16))[name = tensor("op_14676_cast_fp16")]; tensor var_14678_equation_0 = const()[name = tensor("op_14678_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14678_cast_fp16 = einsum(equation = var_14678_equation_0, values = (var_14286_cast_fp16, var_14587_cast_fp16))[name = tensor("op_14678_cast_fp16")]; tensor var_14680_equation_0 = const()[name = tensor("op_14680_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14680_cast_fp16 = einsum(equation = var_14680_equation_0, values = (var_14286_cast_fp16, var_14588_cast_fp16))[name = tensor("op_14680_cast_fp16")]; tensor var_14682_equation_0 = const()[name = tensor("op_14682_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14682_cast_fp16 = einsum(equation = var_14682_equation_0, values = (var_14286_cast_fp16, var_14589_cast_fp16))[name = tensor("op_14682_cast_fp16")]; tensor var_14684_equation_0 = const()[name = tensor("op_14684_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14684_cast_fp16 = einsum(equation = var_14684_equation_0, values = (var_14286_cast_fp16, var_14590_cast_fp16))[name = tensor("op_14684_cast_fp16")]; tensor var_14686_equation_0 = const()[name = tensor("op_14686_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14686_cast_fp16 = einsum(equation = var_14686_equation_0, values = (var_14286_cast_fp16, var_14591_cast_fp16))[name = tensor("op_14686_cast_fp16")]; tensor var_14688_equation_0 = const()[name = tensor("op_14688_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14688_cast_fp16 = einsum(equation = var_14688_equation_0, values = (var_14286_cast_fp16, var_14592_cast_fp16))[name = tensor("op_14688_cast_fp16")]; tensor var_14690_equation_0 = const()[name = tensor("op_14690_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14690_cast_fp16 = einsum(equation = var_14690_equation_0, values = (var_14290_cast_fp16, var_14593_cast_fp16))[name = tensor("op_14690_cast_fp16")]; tensor var_14692_equation_0 = const()[name = tensor("op_14692_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14692_cast_fp16 = einsum(equation = var_14692_equation_0, values = (var_14290_cast_fp16, var_14594_cast_fp16))[name = tensor("op_14692_cast_fp16")]; tensor var_14694_equation_0 = const()[name = tensor("op_14694_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14694_cast_fp16 = einsum(equation = var_14694_equation_0, values = (var_14290_cast_fp16, var_14595_cast_fp16))[name = tensor("op_14694_cast_fp16")]; tensor var_14696_equation_0 = const()[name = tensor("op_14696_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14696_cast_fp16 = einsum(equation = var_14696_equation_0, values = (var_14290_cast_fp16, var_14596_cast_fp16))[name = tensor("op_14696_cast_fp16")]; tensor var_14698_equation_0 = const()[name = tensor("op_14698_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14698_cast_fp16 = einsum(equation = var_14698_equation_0, values = (var_14290_cast_fp16, var_14597_cast_fp16))[name = tensor("op_14698_cast_fp16")]; tensor var_14700_equation_0 = const()[name = tensor("op_14700_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14700_cast_fp16 = einsum(equation = var_14700_equation_0, values = (var_14290_cast_fp16, var_14598_cast_fp16))[name = tensor("op_14700_cast_fp16")]; tensor var_14702_equation_0 = const()[name = tensor("op_14702_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14702_cast_fp16 = einsum(equation = var_14702_equation_0, values = (var_14290_cast_fp16, var_14599_cast_fp16))[name = tensor("op_14702_cast_fp16")]; tensor var_14704_equation_0 = const()[name = tensor("op_14704_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14704_cast_fp16 = einsum(equation = var_14704_equation_0, values = (var_14290_cast_fp16, var_14600_cast_fp16))[name = tensor("op_14704_cast_fp16")]; tensor var_14706_equation_0 = const()[name = tensor("op_14706_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14706_cast_fp16 = einsum(equation = var_14706_equation_0, values = (var_14294_cast_fp16, var_14601_cast_fp16))[name = tensor("op_14706_cast_fp16")]; tensor var_14708_equation_0 = const()[name = tensor("op_14708_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14708_cast_fp16 = einsum(equation = var_14708_equation_0, values = (var_14294_cast_fp16, var_14602_cast_fp16))[name = tensor("op_14708_cast_fp16")]; tensor var_14710_equation_0 = const()[name = tensor("op_14710_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14710_cast_fp16 = einsum(equation = var_14710_equation_0, values = (var_14294_cast_fp16, var_14603_cast_fp16))[name = tensor("op_14710_cast_fp16")]; tensor var_14712_equation_0 = const()[name = tensor("op_14712_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14712_cast_fp16 = einsum(equation = var_14712_equation_0, values = (var_14294_cast_fp16, var_14604_cast_fp16))[name = tensor("op_14712_cast_fp16")]; tensor var_14714_equation_0 = const()[name = tensor("op_14714_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14714_cast_fp16 = einsum(equation = var_14714_equation_0, values = (var_14294_cast_fp16, var_14605_cast_fp16))[name = tensor("op_14714_cast_fp16")]; tensor var_14716_equation_0 = const()[name = tensor("op_14716_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14716_cast_fp16 = einsum(equation = var_14716_equation_0, values = (var_14294_cast_fp16, var_14606_cast_fp16))[name = tensor("op_14716_cast_fp16")]; tensor var_14718_equation_0 = const()[name = tensor("op_14718_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14718_cast_fp16 = einsum(equation = var_14718_equation_0, values = (var_14294_cast_fp16, var_14607_cast_fp16))[name = tensor("op_14718_cast_fp16")]; tensor var_14720_equation_0 = const()[name = tensor("op_14720_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14720_cast_fp16 = einsum(equation = var_14720_equation_0, values = (var_14294_cast_fp16, var_14608_cast_fp16))[name = tensor("op_14720_cast_fp16")]; tensor var_14722_equation_0 = const()[name = tensor("op_14722_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14722_cast_fp16 = einsum(equation = var_14722_equation_0, values = (var_14298_cast_fp16, var_14609_cast_fp16))[name = tensor("op_14722_cast_fp16")]; tensor var_14724_equation_0 = const()[name = tensor("op_14724_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14724_cast_fp16 = einsum(equation = var_14724_equation_0, values = (var_14298_cast_fp16, var_14610_cast_fp16))[name = tensor("op_14724_cast_fp16")]; tensor var_14726_equation_0 = const()[name = tensor("op_14726_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14726_cast_fp16 = einsum(equation = var_14726_equation_0, values = (var_14298_cast_fp16, var_14611_cast_fp16))[name = tensor("op_14726_cast_fp16")]; tensor var_14728_equation_0 = const()[name = tensor("op_14728_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14728_cast_fp16 = einsum(equation = var_14728_equation_0, values = (var_14298_cast_fp16, var_14612_cast_fp16))[name = tensor("op_14728_cast_fp16")]; tensor var_14730_equation_0 = const()[name = tensor("op_14730_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14730_cast_fp16 = einsum(equation = var_14730_equation_0, values = (var_14298_cast_fp16, var_14613_cast_fp16))[name = tensor("op_14730_cast_fp16")]; tensor var_14732_equation_0 = const()[name = tensor("op_14732_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14732_cast_fp16 = einsum(equation = var_14732_equation_0, values = (var_14298_cast_fp16, var_14614_cast_fp16))[name = tensor("op_14732_cast_fp16")]; tensor var_14734_equation_0 = const()[name = tensor("op_14734_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14734_cast_fp16 = einsum(equation = var_14734_equation_0, values = (var_14298_cast_fp16, var_14615_cast_fp16))[name = tensor("op_14734_cast_fp16")]; tensor var_14736_equation_0 = const()[name = tensor("op_14736_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14736_cast_fp16 = einsum(equation = var_14736_equation_0, values = (var_14298_cast_fp16, var_14616_cast_fp16))[name = tensor("op_14736_cast_fp16")]; tensor var_14738_equation_0 = const()[name = tensor("op_14738_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14738_cast_fp16 = einsum(equation = var_14738_equation_0, values = (var_14302_cast_fp16, var_14617_cast_fp16))[name = tensor("op_14738_cast_fp16")]; tensor var_14740_equation_0 = const()[name = tensor("op_14740_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14740_cast_fp16 = einsum(equation = var_14740_equation_0, values = (var_14302_cast_fp16, var_14618_cast_fp16))[name = tensor("op_14740_cast_fp16")]; tensor var_14742_equation_0 = const()[name = tensor("op_14742_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14742_cast_fp16 = einsum(equation = var_14742_equation_0, values = (var_14302_cast_fp16, var_14619_cast_fp16))[name = tensor("op_14742_cast_fp16")]; tensor var_14744_equation_0 = const()[name = tensor("op_14744_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14744_cast_fp16 = einsum(equation = var_14744_equation_0, values = (var_14302_cast_fp16, var_14620_cast_fp16))[name = tensor("op_14744_cast_fp16")]; tensor var_14746_equation_0 = const()[name = tensor("op_14746_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14746_cast_fp16 = einsum(equation = var_14746_equation_0, values = (var_14302_cast_fp16, var_14621_cast_fp16))[name = tensor("op_14746_cast_fp16")]; tensor var_14748_equation_0 = const()[name = tensor("op_14748_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14748_cast_fp16 = einsum(equation = var_14748_equation_0, values = (var_14302_cast_fp16, var_14622_cast_fp16))[name = tensor("op_14748_cast_fp16")]; tensor var_14750_equation_0 = const()[name = tensor("op_14750_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14750_cast_fp16 = einsum(equation = var_14750_equation_0, values = (var_14302_cast_fp16, var_14623_cast_fp16))[name = tensor("op_14750_cast_fp16")]; tensor var_14752_equation_0 = const()[name = tensor("op_14752_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_14752_cast_fp16 = einsum(equation = var_14752_equation_0, values = (var_14302_cast_fp16, var_14624_cast_fp16))[name = tensor("op_14752_cast_fp16")]; tensor var_14754_interleave_0 = const()[name = tensor("op_14754_interleave_0"), val = tensor(false)]; tensor var_14754_cast_fp16 = concat(axis = var_10347, interleave = var_14754_interleave_0, values = (var_14626_cast_fp16, var_14628_cast_fp16, var_14630_cast_fp16, var_14632_cast_fp16, var_14634_cast_fp16, var_14636_cast_fp16, var_14638_cast_fp16, var_14640_cast_fp16))[name = tensor("op_14754_cast_fp16")]; tensor var_14756_interleave_0 = const()[name = tensor("op_14756_interleave_0"), val = tensor(false)]; tensor var_14756_cast_fp16 = concat(axis = var_10347, interleave = var_14756_interleave_0, values = (var_14642_cast_fp16, var_14644_cast_fp16, var_14646_cast_fp16, var_14648_cast_fp16, var_14650_cast_fp16, var_14652_cast_fp16, var_14654_cast_fp16, var_14656_cast_fp16))[name = tensor("op_14756_cast_fp16")]; tensor var_14758_interleave_0 = const()[name = tensor("op_14758_interleave_0"), val = tensor(false)]; tensor var_14758_cast_fp16 = concat(axis = var_10347, interleave = var_14758_interleave_0, values = (var_14658_cast_fp16, var_14660_cast_fp16, var_14662_cast_fp16, var_14664_cast_fp16, var_14666_cast_fp16, var_14668_cast_fp16, var_14670_cast_fp16, var_14672_cast_fp16))[name = tensor("op_14758_cast_fp16")]; tensor var_14760_interleave_0 = const()[name = tensor("op_14760_interleave_0"), val = tensor(false)]; tensor var_14760_cast_fp16 = concat(axis = var_10347, interleave = var_14760_interleave_0, values = (var_14674_cast_fp16, var_14676_cast_fp16, var_14678_cast_fp16, var_14680_cast_fp16, var_14682_cast_fp16, var_14684_cast_fp16, var_14686_cast_fp16, var_14688_cast_fp16))[name = tensor("op_14760_cast_fp16")]; tensor var_14762_interleave_0 = const()[name = tensor("op_14762_interleave_0"), val = tensor(false)]; tensor var_14762_cast_fp16 = concat(axis = var_10347, interleave = var_14762_interleave_0, values = (var_14690_cast_fp16, var_14692_cast_fp16, var_14694_cast_fp16, var_14696_cast_fp16, var_14698_cast_fp16, var_14700_cast_fp16, var_14702_cast_fp16, var_14704_cast_fp16))[name = tensor("op_14762_cast_fp16")]; tensor var_14764_interleave_0 = const()[name = tensor("op_14764_interleave_0"), val = tensor(false)]; tensor var_14764_cast_fp16 = concat(axis = var_10347, interleave = var_14764_interleave_0, values = (var_14706_cast_fp16, var_14708_cast_fp16, var_14710_cast_fp16, var_14712_cast_fp16, var_14714_cast_fp16, var_14716_cast_fp16, var_14718_cast_fp16, var_14720_cast_fp16))[name = tensor("op_14764_cast_fp16")]; tensor var_14766_interleave_0 = const()[name = tensor("op_14766_interleave_0"), val = tensor(false)]; tensor var_14766_cast_fp16 = concat(axis = var_10347, interleave = var_14766_interleave_0, values = (var_14722_cast_fp16, var_14724_cast_fp16, var_14726_cast_fp16, var_14728_cast_fp16, var_14730_cast_fp16, var_14732_cast_fp16, var_14734_cast_fp16, var_14736_cast_fp16))[name = tensor("op_14766_cast_fp16")]; tensor var_14768_interleave_0 = const()[name = tensor("op_14768_interleave_0"), val = tensor(false)]; tensor var_14768_cast_fp16 = concat(axis = var_10347, interleave = var_14768_interleave_0, values = (var_14738_cast_fp16, var_14740_cast_fp16, var_14742_cast_fp16, var_14744_cast_fp16, var_14746_cast_fp16, var_14748_cast_fp16, var_14750_cast_fp16, var_14752_cast_fp16))[name = tensor("op_14768_cast_fp16")]; tensor input_519_interleave_0 = const()[name = tensor("input_519_interleave_0"), val = tensor(false)]; tensor input_519_cast_fp16 = concat(axis = var_10375, interleave = input_519_interleave_0, values = (var_14754_cast_fp16, var_14756_cast_fp16, var_14758_cast_fp16, var_14760_cast_fp16, var_14762_cast_fp16, var_14764_cast_fp16, var_14766_cast_fp16, var_14768_cast_fp16))[name = tensor("input_519_cast_fp16")]; tensor var_14778_pad_type_0 = const()[name = tensor("op_14778_pad_type_0"), val = tensor("valid")]; tensor var_14778_strides_0 = const()[name = tensor("op_14778_strides_0"), val = tensor([1, 1])]; tensor var_14778_pad_0 = const()[name = tensor("op_14778_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_14778_dilations_0 = const()[name = tensor("op_14778_dilations_0"), val = tensor([1, 1])]; tensor var_14778_groups_0 = const()[name = tensor("op_14778_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(828667200)))]; tensor up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(828872064)))]; tensor var_14778_cast_fp16 = conv(bias = up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_14778_dilations_0, groups = var_14778_groups_0, pad = var_14778_pad_0, pad_type = var_14778_pad_type_0, strides = var_14778_strides_0, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_519_cast_fp16)[name = tensor("op_14778_cast_fp16")]; tensor inputs_cast_fp16 = add(x = var_14778_cast_fp16, y = inputs_93_cast_fp16)[name = tensor("inputs_cast_fp16")]; tensor input_521_axes_0 = const()[name = tensor("input_521_axes_0"), val = tensor([1])]; tensor input_521_gamma_0_to_fp16 = const()[name = tensor("input_521_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(828872768)))]; tensor input_521_beta_0_to_fp16 = const()[name = tensor("input_521_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(828873472)))]; tensor var_14788_to_fp16 = const()[name = tensor("op_14788_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_521_cast_fp16 = layer_norm(axes = input_521_axes_0, beta = input_521_beta_0_to_fp16, epsilon = var_14788_to_fp16, gamma = input_521_gamma_0_to_fp16, x = inputs_cast_fp16)[name = tensor("input_521_cast_fp16")]; tensor var_14808_pad_type_0 = const()[name = tensor("op_14808_pad_type_0"), val = tensor("valid")]; tensor var_14808_strides_0 = const()[name = tensor("op_14808_strides_0"), val = tensor([1, 1])]; tensor var_14808_pad_0 = const()[name = tensor("op_14808_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_14808_dilations_0 = const()[name = tensor("op_14808_dilations_0"), val = tensor([1, 1])]; tensor var_14808_groups_0 = const()[name = tensor("op_14808_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(828874176)))]; tensor up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(830512640)))]; tensor var_14808_cast_fp16 = conv(bias = up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_14808_dilations_0, groups = var_14808_groups_0, pad = var_14808_pad_0, pad_type = var_14808_pad_type_0, strides = var_14808_strides_0, weight = up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_521_cast_fp16)[name = tensor("op_14808_cast_fp16")]; tensor var_14809_split_sizes_0 = const()[name = tensor("op_14809_split_sizes_0"), val = tensor([1280, 1280])]; tensor var_14809_axis_0 = const()[name = tensor("op_14809_axis_0"), val = tensor(1)]; tensor var_14809_cast_fp16_0, tensor var_14809_cast_fp16_1 = split(axis = var_14809_axis_0, split_sizes = var_14809_split_sizes_0, x = var_14808_cast_fp16)[name = tensor("op_14809_cast_fp16")]; tensor var_14811_mode_0 = const()[name = tensor("op_14811_mode_0"), val = tensor("EXACT")]; tensor var_14811_cast_fp16 = gelu(mode = var_14811_mode_0, x = var_14809_cast_fp16_1)[name = tensor("op_14811_cast_fp16")]; tensor input_523_cast_fp16 = mul(x = var_14809_cast_fp16_0, y = var_14811_cast_fp16)[name = tensor("input_523_cast_fp16")]; tensor var_14819_pad_type_0 = const()[name = tensor("op_14819_pad_type_0"), val = tensor("valid")]; tensor var_14819_strides_0 = const()[name = tensor("op_14819_strides_0"), val = tensor([1, 1])]; tensor var_14819_pad_0 = const()[name = tensor("op_14819_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_14819_dilations_0 = const()[name = tensor("op_14819_dilations_0"), val = tensor([1, 1])]; tensor var_14819_groups_0 = const()[name = tensor("op_14819_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(830517824)))]; tensor up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(831337088)))]; tensor var_14819_cast_fp16 = conv(bias = up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_14819_dilations_0, groups = var_14819_groups_0, pad = var_14819_pad_0, pad_type = var_14819_pad_type_0, strides = var_14819_strides_0, weight = up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_523_cast_fp16)[name = tensor("op_14819_cast_fp16")]; tensor hidden_states_333_cast_fp16 = add(x = var_14819_cast_fp16, y = inputs_cast_fp16)[name = tensor("hidden_states_333_cast_fp16")]; tensor var_14821 = const()[name = tensor("op_14821"), val = tensor([2, 320, 64, 64])]; tensor input_525_cast_fp16 = reshape(shape = var_14821, x = hidden_states_333_cast_fp16)[name = tensor("input_525_cast_fp16")]; tensor hidden_states_pad_type_0 = const()[name = tensor("hidden_states_pad_type_0"), val = tensor("valid")]; tensor hidden_states_strides_0 = const()[name = tensor("hidden_states_strides_0"), val = tensor([1, 1])]; tensor hidden_states_pad_0 = const()[name = tensor("hidden_states_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_dilations_0 = const()[name = tensor("hidden_states_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_groups_0 = const()[name = tensor("hidden_states_groups_0"), val = tensor(1)]; tensor up_blocks_3_attentions_2_proj_out_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(831337792)))]; tensor up_blocks_3_attentions_2_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(831542656)))]; tensor hidden_states_cast_fp16 = conv(bias = up_blocks_3_attentions_2_proj_out_bias_to_fp16, dilations = hidden_states_dilations_0, groups = hidden_states_groups_0, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = hidden_states_strides_0, weight = up_blocks_3_attentions_2_proj_out_weight_to_fp16, x = input_525_cast_fp16)[name = tensor("hidden_states_cast_fp16")]; tensor input_527_cast_fp16 = add(x = hidden_states_cast_fp16, y = hidden_states_323_cast_fp16)[name = tensor("input_527_cast_fp16")]; tensor reshape_240_shape_0 = const()[name = tensor("reshape_240_shape_0"), val = tensor([2, 32, 10, 64, 64])]; tensor reshape_240_cast_fp16 = reshape(shape = reshape_240_shape_0, x = input_527_cast_fp16)[name = tensor("reshape_240_cast_fp16")]; tensor reduce_mean_180_axes_0 = const()[name = tensor("reduce_mean_180_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_180_keep_dims_0 = const()[name = tensor("reduce_mean_180_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_180_cast_fp16 = reduce_mean(axes = reduce_mean_180_axes_0, keep_dims = reduce_mean_180_keep_dims_0, x = reshape_240_cast_fp16)[name = tensor("reduce_mean_180_cast_fp16")]; tensor sub_120_cast_fp16 = sub(x = reshape_240_cast_fp16, y = reduce_mean_180_cast_fp16)[name = tensor("sub_120_cast_fp16")]; tensor square_60_cast_fp16 = square(x = sub_120_cast_fp16)[name = tensor("square_60_cast_fp16")]; tensor reduce_mean_182_axes_0 = const()[name = tensor("reduce_mean_182_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_182_keep_dims_0 = const()[name = tensor("reduce_mean_182_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_182_cast_fp16 = reduce_mean(axes = reduce_mean_182_axes_0, keep_dims = reduce_mean_182_keep_dims_0, x = square_60_cast_fp16)[name = tensor("reduce_mean_182_cast_fp16")]; tensor add_120_y_0_to_fp16 = const()[name = tensor("add_120_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_120_cast_fp16 = add(x = reduce_mean_182_cast_fp16, y = add_120_y_0_to_fp16)[name = tensor("add_120_cast_fp16")]; tensor sqrt_60_cast_fp16 = sqrt(x = add_120_cast_fp16)[name = tensor("sqrt_60_cast_fp16")]; tensor real_div_60_cast_fp16 = real_div(x = sub_120_cast_fp16, y = sqrt_60_cast_fp16)[name = tensor("real_div_60_cast_fp16")]; tensor reshape_241_shape_0 = const()[name = tensor("reshape_241_shape_0"), val = tensor([2, 320, 64, 64])]; tensor reshape_241_cast_fp16 = reshape(shape = reshape_241_shape_0, x = real_div_60_cast_fp16)[name = tensor("reshape_241_cast_fp16")]; tensor add_121_gamma_0_to_fp16 = const()[name = tensor("add_121_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(831543360)))]; tensor add_121_beta_0_to_fp16 = const()[name = tensor("add_121_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(831544064)))]; tensor add_121_epsilon_0_to_fp16 = const()[name = tensor("add_121_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_121_cast_fp16 = batch_norm(beta = add_121_beta_0_to_fp16, epsilon = add_121_epsilon_0_to_fp16, gamma = add_121_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_241_cast_fp16)[name = tensor("add_121_cast_fp16")]; tensor input_cast_fp16 = silu(x = add_121_cast_fp16)[name = tensor("input_cast_fp16")]; tensor var_14848_pad_type_0 = const()[name = tensor("op_14848_pad_type_0"), val = tensor("custom")]; tensor var_14848_pad_0 = const()[name = tensor("op_14848_pad_0"), val = tensor([1, 1, 1, 1])]; tensor var_14848_strides_0 = const()[name = tensor("op_14848_strides_0"), val = tensor([1, 1])]; tensor var_14848_dilations_0 = const()[name = tensor("op_14848_dilations_0"), val = tensor([1, 1])]; tensor var_14848_groups_0 = const()[name = tensor("op_14848_groups_0"), val = tensor(1)]; tensor conv_out_weight_to_fp16 = const()[name = tensor("conv_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(831544768)))]; tensor conv_out_bias_to_fp16 = const()[name = tensor("conv_out_bias_to_fp16"), val = tensor([-0x1.c6cp-10, -0x1.344p-10, -0x1.d48p-15, -0x1.718p-9])]; tensor var_14848_cast_fp16 = conv(bias = conv_out_bias_to_fp16, dilations = var_14848_dilations_0, groups = var_14848_groups_0, pad = var_14848_pad_0, pad_type = var_14848_pad_type_0, strides = var_14848_strides_0, weight = conv_out_weight_to_fp16, x = input_cast_fp16)[name = tensor("op_14848_cast_fp16")]; tensor var_14848_cast_fp16_to_fp32_dtype_0 = const()[name = tensor("op_14848_cast_fp16_to_fp32_dtype_0"), val = tensor("fp32")]; tensor noise_pred = cast(dtype = var_14848_cast_fp16_to_fp32_dtype_0, x = var_14848_cast_fp16)[name = tensor("cast_0")]; } -> (noise_pred); }