diff --git "a/Resources/UnetChunk1.mlmodelc/model.mil" "b/Resources/UnetChunk1.mlmodelc/model.mil" new file mode 100644--- /dev/null +++ "b/Resources/UnetChunk1.mlmodelc/model.mil" @@ -0,0 +1,7320 @@ +program(1.0) +[buildInfo = dict, tensor>({{"coremlc-component-MIL", "3500.14.1"}, {"coremlc-version", "3500.32.1"}})] +{ + func main(tensor encoder_hidden_states, tensor sample, tensor timestep) { + tensor var_25 = const()[name = tensor("op_25"), val = tensor(-1)]; + tensor var_42_axes_0 = const()[name = tensor("op_42_axes_0"), val = tensor([1])]; + tensor var_42_cast_fp16 = expand_dims(axes = var_42_axes_0, x = timestep)[name = tensor("op_42_cast_fp16")]; + tensor var_44_to_fp16 = const()[name = tensor("op_44_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; + tensor emb_3_cast_fp16 = mul(x = var_42_cast_fp16, y = var_44_to_fp16)[name = tensor("emb_3_cast_fp16")]; + tensor var_49_cast_fp16 = sin(x = emb_3_cast_fp16)[name = tensor("op_49_cast_fp16")]; + tensor var_50_cast_fp16 = cos(x = emb_3_cast_fp16)[name = tensor("op_50_cast_fp16")]; + tensor emb_interleave_0 = const()[name = tensor("emb_interleave_0"), val = tensor(false)]; + tensor emb_cast_fp16 = concat(axis = var_25, interleave = emb_interleave_0, values = (var_49_cast_fp16, var_50_cast_fp16))[name = tensor("emb_cast_fp16")]; + tensor var_54_begin_0 = const()[name = tensor("op_54_begin_0"), val = tensor([0, 160])]; + tensor var_54_end_0 = const()[name = tensor("op_54_end_0"), val = tensor([2, 320])]; + tensor var_54_end_mask_0 = const()[name = tensor("op_54_end_mask_0"), val = tensor([true, true])]; + tensor var_54_cast_fp16 = slice_by_index(begin = var_54_begin_0, end = var_54_end_0, end_mask = var_54_end_mask_0, x = emb_cast_fp16)[name = tensor("op_54_cast_fp16")]; + tensor var_56_begin_0 = const()[name = tensor("op_56_begin_0"), val = tensor([0, 0])]; + tensor var_56_end_0 = const()[name = tensor("op_56_end_0"), val = tensor([2, 160])]; + tensor var_56_end_mask_0 = const()[name = tensor("op_56_end_mask_0"), val = tensor([true, false])]; + tensor var_56_cast_fp16 = slice_by_index(begin = var_56_begin_0, end = var_56_end_0, end_mask = var_56_end_mask_0, x = emb_cast_fp16)[name = tensor("op_56_cast_fp16")]; + tensor sample_interleave_0 = const()[name = tensor("sample_interleave_0"), val = tensor(false)]; + tensor sample_cast_fp16 = concat(axis = var_25, interleave = sample_interleave_0, values = (var_54_cast_fp16, var_56_cast_fp16))[name = tensor("sample_cast_fp16")]; + tensor var_66_axes_0 = const()[name = tensor("op_66_axes_0"), val = tensor([-1])]; + tensor var_66_cast_fp16 = expand_dims(axes = var_66_axes_0, x = sample_cast_fp16)[name = tensor("op_66_cast_fp16")]; + tensor input_1_axes_0 = const()[name = tensor("input_1_axes_0"), val = tensor([-1])]; + tensor input_1_cast_fp16 = expand_dims(axes = input_1_axes_0, x = var_66_cast_fp16)[name = tensor("input_1_cast_fp16")]; + tensor input_3_pad_type_0 = const()[name = tensor("input_3_pad_type_0"), val = tensor("valid")]; + tensor input_3_strides_0 = const()[name = tensor("input_3_strides_0"), val = tensor([1, 1])]; + tensor input_3_pad_0 = const()[name = tensor("input_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_3_dilations_0 = const()[name = tensor("input_3_dilations_0"), val = tensor([1, 1])]; + tensor input_3_groups_0 = const()[name = tensor("input_3_groups_0"), val = tensor(1)]; + tensor time_embedding_linear_1_weight_to_fp16 = const()[name = tensor("time_embedding_linear_1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(448)))]; + tensor time_embedding_linear_1_bias_to_fp16 = const()[name = tensor("time_embedding_linear_1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(819712)))]; + tensor input_3_cast_fp16 = conv(bias = time_embedding_linear_1_bias_to_fp16, dilations = input_3_dilations_0, groups = input_3_groups_0, pad = input_3_pad_0, pad_type = input_3_pad_type_0, strides = input_3_strides_0, weight = time_embedding_linear_1_weight_to_fp16, x = input_1_cast_fp16)[name = tensor("input_3_cast_fp16")]; + tensor input_5_cast_fp16 = silu(x = input_3_cast_fp16)[name = tensor("input_5_cast_fp16")]; + tensor input_13_pad_type_0 = const()[name = tensor("input_13_pad_type_0"), val = tensor("valid")]; + tensor input_13_strides_0 = const()[name = tensor("input_13_strides_0"), val = tensor([1, 1])]; + tensor input_13_pad_0 = const()[name = tensor("input_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_13_dilations_0 = const()[name = tensor("input_13_dilations_0"), val = tensor([1, 1])]; + tensor input_13_groups_0 = const()[name = tensor("input_13_groups_0"), val = tensor(1)]; + tensor time_embedding_linear_2_weight_to_fp16 = const()[name = tensor("time_embedding_linear_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(822336)))]; + tensor time_embedding_linear_2_bias_to_fp16 = const()[name = tensor("time_embedding_linear_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4099200)))]; + tensor input_13_cast_fp16 = conv(bias = time_embedding_linear_2_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = time_embedding_linear_2_weight_to_fp16, x = input_5_cast_fp16)[name = tensor("input_13_cast_fp16")]; + tensor input_7_pad_type_0 = const()[name = tensor("input_7_pad_type_0"), val = tensor("custom")]; + tensor input_7_pad_0 = const()[name = tensor("input_7_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor input_7_strides_0 = const()[name = tensor("input_7_strides_0"), val = tensor([1, 1])]; + tensor input_7_dilations_0 = const()[name = tensor("input_7_dilations_0"), val = tensor([1, 1])]; + tensor input_7_groups_0 = const()[name = tensor("input_7_groups_0"), val = tensor(1)]; + tensor conv_in_weight_to_fp16 = const()[name = tensor("conv_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4101824)))]; + tensor conv_in_bias_to_fp16 = const()[name = tensor("conv_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4124928)))]; + tensor input_7_cast_fp16_1 = conv(bias = conv_in_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = conv_in_weight_to_fp16, x = sample)[name = tensor("input_7_cast_fp16")]; + tensor var_95 = const()[name = tensor("op_95"), val = tensor(3)]; + tensor var_123 = const()[name = tensor("op_123"), val = tensor(1)]; + tensor reshape_0_shape_0 = const()[name = tensor("reshape_0_shape_0"), val = tensor([2, 32, 10, 64, 64])]; + tensor reshape_0_cast_fp16 = reshape(shape = reshape_0_shape_0, x = input_7_cast_fp16_1)[name = tensor("reshape_0_cast_fp16")]; + tensor reduce_mean_0_axes_0 = const()[name = tensor("reduce_mean_0_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_0_keep_dims_0 = const()[name = tensor("reduce_mean_0_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_0_cast_fp16 = reduce_mean(axes = reduce_mean_0_axes_0, keep_dims = reduce_mean_0_keep_dims_0, x = reshape_0_cast_fp16)[name = tensor("reduce_mean_0_cast_fp16")]; + tensor sub_0_cast_fp16 = sub(x = reshape_0_cast_fp16, y = reduce_mean_0_cast_fp16)[name = tensor("sub_0_cast_fp16")]; + tensor square_0_cast_fp16 = square(x = sub_0_cast_fp16)[name = tensor("square_0_cast_fp16")]; + tensor reduce_mean_2_axes_0 = const()[name = tensor("reduce_mean_2_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_2_keep_dims_0 = const()[name = tensor("reduce_mean_2_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_2_cast_fp16 = reduce_mean(axes = reduce_mean_2_axes_0, keep_dims = reduce_mean_2_keep_dims_0, x = square_0_cast_fp16)[name = tensor("reduce_mean_2_cast_fp16")]; + tensor add_0_y_0_to_fp16 = const()[name = tensor("add_0_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_0_cast_fp16 = add(x = reduce_mean_2_cast_fp16, y = add_0_y_0_to_fp16)[name = tensor("add_0_cast_fp16")]; + tensor sqrt_0_cast_fp16 = sqrt(x = add_0_cast_fp16)[name = tensor("sqrt_0_cast_fp16")]; + tensor real_div_0_cast_fp16 = real_div(x = sub_0_cast_fp16, y = sqrt_0_cast_fp16)[name = tensor("real_div_0_cast_fp16")]; + tensor reshape_1_shape_0 = const()[name = tensor("reshape_1_shape_0"), val = tensor([2, 320, 64, 64])]; + tensor reshape_1_cast_fp16 = reshape(shape = reshape_1_shape_0, x = real_div_0_cast_fp16)[name = tensor("reshape_1_cast_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(4125632)))]; + 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(4126336)))]; + tensor add_1_gamma_0_to_fp16 = const()[name = tensor("add_1_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4127040)))]; + tensor add_1_beta_0_to_fp16 = const()[name = tensor("add_1_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4127744)))]; + tensor add_1_epsilon_0_to_fp16 = const()[name = tensor("add_1_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_1_cast_fp16 = batch_norm(beta = add_1_beta_0_to_fp16, epsilon = add_1_epsilon_0_to_fp16, gamma = add_1_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_1_cast_fp16)[name = tensor("add_1_cast_fp16")]; + tensor input_11_cast_fp16 = silu(x = add_1_cast_fp16)[name = tensor("input_11_cast_fp16")]; + tensor hidden_states_1_pad_type_0 = const()[name = tensor("hidden_states_1_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_1_pad_0 = const()[name = tensor("hidden_states_1_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_1_strides_0 = const()[name = tensor("hidden_states_1_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_1_dilations_0 = const()[name = tensor("hidden_states_1_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_1_groups_0 = const()[name = tensor("hidden_states_1_groups_0"), val = tensor(1)]; + tensor down_blocks_0_resnets_0_conv1_weight_to_fp16 = const()[name = tensor("down_blocks_0_resnets_0_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4128448)))]; + tensor down_blocks_0_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_0_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5971712)))]; + tensor hidden_states_1_cast_fp16 = conv(bias = down_blocks_0_resnets_0_conv1_bias_to_fp16, dilations = hidden_states_1_dilations_0, groups = hidden_states_1_groups_0, pad = hidden_states_1_pad_0, pad_type = hidden_states_1_pad_type_0, strides = hidden_states_1_strides_0, weight = down_blocks_0_resnets_0_conv1_weight_to_fp16, x = input_11_cast_fp16)[name = tensor("hidden_states_1_cast_fp16")]; + tensor input_15_cast_fp16_1 = silu(x = input_13_cast_fp16)[name = tensor("input_15_cast_fp16")]; + tensor temb_1_pad_type_0 = const()[name = tensor("temb_1_pad_type_0"), val = tensor("valid")]; + tensor temb_1_strides_0 = const()[name = tensor("temb_1_strides_0"), val = tensor([1, 1])]; + tensor temb_1_pad_0 = const()[name = tensor("temb_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor temb_1_dilations_0 = const()[name = tensor("temb_1_dilations_0"), val = tensor([1, 1])]; + tensor temb_1_groups_0 = const()[name = tensor("temb_1_groups_0"), val = tensor(1)]; + tensor down_blocks_0_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor("down_blocks_0_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5972416)))]; + tensor down_blocks_0_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_0_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6791680)))]; + tensor temb_1_cast_fp16 = conv(bias = down_blocks_0_resnets_0_time_emb_proj_bias_to_fp16, dilations = temb_1_dilations_0, groups = temb_1_groups_0, pad = temb_1_pad_0, pad_type = temb_1_pad_type_0, strides = temb_1_strides_0, weight = down_blocks_0_resnets_0_time_emb_proj_weight_to_fp16, x = input_15_cast_fp16_1)[name = tensor("temb_1_cast_fp16")]; + tensor input_17_cast_fp16 = add(x = hidden_states_1_cast_fp16, y = temb_1_cast_fp16)[name = tensor("input_17_cast_fp16")]; + tensor reshape_4_shape_0 = const()[name = tensor("reshape_4_shape_0"), val = tensor([2, 32, 10, 64, 64])]; + tensor reshape_4_cast_fp16 = reshape(shape = reshape_4_shape_0, x = input_17_cast_fp16)[name = tensor("reshape_4_cast_fp16")]; + tensor reduce_mean_3_axes_0 = const()[name = tensor("reduce_mean_3_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_3_keep_dims_0 = const()[name = tensor("reduce_mean_3_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_3_cast_fp16 = reduce_mean(axes = reduce_mean_3_axes_0, keep_dims = reduce_mean_3_keep_dims_0, x = reshape_4_cast_fp16)[name = tensor("reduce_mean_3_cast_fp16")]; + tensor sub_2_cast_fp16 = sub(x = reshape_4_cast_fp16, y = reduce_mean_3_cast_fp16)[name = tensor("sub_2_cast_fp16")]; + tensor square_1_cast_fp16 = square(x = sub_2_cast_fp16)[name = tensor("square_1_cast_fp16")]; + tensor reduce_mean_5_axes_0 = const()[name = tensor("reduce_mean_5_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_5_keep_dims_0 = const()[name = tensor("reduce_mean_5_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_5_cast_fp16 = reduce_mean(axes = reduce_mean_5_axes_0, keep_dims = reduce_mean_5_keep_dims_0, x = square_1_cast_fp16)[name = tensor("reduce_mean_5_cast_fp16")]; + tensor add_2_y_0_to_fp16 = const()[name = tensor("add_2_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_2_cast_fp16 = add(x = reduce_mean_5_cast_fp16, y = add_2_y_0_to_fp16)[name = tensor("add_2_cast_fp16")]; + tensor sqrt_1_cast_fp16 = sqrt(x = add_2_cast_fp16)[name = tensor("sqrt_1_cast_fp16")]; + tensor real_div_1_cast_fp16 = real_div(x = sub_2_cast_fp16, y = sqrt_1_cast_fp16)[name = tensor("real_div_1_cast_fp16")]; + tensor reshape_5_shape_0 = const()[name = tensor("reshape_5_shape_0"), val = tensor([2, 320, 64, 64])]; + tensor reshape_5_cast_fp16 = reshape(shape = reshape_5_shape_0, x = real_div_1_cast_fp16)[name = tensor("reshape_5_cast_fp16")]; + tensor add_3_gamma_0_to_fp16 = const()[name = tensor("add_3_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6792384)))]; + tensor add_3_beta_0_to_fp16 = const()[name = tensor("add_3_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6793088)))]; + tensor add_3_epsilon_0_to_fp16 = const()[name = tensor("add_3_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_3_cast_fp16 = batch_norm(beta = add_3_beta_0_to_fp16, epsilon = add_3_epsilon_0_to_fp16, gamma = add_3_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_5_cast_fp16)[name = tensor("add_3_cast_fp16")]; + tensor input_21_cast_fp16 = silu(x = add_3_cast_fp16)[name = tensor("input_21_cast_fp16")]; + tensor hidden_states_3_pad_type_0 = const()[name = tensor("hidden_states_3_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_3_pad_0 = const()[name = tensor("hidden_states_3_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_3_strides_0 = const()[name = tensor("hidden_states_3_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_3_dilations_0 = const()[name = tensor("hidden_states_3_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_3_groups_0 = const()[name = tensor("hidden_states_3_groups_0"), val = tensor(1)]; + tensor down_blocks_0_resnets_0_conv2_weight_to_fp16 = const()[name = tensor("down_blocks_0_resnets_0_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6793792)))]; + tensor down_blocks_0_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_0_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8637056)))]; + tensor hidden_states_3_cast_fp16 = conv(bias = down_blocks_0_resnets_0_conv2_bias_to_fp16, dilations = hidden_states_3_dilations_0, groups = hidden_states_3_groups_0, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = hidden_states_3_strides_0, weight = down_blocks_0_resnets_0_conv2_weight_to_fp16, x = input_21_cast_fp16)[name = tensor("hidden_states_3_cast_fp16")]; + tensor hidden_states_5_cast_fp16 = add(x = input_7_cast_fp16_1, y = hidden_states_3_cast_fp16)[name = tensor("hidden_states_5_cast_fp16")]; + tensor reshape_8_shape_0 = const()[name = tensor("reshape_8_shape_0"), val = tensor([2, 32, 10, 64, 64])]; + tensor reshape_8_cast_fp16 = reshape(shape = reshape_8_shape_0, x = hidden_states_5_cast_fp16)[name = tensor("reshape_8_cast_fp16")]; + tensor reduce_mean_6_axes_0 = const()[name = tensor("reduce_mean_6_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_6_keep_dims_0 = const()[name = tensor("reduce_mean_6_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_6_cast_fp16 = reduce_mean(axes = reduce_mean_6_axes_0, keep_dims = reduce_mean_6_keep_dims_0, x = reshape_8_cast_fp16)[name = tensor("reduce_mean_6_cast_fp16")]; + tensor sub_4_cast_fp16 = sub(x = reshape_8_cast_fp16, y = reduce_mean_6_cast_fp16)[name = tensor("sub_4_cast_fp16")]; + tensor square_2_cast_fp16 = square(x = sub_4_cast_fp16)[name = tensor("square_2_cast_fp16")]; + tensor reduce_mean_8_axes_0 = const()[name = tensor("reduce_mean_8_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_8_keep_dims_0 = const()[name = tensor("reduce_mean_8_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_8_cast_fp16 = reduce_mean(axes = reduce_mean_8_axes_0, keep_dims = reduce_mean_8_keep_dims_0, x = square_2_cast_fp16)[name = tensor("reduce_mean_8_cast_fp16")]; + tensor add_4_y_0_to_fp16 = const()[name = tensor("add_4_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_4_cast_fp16 = add(x = reduce_mean_8_cast_fp16, y = add_4_y_0_to_fp16)[name = tensor("add_4_cast_fp16")]; + tensor sqrt_2_cast_fp16 = sqrt(x = add_4_cast_fp16)[name = tensor("sqrt_2_cast_fp16")]; + tensor real_div_2_cast_fp16 = real_div(x = sub_4_cast_fp16, y = sqrt_2_cast_fp16)[name = tensor("real_div_2_cast_fp16")]; + tensor reshape_9_shape_0 = const()[name = tensor("reshape_9_shape_0"), val = tensor([2, 320, 64, 64])]; + tensor reshape_9_cast_fp16 = reshape(shape = reshape_9_shape_0, x = real_div_2_cast_fp16)[name = tensor("reshape_9_cast_fp16")]; + tensor add_5_gamma_0_to_fp16 = const()[name = tensor("add_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8637760)))]; + tensor add_5_beta_0_to_fp16 = const()[name = tensor("add_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8638464)))]; + tensor add_5_epsilon_0_to_fp16 = const()[name = tensor("add_5_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_5_cast_fp16 = batch_norm(beta = add_5_beta_0_to_fp16, epsilon = add_5_epsilon_0_to_fp16, gamma = add_5_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_9_cast_fp16)[name = tensor("add_5_cast_fp16")]; + tensor hidden_states_7_pad_type_0 = const()[name = tensor("hidden_states_7_pad_type_0"), val = tensor("valid")]; + tensor hidden_states_7_strides_0 = const()[name = tensor("hidden_states_7_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_7_pad_0 = const()[name = tensor("hidden_states_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_7_dilations_0 = const()[name = tensor("hidden_states_7_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_7_groups_0 = const()[name = tensor("hidden_states_7_groups_0"), val = tensor(1)]; + tensor down_blocks_0_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor("down_blocks_0_attentions_0_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8639168)))]; + tensor down_blocks_0_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor("down_blocks_0_attentions_0_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8844032)))]; + tensor hidden_states_7_cast_fp16 = conv(bias = down_blocks_0_attentions_0_proj_in_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = down_blocks_0_attentions_0_proj_in_weight_to_fp16, x = add_5_cast_fp16)[name = tensor("hidden_states_7_cast_fp16")]; + tensor var_194 = const()[name = tensor("op_194"), val = tensor([2, 320, 1, 4096])]; + tensor inputs_1_cast_fp16 = reshape(shape = var_194, x = hidden_states_7_cast_fp16)[name = tensor("inputs_1_cast_fp16")]; + tensor hidden_states_9_axes_0 = const()[name = tensor("hidden_states_9_axes_0"), val = tensor([1])]; + tensor hidden_states_9_gamma_0_to_fp16 = const()[name = tensor("hidden_states_9_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8844736)))]; + tensor hidden_states_9_beta_0_to_fp16 = const()[name = tensor("hidden_states_9_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8845440)))]; + tensor var_210_to_fp16 = const()[name = tensor("op_210_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_9_cast_fp16 = layer_norm(axes = hidden_states_9_axes_0, beta = hidden_states_9_beta_0_to_fp16, epsilon = var_210_to_fp16, gamma = hidden_states_9_gamma_0_to_fp16, x = inputs_1_cast_fp16)[name = tensor("hidden_states_9_cast_fp16")]; + tensor q_1_pad_type_0 = const()[name = tensor("q_1_pad_type_0"), val = tensor("valid")]; + tensor q_1_strides_0 = const()[name = tensor("q_1_strides_0"), val = tensor([1, 1])]; + tensor q_1_pad_0 = const()[name = tensor("q_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_1_dilations_0 = const()[name = tensor("q_1_dilations_0"), val = tensor([1, 1])]; + tensor q_1_groups_0 = const()[name = tensor("q_1_groups_0"), val = tensor(1)]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8846144)))]; + tensor q_1_cast_fp16 = conv(dilations = q_1_dilations_0, groups = q_1_groups_0, pad = q_1_pad_0, pad_type = q_1_pad_type_0, strides = q_1_strides_0, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_9_cast_fp16)[name = tensor("q_1_cast_fp16")]; + tensor k_1_pad_type_0 = const()[name = tensor("k_1_pad_type_0"), val = tensor("valid")]; + tensor k_1_strides_0 = const()[name = tensor("k_1_strides_0"), val = tensor([1, 1])]; + tensor k_1_pad_0 = const()[name = tensor("k_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_1_dilations_0 = const()[name = tensor("k_1_dilations_0"), val = tensor([1, 1])]; + tensor k_1_groups_0 = const()[name = tensor("k_1_groups_0"), val = tensor(1)]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9051008)))]; + tensor k_1_cast_fp16 = conv(dilations = k_1_dilations_0, groups = k_1_groups_0, pad = k_1_pad_0, pad_type = k_1_pad_type_0, strides = k_1_strides_0, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_9_cast_fp16)[name = tensor("k_1_cast_fp16")]; + tensor v_1_pad_type_0 = const()[name = tensor("v_1_pad_type_0"), val = tensor("valid")]; + tensor v_1_strides_0 = const()[name = tensor("v_1_strides_0"), val = tensor([1, 1])]; + tensor v_1_pad_0 = const()[name = tensor("v_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_1_dilations_0 = const()[name = tensor("v_1_dilations_0"), val = tensor([1, 1])]; + tensor v_1_groups_0 = const()[name = tensor("v_1_groups_0"), val = tensor(1)]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9255872)))]; + tensor v_1_cast_fp16 = conv(dilations = v_1_dilations_0, groups = v_1_groups_0, pad = v_1_pad_0, pad_type = v_1_pad_type_0, strides = v_1_strides_0, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_9_cast_fp16)[name = tensor("v_1_cast_fp16")]; + tensor var_243_begin_0 = const()[name = tensor("op_243_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_243_end_0 = const()[name = tensor("op_243_end_0"), val = tensor([2, 40, 1, 4096])]; + tensor var_243_end_mask_0 = const()[name = tensor("op_243_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_243_cast_fp16 = slice_by_index(begin = var_243_begin_0, end = var_243_end_0, end_mask = var_243_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_243_cast_fp16")]; + tensor var_247_begin_0 = const()[name = tensor("op_247_begin_0"), val = tensor([0, 40, 0, 0])]; + tensor var_247_end_0 = const()[name = tensor("op_247_end_0"), val = tensor([2, 80, 1, 4096])]; + tensor var_247_end_mask_0 = const()[name = tensor("op_247_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_247_cast_fp16 = slice_by_index(begin = var_247_begin_0, end = var_247_end_0, end_mask = var_247_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_247_cast_fp16")]; + tensor var_251_begin_0 = const()[name = tensor("op_251_begin_0"), val = tensor([0, 80, 0, 0])]; + tensor var_251_end_0 = const()[name = tensor("op_251_end_0"), val = tensor([2, 120, 1, 4096])]; + tensor var_251_end_mask_0 = const()[name = tensor("op_251_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_251_cast_fp16 = slice_by_index(begin = var_251_begin_0, end = var_251_end_0, end_mask = var_251_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_251_cast_fp16")]; + tensor var_255_begin_0 = const()[name = tensor("op_255_begin_0"), val = tensor([0, 120, 0, 0])]; + tensor var_255_end_0 = const()[name = tensor("op_255_end_0"), val = tensor([2, 160, 1, 4096])]; + tensor var_255_end_mask_0 = const()[name = tensor("op_255_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_255_cast_fp16 = slice_by_index(begin = var_255_begin_0, end = var_255_end_0, end_mask = var_255_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_255_cast_fp16")]; + tensor var_259_begin_0 = const()[name = tensor("op_259_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_259_end_0 = const()[name = tensor("op_259_end_0"), val = tensor([2, 200, 1, 4096])]; + tensor var_259_end_mask_0 = const()[name = tensor("op_259_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_259_cast_fp16 = slice_by_index(begin = var_259_begin_0, end = var_259_end_0, end_mask = var_259_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_259_cast_fp16")]; + tensor var_263_begin_0 = const()[name = tensor("op_263_begin_0"), val = tensor([0, 200, 0, 0])]; + tensor var_263_end_0 = const()[name = tensor("op_263_end_0"), val = tensor([2, 240, 1, 4096])]; + tensor var_263_end_mask_0 = const()[name = tensor("op_263_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_263_cast_fp16 = slice_by_index(begin = var_263_begin_0, end = var_263_end_0, end_mask = var_263_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_263_cast_fp16")]; + tensor var_267_begin_0 = const()[name = tensor("op_267_begin_0"), val = tensor([0, 240, 0, 0])]; + tensor var_267_end_0 = const()[name = tensor("op_267_end_0"), val = tensor([2, 280, 1, 4096])]; + tensor var_267_end_mask_0 = const()[name = tensor("op_267_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_267_cast_fp16 = slice_by_index(begin = var_267_begin_0, end = var_267_end_0, end_mask = var_267_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_267_cast_fp16")]; + tensor var_271_begin_0 = const()[name = tensor("op_271_begin_0"), val = tensor([0, 280, 0, 0])]; + tensor var_271_end_0 = const()[name = tensor("op_271_end_0"), val = tensor([2, 1, 1, 4096])]; + tensor var_271_end_mask_0 = const()[name = tensor("op_271_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_271_cast_fp16 = slice_by_index(begin = var_271_begin_0, end = var_271_end_0, end_mask = var_271_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_271_cast_fp16")]; + tensor var_274_begin_0 = const()[name = tensor("op_274_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_274_end_0 = const()[name = tensor("op_274_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_274_end_mask_0 = const()[name = tensor("op_274_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_274_cast_fp16 = slice_by_index(begin = var_274_begin_0, end = var_274_end_0, end_mask = var_274_end_mask_0, x = var_243_cast_fp16)[name = tensor("op_274_cast_fp16")]; + tensor var_275_begin_0 = const()[name = tensor("op_275_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_275_end_0 = const()[name = tensor("op_275_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_275_end_mask_0 = const()[name = tensor("op_275_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_275_cast_fp16 = slice_by_index(begin = var_275_begin_0, end = var_275_end_0, end_mask = var_275_end_mask_0, x = var_243_cast_fp16)[name = tensor("op_275_cast_fp16")]; + tensor var_276_begin_0 = const()[name = tensor("op_276_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_276_end_0 = const()[name = tensor("op_276_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_276_end_mask_0 = const()[name = tensor("op_276_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_276_cast_fp16 = slice_by_index(begin = var_276_begin_0, end = var_276_end_0, end_mask = var_276_end_mask_0, x = var_243_cast_fp16)[name = tensor("op_276_cast_fp16")]; + tensor var_277_begin_0 = const()[name = tensor("op_277_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_277_end_0 = const()[name = tensor("op_277_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_277_end_mask_0 = const()[name = tensor("op_277_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_277_cast_fp16 = slice_by_index(begin = var_277_begin_0, end = var_277_end_0, end_mask = var_277_end_mask_0, x = var_243_cast_fp16)[name = tensor("op_277_cast_fp16")]; + tensor var_278_begin_0 = const()[name = tensor("op_278_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_278_end_0 = const()[name = tensor("op_278_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_278_end_mask_0 = const()[name = tensor("op_278_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_278_cast_fp16 = slice_by_index(begin = var_278_begin_0, end = var_278_end_0, end_mask = var_278_end_mask_0, x = var_243_cast_fp16)[name = tensor("op_278_cast_fp16")]; + tensor var_279_begin_0 = const()[name = tensor("op_279_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_279_end_0 = const()[name = tensor("op_279_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_279_end_mask_0 = const()[name = tensor("op_279_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_279_cast_fp16 = slice_by_index(begin = var_279_begin_0, end = var_279_end_0, end_mask = var_279_end_mask_0, x = var_243_cast_fp16)[name = tensor("op_279_cast_fp16")]; + tensor var_280_begin_0 = const()[name = tensor("op_280_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_280_end_0 = const()[name = tensor("op_280_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_280_end_mask_0 = const()[name = tensor("op_280_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_280_cast_fp16 = slice_by_index(begin = var_280_begin_0, end = var_280_end_0, end_mask = var_280_end_mask_0, x = var_243_cast_fp16)[name = tensor("op_280_cast_fp16")]; + tensor var_281_begin_0 = const()[name = tensor("op_281_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_281_end_0 = const()[name = tensor("op_281_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_281_end_mask_0 = const()[name = tensor("op_281_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_281_cast_fp16 = slice_by_index(begin = var_281_begin_0, end = var_281_end_0, end_mask = var_281_end_mask_0, x = var_243_cast_fp16)[name = tensor("op_281_cast_fp16")]; + tensor var_282_begin_0 = const()[name = tensor("op_282_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_282_end_0 = const()[name = tensor("op_282_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_282_end_mask_0 = const()[name = tensor("op_282_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_282_cast_fp16 = slice_by_index(begin = var_282_begin_0, end = var_282_end_0, end_mask = var_282_end_mask_0, x = var_247_cast_fp16)[name = tensor("op_282_cast_fp16")]; + tensor var_283_begin_0 = const()[name = tensor("op_283_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_283_end_0 = const()[name = tensor("op_283_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_283_end_mask_0 = const()[name = tensor("op_283_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_283_cast_fp16 = slice_by_index(begin = var_283_begin_0, end = var_283_end_0, end_mask = var_283_end_mask_0, x = var_247_cast_fp16)[name = tensor("op_283_cast_fp16")]; + tensor var_284_begin_0 = const()[name = tensor("op_284_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_284_end_0 = const()[name = tensor("op_284_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_284_end_mask_0 = const()[name = tensor("op_284_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_284_cast_fp16 = slice_by_index(begin = var_284_begin_0, end = var_284_end_0, end_mask = var_284_end_mask_0, x = var_247_cast_fp16)[name = tensor("op_284_cast_fp16")]; + tensor var_285_begin_0 = const()[name = tensor("op_285_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_285_end_0 = const()[name = tensor("op_285_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_285_end_mask_0 = const()[name = tensor("op_285_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_285_cast_fp16 = slice_by_index(begin = var_285_begin_0, end = var_285_end_0, end_mask = var_285_end_mask_0, x = var_247_cast_fp16)[name = tensor("op_285_cast_fp16")]; + tensor var_286_begin_0 = const()[name = tensor("op_286_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_286_end_0 = const()[name = tensor("op_286_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_286_end_mask_0 = const()[name = tensor("op_286_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_286_cast_fp16 = slice_by_index(begin = var_286_begin_0, end = var_286_end_0, end_mask = var_286_end_mask_0, x = var_247_cast_fp16)[name = tensor("op_286_cast_fp16")]; + tensor var_287_begin_0 = const()[name = tensor("op_287_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_287_end_0 = const()[name = tensor("op_287_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_287_end_mask_0 = const()[name = tensor("op_287_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_287_cast_fp16 = slice_by_index(begin = var_287_begin_0, end = var_287_end_0, end_mask = var_287_end_mask_0, x = var_247_cast_fp16)[name = tensor("op_287_cast_fp16")]; + tensor var_288_begin_0 = const()[name = tensor("op_288_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_288_end_0 = const()[name = tensor("op_288_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_288_end_mask_0 = const()[name = tensor("op_288_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_288_cast_fp16 = slice_by_index(begin = var_288_begin_0, end = var_288_end_0, end_mask = var_288_end_mask_0, x = var_247_cast_fp16)[name = tensor("op_288_cast_fp16")]; + tensor var_289_begin_0 = const()[name = tensor("op_289_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_289_end_0 = const()[name = tensor("op_289_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_289_end_mask_0 = const()[name = tensor("op_289_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_289_cast_fp16 = slice_by_index(begin = var_289_begin_0, end = var_289_end_0, end_mask = var_289_end_mask_0, x = var_247_cast_fp16)[name = tensor("op_289_cast_fp16")]; + tensor var_290_begin_0 = const()[name = tensor("op_290_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_290_end_0 = const()[name = tensor("op_290_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_290_end_mask_0 = const()[name = tensor("op_290_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_290_cast_fp16 = slice_by_index(begin = var_290_begin_0, end = var_290_end_0, end_mask = var_290_end_mask_0, x = var_251_cast_fp16)[name = tensor("op_290_cast_fp16")]; + tensor var_291_begin_0 = const()[name = tensor("op_291_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_291_end_0 = const()[name = tensor("op_291_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_291_end_mask_0 = const()[name = tensor("op_291_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_291_cast_fp16 = slice_by_index(begin = var_291_begin_0, end = var_291_end_0, end_mask = var_291_end_mask_0, x = var_251_cast_fp16)[name = tensor("op_291_cast_fp16")]; + tensor var_292_begin_0 = const()[name = tensor("op_292_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_292_end_0 = const()[name = tensor("op_292_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_292_end_mask_0 = const()[name = tensor("op_292_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_292_cast_fp16 = slice_by_index(begin = var_292_begin_0, end = var_292_end_0, end_mask = var_292_end_mask_0, x = var_251_cast_fp16)[name = tensor("op_292_cast_fp16")]; + tensor var_293_begin_0 = const()[name = tensor("op_293_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_293_end_0 = const()[name = tensor("op_293_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_293_end_mask_0 = const()[name = tensor("op_293_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_293_cast_fp16 = slice_by_index(begin = var_293_begin_0, end = var_293_end_0, end_mask = var_293_end_mask_0, x = var_251_cast_fp16)[name = tensor("op_293_cast_fp16")]; + tensor var_294_begin_0 = const()[name = tensor("op_294_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_294_end_0 = const()[name = tensor("op_294_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_294_end_mask_0 = const()[name = tensor("op_294_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_294_cast_fp16 = slice_by_index(begin = var_294_begin_0, end = var_294_end_0, end_mask = var_294_end_mask_0, x = var_251_cast_fp16)[name = tensor("op_294_cast_fp16")]; + tensor var_295_begin_0 = const()[name = tensor("op_295_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_295_end_0 = const()[name = tensor("op_295_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_295_end_mask_0 = const()[name = tensor("op_295_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_295_cast_fp16 = slice_by_index(begin = var_295_begin_0, end = var_295_end_0, end_mask = var_295_end_mask_0, x = var_251_cast_fp16)[name = tensor("op_295_cast_fp16")]; + tensor var_296_begin_0 = const()[name = tensor("op_296_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_296_end_0 = const()[name = tensor("op_296_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_296_end_mask_0 = const()[name = tensor("op_296_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_296_cast_fp16 = slice_by_index(begin = var_296_begin_0, end = var_296_end_0, end_mask = var_296_end_mask_0, x = var_251_cast_fp16)[name = tensor("op_296_cast_fp16")]; + tensor var_297_begin_0 = const()[name = tensor("op_297_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_297_end_0 = const()[name = tensor("op_297_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_297_end_mask_0 = const()[name = tensor("op_297_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_297_cast_fp16 = slice_by_index(begin = var_297_begin_0, end = var_297_end_0, end_mask = var_297_end_mask_0, x = var_251_cast_fp16)[name = tensor("op_297_cast_fp16")]; + tensor var_298_begin_0 = const()[name = tensor("op_298_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_298_end_0 = const()[name = tensor("op_298_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_298_end_mask_0 = const()[name = tensor("op_298_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_298_cast_fp16 = slice_by_index(begin = var_298_begin_0, end = var_298_end_0, end_mask = var_298_end_mask_0, x = var_255_cast_fp16)[name = tensor("op_298_cast_fp16")]; + tensor var_299_begin_0 = const()[name = tensor("op_299_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_299_end_0 = const()[name = tensor("op_299_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_299_end_mask_0 = const()[name = tensor("op_299_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_299_cast_fp16 = slice_by_index(begin = var_299_begin_0, end = var_299_end_0, end_mask = var_299_end_mask_0, x = var_255_cast_fp16)[name = tensor("op_299_cast_fp16")]; + tensor var_300_begin_0 = const()[name = tensor("op_300_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_300_end_0 = const()[name = tensor("op_300_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_300_end_mask_0 = const()[name = tensor("op_300_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_300_cast_fp16 = slice_by_index(begin = var_300_begin_0, end = var_300_end_0, end_mask = var_300_end_mask_0, x = var_255_cast_fp16)[name = tensor("op_300_cast_fp16")]; + tensor var_301_begin_0 = const()[name = tensor("op_301_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_301_end_0 = const()[name = tensor("op_301_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_301_end_mask_0 = const()[name = tensor("op_301_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_301_cast_fp16 = slice_by_index(begin = var_301_begin_0, end = var_301_end_0, end_mask = var_301_end_mask_0, x = var_255_cast_fp16)[name = tensor("op_301_cast_fp16")]; + tensor var_302_begin_0 = const()[name = tensor("op_302_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_302_end_0 = const()[name = tensor("op_302_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_302_end_mask_0 = const()[name = tensor("op_302_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_302_cast_fp16 = slice_by_index(begin = var_302_begin_0, end = var_302_end_0, end_mask = var_302_end_mask_0, x = var_255_cast_fp16)[name = tensor("op_302_cast_fp16")]; + tensor var_303_begin_0 = const()[name = tensor("op_303_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_303_end_0 = const()[name = tensor("op_303_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_303_end_mask_0 = const()[name = tensor("op_303_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_303_cast_fp16 = slice_by_index(begin = var_303_begin_0, end = var_303_end_0, end_mask = var_303_end_mask_0, x = var_255_cast_fp16)[name = tensor("op_303_cast_fp16")]; + tensor var_304_begin_0 = const()[name = tensor("op_304_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_304_end_0 = const()[name = tensor("op_304_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_304_end_mask_0 = const()[name = tensor("op_304_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_304_cast_fp16 = slice_by_index(begin = var_304_begin_0, end = var_304_end_0, end_mask = var_304_end_mask_0, x = var_255_cast_fp16)[name = tensor("op_304_cast_fp16")]; + tensor var_305_begin_0 = const()[name = tensor("op_305_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_305_end_0 = const()[name = tensor("op_305_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_305_end_mask_0 = const()[name = tensor("op_305_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_305_cast_fp16 = slice_by_index(begin = var_305_begin_0, end = var_305_end_0, end_mask = var_305_end_mask_0, x = var_255_cast_fp16)[name = tensor("op_305_cast_fp16")]; + tensor var_306_begin_0 = const()[name = tensor("op_306_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_306_end_0 = const()[name = tensor("op_306_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_306_end_mask_0 = const()[name = tensor("op_306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_306_cast_fp16 = slice_by_index(begin = var_306_begin_0, end = var_306_end_0, end_mask = var_306_end_mask_0, x = var_259_cast_fp16)[name = tensor("op_306_cast_fp16")]; + tensor var_307_begin_0 = const()[name = tensor("op_307_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_307_end_0 = const()[name = tensor("op_307_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_307_end_mask_0 = const()[name = tensor("op_307_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_307_cast_fp16 = slice_by_index(begin = var_307_begin_0, end = var_307_end_0, end_mask = var_307_end_mask_0, x = var_259_cast_fp16)[name = tensor("op_307_cast_fp16")]; + tensor var_308_begin_0 = const()[name = tensor("op_308_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_308_end_0 = const()[name = tensor("op_308_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_308_end_mask_0 = const()[name = tensor("op_308_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_308_cast_fp16 = slice_by_index(begin = var_308_begin_0, end = var_308_end_0, end_mask = var_308_end_mask_0, x = var_259_cast_fp16)[name = tensor("op_308_cast_fp16")]; + tensor var_309_begin_0 = const()[name = tensor("op_309_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_309_end_0 = const()[name = tensor("op_309_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_309_end_mask_0 = const()[name = tensor("op_309_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_309_cast_fp16 = slice_by_index(begin = var_309_begin_0, end = var_309_end_0, end_mask = var_309_end_mask_0, x = var_259_cast_fp16)[name = tensor("op_309_cast_fp16")]; + tensor var_310_begin_0 = const()[name = tensor("op_310_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_310_end_0 = const()[name = tensor("op_310_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_310_end_mask_0 = const()[name = tensor("op_310_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_310_cast_fp16 = slice_by_index(begin = var_310_begin_0, end = var_310_end_0, end_mask = var_310_end_mask_0, x = var_259_cast_fp16)[name = tensor("op_310_cast_fp16")]; + tensor var_311_begin_0 = const()[name = tensor("op_311_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_311_end_0 = const()[name = tensor("op_311_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_311_end_mask_0 = const()[name = tensor("op_311_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_311_cast_fp16 = slice_by_index(begin = var_311_begin_0, end = var_311_end_0, end_mask = var_311_end_mask_0, x = var_259_cast_fp16)[name = tensor("op_311_cast_fp16")]; + tensor var_312_begin_0 = const()[name = tensor("op_312_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_312_end_0 = const()[name = tensor("op_312_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_312_end_mask_0 = const()[name = tensor("op_312_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_312_cast_fp16 = slice_by_index(begin = var_312_begin_0, end = var_312_end_0, end_mask = var_312_end_mask_0, x = var_259_cast_fp16)[name = tensor("op_312_cast_fp16")]; + tensor var_313_begin_0 = const()[name = tensor("op_313_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_313_end_0 = const()[name = tensor("op_313_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_313_end_mask_0 = const()[name = tensor("op_313_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_313_cast_fp16 = slice_by_index(begin = var_313_begin_0, end = var_313_end_0, end_mask = var_313_end_mask_0, x = var_259_cast_fp16)[name = tensor("op_313_cast_fp16")]; + tensor var_314_begin_0 = const()[name = tensor("op_314_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_314_end_0 = const()[name = tensor("op_314_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_314_end_mask_0 = const()[name = tensor("op_314_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_314_cast_fp16 = slice_by_index(begin = var_314_begin_0, end = var_314_end_0, end_mask = var_314_end_mask_0, x = var_263_cast_fp16)[name = tensor("op_314_cast_fp16")]; + tensor var_315_begin_0 = const()[name = tensor("op_315_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_315_end_0 = const()[name = tensor("op_315_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_315_end_mask_0 = const()[name = tensor("op_315_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_315_cast_fp16 = slice_by_index(begin = var_315_begin_0, end = var_315_end_0, end_mask = var_315_end_mask_0, x = var_263_cast_fp16)[name = tensor("op_315_cast_fp16")]; + tensor var_316_begin_0 = const()[name = tensor("op_316_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_316_end_0 = const()[name = tensor("op_316_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_316_end_mask_0 = const()[name = tensor("op_316_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_316_cast_fp16 = slice_by_index(begin = var_316_begin_0, end = var_316_end_0, end_mask = var_316_end_mask_0, x = var_263_cast_fp16)[name = tensor("op_316_cast_fp16")]; + tensor var_317_begin_0 = const()[name = tensor("op_317_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_317_end_0 = const()[name = tensor("op_317_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_317_end_mask_0 = const()[name = tensor("op_317_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_317_cast_fp16 = slice_by_index(begin = var_317_begin_0, end = var_317_end_0, end_mask = var_317_end_mask_0, x = var_263_cast_fp16)[name = tensor("op_317_cast_fp16")]; + tensor var_318_begin_0 = const()[name = tensor("op_318_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_318_end_0 = const()[name = tensor("op_318_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_318_end_mask_0 = const()[name = tensor("op_318_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_318_cast_fp16 = slice_by_index(begin = var_318_begin_0, end = var_318_end_0, end_mask = var_318_end_mask_0, x = var_263_cast_fp16)[name = tensor("op_318_cast_fp16")]; + tensor var_319_begin_0 = const()[name = tensor("op_319_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_319_end_0 = const()[name = tensor("op_319_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_319_end_mask_0 = const()[name = tensor("op_319_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_319_cast_fp16 = slice_by_index(begin = var_319_begin_0, end = var_319_end_0, end_mask = var_319_end_mask_0, x = var_263_cast_fp16)[name = tensor("op_319_cast_fp16")]; + tensor var_320_begin_0 = const()[name = tensor("op_320_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_320_end_0 = const()[name = tensor("op_320_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_320_end_mask_0 = const()[name = tensor("op_320_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_320_cast_fp16 = slice_by_index(begin = var_320_begin_0, end = var_320_end_0, end_mask = var_320_end_mask_0, x = var_263_cast_fp16)[name = tensor("op_320_cast_fp16")]; + tensor var_321_begin_0 = const()[name = tensor("op_321_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_321_end_0 = const()[name = tensor("op_321_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_321_end_mask_0 = const()[name = tensor("op_321_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_321_cast_fp16 = slice_by_index(begin = var_321_begin_0, end = var_321_end_0, end_mask = var_321_end_mask_0, x = var_263_cast_fp16)[name = tensor("op_321_cast_fp16")]; + tensor var_322_begin_0 = const()[name = tensor("op_322_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_322_end_0 = const()[name = tensor("op_322_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_322_end_mask_0 = const()[name = tensor("op_322_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_322_cast_fp16 = slice_by_index(begin = var_322_begin_0, end = var_322_end_0, end_mask = var_322_end_mask_0, x = var_267_cast_fp16)[name = tensor("op_322_cast_fp16")]; + tensor var_323_begin_0 = const()[name = tensor("op_323_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_323_end_0 = const()[name = tensor("op_323_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_323_end_mask_0 = const()[name = tensor("op_323_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_323_cast_fp16 = slice_by_index(begin = var_323_begin_0, end = var_323_end_0, end_mask = var_323_end_mask_0, x = var_267_cast_fp16)[name = tensor("op_323_cast_fp16")]; + tensor var_324_begin_0 = const()[name = tensor("op_324_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_324_end_0 = const()[name = tensor("op_324_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_324_end_mask_0 = const()[name = tensor("op_324_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_324_cast_fp16 = slice_by_index(begin = var_324_begin_0, end = var_324_end_0, end_mask = var_324_end_mask_0, x = var_267_cast_fp16)[name = tensor("op_324_cast_fp16")]; + tensor var_325_begin_0 = const()[name = tensor("op_325_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_325_end_0 = const()[name = tensor("op_325_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_325_end_mask_0 = const()[name = tensor("op_325_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_325_cast_fp16 = slice_by_index(begin = var_325_begin_0, end = var_325_end_0, end_mask = var_325_end_mask_0, x = var_267_cast_fp16)[name = tensor("op_325_cast_fp16")]; + tensor var_326_begin_0 = const()[name = tensor("op_326_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_326_end_0 = const()[name = tensor("op_326_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_326_end_mask_0 = const()[name = tensor("op_326_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_326_cast_fp16 = slice_by_index(begin = var_326_begin_0, end = var_326_end_0, end_mask = var_326_end_mask_0, x = var_267_cast_fp16)[name = tensor("op_326_cast_fp16")]; + tensor var_327_begin_0 = const()[name = tensor("op_327_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_327_end_0 = const()[name = tensor("op_327_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_327_end_mask_0 = const()[name = tensor("op_327_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_327_cast_fp16 = slice_by_index(begin = var_327_begin_0, end = var_327_end_0, end_mask = var_327_end_mask_0, x = var_267_cast_fp16)[name = tensor("op_327_cast_fp16")]; + tensor var_328_begin_0 = const()[name = tensor("op_328_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_328_end_0 = const()[name = tensor("op_328_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_328_end_mask_0 = const()[name = tensor("op_328_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_328_cast_fp16 = slice_by_index(begin = var_328_begin_0, end = var_328_end_0, end_mask = var_328_end_mask_0, x = var_267_cast_fp16)[name = tensor("op_328_cast_fp16")]; + tensor var_329_begin_0 = const()[name = tensor("op_329_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_329_end_0 = const()[name = tensor("op_329_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_329_end_mask_0 = const()[name = tensor("op_329_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_329_cast_fp16 = slice_by_index(begin = var_329_begin_0, end = var_329_end_0, end_mask = var_329_end_mask_0, x = var_267_cast_fp16)[name = tensor("op_329_cast_fp16")]; + tensor var_330_begin_0 = const()[name = tensor("op_330_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_330_end_0 = const()[name = tensor("op_330_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_330_end_mask_0 = const()[name = tensor("op_330_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_330_cast_fp16 = slice_by_index(begin = var_330_begin_0, end = var_330_end_0, end_mask = var_330_end_mask_0, x = var_271_cast_fp16)[name = tensor("op_330_cast_fp16")]; + tensor var_331_begin_0 = const()[name = tensor("op_331_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_331_end_0 = const()[name = tensor("op_331_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_331_end_mask_0 = const()[name = tensor("op_331_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_331_cast_fp16 = slice_by_index(begin = var_331_begin_0, end = var_331_end_0, end_mask = var_331_end_mask_0, x = var_271_cast_fp16)[name = tensor("op_331_cast_fp16")]; + tensor var_332_begin_0 = const()[name = tensor("op_332_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_332_end_0 = const()[name = tensor("op_332_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_332_end_mask_0 = const()[name = tensor("op_332_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_332_cast_fp16 = slice_by_index(begin = var_332_begin_0, end = var_332_end_0, end_mask = var_332_end_mask_0, x = var_271_cast_fp16)[name = tensor("op_332_cast_fp16")]; + tensor var_333_begin_0 = const()[name = tensor("op_333_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_333_end_0 = const()[name = tensor("op_333_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_333_end_mask_0 = const()[name = tensor("op_333_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_333_cast_fp16 = slice_by_index(begin = var_333_begin_0, end = var_333_end_0, end_mask = var_333_end_mask_0, x = var_271_cast_fp16)[name = tensor("op_333_cast_fp16")]; + tensor var_334_begin_0 = const()[name = tensor("op_334_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_334_end_0 = const()[name = tensor("op_334_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_334_end_mask_0 = const()[name = tensor("op_334_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_334_cast_fp16 = slice_by_index(begin = var_334_begin_0, end = var_334_end_0, end_mask = var_334_end_mask_0, x = var_271_cast_fp16)[name = tensor("op_334_cast_fp16")]; + tensor var_335_begin_0 = const()[name = tensor("op_335_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_335_end_0 = const()[name = tensor("op_335_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_335_end_mask_0 = const()[name = tensor("op_335_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_335_cast_fp16 = slice_by_index(begin = var_335_begin_0, end = var_335_end_0, end_mask = var_335_end_mask_0, x = var_271_cast_fp16)[name = tensor("op_335_cast_fp16")]; + tensor var_336_begin_0 = const()[name = tensor("op_336_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_336_end_0 = const()[name = tensor("op_336_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_336_end_mask_0 = const()[name = tensor("op_336_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_336_cast_fp16 = slice_by_index(begin = var_336_begin_0, end = var_336_end_0, end_mask = var_336_end_mask_0, x = var_271_cast_fp16)[name = tensor("op_336_cast_fp16")]; + tensor var_337_begin_0 = const()[name = tensor("op_337_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_337_end_0 = const()[name = tensor("op_337_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_337_end_mask_0 = const()[name = tensor("op_337_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_337_cast_fp16 = slice_by_index(begin = var_337_begin_0, end = var_337_end_0, end_mask = var_337_end_mask_0, x = var_271_cast_fp16)[name = tensor("op_337_cast_fp16")]; + tensor k_3_perm_0 = const()[name = tensor("k_3_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_342_begin_0 = const()[name = tensor("op_342_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_342_end_0 = const()[name = tensor("op_342_end_0"), val = tensor([2, 4096, 1, 40])]; + tensor var_342_end_mask_0 = const()[name = tensor("op_342_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_3_cast_fp16 = transpose(perm = k_3_perm_0, x = k_1_cast_fp16)[name = tensor("transpose_13")]; + tensor var_342_cast_fp16 = slice_by_index(begin = var_342_begin_0, end = var_342_end_0, end_mask = var_342_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_342_cast_fp16")]; + tensor var_346_begin_0 = const()[name = tensor("op_346_begin_0"), val = tensor([0, 0, 0, 40])]; + tensor var_346_end_0 = const()[name = tensor("op_346_end_0"), val = tensor([2, 4096, 1, 80])]; + tensor var_346_end_mask_0 = const()[name = tensor("op_346_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_346_cast_fp16 = slice_by_index(begin = var_346_begin_0, end = var_346_end_0, end_mask = var_346_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_346_cast_fp16")]; + tensor var_350_begin_0 = const()[name = tensor("op_350_begin_0"), val = tensor([0, 0, 0, 80])]; + tensor var_350_end_0 = const()[name = tensor("op_350_end_0"), val = tensor([2, 4096, 1, 120])]; + tensor var_350_end_mask_0 = const()[name = tensor("op_350_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_350_cast_fp16 = slice_by_index(begin = var_350_begin_0, end = var_350_end_0, end_mask = var_350_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_350_cast_fp16")]; + tensor var_354_begin_0 = const()[name = tensor("op_354_begin_0"), val = tensor([0, 0, 0, 120])]; + tensor var_354_end_0 = const()[name = tensor("op_354_end_0"), val = tensor([2, 4096, 1, 160])]; + tensor var_354_end_mask_0 = const()[name = tensor("op_354_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_354_cast_fp16 = slice_by_index(begin = var_354_begin_0, end = var_354_end_0, end_mask = var_354_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_354_cast_fp16")]; + tensor var_358_begin_0 = const()[name = tensor("op_358_begin_0"), val = tensor([0, 0, 0, 160])]; + tensor var_358_end_0 = const()[name = tensor("op_358_end_0"), val = tensor([2, 4096, 1, 200])]; + tensor var_358_end_mask_0 = const()[name = tensor("op_358_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_358_cast_fp16 = slice_by_index(begin = var_358_begin_0, end = var_358_end_0, end_mask = var_358_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_358_cast_fp16")]; + tensor var_362_begin_0 = const()[name = tensor("op_362_begin_0"), val = tensor([0, 0, 0, 200])]; + tensor var_362_end_0 = const()[name = tensor("op_362_end_0"), val = tensor([2, 4096, 1, 240])]; + tensor var_362_end_mask_0 = const()[name = tensor("op_362_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_362_cast_fp16 = slice_by_index(begin = var_362_begin_0, end = var_362_end_0, end_mask = var_362_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_362_cast_fp16")]; + tensor var_366_begin_0 = const()[name = tensor("op_366_begin_0"), val = tensor([0, 0, 0, 240])]; + tensor var_366_end_0 = const()[name = tensor("op_366_end_0"), val = tensor([2, 4096, 1, 280])]; + tensor var_366_end_mask_0 = const()[name = tensor("op_366_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_366_cast_fp16 = slice_by_index(begin = var_366_begin_0, end = var_366_end_0, end_mask = var_366_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_366_cast_fp16")]; + tensor var_370_begin_0 = const()[name = tensor("op_370_begin_0"), val = tensor([0, 0, 0, 280])]; + tensor var_370_end_0 = const()[name = tensor("op_370_end_0"), val = tensor([2, 4096, 1, 1])]; + tensor var_370_end_mask_0 = const()[name = tensor("op_370_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_370_cast_fp16 = slice_by_index(begin = var_370_begin_0, end = var_370_end_0, end_mask = var_370_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_370_cast_fp16")]; + tensor var_372_begin_0 = const()[name = tensor("op_372_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_372_end_0 = const()[name = tensor("op_372_end_0"), val = tensor([2, 40, 1, 4096])]; + tensor var_372_end_mask_0 = const()[name = tensor("op_372_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_372_cast_fp16 = slice_by_index(begin = var_372_begin_0, end = var_372_end_0, end_mask = var_372_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_372_cast_fp16")]; + tensor var_376_begin_0 = const()[name = tensor("op_376_begin_0"), val = tensor([0, 40, 0, 0])]; + tensor var_376_end_0 = const()[name = tensor("op_376_end_0"), val = tensor([2, 80, 1, 4096])]; + tensor var_376_end_mask_0 = const()[name = tensor("op_376_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_376_cast_fp16 = slice_by_index(begin = var_376_begin_0, end = var_376_end_0, end_mask = var_376_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_376_cast_fp16")]; + tensor var_380_begin_0 = const()[name = tensor("op_380_begin_0"), val = tensor([0, 80, 0, 0])]; + tensor var_380_end_0 = const()[name = tensor("op_380_end_0"), val = tensor([2, 120, 1, 4096])]; + tensor var_380_end_mask_0 = const()[name = tensor("op_380_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_380_cast_fp16 = slice_by_index(begin = var_380_begin_0, end = var_380_end_0, end_mask = var_380_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_380_cast_fp16")]; + tensor var_384_begin_0 = const()[name = tensor("op_384_begin_0"), val = tensor([0, 120, 0, 0])]; + tensor var_384_end_0 = const()[name = tensor("op_384_end_0"), val = tensor([2, 160, 1, 4096])]; + tensor var_384_end_mask_0 = const()[name = tensor("op_384_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_384_cast_fp16 = slice_by_index(begin = var_384_begin_0, end = var_384_end_0, end_mask = var_384_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_384_cast_fp16")]; + tensor var_388_begin_0 = const()[name = tensor("op_388_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_388_end_0 = const()[name = tensor("op_388_end_0"), val = tensor([2, 200, 1, 4096])]; + tensor var_388_end_mask_0 = const()[name = tensor("op_388_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_388_cast_fp16 = slice_by_index(begin = var_388_begin_0, end = var_388_end_0, end_mask = var_388_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_388_cast_fp16")]; + tensor var_392_begin_0 = const()[name = tensor("op_392_begin_0"), val = tensor([0, 200, 0, 0])]; + tensor var_392_end_0 = const()[name = tensor("op_392_end_0"), val = tensor([2, 240, 1, 4096])]; + tensor var_392_end_mask_0 = const()[name = tensor("op_392_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_392_cast_fp16 = slice_by_index(begin = var_392_begin_0, end = var_392_end_0, end_mask = var_392_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_392_cast_fp16")]; + tensor var_396_begin_0 = const()[name = tensor("op_396_begin_0"), val = tensor([0, 240, 0, 0])]; + tensor var_396_end_0 = const()[name = tensor("op_396_end_0"), val = tensor([2, 280, 1, 4096])]; + tensor var_396_end_mask_0 = const()[name = tensor("op_396_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_396_cast_fp16 = slice_by_index(begin = var_396_begin_0, end = var_396_end_0, end_mask = var_396_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_396_cast_fp16")]; + tensor var_400_begin_0 = const()[name = tensor("op_400_begin_0"), val = tensor([0, 280, 0, 0])]; + tensor var_400_end_0 = const()[name = tensor("op_400_end_0"), val = tensor([2, 1, 1, 4096])]; + tensor var_400_end_mask_0 = const()[name = tensor("op_400_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_400_cast_fp16 = slice_by_index(begin = var_400_begin_0, end = var_400_end_0, end_mask = var_400_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_400_cast_fp16")]; + tensor var_404_equation_0 = const()[name = tensor("op_404_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_404_cast_fp16 = einsum(equation = var_404_equation_0, values = (var_342_cast_fp16, var_274_cast_fp16))[name = tensor("op_404_cast_fp16")]; + tensor var_405_to_fp16 = const()[name = tensor("op_405_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_1_cast_fp16 = mul(x = var_404_cast_fp16, y = var_405_to_fp16)[name = tensor("aw_chunk_1_cast_fp16")]; + tensor var_408_equation_0 = const()[name = tensor("op_408_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_408_cast_fp16 = einsum(equation = var_408_equation_0, values = (var_342_cast_fp16, var_275_cast_fp16))[name = tensor("op_408_cast_fp16")]; + tensor var_409_to_fp16 = const()[name = tensor("op_409_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_3_cast_fp16 = mul(x = var_408_cast_fp16, y = var_409_to_fp16)[name = tensor("aw_chunk_3_cast_fp16")]; + tensor var_412_equation_0 = const()[name = tensor("op_412_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_412_cast_fp16 = einsum(equation = var_412_equation_0, values = (var_342_cast_fp16, var_276_cast_fp16))[name = tensor("op_412_cast_fp16")]; + tensor var_413_to_fp16 = const()[name = tensor("op_413_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_5_cast_fp16 = mul(x = var_412_cast_fp16, y = var_413_to_fp16)[name = tensor("aw_chunk_5_cast_fp16")]; + tensor var_416_equation_0 = const()[name = tensor("op_416_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_416_cast_fp16 = einsum(equation = var_416_equation_0, values = (var_342_cast_fp16, var_277_cast_fp16))[name = tensor("op_416_cast_fp16")]; + tensor var_417_to_fp16 = const()[name = tensor("op_417_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_7_cast_fp16 = mul(x = var_416_cast_fp16, y = var_417_to_fp16)[name = tensor("aw_chunk_7_cast_fp16")]; + tensor var_420_equation_0 = const()[name = tensor("op_420_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_420_cast_fp16 = einsum(equation = var_420_equation_0, values = (var_342_cast_fp16, var_278_cast_fp16))[name = tensor("op_420_cast_fp16")]; + tensor var_421_to_fp16 = const()[name = tensor("op_421_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_9_cast_fp16 = mul(x = var_420_cast_fp16, y = var_421_to_fp16)[name = tensor("aw_chunk_9_cast_fp16")]; + tensor var_424_equation_0 = const()[name = tensor("op_424_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_424_cast_fp16 = einsum(equation = var_424_equation_0, values = (var_342_cast_fp16, var_279_cast_fp16))[name = tensor("op_424_cast_fp16")]; + tensor var_425_to_fp16 = const()[name = tensor("op_425_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_11_cast_fp16 = mul(x = var_424_cast_fp16, y = var_425_to_fp16)[name = tensor("aw_chunk_11_cast_fp16")]; + tensor var_428_equation_0 = const()[name = tensor("op_428_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_428_cast_fp16 = einsum(equation = var_428_equation_0, values = (var_342_cast_fp16, var_280_cast_fp16))[name = tensor("op_428_cast_fp16")]; + tensor var_429_to_fp16 = const()[name = tensor("op_429_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_13_cast_fp16 = mul(x = var_428_cast_fp16, y = var_429_to_fp16)[name = tensor("aw_chunk_13_cast_fp16")]; + tensor var_432_equation_0 = const()[name = tensor("op_432_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_432_cast_fp16 = einsum(equation = var_432_equation_0, values = (var_342_cast_fp16, var_281_cast_fp16))[name = tensor("op_432_cast_fp16")]; + tensor var_433_to_fp16 = const()[name = tensor("op_433_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_15_cast_fp16 = mul(x = var_432_cast_fp16, y = var_433_to_fp16)[name = tensor("aw_chunk_15_cast_fp16")]; + tensor var_436_equation_0 = const()[name = tensor("op_436_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_436_cast_fp16 = einsum(equation = var_436_equation_0, values = (var_346_cast_fp16, var_282_cast_fp16))[name = tensor("op_436_cast_fp16")]; + tensor var_437_to_fp16 = const()[name = tensor("op_437_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_17_cast_fp16 = mul(x = var_436_cast_fp16, y = var_437_to_fp16)[name = tensor("aw_chunk_17_cast_fp16")]; + tensor var_440_equation_0 = const()[name = tensor("op_440_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_440_cast_fp16 = einsum(equation = var_440_equation_0, values = (var_346_cast_fp16, var_283_cast_fp16))[name = tensor("op_440_cast_fp16")]; + tensor var_441_to_fp16 = const()[name = tensor("op_441_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_19_cast_fp16 = mul(x = var_440_cast_fp16, y = var_441_to_fp16)[name = tensor("aw_chunk_19_cast_fp16")]; + tensor var_444_equation_0 = const()[name = tensor("op_444_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_444_cast_fp16 = einsum(equation = var_444_equation_0, values = (var_346_cast_fp16, var_284_cast_fp16))[name = tensor("op_444_cast_fp16")]; + tensor var_445_to_fp16 = const()[name = tensor("op_445_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_21_cast_fp16 = mul(x = var_444_cast_fp16, y = var_445_to_fp16)[name = tensor("aw_chunk_21_cast_fp16")]; + tensor var_448_equation_0 = const()[name = tensor("op_448_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_448_cast_fp16 = einsum(equation = var_448_equation_0, values = (var_346_cast_fp16, var_285_cast_fp16))[name = tensor("op_448_cast_fp16")]; + tensor var_449_to_fp16 = const()[name = tensor("op_449_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_23_cast_fp16 = mul(x = var_448_cast_fp16, y = var_449_to_fp16)[name = tensor("aw_chunk_23_cast_fp16")]; + tensor var_452_equation_0 = const()[name = tensor("op_452_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_452_cast_fp16 = einsum(equation = var_452_equation_0, values = (var_346_cast_fp16, var_286_cast_fp16))[name = tensor("op_452_cast_fp16")]; + tensor var_453_to_fp16 = const()[name = tensor("op_453_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_25_cast_fp16 = mul(x = var_452_cast_fp16, y = var_453_to_fp16)[name = tensor("aw_chunk_25_cast_fp16")]; + tensor var_456_equation_0 = const()[name = tensor("op_456_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_456_cast_fp16 = einsum(equation = var_456_equation_0, values = (var_346_cast_fp16, var_287_cast_fp16))[name = tensor("op_456_cast_fp16")]; + tensor var_457_to_fp16 = const()[name = tensor("op_457_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_27_cast_fp16 = mul(x = var_456_cast_fp16, y = var_457_to_fp16)[name = tensor("aw_chunk_27_cast_fp16")]; + tensor var_460_equation_0 = const()[name = tensor("op_460_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_460_cast_fp16 = einsum(equation = var_460_equation_0, values = (var_346_cast_fp16, var_288_cast_fp16))[name = tensor("op_460_cast_fp16")]; + tensor var_461_to_fp16 = const()[name = tensor("op_461_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_29_cast_fp16 = mul(x = var_460_cast_fp16, y = var_461_to_fp16)[name = tensor("aw_chunk_29_cast_fp16")]; + tensor var_464_equation_0 = const()[name = tensor("op_464_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_464_cast_fp16 = einsum(equation = var_464_equation_0, values = (var_346_cast_fp16, var_289_cast_fp16))[name = tensor("op_464_cast_fp16")]; + tensor var_465_to_fp16 = const()[name = tensor("op_465_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_31_cast_fp16 = mul(x = var_464_cast_fp16, y = var_465_to_fp16)[name = tensor("aw_chunk_31_cast_fp16")]; + tensor var_468_equation_0 = const()[name = tensor("op_468_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_468_cast_fp16 = einsum(equation = var_468_equation_0, values = (var_350_cast_fp16, var_290_cast_fp16))[name = tensor("op_468_cast_fp16")]; + tensor var_469_to_fp16 = const()[name = tensor("op_469_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_33_cast_fp16 = mul(x = var_468_cast_fp16, y = var_469_to_fp16)[name = tensor("aw_chunk_33_cast_fp16")]; + tensor var_472_equation_0 = const()[name = tensor("op_472_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_472_cast_fp16 = einsum(equation = var_472_equation_0, values = (var_350_cast_fp16, var_291_cast_fp16))[name = tensor("op_472_cast_fp16")]; + tensor var_473_to_fp16 = const()[name = tensor("op_473_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_35_cast_fp16 = mul(x = var_472_cast_fp16, y = var_473_to_fp16)[name = tensor("aw_chunk_35_cast_fp16")]; + tensor var_476_equation_0 = const()[name = tensor("op_476_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_476_cast_fp16 = einsum(equation = var_476_equation_0, values = (var_350_cast_fp16, var_292_cast_fp16))[name = tensor("op_476_cast_fp16")]; + tensor var_477_to_fp16 = const()[name = tensor("op_477_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_37_cast_fp16 = mul(x = var_476_cast_fp16, y = var_477_to_fp16)[name = tensor("aw_chunk_37_cast_fp16")]; + tensor var_480_equation_0 = const()[name = tensor("op_480_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_480_cast_fp16 = einsum(equation = var_480_equation_0, values = (var_350_cast_fp16, var_293_cast_fp16))[name = tensor("op_480_cast_fp16")]; + tensor var_481_to_fp16 = const()[name = tensor("op_481_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_39_cast_fp16 = mul(x = var_480_cast_fp16, y = var_481_to_fp16)[name = tensor("aw_chunk_39_cast_fp16")]; + tensor var_484_equation_0 = const()[name = tensor("op_484_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_484_cast_fp16 = einsum(equation = var_484_equation_0, values = (var_350_cast_fp16, var_294_cast_fp16))[name = tensor("op_484_cast_fp16")]; + tensor var_485_to_fp16 = const()[name = tensor("op_485_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_41_cast_fp16 = mul(x = var_484_cast_fp16, y = var_485_to_fp16)[name = tensor("aw_chunk_41_cast_fp16")]; + tensor var_488_equation_0 = const()[name = tensor("op_488_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_488_cast_fp16 = einsum(equation = var_488_equation_0, values = (var_350_cast_fp16, var_295_cast_fp16))[name = tensor("op_488_cast_fp16")]; + tensor var_489_to_fp16 = const()[name = tensor("op_489_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_43_cast_fp16 = mul(x = var_488_cast_fp16, y = var_489_to_fp16)[name = tensor("aw_chunk_43_cast_fp16")]; + tensor var_492_equation_0 = const()[name = tensor("op_492_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_492_cast_fp16 = einsum(equation = var_492_equation_0, values = (var_350_cast_fp16, var_296_cast_fp16))[name = tensor("op_492_cast_fp16")]; + tensor var_493_to_fp16 = const()[name = tensor("op_493_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_45_cast_fp16 = mul(x = var_492_cast_fp16, y = var_493_to_fp16)[name = tensor("aw_chunk_45_cast_fp16")]; + tensor var_496_equation_0 = const()[name = tensor("op_496_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_496_cast_fp16 = einsum(equation = var_496_equation_0, values = (var_350_cast_fp16, var_297_cast_fp16))[name = tensor("op_496_cast_fp16")]; + tensor var_497_to_fp16 = const()[name = tensor("op_497_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_47_cast_fp16 = mul(x = var_496_cast_fp16, y = var_497_to_fp16)[name = tensor("aw_chunk_47_cast_fp16")]; + tensor var_500_equation_0 = const()[name = tensor("op_500_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_500_cast_fp16 = einsum(equation = var_500_equation_0, values = (var_354_cast_fp16, var_298_cast_fp16))[name = tensor("op_500_cast_fp16")]; + tensor var_501_to_fp16 = const()[name = tensor("op_501_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_49_cast_fp16 = mul(x = var_500_cast_fp16, y = var_501_to_fp16)[name = tensor("aw_chunk_49_cast_fp16")]; + tensor var_504_equation_0 = const()[name = tensor("op_504_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_504_cast_fp16 = einsum(equation = var_504_equation_0, values = (var_354_cast_fp16, var_299_cast_fp16))[name = tensor("op_504_cast_fp16")]; + tensor var_505_to_fp16 = const()[name = tensor("op_505_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_51_cast_fp16 = mul(x = var_504_cast_fp16, y = var_505_to_fp16)[name = tensor("aw_chunk_51_cast_fp16")]; + tensor var_508_equation_0 = const()[name = tensor("op_508_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_508_cast_fp16 = einsum(equation = var_508_equation_0, values = (var_354_cast_fp16, var_300_cast_fp16))[name = tensor("op_508_cast_fp16")]; + tensor var_509_to_fp16 = const()[name = tensor("op_509_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_53_cast_fp16 = mul(x = var_508_cast_fp16, y = var_509_to_fp16)[name = tensor("aw_chunk_53_cast_fp16")]; + tensor var_512_equation_0 = const()[name = tensor("op_512_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_512_cast_fp16 = einsum(equation = var_512_equation_0, values = (var_354_cast_fp16, var_301_cast_fp16))[name = tensor("op_512_cast_fp16")]; + tensor var_513_to_fp16 = const()[name = tensor("op_513_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_55_cast_fp16 = mul(x = var_512_cast_fp16, y = var_513_to_fp16)[name = tensor("aw_chunk_55_cast_fp16")]; + tensor var_516_equation_0 = const()[name = tensor("op_516_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_516_cast_fp16 = einsum(equation = var_516_equation_0, values = (var_354_cast_fp16, var_302_cast_fp16))[name = tensor("op_516_cast_fp16")]; + tensor var_517_to_fp16 = const()[name = tensor("op_517_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_57_cast_fp16 = mul(x = var_516_cast_fp16, y = var_517_to_fp16)[name = tensor("aw_chunk_57_cast_fp16")]; + tensor var_520_equation_0 = const()[name = tensor("op_520_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_520_cast_fp16 = einsum(equation = var_520_equation_0, values = (var_354_cast_fp16, var_303_cast_fp16))[name = tensor("op_520_cast_fp16")]; + tensor var_521_to_fp16 = const()[name = tensor("op_521_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_59_cast_fp16 = mul(x = var_520_cast_fp16, y = var_521_to_fp16)[name = tensor("aw_chunk_59_cast_fp16")]; + tensor var_524_equation_0 = const()[name = tensor("op_524_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_524_cast_fp16 = einsum(equation = var_524_equation_0, values = (var_354_cast_fp16, var_304_cast_fp16))[name = tensor("op_524_cast_fp16")]; + tensor var_525_to_fp16 = const()[name = tensor("op_525_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_61_cast_fp16 = mul(x = var_524_cast_fp16, y = var_525_to_fp16)[name = tensor("aw_chunk_61_cast_fp16")]; + tensor var_528_equation_0 = const()[name = tensor("op_528_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_528_cast_fp16 = einsum(equation = var_528_equation_0, values = (var_354_cast_fp16, var_305_cast_fp16))[name = tensor("op_528_cast_fp16")]; + tensor var_529_to_fp16 = const()[name = tensor("op_529_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_63_cast_fp16 = mul(x = var_528_cast_fp16, y = var_529_to_fp16)[name = tensor("aw_chunk_63_cast_fp16")]; + tensor var_532_equation_0 = const()[name = tensor("op_532_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_532_cast_fp16 = einsum(equation = var_532_equation_0, values = (var_358_cast_fp16, var_306_cast_fp16))[name = tensor("op_532_cast_fp16")]; + tensor var_533_to_fp16 = const()[name = tensor("op_533_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_65_cast_fp16 = mul(x = var_532_cast_fp16, y = var_533_to_fp16)[name = tensor("aw_chunk_65_cast_fp16")]; + tensor var_536_equation_0 = const()[name = tensor("op_536_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_536_cast_fp16 = einsum(equation = var_536_equation_0, values = (var_358_cast_fp16, var_307_cast_fp16))[name = tensor("op_536_cast_fp16")]; + tensor var_537_to_fp16 = const()[name = tensor("op_537_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_67_cast_fp16 = mul(x = var_536_cast_fp16, y = var_537_to_fp16)[name = tensor("aw_chunk_67_cast_fp16")]; + tensor var_540_equation_0 = const()[name = tensor("op_540_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_540_cast_fp16 = einsum(equation = var_540_equation_0, values = (var_358_cast_fp16, var_308_cast_fp16))[name = tensor("op_540_cast_fp16")]; + tensor var_541_to_fp16 = const()[name = tensor("op_541_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_69_cast_fp16 = mul(x = var_540_cast_fp16, y = var_541_to_fp16)[name = tensor("aw_chunk_69_cast_fp16")]; + tensor var_544_equation_0 = const()[name = tensor("op_544_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_544_cast_fp16 = einsum(equation = var_544_equation_0, values = (var_358_cast_fp16, var_309_cast_fp16))[name = tensor("op_544_cast_fp16")]; + tensor var_545_to_fp16 = const()[name = tensor("op_545_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_71_cast_fp16 = mul(x = var_544_cast_fp16, y = var_545_to_fp16)[name = tensor("aw_chunk_71_cast_fp16")]; + tensor var_548_equation_0 = const()[name = tensor("op_548_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_548_cast_fp16 = einsum(equation = var_548_equation_0, values = (var_358_cast_fp16, var_310_cast_fp16))[name = tensor("op_548_cast_fp16")]; + tensor var_549_to_fp16 = const()[name = tensor("op_549_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_73_cast_fp16 = mul(x = var_548_cast_fp16, y = var_549_to_fp16)[name = tensor("aw_chunk_73_cast_fp16")]; + tensor var_552_equation_0 = const()[name = tensor("op_552_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_552_cast_fp16 = einsum(equation = var_552_equation_0, values = (var_358_cast_fp16, var_311_cast_fp16))[name = tensor("op_552_cast_fp16")]; + tensor var_553_to_fp16 = const()[name = tensor("op_553_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_75_cast_fp16 = mul(x = var_552_cast_fp16, y = var_553_to_fp16)[name = tensor("aw_chunk_75_cast_fp16")]; + tensor var_556_equation_0 = const()[name = tensor("op_556_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_556_cast_fp16 = einsum(equation = var_556_equation_0, values = (var_358_cast_fp16, var_312_cast_fp16))[name = tensor("op_556_cast_fp16")]; + tensor var_557_to_fp16 = const()[name = tensor("op_557_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_77_cast_fp16 = mul(x = var_556_cast_fp16, y = var_557_to_fp16)[name = tensor("aw_chunk_77_cast_fp16")]; + tensor var_560_equation_0 = const()[name = tensor("op_560_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_560_cast_fp16 = einsum(equation = var_560_equation_0, values = (var_358_cast_fp16, var_313_cast_fp16))[name = tensor("op_560_cast_fp16")]; + tensor var_561_to_fp16 = const()[name = tensor("op_561_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_79_cast_fp16 = mul(x = var_560_cast_fp16, y = var_561_to_fp16)[name = tensor("aw_chunk_79_cast_fp16")]; + tensor var_564_equation_0 = const()[name = tensor("op_564_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_564_cast_fp16 = einsum(equation = var_564_equation_0, values = (var_362_cast_fp16, var_314_cast_fp16))[name = tensor("op_564_cast_fp16")]; + tensor var_565_to_fp16 = const()[name = tensor("op_565_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_81_cast_fp16 = mul(x = var_564_cast_fp16, y = var_565_to_fp16)[name = tensor("aw_chunk_81_cast_fp16")]; + tensor var_568_equation_0 = const()[name = tensor("op_568_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_568_cast_fp16 = einsum(equation = var_568_equation_0, values = (var_362_cast_fp16, var_315_cast_fp16))[name = tensor("op_568_cast_fp16")]; + tensor var_569_to_fp16 = const()[name = tensor("op_569_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_83_cast_fp16 = mul(x = var_568_cast_fp16, y = var_569_to_fp16)[name = tensor("aw_chunk_83_cast_fp16")]; + tensor var_572_equation_0 = const()[name = tensor("op_572_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_572_cast_fp16 = einsum(equation = var_572_equation_0, values = (var_362_cast_fp16, var_316_cast_fp16))[name = tensor("op_572_cast_fp16")]; + tensor var_573_to_fp16 = const()[name = tensor("op_573_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_85_cast_fp16 = mul(x = var_572_cast_fp16, y = var_573_to_fp16)[name = tensor("aw_chunk_85_cast_fp16")]; + tensor var_576_equation_0 = const()[name = tensor("op_576_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_576_cast_fp16 = einsum(equation = var_576_equation_0, values = (var_362_cast_fp16, var_317_cast_fp16))[name = tensor("op_576_cast_fp16")]; + tensor var_577_to_fp16 = const()[name = tensor("op_577_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_87_cast_fp16 = mul(x = var_576_cast_fp16, y = var_577_to_fp16)[name = tensor("aw_chunk_87_cast_fp16")]; + tensor var_580_equation_0 = const()[name = tensor("op_580_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_580_cast_fp16 = einsum(equation = var_580_equation_0, values = (var_362_cast_fp16, var_318_cast_fp16))[name = tensor("op_580_cast_fp16")]; + tensor var_581_to_fp16 = const()[name = tensor("op_581_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_89_cast_fp16 = mul(x = var_580_cast_fp16, y = var_581_to_fp16)[name = tensor("aw_chunk_89_cast_fp16")]; + tensor var_584_equation_0 = const()[name = tensor("op_584_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_584_cast_fp16 = einsum(equation = var_584_equation_0, values = (var_362_cast_fp16, var_319_cast_fp16))[name = tensor("op_584_cast_fp16")]; + tensor var_585_to_fp16 = const()[name = tensor("op_585_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_91_cast_fp16 = mul(x = var_584_cast_fp16, y = var_585_to_fp16)[name = tensor("aw_chunk_91_cast_fp16")]; + tensor var_588_equation_0 = const()[name = tensor("op_588_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_588_cast_fp16 = einsum(equation = var_588_equation_0, values = (var_362_cast_fp16, var_320_cast_fp16))[name = tensor("op_588_cast_fp16")]; + tensor var_589_to_fp16 = const()[name = tensor("op_589_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_93_cast_fp16 = mul(x = var_588_cast_fp16, y = var_589_to_fp16)[name = tensor("aw_chunk_93_cast_fp16")]; + tensor var_592_equation_0 = const()[name = tensor("op_592_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_592_cast_fp16 = einsum(equation = var_592_equation_0, values = (var_362_cast_fp16, var_321_cast_fp16))[name = tensor("op_592_cast_fp16")]; + tensor var_593_to_fp16 = const()[name = tensor("op_593_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_95_cast_fp16 = mul(x = var_592_cast_fp16, y = var_593_to_fp16)[name = tensor("aw_chunk_95_cast_fp16")]; + tensor var_596_equation_0 = const()[name = tensor("op_596_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_596_cast_fp16 = einsum(equation = var_596_equation_0, values = (var_366_cast_fp16, var_322_cast_fp16))[name = tensor("op_596_cast_fp16")]; + tensor var_597_to_fp16 = const()[name = tensor("op_597_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_97_cast_fp16 = mul(x = var_596_cast_fp16, y = var_597_to_fp16)[name = tensor("aw_chunk_97_cast_fp16")]; + tensor var_600_equation_0 = const()[name = tensor("op_600_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_600_cast_fp16 = einsum(equation = var_600_equation_0, values = (var_366_cast_fp16, var_323_cast_fp16))[name = tensor("op_600_cast_fp16")]; + tensor var_601_to_fp16 = const()[name = tensor("op_601_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_99_cast_fp16 = mul(x = var_600_cast_fp16, y = var_601_to_fp16)[name = tensor("aw_chunk_99_cast_fp16")]; + tensor var_604_equation_0 = const()[name = tensor("op_604_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_604_cast_fp16 = einsum(equation = var_604_equation_0, values = (var_366_cast_fp16, var_324_cast_fp16))[name = tensor("op_604_cast_fp16")]; + tensor var_605_to_fp16 = const()[name = tensor("op_605_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_101_cast_fp16 = mul(x = var_604_cast_fp16, y = var_605_to_fp16)[name = tensor("aw_chunk_101_cast_fp16")]; + tensor var_608_equation_0 = const()[name = tensor("op_608_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_608_cast_fp16 = einsum(equation = var_608_equation_0, values = (var_366_cast_fp16, var_325_cast_fp16))[name = tensor("op_608_cast_fp16")]; + tensor var_609_to_fp16 = const()[name = tensor("op_609_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_103_cast_fp16 = mul(x = var_608_cast_fp16, y = var_609_to_fp16)[name = tensor("aw_chunk_103_cast_fp16")]; + tensor var_612_equation_0 = const()[name = tensor("op_612_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_612_cast_fp16 = einsum(equation = var_612_equation_0, values = (var_366_cast_fp16, var_326_cast_fp16))[name = tensor("op_612_cast_fp16")]; + tensor var_613_to_fp16 = const()[name = tensor("op_613_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_105_cast_fp16 = mul(x = var_612_cast_fp16, y = var_613_to_fp16)[name = tensor("aw_chunk_105_cast_fp16")]; + tensor var_616_equation_0 = const()[name = tensor("op_616_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_616_cast_fp16 = einsum(equation = var_616_equation_0, values = (var_366_cast_fp16, var_327_cast_fp16))[name = tensor("op_616_cast_fp16")]; + tensor var_617_to_fp16 = const()[name = tensor("op_617_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_107_cast_fp16 = mul(x = var_616_cast_fp16, y = var_617_to_fp16)[name = tensor("aw_chunk_107_cast_fp16")]; + tensor var_620_equation_0 = const()[name = tensor("op_620_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_620_cast_fp16 = einsum(equation = var_620_equation_0, values = (var_366_cast_fp16, var_328_cast_fp16))[name = tensor("op_620_cast_fp16")]; + tensor var_621_to_fp16 = const()[name = tensor("op_621_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_109_cast_fp16 = mul(x = var_620_cast_fp16, y = var_621_to_fp16)[name = tensor("aw_chunk_109_cast_fp16")]; + tensor var_624_equation_0 = const()[name = tensor("op_624_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_624_cast_fp16 = einsum(equation = var_624_equation_0, values = (var_366_cast_fp16, var_329_cast_fp16))[name = tensor("op_624_cast_fp16")]; + tensor var_625_to_fp16 = const()[name = tensor("op_625_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_111_cast_fp16 = mul(x = var_624_cast_fp16, y = var_625_to_fp16)[name = tensor("aw_chunk_111_cast_fp16")]; + tensor var_628_equation_0 = const()[name = tensor("op_628_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_628_cast_fp16 = einsum(equation = var_628_equation_0, values = (var_370_cast_fp16, var_330_cast_fp16))[name = tensor("op_628_cast_fp16")]; + tensor var_629_to_fp16 = const()[name = tensor("op_629_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_113_cast_fp16 = mul(x = var_628_cast_fp16, y = var_629_to_fp16)[name = tensor("aw_chunk_113_cast_fp16")]; + tensor var_632_equation_0 = const()[name = tensor("op_632_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_632_cast_fp16 = einsum(equation = var_632_equation_0, values = (var_370_cast_fp16, var_331_cast_fp16))[name = tensor("op_632_cast_fp16")]; + tensor var_633_to_fp16 = const()[name = tensor("op_633_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_115_cast_fp16 = mul(x = var_632_cast_fp16, y = var_633_to_fp16)[name = tensor("aw_chunk_115_cast_fp16")]; + tensor var_636_equation_0 = const()[name = tensor("op_636_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_636_cast_fp16 = einsum(equation = var_636_equation_0, values = (var_370_cast_fp16, var_332_cast_fp16))[name = tensor("op_636_cast_fp16")]; + tensor var_637_to_fp16 = const()[name = tensor("op_637_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_117_cast_fp16 = mul(x = var_636_cast_fp16, y = var_637_to_fp16)[name = tensor("aw_chunk_117_cast_fp16")]; + tensor var_640_equation_0 = const()[name = tensor("op_640_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_640_cast_fp16 = einsum(equation = var_640_equation_0, values = (var_370_cast_fp16, var_333_cast_fp16))[name = tensor("op_640_cast_fp16")]; + tensor var_641_to_fp16 = const()[name = tensor("op_641_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_119_cast_fp16 = mul(x = var_640_cast_fp16, y = var_641_to_fp16)[name = tensor("aw_chunk_119_cast_fp16")]; + tensor var_644_equation_0 = const()[name = tensor("op_644_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_644_cast_fp16 = einsum(equation = var_644_equation_0, values = (var_370_cast_fp16, var_334_cast_fp16))[name = tensor("op_644_cast_fp16")]; + tensor var_645_to_fp16 = const()[name = tensor("op_645_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_121_cast_fp16 = mul(x = var_644_cast_fp16, y = var_645_to_fp16)[name = tensor("aw_chunk_121_cast_fp16")]; + tensor var_648_equation_0 = const()[name = tensor("op_648_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_648_cast_fp16 = einsum(equation = var_648_equation_0, values = (var_370_cast_fp16, var_335_cast_fp16))[name = tensor("op_648_cast_fp16")]; + tensor var_649_to_fp16 = const()[name = tensor("op_649_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_123_cast_fp16 = mul(x = var_648_cast_fp16, y = var_649_to_fp16)[name = tensor("aw_chunk_123_cast_fp16")]; + tensor var_652_equation_0 = const()[name = tensor("op_652_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_652_cast_fp16 = einsum(equation = var_652_equation_0, values = (var_370_cast_fp16, var_336_cast_fp16))[name = tensor("op_652_cast_fp16")]; + tensor var_653_to_fp16 = const()[name = tensor("op_653_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_125_cast_fp16 = mul(x = var_652_cast_fp16, y = var_653_to_fp16)[name = tensor("aw_chunk_125_cast_fp16")]; + tensor var_656_equation_0 = const()[name = tensor("op_656_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_656_cast_fp16 = einsum(equation = var_656_equation_0, values = (var_370_cast_fp16, var_337_cast_fp16))[name = tensor("op_656_cast_fp16")]; + tensor var_657_to_fp16 = const()[name = tensor("op_657_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_127_cast_fp16 = mul(x = var_656_cast_fp16, y = var_657_to_fp16)[name = tensor("aw_chunk_127_cast_fp16")]; + tensor var_659_cast_fp16 = softmax(axis = var_123, x = aw_chunk_1_cast_fp16)[name = tensor("op_659_cast_fp16")]; + tensor var_660_cast_fp16 = softmax(axis = var_123, x = aw_chunk_3_cast_fp16)[name = tensor("op_660_cast_fp16")]; + tensor var_661_cast_fp16 = softmax(axis = var_123, x = aw_chunk_5_cast_fp16)[name = tensor("op_661_cast_fp16")]; + tensor var_662_cast_fp16 = softmax(axis = var_123, x = aw_chunk_7_cast_fp16)[name = tensor("op_662_cast_fp16")]; + tensor var_663_cast_fp16 = softmax(axis = var_123, x = aw_chunk_9_cast_fp16)[name = tensor("op_663_cast_fp16")]; + tensor var_664_cast_fp16 = softmax(axis = var_123, x = aw_chunk_11_cast_fp16)[name = tensor("op_664_cast_fp16")]; + tensor var_665_cast_fp16 = softmax(axis = var_123, x = aw_chunk_13_cast_fp16)[name = tensor("op_665_cast_fp16")]; + tensor var_666_cast_fp16 = softmax(axis = var_123, x = aw_chunk_15_cast_fp16)[name = tensor("op_666_cast_fp16")]; + tensor var_667_cast_fp16 = softmax(axis = var_123, x = aw_chunk_17_cast_fp16)[name = tensor("op_667_cast_fp16")]; + tensor var_668_cast_fp16 = softmax(axis = var_123, x = aw_chunk_19_cast_fp16)[name = tensor("op_668_cast_fp16")]; + tensor var_669_cast_fp16 = softmax(axis = var_123, x = aw_chunk_21_cast_fp16)[name = tensor("op_669_cast_fp16")]; + tensor var_670_cast_fp16 = softmax(axis = var_123, x = aw_chunk_23_cast_fp16)[name = tensor("op_670_cast_fp16")]; + tensor var_671_cast_fp16 = softmax(axis = var_123, x = aw_chunk_25_cast_fp16)[name = tensor("op_671_cast_fp16")]; + tensor var_672_cast_fp16 = softmax(axis = var_123, x = aw_chunk_27_cast_fp16)[name = tensor("op_672_cast_fp16")]; + tensor var_673_cast_fp16 = softmax(axis = var_123, x = aw_chunk_29_cast_fp16)[name = tensor("op_673_cast_fp16")]; + tensor var_674_cast_fp16 = softmax(axis = var_123, x = aw_chunk_31_cast_fp16)[name = tensor("op_674_cast_fp16")]; + tensor var_675_cast_fp16 = softmax(axis = var_123, x = aw_chunk_33_cast_fp16)[name = tensor("op_675_cast_fp16")]; + tensor var_676_cast_fp16 = softmax(axis = var_123, x = aw_chunk_35_cast_fp16)[name = tensor("op_676_cast_fp16")]; + tensor var_677_cast_fp16 = softmax(axis = var_123, x = aw_chunk_37_cast_fp16)[name = tensor("op_677_cast_fp16")]; + tensor var_678_cast_fp16 = softmax(axis = var_123, x = aw_chunk_39_cast_fp16)[name = tensor("op_678_cast_fp16")]; + tensor var_679_cast_fp16 = softmax(axis = var_123, x = aw_chunk_41_cast_fp16)[name = tensor("op_679_cast_fp16")]; + tensor var_680_cast_fp16 = softmax(axis = var_123, x = aw_chunk_43_cast_fp16)[name = tensor("op_680_cast_fp16")]; + tensor var_681_cast_fp16 = softmax(axis = var_123, x = aw_chunk_45_cast_fp16)[name = tensor("op_681_cast_fp16")]; + tensor var_682_cast_fp16 = softmax(axis = var_123, x = aw_chunk_47_cast_fp16)[name = tensor("op_682_cast_fp16")]; + tensor var_683_cast_fp16 = softmax(axis = var_123, x = aw_chunk_49_cast_fp16)[name = tensor("op_683_cast_fp16")]; + tensor var_684_cast_fp16 = softmax(axis = var_123, x = aw_chunk_51_cast_fp16)[name = tensor("op_684_cast_fp16")]; + tensor var_685_cast_fp16 = softmax(axis = var_123, x = aw_chunk_53_cast_fp16)[name = tensor("op_685_cast_fp16")]; + tensor var_686_cast_fp16 = softmax(axis = var_123, x = aw_chunk_55_cast_fp16)[name = tensor("op_686_cast_fp16")]; + tensor var_687_cast_fp16 = softmax(axis = var_123, x = aw_chunk_57_cast_fp16)[name = tensor("op_687_cast_fp16")]; + tensor var_688_cast_fp16 = softmax(axis = var_123, x = aw_chunk_59_cast_fp16)[name = tensor("op_688_cast_fp16")]; + tensor var_689_cast_fp16 = softmax(axis = var_123, x = aw_chunk_61_cast_fp16)[name = tensor("op_689_cast_fp16")]; + tensor var_690_cast_fp16 = softmax(axis = var_123, x = aw_chunk_63_cast_fp16)[name = tensor("op_690_cast_fp16")]; + tensor var_691_cast_fp16 = softmax(axis = var_123, x = aw_chunk_65_cast_fp16)[name = tensor("op_691_cast_fp16")]; + tensor var_692_cast_fp16 = softmax(axis = var_123, x = aw_chunk_67_cast_fp16)[name = tensor("op_692_cast_fp16")]; + tensor var_693_cast_fp16 = softmax(axis = var_123, x = aw_chunk_69_cast_fp16)[name = tensor("op_693_cast_fp16")]; + tensor var_694_cast_fp16 = softmax(axis = var_123, x = aw_chunk_71_cast_fp16)[name = tensor("op_694_cast_fp16")]; + tensor var_695_cast_fp16 = softmax(axis = var_123, x = aw_chunk_73_cast_fp16)[name = tensor("op_695_cast_fp16")]; + tensor var_696_cast_fp16 = softmax(axis = var_123, x = aw_chunk_75_cast_fp16)[name = tensor("op_696_cast_fp16")]; + tensor var_697_cast_fp16 = softmax(axis = var_123, x = aw_chunk_77_cast_fp16)[name = tensor("op_697_cast_fp16")]; + tensor var_698_cast_fp16 = softmax(axis = var_123, x = aw_chunk_79_cast_fp16)[name = tensor("op_698_cast_fp16")]; + tensor var_699_cast_fp16 = softmax(axis = var_123, x = aw_chunk_81_cast_fp16)[name = tensor("op_699_cast_fp16")]; + tensor var_700_cast_fp16 = softmax(axis = var_123, x = aw_chunk_83_cast_fp16)[name = tensor("op_700_cast_fp16")]; + tensor var_701_cast_fp16 = softmax(axis = var_123, x = aw_chunk_85_cast_fp16)[name = tensor("op_701_cast_fp16")]; + tensor var_702_cast_fp16 = softmax(axis = var_123, x = aw_chunk_87_cast_fp16)[name = tensor("op_702_cast_fp16")]; + tensor var_703_cast_fp16 = softmax(axis = var_123, x = aw_chunk_89_cast_fp16)[name = tensor("op_703_cast_fp16")]; + tensor var_704_cast_fp16 = softmax(axis = var_123, x = aw_chunk_91_cast_fp16)[name = tensor("op_704_cast_fp16")]; + tensor var_705_cast_fp16 = softmax(axis = var_123, x = aw_chunk_93_cast_fp16)[name = tensor("op_705_cast_fp16")]; + tensor var_706_cast_fp16 = softmax(axis = var_123, x = aw_chunk_95_cast_fp16)[name = tensor("op_706_cast_fp16")]; + tensor var_707_cast_fp16 = softmax(axis = var_123, x = aw_chunk_97_cast_fp16)[name = tensor("op_707_cast_fp16")]; + tensor var_708_cast_fp16 = softmax(axis = var_123, x = aw_chunk_99_cast_fp16)[name = tensor("op_708_cast_fp16")]; + tensor var_709_cast_fp16 = softmax(axis = var_123, x = aw_chunk_101_cast_fp16)[name = tensor("op_709_cast_fp16")]; + tensor var_710_cast_fp16 = softmax(axis = var_123, x = aw_chunk_103_cast_fp16)[name = tensor("op_710_cast_fp16")]; + tensor var_711_cast_fp16 = softmax(axis = var_123, x = aw_chunk_105_cast_fp16)[name = tensor("op_711_cast_fp16")]; + tensor var_712_cast_fp16 = softmax(axis = var_123, x = aw_chunk_107_cast_fp16)[name = tensor("op_712_cast_fp16")]; + tensor var_713_cast_fp16 = softmax(axis = var_123, x = aw_chunk_109_cast_fp16)[name = tensor("op_713_cast_fp16")]; + tensor var_714_cast_fp16 = softmax(axis = var_123, x = aw_chunk_111_cast_fp16)[name = tensor("op_714_cast_fp16")]; + tensor var_715_cast_fp16 = softmax(axis = var_123, x = aw_chunk_113_cast_fp16)[name = tensor("op_715_cast_fp16")]; + tensor var_716_cast_fp16 = softmax(axis = var_123, x = aw_chunk_115_cast_fp16)[name = tensor("op_716_cast_fp16")]; + tensor var_717_cast_fp16 = softmax(axis = var_123, x = aw_chunk_117_cast_fp16)[name = tensor("op_717_cast_fp16")]; + tensor var_718_cast_fp16 = softmax(axis = var_123, x = aw_chunk_119_cast_fp16)[name = tensor("op_718_cast_fp16")]; + tensor var_719_cast_fp16 = softmax(axis = var_123, x = aw_chunk_121_cast_fp16)[name = tensor("op_719_cast_fp16")]; + tensor var_720_cast_fp16 = softmax(axis = var_123, x = aw_chunk_123_cast_fp16)[name = tensor("op_720_cast_fp16")]; + tensor var_721_cast_fp16 = softmax(axis = var_123, x = aw_chunk_125_cast_fp16)[name = tensor("op_721_cast_fp16")]; + tensor var_722_cast_fp16 = softmax(axis = var_123, x = aw_chunk_127_cast_fp16)[name = tensor("op_722_cast_fp16")]; + tensor var_724_equation_0 = const()[name = tensor("op_724_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_724_cast_fp16 = einsum(equation = var_724_equation_0, values = (var_372_cast_fp16, var_659_cast_fp16))[name = tensor("op_724_cast_fp16")]; + tensor var_726_equation_0 = const()[name = tensor("op_726_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_726_cast_fp16 = einsum(equation = var_726_equation_0, values = (var_372_cast_fp16, var_660_cast_fp16))[name = tensor("op_726_cast_fp16")]; + tensor var_728_equation_0 = const()[name = tensor("op_728_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_728_cast_fp16 = einsum(equation = var_728_equation_0, values = (var_372_cast_fp16, var_661_cast_fp16))[name = tensor("op_728_cast_fp16")]; + tensor var_730_equation_0 = const()[name = tensor("op_730_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_730_cast_fp16 = einsum(equation = var_730_equation_0, values = (var_372_cast_fp16, var_662_cast_fp16))[name = tensor("op_730_cast_fp16")]; + tensor var_732_equation_0 = const()[name = tensor("op_732_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_732_cast_fp16 = einsum(equation = var_732_equation_0, values = (var_372_cast_fp16, var_663_cast_fp16))[name = tensor("op_732_cast_fp16")]; + tensor var_734_equation_0 = const()[name = tensor("op_734_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_734_cast_fp16 = einsum(equation = var_734_equation_0, values = (var_372_cast_fp16, var_664_cast_fp16))[name = tensor("op_734_cast_fp16")]; + tensor var_736_equation_0 = const()[name = tensor("op_736_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_736_cast_fp16 = einsum(equation = var_736_equation_0, values = (var_372_cast_fp16, var_665_cast_fp16))[name = tensor("op_736_cast_fp16")]; + tensor var_738_equation_0 = const()[name = tensor("op_738_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_738_cast_fp16 = einsum(equation = var_738_equation_0, values = (var_372_cast_fp16, var_666_cast_fp16))[name = tensor("op_738_cast_fp16")]; + tensor var_740_equation_0 = const()[name = tensor("op_740_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_740_cast_fp16 = einsum(equation = var_740_equation_0, values = (var_376_cast_fp16, var_667_cast_fp16))[name = tensor("op_740_cast_fp16")]; + tensor var_742_equation_0 = const()[name = tensor("op_742_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_742_cast_fp16 = einsum(equation = var_742_equation_0, values = (var_376_cast_fp16, var_668_cast_fp16))[name = tensor("op_742_cast_fp16")]; + tensor var_744_equation_0 = const()[name = tensor("op_744_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_744_cast_fp16 = einsum(equation = var_744_equation_0, values = (var_376_cast_fp16, var_669_cast_fp16))[name = tensor("op_744_cast_fp16")]; + tensor var_746_equation_0 = const()[name = tensor("op_746_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_746_cast_fp16 = einsum(equation = var_746_equation_0, values = (var_376_cast_fp16, var_670_cast_fp16))[name = tensor("op_746_cast_fp16")]; + tensor var_748_equation_0 = const()[name = tensor("op_748_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_748_cast_fp16 = einsum(equation = var_748_equation_0, values = (var_376_cast_fp16, var_671_cast_fp16))[name = tensor("op_748_cast_fp16")]; + tensor var_750_equation_0 = const()[name = tensor("op_750_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_750_cast_fp16 = einsum(equation = var_750_equation_0, values = (var_376_cast_fp16, var_672_cast_fp16))[name = tensor("op_750_cast_fp16")]; + tensor var_752_equation_0 = const()[name = tensor("op_752_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_752_cast_fp16 = einsum(equation = var_752_equation_0, values = (var_376_cast_fp16, var_673_cast_fp16))[name = tensor("op_752_cast_fp16")]; + tensor var_754_equation_0 = const()[name = tensor("op_754_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_754_cast_fp16 = einsum(equation = var_754_equation_0, values = (var_376_cast_fp16, var_674_cast_fp16))[name = tensor("op_754_cast_fp16")]; + tensor var_756_equation_0 = const()[name = tensor("op_756_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_756_cast_fp16 = einsum(equation = var_756_equation_0, values = (var_380_cast_fp16, var_675_cast_fp16))[name = tensor("op_756_cast_fp16")]; + tensor var_758_equation_0 = const()[name = tensor("op_758_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_758_cast_fp16 = einsum(equation = var_758_equation_0, values = (var_380_cast_fp16, var_676_cast_fp16))[name = tensor("op_758_cast_fp16")]; + tensor var_760_equation_0 = const()[name = tensor("op_760_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_760_cast_fp16 = einsum(equation = var_760_equation_0, values = (var_380_cast_fp16, var_677_cast_fp16))[name = tensor("op_760_cast_fp16")]; + tensor var_762_equation_0 = const()[name = tensor("op_762_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_762_cast_fp16 = einsum(equation = var_762_equation_0, values = (var_380_cast_fp16, var_678_cast_fp16))[name = tensor("op_762_cast_fp16")]; + tensor var_764_equation_0 = const()[name = tensor("op_764_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_764_cast_fp16 = einsum(equation = var_764_equation_0, values = (var_380_cast_fp16, var_679_cast_fp16))[name = tensor("op_764_cast_fp16")]; + tensor var_766_equation_0 = const()[name = tensor("op_766_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_766_cast_fp16 = einsum(equation = var_766_equation_0, values = (var_380_cast_fp16, var_680_cast_fp16))[name = tensor("op_766_cast_fp16")]; + tensor var_768_equation_0 = const()[name = tensor("op_768_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_768_cast_fp16 = einsum(equation = var_768_equation_0, values = (var_380_cast_fp16, var_681_cast_fp16))[name = tensor("op_768_cast_fp16")]; + tensor var_770_equation_0 = const()[name = tensor("op_770_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_770_cast_fp16 = einsum(equation = var_770_equation_0, values = (var_380_cast_fp16, var_682_cast_fp16))[name = tensor("op_770_cast_fp16")]; + tensor var_772_equation_0 = const()[name = tensor("op_772_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_772_cast_fp16 = einsum(equation = var_772_equation_0, values = (var_384_cast_fp16, var_683_cast_fp16))[name = tensor("op_772_cast_fp16")]; + tensor var_774_equation_0 = const()[name = tensor("op_774_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_774_cast_fp16 = einsum(equation = var_774_equation_0, values = (var_384_cast_fp16, var_684_cast_fp16))[name = tensor("op_774_cast_fp16")]; + tensor var_776_equation_0 = const()[name = tensor("op_776_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_776_cast_fp16 = einsum(equation = var_776_equation_0, values = (var_384_cast_fp16, var_685_cast_fp16))[name = tensor("op_776_cast_fp16")]; + tensor var_778_equation_0 = const()[name = tensor("op_778_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_778_cast_fp16 = einsum(equation = var_778_equation_0, values = (var_384_cast_fp16, var_686_cast_fp16))[name = tensor("op_778_cast_fp16")]; + tensor var_780_equation_0 = const()[name = tensor("op_780_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_780_cast_fp16 = einsum(equation = var_780_equation_0, values = (var_384_cast_fp16, var_687_cast_fp16))[name = tensor("op_780_cast_fp16")]; + tensor var_782_equation_0 = const()[name = tensor("op_782_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_782_cast_fp16 = einsum(equation = var_782_equation_0, values = (var_384_cast_fp16, var_688_cast_fp16))[name = tensor("op_782_cast_fp16")]; + tensor var_784_equation_0 = const()[name = tensor("op_784_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_784_cast_fp16 = einsum(equation = var_784_equation_0, values = (var_384_cast_fp16, var_689_cast_fp16))[name = tensor("op_784_cast_fp16")]; + tensor var_786_equation_0 = const()[name = tensor("op_786_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_786_cast_fp16 = einsum(equation = var_786_equation_0, values = (var_384_cast_fp16, var_690_cast_fp16))[name = tensor("op_786_cast_fp16")]; + tensor var_788_equation_0 = const()[name = tensor("op_788_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_788_cast_fp16 = einsum(equation = var_788_equation_0, values = (var_388_cast_fp16, var_691_cast_fp16))[name = tensor("op_788_cast_fp16")]; + tensor var_790_equation_0 = const()[name = tensor("op_790_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_790_cast_fp16 = einsum(equation = var_790_equation_0, values = (var_388_cast_fp16, var_692_cast_fp16))[name = tensor("op_790_cast_fp16")]; + tensor var_792_equation_0 = const()[name = tensor("op_792_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_792_cast_fp16 = einsum(equation = var_792_equation_0, values = (var_388_cast_fp16, var_693_cast_fp16))[name = tensor("op_792_cast_fp16")]; + tensor var_794_equation_0 = const()[name = tensor("op_794_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_794_cast_fp16 = einsum(equation = var_794_equation_0, values = (var_388_cast_fp16, var_694_cast_fp16))[name = tensor("op_794_cast_fp16")]; + tensor var_796_equation_0 = const()[name = tensor("op_796_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_796_cast_fp16 = einsum(equation = var_796_equation_0, values = (var_388_cast_fp16, var_695_cast_fp16))[name = tensor("op_796_cast_fp16")]; + tensor var_798_equation_0 = const()[name = tensor("op_798_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_798_cast_fp16 = einsum(equation = var_798_equation_0, values = (var_388_cast_fp16, var_696_cast_fp16))[name = tensor("op_798_cast_fp16")]; + tensor var_800_equation_0 = const()[name = tensor("op_800_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_800_cast_fp16 = einsum(equation = var_800_equation_0, values = (var_388_cast_fp16, var_697_cast_fp16))[name = tensor("op_800_cast_fp16")]; + tensor var_802_equation_0 = const()[name = tensor("op_802_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_802_cast_fp16 = einsum(equation = var_802_equation_0, values = (var_388_cast_fp16, var_698_cast_fp16))[name = tensor("op_802_cast_fp16")]; + tensor var_804_equation_0 = const()[name = tensor("op_804_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_804_cast_fp16 = einsum(equation = var_804_equation_0, values = (var_392_cast_fp16, var_699_cast_fp16))[name = tensor("op_804_cast_fp16")]; + tensor var_806_equation_0 = const()[name = tensor("op_806_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_806_cast_fp16 = einsum(equation = var_806_equation_0, values = (var_392_cast_fp16, var_700_cast_fp16))[name = tensor("op_806_cast_fp16")]; + tensor var_808_equation_0 = const()[name = tensor("op_808_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_808_cast_fp16 = einsum(equation = var_808_equation_0, values = (var_392_cast_fp16, var_701_cast_fp16))[name = tensor("op_808_cast_fp16")]; + tensor var_810_equation_0 = const()[name = tensor("op_810_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_810_cast_fp16 = einsum(equation = var_810_equation_0, values = (var_392_cast_fp16, var_702_cast_fp16))[name = tensor("op_810_cast_fp16")]; + tensor var_812_equation_0 = const()[name = tensor("op_812_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_812_cast_fp16 = einsum(equation = var_812_equation_0, values = (var_392_cast_fp16, var_703_cast_fp16))[name = tensor("op_812_cast_fp16")]; + tensor var_814_equation_0 = const()[name = tensor("op_814_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_814_cast_fp16 = einsum(equation = var_814_equation_0, values = (var_392_cast_fp16, var_704_cast_fp16))[name = tensor("op_814_cast_fp16")]; + tensor var_816_equation_0 = const()[name = tensor("op_816_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_816_cast_fp16 = einsum(equation = var_816_equation_0, values = (var_392_cast_fp16, var_705_cast_fp16))[name = tensor("op_816_cast_fp16")]; + tensor var_818_equation_0 = const()[name = tensor("op_818_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_818_cast_fp16 = einsum(equation = var_818_equation_0, values = (var_392_cast_fp16, var_706_cast_fp16))[name = tensor("op_818_cast_fp16")]; + tensor var_820_equation_0 = const()[name = tensor("op_820_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_820_cast_fp16 = einsum(equation = var_820_equation_0, values = (var_396_cast_fp16, var_707_cast_fp16))[name = tensor("op_820_cast_fp16")]; + tensor var_822_equation_0 = const()[name = tensor("op_822_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_822_cast_fp16 = einsum(equation = var_822_equation_0, values = (var_396_cast_fp16, var_708_cast_fp16))[name = tensor("op_822_cast_fp16")]; + tensor var_824_equation_0 = const()[name = tensor("op_824_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_824_cast_fp16 = einsum(equation = var_824_equation_0, values = (var_396_cast_fp16, var_709_cast_fp16))[name = tensor("op_824_cast_fp16")]; + tensor var_826_equation_0 = const()[name = tensor("op_826_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_826_cast_fp16 = einsum(equation = var_826_equation_0, values = (var_396_cast_fp16, var_710_cast_fp16))[name = tensor("op_826_cast_fp16")]; + tensor var_828_equation_0 = const()[name = tensor("op_828_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_828_cast_fp16 = einsum(equation = var_828_equation_0, values = (var_396_cast_fp16, var_711_cast_fp16))[name = tensor("op_828_cast_fp16")]; + tensor var_830_equation_0 = const()[name = tensor("op_830_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_830_cast_fp16 = einsum(equation = var_830_equation_0, values = (var_396_cast_fp16, var_712_cast_fp16))[name = tensor("op_830_cast_fp16")]; + tensor var_832_equation_0 = const()[name = tensor("op_832_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_832_cast_fp16 = einsum(equation = var_832_equation_0, values = (var_396_cast_fp16, var_713_cast_fp16))[name = tensor("op_832_cast_fp16")]; + tensor var_834_equation_0 = const()[name = tensor("op_834_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_834_cast_fp16 = einsum(equation = var_834_equation_0, values = (var_396_cast_fp16, var_714_cast_fp16))[name = tensor("op_834_cast_fp16")]; + tensor var_836_equation_0 = const()[name = tensor("op_836_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_836_cast_fp16 = einsum(equation = var_836_equation_0, values = (var_400_cast_fp16, var_715_cast_fp16))[name = tensor("op_836_cast_fp16")]; + tensor var_838_equation_0 = const()[name = tensor("op_838_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_838_cast_fp16 = einsum(equation = var_838_equation_0, values = (var_400_cast_fp16, var_716_cast_fp16))[name = tensor("op_838_cast_fp16")]; + tensor var_840_equation_0 = const()[name = tensor("op_840_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_840_cast_fp16 = einsum(equation = var_840_equation_0, values = (var_400_cast_fp16, var_717_cast_fp16))[name = tensor("op_840_cast_fp16")]; + tensor var_842_equation_0 = const()[name = tensor("op_842_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_842_cast_fp16 = einsum(equation = var_842_equation_0, values = (var_400_cast_fp16, var_718_cast_fp16))[name = tensor("op_842_cast_fp16")]; + tensor var_844_equation_0 = const()[name = tensor("op_844_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_844_cast_fp16 = einsum(equation = var_844_equation_0, values = (var_400_cast_fp16, var_719_cast_fp16))[name = tensor("op_844_cast_fp16")]; + tensor var_846_equation_0 = const()[name = tensor("op_846_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_846_cast_fp16 = einsum(equation = var_846_equation_0, values = (var_400_cast_fp16, var_720_cast_fp16))[name = tensor("op_846_cast_fp16")]; + tensor var_848_equation_0 = const()[name = tensor("op_848_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_848_cast_fp16 = einsum(equation = var_848_equation_0, values = (var_400_cast_fp16, var_721_cast_fp16))[name = tensor("op_848_cast_fp16")]; + tensor var_850_equation_0 = const()[name = tensor("op_850_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_850_cast_fp16 = einsum(equation = var_850_equation_0, values = (var_400_cast_fp16, var_722_cast_fp16))[name = tensor("op_850_cast_fp16")]; + tensor var_852_interleave_0 = const()[name = tensor("op_852_interleave_0"), val = tensor(false)]; + tensor var_852_cast_fp16 = concat(axis = var_95, interleave = var_852_interleave_0, values = (var_724_cast_fp16, var_726_cast_fp16, var_728_cast_fp16, var_730_cast_fp16, var_732_cast_fp16, var_734_cast_fp16, var_736_cast_fp16, var_738_cast_fp16))[name = tensor("op_852_cast_fp16")]; + tensor var_854_interleave_0 = const()[name = tensor("op_854_interleave_0"), val = tensor(false)]; + tensor var_854_cast_fp16 = concat(axis = var_95, interleave = var_854_interleave_0, values = (var_740_cast_fp16, var_742_cast_fp16, var_744_cast_fp16, var_746_cast_fp16, var_748_cast_fp16, var_750_cast_fp16, var_752_cast_fp16, var_754_cast_fp16))[name = tensor("op_854_cast_fp16")]; + tensor var_856_interleave_0 = const()[name = tensor("op_856_interleave_0"), val = tensor(false)]; + tensor var_856_cast_fp16 = concat(axis = var_95, interleave = var_856_interleave_0, values = (var_756_cast_fp16, var_758_cast_fp16, var_760_cast_fp16, var_762_cast_fp16, var_764_cast_fp16, var_766_cast_fp16, var_768_cast_fp16, var_770_cast_fp16))[name = tensor("op_856_cast_fp16")]; + tensor var_858_interleave_0 = const()[name = tensor("op_858_interleave_0"), val = tensor(false)]; + tensor var_858_cast_fp16 = concat(axis = var_95, interleave = var_858_interleave_0, values = (var_772_cast_fp16, var_774_cast_fp16, var_776_cast_fp16, var_778_cast_fp16, var_780_cast_fp16, var_782_cast_fp16, var_784_cast_fp16, var_786_cast_fp16))[name = tensor("op_858_cast_fp16")]; + tensor var_860_interleave_0 = const()[name = tensor("op_860_interleave_0"), val = tensor(false)]; + tensor var_860_cast_fp16 = concat(axis = var_95, interleave = var_860_interleave_0, values = (var_788_cast_fp16, var_790_cast_fp16, var_792_cast_fp16, var_794_cast_fp16, var_796_cast_fp16, var_798_cast_fp16, var_800_cast_fp16, var_802_cast_fp16))[name = tensor("op_860_cast_fp16")]; + tensor var_862_interleave_0 = const()[name = tensor("op_862_interleave_0"), val = tensor(false)]; + tensor var_862_cast_fp16 = concat(axis = var_95, interleave = var_862_interleave_0, values = (var_804_cast_fp16, var_806_cast_fp16, var_808_cast_fp16, var_810_cast_fp16, var_812_cast_fp16, var_814_cast_fp16, var_816_cast_fp16, var_818_cast_fp16))[name = tensor("op_862_cast_fp16")]; + tensor var_864_interleave_0 = const()[name = tensor("op_864_interleave_0"), val = tensor(false)]; + tensor var_864_cast_fp16 = concat(axis = var_95, interleave = var_864_interleave_0, values = (var_820_cast_fp16, var_822_cast_fp16, var_824_cast_fp16, var_826_cast_fp16, var_828_cast_fp16, var_830_cast_fp16, var_832_cast_fp16, var_834_cast_fp16))[name = tensor("op_864_cast_fp16")]; + tensor var_866_interleave_0 = const()[name = tensor("op_866_interleave_0"), val = tensor(false)]; + tensor var_866_cast_fp16 = concat(axis = var_95, interleave = var_866_interleave_0, values = (var_836_cast_fp16, var_838_cast_fp16, var_840_cast_fp16, var_842_cast_fp16, var_844_cast_fp16, var_846_cast_fp16, var_848_cast_fp16, var_850_cast_fp16))[name = tensor("op_866_cast_fp16")]; + tensor input_25_interleave_0 = const()[name = tensor("input_25_interleave_0"), val = tensor(false)]; + tensor input_25_cast_fp16 = concat(axis = var_123, interleave = input_25_interleave_0, values = (var_852_cast_fp16, var_854_cast_fp16, var_856_cast_fp16, var_858_cast_fp16, var_860_cast_fp16, var_862_cast_fp16, var_864_cast_fp16, var_866_cast_fp16))[name = tensor("input_25_cast_fp16")]; + tensor var_876_pad_type_0 = const()[name = tensor("op_876_pad_type_0"), val = tensor("valid")]; + tensor var_876_strides_0 = const()[name = tensor("op_876_strides_0"), val = tensor([1, 1])]; + tensor var_876_pad_0 = const()[name = tensor("op_876_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_876_dilations_0 = const()[name = tensor("op_876_dilations_0"), val = tensor([1, 1])]; + tensor var_876_groups_0 = const()[name = tensor("op_876_groups_0"), val = tensor(1)]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9460736)))]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9665600)))]; + tensor var_876_cast_fp16 = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_876_dilations_0, groups = var_876_groups_0, pad = var_876_pad_0, pad_type = var_876_pad_type_0, strides = var_876_strides_0, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_25_cast_fp16)[name = tensor("op_876_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = var_876_cast_fp16, y = inputs_1_cast_fp16)[name = tensor("inputs_3_cast_fp16")]; + tensor hidden_states_11_axes_0 = const()[name = tensor("hidden_states_11_axes_0"), val = tensor([1])]; + tensor hidden_states_11_gamma_0_to_fp16 = const()[name = tensor("hidden_states_11_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9666304)))]; + tensor hidden_states_11_beta_0_to_fp16 = const()[name = tensor("hidden_states_11_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9667008)))]; + tensor var_886_to_fp16 = const()[name = tensor("op_886_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_11_cast_fp16 = layer_norm(axes = hidden_states_11_axes_0, beta = hidden_states_11_beta_0_to_fp16, epsilon = var_886_to_fp16, gamma = hidden_states_11_gamma_0_to_fp16, x = inputs_3_cast_fp16)[name = tensor("hidden_states_11_cast_fp16")]; + tensor q_3_pad_type_0 = const()[name = tensor("q_3_pad_type_0"), val = tensor("valid")]; + tensor q_3_strides_0 = const()[name = tensor("q_3_strides_0"), val = tensor([1, 1])]; + tensor q_3_pad_0 = const()[name = tensor("q_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_3_dilations_0 = const()[name = tensor("q_3_dilations_0"), val = tensor([1, 1])]; + tensor q_3_groups_0 = const()[name = tensor("q_3_groups_0"), val = tensor(1)]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9667712)))]; + tensor q_3_cast_fp16 = conv(dilations = q_3_dilations_0, groups = q_3_groups_0, pad = q_3_pad_0, pad_type = q_3_pad_type_0, strides = q_3_strides_0, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_11_cast_fp16)[name = tensor("q_3_cast_fp16")]; + tensor k_5_pad_type_0 = const()[name = tensor("k_5_pad_type_0"), val = tensor("valid")]; + tensor k_5_strides_0 = const()[name = tensor("k_5_strides_0"), val = tensor([1, 1])]; + tensor k_5_pad_0 = const()[name = tensor("k_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_5_dilations_0 = const()[name = tensor("k_5_dilations_0"), val = tensor([1, 1])]; + tensor k_5_groups_0 = const()[name = tensor("k_5_groups_0"), val = tensor(1)]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9872576)))]; + tensor k_5_cast_fp16 = conv(dilations = k_5_dilations_0, groups = k_5_groups_0, pad = k_5_pad_0, pad_type = k_5_pad_type_0, strides = k_5_strides_0, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_5_cast_fp16")]; + tensor v_3_pad_type_0 = const()[name = tensor("v_3_pad_type_0"), val = tensor("valid")]; + tensor v_3_strides_0 = const()[name = tensor("v_3_strides_0"), val = tensor([1, 1])]; + tensor v_3_pad_0 = const()[name = tensor("v_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_3_dilations_0 = const()[name = tensor("v_3_dilations_0"), val = tensor([1, 1])]; + tensor v_3_groups_0 = const()[name = tensor("v_3_groups_0"), val = tensor(1)]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10364160)))]; + tensor v_3_cast_fp16 = conv(dilations = v_3_dilations_0, groups = v_3_groups_0, pad = v_3_pad_0, pad_type = v_3_pad_type_0, strides = v_3_strides_0, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_3_cast_fp16")]; + tensor var_919_begin_0 = const()[name = tensor("op_919_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_919_end_0 = const()[name = tensor("op_919_end_0"), val = tensor([2, 40, 1, 4096])]; + tensor var_919_end_mask_0 = const()[name = tensor("op_919_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_919_cast_fp16 = slice_by_index(begin = var_919_begin_0, end = var_919_end_0, end_mask = var_919_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_919_cast_fp16")]; + tensor var_923_begin_0 = const()[name = tensor("op_923_begin_0"), val = tensor([0, 40, 0, 0])]; + tensor var_923_end_0 = const()[name = tensor("op_923_end_0"), val = tensor([2, 80, 1, 4096])]; + tensor var_923_end_mask_0 = const()[name = tensor("op_923_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_923_cast_fp16 = slice_by_index(begin = var_923_begin_0, end = var_923_end_0, end_mask = var_923_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_923_cast_fp16")]; + tensor var_927_begin_0 = const()[name = tensor("op_927_begin_0"), val = tensor([0, 80, 0, 0])]; + tensor var_927_end_0 = const()[name = tensor("op_927_end_0"), val = tensor([2, 120, 1, 4096])]; + tensor var_927_end_mask_0 = const()[name = tensor("op_927_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_927_cast_fp16 = slice_by_index(begin = var_927_begin_0, end = var_927_end_0, end_mask = var_927_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_927_cast_fp16")]; + tensor var_931_begin_0 = const()[name = tensor("op_931_begin_0"), val = tensor([0, 120, 0, 0])]; + tensor var_931_end_0 = const()[name = tensor("op_931_end_0"), val = tensor([2, 160, 1, 4096])]; + tensor var_931_end_mask_0 = const()[name = tensor("op_931_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_931_cast_fp16 = slice_by_index(begin = var_931_begin_0, end = var_931_end_0, end_mask = var_931_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_931_cast_fp16")]; + tensor var_935_begin_0 = const()[name = tensor("op_935_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_935_end_0 = const()[name = tensor("op_935_end_0"), val = tensor([2, 200, 1, 4096])]; + tensor var_935_end_mask_0 = const()[name = tensor("op_935_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_935_cast_fp16 = slice_by_index(begin = var_935_begin_0, end = var_935_end_0, end_mask = var_935_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_935_cast_fp16")]; + tensor var_939_begin_0 = const()[name = tensor("op_939_begin_0"), val = tensor([0, 200, 0, 0])]; + tensor var_939_end_0 = const()[name = tensor("op_939_end_0"), val = tensor([2, 240, 1, 4096])]; + tensor var_939_end_mask_0 = const()[name = tensor("op_939_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_939_cast_fp16 = slice_by_index(begin = var_939_begin_0, end = var_939_end_0, end_mask = var_939_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_939_cast_fp16")]; + tensor var_943_begin_0 = const()[name = tensor("op_943_begin_0"), val = tensor([0, 240, 0, 0])]; + tensor var_943_end_0 = const()[name = tensor("op_943_end_0"), val = tensor([2, 280, 1, 4096])]; + tensor var_943_end_mask_0 = const()[name = tensor("op_943_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_943_cast_fp16 = slice_by_index(begin = var_943_begin_0, end = var_943_end_0, end_mask = var_943_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_943_cast_fp16")]; + tensor var_947_begin_0 = const()[name = tensor("op_947_begin_0"), val = tensor([0, 280, 0, 0])]; + tensor var_947_end_0 = const()[name = tensor("op_947_end_0"), val = tensor([2, 1, 1, 4096])]; + tensor var_947_end_mask_0 = const()[name = tensor("op_947_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_947_cast_fp16 = slice_by_index(begin = var_947_begin_0, end = var_947_end_0, end_mask = var_947_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_947_cast_fp16")]; + tensor var_950_begin_0 = const()[name = tensor("op_950_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_950_end_0 = const()[name = tensor("op_950_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_950_end_mask_0 = const()[name = tensor("op_950_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_950_cast_fp16 = slice_by_index(begin = var_950_begin_0, end = var_950_end_0, end_mask = var_950_end_mask_0, x = var_919_cast_fp16)[name = tensor("op_950_cast_fp16")]; + tensor var_951_begin_0 = const()[name = tensor("op_951_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_951_end_0 = const()[name = tensor("op_951_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_951_end_mask_0 = const()[name = tensor("op_951_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_951_cast_fp16 = slice_by_index(begin = var_951_begin_0, end = var_951_end_0, end_mask = var_951_end_mask_0, x = var_919_cast_fp16)[name = tensor("op_951_cast_fp16")]; + tensor var_952_begin_0 = const()[name = tensor("op_952_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_952_end_0 = const()[name = tensor("op_952_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_952_end_mask_0 = const()[name = tensor("op_952_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_952_cast_fp16 = slice_by_index(begin = var_952_begin_0, end = var_952_end_0, end_mask = var_952_end_mask_0, x = var_919_cast_fp16)[name = tensor("op_952_cast_fp16")]; + tensor var_953_begin_0 = const()[name = tensor("op_953_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_953_end_0 = const()[name = tensor("op_953_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_953_end_mask_0 = const()[name = tensor("op_953_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_953_cast_fp16 = slice_by_index(begin = var_953_begin_0, end = var_953_end_0, end_mask = var_953_end_mask_0, x = var_919_cast_fp16)[name = tensor("op_953_cast_fp16")]; + tensor var_954_begin_0 = const()[name = tensor("op_954_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_954_end_0 = const()[name = tensor("op_954_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_954_end_mask_0 = const()[name = tensor("op_954_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_954_cast_fp16 = slice_by_index(begin = var_954_begin_0, end = var_954_end_0, end_mask = var_954_end_mask_0, x = var_919_cast_fp16)[name = tensor("op_954_cast_fp16")]; + tensor var_955_begin_0 = const()[name = tensor("op_955_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_955_end_0 = const()[name = tensor("op_955_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_955_end_mask_0 = const()[name = tensor("op_955_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_955_cast_fp16 = slice_by_index(begin = var_955_begin_0, end = var_955_end_0, end_mask = var_955_end_mask_0, x = var_919_cast_fp16)[name = tensor("op_955_cast_fp16")]; + tensor var_956_begin_0 = const()[name = tensor("op_956_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_956_end_0 = const()[name = tensor("op_956_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_956_end_mask_0 = const()[name = tensor("op_956_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_956_cast_fp16 = slice_by_index(begin = var_956_begin_0, end = var_956_end_0, end_mask = var_956_end_mask_0, x = var_919_cast_fp16)[name = tensor("op_956_cast_fp16")]; + tensor var_957_begin_0 = const()[name = tensor("op_957_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_957_end_0 = const()[name = tensor("op_957_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_957_end_mask_0 = const()[name = tensor("op_957_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_957_cast_fp16 = slice_by_index(begin = var_957_begin_0, end = var_957_end_0, end_mask = var_957_end_mask_0, x = var_919_cast_fp16)[name = tensor("op_957_cast_fp16")]; + tensor var_958_begin_0 = const()[name = tensor("op_958_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_958_end_0 = const()[name = tensor("op_958_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_958_end_mask_0 = const()[name = tensor("op_958_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_958_cast_fp16 = slice_by_index(begin = var_958_begin_0, end = var_958_end_0, end_mask = var_958_end_mask_0, x = var_923_cast_fp16)[name = tensor("op_958_cast_fp16")]; + tensor var_959_begin_0 = const()[name = tensor("op_959_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_959_end_0 = const()[name = tensor("op_959_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_959_end_mask_0 = const()[name = tensor("op_959_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_959_cast_fp16 = slice_by_index(begin = var_959_begin_0, end = var_959_end_0, end_mask = var_959_end_mask_0, x = var_923_cast_fp16)[name = tensor("op_959_cast_fp16")]; + tensor var_960_begin_0 = const()[name = tensor("op_960_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_960_end_0 = const()[name = tensor("op_960_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_960_end_mask_0 = const()[name = tensor("op_960_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_960_cast_fp16 = slice_by_index(begin = var_960_begin_0, end = var_960_end_0, end_mask = var_960_end_mask_0, x = var_923_cast_fp16)[name = tensor("op_960_cast_fp16")]; + tensor var_961_begin_0 = const()[name = tensor("op_961_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_961_end_0 = const()[name = tensor("op_961_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_961_end_mask_0 = const()[name = tensor("op_961_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_961_cast_fp16 = slice_by_index(begin = var_961_begin_0, end = var_961_end_0, end_mask = var_961_end_mask_0, x = var_923_cast_fp16)[name = tensor("op_961_cast_fp16")]; + tensor var_962_begin_0 = const()[name = tensor("op_962_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_962_end_0 = const()[name = tensor("op_962_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_962_end_mask_0 = const()[name = tensor("op_962_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_962_cast_fp16 = slice_by_index(begin = var_962_begin_0, end = var_962_end_0, end_mask = var_962_end_mask_0, x = var_923_cast_fp16)[name = tensor("op_962_cast_fp16")]; + tensor var_963_begin_0 = const()[name = tensor("op_963_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_963_end_0 = const()[name = tensor("op_963_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_963_end_mask_0 = const()[name = tensor("op_963_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_963_cast_fp16 = slice_by_index(begin = var_963_begin_0, end = var_963_end_0, end_mask = var_963_end_mask_0, x = var_923_cast_fp16)[name = tensor("op_963_cast_fp16")]; + tensor var_964_begin_0 = const()[name = tensor("op_964_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_964_end_0 = const()[name = tensor("op_964_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_964_end_mask_0 = const()[name = tensor("op_964_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_964_cast_fp16 = slice_by_index(begin = var_964_begin_0, end = var_964_end_0, end_mask = var_964_end_mask_0, x = var_923_cast_fp16)[name = tensor("op_964_cast_fp16")]; + tensor var_965_begin_0 = const()[name = tensor("op_965_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_965_end_0 = const()[name = tensor("op_965_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_965_end_mask_0 = const()[name = tensor("op_965_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_965_cast_fp16 = slice_by_index(begin = var_965_begin_0, end = var_965_end_0, end_mask = var_965_end_mask_0, x = var_923_cast_fp16)[name = tensor("op_965_cast_fp16")]; + tensor var_966_begin_0 = const()[name = tensor("op_966_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_966_end_0 = const()[name = tensor("op_966_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_966_end_mask_0 = const()[name = tensor("op_966_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_966_cast_fp16 = slice_by_index(begin = var_966_begin_0, end = var_966_end_0, end_mask = var_966_end_mask_0, x = var_927_cast_fp16)[name = tensor("op_966_cast_fp16")]; + tensor var_967_begin_0 = const()[name = tensor("op_967_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_967_end_0 = const()[name = tensor("op_967_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_967_end_mask_0 = const()[name = tensor("op_967_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_967_cast_fp16 = slice_by_index(begin = var_967_begin_0, end = var_967_end_0, end_mask = var_967_end_mask_0, x = var_927_cast_fp16)[name = tensor("op_967_cast_fp16")]; + tensor var_968_begin_0 = const()[name = tensor("op_968_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_968_end_0 = const()[name = tensor("op_968_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_968_end_mask_0 = const()[name = tensor("op_968_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_968_cast_fp16 = slice_by_index(begin = var_968_begin_0, end = var_968_end_0, end_mask = var_968_end_mask_0, x = var_927_cast_fp16)[name = tensor("op_968_cast_fp16")]; + tensor var_969_begin_0 = const()[name = tensor("op_969_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_969_end_0 = const()[name = tensor("op_969_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_969_end_mask_0 = const()[name = tensor("op_969_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_969_cast_fp16 = slice_by_index(begin = var_969_begin_0, end = var_969_end_0, end_mask = var_969_end_mask_0, x = var_927_cast_fp16)[name = tensor("op_969_cast_fp16")]; + tensor var_970_begin_0 = const()[name = tensor("op_970_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_970_end_0 = const()[name = tensor("op_970_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_970_end_mask_0 = const()[name = tensor("op_970_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_970_cast_fp16 = slice_by_index(begin = var_970_begin_0, end = var_970_end_0, end_mask = var_970_end_mask_0, x = var_927_cast_fp16)[name = tensor("op_970_cast_fp16")]; + tensor var_971_begin_0 = const()[name = tensor("op_971_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_971_end_0 = const()[name = tensor("op_971_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_971_end_mask_0 = const()[name = tensor("op_971_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_971_cast_fp16 = slice_by_index(begin = var_971_begin_0, end = var_971_end_0, end_mask = var_971_end_mask_0, x = var_927_cast_fp16)[name = tensor("op_971_cast_fp16")]; + tensor var_972_begin_0 = const()[name = tensor("op_972_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_972_end_0 = const()[name = tensor("op_972_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_972_end_mask_0 = const()[name = tensor("op_972_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_972_cast_fp16 = slice_by_index(begin = var_972_begin_0, end = var_972_end_0, end_mask = var_972_end_mask_0, x = var_927_cast_fp16)[name = tensor("op_972_cast_fp16")]; + tensor var_973_begin_0 = const()[name = tensor("op_973_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_973_end_0 = const()[name = tensor("op_973_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_973_end_mask_0 = const()[name = tensor("op_973_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_973_cast_fp16 = slice_by_index(begin = var_973_begin_0, end = var_973_end_0, end_mask = var_973_end_mask_0, x = var_927_cast_fp16)[name = tensor("op_973_cast_fp16")]; + tensor var_974_begin_0 = const()[name = tensor("op_974_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_974_end_0 = const()[name = tensor("op_974_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_974_end_mask_0 = const()[name = tensor("op_974_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_974_cast_fp16 = slice_by_index(begin = var_974_begin_0, end = var_974_end_0, end_mask = var_974_end_mask_0, x = var_931_cast_fp16)[name = tensor("op_974_cast_fp16")]; + tensor var_975_begin_0 = const()[name = tensor("op_975_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_975_end_0 = const()[name = tensor("op_975_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_975_end_mask_0 = const()[name = tensor("op_975_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_975_cast_fp16 = slice_by_index(begin = var_975_begin_0, end = var_975_end_0, end_mask = var_975_end_mask_0, x = var_931_cast_fp16)[name = tensor("op_975_cast_fp16")]; + tensor var_976_begin_0 = const()[name = tensor("op_976_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_976_end_0 = const()[name = tensor("op_976_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_976_end_mask_0 = const()[name = tensor("op_976_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_976_cast_fp16 = slice_by_index(begin = var_976_begin_0, end = var_976_end_0, end_mask = var_976_end_mask_0, x = var_931_cast_fp16)[name = tensor("op_976_cast_fp16")]; + tensor var_977_begin_0 = const()[name = tensor("op_977_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_977_end_0 = const()[name = tensor("op_977_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_977_end_mask_0 = const()[name = tensor("op_977_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_977_cast_fp16 = slice_by_index(begin = var_977_begin_0, end = var_977_end_0, end_mask = var_977_end_mask_0, x = var_931_cast_fp16)[name = tensor("op_977_cast_fp16")]; + tensor var_978_begin_0 = const()[name = tensor("op_978_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_978_end_0 = const()[name = tensor("op_978_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_978_end_mask_0 = const()[name = tensor("op_978_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_978_cast_fp16 = slice_by_index(begin = var_978_begin_0, end = var_978_end_0, end_mask = var_978_end_mask_0, x = var_931_cast_fp16)[name = tensor("op_978_cast_fp16")]; + tensor var_979_begin_0 = const()[name = tensor("op_979_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_979_end_0 = const()[name = tensor("op_979_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_979_end_mask_0 = const()[name = tensor("op_979_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_979_cast_fp16 = slice_by_index(begin = var_979_begin_0, end = var_979_end_0, end_mask = var_979_end_mask_0, x = var_931_cast_fp16)[name = tensor("op_979_cast_fp16")]; + tensor var_980_begin_0 = const()[name = tensor("op_980_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_980_end_0 = const()[name = tensor("op_980_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_980_end_mask_0 = const()[name = tensor("op_980_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_980_cast_fp16 = slice_by_index(begin = var_980_begin_0, end = var_980_end_0, end_mask = var_980_end_mask_0, x = var_931_cast_fp16)[name = tensor("op_980_cast_fp16")]; + tensor var_981_begin_0 = const()[name = tensor("op_981_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_981_end_0 = const()[name = tensor("op_981_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_981_end_mask_0 = const()[name = tensor("op_981_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_981_cast_fp16 = slice_by_index(begin = var_981_begin_0, end = var_981_end_0, end_mask = var_981_end_mask_0, x = var_931_cast_fp16)[name = tensor("op_981_cast_fp16")]; + tensor var_982_begin_0 = const()[name = tensor("op_982_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_982_end_0 = const()[name = tensor("op_982_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_982_end_mask_0 = const()[name = tensor("op_982_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_982_cast_fp16 = slice_by_index(begin = var_982_begin_0, end = var_982_end_0, end_mask = var_982_end_mask_0, x = var_935_cast_fp16)[name = tensor("op_982_cast_fp16")]; + tensor var_983_begin_0 = const()[name = tensor("op_983_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_983_end_0 = const()[name = tensor("op_983_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_983_end_mask_0 = const()[name = tensor("op_983_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_983_cast_fp16 = slice_by_index(begin = var_983_begin_0, end = var_983_end_0, end_mask = var_983_end_mask_0, x = var_935_cast_fp16)[name = tensor("op_983_cast_fp16")]; + tensor var_984_begin_0 = const()[name = tensor("op_984_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_984_end_0 = const()[name = tensor("op_984_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_984_end_mask_0 = const()[name = tensor("op_984_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_984_cast_fp16 = slice_by_index(begin = var_984_begin_0, end = var_984_end_0, end_mask = var_984_end_mask_0, x = var_935_cast_fp16)[name = tensor("op_984_cast_fp16")]; + tensor var_985_begin_0 = const()[name = tensor("op_985_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_985_end_0 = const()[name = tensor("op_985_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_985_end_mask_0 = const()[name = tensor("op_985_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_985_cast_fp16 = slice_by_index(begin = var_985_begin_0, end = var_985_end_0, end_mask = var_985_end_mask_0, x = var_935_cast_fp16)[name = tensor("op_985_cast_fp16")]; + tensor var_986_begin_0 = const()[name = tensor("op_986_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_986_end_0 = const()[name = tensor("op_986_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_986_end_mask_0 = const()[name = tensor("op_986_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_986_cast_fp16 = slice_by_index(begin = var_986_begin_0, end = var_986_end_0, end_mask = var_986_end_mask_0, x = var_935_cast_fp16)[name = tensor("op_986_cast_fp16")]; + tensor var_987_begin_0 = const()[name = tensor("op_987_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_987_end_0 = const()[name = tensor("op_987_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_987_end_mask_0 = const()[name = tensor("op_987_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_987_cast_fp16 = slice_by_index(begin = var_987_begin_0, end = var_987_end_0, end_mask = var_987_end_mask_0, x = var_935_cast_fp16)[name = tensor("op_987_cast_fp16")]; + tensor var_988_begin_0 = const()[name = tensor("op_988_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_988_end_0 = const()[name = tensor("op_988_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_988_end_mask_0 = const()[name = tensor("op_988_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_988_cast_fp16 = slice_by_index(begin = var_988_begin_0, end = var_988_end_0, end_mask = var_988_end_mask_0, x = var_935_cast_fp16)[name = tensor("op_988_cast_fp16")]; + tensor var_989_begin_0 = const()[name = tensor("op_989_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_989_end_0 = const()[name = tensor("op_989_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_989_end_mask_0 = const()[name = tensor("op_989_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_989_cast_fp16 = slice_by_index(begin = var_989_begin_0, end = var_989_end_0, end_mask = var_989_end_mask_0, x = var_935_cast_fp16)[name = tensor("op_989_cast_fp16")]; + tensor var_990_begin_0 = const()[name = tensor("op_990_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_990_end_0 = const()[name = tensor("op_990_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_990_end_mask_0 = const()[name = tensor("op_990_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_990_cast_fp16 = slice_by_index(begin = var_990_begin_0, end = var_990_end_0, end_mask = var_990_end_mask_0, x = var_939_cast_fp16)[name = tensor("op_990_cast_fp16")]; + tensor var_991_begin_0 = const()[name = tensor("op_991_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_991_end_0 = const()[name = tensor("op_991_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_991_end_mask_0 = const()[name = tensor("op_991_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_991_cast_fp16 = slice_by_index(begin = var_991_begin_0, end = var_991_end_0, end_mask = var_991_end_mask_0, x = var_939_cast_fp16)[name = tensor("op_991_cast_fp16")]; + tensor var_992_begin_0 = const()[name = tensor("op_992_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_992_end_0 = const()[name = tensor("op_992_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_992_end_mask_0 = const()[name = tensor("op_992_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_992_cast_fp16 = slice_by_index(begin = var_992_begin_0, end = var_992_end_0, end_mask = var_992_end_mask_0, x = var_939_cast_fp16)[name = tensor("op_992_cast_fp16")]; + tensor var_993_begin_0 = const()[name = tensor("op_993_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_993_end_0 = const()[name = tensor("op_993_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_993_end_mask_0 = const()[name = tensor("op_993_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_993_cast_fp16 = slice_by_index(begin = var_993_begin_0, end = var_993_end_0, end_mask = var_993_end_mask_0, x = var_939_cast_fp16)[name = tensor("op_993_cast_fp16")]; + tensor var_994_begin_0 = const()[name = tensor("op_994_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_994_end_0 = const()[name = tensor("op_994_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_994_end_mask_0 = const()[name = tensor("op_994_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_994_cast_fp16 = slice_by_index(begin = var_994_begin_0, end = var_994_end_0, end_mask = var_994_end_mask_0, x = var_939_cast_fp16)[name = tensor("op_994_cast_fp16")]; + tensor var_995_begin_0 = const()[name = tensor("op_995_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_995_end_0 = const()[name = tensor("op_995_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_995_end_mask_0 = const()[name = tensor("op_995_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_995_cast_fp16 = slice_by_index(begin = var_995_begin_0, end = var_995_end_0, end_mask = var_995_end_mask_0, x = var_939_cast_fp16)[name = tensor("op_995_cast_fp16")]; + tensor var_996_begin_0 = const()[name = tensor("op_996_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_996_end_0 = const()[name = tensor("op_996_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_996_end_mask_0 = const()[name = tensor("op_996_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_996_cast_fp16 = slice_by_index(begin = var_996_begin_0, end = var_996_end_0, end_mask = var_996_end_mask_0, x = var_939_cast_fp16)[name = tensor("op_996_cast_fp16")]; + tensor var_997_begin_0 = const()[name = tensor("op_997_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_997_end_0 = const()[name = tensor("op_997_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_997_end_mask_0 = const()[name = tensor("op_997_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_997_cast_fp16 = slice_by_index(begin = var_997_begin_0, end = var_997_end_0, end_mask = var_997_end_mask_0, x = var_939_cast_fp16)[name = tensor("op_997_cast_fp16")]; + tensor var_998_begin_0 = const()[name = tensor("op_998_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_998_end_0 = const()[name = tensor("op_998_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_998_end_mask_0 = const()[name = tensor("op_998_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_998_cast_fp16 = slice_by_index(begin = var_998_begin_0, end = var_998_end_0, end_mask = var_998_end_mask_0, x = var_943_cast_fp16)[name = tensor("op_998_cast_fp16")]; + tensor var_999_begin_0 = const()[name = tensor("op_999_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_999_end_0 = const()[name = tensor("op_999_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_999_end_mask_0 = const()[name = tensor("op_999_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_999_cast_fp16 = slice_by_index(begin = var_999_begin_0, end = var_999_end_0, end_mask = var_999_end_mask_0, x = var_943_cast_fp16)[name = tensor("op_999_cast_fp16")]; + tensor var_1000_begin_0 = const()[name = tensor("op_1000_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_1000_end_0 = const()[name = tensor("op_1000_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_1000_end_mask_0 = const()[name = tensor("op_1000_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1000_cast_fp16 = slice_by_index(begin = var_1000_begin_0, end = var_1000_end_0, end_mask = var_1000_end_mask_0, x = var_943_cast_fp16)[name = tensor("op_1000_cast_fp16")]; + tensor var_1001_begin_0 = const()[name = tensor("op_1001_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_1001_end_0 = const()[name = tensor("op_1001_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_1001_end_mask_0 = const()[name = tensor("op_1001_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1001_cast_fp16 = slice_by_index(begin = var_1001_begin_0, end = var_1001_end_0, end_mask = var_1001_end_mask_0, x = var_943_cast_fp16)[name = tensor("op_1001_cast_fp16")]; + tensor var_1002_begin_0 = const()[name = tensor("op_1002_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_1002_end_0 = const()[name = tensor("op_1002_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_1002_end_mask_0 = const()[name = tensor("op_1002_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1002_cast_fp16 = slice_by_index(begin = var_1002_begin_0, end = var_1002_end_0, end_mask = var_1002_end_mask_0, x = var_943_cast_fp16)[name = tensor("op_1002_cast_fp16")]; + tensor var_1003_begin_0 = const()[name = tensor("op_1003_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_1003_end_0 = const()[name = tensor("op_1003_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_1003_end_mask_0 = const()[name = tensor("op_1003_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1003_cast_fp16 = slice_by_index(begin = var_1003_begin_0, end = var_1003_end_0, end_mask = var_1003_end_mask_0, x = var_943_cast_fp16)[name = tensor("op_1003_cast_fp16")]; + tensor var_1004_begin_0 = const()[name = tensor("op_1004_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_1004_end_0 = const()[name = tensor("op_1004_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_1004_end_mask_0 = const()[name = tensor("op_1004_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1004_cast_fp16 = slice_by_index(begin = var_1004_begin_0, end = var_1004_end_0, end_mask = var_1004_end_mask_0, x = var_943_cast_fp16)[name = tensor("op_1004_cast_fp16")]; + tensor var_1005_begin_0 = const()[name = tensor("op_1005_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_1005_end_0 = const()[name = tensor("op_1005_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_1005_end_mask_0 = const()[name = tensor("op_1005_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1005_cast_fp16 = slice_by_index(begin = var_1005_begin_0, end = var_1005_end_0, end_mask = var_1005_end_mask_0, x = var_943_cast_fp16)[name = tensor("op_1005_cast_fp16")]; + tensor var_1006_begin_0 = const()[name = tensor("op_1006_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1006_end_0 = const()[name = tensor("op_1006_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_1006_end_mask_0 = const()[name = tensor("op_1006_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1006_cast_fp16 = slice_by_index(begin = var_1006_begin_0, end = var_1006_end_0, end_mask = var_1006_end_mask_0, x = var_947_cast_fp16)[name = tensor("op_1006_cast_fp16")]; + tensor var_1007_begin_0 = const()[name = tensor("op_1007_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1007_end_0 = const()[name = tensor("op_1007_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_1007_end_mask_0 = const()[name = tensor("op_1007_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1007_cast_fp16 = slice_by_index(begin = var_1007_begin_0, end = var_1007_end_0, end_mask = var_1007_end_mask_0, x = var_947_cast_fp16)[name = tensor("op_1007_cast_fp16")]; + tensor var_1008_begin_0 = const()[name = tensor("op_1008_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_1008_end_0 = const()[name = tensor("op_1008_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_1008_end_mask_0 = const()[name = tensor("op_1008_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1008_cast_fp16 = slice_by_index(begin = var_1008_begin_0, end = var_1008_end_0, end_mask = var_1008_end_mask_0, x = var_947_cast_fp16)[name = tensor("op_1008_cast_fp16")]; + tensor var_1009_begin_0 = const()[name = tensor("op_1009_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_1009_end_0 = const()[name = tensor("op_1009_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_1009_end_mask_0 = const()[name = tensor("op_1009_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1009_cast_fp16 = slice_by_index(begin = var_1009_begin_0, end = var_1009_end_0, end_mask = var_1009_end_mask_0, x = var_947_cast_fp16)[name = tensor("op_1009_cast_fp16")]; + tensor var_1010_begin_0 = const()[name = tensor("op_1010_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_1010_end_0 = const()[name = tensor("op_1010_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_1010_end_mask_0 = const()[name = tensor("op_1010_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1010_cast_fp16 = slice_by_index(begin = var_1010_begin_0, end = var_1010_end_0, end_mask = var_1010_end_mask_0, x = var_947_cast_fp16)[name = tensor("op_1010_cast_fp16")]; + tensor var_1011_begin_0 = const()[name = tensor("op_1011_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_1011_end_0 = const()[name = tensor("op_1011_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_1011_end_mask_0 = const()[name = tensor("op_1011_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1011_cast_fp16 = slice_by_index(begin = var_1011_begin_0, end = var_1011_end_0, end_mask = var_1011_end_mask_0, x = var_947_cast_fp16)[name = tensor("op_1011_cast_fp16")]; + tensor var_1012_begin_0 = const()[name = tensor("op_1012_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_1012_end_0 = const()[name = tensor("op_1012_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_1012_end_mask_0 = const()[name = tensor("op_1012_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1012_cast_fp16 = slice_by_index(begin = var_1012_begin_0, end = var_1012_end_0, end_mask = var_1012_end_mask_0, x = var_947_cast_fp16)[name = tensor("op_1012_cast_fp16")]; + tensor var_1013_begin_0 = const()[name = tensor("op_1013_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_1013_end_0 = const()[name = tensor("op_1013_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_1013_end_mask_0 = const()[name = tensor("op_1013_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1013_cast_fp16 = slice_by_index(begin = var_1013_begin_0, end = var_1013_end_0, end_mask = var_1013_end_mask_0, x = var_947_cast_fp16)[name = tensor("op_1013_cast_fp16")]; + tensor k_7_perm_0 = const()[name = tensor("k_7_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_1018_begin_0 = const()[name = tensor("op_1018_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1018_end_0 = const()[name = tensor("op_1018_end_0"), val = tensor([2, 77, 1, 40])]; + tensor var_1018_end_mask_0 = const()[name = tensor("op_1018_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_7_cast_fp16 = transpose(perm = k_7_perm_0, x = k_5_cast_fp16)[name = tensor("transpose_12")]; + tensor var_1018_cast_fp16 = slice_by_index(begin = var_1018_begin_0, end = var_1018_end_0, end_mask = var_1018_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_1018_cast_fp16")]; + tensor var_1022_begin_0 = const()[name = tensor("op_1022_begin_0"), val = tensor([0, 0, 0, 40])]; + tensor var_1022_end_0 = const()[name = tensor("op_1022_end_0"), val = tensor([2, 77, 1, 80])]; + tensor var_1022_end_mask_0 = const()[name = tensor("op_1022_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1022_cast_fp16 = slice_by_index(begin = var_1022_begin_0, end = var_1022_end_0, end_mask = var_1022_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_1022_cast_fp16")]; + tensor var_1026_begin_0 = const()[name = tensor("op_1026_begin_0"), val = tensor([0, 0, 0, 80])]; + tensor var_1026_end_0 = const()[name = tensor("op_1026_end_0"), val = tensor([2, 77, 1, 120])]; + tensor var_1026_end_mask_0 = const()[name = tensor("op_1026_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1026_cast_fp16 = slice_by_index(begin = var_1026_begin_0, end = var_1026_end_0, end_mask = var_1026_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_1026_cast_fp16")]; + tensor var_1030_begin_0 = const()[name = tensor("op_1030_begin_0"), val = tensor([0, 0, 0, 120])]; + tensor var_1030_end_0 = const()[name = tensor("op_1030_end_0"), val = tensor([2, 77, 1, 160])]; + tensor var_1030_end_mask_0 = const()[name = tensor("op_1030_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1030_cast_fp16 = slice_by_index(begin = var_1030_begin_0, end = var_1030_end_0, end_mask = var_1030_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_1030_cast_fp16")]; + tensor var_1034_begin_0 = const()[name = tensor("op_1034_begin_0"), val = tensor([0, 0, 0, 160])]; + tensor var_1034_end_0 = const()[name = tensor("op_1034_end_0"), val = tensor([2, 77, 1, 200])]; + tensor var_1034_end_mask_0 = const()[name = tensor("op_1034_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1034_cast_fp16 = slice_by_index(begin = var_1034_begin_0, end = var_1034_end_0, end_mask = var_1034_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_1034_cast_fp16")]; + tensor var_1038_begin_0 = const()[name = tensor("op_1038_begin_0"), val = tensor([0, 0, 0, 200])]; + tensor var_1038_end_0 = const()[name = tensor("op_1038_end_0"), val = tensor([2, 77, 1, 240])]; + tensor var_1038_end_mask_0 = const()[name = tensor("op_1038_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1038_cast_fp16 = slice_by_index(begin = var_1038_begin_0, end = var_1038_end_0, end_mask = var_1038_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_1038_cast_fp16")]; + tensor var_1042_begin_0 = const()[name = tensor("op_1042_begin_0"), val = tensor([0, 0, 0, 240])]; + tensor var_1042_end_0 = const()[name = tensor("op_1042_end_0"), val = tensor([2, 77, 1, 280])]; + tensor var_1042_end_mask_0 = const()[name = tensor("op_1042_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1042_cast_fp16 = slice_by_index(begin = var_1042_begin_0, end = var_1042_end_0, end_mask = var_1042_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_1042_cast_fp16")]; + tensor var_1046_begin_0 = const()[name = tensor("op_1046_begin_0"), val = tensor([0, 0, 0, 280])]; + tensor var_1046_end_0 = const()[name = tensor("op_1046_end_0"), val = tensor([2, 77, 1, 1])]; + tensor var_1046_end_mask_0 = const()[name = tensor("op_1046_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1046_cast_fp16 = slice_by_index(begin = var_1046_begin_0, end = var_1046_end_0, end_mask = var_1046_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_1046_cast_fp16")]; + tensor var_1048_begin_0 = const()[name = tensor("op_1048_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1048_end_0 = const()[name = tensor("op_1048_end_0"), val = tensor([2, 40, 1, 77])]; + tensor var_1048_end_mask_0 = const()[name = tensor("op_1048_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1048_cast_fp16 = slice_by_index(begin = var_1048_begin_0, end = var_1048_end_0, end_mask = var_1048_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_1048_cast_fp16")]; + tensor var_1052_begin_0 = const()[name = tensor("op_1052_begin_0"), val = tensor([0, 40, 0, 0])]; + tensor var_1052_end_0 = const()[name = tensor("op_1052_end_0"), val = tensor([2, 80, 1, 77])]; + tensor var_1052_end_mask_0 = const()[name = tensor("op_1052_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1052_cast_fp16 = slice_by_index(begin = var_1052_begin_0, end = var_1052_end_0, end_mask = var_1052_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_1052_cast_fp16")]; + tensor var_1056_begin_0 = const()[name = tensor("op_1056_begin_0"), val = tensor([0, 80, 0, 0])]; + tensor var_1056_end_0 = const()[name = tensor("op_1056_end_0"), val = tensor([2, 120, 1, 77])]; + tensor var_1056_end_mask_0 = const()[name = tensor("op_1056_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1056_cast_fp16 = slice_by_index(begin = var_1056_begin_0, end = var_1056_end_0, end_mask = var_1056_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_1056_cast_fp16")]; + tensor var_1060_begin_0 = const()[name = tensor("op_1060_begin_0"), val = tensor([0, 120, 0, 0])]; + tensor var_1060_end_0 = const()[name = tensor("op_1060_end_0"), val = tensor([2, 160, 1, 77])]; + tensor var_1060_end_mask_0 = const()[name = tensor("op_1060_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1060_cast_fp16 = slice_by_index(begin = var_1060_begin_0, end = var_1060_end_0, end_mask = var_1060_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_1060_cast_fp16")]; + tensor var_1064_begin_0 = const()[name = tensor("op_1064_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_1064_end_0 = const()[name = tensor("op_1064_end_0"), val = tensor([2, 200, 1, 77])]; + tensor var_1064_end_mask_0 = const()[name = tensor("op_1064_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1064_cast_fp16 = slice_by_index(begin = var_1064_begin_0, end = var_1064_end_0, end_mask = var_1064_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_1064_cast_fp16")]; + tensor var_1068_begin_0 = const()[name = tensor("op_1068_begin_0"), val = tensor([0, 200, 0, 0])]; + tensor var_1068_end_0 = const()[name = tensor("op_1068_end_0"), val = tensor([2, 240, 1, 77])]; + tensor var_1068_end_mask_0 = const()[name = tensor("op_1068_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1068_cast_fp16 = slice_by_index(begin = var_1068_begin_0, end = var_1068_end_0, end_mask = var_1068_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_1068_cast_fp16")]; + tensor var_1072_begin_0 = const()[name = tensor("op_1072_begin_0"), val = tensor([0, 240, 0, 0])]; + tensor var_1072_end_0 = const()[name = tensor("op_1072_end_0"), val = tensor([2, 280, 1, 77])]; + tensor var_1072_end_mask_0 = const()[name = tensor("op_1072_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1072_cast_fp16 = slice_by_index(begin = var_1072_begin_0, end = var_1072_end_0, end_mask = var_1072_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_1072_cast_fp16")]; + tensor var_1076_begin_0 = const()[name = tensor("op_1076_begin_0"), val = tensor([0, 280, 0, 0])]; + tensor var_1076_end_0 = const()[name = tensor("op_1076_end_0"), val = tensor([2, 1, 1, 77])]; + tensor var_1076_end_mask_0 = const()[name = tensor("op_1076_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1076_cast_fp16 = slice_by_index(begin = var_1076_begin_0, end = var_1076_end_0, end_mask = var_1076_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_1076_cast_fp16")]; + tensor var_1080_equation_0 = const()[name = tensor("op_1080_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1080_cast_fp16 = einsum(equation = var_1080_equation_0, values = (var_1018_cast_fp16, var_950_cast_fp16))[name = tensor("op_1080_cast_fp16")]; + tensor var_1081_to_fp16 = const()[name = tensor("op_1081_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_129_cast_fp16 = mul(x = var_1080_cast_fp16, y = var_1081_to_fp16)[name = tensor("aw_chunk_129_cast_fp16")]; + tensor var_1084_equation_0 = const()[name = tensor("op_1084_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1084_cast_fp16 = einsum(equation = var_1084_equation_0, values = (var_1018_cast_fp16, var_951_cast_fp16))[name = tensor("op_1084_cast_fp16")]; + tensor var_1085_to_fp16 = const()[name = tensor("op_1085_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_131_cast_fp16 = mul(x = var_1084_cast_fp16, y = var_1085_to_fp16)[name = tensor("aw_chunk_131_cast_fp16")]; + tensor var_1088_equation_0 = const()[name = tensor("op_1088_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1088_cast_fp16 = einsum(equation = var_1088_equation_0, values = (var_1018_cast_fp16, var_952_cast_fp16))[name = tensor("op_1088_cast_fp16")]; + tensor var_1089_to_fp16 = const()[name = tensor("op_1089_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_133_cast_fp16 = mul(x = var_1088_cast_fp16, y = var_1089_to_fp16)[name = tensor("aw_chunk_133_cast_fp16")]; + tensor var_1092_equation_0 = const()[name = tensor("op_1092_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1092_cast_fp16 = einsum(equation = var_1092_equation_0, values = (var_1018_cast_fp16, var_953_cast_fp16))[name = tensor("op_1092_cast_fp16")]; + tensor var_1093_to_fp16 = const()[name = tensor("op_1093_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_135_cast_fp16 = mul(x = var_1092_cast_fp16, y = var_1093_to_fp16)[name = tensor("aw_chunk_135_cast_fp16")]; + tensor var_1096_equation_0 = const()[name = tensor("op_1096_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1096_cast_fp16 = einsum(equation = var_1096_equation_0, values = (var_1018_cast_fp16, var_954_cast_fp16))[name = tensor("op_1096_cast_fp16")]; + tensor var_1097_to_fp16 = const()[name = tensor("op_1097_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_137_cast_fp16 = mul(x = var_1096_cast_fp16, y = var_1097_to_fp16)[name = tensor("aw_chunk_137_cast_fp16")]; + tensor var_1100_equation_0 = const()[name = tensor("op_1100_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1100_cast_fp16 = einsum(equation = var_1100_equation_0, values = (var_1018_cast_fp16, var_955_cast_fp16))[name = tensor("op_1100_cast_fp16")]; + tensor var_1101_to_fp16 = const()[name = tensor("op_1101_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_139_cast_fp16 = mul(x = var_1100_cast_fp16, y = var_1101_to_fp16)[name = tensor("aw_chunk_139_cast_fp16")]; + tensor var_1104_equation_0 = const()[name = tensor("op_1104_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1104_cast_fp16 = einsum(equation = var_1104_equation_0, values = (var_1018_cast_fp16, var_956_cast_fp16))[name = tensor("op_1104_cast_fp16")]; + tensor var_1105_to_fp16 = const()[name = tensor("op_1105_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_141_cast_fp16 = mul(x = var_1104_cast_fp16, y = var_1105_to_fp16)[name = tensor("aw_chunk_141_cast_fp16")]; + tensor var_1108_equation_0 = const()[name = tensor("op_1108_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1108_cast_fp16 = einsum(equation = var_1108_equation_0, values = (var_1018_cast_fp16, var_957_cast_fp16))[name = tensor("op_1108_cast_fp16")]; + tensor var_1109_to_fp16 = const()[name = tensor("op_1109_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_143_cast_fp16 = mul(x = var_1108_cast_fp16, y = var_1109_to_fp16)[name = tensor("aw_chunk_143_cast_fp16")]; + tensor var_1112_equation_0 = const()[name = tensor("op_1112_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1112_cast_fp16 = einsum(equation = var_1112_equation_0, values = (var_1022_cast_fp16, var_958_cast_fp16))[name = tensor("op_1112_cast_fp16")]; + tensor var_1113_to_fp16 = const()[name = tensor("op_1113_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_145_cast_fp16 = mul(x = var_1112_cast_fp16, y = var_1113_to_fp16)[name = tensor("aw_chunk_145_cast_fp16")]; + tensor var_1116_equation_0 = const()[name = tensor("op_1116_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1116_cast_fp16 = einsum(equation = var_1116_equation_0, values = (var_1022_cast_fp16, var_959_cast_fp16))[name = tensor("op_1116_cast_fp16")]; + tensor var_1117_to_fp16 = const()[name = tensor("op_1117_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_147_cast_fp16 = mul(x = var_1116_cast_fp16, y = var_1117_to_fp16)[name = tensor("aw_chunk_147_cast_fp16")]; + tensor var_1120_equation_0 = const()[name = tensor("op_1120_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1120_cast_fp16 = einsum(equation = var_1120_equation_0, values = (var_1022_cast_fp16, var_960_cast_fp16))[name = tensor("op_1120_cast_fp16")]; + tensor var_1121_to_fp16 = const()[name = tensor("op_1121_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_149_cast_fp16 = mul(x = var_1120_cast_fp16, y = var_1121_to_fp16)[name = tensor("aw_chunk_149_cast_fp16")]; + tensor var_1124_equation_0 = const()[name = tensor("op_1124_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1124_cast_fp16 = einsum(equation = var_1124_equation_0, values = (var_1022_cast_fp16, var_961_cast_fp16))[name = tensor("op_1124_cast_fp16")]; + tensor var_1125_to_fp16 = const()[name = tensor("op_1125_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_151_cast_fp16 = mul(x = var_1124_cast_fp16, y = var_1125_to_fp16)[name = tensor("aw_chunk_151_cast_fp16")]; + tensor var_1128_equation_0 = const()[name = tensor("op_1128_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1128_cast_fp16 = einsum(equation = var_1128_equation_0, values = (var_1022_cast_fp16, var_962_cast_fp16))[name = tensor("op_1128_cast_fp16")]; + tensor var_1129_to_fp16 = const()[name = tensor("op_1129_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_153_cast_fp16 = mul(x = var_1128_cast_fp16, y = var_1129_to_fp16)[name = tensor("aw_chunk_153_cast_fp16")]; + tensor var_1132_equation_0 = const()[name = tensor("op_1132_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1132_cast_fp16 = einsum(equation = var_1132_equation_0, values = (var_1022_cast_fp16, var_963_cast_fp16))[name = tensor("op_1132_cast_fp16")]; + tensor var_1133_to_fp16 = const()[name = tensor("op_1133_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_155_cast_fp16 = mul(x = var_1132_cast_fp16, y = var_1133_to_fp16)[name = tensor("aw_chunk_155_cast_fp16")]; + tensor var_1136_equation_0 = const()[name = tensor("op_1136_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1136_cast_fp16 = einsum(equation = var_1136_equation_0, values = (var_1022_cast_fp16, var_964_cast_fp16))[name = tensor("op_1136_cast_fp16")]; + tensor var_1137_to_fp16 = const()[name = tensor("op_1137_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_157_cast_fp16 = mul(x = var_1136_cast_fp16, y = var_1137_to_fp16)[name = tensor("aw_chunk_157_cast_fp16")]; + tensor var_1140_equation_0 = const()[name = tensor("op_1140_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1140_cast_fp16 = einsum(equation = var_1140_equation_0, values = (var_1022_cast_fp16, var_965_cast_fp16))[name = tensor("op_1140_cast_fp16")]; + tensor var_1141_to_fp16 = const()[name = tensor("op_1141_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_159_cast_fp16 = mul(x = var_1140_cast_fp16, y = var_1141_to_fp16)[name = tensor("aw_chunk_159_cast_fp16")]; + tensor var_1144_equation_0 = const()[name = tensor("op_1144_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1144_cast_fp16 = einsum(equation = var_1144_equation_0, values = (var_1026_cast_fp16, var_966_cast_fp16))[name = tensor("op_1144_cast_fp16")]; + tensor var_1145_to_fp16 = const()[name = tensor("op_1145_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_161_cast_fp16 = mul(x = var_1144_cast_fp16, y = var_1145_to_fp16)[name = tensor("aw_chunk_161_cast_fp16")]; + tensor var_1148_equation_0 = const()[name = tensor("op_1148_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1148_cast_fp16 = einsum(equation = var_1148_equation_0, values = (var_1026_cast_fp16, var_967_cast_fp16))[name = tensor("op_1148_cast_fp16")]; + tensor var_1149_to_fp16 = const()[name = tensor("op_1149_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_163_cast_fp16 = mul(x = var_1148_cast_fp16, y = var_1149_to_fp16)[name = tensor("aw_chunk_163_cast_fp16")]; + tensor var_1152_equation_0 = const()[name = tensor("op_1152_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1152_cast_fp16 = einsum(equation = var_1152_equation_0, values = (var_1026_cast_fp16, var_968_cast_fp16))[name = tensor("op_1152_cast_fp16")]; + tensor var_1153_to_fp16 = const()[name = tensor("op_1153_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_165_cast_fp16 = mul(x = var_1152_cast_fp16, y = var_1153_to_fp16)[name = tensor("aw_chunk_165_cast_fp16")]; + tensor var_1156_equation_0 = const()[name = tensor("op_1156_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1156_cast_fp16 = einsum(equation = var_1156_equation_0, values = (var_1026_cast_fp16, var_969_cast_fp16))[name = tensor("op_1156_cast_fp16")]; + tensor var_1157_to_fp16 = const()[name = tensor("op_1157_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_167_cast_fp16 = mul(x = var_1156_cast_fp16, y = var_1157_to_fp16)[name = tensor("aw_chunk_167_cast_fp16")]; + tensor var_1160_equation_0 = const()[name = tensor("op_1160_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1160_cast_fp16 = einsum(equation = var_1160_equation_0, values = (var_1026_cast_fp16, var_970_cast_fp16))[name = tensor("op_1160_cast_fp16")]; + tensor var_1161_to_fp16 = const()[name = tensor("op_1161_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_169_cast_fp16 = mul(x = var_1160_cast_fp16, y = var_1161_to_fp16)[name = tensor("aw_chunk_169_cast_fp16")]; + tensor var_1164_equation_0 = const()[name = tensor("op_1164_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1164_cast_fp16 = einsum(equation = var_1164_equation_0, values = (var_1026_cast_fp16, var_971_cast_fp16))[name = tensor("op_1164_cast_fp16")]; + tensor var_1165_to_fp16 = const()[name = tensor("op_1165_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_171_cast_fp16 = mul(x = var_1164_cast_fp16, y = var_1165_to_fp16)[name = tensor("aw_chunk_171_cast_fp16")]; + tensor var_1168_equation_0 = const()[name = tensor("op_1168_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1168_cast_fp16 = einsum(equation = var_1168_equation_0, values = (var_1026_cast_fp16, var_972_cast_fp16))[name = tensor("op_1168_cast_fp16")]; + tensor var_1169_to_fp16 = const()[name = tensor("op_1169_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_173_cast_fp16 = mul(x = var_1168_cast_fp16, y = var_1169_to_fp16)[name = tensor("aw_chunk_173_cast_fp16")]; + tensor var_1172_equation_0 = const()[name = tensor("op_1172_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1172_cast_fp16 = einsum(equation = var_1172_equation_0, values = (var_1026_cast_fp16, var_973_cast_fp16))[name = tensor("op_1172_cast_fp16")]; + tensor var_1173_to_fp16 = const()[name = tensor("op_1173_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_175_cast_fp16 = mul(x = var_1172_cast_fp16, y = var_1173_to_fp16)[name = tensor("aw_chunk_175_cast_fp16")]; + tensor var_1176_equation_0 = const()[name = tensor("op_1176_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1176_cast_fp16 = einsum(equation = var_1176_equation_0, values = (var_1030_cast_fp16, var_974_cast_fp16))[name = tensor("op_1176_cast_fp16")]; + tensor var_1177_to_fp16 = const()[name = tensor("op_1177_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_177_cast_fp16 = mul(x = var_1176_cast_fp16, y = var_1177_to_fp16)[name = tensor("aw_chunk_177_cast_fp16")]; + tensor var_1180_equation_0 = const()[name = tensor("op_1180_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1180_cast_fp16 = einsum(equation = var_1180_equation_0, values = (var_1030_cast_fp16, var_975_cast_fp16))[name = tensor("op_1180_cast_fp16")]; + tensor var_1181_to_fp16 = const()[name = tensor("op_1181_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_179_cast_fp16 = mul(x = var_1180_cast_fp16, y = var_1181_to_fp16)[name = tensor("aw_chunk_179_cast_fp16")]; + tensor var_1184_equation_0 = const()[name = tensor("op_1184_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1184_cast_fp16 = einsum(equation = var_1184_equation_0, values = (var_1030_cast_fp16, var_976_cast_fp16))[name = tensor("op_1184_cast_fp16")]; + tensor var_1185_to_fp16 = const()[name = tensor("op_1185_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_181_cast_fp16 = mul(x = var_1184_cast_fp16, y = var_1185_to_fp16)[name = tensor("aw_chunk_181_cast_fp16")]; + tensor var_1188_equation_0 = const()[name = tensor("op_1188_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1188_cast_fp16 = einsum(equation = var_1188_equation_0, values = (var_1030_cast_fp16, var_977_cast_fp16))[name = tensor("op_1188_cast_fp16")]; + tensor var_1189_to_fp16 = const()[name = tensor("op_1189_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_183_cast_fp16 = mul(x = var_1188_cast_fp16, y = var_1189_to_fp16)[name = tensor("aw_chunk_183_cast_fp16")]; + tensor var_1192_equation_0 = const()[name = tensor("op_1192_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1192_cast_fp16 = einsum(equation = var_1192_equation_0, values = (var_1030_cast_fp16, var_978_cast_fp16))[name = tensor("op_1192_cast_fp16")]; + tensor var_1193_to_fp16 = const()[name = tensor("op_1193_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_185_cast_fp16 = mul(x = var_1192_cast_fp16, y = var_1193_to_fp16)[name = tensor("aw_chunk_185_cast_fp16")]; + tensor var_1196_equation_0 = const()[name = tensor("op_1196_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1196_cast_fp16 = einsum(equation = var_1196_equation_0, values = (var_1030_cast_fp16, var_979_cast_fp16))[name = tensor("op_1196_cast_fp16")]; + tensor var_1197_to_fp16 = const()[name = tensor("op_1197_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_187_cast_fp16 = mul(x = var_1196_cast_fp16, y = var_1197_to_fp16)[name = tensor("aw_chunk_187_cast_fp16")]; + tensor var_1200_equation_0 = const()[name = tensor("op_1200_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1200_cast_fp16 = einsum(equation = var_1200_equation_0, values = (var_1030_cast_fp16, var_980_cast_fp16))[name = tensor("op_1200_cast_fp16")]; + tensor var_1201_to_fp16 = const()[name = tensor("op_1201_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_189_cast_fp16 = mul(x = var_1200_cast_fp16, y = var_1201_to_fp16)[name = tensor("aw_chunk_189_cast_fp16")]; + tensor var_1204_equation_0 = const()[name = tensor("op_1204_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1204_cast_fp16 = einsum(equation = var_1204_equation_0, values = (var_1030_cast_fp16, var_981_cast_fp16))[name = tensor("op_1204_cast_fp16")]; + tensor var_1205_to_fp16 = const()[name = tensor("op_1205_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_191_cast_fp16 = mul(x = var_1204_cast_fp16, y = var_1205_to_fp16)[name = tensor("aw_chunk_191_cast_fp16")]; + tensor var_1208_equation_0 = const()[name = tensor("op_1208_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1208_cast_fp16 = einsum(equation = var_1208_equation_0, values = (var_1034_cast_fp16, var_982_cast_fp16))[name = tensor("op_1208_cast_fp16")]; + tensor var_1209_to_fp16 = const()[name = tensor("op_1209_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_193_cast_fp16 = mul(x = var_1208_cast_fp16, y = var_1209_to_fp16)[name = tensor("aw_chunk_193_cast_fp16")]; + tensor var_1212_equation_0 = const()[name = tensor("op_1212_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1212_cast_fp16 = einsum(equation = var_1212_equation_0, values = (var_1034_cast_fp16, var_983_cast_fp16))[name = tensor("op_1212_cast_fp16")]; + tensor var_1213_to_fp16 = const()[name = tensor("op_1213_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_195_cast_fp16 = mul(x = var_1212_cast_fp16, y = var_1213_to_fp16)[name = tensor("aw_chunk_195_cast_fp16")]; + tensor var_1216_equation_0 = const()[name = tensor("op_1216_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1216_cast_fp16 = einsum(equation = var_1216_equation_0, values = (var_1034_cast_fp16, var_984_cast_fp16))[name = tensor("op_1216_cast_fp16")]; + tensor var_1217_to_fp16 = const()[name = tensor("op_1217_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_197_cast_fp16 = mul(x = var_1216_cast_fp16, y = var_1217_to_fp16)[name = tensor("aw_chunk_197_cast_fp16")]; + tensor var_1220_equation_0 = const()[name = tensor("op_1220_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1220_cast_fp16 = einsum(equation = var_1220_equation_0, values = (var_1034_cast_fp16, var_985_cast_fp16))[name = tensor("op_1220_cast_fp16")]; + tensor var_1221_to_fp16 = const()[name = tensor("op_1221_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_199_cast_fp16 = mul(x = var_1220_cast_fp16, y = var_1221_to_fp16)[name = tensor("aw_chunk_199_cast_fp16")]; + tensor var_1224_equation_0 = const()[name = tensor("op_1224_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1224_cast_fp16 = einsum(equation = var_1224_equation_0, values = (var_1034_cast_fp16, var_986_cast_fp16))[name = tensor("op_1224_cast_fp16")]; + tensor var_1225_to_fp16 = const()[name = tensor("op_1225_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_201_cast_fp16 = mul(x = var_1224_cast_fp16, y = var_1225_to_fp16)[name = tensor("aw_chunk_201_cast_fp16")]; + tensor var_1228_equation_0 = const()[name = tensor("op_1228_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1228_cast_fp16 = einsum(equation = var_1228_equation_0, values = (var_1034_cast_fp16, var_987_cast_fp16))[name = tensor("op_1228_cast_fp16")]; + tensor var_1229_to_fp16 = const()[name = tensor("op_1229_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_203_cast_fp16 = mul(x = var_1228_cast_fp16, y = var_1229_to_fp16)[name = tensor("aw_chunk_203_cast_fp16")]; + tensor var_1232_equation_0 = const()[name = tensor("op_1232_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1232_cast_fp16 = einsum(equation = var_1232_equation_0, values = (var_1034_cast_fp16, var_988_cast_fp16))[name = tensor("op_1232_cast_fp16")]; + tensor var_1233_to_fp16 = const()[name = tensor("op_1233_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_205_cast_fp16 = mul(x = var_1232_cast_fp16, y = var_1233_to_fp16)[name = tensor("aw_chunk_205_cast_fp16")]; + tensor var_1236_equation_0 = const()[name = tensor("op_1236_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1236_cast_fp16 = einsum(equation = var_1236_equation_0, values = (var_1034_cast_fp16, var_989_cast_fp16))[name = tensor("op_1236_cast_fp16")]; + tensor var_1237_to_fp16 = const()[name = tensor("op_1237_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_207_cast_fp16 = mul(x = var_1236_cast_fp16, y = var_1237_to_fp16)[name = tensor("aw_chunk_207_cast_fp16")]; + tensor var_1240_equation_0 = const()[name = tensor("op_1240_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1240_cast_fp16 = einsum(equation = var_1240_equation_0, values = (var_1038_cast_fp16, var_990_cast_fp16))[name = tensor("op_1240_cast_fp16")]; + tensor var_1241_to_fp16 = const()[name = tensor("op_1241_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_209_cast_fp16 = mul(x = var_1240_cast_fp16, y = var_1241_to_fp16)[name = tensor("aw_chunk_209_cast_fp16")]; + tensor var_1244_equation_0 = const()[name = tensor("op_1244_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1244_cast_fp16 = einsum(equation = var_1244_equation_0, values = (var_1038_cast_fp16, var_991_cast_fp16))[name = tensor("op_1244_cast_fp16")]; + tensor var_1245_to_fp16 = const()[name = tensor("op_1245_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_211_cast_fp16 = mul(x = var_1244_cast_fp16, y = var_1245_to_fp16)[name = tensor("aw_chunk_211_cast_fp16")]; + tensor var_1248_equation_0 = const()[name = tensor("op_1248_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1248_cast_fp16 = einsum(equation = var_1248_equation_0, values = (var_1038_cast_fp16, var_992_cast_fp16))[name = tensor("op_1248_cast_fp16")]; + tensor var_1249_to_fp16 = const()[name = tensor("op_1249_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_213_cast_fp16 = mul(x = var_1248_cast_fp16, y = var_1249_to_fp16)[name = tensor("aw_chunk_213_cast_fp16")]; + tensor var_1252_equation_0 = const()[name = tensor("op_1252_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1252_cast_fp16 = einsum(equation = var_1252_equation_0, values = (var_1038_cast_fp16, var_993_cast_fp16))[name = tensor("op_1252_cast_fp16")]; + tensor var_1253_to_fp16 = const()[name = tensor("op_1253_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_215_cast_fp16 = mul(x = var_1252_cast_fp16, y = var_1253_to_fp16)[name = tensor("aw_chunk_215_cast_fp16")]; + tensor var_1256_equation_0 = const()[name = tensor("op_1256_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1256_cast_fp16 = einsum(equation = var_1256_equation_0, values = (var_1038_cast_fp16, var_994_cast_fp16))[name = tensor("op_1256_cast_fp16")]; + tensor var_1257_to_fp16 = const()[name = tensor("op_1257_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_217_cast_fp16 = mul(x = var_1256_cast_fp16, y = var_1257_to_fp16)[name = tensor("aw_chunk_217_cast_fp16")]; + tensor var_1260_equation_0 = const()[name = tensor("op_1260_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1260_cast_fp16 = einsum(equation = var_1260_equation_0, values = (var_1038_cast_fp16, var_995_cast_fp16))[name = tensor("op_1260_cast_fp16")]; + tensor var_1261_to_fp16 = const()[name = tensor("op_1261_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_219_cast_fp16 = mul(x = var_1260_cast_fp16, y = var_1261_to_fp16)[name = tensor("aw_chunk_219_cast_fp16")]; + tensor var_1264_equation_0 = const()[name = tensor("op_1264_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1264_cast_fp16 = einsum(equation = var_1264_equation_0, values = (var_1038_cast_fp16, var_996_cast_fp16))[name = tensor("op_1264_cast_fp16")]; + tensor var_1265_to_fp16 = const()[name = tensor("op_1265_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_221_cast_fp16 = mul(x = var_1264_cast_fp16, y = var_1265_to_fp16)[name = tensor("aw_chunk_221_cast_fp16")]; + tensor var_1268_equation_0 = const()[name = tensor("op_1268_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1268_cast_fp16 = einsum(equation = var_1268_equation_0, values = (var_1038_cast_fp16, var_997_cast_fp16))[name = tensor("op_1268_cast_fp16")]; + tensor var_1269_to_fp16 = const()[name = tensor("op_1269_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_223_cast_fp16 = mul(x = var_1268_cast_fp16, y = var_1269_to_fp16)[name = tensor("aw_chunk_223_cast_fp16")]; + tensor var_1272_equation_0 = const()[name = tensor("op_1272_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1272_cast_fp16 = einsum(equation = var_1272_equation_0, values = (var_1042_cast_fp16, var_998_cast_fp16))[name = tensor("op_1272_cast_fp16")]; + tensor var_1273_to_fp16 = const()[name = tensor("op_1273_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_225_cast_fp16 = mul(x = var_1272_cast_fp16, y = var_1273_to_fp16)[name = tensor("aw_chunk_225_cast_fp16")]; + tensor var_1276_equation_0 = const()[name = tensor("op_1276_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1276_cast_fp16 = einsum(equation = var_1276_equation_0, values = (var_1042_cast_fp16, var_999_cast_fp16))[name = tensor("op_1276_cast_fp16")]; + tensor var_1277_to_fp16 = const()[name = tensor("op_1277_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_227_cast_fp16 = mul(x = var_1276_cast_fp16, y = var_1277_to_fp16)[name = tensor("aw_chunk_227_cast_fp16")]; + tensor var_1280_equation_0 = const()[name = tensor("op_1280_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1280_cast_fp16 = einsum(equation = var_1280_equation_0, values = (var_1042_cast_fp16, var_1000_cast_fp16))[name = tensor("op_1280_cast_fp16")]; + tensor var_1281_to_fp16 = const()[name = tensor("op_1281_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_229_cast_fp16 = mul(x = var_1280_cast_fp16, y = var_1281_to_fp16)[name = tensor("aw_chunk_229_cast_fp16")]; + tensor var_1284_equation_0 = const()[name = tensor("op_1284_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1284_cast_fp16 = einsum(equation = var_1284_equation_0, values = (var_1042_cast_fp16, var_1001_cast_fp16))[name = tensor("op_1284_cast_fp16")]; + tensor var_1285_to_fp16 = const()[name = tensor("op_1285_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_231_cast_fp16 = mul(x = var_1284_cast_fp16, y = var_1285_to_fp16)[name = tensor("aw_chunk_231_cast_fp16")]; + tensor var_1288_equation_0 = const()[name = tensor("op_1288_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1288_cast_fp16 = einsum(equation = var_1288_equation_0, values = (var_1042_cast_fp16, var_1002_cast_fp16))[name = tensor("op_1288_cast_fp16")]; + tensor var_1289_to_fp16 = const()[name = tensor("op_1289_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_233_cast_fp16 = mul(x = var_1288_cast_fp16, y = var_1289_to_fp16)[name = tensor("aw_chunk_233_cast_fp16")]; + tensor var_1292_equation_0 = const()[name = tensor("op_1292_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1292_cast_fp16 = einsum(equation = var_1292_equation_0, values = (var_1042_cast_fp16, var_1003_cast_fp16))[name = tensor("op_1292_cast_fp16")]; + tensor var_1293_to_fp16 = const()[name = tensor("op_1293_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_235_cast_fp16 = mul(x = var_1292_cast_fp16, y = var_1293_to_fp16)[name = tensor("aw_chunk_235_cast_fp16")]; + tensor var_1296_equation_0 = const()[name = tensor("op_1296_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1296_cast_fp16 = einsum(equation = var_1296_equation_0, values = (var_1042_cast_fp16, var_1004_cast_fp16))[name = tensor("op_1296_cast_fp16")]; + tensor var_1297_to_fp16 = const()[name = tensor("op_1297_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_237_cast_fp16 = mul(x = var_1296_cast_fp16, y = var_1297_to_fp16)[name = tensor("aw_chunk_237_cast_fp16")]; + tensor var_1300_equation_0 = const()[name = tensor("op_1300_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1300_cast_fp16 = einsum(equation = var_1300_equation_0, values = (var_1042_cast_fp16, var_1005_cast_fp16))[name = tensor("op_1300_cast_fp16")]; + tensor var_1301_to_fp16 = const()[name = tensor("op_1301_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_239_cast_fp16 = mul(x = var_1300_cast_fp16, y = var_1301_to_fp16)[name = tensor("aw_chunk_239_cast_fp16")]; + tensor var_1304_equation_0 = const()[name = tensor("op_1304_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1304_cast_fp16 = einsum(equation = var_1304_equation_0, values = (var_1046_cast_fp16, var_1006_cast_fp16))[name = tensor("op_1304_cast_fp16")]; + tensor var_1305_to_fp16 = const()[name = tensor("op_1305_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_241_cast_fp16 = mul(x = var_1304_cast_fp16, y = var_1305_to_fp16)[name = tensor("aw_chunk_241_cast_fp16")]; + tensor var_1308_equation_0 = const()[name = tensor("op_1308_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1308_cast_fp16 = einsum(equation = var_1308_equation_0, values = (var_1046_cast_fp16, var_1007_cast_fp16))[name = tensor("op_1308_cast_fp16")]; + tensor var_1309_to_fp16 = const()[name = tensor("op_1309_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_243_cast_fp16 = mul(x = var_1308_cast_fp16, y = var_1309_to_fp16)[name = tensor("aw_chunk_243_cast_fp16")]; + tensor var_1312_equation_0 = const()[name = tensor("op_1312_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1312_cast_fp16 = einsum(equation = var_1312_equation_0, values = (var_1046_cast_fp16, var_1008_cast_fp16))[name = tensor("op_1312_cast_fp16")]; + tensor var_1313_to_fp16 = const()[name = tensor("op_1313_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_245_cast_fp16 = mul(x = var_1312_cast_fp16, y = var_1313_to_fp16)[name = tensor("aw_chunk_245_cast_fp16")]; + tensor var_1316_equation_0 = const()[name = tensor("op_1316_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1316_cast_fp16 = einsum(equation = var_1316_equation_0, values = (var_1046_cast_fp16, var_1009_cast_fp16))[name = tensor("op_1316_cast_fp16")]; + tensor var_1317_to_fp16 = const()[name = tensor("op_1317_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_247_cast_fp16 = mul(x = var_1316_cast_fp16, y = var_1317_to_fp16)[name = tensor("aw_chunk_247_cast_fp16")]; + tensor var_1320_equation_0 = const()[name = tensor("op_1320_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1320_cast_fp16 = einsum(equation = var_1320_equation_0, values = (var_1046_cast_fp16, var_1010_cast_fp16))[name = tensor("op_1320_cast_fp16")]; + tensor var_1321_to_fp16 = const()[name = tensor("op_1321_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_249_cast_fp16 = mul(x = var_1320_cast_fp16, y = var_1321_to_fp16)[name = tensor("aw_chunk_249_cast_fp16")]; + tensor var_1324_equation_0 = const()[name = tensor("op_1324_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1324_cast_fp16 = einsum(equation = var_1324_equation_0, values = (var_1046_cast_fp16, var_1011_cast_fp16))[name = tensor("op_1324_cast_fp16")]; + tensor var_1325_to_fp16 = const()[name = tensor("op_1325_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_251_cast_fp16 = mul(x = var_1324_cast_fp16, y = var_1325_to_fp16)[name = tensor("aw_chunk_251_cast_fp16")]; + tensor var_1328_equation_0 = const()[name = tensor("op_1328_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1328_cast_fp16 = einsum(equation = var_1328_equation_0, values = (var_1046_cast_fp16, var_1012_cast_fp16))[name = tensor("op_1328_cast_fp16")]; + tensor var_1329_to_fp16 = const()[name = tensor("op_1329_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_253_cast_fp16 = mul(x = var_1328_cast_fp16, y = var_1329_to_fp16)[name = tensor("aw_chunk_253_cast_fp16")]; + tensor var_1332_equation_0 = const()[name = tensor("op_1332_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1332_cast_fp16 = einsum(equation = var_1332_equation_0, values = (var_1046_cast_fp16, var_1013_cast_fp16))[name = tensor("op_1332_cast_fp16")]; + tensor var_1333_to_fp16 = const()[name = tensor("op_1333_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_255_cast_fp16 = mul(x = var_1332_cast_fp16, y = var_1333_to_fp16)[name = tensor("aw_chunk_255_cast_fp16")]; + tensor var_1335_cast_fp16 = softmax(axis = var_123, x = aw_chunk_129_cast_fp16)[name = tensor("op_1335_cast_fp16")]; + tensor var_1336_cast_fp16 = softmax(axis = var_123, x = aw_chunk_131_cast_fp16)[name = tensor("op_1336_cast_fp16")]; + tensor var_1337_cast_fp16 = softmax(axis = var_123, x = aw_chunk_133_cast_fp16)[name = tensor("op_1337_cast_fp16")]; + tensor var_1338_cast_fp16 = softmax(axis = var_123, x = aw_chunk_135_cast_fp16)[name = tensor("op_1338_cast_fp16")]; + tensor var_1339_cast_fp16 = softmax(axis = var_123, x = aw_chunk_137_cast_fp16)[name = tensor("op_1339_cast_fp16")]; + tensor var_1340_cast_fp16 = softmax(axis = var_123, x = aw_chunk_139_cast_fp16)[name = tensor("op_1340_cast_fp16")]; + tensor var_1341_cast_fp16 = softmax(axis = var_123, x = aw_chunk_141_cast_fp16)[name = tensor("op_1341_cast_fp16")]; + tensor var_1342_cast_fp16 = softmax(axis = var_123, x = aw_chunk_143_cast_fp16)[name = tensor("op_1342_cast_fp16")]; + tensor var_1343_cast_fp16 = softmax(axis = var_123, x = aw_chunk_145_cast_fp16)[name = tensor("op_1343_cast_fp16")]; + tensor var_1344_cast_fp16 = softmax(axis = var_123, x = aw_chunk_147_cast_fp16)[name = tensor("op_1344_cast_fp16")]; + tensor var_1345_cast_fp16 = softmax(axis = var_123, x = aw_chunk_149_cast_fp16)[name = tensor("op_1345_cast_fp16")]; + tensor var_1346_cast_fp16 = softmax(axis = var_123, x = aw_chunk_151_cast_fp16)[name = tensor("op_1346_cast_fp16")]; + tensor var_1347_cast_fp16 = softmax(axis = var_123, x = aw_chunk_153_cast_fp16)[name = tensor("op_1347_cast_fp16")]; + tensor var_1348_cast_fp16 = softmax(axis = var_123, x = aw_chunk_155_cast_fp16)[name = tensor("op_1348_cast_fp16")]; + tensor var_1349_cast_fp16 = softmax(axis = var_123, x = aw_chunk_157_cast_fp16)[name = tensor("op_1349_cast_fp16")]; + tensor var_1350_cast_fp16 = softmax(axis = var_123, x = aw_chunk_159_cast_fp16)[name = tensor("op_1350_cast_fp16")]; + tensor var_1351_cast_fp16 = softmax(axis = var_123, x = aw_chunk_161_cast_fp16)[name = tensor("op_1351_cast_fp16")]; + tensor var_1352_cast_fp16 = softmax(axis = var_123, x = aw_chunk_163_cast_fp16)[name = tensor("op_1352_cast_fp16")]; + tensor var_1353_cast_fp16 = softmax(axis = var_123, x = aw_chunk_165_cast_fp16)[name = tensor("op_1353_cast_fp16")]; + tensor var_1354_cast_fp16 = softmax(axis = var_123, x = aw_chunk_167_cast_fp16)[name = tensor("op_1354_cast_fp16")]; + tensor var_1355_cast_fp16 = softmax(axis = var_123, x = aw_chunk_169_cast_fp16)[name = tensor("op_1355_cast_fp16")]; + tensor var_1356_cast_fp16 = softmax(axis = var_123, x = aw_chunk_171_cast_fp16)[name = tensor("op_1356_cast_fp16")]; + tensor var_1357_cast_fp16 = softmax(axis = var_123, x = aw_chunk_173_cast_fp16)[name = tensor("op_1357_cast_fp16")]; + tensor var_1358_cast_fp16 = softmax(axis = var_123, x = aw_chunk_175_cast_fp16)[name = tensor("op_1358_cast_fp16")]; + tensor var_1359_cast_fp16 = softmax(axis = var_123, x = aw_chunk_177_cast_fp16)[name = tensor("op_1359_cast_fp16")]; + tensor var_1360_cast_fp16 = softmax(axis = var_123, x = aw_chunk_179_cast_fp16)[name = tensor("op_1360_cast_fp16")]; + tensor var_1361_cast_fp16 = softmax(axis = var_123, x = aw_chunk_181_cast_fp16)[name = tensor("op_1361_cast_fp16")]; + tensor var_1362_cast_fp16 = softmax(axis = var_123, x = aw_chunk_183_cast_fp16)[name = tensor("op_1362_cast_fp16")]; + tensor var_1363_cast_fp16 = softmax(axis = var_123, x = aw_chunk_185_cast_fp16)[name = tensor("op_1363_cast_fp16")]; + tensor var_1364_cast_fp16 = softmax(axis = var_123, x = aw_chunk_187_cast_fp16)[name = tensor("op_1364_cast_fp16")]; + tensor var_1365_cast_fp16 = softmax(axis = var_123, x = aw_chunk_189_cast_fp16)[name = tensor("op_1365_cast_fp16")]; + tensor var_1366_cast_fp16 = softmax(axis = var_123, x = aw_chunk_191_cast_fp16)[name = tensor("op_1366_cast_fp16")]; + tensor var_1367_cast_fp16 = softmax(axis = var_123, x = aw_chunk_193_cast_fp16)[name = tensor("op_1367_cast_fp16")]; + tensor var_1368_cast_fp16 = softmax(axis = var_123, x = aw_chunk_195_cast_fp16)[name = tensor("op_1368_cast_fp16")]; + tensor var_1369_cast_fp16 = softmax(axis = var_123, x = aw_chunk_197_cast_fp16)[name = tensor("op_1369_cast_fp16")]; + tensor var_1370_cast_fp16 = softmax(axis = var_123, x = aw_chunk_199_cast_fp16)[name = tensor("op_1370_cast_fp16")]; + tensor var_1371_cast_fp16 = softmax(axis = var_123, x = aw_chunk_201_cast_fp16)[name = tensor("op_1371_cast_fp16")]; + tensor var_1372_cast_fp16 = softmax(axis = var_123, x = aw_chunk_203_cast_fp16)[name = tensor("op_1372_cast_fp16")]; + tensor var_1373_cast_fp16 = softmax(axis = var_123, x = aw_chunk_205_cast_fp16)[name = tensor("op_1373_cast_fp16")]; + tensor var_1374_cast_fp16 = softmax(axis = var_123, x = aw_chunk_207_cast_fp16)[name = tensor("op_1374_cast_fp16")]; + tensor var_1375_cast_fp16 = softmax(axis = var_123, x = aw_chunk_209_cast_fp16)[name = tensor("op_1375_cast_fp16")]; + tensor var_1376_cast_fp16 = softmax(axis = var_123, x = aw_chunk_211_cast_fp16)[name = tensor("op_1376_cast_fp16")]; + tensor var_1377_cast_fp16 = softmax(axis = var_123, x = aw_chunk_213_cast_fp16)[name = tensor("op_1377_cast_fp16")]; + tensor var_1378_cast_fp16 = softmax(axis = var_123, x = aw_chunk_215_cast_fp16)[name = tensor("op_1378_cast_fp16")]; + tensor var_1379_cast_fp16 = softmax(axis = var_123, x = aw_chunk_217_cast_fp16)[name = tensor("op_1379_cast_fp16")]; + tensor var_1380_cast_fp16 = softmax(axis = var_123, x = aw_chunk_219_cast_fp16)[name = tensor("op_1380_cast_fp16")]; + tensor var_1381_cast_fp16 = softmax(axis = var_123, x = aw_chunk_221_cast_fp16)[name = tensor("op_1381_cast_fp16")]; + tensor var_1382_cast_fp16 = softmax(axis = var_123, x = aw_chunk_223_cast_fp16)[name = tensor("op_1382_cast_fp16")]; + tensor var_1383_cast_fp16 = softmax(axis = var_123, x = aw_chunk_225_cast_fp16)[name = tensor("op_1383_cast_fp16")]; + tensor var_1384_cast_fp16 = softmax(axis = var_123, x = aw_chunk_227_cast_fp16)[name = tensor("op_1384_cast_fp16")]; + tensor var_1385_cast_fp16 = softmax(axis = var_123, x = aw_chunk_229_cast_fp16)[name = tensor("op_1385_cast_fp16")]; + tensor var_1386_cast_fp16 = softmax(axis = var_123, x = aw_chunk_231_cast_fp16)[name = tensor("op_1386_cast_fp16")]; + tensor var_1387_cast_fp16 = softmax(axis = var_123, x = aw_chunk_233_cast_fp16)[name = tensor("op_1387_cast_fp16")]; + tensor var_1388_cast_fp16 = softmax(axis = var_123, x = aw_chunk_235_cast_fp16)[name = tensor("op_1388_cast_fp16")]; + tensor var_1389_cast_fp16 = softmax(axis = var_123, x = aw_chunk_237_cast_fp16)[name = tensor("op_1389_cast_fp16")]; + tensor var_1390_cast_fp16 = softmax(axis = var_123, x = aw_chunk_239_cast_fp16)[name = tensor("op_1390_cast_fp16")]; + tensor var_1391_cast_fp16 = softmax(axis = var_123, x = aw_chunk_241_cast_fp16)[name = tensor("op_1391_cast_fp16")]; + tensor var_1392_cast_fp16 = softmax(axis = var_123, x = aw_chunk_243_cast_fp16)[name = tensor("op_1392_cast_fp16")]; + tensor var_1393_cast_fp16 = softmax(axis = var_123, x = aw_chunk_245_cast_fp16)[name = tensor("op_1393_cast_fp16")]; + tensor var_1394_cast_fp16 = softmax(axis = var_123, x = aw_chunk_247_cast_fp16)[name = tensor("op_1394_cast_fp16")]; + tensor var_1395_cast_fp16 = softmax(axis = var_123, x = aw_chunk_249_cast_fp16)[name = tensor("op_1395_cast_fp16")]; + tensor var_1396_cast_fp16 = softmax(axis = var_123, x = aw_chunk_251_cast_fp16)[name = tensor("op_1396_cast_fp16")]; + tensor var_1397_cast_fp16 = softmax(axis = var_123, x = aw_chunk_253_cast_fp16)[name = tensor("op_1397_cast_fp16")]; + tensor var_1398_cast_fp16 = softmax(axis = var_123, x = aw_chunk_255_cast_fp16)[name = tensor("op_1398_cast_fp16")]; + tensor var_1400_equation_0 = const()[name = tensor("op_1400_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1400_cast_fp16 = einsum(equation = var_1400_equation_0, values = (var_1048_cast_fp16, var_1335_cast_fp16))[name = tensor("op_1400_cast_fp16")]; + tensor var_1402_equation_0 = const()[name = tensor("op_1402_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1402_cast_fp16 = einsum(equation = var_1402_equation_0, values = (var_1048_cast_fp16, var_1336_cast_fp16))[name = tensor("op_1402_cast_fp16")]; + tensor var_1404_equation_0 = const()[name = tensor("op_1404_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1404_cast_fp16 = einsum(equation = var_1404_equation_0, values = (var_1048_cast_fp16, var_1337_cast_fp16))[name = tensor("op_1404_cast_fp16")]; + tensor var_1406_equation_0 = const()[name = tensor("op_1406_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1406_cast_fp16 = einsum(equation = var_1406_equation_0, values = (var_1048_cast_fp16, var_1338_cast_fp16))[name = tensor("op_1406_cast_fp16")]; + tensor var_1408_equation_0 = const()[name = tensor("op_1408_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1408_cast_fp16 = einsum(equation = var_1408_equation_0, values = (var_1048_cast_fp16, var_1339_cast_fp16))[name = tensor("op_1408_cast_fp16")]; + tensor var_1410_equation_0 = const()[name = tensor("op_1410_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1410_cast_fp16 = einsum(equation = var_1410_equation_0, values = (var_1048_cast_fp16, var_1340_cast_fp16))[name = tensor("op_1410_cast_fp16")]; + tensor var_1412_equation_0 = const()[name = tensor("op_1412_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1412_cast_fp16 = einsum(equation = var_1412_equation_0, values = (var_1048_cast_fp16, var_1341_cast_fp16))[name = tensor("op_1412_cast_fp16")]; + tensor var_1414_equation_0 = const()[name = tensor("op_1414_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1414_cast_fp16 = einsum(equation = var_1414_equation_0, values = (var_1048_cast_fp16, var_1342_cast_fp16))[name = tensor("op_1414_cast_fp16")]; + tensor var_1416_equation_0 = const()[name = tensor("op_1416_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1416_cast_fp16 = einsum(equation = var_1416_equation_0, values = (var_1052_cast_fp16, var_1343_cast_fp16))[name = tensor("op_1416_cast_fp16")]; + tensor var_1418_equation_0 = const()[name = tensor("op_1418_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1418_cast_fp16 = einsum(equation = var_1418_equation_0, values = (var_1052_cast_fp16, var_1344_cast_fp16))[name = tensor("op_1418_cast_fp16")]; + tensor var_1420_equation_0 = const()[name = tensor("op_1420_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1420_cast_fp16 = einsum(equation = var_1420_equation_0, values = (var_1052_cast_fp16, var_1345_cast_fp16))[name = tensor("op_1420_cast_fp16")]; + tensor var_1422_equation_0 = const()[name = tensor("op_1422_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1422_cast_fp16 = einsum(equation = var_1422_equation_0, values = (var_1052_cast_fp16, var_1346_cast_fp16))[name = tensor("op_1422_cast_fp16")]; + tensor var_1424_equation_0 = const()[name = tensor("op_1424_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1424_cast_fp16 = einsum(equation = var_1424_equation_0, values = (var_1052_cast_fp16, var_1347_cast_fp16))[name = tensor("op_1424_cast_fp16")]; + tensor var_1426_equation_0 = const()[name = tensor("op_1426_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1426_cast_fp16 = einsum(equation = var_1426_equation_0, values = (var_1052_cast_fp16, var_1348_cast_fp16))[name = tensor("op_1426_cast_fp16")]; + tensor var_1428_equation_0 = const()[name = tensor("op_1428_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1428_cast_fp16 = einsum(equation = var_1428_equation_0, values = (var_1052_cast_fp16, var_1349_cast_fp16))[name = tensor("op_1428_cast_fp16")]; + tensor var_1430_equation_0 = const()[name = tensor("op_1430_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1430_cast_fp16 = einsum(equation = var_1430_equation_0, values = (var_1052_cast_fp16, var_1350_cast_fp16))[name = tensor("op_1430_cast_fp16")]; + tensor var_1432_equation_0 = const()[name = tensor("op_1432_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1432_cast_fp16 = einsum(equation = var_1432_equation_0, values = (var_1056_cast_fp16, var_1351_cast_fp16))[name = tensor("op_1432_cast_fp16")]; + tensor var_1434_equation_0 = const()[name = tensor("op_1434_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1434_cast_fp16 = einsum(equation = var_1434_equation_0, values = (var_1056_cast_fp16, var_1352_cast_fp16))[name = tensor("op_1434_cast_fp16")]; + tensor var_1436_equation_0 = const()[name = tensor("op_1436_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1436_cast_fp16 = einsum(equation = var_1436_equation_0, values = (var_1056_cast_fp16, var_1353_cast_fp16))[name = tensor("op_1436_cast_fp16")]; + tensor var_1438_equation_0 = const()[name = tensor("op_1438_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1438_cast_fp16 = einsum(equation = var_1438_equation_0, values = (var_1056_cast_fp16, var_1354_cast_fp16))[name = tensor("op_1438_cast_fp16")]; + tensor var_1440_equation_0 = const()[name = tensor("op_1440_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1440_cast_fp16 = einsum(equation = var_1440_equation_0, values = (var_1056_cast_fp16, var_1355_cast_fp16))[name = tensor("op_1440_cast_fp16")]; + tensor var_1442_equation_0 = const()[name = tensor("op_1442_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1442_cast_fp16 = einsum(equation = var_1442_equation_0, values = (var_1056_cast_fp16, var_1356_cast_fp16))[name = tensor("op_1442_cast_fp16")]; + tensor var_1444_equation_0 = const()[name = tensor("op_1444_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1444_cast_fp16 = einsum(equation = var_1444_equation_0, values = (var_1056_cast_fp16, var_1357_cast_fp16))[name = tensor("op_1444_cast_fp16")]; + tensor var_1446_equation_0 = const()[name = tensor("op_1446_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1446_cast_fp16 = einsum(equation = var_1446_equation_0, values = (var_1056_cast_fp16, var_1358_cast_fp16))[name = tensor("op_1446_cast_fp16")]; + tensor var_1448_equation_0 = const()[name = tensor("op_1448_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1448_cast_fp16 = einsum(equation = var_1448_equation_0, values = (var_1060_cast_fp16, var_1359_cast_fp16))[name = tensor("op_1448_cast_fp16")]; + tensor var_1450_equation_0 = const()[name = tensor("op_1450_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1450_cast_fp16 = einsum(equation = var_1450_equation_0, values = (var_1060_cast_fp16, var_1360_cast_fp16))[name = tensor("op_1450_cast_fp16")]; + tensor var_1452_equation_0 = const()[name = tensor("op_1452_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1452_cast_fp16 = einsum(equation = var_1452_equation_0, values = (var_1060_cast_fp16, var_1361_cast_fp16))[name = tensor("op_1452_cast_fp16")]; + tensor var_1454_equation_0 = const()[name = tensor("op_1454_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1454_cast_fp16 = einsum(equation = var_1454_equation_0, values = (var_1060_cast_fp16, var_1362_cast_fp16))[name = tensor("op_1454_cast_fp16")]; + tensor var_1456_equation_0 = const()[name = tensor("op_1456_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1456_cast_fp16 = einsum(equation = var_1456_equation_0, values = (var_1060_cast_fp16, var_1363_cast_fp16))[name = tensor("op_1456_cast_fp16")]; + tensor var_1458_equation_0 = const()[name = tensor("op_1458_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1458_cast_fp16 = einsum(equation = var_1458_equation_0, values = (var_1060_cast_fp16, var_1364_cast_fp16))[name = tensor("op_1458_cast_fp16")]; + tensor var_1460_equation_0 = const()[name = tensor("op_1460_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1460_cast_fp16 = einsum(equation = var_1460_equation_0, values = (var_1060_cast_fp16, var_1365_cast_fp16))[name = tensor("op_1460_cast_fp16")]; + tensor var_1462_equation_0 = const()[name = tensor("op_1462_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1462_cast_fp16 = einsum(equation = var_1462_equation_0, values = (var_1060_cast_fp16, var_1366_cast_fp16))[name = tensor("op_1462_cast_fp16")]; + tensor var_1464_equation_0 = const()[name = tensor("op_1464_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1464_cast_fp16 = einsum(equation = var_1464_equation_0, values = (var_1064_cast_fp16, var_1367_cast_fp16))[name = tensor("op_1464_cast_fp16")]; + tensor var_1466_equation_0 = const()[name = tensor("op_1466_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1466_cast_fp16 = einsum(equation = var_1466_equation_0, values = (var_1064_cast_fp16, var_1368_cast_fp16))[name = tensor("op_1466_cast_fp16")]; + tensor var_1468_equation_0 = const()[name = tensor("op_1468_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1468_cast_fp16 = einsum(equation = var_1468_equation_0, values = (var_1064_cast_fp16, var_1369_cast_fp16))[name = tensor("op_1468_cast_fp16")]; + tensor var_1470_equation_0 = const()[name = tensor("op_1470_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1470_cast_fp16 = einsum(equation = var_1470_equation_0, values = (var_1064_cast_fp16, var_1370_cast_fp16))[name = tensor("op_1470_cast_fp16")]; + tensor var_1472_equation_0 = const()[name = tensor("op_1472_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1472_cast_fp16 = einsum(equation = var_1472_equation_0, values = (var_1064_cast_fp16, var_1371_cast_fp16))[name = tensor("op_1472_cast_fp16")]; + tensor var_1474_equation_0 = const()[name = tensor("op_1474_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1474_cast_fp16 = einsum(equation = var_1474_equation_0, values = (var_1064_cast_fp16, var_1372_cast_fp16))[name = tensor("op_1474_cast_fp16")]; + tensor var_1476_equation_0 = const()[name = tensor("op_1476_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1476_cast_fp16 = einsum(equation = var_1476_equation_0, values = (var_1064_cast_fp16, var_1373_cast_fp16))[name = tensor("op_1476_cast_fp16")]; + tensor var_1478_equation_0 = const()[name = tensor("op_1478_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1478_cast_fp16 = einsum(equation = var_1478_equation_0, values = (var_1064_cast_fp16, var_1374_cast_fp16))[name = tensor("op_1478_cast_fp16")]; + tensor var_1480_equation_0 = const()[name = tensor("op_1480_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1480_cast_fp16 = einsum(equation = var_1480_equation_0, values = (var_1068_cast_fp16, var_1375_cast_fp16))[name = tensor("op_1480_cast_fp16")]; + tensor var_1482_equation_0 = const()[name = tensor("op_1482_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1482_cast_fp16 = einsum(equation = var_1482_equation_0, values = (var_1068_cast_fp16, var_1376_cast_fp16))[name = tensor("op_1482_cast_fp16")]; + tensor var_1484_equation_0 = const()[name = tensor("op_1484_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1484_cast_fp16 = einsum(equation = var_1484_equation_0, values = (var_1068_cast_fp16, var_1377_cast_fp16))[name = tensor("op_1484_cast_fp16")]; + tensor var_1486_equation_0 = const()[name = tensor("op_1486_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1486_cast_fp16 = einsum(equation = var_1486_equation_0, values = (var_1068_cast_fp16, var_1378_cast_fp16))[name = tensor("op_1486_cast_fp16")]; + tensor var_1488_equation_0 = const()[name = tensor("op_1488_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1488_cast_fp16 = einsum(equation = var_1488_equation_0, values = (var_1068_cast_fp16, var_1379_cast_fp16))[name = tensor("op_1488_cast_fp16")]; + tensor var_1490_equation_0 = const()[name = tensor("op_1490_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1490_cast_fp16 = einsum(equation = var_1490_equation_0, values = (var_1068_cast_fp16, var_1380_cast_fp16))[name = tensor("op_1490_cast_fp16")]; + tensor var_1492_equation_0 = const()[name = tensor("op_1492_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1492_cast_fp16 = einsum(equation = var_1492_equation_0, values = (var_1068_cast_fp16, var_1381_cast_fp16))[name = tensor("op_1492_cast_fp16")]; + tensor var_1494_equation_0 = const()[name = tensor("op_1494_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1494_cast_fp16 = einsum(equation = var_1494_equation_0, values = (var_1068_cast_fp16, var_1382_cast_fp16))[name = tensor("op_1494_cast_fp16")]; + tensor var_1496_equation_0 = const()[name = tensor("op_1496_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1496_cast_fp16 = einsum(equation = var_1496_equation_0, values = (var_1072_cast_fp16, var_1383_cast_fp16))[name = tensor("op_1496_cast_fp16")]; + tensor var_1498_equation_0 = const()[name = tensor("op_1498_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1498_cast_fp16 = einsum(equation = var_1498_equation_0, values = (var_1072_cast_fp16, var_1384_cast_fp16))[name = tensor("op_1498_cast_fp16")]; + tensor var_1500_equation_0 = const()[name = tensor("op_1500_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1500_cast_fp16 = einsum(equation = var_1500_equation_0, values = (var_1072_cast_fp16, var_1385_cast_fp16))[name = tensor("op_1500_cast_fp16")]; + tensor var_1502_equation_0 = const()[name = tensor("op_1502_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1502_cast_fp16 = einsum(equation = var_1502_equation_0, values = (var_1072_cast_fp16, var_1386_cast_fp16))[name = tensor("op_1502_cast_fp16")]; + tensor var_1504_equation_0 = const()[name = tensor("op_1504_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1504_cast_fp16 = einsum(equation = var_1504_equation_0, values = (var_1072_cast_fp16, var_1387_cast_fp16))[name = tensor("op_1504_cast_fp16")]; + tensor var_1506_equation_0 = const()[name = tensor("op_1506_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1506_cast_fp16 = einsum(equation = var_1506_equation_0, values = (var_1072_cast_fp16, var_1388_cast_fp16))[name = tensor("op_1506_cast_fp16")]; + tensor var_1508_equation_0 = const()[name = tensor("op_1508_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1508_cast_fp16 = einsum(equation = var_1508_equation_0, values = (var_1072_cast_fp16, var_1389_cast_fp16))[name = tensor("op_1508_cast_fp16")]; + tensor var_1510_equation_0 = const()[name = tensor("op_1510_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1510_cast_fp16 = einsum(equation = var_1510_equation_0, values = (var_1072_cast_fp16, var_1390_cast_fp16))[name = tensor("op_1510_cast_fp16")]; + tensor var_1512_equation_0 = const()[name = tensor("op_1512_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1512_cast_fp16 = einsum(equation = var_1512_equation_0, values = (var_1076_cast_fp16, var_1391_cast_fp16))[name = tensor("op_1512_cast_fp16")]; + tensor var_1514_equation_0 = const()[name = tensor("op_1514_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1514_cast_fp16 = einsum(equation = var_1514_equation_0, values = (var_1076_cast_fp16, var_1392_cast_fp16))[name = tensor("op_1514_cast_fp16")]; + tensor var_1516_equation_0 = const()[name = tensor("op_1516_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1516_cast_fp16 = einsum(equation = var_1516_equation_0, values = (var_1076_cast_fp16, var_1393_cast_fp16))[name = tensor("op_1516_cast_fp16")]; + tensor var_1518_equation_0 = const()[name = tensor("op_1518_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1518_cast_fp16 = einsum(equation = var_1518_equation_0, values = (var_1076_cast_fp16, var_1394_cast_fp16))[name = tensor("op_1518_cast_fp16")]; + tensor var_1520_equation_0 = const()[name = tensor("op_1520_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1520_cast_fp16 = einsum(equation = var_1520_equation_0, values = (var_1076_cast_fp16, var_1395_cast_fp16))[name = tensor("op_1520_cast_fp16")]; + tensor var_1522_equation_0 = const()[name = tensor("op_1522_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1522_cast_fp16 = einsum(equation = var_1522_equation_0, values = (var_1076_cast_fp16, var_1396_cast_fp16))[name = tensor("op_1522_cast_fp16")]; + tensor var_1524_equation_0 = const()[name = tensor("op_1524_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1524_cast_fp16 = einsum(equation = var_1524_equation_0, values = (var_1076_cast_fp16, var_1397_cast_fp16))[name = tensor("op_1524_cast_fp16")]; + tensor var_1526_equation_0 = const()[name = tensor("op_1526_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1526_cast_fp16 = einsum(equation = var_1526_equation_0, values = (var_1076_cast_fp16, var_1398_cast_fp16))[name = tensor("op_1526_cast_fp16")]; + tensor var_1528_interleave_0 = const()[name = tensor("op_1528_interleave_0"), val = tensor(false)]; + tensor var_1528_cast_fp16 = concat(axis = var_95, interleave = var_1528_interleave_0, values = (var_1400_cast_fp16, var_1402_cast_fp16, var_1404_cast_fp16, var_1406_cast_fp16, var_1408_cast_fp16, var_1410_cast_fp16, var_1412_cast_fp16, var_1414_cast_fp16))[name = tensor("op_1528_cast_fp16")]; + tensor var_1530_interleave_0 = const()[name = tensor("op_1530_interleave_0"), val = tensor(false)]; + tensor var_1530_cast_fp16 = concat(axis = var_95, interleave = var_1530_interleave_0, values = (var_1416_cast_fp16, var_1418_cast_fp16, var_1420_cast_fp16, var_1422_cast_fp16, var_1424_cast_fp16, var_1426_cast_fp16, var_1428_cast_fp16, var_1430_cast_fp16))[name = tensor("op_1530_cast_fp16")]; + tensor var_1532_interleave_0 = const()[name = tensor("op_1532_interleave_0"), val = tensor(false)]; + tensor var_1532_cast_fp16 = concat(axis = var_95, interleave = var_1532_interleave_0, values = (var_1432_cast_fp16, var_1434_cast_fp16, var_1436_cast_fp16, var_1438_cast_fp16, var_1440_cast_fp16, var_1442_cast_fp16, var_1444_cast_fp16, var_1446_cast_fp16))[name = tensor("op_1532_cast_fp16")]; + tensor var_1534_interleave_0 = const()[name = tensor("op_1534_interleave_0"), val = tensor(false)]; + tensor var_1534_cast_fp16 = concat(axis = var_95, interleave = var_1534_interleave_0, values = (var_1448_cast_fp16, var_1450_cast_fp16, var_1452_cast_fp16, var_1454_cast_fp16, var_1456_cast_fp16, var_1458_cast_fp16, var_1460_cast_fp16, var_1462_cast_fp16))[name = tensor("op_1534_cast_fp16")]; + tensor var_1536_interleave_0 = const()[name = tensor("op_1536_interleave_0"), val = tensor(false)]; + tensor var_1536_cast_fp16 = concat(axis = var_95, interleave = var_1536_interleave_0, values = (var_1464_cast_fp16, var_1466_cast_fp16, var_1468_cast_fp16, var_1470_cast_fp16, var_1472_cast_fp16, var_1474_cast_fp16, var_1476_cast_fp16, var_1478_cast_fp16))[name = tensor("op_1536_cast_fp16")]; + tensor var_1538_interleave_0 = const()[name = tensor("op_1538_interleave_0"), val = tensor(false)]; + tensor var_1538_cast_fp16 = concat(axis = var_95, interleave = var_1538_interleave_0, values = (var_1480_cast_fp16, var_1482_cast_fp16, var_1484_cast_fp16, var_1486_cast_fp16, var_1488_cast_fp16, var_1490_cast_fp16, var_1492_cast_fp16, var_1494_cast_fp16))[name = tensor("op_1538_cast_fp16")]; + tensor var_1540_interleave_0 = const()[name = tensor("op_1540_interleave_0"), val = tensor(false)]; + tensor var_1540_cast_fp16 = concat(axis = var_95, interleave = var_1540_interleave_0, values = (var_1496_cast_fp16, var_1498_cast_fp16, var_1500_cast_fp16, var_1502_cast_fp16, var_1504_cast_fp16, var_1506_cast_fp16, var_1508_cast_fp16, var_1510_cast_fp16))[name = tensor("op_1540_cast_fp16")]; + tensor var_1542_interleave_0 = const()[name = tensor("op_1542_interleave_0"), val = tensor(false)]; + tensor var_1542_cast_fp16 = concat(axis = var_95, interleave = var_1542_interleave_0, values = (var_1512_cast_fp16, var_1514_cast_fp16, var_1516_cast_fp16, var_1518_cast_fp16, var_1520_cast_fp16, var_1522_cast_fp16, var_1524_cast_fp16, var_1526_cast_fp16))[name = tensor("op_1542_cast_fp16")]; + tensor input_27_interleave_0 = const()[name = tensor("input_27_interleave_0"), val = tensor(false)]; + tensor input_27_cast_fp16 = concat(axis = var_123, interleave = input_27_interleave_0, values = (var_1528_cast_fp16, var_1530_cast_fp16, var_1532_cast_fp16, var_1534_cast_fp16, var_1536_cast_fp16, var_1538_cast_fp16, var_1540_cast_fp16, var_1542_cast_fp16))[name = tensor("input_27_cast_fp16")]; + tensor var_1552_pad_type_0 = const()[name = tensor("op_1552_pad_type_0"), val = tensor("valid")]; + tensor var_1552_strides_0 = const()[name = tensor("op_1552_strides_0"), val = tensor([1, 1])]; + tensor var_1552_pad_0 = const()[name = tensor("op_1552_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1552_dilations_0 = const()[name = tensor("op_1552_dilations_0"), val = tensor([1, 1])]; + tensor var_1552_groups_0 = const()[name = tensor("op_1552_groups_0"), val = tensor(1)]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10855744)))]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11060608)))]; + tensor var_1552_cast_fp16 = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_1552_dilations_0, groups = var_1552_groups_0, pad = var_1552_pad_0, pad_type = var_1552_pad_type_0, strides = var_1552_strides_0, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_27_cast_fp16)[name = tensor("op_1552_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = var_1552_cast_fp16, y = inputs_3_cast_fp16)[name = tensor("inputs_5_cast_fp16")]; + tensor input_29_axes_0 = const()[name = tensor("input_29_axes_0"), val = tensor([1])]; + tensor input_29_gamma_0_to_fp16 = const()[name = tensor("input_29_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11061312)))]; + tensor input_29_beta_0_to_fp16 = const()[name = tensor("input_29_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11062016)))]; + tensor var_1562_to_fp16 = const()[name = tensor("op_1562_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_29_cast_fp16 = layer_norm(axes = input_29_axes_0, beta = input_29_beta_0_to_fp16, epsilon = var_1562_to_fp16, gamma = input_29_gamma_0_to_fp16, x = inputs_5_cast_fp16)[name = tensor("input_29_cast_fp16")]; + tensor var_1582_pad_type_0 = const()[name = tensor("op_1582_pad_type_0"), val = tensor("valid")]; + tensor var_1582_strides_0 = const()[name = tensor("op_1582_strides_0"), val = tensor([1, 1])]; + tensor var_1582_pad_0 = const()[name = tensor("op_1582_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1582_dilations_0 = const()[name = tensor("op_1582_dilations_0"), val = tensor([1, 1])]; + tensor var_1582_groups_0 = const()[name = tensor("op_1582_groups_0"), val = tensor(1)]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11062720)))]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12701184)))]; + tensor var_1582_cast_fp16 = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_1582_dilations_0, groups = var_1582_groups_0, pad = var_1582_pad_0, pad_type = var_1582_pad_type_0, strides = var_1582_strides_0, weight = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_29_cast_fp16)[name = tensor("op_1582_cast_fp16")]; + tensor var_1583_split_sizes_0 = const()[name = tensor("op_1583_split_sizes_0"), val = tensor([1280, 1280])]; + tensor var_1583_axis_0 = const()[name = tensor("op_1583_axis_0"), val = tensor(1)]; + tensor var_1583_cast_fp16_0, tensor var_1583_cast_fp16_1 = split(axis = var_1583_axis_0, split_sizes = var_1583_split_sizes_0, x = var_1582_cast_fp16)[name = tensor("op_1583_cast_fp16")]; + tensor var_1585_mode_0 = const()[name = tensor("op_1585_mode_0"), val = tensor("EXACT")]; + tensor var_1585_cast_fp16 = gelu(mode = var_1585_mode_0, x = var_1583_cast_fp16_1)[name = tensor("op_1585_cast_fp16")]; + tensor input_31_cast_fp16 = mul(x = var_1583_cast_fp16_0, y = var_1585_cast_fp16)[name = tensor("input_31_cast_fp16")]; + tensor var_1593_pad_type_0 = const()[name = tensor("op_1593_pad_type_0"), val = tensor("valid")]; + tensor var_1593_strides_0 = const()[name = tensor("op_1593_strides_0"), val = tensor([1, 1])]; + tensor var_1593_pad_0 = const()[name = tensor("op_1593_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1593_dilations_0 = const()[name = tensor("op_1593_dilations_0"), val = tensor([1, 1])]; + tensor var_1593_groups_0 = const()[name = tensor("op_1593_groups_0"), val = tensor(1)]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12706368)))]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13525632)))]; + tensor var_1593_cast_fp16 = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_1593_dilations_0, groups = var_1593_groups_0, pad = var_1593_pad_0, pad_type = var_1593_pad_type_0, strides = var_1593_strides_0, weight = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_31_cast_fp16)[name = tensor("op_1593_cast_fp16")]; + tensor hidden_states_15_cast_fp16 = add(x = var_1593_cast_fp16, y = inputs_5_cast_fp16)[name = tensor("hidden_states_15_cast_fp16")]; + tensor var_1595 = const()[name = tensor("op_1595"), val = tensor([2, 320, 64, 64])]; + tensor input_33_cast_fp16 = reshape(shape = var_1595, x = hidden_states_15_cast_fp16)[name = tensor("input_33_cast_fp16")]; + tensor hidden_states_17_pad_type_0 = const()[name = tensor("hidden_states_17_pad_type_0"), val = tensor("valid")]; + tensor hidden_states_17_strides_0 = const()[name = tensor("hidden_states_17_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_17_pad_0 = const()[name = tensor("hidden_states_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_17_dilations_0 = const()[name = tensor("hidden_states_17_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_17_groups_0 = const()[name = tensor("hidden_states_17_groups_0"), val = tensor(1)]; + tensor down_blocks_0_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor("down_blocks_0_attentions_0_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13526336)))]; + tensor down_blocks_0_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor("down_blocks_0_attentions_0_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13731200)))]; + tensor hidden_states_17_cast_fp16 = conv(bias = down_blocks_0_attentions_0_proj_out_bias_to_fp16, dilations = hidden_states_17_dilations_0, groups = hidden_states_17_groups_0, pad = hidden_states_17_pad_0, pad_type = hidden_states_17_pad_type_0, strides = hidden_states_17_strides_0, weight = down_blocks_0_attentions_0_proj_out_weight_to_fp16, x = input_33_cast_fp16)[name = tensor("hidden_states_17_cast_fp16")]; + tensor input_35_cast_fp16_1 = add(x = hidden_states_17_cast_fp16, y = hidden_states_5_cast_fp16)[name = tensor("input_35_cast_fp16")]; + tensor reshape_12_shape_0 = const()[name = tensor("reshape_12_shape_0"), val = tensor([2, 32, 10, 64, 64])]; + tensor reshape_12_cast_fp16 = reshape(shape = reshape_12_shape_0, x = input_35_cast_fp16_1)[name = tensor("reshape_12_cast_fp16")]; + tensor reduce_mean_9_axes_0 = const()[name = tensor("reduce_mean_9_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_9_keep_dims_0 = const()[name = tensor("reduce_mean_9_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_9_cast_fp16 = reduce_mean(axes = reduce_mean_9_axes_0, keep_dims = reduce_mean_9_keep_dims_0, x = reshape_12_cast_fp16)[name = tensor("reduce_mean_9_cast_fp16")]; + tensor sub_6_cast_fp16 = sub(x = reshape_12_cast_fp16, y = reduce_mean_9_cast_fp16)[name = tensor("sub_6_cast_fp16")]; + tensor square_3_cast_fp16 = square(x = sub_6_cast_fp16)[name = tensor("square_3_cast_fp16")]; + tensor reduce_mean_11_axes_0 = const()[name = tensor("reduce_mean_11_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_11_keep_dims_0 = const()[name = tensor("reduce_mean_11_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_11_cast_fp16 = reduce_mean(axes = reduce_mean_11_axes_0, keep_dims = reduce_mean_11_keep_dims_0, x = square_3_cast_fp16)[name = tensor("reduce_mean_11_cast_fp16")]; + tensor add_6_y_0_to_fp16 = const()[name = tensor("add_6_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_6_cast_fp16 = add(x = reduce_mean_11_cast_fp16, y = add_6_y_0_to_fp16)[name = tensor("add_6_cast_fp16")]; + tensor sqrt_3_cast_fp16 = sqrt(x = add_6_cast_fp16)[name = tensor("sqrt_3_cast_fp16")]; + tensor real_div_3_cast_fp16 = real_div(x = sub_6_cast_fp16, y = sqrt_3_cast_fp16)[name = tensor("real_div_3_cast_fp16")]; + tensor reshape_13_shape_0 = const()[name = tensor("reshape_13_shape_0"), val = tensor([2, 320, 64, 64])]; + tensor reshape_13_cast_fp16 = reshape(shape = reshape_13_shape_0, x = real_div_3_cast_fp16)[name = tensor("reshape_13_cast_fp16")]; + tensor add_7_gamma_0_to_fp16 = const()[name = tensor("add_7_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13731904)))]; + tensor add_7_beta_0_to_fp16 = const()[name = tensor("add_7_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13732608)))]; + tensor add_7_epsilon_0_to_fp16 = const()[name = tensor("add_7_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_7_cast_fp16 = batch_norm(beta = add_7_beta_0_to_fp16, epsilon = add_7_epsilon_0_to_fp16, gamma = add_7_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_13_cast_fp16)[name = tensor("add_7_cast_fp16")]; + tensor input_39_cast_fp16 = silu(x = add_7_cast_fp16)[name = tensor("input_39_cast_fp16")]; + tensor hidden_states_19_pad_type_0 = const()[name = tensor("hidden_states_19_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_19_pad_0 = const()[name = tensor("hidden_states_19_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_19_strides_0 = const()[name = tensor("hidden_states_19_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_19_dilations_0 = const()[name = tensor("hidden_states_19_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_19_groups_0 = const()[name = tensor("hidden_states_19_groups_0"), val = tensor(1)]; + tensor down_blocks_0_resnets_1_conv1_weight_to_fp16 = const()[name = tensor("down_blocks_0_resnets_1_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13733312)))]; + tensor down_blocks_0_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_0_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15576576)))]; + tensor hidden_states_19_cast_fp16 = conv(bias = down_blocks_0_resnets_1_conv1_bias_to_fp16, dilations = hidden_states_19_dilations_0, groups = hidden_states_19_groups_0, pad = hidden_states_19_pad_0, pad_type = hidden_states_19_pad_type_0, strides = hidden_states_19_strides_0, weight = down_blocks_0_resnets_1_conv1_weight_to_fp16, x = input_39_cast_fp16)[name = tensor("hidden_states_19_cast_fp16")]; + tensor temb_3_pad_type_0 = const()[name = tensor("temb_3_pad_type_0"), val = tensor("valid")]; + tensor temb_3_strides_0 = const()[name = tensor("temb_3_strides_0"), val = tensor([1, 1])]; + tensor temb_3_pad_0 = const()[name = tensor("temb_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor temb_3_dilations_0 = const()[name = tensor("temb_3_dilations_0"), val = tensor([1, 1])]; + tensor temb_3_groups_0 = const()[name = tensor("temb_3_groups_0"), val = tensor(1)]; + tensor down_blocks_0_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor("down_blocks_0_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15577280)))]; + tensor down_blocks_0_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_0_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16396544)))]; + tensor temb_3_cast_fp16 = conv(bias = down_blocks_0_resnets_1_time_emb_proj_bias_to_fp16, dilations = temb_3_dilations_0, groups = temb_3_groups_0, pad = temb_3_pad_0, pad_type = temb_3_pad_type_0, strides = temb_3_strides_0, weight = down_blocks_0_resnets_1_time_emb_proj_weight_to_fp16, x = input_15_cast_fp16_1)[name = tensor("temb_3_cast_fp16")]; + tensor input_43_cast_fp16 = add(x = hidden_states_19_cast_fp16, y = temb_3_cast_fp16)[name = tensor("input_43_cast_fp16")]; + tensor reshape_16_shape_0 = const()[name = tensor("reshape_16_shape_0"), val = tensor([2, 32, 10, 64, 64])]; + tensor reshape_16_cast_fp16 = reshape(shape = reshape_16_shape_0, x = input_43_cast_fp16)[name = tensor("reshape_16_cast_fp16")]; + tensor reduce_mean_12_axes_0 = const()[name = tensor("reduce_mean_12_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_12_keep_dims_0 = const()[name = tensor("reduce_mean_12_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_12_cast_fp16 = reduce_mean(axes = reduce_mean_12_axes_0, keep_dims = reduce_mean_12_keep_dims_0, x = reshape_16_cast_fp16)[name = tensor("reduce_mean_12_cast_fp16")]; + tensor sub_8_cast_fp16 = sub(x = reshape_16_cast_fp16, y = reduce_mean_12_cast_fp16)[name = tensor("sub_8_cast_fp16")]; + tensor square_4_cast_fp16 = square(x = sub_8_cast_fp16)[name = tensor("square_4_cast_fp16")]; + tensor reduce_mean_14_axes_0 = const()[name = tensor("reduce_mean_14_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_14_keep_dims_0 = const()[name = tensor("reduce_mean_14_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_14_cast_fp16 = reduce_mean(axes = reduce_mean_14_axes_0, keep_dims = reduce_mean_14_keep_dims_0, x = square_4_cast_fp16)[name = tensor("reduce_mean_14_cast_fp16")]; + tensor add_8_y_0_to_fp16 = const()[name = tensor("add_8_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_8_cast_fp16 = add(x = reduce_mean_14_cast_fp16, y = add_8_y_0_to_fp16)[name = tensor("add_8_cast_fp16")]; + tensor sqrt_4_cast_fp16 = sqrt(x = add_8_cast_fp16)[name = tensor("sqrt_4_cast_fp16")]; + tensor real_div_4_cast_fp16 = real_div(x = sub_8_cast_fp16, y = sqrt_4_cast_fp16)[name = tensor("real_div_4_cast_fp16")]; + tensor reshape_17_shape_0 = const()[name = tensor("reshape_17_shape_0"), val = tensor([2, 320, 64, 64])]; + tensor reshape_17_cast_fp16 = reshape(shape = reshape_17_shape_0, x = real_div_4_cast_fp16)[name = tensor("reshape_17_cast_fp16")]; + tensor add_9_gamma_0_to_fp16 = const()[name = tensor("add_9_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16397248)))]; + tensor add_9_beta_0_to_fp16 = const()[name = tensor("add_9_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16397952)))]; + tensor add_9_epsilon_0_to_fp16 = const()[name = tensor("add_9_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_9_cast_fp16 = batch_norm(beta = add_9_beta_0_to_fp16, epsilon = add_9_epsilon_0_to_fp16, gamma = add_9_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_17_cast_fp16)[name = tensor("add_9_cast_fp16")]; + tensor input_47_cast_fp16 = silu(x = add_9_cast_fp16)[name = tensor("input_47_cast_fp16")]; + tensor hidden_states_21_pad_type_0 = const()[name = tensor("hidden_states_21_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_21_pad_0 = const()[name = tensor("hidden_states_21_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_21_strides_0 = const()[name = tensor("hidden_states_21_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_21_dilations_0 = const()[name = tensor("hidden_states_21_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_21_groups_0 = const()[name = tensor("hidden_states_21_groups_0"), val = tensor(1)]; + tensor down_blocks_0_resnets_1_conv2_weight_to_fp16 = const()[name = tensor("down_blocks_0_resnets_1_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16398656)))]; + tensor down_blocks_0_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_0_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18241920)))]; + tensor hidden_states_21_cast_fp16 = conv(bias = down_blocks_0_resnets_1_conv2_bias_to_fp16, dilations = hidden_states_21_dilations_0, groups = hidden_states_21_groups_0, pad = hidden_states_21_pad_0, pad_type = hidden_states_21_pad_type_0, strides = hidden_states_21_strides_0, weight = down_blocks_0_resnets_1_conv2_weight_to_fp16, x = input_47_cast_fp16)[name = tensor("hidden_states_21_cast_fp16")]; + tensor hidden_states_23_cast_fp16 = add(x = input_35_cast_fp16_1, y = hidden_states_21_cast_fp16)[name = tensor("hidden_states_23_cast_fp16")]; + tensor reshape_20_shape_0 = const()[name = tensor("reshape_20_shape_0"), val = tensor([2, 32, 10, 64, 64])]; + tensor reshape_20_cast_fp16 = reshape(shape = reshape_20_shape_0, x = hidden_states_23_cast_fp16)[name = tensor("reshape_20_cast_fp16")]; + tensor reduce_mean_15_axes_0 = const()[name = tensor("reduce_mean_15_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_15_keep_dims_0 = const()[name = tensor("reduce_mean_15_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_15_cast_fp16 = reduce_mean(axes = reduce_mean_15_axes_0, keep_dims = reduce_mean_15_keep_dims_0, x = reshape_20_cast_fp16)[name = tensor("reduce_mean_15_cast_fp16")]; + tensor sub_10_cast_fp16 = sub(x = reshape_20_cast_fp16, y = reduce_mean_15_cast_fp16)[name = tensor("sub_10_cast_fp16")]; + tensor square_5_cast_fp16 = square(x = sub_10_cast_fp16)[name = tensor("square_5_cast_fp16")]; + tensor reduce_mean_17_axes_0 = const()[name = tensor("reduce_mean_17_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_17_keep_dims_0 = const()[name = tensor("reduce_mean_17_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_17_cast_fp16 = reduce_mean(axes = reduce_mean_17_axes_0, keep_dims = reduce_mean_17_keep_dims_0, x = square_5_cast_fp16)[name = tensor("reduce_mean_17_cast_fp16")]; + tensor add_10_y_0_to_fp16 = const()[name = tensor("add_10_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_10_cast_fp16 = add(x = reduce_mean_17_cast_fp16, y = add_10_y_0_to_fp16)[name = tensor("add_10_cast_fp16")]; + tensor sqrt_5_cast_fp16 = sqrt(x = add_10_cast_fp16)[name = tensor("sqrt_5_cast_fp16")]; + tensor real_div_5_cast_fp16 = real_div(x = sub_10_cast_fp16, y = sqrt_5_cast_fp16)[name = tensor("real_div_5_cast_fp16")]; + tensor reshape_21_shape_0 = const()[name = tensor("reshape_21_shape_0"), val = tensor([2, 320, 64, 64])]; + tensor reshape_21_cast_fp16 = reshape(shape = reshape_21_shape_0, x = real_div_5_cast_fp16)[name = tensor("reshape_21_cast_fp16")]; + tensor add_11_gamma_0_to_fp16 = const()[name = tensor("add_11_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18242624)))]; + tensor add_11_beta_0_to_fp16 = const()[name = tensor("add_11_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18243328)))]; + tensor add_11_epsilon_0_to_fp16 = const()[name = tensor("add_11_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_11_cast_fp16 = batch_norm(beta = add_11_beta_0_to_fp16, epsilon = add_11_epsilon_0_to_fp16, gamma = add_11_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_21_cast_fp16)[name = tensor("add_11_cast_fp16")]; + tensor hidden_states_25_pad_type_0 = const()[name = tensor("hidden_states_25_pad_type_0"), val = tensor("valid")]; + tensor hidden_states_25_strides_0 = const()[name = tensor("hidden_states_25_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_25_pad_0 = const()[name = tensor("hidden_states_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_25_dilations_0 = const()[name = tensor("hidden_states_25_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_25_groups_0 = const()[name = tensor("hidden_states_25_groups_0"), val = tensor(1)]; + tensor down_blocks_0_attentions_1_proj_in_weight_to_fp16 = const()[name = tensor("down_blocks_0_attentions_1_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18244032)))]; + tensor down_blocks_0_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor("down_blocks_0_attentions_1_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18448896)))]; + tensor hidden_states_25_cast_fp16 = conv(bias = down_blocks_0_attentions_1_proj_in_bias_to_fp16, dilations = hidden_states_25_dilations_0, groups = hidden_states_25_groups_0, pad = hidden_states_25_pad_0, pad_type = hidden_states_25_pad_type_0, strides = hidden_states_25_strides_0, weight = down_blocks_0_attentions_1_proj_in_weight_to_fp16, x = add_11_cast_fp16)[name = tensor("hidden_states_25_cast_fp16")]; + tensor var_1665 = const()[name = tensor("op_1665"), val = tensor([2, 320, 1, 4096])]; + tensor inputs_7_cast_fp16 = reshape(shape = var_1665, x = hidden_states_25_cast_fp16)[name = tensor("inputs_7_cast_fp16")]; + tensor hidden_states_27_axes_0 = const()[name = tensor("hidden_states_27_axes_0"), val = tensor([1])]; + tensor hidden_states_27_gamma_0_to_fp16 = const()[name = tensor("hidden_states_27_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18449600)))]; + tensor hidden_states_27_beta_0_to_fp16 = const()[name = tensor("hidden_states_27_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18450304)))]; + tensor var_1681_to_fp16 = const()[name = tensor("op_1681_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_27_cast_fp16 = layer_norm(axes = hidden_states_27_axes_0, beta = hidden_states_27_beta_0_to_fp16, epsilon = var_1681_to_fp16, gamma = hidden_states_27_gamma_0_to_fp16, x = inputs_7_cast_fp16)[name = tensor("hidden_states_27_cast_fp16")]; + tensor q_5_pad_type_0 = const()[name = tensor("q_5_pad_type_0"), val = tensor("valid")]; + tensor q_5_strides_0 = const()[name = tensor("q_5_strides_0"), val = tensor([1, 1])]; + tensor q_5_pad_0 = const()[name = tensor("q_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_5_dilations_0 = const()[name = tensor("q_5_dilations_0"), val = tensor([1, 1])]; + tensor q_5_groups_0 = const()[name = tensor("q_5_groups_0"), val = tensor(1)]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18451008)))]; + tensor q_5_cast_fp16 = conv(dilations = q_5_dilations_0, groups = q_5_groups_0, pad = q_5_pad_0, pad_type = q_5_pad_type_0, strides = q_5_strides_0, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_27_cast_fp16)[name = tensor("q_5_cast_fp16")]; + tensor k_9_pad_type_0 = const()[name = tensor("k_9_pad_type_0"), val = tensor("valid")]; + tensor k_9_strides_0 = const()[name = tensor("k_9_strides_0"), val = tensor([1, 1])]; + tensor k_9_pad_0 = const()[name = tensor("k_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_9_dilations_0 = const()[name = tensor("k_9_dilations_0"), val = tensor([1, 1])]; + tensor k_9_groups_0 = const()[name = tensor("k_9_groups_0"), val = tensor(1)]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18655872)))]; + tensor k_9_cast_fp16 = conv(dilations = k_9_dilations_0, groups = k_9_groups_0, pad = k_9_pad_0, pad_type = k_9_pad_type_0, strides = k_9_strides_0, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_27_cast_fp16)[name = tensor("k_9_cast_fp16")]; + tensor v_5_pad_type_0 = const()[name = tensor("v_5_pad_type_0"), val = tensor("valid")]; + tensor v_5_strides_0 = const()[name = tensor("v_5_strides_0"), val = tensor([1, 1])]; + tensor v_5_pad_0 = const()[name = tensor("v_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_5_dilations_0 = const()[name = tensor("v_5_dilations_0"), val = tensor([1, 1])]; + tensor v_5_groups_0 = const()[name = tensor("v_5_groups_0"), val = tensor(1)]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18860736)))]; + tensor v_5_cast_fp16 = conv(dilations = v_5_dilations_0, groups = v_5_groups_0, pad = v_5_pad_0, pad_type = v_5_pad_type_0, strides = v_5_strides_0, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_27_cast_fp16)[name = tensor("v_5_cast_fp16")]; + tensor var_1714_begin_0 = const()[name = tensor("op_1714_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1714_end_0 = const()[name = tensor("op_1714_end_0"), val = tensor([2, 40, 1, 4096])]; + tensor var_1714_end_mask_0 = const()[name = tensor("op_1714_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1714_cast_fp16 = slice_by_index(begin = var_1714_begin_0, end = var_1714_end_0, end_mask = var_1714_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_1714_cast_fp16")]; + tensor var_1718_begin_0 = const()[name = tensor("op_1718_begin_0"), val = tensor([0, 40, 0, 0])]; + tensor var_1718_end_0 = const()[name = tensor("op_1718_end_0"), val = tensor([2, 80, 1, 4096])]; + tensor var_1718_end_mask_0 = const()[name = tensor("op_1718_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1718_cast_fp16 = slice_by_index(begin = var_1718_begin_0, end = var_1718_end_0, end_mask = var_1718_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_1718_cast_fp16")]; + tensor var_1722_begin_0 = const()[name = tensor("op_1722_begin_0"), val = tensor([0, 80, 0, 0])]; + tensor var_1722_end_0 = const()[name = tensor("op_1722_end_0"), val = tensor([2, 120, 1, 4096])]; + tensor var_1722_end_mask_0 = const()[name = tensor("op_1722_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1722_cast_fp16 = slice_by_index(begin = var_1722_begin_0, end = var_1722_end_0, end_mask = var_1722_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_1722_cast_fp16")]; + tensor var_1726_begin_0 = const()[name = tensor("op_1726_begin_0"), val = tensor([0, 120, 0, 0])]; + tensor var_1726_end_0 = const()[name = tensor("op_1726_end_0"), val = tensor([2, 160, 1, 4096])]; + tensor var_1726_end_mask_0 = const()[name = tensor("op_1726_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1726_cast_fp16 = slice_by_index(begin = var_1726_begin_0, end = var_1726_end_0, end_mask = var_1726_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_1726_cast_fp16")]; + tensor var_1730_begin_0 = const()[name = tensor("op_1730_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_1730_end_0 = const()[name = tensor("op_1730_end_0"), val = tensor([2, 200, 1, 4096])]; + tensor var_1730_end_mask_0 = const()[name = tensor("op_1730_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1730_cast_fp16 = slice_by_index(begin = var_1730_begin_0, end = var_1730_end_0, end_mask = var_1730_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_1730_cast_fp16")]; + tensor var_1734_begin_0 = const()[name = tensor("op_1734_begin_0"), val = tensor([0, 200, 0, 0])]; + tensor var_1734_end_0 = const()[name = tensor("op_1734_end_0"), val = tensor([2, 240, 1, 4096])]; + tensor var_1734_end_mask_0 = const()[name = tensor("op_1734_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1734_cast_fp16 = slice_by_index(begin = var_1734_begin_0, end = var_1734_end_0, end_mask = var_1734_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_1734_cast_fp16")]; + tensor var_1738_begin_0 = const()[name = tensor("op_1738_begin_0"), val = tensor([0, 240, 0, 0])]; + tensor var_1738_end_0 = const()[name = tensor("op_1738_end_0"), val = tensor([2, 280, 1, 4096])]; + tensor var_1738_end_mask_0 = const()[name = tensor("op_1738_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1738_cast_fp16 = slice_by_index(begin = var_1738_begin_0, end = var_1738_end_0, end_mask = var_1738_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_1738_cast_fp16")]; + tensor var_1742_begin_0 = const()[name = tensor("op_1742_begin_0"), val = tensor([0, 280, 0, 0])]; + tensor var_1742_end_0 = const()[name = tensor("op_1742_end_0"), val = tensor([2, 1, 1, 4096])]; + tensor var_1742_end_mask_0 = const()[name = tensor("op_1742_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1742_cast_fp16 = slice_by_index(begin = var_1742_begin_0, end = var_1742_end_0, end_mask = var_1742_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_1742_cast_fp16")]; + tensor var_1745_begin_0 = const()[name = tensor("op_1745_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1745_end_0 = const()[name = tensor("op_1745_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_1745_end_mask_0 = const()[name = tensor("op_1745_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1745_cast_fp16 = slice_by_index(begin = var_1745_begin_0, end = var_1745_end_0, end_mask = var_1745_end_mask_0, x = var_1714_cast_fp16)[name = tensor("op_1745_cast_fp16")]; + tensor var_1746_begin_0 = const()[name = tensor("op_1746_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1746_end_0 = const()[name = tensor("op_1746_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_1746_end_mask_0 = const()[name = tensor("op_1746_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1746_cast_fp16 = slice_by_index(begin = var_1746_begin_0, end = var_1746_end_0, end_mask = var_1746_end_mask_0, x = var_1714_cast_fp16)[name = tensor("op_1746_cast_fp16")]; + tensor var_1747_begin_0 = const()[name = tensor("op_1747_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_1747_end_0 = const()[name = tensor("op_1747_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_1747_end_mask_0 = const()[name = tensor("op_1747_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1747_cast_fp16 = slice_by_index(begin = var_1747_begin_0, end = var_1747_end_0, end_mask = var_1747_end_mask_0, x = var_1714_cast_fp16)[name = tensor("op_1747_cast_fp16")]; + tensor var_1748_begin_0 = const()[name = tensor("op_1748_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_1748_end_0 = const()[name = tensor("op_1748_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_1748_end_mask_0 = const()[name = tensor("op_1748_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1748_cast_fp16 = slice_by_index(begin = var_1748_begin_0, end = var_1748_end_0, end_mask = var_1748_end_mask_0, x = var_1714_cast_fp16)[name = tensor("op_1748_cast_fp16")]; + tensor var_1749_begin_0 = const()[name = tensor("op_1749_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_1749_end_0 = const()[name = tensor("op_1749_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_1749_end_mask_0 = const()[name = tensor("op_1749_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1749_cast_fp16 = slice_by_index(begin = var_1749_begin_0, end = var_1749_end_0, end_mask = var_1749_end_mask_0, x = var_1714_cast_fp16)[name = tensor("op_1749_cast_fp16")]; + tensor var_1750_begin_0 = const()[name = tensor("op_1750_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_1750_end_0 = const()[name = tensor("op_1750_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_1750_end_mask_0 = const()[name = tensor("op_1750_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1750_cast_fp16 = slice_by_index(begin = var_1750_begin_0, end = var_1750_end_0, end_mask = var_1750_end_mask_0, x = var_1714_cast_fp16)[name = tensor("op_1750_cast_fp16")]; + tensor var_1751_begin_0 = const()[name = tensor("op_1751_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_1751_end_0 = const()[name = tensor("op_1751_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_1751_end_mask_0 = const()[name = tensor("op_1751_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1751_cast_fp16 = slice_by_index(begin = var_1751_begin_0, end = var_1751_end_0, end_mask = var_1751_end_mask_0, x = var_1714_cast_fp16)[name = tensor("op_1751_cast_fp16")]; + tensor var_1752_begin_0 = const()[name = tensor("op_1752_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_1752_end_0 = const()[name = tensor("op_1752_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_1752_end_mask_0 = const()[name = tensor("op_1752_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1752_cast_fp16 = slice_by_index(begin = var_1752_begin_0, end = var_1752_end_0, end_mask = var_1752_end_mask_0, x = var_1714_cast_fp16)[name = tensor("op_1752_cast_fp16")]; + tensor var_1753_begin_0 = const()[name = tensor("op_1753_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1753_end_0 = const()[name = tensor("op_1753_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_1753_end_mask_0 = const()[name = tensor("op_1753_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1753_cast_fp16 = slice_by_index(begin = var_1753_begin_0, end = var_1753_end_0, end_mask = var_1753_end_mask_0, x = var_1718_cast_fp16)[name = tensor("op_1753_cast_fp16")]; + tensor var_1754_begin_0 = const()[name = tensor("op_1754_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1754_end_0 = const()[name = tensor("op_1754_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_1754_end_mask_0 = const()[name = tensor("op_1754_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1754_cast_fp16 = slice_by_index(begin = var_1754_begin_0, end = var_1754_end_0, end_mask = var_1754_end_mask_0, x = var_1718_cast_fp16)[name = tensor("op_1754_cast_fp16")]; + tensor var_1755_begin_0 = const()[name = tensor("op_1755_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_1755_end_0 = const()[name = tensor("op_1755_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_1755_end_mask_0 = const()[name = tensor("op_1755_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1755_cast_fp16 = slice_by_index(begin = var_1755_begin_0, end = var_1755_end_0, end_mask = var_1755_end_mask_0, x = var_1718_cast_fp16)[name = tensor("op_1755_cast_fp16")]; + tensor var_1756_begin_0 = const()[name = tensor("op_1756_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_1756_end_0 = const()[name = tensor("op_1756_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_1756_end_mask_0 = const()[name = tensor("op_1756_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1756_cast_fp16 = slice_by_index(begin = var_1756_begin_0, end = var_1756_end_0, end_mask = var_1756_end_mask_0, x = var_1718_cast_fp16)[name = tensor("op_1756_cast_fp16")]; + tensor var_1757_begin_0 = const()[name = tensor("op_1757_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_1757_end_0 = const()[name = tensor("op_1757_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_1757_end_mask_0 = const()[name = tensor("op_1757_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1757_cast_fp16 = slice_by_index(begin = var_1757_begin_0, end = var_1757_end_0, end_mask = var_1757_end_mask_0, x = var_1718_cast_fp16)[name = tensor("op_1757_cast_fp16")]; + tensor var_1758_begin_0 = const()[name = tensor("op_1758_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_1758_end_0 = const()[name = tensor("op_1758_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_1758_end_mask_0 = const()[name = tensor("op_1758_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1758_cast_fp16 = slice_by_index(begin = var_1758_begin_0, end = var_1758_end_0, end_mask = var_1758_end_mask_0, x = var_1718_cast_fp16)[name = tensor("op_1758_cast_fp16")]; + tensor var_1759_begin_0 = const()[name = tensor("op_1759_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_1759_end_0 = const()[name = tensor("op_1759_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_1759_end_mask_0 = const()[name = tensor("op_1759_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1759_cast_fp16 = slice_by_index(begin = var_1759_begin_0, end = var_1759_end_0, end_mask = var_1759_end_mask_0, x = var_1718_cast_fp16)[name = tensor("op_1759_cast_fp16")]; + tensor var_1760_begin_0 = const()[name = tensor("op_1760_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_1760_end_0 = const()[name = tensor("op_1760_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_1760_end_mask_0 = const()[name = tensor("op_1760_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1760_cast_fp16 = slice_by_index(begin = var_1760_begin_0, end = var_1760_end_0, end_mask = var_1760_end_mask_0, x = var_1718_cast_fp16)[name = tensor("op_1760_cast_fp16")]; + tensor var_1761_begin_0 = const()[name = tensor("op_1761_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1761_end_0 = const()[name = tensor("op_1761_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_1761_end_mask_0 = const()[name = tensor("op_1761_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1761_cast_fp16 = slice_by_index(begin = var_1761_begin_0, end = var_1761_end_0, end_mask = var_1761_end_mask_0, x = var_1722_cast_fp16)[name = tensor("op_1761_cast_fp16")]; + tensor var_1762_begin_0 = const()[name = tensor("op_1762_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1762_end_0 = const()[name = tensor("op_1762_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_1762_end_mask_0 = const()[name = tensor("op_1762_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1762_cast_fp16 = slice_by_index(begin = var_1762_begin_0, end = var_1762_end_0, end_mask = var_1762_end_mask_0, x = var_1722_cast_fp16)[name = tensor("op_1762_cast_fp16")]; + tensor var_1763_begin_0 = const()[name = tensor("op_1763_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_1763_end_0 = const()[name = tensor("op_1763_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_1763_end_mask_0 = const()[name = tensor("op_1763_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1763_cast_fp16 = slice_by_index(begin = var_1763_begin_0, end = var_1763_end_0, end_mask = var_1763_end_mask_0, x = var_1722_cast_fp16)[name = tensor("op_1763_cast_fp16")]; + tensor var_1764_begin_0 = const()[name = tensor("op_1764_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_1764_end_0 = const()[name = tensor("op_1764_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_1764_end_mask_0 = const()[name = tensor("op_1764_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1764_cast_fp16 = slice_by_index(begin = var_1764_begin_0, end = var_1764_end_0, end_mask = var_1764_end_mask_0, x = var_1722_cast_fp16)[name = tensor("op_1764_cast_fp16")]; + tensor var_1765_begin_0 = const()[name = tensor("op_1765_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_1765_end_0 = const()[name = tensor("op_1765_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_1765_end_mask_0 = const()[name = tensor("op_1765_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1765_cast_fp16 = slice_by_index(begin = var_1765_begin_0, end = var_1765_end_0, end_mask = var_1765_end_mask_0, x = var_1722_cast_fp16)[name = tensor("op_1765_cast_fp16")]; + tensor var_1766_begin_0 = const()[name = tensor("op_1766_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_1766_end_0 = const()[name = tensor("op_1766_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_1766_end_mask_0 = const()[name = tensor("op_1766_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1766_cast_fp16 = slice_by_index(begin = var_1766_begin_0, end = var_1766_end_0, end_mask = var_1766_end_mask_0, x = var_1722_cast_fp16)[name = tensor("op_1766_cast_fp16")]; + tensor var_1767_begin_0 = const()[name = tensor("op_1767_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_1767_end_0 = const()[name = tensor("op_1767_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_1767_end_mask_0 = const()[name = tensor("op_1767_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1767_cast_fp16 = slice_by_index(begin = var_1767_begin_0, end = var_1767_end_0, end_mask = var_1767_end_mask_0, x = var_1722_cast_fp16)[name = tensor("op_1767_cast_fp16")]; + tensor var_1768_begin_0 = const()[name = tensor("op_1768_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_1768_end_0 = const()[name = tensor("op_1768_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_1768_end_mask_0 = const()[name = tensor("op_1768_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1768_cast_fp16 = slice_by_index(begin = var_1768_begin_0, end = var_1768_end_0, end_mask = var_1768_end_mask_0, x = var_1722_cast_fp16)[name = tensor("op_1768_cast_fp16")]; + tensor var_1769_begin_0 = const()[name = tensor("op_1769_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1769_end_0 = const()[name = tensor("op_1769_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_1769_end_mask_0 = const()[name = tensor("op_1769_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1769_cast_fp16 = slice_by_index(begin = var_1769_begin_0, end = var_1769_end_0, end_mask = var_1769_end_mask_0, x = var_1726_cast_fp16)[name = tensor("op_1769_cast_fp16")]; + tensor var_1770_begin_0 = const()[name = tensor("op_1770_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1770_end_0 = const()[name = tensor("op_1770_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_1770_end_mask_0 = const()[name = tensor("op_1770_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1770_cast_fp16 = slice_by_index(begin = var_1770_begin_0, end = var_1770_end_0, end_mask = var_1770_end_mask_0, x = var_1726_cast_fp16)[name = tensor("op_1770_cast_fp16")]; + tensor var_1771_begin_0 = const()[name = tensor("op_1771_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_1771_end_0 = const()[name = tensor("op_1771_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_1771_end_mask_0 = const()[name = tensor("op_1771_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1771_cast_fp16 = slice_by_index(begin = var_1771_begin_0, end = var_1771_end_0, end_mask = var_1771_end_mask_0, x = var_1726_cast_fp16)[name = tensor("op_1771_cast_fp16")]; + tensor var_1772_begin_0 = const()[name = tensor("op_1772_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_1772_end_0 = const()[name = tensor("op_1772_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_1772_end_mask_0 = const()[name = tensor("op_1772_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1772_cast_fp16 = slice_by_index(begin = var_1772_begin_0, end = var_1772_end_0, end_mask = var_1772_end_mask_0, x = var_1726_cast_fp16)[name = tensor("op_1772_cast_fp16")]; + tensor var_1773_begin_0 = const()[name = tensor("op_1773_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_1773_end_0 = const()[name = tensor("op_1773_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_1773_end_mask_0 = const()[name = tensor("op_1773_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1773_cast_fp16 = slice_by_index(begin = var_1773_begin_0, end = var_1773_end_0, end_mask = var_1773_end_mask_0, x = var_1726_cast_fp16)[name = tensor("op_1773_cast_fp16")]; + tensor var_1774_begin_0 = const()[name = tensor("op_1774_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_1774_end_0 = const()[name = tensor("op_1774_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_1774_end_mask_0 = const()[name = tensor("op_1774_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1774_cast_fp16 = slice_by_index(begin = var_1774_begin_0, end = var_1774_end_0, end_mask = var_1774_end_mask_0, x = var_1726_cast_fp16)[name = tensor("op_1774_cast_fp16")]; + tensor var_1775_begin_0 = const()[name = tensor("op_1775_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_1775_end_0 = const()[name = tensor("op_1775_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_1775_end_mask_0 = const()[name = tensor("op_1775_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1775_cast_fp16 = slice_by_index(begin = var_1775_begin_0, end = var_1775_end_0, end_mask = var_1775_end_mask_0, x = var_1726_cast_fp16)[name = tensor("op_1775_cast_fp16")]; + tensor var_1776_begin_0 = const()[name = tensor("op_1776_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_1776_end_0 = const()[name = tensor("op_1776_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_1776_end_mask_0 = const()[name = tensor("op_1776_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1776_cast_fp16 = slice_by_index(begin = var_1776_begin_0, end = var_1776_end_0, end_mask = var_1776_end_mask_0, x = var_1726_cast_fp16)[name = tensor("op_1776_cast_fp16")]; + tensor var_1777_begin_0 = const()[name = tensor("op_1777_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1777_end_0 = const()[name = tensor("op_1777_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_1777_end_mask_0 = const()[name = tensor("op_1777_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1777_cast_fp16 = slice_by_index(begin = var_1777_begin_0, end = var_1777_end_0, end_mask = var_1777_end_mask_0, x = var_1730_cast_fp16)[name = tensor("op_1777_cast_fp16")]; + tensor var_1778_begin_0 = const()[name = tensor("op_1778_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1778_end_0 = const()[name = tensor("op_1778_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_1778_end_mask_0 = const()[name = tensor("op_1778_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1778_cast_fp16 = slice_by_index(begin = var_1778_begin_0, end = var_1778_end_0, end_mask = var_1778_end_mask_0, x = var_1730_cast_fp16)[name = tensor("op_1778_cast_fp16")]; + tensor var_1779_begin_0 = const()[name = tensor("op_1779_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_1779_end_0 = const()[name = tensor("op_1779_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_1779_end_mask_0 = const()[name = tensor("op_1779_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1779_cast_fp16 = slice_by_index(begin = var_1779_begin_0, end = var_1779_end_0, end_mask = var_1779_end_mask_0, x = var_1730_cast_fp16)[name = tensor("op_1779_cast_fp16")]; + tensor var_1780_begin_0 = const()[name = tensor("op_1780_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_1780_end_0 = const()[name = tensor("op_1780_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_1780_end_mask_0 = const()[name = tensor("op_1780_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1780_cast_fp16 = slice_by_index(begin = var_1780_begin_0, end = var_1780_end_0, end_mask = var_1780_end_mask_0, x = var_1730_cast_fp16)[name = tensor("op_1780_cast_fp16")]; + tensor var_1781_begin_0 = const()[name = tensor("op_1781_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_1781_end_0 = const()[name = tensor("op_1781_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_1781_end_mask_0 = const()[name = tensor("op_1781_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1781_cast_fp16 = slice_by_index(begin = var_1781_begin_0, end = var_1781_end_0, end_mask = var_1781_end_mask_0, x = var_1730_cast_fp16)[name = tensor("op_1781_cast_fp16")]; + tensor var_1782_begin_0 = const()[name = tensor("op_1782_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_1782_end_0 = const()[name = tensor("op_1782_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_1782_end_mask_0 = const()[name = tensor("op_1782_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1782_cast_fp16 = slice_by_index(begin = var_1782_begin_0, end = var_1782_end_0, end_mask = var_1782_end_mask_0, x = var_1730_cast_fp16)[name = tensor("op_1782_cast_fp16")]; + tensor var_1783_begin_0 = const()[name = tensor("op_1783_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_1783_end_0 = const()[name = tensor("op_1783_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_1783_end_mask_0 = const()[name = tensor("op_1783_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1783_cast_fp16 = slice_by_index(begin = var_1783_begin_0, end = var_1783_end_0, end_mask = var_1783_end_mask_0, x = var_1730_cast_fp16)[name = tensor("op_1783_cast_fp16")]; + tensor var_1784_begin_0 = const()[name = tensor("op_1784_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_1784_end_0 = const()[name = tensor("op_1784_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_1784_end_mask_0 = const()[name = tensor("op_1784_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1784_cast_fp16 = slice_by_index(begin = var_1784_begin_0, end = var_1784_end_0, end_mask = var_1784_end_mask_0, x = var_1730_cast_fp16)[name = tensor("op_1784_cast_fp16")]; + tensor var_1785_begin_0 = const()[name = tensor("op_1785_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1785_end_0 = const()[name = tensor("op_1785_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_1785_end_mask_0 = const()[name = tensor("op_1785_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1785_cast_fp16 = slice_by_index(begin = var_1785_begin_0, end = var_1785_end_0, end_mask = var_1785_end_mask_0, x = var_1734_cast_fp16)[name = tensor("op_1785_cast_fp16")]; + tensor var_1786_begin_0 = const()[name = tensor("op_1786_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1786_end_0 = const()[name = tensor("op_1786_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_1786_end_mask_0 = const()[name = tensor("op_1786_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1786_cast_fp16 = slice_by_index(begin = var_1786_begin_0, end = var_1786_end_0, end_mask = var_1786_end_mask_0, x = var_1734_cast_fp16)[name = tensor("op_1786_cast_fp16")]; + tensor var_1787_begin_0 = const()[name = tensor("op_1787_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_1787_end_0 = const()[name = tensor("op_1787_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_1787_end_mask_0 = const()[name = tensor("op_1787_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1787_cast_fp16 = slice_by_index(begin = var_1787_begin_0, end = var_1787_end_0, end_mask = var_1787_end_mask_0, x = var_1734_cast_fp16)[name = tensor("op_1787_cast_fp16")]; + tensor var_1788_begin_0 = const()[name = tensor("op_1788_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_1788_end_0 = const()[name = tensor("op_1788_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_1788_end_mask_0 = const()[name = tensor("op_1788_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1788_cast_fp16 = slice_by_index(begin = var_1788_begin_0, end = var_1788_end_0, end_mask = var_1788_end_mask_0, x = var_1734_cast_fp16)[name = tensor("op_1788_cast_fp16")]; + tensor var_1789_begin_0 = const()[name = tensor("op_1789_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_1789_end_0 = const()[name = tensor("op_1789_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_1789_end_mask_0 = const()[name = tensor("op_1789_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1789_cast_fp16 = slice_by_index(begin = var_1789_begin_0, end = var_1789_end_0, end_mask = var_1789_end_mask_0, x = var_1734_cast_fp16)[name = tensor("op_1789_cast_fp16")]; + tensor var_1790_begin_0 = const()[name = tensor("op_1790_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_1790_end_0 = const()[name = tensor("op_1790_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_1790_end_mask_0 = const()[name = tensor("op_1790_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1790_cast_fp16 = slice_by_index(begin = var_1790_begin_0, end = var_1790_end_0, end_mask = var_1790_end_mask_0, x = var_1734_cast_fp16)[name = tensor("op_1790_cast_fp16")]; + tensor var_1791_begin_0 = const()[name = tensor("op_1791_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_1791_end_0 = const()[name = tensor("op_1791_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_1791_end_mask_0 = const()[name = tensor("op_1791_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1791_cast_fp16 = slice_by_index(begin = var_1791_begin_0, end = var_1791_end_0, end_mask = var_1791_end_mask_0, x = var_1734_cast_fp16)[name = tensor("op_1791_cast_fp16")]; + tensor var_1792_begin_0 = const()[name = tensor("op_1792_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_1792_end_0 = const()[name = tensor("op_1792_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_1792_end_mask_0 = const()[name = tensor("op_1792_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1792_cast_fp16 = slice_by_index(begin = var_1792_begin_0, end = var_1792_end_0, end_mask = var_1792_end_mask_0, x = var_1734_cast_fp16)[name = tensor("op_1792_cast_fp16")]; + tensor var_1793_begin_0 = const()[name = tensor("op_1793_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1793_end_0 = const()[name = tensor("op_1793_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_1793_end_mask_0 = const()[name = tensor("op_1793_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1793_cast_fp16 = slice_by_index(begin = var_1793_begin_0, end = var_1793_end_0, end_mask = var_1793_end_mask_0, x = var_1738_cast_fp16)[name = tensor("op_1793_cast_fp16")]; + tensor var_1794_begin_0 = const()[name = tensor("op_1794_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1794_end_0 = const()[name = tensor("op_1794_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_1794_end_mask_0 = const()[name = tensor("op_1794_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1794_cast_fp16 = slice_by_index(begin = var_1794_begin_0, end = var_1794_end_0, end_mask = var_1794_end_mask_0, x = var_1738_cast_fp16)[name = tensor("op_1794_cast_fp16")]; + tensor var_1795_begin_0 = const()[name = tensor("op_1795_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_1795_end_0 = const()[name = tensor("op_1795_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_1795_end_mask_0 = const()[name = tensor("op_1795_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1795_cast_fp16 = slice_by_index(begin = var_1795_begin_0, end = var_1795_end_0, end_mask = var_1795_end_mask_0, x = var_1738_cast_fp16)[name = tensor("op_1795_cast_fp16")]; + tensor var_1796_begin_0 = const()[name = tensor("op_1796_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_1796_end_0 = const()[name = tensor("op_1796_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_1796_end_mask_0 = const()[name = tensor("op_1796_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1796_cast_fp16 = slice_by_index(begin = var_1796_begin_0, end = var_1796_end_0, end_mask = var_1796_end_mask_0, x = var_1738_cast_fp16)[name = tensor("op_1796_cast_fp16")]; + tensor var_1797_begin_0 = const()[name = tensor("op_1797_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_1797_end_0 = const()[name = tensor("op_1797_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_1797_end_mask_0 = const()[name = tensor("op_1797_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1797_cast_fp16 = slice_by_index(begin = var_1797_begin_0, end = var_1797_end_0, end_mask = var_1797_end_mask_0, x = var_1738_cast_fp16)[name = tensor("op_1797_cast_fp16")]; + tensor var_1798_begin_0 = const()[name = tensor("op_1798_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_1798_end_0 = const()[name = tensor("op_1798_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_1798_end_mask_0 = const()[name = tensor("op_1798_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1798_cast_fp16 = slice_by_index(begin = var_1798_begin_0, end = var_1798_end_0, end_mask = var_1798_end_mask_0, x = var_1738_cast_fp16)[name = tensor("op_1798_cast_fp16")]; + tensor var_1799_begin_0 = const()[name = tensor("op_1799_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_1799_end_0 = const()[name = tensor("op_1799_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_1799_end_mask_0 = const()[name = tensor("op_1799_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1799_cast_fp16 = slice_by_index(begin = var_1799_begin_0, end = var_1799_end_0, end_mask = var_1799_end_mask_0, x = var_1738_cast_fp16)[name = tensor("op_1799_cast_fp16")]; + tensor var_1800_begin_0 = const()[name = tensor("op_1800_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_1800_end_0 = const()[name = tensor("op_1800_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_1800_end_mask_0 = const()[name = tensor("op_1800_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1800_cast_fp16 = slice_by_index(begin = var_1800_begin_0, end = var_1800_end_0, end_mask = var_1800_end_mask_0, x = var_1738_cast_fp16)[name = tensor("op_1800_cast_fp16")]; + tensor var_1801_begin_0 = const()[name = tensor("op_1801_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1801_end_0 = const()[name = tensor("op_1801_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_1801_end_mask_0 = const()[name = tensor("op_1801_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1801_cast_fp16 = slice_by_index(begin = var_1801_begin_0, end = var_1801_end_0, end_mask = var_1801_end_mask_0, x = var_1742_cast_fp16)[name = tensor("op_1801_cast_fp16")]; + tensor var_1802_begin_0 = const()[name = tensor("op_1802_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1802_end_0 = const()[name = tensor("op_1802_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_1802_end_mask_0 = const()[name = tensor("op_1802_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1802_cast_fp16 = slice_by_index(begin = var_1802_begin_0, end = var_1802_end_0, end_mask = var_1802_end_mask_0, x = var_1742_cast_fp16)[name = tensor("op_1802_cast_fp16")]; + tensor var_1803_begin_0 = const()[name = tensor("op_1803_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_1803_end_0 = const()[name = tensor("op_1803_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_1803_end_mask_0 = const()[name = tensor("op_1803_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1803_cast_fp16 = slice_by_index(begin = var_1803_begin_0, end = var_1803_end_0, end_mask = var_1803_end_mask_0, x = var_1742_cast_fp16)[name = tensor("op_1803_cast_fp16")]; + tensor var_1804_begin_0 = const()[name = tensor("op_1804_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_1804_end_0 = const()[name = tensor("op_1804_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_1804_end_mask_0 = const()[name = tensor("op_1804_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1804_cast_fp16 = slice_by_index(begin = var_1804_begin_0, end = var_1804_end_0, end_mask = var_1804_end_mask_0, x = var_1742_cast_fp16)[name = tensor("op_1804_cast_fp16")]; + tensor var_1805_begin_0 = const()[name = tensor("op_1805_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_1805_end_0 = const()[name = tensor("op_1805_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_1805_end_mask_0 = const()[name = tensor("op_1805_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1805_cast_fp16 = slice_by_index(begin = var_1805_begin_0, end = var_1805_end_0, end_mask = var_1805_end_mask_0, x = var_1742_cast_fp16)[name = tensor("op_1805_cast_fp16")]; + tensor var_1806_begin_0 = const()[name = tensor("op_1806_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_1806_end_0 = const()[name = tensor("op_1806_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_1806_end_mask_0 = const()[name = tensor("op_1806_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1806_cast_fp16 = slice_by_index(begin = var_1806_begin_0, end = var_1806_end_0, end_mask = var_1806_end_mask_0, x = var_1742_cast_fp16)[name = tensor("op_1806_cast_fp16")]; + tensor var_1807_begin_0 = const()[name = tensor("op_1807_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_1807_end_0 = const()[name = tensor("op_1807_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_1807_end_mask_0 = const()[name = tensor("op_1807_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1807_cast_fp16 = slice_by_index(begin = var_1807_begin_0, end = var_1807_end_0, end_mask = var_1807_end_mask_0, x = var_1742_cast_fp16)[name = tensor("op_1807_cast_fp16")]; + tensor var_1808_begin_0 = const()[name = tensor("op_1808_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_1808_end_0 = const()[name = tensor("op_1808_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_1808_end_mask_0 = const()[name = tensor("op_1808_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1808_cast_fp16 = slice_by_index(begin = var_1808_begin_0, end = var_1808_end_0, end_mask = var_1808_end_mask_0, x = var_1742_cast_fp16)[name = tensor("op_1808_cast_fp16")]; + tensor k_11_perm_0 = const()[name = tensor("k_11_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_1813_begin_0 = const()[name = tensor("op_1813_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1813_end_0 = const()[name = tensor("op_1813_end_0"), val = tensor([2, 4096, 1, 40])]; + tensor var_1813_end_mask_0 = const()[name = tensor("op_1813_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_11_cast_fp16 = transpose(perm = k_11_perm_0, x = k_9_cast_fp16)[name = tensor("transpose_11")]; + tensor var_1813_cast_fp16 = slice_by_index(begin = var_1813_begin_0, end = var_1813_end_0, end_mask = var_1813_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_1813_cast_fp16")]; + tensor var_1817_begin_0 = const()[name = tensor("op_1817_begin_0"), val = tensor([0, 0, 0, 40])]; + tensor var_1817_end_0 = const()[name = tensor("op_1817_end_0"), val = tensor([2, 4096, 1, 80])]; + tensor var_1817_end_mask_0 = const()[name = tensor("op_1817_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1817_cast_fp16 = slice_by_index(begin = var_1817_begin_0, end = var_1817_end_0, end_mask = var_1817_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_1817_cast_fp16")]; + tensor var_1821_begin_0 = const()[name = tensor("op_1821_begin_0"), val = tensor([0, 0, 0, 80])]; + tensor var_1821_end_0 = const()[name = tensor("op_1821_end_0"), val = tensor([2, 4096, 1, 120])]; + tensor var_1821_end_mask_0 = const()[name = tensor("op_1821_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1821_cast_fp16 = slice_by_index(begin = var_1821_begin_0, end = var_1821_end_0, end_mask = var_1821_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_1821_cast_fp16")]; + tensor var_1825_begin_0 = const()[name = tensor("op_1825_begin_0"), val = tensor([0, 0, 0, 120])]; + tensor var_1825_end_0 = const()[name = tensor("op_1825_end_0"), val = tensor([2, 4096, 1, 160])]; + tensor var_1825_end_mask_0 = const()[name = tensor("op_1825_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1825_cast_fp16 = slice_by_index(begin = var_1825_begin_0, end = var_1825_end_0, end_mask = var_1825_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_1825_cast_fp16")]; + tensor var_1829_begin_0 = const()[name = tensor("op_1829_begin_0"), val = tensor([0, 0, 0, 160])]; + tensor var_1829_end_0 = const()[name = tensor("op_1829_end_0"), val = tensor([2, 4096, 1, 200])]; + tensor var_1829_end_mask_0 = const()[name = tensor("op_1829_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1829_cast_fp16 = slice_by_index(begin = var_1829_begin_0, end = var_1829_end_0, end_mask = var_1829_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_1829_cast_fp16")]; + tensor var_1833_begin_0 = const()[name = tensor("op_1833_begin_0"), val = tensor([0, 0, 0, 200])]; + tensor var_1833_end_0 = const()[name = tensor("op_1833_end_0"), val = tensor([2, 4096, 1, 240])]; + tensor var_1833_end_mask_0 = const()[name = tensor("op_1833_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1833_cast_fp16 = slice_by_index(begin = var_1833_begin_0, end = var_1833_end_0, end_mask = var_1833_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_1833_cast_fp16")]; + tensor var_1837_begin_0 = const()[name = tensor("op_1837_begin_0"), val = tensor([0, 0, 0, 240])]; + tensor var_1837_end_0 = const()[name = tensor("op_1837_end_0"), val = tensor([2, 4096, 1, 280])]; + tensor var_1837_end_mask_0 = const()[name = tensor("op_1837_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1837_cast_fp16 = slice_by_index(begin = var_1837_begin_0, end = var_1837_end_0, end_mask = var_1837_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_1837_cast_fp16")]; + tensor var_1841_begin_0 = const()[name = tensor("op_1841_begin_0"), val = tensor([0, 0, 0, 280])]; + tensor var_1841_end_0 = const()[name = tensor("op_1841_end_0"), val = tensor([2, 4096, 1, 1])]; + tensor var_1841_end_mask_0 = const()[name = tensor("op_1841_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1841_cast_fp16 = slice_by_index(begin = var_1841_begin_0, end = var_1841_end_0, end_mask = var_1841_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_1841_cast_fp16")]; + tensor var_1843_begin_0 = const()[name = tensor("op_1843_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1843_end_0 = const()[name = tensor("op_1843_end_0"), val = tensor([2, 40, 1, 4096])]; + tensor var_1843_end_mask_0 = const()[name = tensor("op_1843_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1843_cast_fp16 = slice_by_index(begin = var_1843_begin_0, end = var_1843_end_0, end_mask = var_1843_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1843_cast_fp16")]; + tensor var_1847_begin_0 = const()[name = tensor("op_1847_begin_0"), val = tensor([0, 40, 0, 0])]; + tensor var_1847_end_0 = const()[name = tensor("op_1847_end_0"), val = tensor([2, 80, 1, 4096])]; + tensor var_1847_end_mask_0 = const()[name = tensor("op_1847_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1847_cast_fp16 = slice_by_index(begin = var_1847_begin_0, end = var_1847_end_0, end_mask = var_1847_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1847_cast_fp16")]; + tensor var_1851_begin_0 = const()[name = tensor("op_1851_begin_0"), val = tensor([0, 80, 0, 0])]; + tensor var_1851_end_0 = const()[name = tensor("op_1851_end_0"), val = tensor([2, 120, 1, 4096])]; + tensor var_1851_end_mask_0 = const()[name = tensor("op_1851_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1851_cast_fp16 = slice_by_index(begin = var_1851_begin_0, end = var_1851_end_0, end_mask = var_1851_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1851_cast_fp16")]; + tensor var_1855_begin_0 = const()[name = tensor("op_1855_begin_0"), val = tensor([0, 120, 0, 0])]; + tensor var_1855_end_0 = const()[name = tensor("op_1855_end_0"), val = tensor([2, 160, 1, 4096])]; + tensor var_1855_end_mask_0 = const()[name = tensor("op_1855_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1855_cast_fp16 = slice_by_index(begin = var_1855_begin_0, end = var_1855_end_0, end_mask = var_1855_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1855_cast_fp16")]; + tensor var_1859_begin_0 = const()[name = tensor("op_1859_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_1859_end_0 = const()[name = tensor("op_1859_end_0"), val = tensor([2, 200, 1, 4096])]; + tensor var_1859_end_mask_0 = const()[name = tensor("op_1859_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1859_cast_fp16 = slice_by_index(begin = var_1859_begin_0, end = var_1859_end_0, end_mask = var_1859_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1859_cast_fp16")]; + tensor var_1863_begin_0 = const()[name = tensor("op_1863_begin_0"), val = tensor([0, 200, 0, 0])]; + tensor var_1863_end_0 = const()[name = tensor("op_1863_end_0"), val = tensor([2, 240, 1, 4096])]; + tensor var_1863_end_mask_0 = const()[name = tensor("op_1863_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1863_cast_fp16 = slice_by_index(begin = var_1863_begin_0, end = var_1863_end_0, end_mask = var_1863_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1863_cast_fp16")]; + tensor var_1867_begin_0 = const()[name = tensor("op_1867_begin_0"), val = tensor([0, 240, 0, 0])]; + tensor var_1867_end_0 = const()[name = tensor("op_1867_end_0"), val = tensor([2, 280, 1, 4096])]; + tensor var_1867_end_mask_0 = const()[name = tensor("op_1867_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1867_cast_fp16 = slice_by_index(begin = var_1867_begin_0, end = var_1867_end_0, end_mask = var_1867_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1867_cast_fp16")]; + tensor var_1871_begin_0 = const()[name = tensor("op_1871_begin_0"), val = tensor([0, 280, 0, 0])]; + tensor var_1871_end_0 = const()[name = tensor("op_1871_end_0"), val = tensor([2, 1, 1, 4096])]; + tensor var_1871_end_mask_0 = const()[name = tensor("op_1871_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1871_cast_fp16 = slice_by_index(begin = var_1871_begin_0, end = var_1871_end_0, end_mask = var_1871_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1871_cast_fp16")]; + tensor var_1875_equation_0 = const()[name = tensor("op_1875_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1875_cast_fp16 = einsum(equation = var_1875_equation_0, values = (var_1813_cast_fp16, var_1745_cast_fp16))[name = tensor("op_1875_cast_fp16")]; + tensor var_1876_to_fp16 = const()[name = tensor("op_1876_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_257_cast_fp16 = mul(x = var_1875_cast_fp16, y = var_1876_to_fp16)[name = tensor("aw_chunk_257_cast_fp16")]; + tensor var_1879_equation_0 = const()[name = tensor("op_1879_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1879_cast_fp16 = einsum(equation = var_1879_equation_0, values = (var_1813_cast_fp16, var_1746_cast_fp16))[name = tensor("op_1879_cast_fp16")]; + tensor var_1880_to_fp16 = const()[name = tensor("op_1880_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_259_cast_fp16 = mul(x = var_1879_cast_fp16, y = var_1880_to_fp16)[name = tensor("aw_chunk_259_cast_fp16")]; + tensor var_1883_equation_0 = const()[name = tensor("op_1883_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1883_cast_fp16 = einsum(equation = var_1883_equation_0, values = (var_1813_cast_fp16, var_1747_cast_fp16))[name = tensor("op_1883_cast_fp16")]; + tensor var_1884_to_fp16 = const()[name = tensor("op_1884_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_261_cast_fp16 = mul(x = var_1883_cast_fp16, y = var_1884_to_fp16)[name = tensor("aw_chunk_261_cast_fp16")]; + tensor var_1887_equation_0 = const()[name = tensor("op_1887_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1887_cast_fp16 = einsum(equation = var_1887_equation_0, values = (var_1813_cast_fp16, var_1748_cast_fp16))[name = tensor("op_1887_cast_fp16")]; + tensor var_1888_to_fp16 = const()[name = tensor("op_1888_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_263_cast_fp16 = mul(x = var_1887_cast_fp16, y = var_1888_to_fp16)[name = tensor("aw_chunk_263_cast_fp16")]; + tensor var_1891_equation_0 = const()[name = tensor("op_1891_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1891_cast_fp16 = einsum(equation = var_1891_equation_0, values = (var_1813_cast_fp16, var_1749_cast_fp16))[name = tensor("op_1891_cast_fp16")]; + tensor var_1892_to_fp16 = const()[name = tensor("op_1892_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_265_cast_fp16 = mul(x = var_1891_cast_fp16, y = var_1892_to_fp16)[name = tensor("aw_chunk_265_cast_fp16")]; + tensor var_1895_equation_0 = const()[name = tensor("op_1895_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1895_cast_fp16 = einsum(equation = var_1895_equation_0, values = (var_1813_cast_fp16, var_1750_cast_fp16))[name = tensor("op_1895_cast_fp16")]; + tensor var_1896_to_fp16 = const()[name = tensor("op_1896_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_267_cast_fp16 = mul(x = var_1895_cast_fp16, y = var_1896_to_fp16)[name = tensor("aw_chunk_267_cast_fp16")]; + tensor var_1899_equation_0 = const()[name = tensor("op_1899_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1899_cast_fp16 = einsum(equation = var_1899_equation_0, values = (var_1813_cast_fp16, var_1751_cast_fp16))[name = tensor("op_1899_cast_fp16")]; + tensor var_1900_to_fp16 = const()[name = tensor("op_1900_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_269_cast_fp16 = mul(x = var_1899_cast_fp16, y = var_1900_to_fp16)[name = tensor("aw_chunk_269_cast_fp16")]; + tensor var_1903_equation_0 = const()[name = tensor("op_1903_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1903_cast_fp16 = einsum(equation = var_1903_equation_0, values = (var_1813_cast_fp16, var_1752_cast_fp16))[name = tensor("op_1903_cast_fp16")]; + tensor var_1904_to_fp16 = const()[name = tensor("op_1904_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_271_cast_fp16 = mul(x = var_1903_cast_fp16, y = var_1904_to_fp16)[name = tensor("aw_chunk_271_cast_fp16")]; + tensor var_1907_equation_0 = const()[name = tensor("op_1907_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1907_cast_fp16 = einsum(equation = var_1907_equation_0, values = (var_1817_cast_fp16, var_1753_cast_fp16))[name = tensor("op_1907_cast_fp16")]; + tensor var_1908_to_fp16 = const()[name = tensor("op_1908_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_273_cast_fp16 = mul(x = var_1907_cast_fp16, y = var_1908_to_fp16)[name = tensor("aw_chunk_273_cast_fp16")]; + tensor var_1911_equation_0 = const()[name = tensor("op_1911_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1911_cast_fp16 = einsum(equation = var_1911_equation_0, values = (var_1817_cast_fp16, var_1754_cast_fp16))[name = tensor("op_1911_cast_fp16")]; + tensor var_1912_to_fp16 = const()[name = tensor("op_1912_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_275_cast_fp16 = mul(x = var_1911_cast_fp16, y = var_1912_to_fp16)[name = tensor("aw_chunk_275_cast_fp16")]; + tensor var_1915_equation_0 = const()[name = tensor("op_1915_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1915_cast_fp16 = einsum(equation = var_1915_equation_0, values = (var_1817_cast_fp16, var_1755_cast_fp16))[name = tensor("op_1915_cast_fp16")]; + tensor var_1916_to_fp16 = const()[name = tensor("op_1916_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_277_cast_fp16 = mul(x = var_1915_cast_fp16, y = var_1916_to_fp16)[name = tensor("aw_chunk_277_cast_fp16")]; + tensor var_1919_equation_0 = const()[name = tensor("op_1919_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1919_cast_fp16 = einsum(equation = var_1919_equation_0, values = (var_1817_cast_fp16, var_1756_cast_fp16))[name = tensor("op_1919_cast_fp16")]; + tensor var_1920_to_fp16 = const()[name = tensor("op_1920_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_279_cast_fp16 = mul(x = var_1919_cast_fp16, y = var_1920_to_fp16)[name = tensor("aw_chunk_279_cast_fp16")]; + tensor var_1923_equation_0 = const()[name = tensor("op_1923_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1923_cast_fp16 = einsum(equation = var_1923_equation_0, values = (var_1817_cast_fp16, var_1757_cast_fp16))[name = tensor("op_1923_cast_fp16")]; + tensor var_1924_to_fp16 = const()[name = tensor("op_1924_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_281_cast_fp16 = mul(x = var_1923_cast_fp16, y = var_1924_to_fp16)[name = tensor("aw_chunk_281_cast_fp16")]; + tensor var_1927_equation_0 = const()[name = tensor("op_1927_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1927_cast_fp16 = einsum(equation = var_1927_equation_0, values = (var_1817_cast_fp16, var_1758_cast_fp16))[name = tensor("op_1927_cast_fp16")]; + tensor var_1928_to_fp16 = const()[name = tensor("op_1928_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_283_cast_fp16 = mul(x = var_1927_cast_fp16, y = var_1928_to_fp16)[name = tensor("aw_chunk_283_cast_fp16")]; + tensor var_1931_equation_0 = const()[name = tensor("op_1931_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1931_cast_fp16 = einsum(equation = var_1931_equation_0, values = (var_1817_cast_fp16, var_1759_cast_fp16))[name = tensor("op_1931_cast_fp16")]; + tensor var_1932_to_fp16 = const()[name = tensor("op_1932_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_285_cast_fp16 = mul(x = var_1931_cast_fp16, y = var_1932_to_fp16)[name = tensor("aw_chunk_285_cast_fp16")]; + tensor var_1935_equation_0 = const()[name = tensor("op_1935_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1935_cast_fp16 = einsum(equation = var_1935_equation_0, values = (var_1817_cast_fp16, var_1760_cast_fp16))[name = tensor("op_1935_cast_fp16")]; + tensor var_1936_to_fp16 = const()[name = tensor("op_1936_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_287_cast_fp16 = mul(x = var_1935_cast_fp16, y = var_1936_to_fp16)[name = tensor("aw_chunk_287_cast_fp16")]; + tensor var_1939_equation_0 = const()[name = tensor("op_1939_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1939_cast_fp16 = einsum(equation = var_1939_equation_0, values = (var_1821_cast_fp16, var_1761_cast_fp16))[name = tensor("op_1939_cast_fp16")]; + tensor var_1940_to_fp16 = const()[name = tensor("op_1940_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_289_cast_fp16 = mul(x = var_1939_cast_fp16, y = var_1940_to_fp16)[name = tensor("aw_chunk_289_cast_fp16")]; + tensor var_1943_equation_0 = const()[name = tensor("op_1943_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1943_cast_fp16 = einsum(equation = var_1943_equation_0, values = (var_1821_cast_fp16, var_1762_cast_fp16))[name = tensor("op_1943_cast_fp16")]; + tensor var_1944_to_fp16 = const()[name = tensor("op_1944_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_291_cast_fp16 = mul(x = var_1943_cast_fp16, y = var_1944_to_fp16)[name = tensor("aw_chunk_291_cast_fp16")]; + tensor var_1947_equation_0 = const()[name = tensor("op_1947_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1947_cast_fp16 = einsum(equation = var_1947_equation_0, values = (var_1821_cast_fp16, var_1763_cast_fp16))[name = tensor("op_1947_cast_fp16")]; + tensor var_1948_to_fp16 = const()[name = tensor("op_1948_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_293_cast_fp16 = mul(x = var_1947_cast_fp16, y = var_1948_to_fp16)[name = tensor("aw_chunk_293_cast_fp16")]; + tensor var_1951_equation_0 = const()[name = tensor("op_1951_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1951_cast_fp16 = einsum(equation = var_1951_equation_0, values = (var_1821_cast_fp16, var_1764_cast_fp16))[name = tensor("op_1951_cast_fp16")]; + tensor var_1952_to_fp16 = const()[name = tensor("op_1952_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_295_cast_fp16 = mul(x = var_1951_cast_fp16, y = var_1952_to_fp16)[name = tensor("aw_chunk_295_cast_fp16")]; + tensor var_1955_equation_0 = const()[name = tensor("op_1955_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1955_cast_fp16 = einsum(equation = var_1955_equation_0, values = (var_1821_cast_fp16, var_1765_cast_fp16))[name = tensor("op_1955_cast_fp16")]; + tensor var_1956_to_fp16 = const()[name = tensor("op_1956_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_297_cast_fp16 = mul(x = var_1955_cast_fp16, y = var_1956_to_fp16)[name = tensor("aw_chunk_297_cast_fp16")]; + tensor var_1959_equation_0 = const()[name = tensor("op_1959_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1959_cast_fp16 = einsum(equation = var_1959_equation_0, values = (var_1821_cast_fp16, var_1766_cast_fp16))[name = tensor("op_1959_cast_fp16")]; + tensor var_1960_to_fp16 = const()[name = tensor("op_1960_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_299_cast_fp16 = mul(x = var_1959_cast_fp16, y = var_1960_to_fp16)[name = tensor("aw_chunk_299_cast_fp16")]; + tensor var_1963_equation_0 = const()[name = tensor("op_1963_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1963_cast_fp16 = einsum(equation = var_1963_equation_0, values = (var_1821_cast_fp16, var_1767_cast_fp16))[name = tensor("op_1963_cast_fp16")]; + tensor var_1964_to_fp16 = const()[name = tensor("op_1964_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_301_cast_fp16 = mul(x = var_1963_cast_fp16, y = var_1964_to_fp16)[name = tensor("aw_chunk_301_cast_fp16")]; + tensor var_1967_equation_0 = const()[name = tensor("op_1967_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1967_cast_fp16 = einsum(equation = var_1967_equation_0, values = (var_1821_cast_fp16, var_1768_cast_fp16))[name = tensor("op_1967_cast_fp16")]; + tensor var_1968_to_fp16 = const()[name = tensor("op_1968_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_303_cast_fp16 = mul(x = var_1967_cast_fp16, y = var_1968_to_fp16)[name = tensor("aw_chunk_303_cast_fp16")]; + tensor var_1971_equation_0 = const()[name = tensor("op_1971_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1971_cast_fp16 = einsum(equation = var_1971_equation_0, values = (var_1825_cast_fp16, var_1769_cast_fp16))[name = tensor("op_1971_cast_fp16")]; + tensor var_1972_to_fp16 = const()[name = tensor("op_1972_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_305_cast_fp16 = mul(x = var_1971_cast_fp16, y = var_1972_to_fp16)[name = tensor("aw_chunk_305_cast_fp16")]; + tensor var_1975_equation_0 = const()[name = tensor("op_1975_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1975_cast_fp16 = einsum(equation = var_1975_equation_0, values = (var_1825_cast_fp16, var_1770_cast_fp16))[name = tensor("op_1975_cast_fp16")]; + tensor var_1976_to_fp16 = const()[name = tensor("op_1976_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_307_cast_fp16 = mul(x = var_1975_cast_fp16, y = var_1976_to_fp16)[name = tensor("aw_chunk_307_cast_fp16")]; + tensor var_1979_equation_0 = const()[name = tensor("op_1979_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1979_cast_fp16 = einsum(equation = var_1979_equation_0, values = (var_1825_cast_fp16, var_1771_cast_fp16))[name = tensor("op_1979_cast_fp16")]; + tensor var_1980_to_fp16 = const()[name = tensor("op_1980_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_309_cast_fp16 = mul(x = var_1979_cast_fp16, y = var_1980_to_fp16)[name = tensor("aw_chunk_309_cast_fp16")]; + tensor var_1983_equation_0 = const()[name = tensor("op_1983_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1983_cast_fp16 = einsum(equation = var_1983_equation_0, values = (var_1825_cast_fp16, var_1772_cast_fp16))[name = tensor("op_1983_cast_fp16")]; + tensor var_1984_to_fp16 = const()[name = tensor("op_1984_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_311_cast_fp16 = mul(x = var_1983_cast_fp16, y = var_1984_to_fp16)[name = tensor("aw_chunk_311_cast_fp16")]; + tensor var_1987_equation_0 = const()[name = tensor("op_1987_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1987_cast_fp16 = einsum(equation = var_1987_equation_0, values = (var_1825_cast_fp16, var_1773_cast_fp16))[name = tensor("op_1987_cast_fp16")]; + tensor var_1988_to_fp16 = const()[name = tensor("op_1988_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_313_cast_fp16 = mul(x = var_1987_cast_fp16, y = var_1988_to_fp16)[name = tensor("aw_chunk_313_cast_fp16")]; + tensor var_1991_equation_0 = const()[name = tensor("op_1991_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1991_cast_fp16 = einsum(equation = var_1991_equation_0, values = (var_1825_cast_fp16, var_1774_cast_fp16))[name = tensor("op_1991_cast_fp16")]; + tensor var_1992_to_fp16 = const()[name = tensor("op_1992_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_315_cast_fp16 = mul(x = var_1991_cast_fp16, y = var_1992_to_fp16)[name = tensor("aw_chunk_315_cast_fp16")]; + tensor var_1995_equation_0 = const()[name = tensor("op_1995_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1995_cast_fp16 = einsum(equation = var_1995_equation_0, values = (var_1825_cast_fp16, var_1775_cast_fp16))[name = tensor("op_1995_cast_fp16")]; + tensor var_1996_to_fp16 = const()[name = tensor("op_1996_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_317_cast_fp16 = mul(x = var_1995_cast_fp16, y = var_1996_to_fp16)[name = tensor("aw_chunk_317_cast_fp16")]; + tensor var_1999_equation_0 = const()[name = tensor("op_1999_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1999_cast_fp16 = einsum(equation = var_1999_equation_0, values = (var_1825_cast_fp16, var_1776_cast_fp16))[name = tensor("op_1999_cast_fp16")]; + tensor var_2000_to_fp16 = const()[name = tensor("op_2000_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_319_cast_fp16 = mul(x = var_1999_cast_fp16, y = var_2000_to_fp16)[name = tensor("aw_chunk_319_cast_fp16")]; + tensor var_2003_equation_0 = const()[name = tensor("op_2003_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2003_cast_fp16 = einsum(equation = var_2003_equation_0, values = (var_1829_cast_fp16, var_1777_cast_fp16))[name = tensor("op_2003_cast_fp16")]; + tensor var_2004_to_fp16 = const()[name = tensor("op_2004_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_321_cast_fp16 = mul(x = var_2003_cast_fp16, y = var_2004_to_fp16)[name = tensor("aw_chunk_321_cast_fp16")]; + tensor var_2007_equation_0 = const()[name = tensor("op_2007_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2007_cast_fp16 = einsum(equation = var_2007_equation_0, values = (var_1829_cast_fp16, var_1778_cast_fp16))[name = tensor("op_2007_cast_fp16")]; + tensor var_2008_to_fp16 = const()[name = tensor("op_2008_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_323_cast_fp16 = mul(x = var_2007_cast_fp16, y = var_2008_to_fp16)[name = tensor("aw_chunk_323_cast_fp16")]; + tensor var_2011_equation_0 = const()[name = tensor("op_2011_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2011_cast_fp16 = einsum(equation = var_2011_equation_0, values = (var_1829_cast_fp16, var_1779_cast_fp16))[name = tensor("op_2011_cast_fp16")]; + tensor var_2012_to_fp16 = const()[name = tensor("op_2012_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_325_cast_fp16 = mul(x = var_2011_cast_fp16, y = var_2012_to_fp16)[name = tensor("aw_chunk_325_cast_fp16")]; + tensor var_2015_equation_0 = const()[name = tensor("op_2015_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2015_cast_fp16 = einsum(equation = var_2015_equation_0, values = (var_1829_cast_fp16, var_1780_cast_fp16))[name = tensor("op_2015_cast_fp16")]; + tensor var_2016_to_fp16 = const()[name = tensor("op_2016_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_327_cast_fp16 = mul(x = var_2015_cast_fp16, y = var_2016_to_fp16)[name = tensor("aw_chunk_327_cast_fp16")]; + tensor var_2019_equation_0 = const()[name = tensor("op_2019_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2019_cast_fp16 = einsum(equation = var_2019_equation_0, values = (var_1829_cast_fp16, var_1781_cast_fp16))[name = tensor("op_2019_cast_fp16")]; + tensor var_2020_to_fp16 = const()[name = tensor("op_2020_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_329_cast_fp16 = mul(x = var_2019_cast_fp16, y = var_2020_to_fp16)[name = tensor("aw_chunk_329_cast_fp16")]; + tensor var_2023_equation_0 = const()[name = tensor("op_2023_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2023_cast_fp16 = einsum(equation = var_2023_equation_0, values = (var_1829_cast_fp16, var_1782_cast_fp16))[name = tensor("op_2023_cast_fp16")]; + tensor var_2024_to_fp16 = const()[name = tensor("op_2024_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_331_cast_fp16 = mul(x = var_2023_cast_fp16, y = var_2024_to_fp16)[name = tensor("aw_chunk_331_cast_fp16")]; + tensor var_2027_equation_0 = const()[name = tensor("op_2027_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2027_cast_fp16 = einsum(equation = var_2027_equation_0, values = (var_1829_cast_fp16, var_1783_cast_fp16))[name = tensor("op_2027_cast_fp16")]; + tensor var_2028_to_fp16 = const()[name = tensor("op_2028_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_333_cast_fp16 = mul(x = var_2027_cast_fp16, y = var_2028_to_fp16)[name = tensor("aw_chunk_333_cast_fp16")]; + tensor var_2031_equation_0 = const()[name = tensor("op_2031_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2031_cast_fp16 = einsum(equation = var_2031_equation_0, values = (var_1829_cast_fp16, var_1784_cast_fp16))[name = tensor("op_2031_cast_fp16")]; + tensor var_2032_to_fp16 = const()[name = tensor("op_2032_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_335_cast_fp16 = mul(x = var_2031_cast_fp16, y = var_2032_to_fp16)[name = tensor("aw_chunk_335_cast_fp16")]; + tensor var_2035_equation_0 = const()[name = tensor("op_2035_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2035_cast_fp16 = einsum(equation = var_2035_equation_0, values = (var_1833_cast_fp16, var_1785_cast_fp16))[name = tensor("op_2035_cast_fp16")]; + tensor var_2036_to_fp16 = const()[name = tensor("op_2036_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_337_cast_fp16 = mul(x = var_2035_cast_fp16, y = var_2036_to_fp16)[name = tensor("aw_chunk_337_cast_fp16")]; + tensor var_2039_equation_0 = const()[name = tensor("op_2039_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2039_cast_fp16 = einsum(equation = var_2039_equation_0, values = (var_1833_cast_fp16, var_1786_cast_fp16))[name = tensor("op_2039_cast_fp16")]; + tensor var_2040_to_fp16 = const()[name = tensor("op_2040_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_339_cast_fp16 = mul(x = var_2039_cast_fp16, y = var_2040_to_fp16)[name = tensor("aw_chunk_339_cast_fp16")]; + tensor var_2043_equation_0 = const()[name = tensor("op_2043_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2043_cast_fp16 = einsum(equation = var_2043_equation_0, values = (var_1833_cast_fp16, var_1787_cast_fp16))[name = tensor("op_2043_cast_fp16")]; + tensor var_2044_to_fp16 = const()[name = tensor("op_2044_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_341_cast_fp16 = mul(x = var_2043_cast_fp16, y = var_2044_to_fp16)[name = tensor("aw_chunk_341_cast_fp16")]; + tensor var_2047_equation_0 = const()[name = tensor("op_2047_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2047_cast_fp16 = einsum(equation = var_2047_equation_0, values = (var_1833_cast_fp16, var_1788_cast_fp16))[name = tensor("op_2047_cast_fp16")]; + tensor var_2048_to_fp16 = const()[name = tensor("op_2048_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_343_cast_fp16 = mul(x = var_2047_cast_fp16, y = var_2048_to_fp16)[name = tensor("aw_chunk_343_cast_fp16")]; + tensor var_2051_equation_0 = const()[name = tensor("op_2051_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2051_cast_fp16 = einsum(equation = var_2051_equation_0, values = (var_1833_cast_fp16, var_1789_cast_fp16))[name = tensor("op_2051_cast_fp16")]; + tensor var_2052_to_fp16 = const()[name = tensor("op_2052_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_345_cast_fp16 = mul(x = var_2051_cast_fp16, y = var_2052_to_fp16)[name = tensor("aw_chunk_345_cast_fp16")]; + tensor var_2055_equation_0 = const()[name = tensor("op_2055_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2055_cast_fp16 = einsum(equation = var_2055_equation_0, values = (var_1833_cast_fp16, var_1790_cast_fp16))[name = tensor("op_2055_cast_fp16")]; + tensor var_2056_to_fp16 = const()[name = tensor("op_2056_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_347_cast_fp16 = mul(x = var_2055_cast_fp16, y = var_2056_to_fp16)[name = tensor("aw_chunk_347_cast_fp16")]; + tensor var_2059_equation_0 = const()[name = tensor("op_2059_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2059_cast_fp16 = einsum(equation = var_2059_equation_0, values = (var_1833_cast_fp16, var_1791_cast_fp16))[name = tensor("op_2059_cast_fp16")]; + tensor var_2060_to_fp16 = const()[name = tensor("op_2060_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_349_cast_fp16 = mul(x = var_2059_cast_fp16, y = var_2060_to_fp16)[name = tensor("aw_chunk_349_cast_fp16")]; + tensor var_2063_equation_0 = const()[name = tensor("op_2063_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2063_cast_fp16 = einsum(equation = var_2063_equation_0, values = (var_1833_cast_fp16, var_1792_cast_fp16))[name = tensor("op_2063_cast_fp16")]; + tensor var_2064_to_fp16 = const()[name = tensor("op_2064_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_351_cast_fp16 = mul(x = var_2063_cast_fp16, y = var_2064_to_fp16)[name = tensor("aw_chunk_351_cast_fp16")]; + tensor var_2067_equation_0 = const()[name = tensor("op_2067_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2067_cast_fp16 = einsum(equation = var_2067_equation_0, values = (var_1837_cast_fp16, var_1793_cast_fp16))[name = tensor("op_2067_cast_fp16")]; + tensor var_2068_to_fp16 = const()[name = tensor("op_2068_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_353_cast_fp16 = mul(x = var_2067_cast_fp16, y = var_2068_to_fp16)[name = tensor("aw_chunk_353_cast_fp16")]; + tensor var_2071_equation_0 = const()[name = tensor("op_2071_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2071_cast_fp16 = einsum(equation = var_2071_equation_0, values = (var_1837_cast_fp16, var_1794_cast_fp16))[name = tensor("op_2071_cast_fp16")]; + tensor var_2072_to_fp16 = const()[name = tensor("op_2072_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_355_cast_fp16 = mul(x = var_2071_cast_fp16, y = var_2072_to_fp16)[name = tensor("aw_chunk_355_cast_fp16")]; + tensor var_2075_equation_0 = const()[name = tensor("op_2075_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2075_cast_fp16 = einsum(equation = var_2075_equation_0, values = (var_1837_cast_fp16, var_1795_cast_fp16))[name = tensor("op_2075_cast_fp16")]; + tensor var_2076_to_fp16 = const()[name = tensor("op_2076_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_357_cast_fp16 = mul(x = var_2075_cast_fp16, y = var_2076_to_fp16)[name = tensor("aw_chunk_357_cast_fp16")]; + tensor var_2079_equation_0 = const()[name = tensor("op_2079_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2079_cast_fp16 = einsum(equation = var_2079_equation_0, values = (var_1837_cast_fp16, var_1796_cast_fp16))[name = tensor("op_2079_cast_fp16")]; + tensor var_2080_to_fp16 = const()[name = tensor("op_2080_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_359_cast_fp16 = mul(x = var_2079_cast_fp16, y = var_2080_to_fp16)[name = tensor("aw_chunk_359_cast_fp16")]; + tensor var_2083_equation_0 = const()[name = tensor("op_2083_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2083_cast_fp16 = einsum(equation = var_2083_equation_0, values = (var_1837_cast_fp16, var_1797_cast_fp16))[name = tensor("op_2083_cast_fp16")]; + tensor var_2084_to_fp16 = const()[name = tensor("op_2084_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_361_cast_fp16 = mul(x = var_2083_cast_fp16, y = var_2084_to_fp16)[name = tensor("aw_chunk_361_cast_fp16")]; + tensor var_2087_equation_0 = const()[name = tensor("op_2087_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2087_cast_fp16 = einsum(equation = var_2087_equation_0, values = (var_1837_cast_fp16, var_1798_cast_fp16))[name = tensor("op_2087_cast_fp16")]; + tensor var_2088_to_fp16 = const()[name = tensor("op_2088_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_363_cast_fp16 = mul(x = var_2087_cast_fp16, y = var_2088_to_fp16)[name = tensor("aw_chunk_363_cast_fp16")]; + tensor var_2091_equation_0 = const()[name = tensor("op_2091_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2091_cast_fp16 = einsum(equation = var_2091_equation_0, values = (var_1837_cast_fp16, var_1799_cast_fp16))[name = tensor("op_2091_cast_fp16")]; + tensor var_2092_to_fp16 = const()[name = tensor("op_2092_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_365_cast_fp16 = mul(x = var_2091_cast_fp16, y = var_2092_to_fp16)[name = tensor("aw_chunk_365_cast_fp16")]; + tensor var_2095_equation_0 = const()[name = tensor("op_2095_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2095_cast_fp16 = einsum(equation = var_2095_equation_0, values = (var_1837_cast_fp16, var_1800_cast_fp16))[name = tensor("op_2095_cast_fp16")]; + tensor var_2096_to_fp16 = const()[name = tensor("op_2096_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_367_cast_fp16 = mul(x = var_2095_cast_fp16, y = var_2096_to_fp16)[name = tensor("aw_chunk_367_cast_fp16")]; + tensor var_2099_equation_0 = const()[name = tensor("op_2099_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2099_cast_fp16 = einsum(equation = var_2099_equation_0, values = (var_1841_cast_fp16, var_1801_cast_fp16))[name = tensor("op_2099_cast_fp16")]; + tensor var_2100_to_fp16 = const()[name = tensor("op_2100_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_369_cast_fp16 = mul(x = var_2099_cast_fp16, y = var_2100_to_fp16)[name = tensor("aw_chunk_369_cast_fp16")]; + tensor var_2103_equation_0 = const()[name = tensor("op_2103_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2103_cast_fp16 = einsum(equation = var_2103_equation_0, values = (var_1841_cast_fp16, var_1802_cast_fp16))[name = tensor("op_2103_cast_fp16")]; + tensor var_2104_to_fp16 = const()[name = tensor("op_2104_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_371_cast_fp16 = mul(x = var_2103_cast_fp16, y = var_2104_to_fp16)[name = tensor("aw_chunk_371_cast_fp16")]; + tensor var_2107_equation_0 = const()[name = tensor("op_2107_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2107_cast_fp16 = einsum(equation = var_2107_equation_0, values = (var_1841_cast_fp16, var_1803_cast_fp16))[name = tensor("op_2107_cast_fp16")]; + tensor var_2108_to_fp16 = const()[name = tensor("op_2108_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_373_cast_fp16 = mul(x = var_2107_cast_fp16, y = var_2108_to_fp16)[name = tensor("aw_chunk_373_cast_fp16")]; + tensor var_2111_equation_0 = const()[name = tensor("op_2111_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2111_cast_fp16 = einsum(equation = var_2111_equation_0, values = (var_1841_cast_fp16, var_1804_cast_fp16))[name = tensor("op_2111_cast_fp16")]; + tensor var_2112_to_fp16 = const()[name = tensor("op_2112_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_375_cast_fp16 = mul(x = var_2111_cast_fp16, y = var_2112_to_fp16)[name = tensor("aw_chunk_375_cast_fp16")]; + tensor var_2115_equation_0 = const()[name = tensor("op_2115_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2115_cast_fp16 = einsum(equation = var_2115_equation_0, values = (var_1841_cast_fp16, var_1805_cast_fp16))[name = tensor("op_2115_cast_fp16")]; + tensor var_2116_to_fp16 = const()[name = tensor("op_2116_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_377_cast_fp16 = mul(x = var_2115_cast_fp16, y = var_2116_to_fp16)[name = tensor("aw_chunk_377_cast_fp16")]; + tensor var_2119_equation_0 = const()[name = tensor("op_2119_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2119_cast_fp16 = einsum(equation = var_2119_equation_0, values = (var_1841_cast_fp16, var_1806_cast_fp16))[name = tensor("op_2119_cast_fp16")]; + tensor var_2120_to_fp16 = const()[name = tensor("op_2120_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_379_cast_fp16 = mul(x = var_2119_cast_fp16, y = var_2120_to_fp16)[name = tensor("aw_chunk_379_cast_fp16")]; + tensor var_2123_equation_0 = const()[name = tensor("op_2123_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2123_cast_fp16 = einsum(equation = var_2123_equation_0, values = (var_1841_cast_fp16, var_1807_cast_fp16))[name = tensor("op_2123_cast_fp16")]; + tensor var_2124_to_fp16 = const()[name = tensor("op_2124_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_381_cast_fp16 = mul(x = var_2123_cast_fp16, y = var_2124_to_fp16)[name = tensor("aw_chunk_381_cast_fp16")]; + tensor var_2127_equation_0 = const()[name = tensor("op_2127_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2127_cast_fp16 = einsum(equation = var_2127_equation_0, values = (var_1841_cast_fp16, var_1808_cast_fp16))[name = tensor("op_2127_cast_fp16")]; + tensor var_2128_to_fp16 = const()[name = tensor("op_2128_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_383_cast_fp16 = mul(x = var_2127_cast_fp16, y = var_2128_to_fp16)[name = tensor("aw_chunk_383_cast_fp16")]; + tensor var_2130_cast_fp16 = softmax(axis = var_123, x = aw_chunk_257_cast_fp16)[name = tensor("op_2130_cast_fp16")]; + tensor var_2131_cast_fp16 = softmax(axis = var_123, x = aw_chunk_259_cast_fp16)[name = tensor("op_2131_cast_fp16")]; + tensor var_2132_cast_fp16 = softmax(axis = var_123, x = aw_chunk_261_cast_fp16)[name = tensor("op_2132_cast_fp16")]; + tensor var_2133_cast_fp16 = softmax(axis = var_123, x = aw_chunk_263_cast_fp16)[name = tensor("op_2133_cast_fp16")]; + tensor var_2134_cast_fp16 = softmax(axis = var_123, x = aw_chunk_265_cast_fp16)[name = tensor("op_2134_cast_fp16")]; + tensor var_2135_cast_fp16 = softmax(axis = var_123, x = aw_chunk_267_cast_fp16)[name = tensor("op_2135_cast_fp16")]; + tensor var_2136_cast_fp16 = softmax(axis = var_123, x = aw_chunk_269_cast_fp16)[name = tensor("op_2136_cast_fp16")]; + tensor var_2137_cast_fp16 = softmax(axis = var_123, x = aw_chunk_271_cast_fp16)[name = tensor("op_2137_cast_fp16")]; + tensor var_2138_cast_fp16 = softmax(axis = var_123, x = aw_chunk_273_cast_fp16)[name = tensor("op_2138_cast_fp16")]; + tensor var_2139_cast_fp16 = softmax(axis = var_123, x = aw_chunk_275_cast_fp16)[name = tensor("op_2139_cast_fp16")]; + tensor var_2140_cast_fp16 = softmax(axis = var_123, x = aw_chunk_277_cast_fp16)[name = tensor("op_2140_cast_fp16")]; + tensor var_2141_cast_fp16 = softmax(axis = var_123, x = aw_chunk_279_cast_fp16)[name = tensor("op_2141_cast_fp16")]; + tensor var_2142_cast_fp16 = softmax(axis = var_123, x = aw_chunk_281_cast_fp16)[name = tensor("op_2142_cast_fp16")]; + tensor var_2143_cast_fp16 = softmax(axis = var_123, x = aw_chunk_283_cast_fp16)[name = tensor("op_2143_cast_fp16")]; + tensor var_2144_cast_fp16 = softmax(axis = var_123, x = aw_chunk_285_cast_fp16)[name = tensor("op_2144_cast_fp16")]; + tensor var_2145_cast_fp16 = softmax(axis = var_123, x = aw_chunk_287_cast_fp16)[name = tensor("op_2145_cast_fp16")]; + tensor var_2146_cast_fp16 = softmax(axis = var_123, x = aw_chunk_289_cast_fp16)[name = tensor("op_2146_cast_fp16")]; + tensor var_2147_cast_fp16 = softmax(axis = var_123, x = aw_chunk_291_cast_fp16)[name = tensor("op_2147_cast_fp16")]; + tensor var_2148_cast_fp16 = softmax(axis = var_123, x = aw_chunk_293_cast_fp16)[name = tensor("op_2148_cast_fp16")]; + tensor var_2149_cast_fp16 = softmax(axis = var_123, x = aw_chunk_295_cast_fp16)[name = tensor("op_2149_cast_fp16")]; + tensor var_2150_cast_fp16 = softmax(axis = var_123, x = aw_chunk_297_cast_fp16)[name = tensor("op_2150_cast_fp16")]; + tensor var_2151_cast_fp16 = softmax(axis = var_123, x = aw_chunk_299_cast_fp16)[name = tensor("op_2151_cast_fp16")]; + tensor var_2152_cast_fp16 = softmax(axis = var_123, x = aw_chunk_301_cast_fp16)[name = tensor("op_2152_cast_fp16")]; + tensor var_2153_cast_fp16 = softmax(axis = var_123, x = aw_chunk_303_cast_fp16)[name = tensor("op_2153_cast_fp16")]; + tensor var_2154_cast_fp16 = softmax(axis = var_123, x = aw_chunk_305_cast_fp16)[name = tensor("op_2154_cast_fp16")]; + tensor var_2155_cast_fp16 = softmax(axis = var_123, x = aw_chunk_307_cast_fp16)[name = tensor("op_2155_cast_fp16")]; + tensor var_2156_cast_fp16 = softmax(axis = var_123, x = aw_chunk_309_cast_fp16)[name = tensor("op_2156_cast_fp16")]; + tensor var_2157_cast_fp16 = softmax(axis = var_123, x = aw_chunk_311_cast_fp16)[name = tensor("op_2157_cast_fp16")]; + tensor var_2158_cast_fp16 = softmax(axis = var_123, x = aw_chunk_313_cast_fp16)[name = tensor("op_2158_cast_fp16")]; + tensor var_2159_cast_fp16 = softmax(axis = var_123, x = aw_chunk_315_cast_fp16)[name = tensor("op_2159_cast_fp16")]; + tensor var_2160_cast_fp16 = softmax(axis = var_123, x = aw_chunk_317_cast_fp16)[name = tensor("op_2160_cast_fp16")]; + tensor var_2161_cast_fp16 = softmax(axis = var_123, x = aw_chunk_319_cast_fp16)[name = tensor("op_2161_cast_fp16")]; + tensor var_2162_cast_fp16 = softmax(axis = var_123, x = aw_chunk_321_cast_fp16)[name = tensor("op_2162_cast_fp16")]; + tensor var_2163_cast_fp16 = softmax(axis = var_123, x = aw_chunk_323_cast_fp16)[name = tensor("op_2163_cast_fp16")]; + tensor var_2164_cast_fp16 = softmax(axis = var_123, x = aw_chunk_325_cast_fp16)[name = tensor("op_2164_cast_fp16")]; + tensor var_2165_cast_fp16 = softmax(axis = var_123, x = aw_chunk_327_cast_fp16)[name = tensor("op_2165_cast_fp16")]; + tensor var_2166_cast_fp16 = softmax(axis = var_123, x = aw_chunk_329_cast_fp16)[name = tensor("op_2166_cast_fp16")]; + tensor var_2167_cast_fp16 = softmax(axis = var_123, x = aw_chunk_331_cast_fp16)[name = tensor("op_2167_cast_fp16")]; + tensor var_2168_cast_fp16 = softmax(axis = var_123, x = aw_chunk_333_cast_fp16)[name = tensor("op_2168_cast_fp16")]; + tensor var_2169_cast_fp16 = softmax(axis = var_123, x = aw_chunk_335_cast_fp16)[name = tensor("op_2169_cast_fp16")]; + tensor var_2170_cast_fp16 = softmax(axis = var_123, x = aw_chunk_337_cast_fp16)[name = tensor("op_2170_cast_fp16")]; + tensor var_2171_cast_fp16 = softmax(axis = var_123, x = aw_chunk_339_cast_fp16)[name = tensor("op_2171_cast_fp16")]; + tensor var_2172_cast_fp16 = softmax(axis = var_123, x = aw_chunk_341_cast_fp16)[name = tensor("op_2172_cast_fp16")]; + tensor var_2173_cast_fp16 = softmax(axis = var_123, x = aw_chunk_343_cast_fp16)[name = tensor("op_2173_cast_fp16")]; + tensor var_2174_cast_fp16 = softmax(axis = var_123, x = aw_chunk_345_cast_fp16)[name = tensor("op_2174_cast_fp16")]; + tensor var_2175_cast_fp16 = softmax(axis = var_123, x = aw_chunk_347_cast_fp16)[name = tensor("op_2175_cast_fp16")]; + tensor var_2176_cast_fp16 = softmax(axis = var_123, x = aw_chunk_349_cast_fp16)[name = tensor("op_2176_cast_fp16")]; + tensor var_2177_cast_fp16 = softmax(axis = var_123, x = aw_chunk_351_cast_fp16)[name = tensor("op_2177_cast_fp16")]; + tensor var_2178_cast_fp16 = softmax(axis = var_123, x = aw_chunk_353_cast_fp16)[name = tensor("op_2178_cast_fp16")]; + tensor var_2179_cast_fp16 = softmax(axis = var_123, x = aw_chunk_355_cast_fp16)[name = tensor("op_2179_cast_fp16")]; + tensor var_2180_cast_fp16 = softmax(axis = var_123, x = aw_chunk_357_cast_fp16)[name = tensor("op_2180_cast_fp16")]; + tensor var_2181_cast_fp16 = softmax(axis = var_123, x = aw_chunk_359_cast_fp16)[name = tensor("op_2181_cast_fp16")]; + tensor var_2182_cast_fp16 = softmax(axis = var_123, x = aw_chunk_361_cast_fp16)[name = tensor("op_2182_cast_fp16")]; + tensor var_2183_cast_fp16 = softmax(axis = var_123, x = aw_chunk_363_cast_fp16)[name = tensor("op_2183_cast_fp16")]; + tensor var_2184_cast_fp16 = softmax(axis = var_123, x = aw_chunk_365_cast_fp16)[name = tensor("op_2184_cast_fp16")]; + tensor var_2185_cast_fp16 = softmax(axis = var_123, x = aw_chunk_367_cast_fp16)[name = tensor("op_2185_cast_fp16")]; + tensor var_2186_cast_fp16 = softmax(axis = var_123, x = aw_chunk_369_cast_fp16)[name = tensor("op_2186_cast_fp16")]; + tensor var_2187_cast_fp16 = softmax(axis = var_123, x = aw_chunk_371_cast_fp16)[name = tensor("op_2187_cast_fp16")]; + tensor var_2188_cast_fp16 = softmax(axis = var_123, x = aw_chunk_373_cast_fp16)[name = tensor("op_2188_cast_fp16")]; + tensor var_2189_cast_fp16 = softmax(axis = var_123, x = aw_chunk_375_cast_fp16)[name = tensor("op_2189_cast_fp16")]; + tensor var_2190_cast_fp16 = softmax(axis = var_123, x = aw_chunk_377_cast_fp16)[name = tensor("op_2190_cast_fp16")]; + tensor var_2191_cast_fp16 = softmax(axis = var_123, x = aw_chunk_379_cast_fp16)[name = tensor("op_2191_cast_fp16")]; + tensor var_2192_cast_fp16 = softmax(axis = var_123, x = aw_chunk_381_cast_fp16)[name = tensor("op_2192_cast_fp16")]; + tensor var_2193_cast_fp16 = softmax(axis = var_123, x = aw_chunk_383_cast_fp16)[name = tensor("op_2193_cast_fp16")]; + tensor var_2195_equation_0 = const()[name = tensor("op_2195_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2195_cast_fp16 = einsum(equation = var_2195_equation_0, values = (var_1843_cast_fp16, var_2130_cast_fp16))[name = tensor("op_2195_cast_fp16")]; + tensor var_2197_equation_0 = const()[name = tensor("op_2197_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2197_cast_fp16 = einsum(equation = var_2197_equation_0, values = (var_1843_cast_fp16, var_2131_cast_fp16))[name = tensor("op_2197_cast_fp16")]; + tensor var_2199_equation_0 = const()[name = tensor("op_2199_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2199_cast_fp16 = einsum(equation = var_2199_equation_0, values = (var_1843_cast_fp16, var_2132_cast_fp16))[name = tensor("op_2199_cast_fp16")]; + tensor var_2201_equation_0 = const()[name = tensor("op_2201_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2201_cast_fp16 = einsum(equation = var_2201_equation_0, values = (var_1843_cast_fp16, var_2133_cast_fp16))[name = tensor("op_2201_cast_fp16")]; + tensor var_2203_equation_0 = const()[name = tensor("op_2203_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2203_cast_fp16 = einsum(equation = var_2203_equation_0, values = (var_1843_cast_fp16, var_2134_cast_fp16))[name = tensor("op_2203_cast_fp16")]; + tensor var_2205_equation_0 = const()[name = tensor("op_2205_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2205_cast_fp16 = einsum(equation = var_2205_equation_0, values = (var_1843_cast_fp16, var_2135_cast_fp16))[name = tensor("op_2205_cast_fp16")]; + tensor var_2207_equation_0 = const()[name = tensor("op_2207_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2207_cast_fp16 = einsum(equation = var_2207_equation_0, values = (var_1843_cast_fp16, var_2136_cast_fp16))[name = tensor("op_2207_cast_fp16")]; + tensor var_2209_equation_0 = const()[name = tensor("op_2209_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2209_cast_fp16 = einsum(equation = var_2209_equation_0, values = (var_1843_cast_fp16, var_2137_cast_fp16))[name = tensor("op_2209_cast_fp16")]; + tensor var_2211_equation_0 = const()[name = tensor("op_2211_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2211_cast_fp16 = einsum(equation = var_2211_equation_0, values = (var_1847_cast_fp16, var_2138_cast_fp16))[name = tensor("op_2211_cast_fp16")]; + tensor var_2213_equation_0 = const()[name = tensor("op_2213_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2213_cast_fp16 = einsum(equation = var_2213_equation_0, values = (var_1847_cast_fp16, var_2139_cast_fp16))[name = tensor("op_2213_cast_fp16")]; + tensor var_2215_equation_0 = const()[name = tensor("op_2215_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2215_cast_fp16 = einsum(equation = var_2215_equation_0, values = (var_1847_cast_fp16, var_2140_cast_fp16))[name = tensor("op_2215_cast_fp16")]; + tensor var_2217_equation_0 = const()[name = tensor("op_2217_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2217_cast_fp16 = einsum(equation = var_2217_equation_0, values = (var_1847_cast_fp16, var_2141_cast_fp16))[name = tensor("op_2217_cast_fp16")]; + tensor var_2219_equation_0 = const()[name = tensor("op_2219_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2219_cast_fp16 = einsum(equation = var_2219_equation_0, values = (var_1847_cast_fp16, var_2142_cast_fp16))[name = tensor("op_2219_cast_fp16")]; + tensor var_2221_equation_0 = const()[name = tensor("op_2221_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2221_cast_fp16 = einsum(equation = var_2221_equation_0, values = (var_1847_cast_fp16, var_2143_cast_fp16))[name = tensor("op_2221_cast_fp16")]; + tensor var_2223_equation_0 = const()[name = tensor("op_2223_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2223_cast_fp16 = einsum(equation = var_2223_equation_0, values = (var_1847_cast_fp16, var_2144_cast_fp16))[name = tensor("op_2223_cast_fp16")]; + tensor var_2225_equation_0 = const()[name = tensor("op_2225_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2225_cast_fp16 = einsum(equation = var_2225_equation_0, values = (var_1847_cast_fp16, var_2145_cast_fp16))[name = tensor("op_2225_cast_fp16")]; + tensor var_2227_equation_0 = const()[name = tensor("op_2227_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2227_cast_fp16 = einsum(equation = var_2227_equation_0, values = (var_1851_cast_fp16, var_2146_cast_fp16))[name = tensor("op_2227_cast_fp16")]; + tensor var_2229_equation_0 = const()[name = tensor("op_2229_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2229_cast_fp16 = einsum(equation = var_2229_equation_0, values = (var_1851_cast_fp16, var_2147_cast_fp16))[name = tensor("op_2229_cast_fp16")]; + tensor var_2231_equation_0 = const()[name = tensor("op_2231_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2231_cast_fp16 = einsum(equation = var_2231_equation_0, values = (var_1851_cast_fp16, var_2148_cast_fp16))[name = tensor("op_2231_cast_fp16")]; + tensor var_2233_equation_0 = const()[name = tensor("op_2233_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2233_cast_fp16 = einsum(equation = var_2233_equation_0, values = (var_1851_cast_fp16, var_2149_cast_fp16))[name = tensor("op_2233_cast_fp16")]; + tensor var_2235_equation_0 = const()[name = tensor("op_2235_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2235_cast_fp16 = einsum(equation = var_2235_equation_0, values = (var_1851_cast_fp16, var_2150_cast_fp16))[name = tensor("op_2235_cast_fp16")]; + tensor var_2237_equation_0 = const()[name = tensor("op_2237_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2237_cast_fp16 = einsum(equation = var_2237_equation_0, values = (var_1851_cast_fp16, var_2151_cast_fp16))[name = tensor("op_2237_cast_fp16")]; + tensor var_2239_equation_0 = const()[name = tensor("op_2239_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2239_cast_fp16 = einsum(equation = var_2239_equation_0, values = (var_1851_cast_fp16, var_2152_cast_fp16))[name = tensor("op_2239_cast_fp16")]; + tensor var_2241_equation_0 = const()[name = tensor("op_2241_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2241_cast_fp16 = einsum(equation = var_2241_equation_0, values = (var_1851_cast_fp16, var_2153_cast_fp16))[name = tensor("op_2241_cast_fp16")]; + tensor var_2243_equation_0 = const()[name = tensor("op_2243_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2243_cast_fp16 = einsum(equation = var_2243_equation_0, values = (var_1855_cast_fp16, var_2154_cast_fp16))[name = tensor("op_2243_cast_fp16")]; + tensor var_2245_equation_0 = const()[name = tensor("op_2245_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2245_cast_fp16 = einsum(equation = var_2245_equation_0, values = (var_1855_cast_fp16, var_2155_cast_fp16))[name = tensor("op_2245_cast_fp16")]; + tensor var_2247_equation_0 = const()[name = tensor("op_2247_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2247_cast_fp16 = einsum(equation = var_2247_equation_0, values = (var_1855_cast_fp16, var_2156_cast_fp16))[name = tensor("op_2247_cast_fp16")]; + tensor var_2249_equation_0 = const()[name = tensor("op_2249_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2249_cast_fp16 = einsum(equation = var_2249_equation_0, values = (var_1855_cast_fp16, var_2157_cast_fp16))[name = tensor("op_2249_cast_fp16")]; + tensor var_2251_equation_0 = const()[name = tensor("op_2251_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2251_cast_fp16 = einsum(equation = var_2251_equation_0, values = (var_1855_cast_fp16, var_2158_cast_fp16))[name = tensor("op_2251_cast_fp16")]; + tensor var_2253_equation_0 = const()[name = tensor("op_2253_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2253_cast_fp16 = einsum(equation = var_2253_equation_0, values = (var_1855_cast_fp16, var_2159_cast_fp16))[name = tensor("op_2253_cast_fp16")]; + tensor var_2255_equation_0 = const()[name = tensor("op_2255_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2255_cast_fp16 = einsum(equation = var_2255_equation_0, values = (var_1855_cast_fp16, var_2160_cast_fp16))[name = tensor("op_2255_cast_fp16")]; + tensor var_2257_equation_0 = const()[name = tensor("op_2257_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2257_cast_fp16 = einsum(equation = var_2257_equation_0, values = (var_1855_cast_fp16, var_2161_cast_fp16))[name = tensor("op_2257_cast_fp16")]; + tensor var_2259_equation_0 = const()[name = tensor("op_2259_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2259_cast_fp16 = einsum(equation = var_2259_equation_0, values = (var_1859_cast_fp16, var_2162_cast_fp16))[name = tensor("op_2259_cast_fp16")]; + tensor var_2261_equation_0 = const()[name = tensor("op_2261_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2261_cast_fp16 = einsum(equation = var_2261_equation_0, values = (var_1859_cast_fp16, var_2163_cast_fp16))[name = tensor("op_2261_cast_fp16")]; + tensor var_2263_equation_0 = const()[name = tensor("op_2263_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2263_cast_fp16 = einsum(equation = var_2263_equation_0, values = (var_1859_cast_fp16, var_2164_cast_fp16))[name = tensor("op_2263_cast_fp16")]; + tensor var_2265_equation_0 = const()[name = tensor("op_2265_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2265_cast_fp16 = einsum(equation = var_2265_equation_0, values = (var_1859_cast_fp16, var_2165_cast_fp16))[name = tensor("op_2265_cast_fp16")]; + tensor var_2267_equation_0 = const()[name = tensor("op_2267_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2267_cast_fp16 = einsum(equation = var_2267_equation_0, values = (var_1859_cast_fp16, var_2166_cast_fp16))[name = tensor("op_2267_cast_fp16")]; + tensor var_2269_equation_0 = const()[name = tensor("op_2269_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2269_cast_fp16 = einsum(equation = var_2269_equation_0, values = (var_1859_cast_fp16, var_2167_cast_fp16))[name = tensor("op_2269_cast_fp16")]; + tensor var_2271_equation_0 = const()[name = tensor("op_2271_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2271_cast_fp16 = einsum(equation = var_2271_equation_0, values = (var_1859_cast_fp16, var_2168_cast_fp16))[name = tensor("op_2271_cast_fp16")]; + tensor var_2273_equation_0 = const()[name = tensor("op_2273_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2273_cast_fp16 = einsum(equation = var_2273_equation_0, values = (var_1859_cast_fp16, var_2169_cast_fp16))[name = tensor("op_2273_cast_fp16")]; + tensor var_2275_equation_0 = const()[name = tensor("op_2275_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2275_cast_fp16 = einsum(equation = var_2275_equation_0, values = (var_1863_cast_fp16, var_2170_cast_fp16))[name = tensor("op_2275_cast_fp16")]; + tensor var_2277_equation_0 = const()[name = tensor("op_2277_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2277_cast_fp16 = einsum(equation = var_2277_equation_0, values = (var_1863_cast_fp16, var_2171_cast_fp16))[name = tensor("op_2277_cast_fp16")]; + tensor var_2279_equation_0 = const()[name = tensor("op_2279_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2279_cast_fp16 = einsum(equation = var_2279_equation_0, values = (var_1863_cast_fp16, var_2172_cast_fp16))[name = tensor("op_2279_cast_fp16")]; + tensor var_2281_equation_0 = const()[name = tensor("op_2281_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2281_cast_fp16 = einsum(equation = var_2281_equation_0, values = (var_1863_cast_fp16, var_2173_cast_fp16))[name = tensor("op_2281_cast_fp16")]; + tensor var_2283_equation_0 = const()[name = tensor("op_2283_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2283_cast_fp16 = einsum(equation = var_2283_equation_0, values = (var_1863_cast_fp16, var_2174_cast_fp16))[name = tensor("op_2283_cast_fp16")]; + tensor var_2285_equation_0 = const()[name = tensor("op_2285_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2285_cast_fp16 = einsum(equation = var_2285_equation_0, values = (var_1863_cast_fp16, var_2175_cast_fp16))[name = tensor("op_2285_cast_fp16")]; + tensor var_2287_equation_0 = const()[name = tensor("op_2287_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2287_cast_fp16 = einsum(equation = var_2287_equation_0, values = (var_1863_cast_fp16, var_2176_cast_fp16))[name = tensor("op_2287_cast_fp16")]; + tensor var_2289_equation_0 = const()[name = tensor("op_2289_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2289_cast_fp16 = einsum(equation = var_2289_equation_0, values = (var_1863_cast_fp16, var_2177_cast_fp16))[name = tensor("op_2289_cast_fp16")]; + tensor var_2291_equation_0 = const()[name = tensor("op_2291_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2291_cast_fp16 = einsum(equation = var_2291_equation_0, values = (var_1867_cast_fp16, var_2178_cast_fp16))[name = tensor("op_2291_cast_fp16")]; + tensor var_2293_equation_0 = const()[name = tensor("op_2293_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2293_cast_fp16 = einsum(equation = var_2293_equation_0, values = (var_1867_cast_fp16, var_2179_cast_fp16))[name = tensor("op_2293_cast_fp16")]; + tensor var_2295_equation_0 = const()[name = tensor("op_2295_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2295_cast_fp16 = einsum(equation = var_2295_equation_0, values = (var_1867_cast_fp16, var_2180_cast_fp16))[name = tensor("op_2295_cast_fp16")]; + tensor var_2297_equation_0 = const()[name = tensor("op_2297_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2297_cast_fp16 = einsum(equation = var_2297_equation_0, values = (var_1867_cast_fp16, var_2181_cast_fp16))[name = tensor("op_2297_cast_fp16")]; + tensor var_2299_equation_0 = const()[name = tensor("op_2299_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2299_cast_fp16 = einsum(equation = var_2299_equation_0, values = (var_1867_cast_fp16, var_2182_cast_fp16))[name = tensor("op_2299_cast_fp16")]; + tensor var_2301_equation_0 = const()[name = tensor("op_2301_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2301_cast_fp16 = einsum(equation = var_2301_equation_0, values = (var_1867_cast_fp16, var_2183_cast_fp16))[name = tensor("op_2301_cast_fp16")]; + tensor var_2303_equation_0 = const()[name = tensor("op_2303_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2303_cast_fp16 = einsum(equation = var_2303_equation_0, values = (var_1867_cast_fp16, var_2184_cast_fp16))[name = tensor("op_2303_cast_fp16")]; + tensor var_2305_equation_0 = const()[name = tensor("op_2305_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2305_cast_fp16 = einsum(equation = var_2305_equation_0, values = (var_1867_cast_fp16, var_2185_cast_fp16))[name = tensor("op_2305_cast_fp16")]; + tensor var_2307_equation_0 = const()[name = tensor("op_2307_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2307_cast_fp16 = einsum(equation = var_2307_equation_0, values = (var_1871_cast_fp16, var_2186_cast_fp16))[name = tensor("op_2307_cast_fp16")]; + tensor var_2309_equation_0 = const()[name = tensor("op_2309_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2309_cast_fp16 = einsum(equation = var_2309_equation_0, values = (var_1871_cast_fp16, var_2187_cast_fp16))[name = tensor("op_2309_cast_fp16")]; + tensor var_2311_equation_0 = const()[name = tensor("op_2311_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2311_cast_fp16 = einsum(equation = var_2311_equation_0, values = (var_1871_cast_fp16, var_2188_cast_fp16))[name = tensor("op_2311_cast_fp16")]; + tensor var_2313_equation_0 = const()[name = tensor("op_2313_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2313_cast_fp16 = einsum(equation = var_2313_equation_0, values = (var_1871_cast_fp16, var_2189_cast_fp16))[name = tensor("op_2313_cast_fp16")]; + tensor var_2315_equation_0 = const()[name = tensor("op_2315_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2315_cast_fp16 = einsum(equation = var_2315_equation_0, values = (var_1871_cast_fp16, var_2190_cast_fp16))[name = tensor("op_2315_cast_fp16")]; + tensor var_2317_equation_0 = const()[name = tensor("op_2317_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2317_cast_fp16 = einsum(equation = var_2317_equation_0, values = (var_1871_cast_fp16, var_2191_cast_fp16))[name = tensor("op_2317_cast_fp16")]; + tensor var_2319_equation_0 = const()[name = tensor("op_2319_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2319_cast_fp16 = einsum(equation = var_2319_equation_0, values = (var_1871_cast_fp16, var_2192_cast_fp16))[name = tensor("op_2319_cast_fp16")]; + tensor var_2321_equation_0 = const()[name = tensor("op_2321_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2321_cast_fp16 = einsum(equation = var_2321_equation_0, values = (var_1871_cast_fp16, var_2193_cast_fp16))[name = tensor("op_2321_cast_fp16")]; + tensor var_2323_interleave_0 = const()[name = tensor("op_2323_interleave_0"), val = tensor(false)]; + tensor var_2323_cast_fp16 = concat(axis = var_95, interleave = var_2323_interleave_0, values = (var_2195_cast_fp16, var_2197_cast_fp16, var_2199_cast_fp16, var_2201_cast_fp16, var_2203_cast_fp16, var_2205_cast_fp16, var_2207_cast_fp16, var_2209_cast_fp16))[name = tensor("op_2323_cast_fp16")]; + tensor var_2325_interleave_0 = const()[name = tensor("op_2325_interleave_0"), val = tensor(false)]; + tensor var_2325_cast_fp16 = concat(axis = var_95, interleave = var_2325_interleave_0, values = (var_2211_cast_fp16, var_2213_cast_fp16, var_2215_cast_fp16, var_2217_cast_fp16, var_2219_cast_fp16, var_2221_cast_fp16, var_2223_cast_fp16, var_2225_cast_fp16))[name = tensor("op_2325_cast_fp16")]; + tensor var_2327_interleave_0 = const()[name = tensor("op_2327_interleave_0"), val = tensor(false)]; + tensor var_2327_cast_fp16 = concat(axis = var_95, interleave = var_2327_interleave_0, values = (var_2227_cast_fp16, var_2229_cast_fp16, var_2231_cast_fp16, var_2233_cast_fp16, var_2235_cast_fp16, var_2237_cast_fp16, var_2239_cast_fp16, var_2241_cast_fp16))[name = tensor("op_2327_cast_fp16")]; + tensor var_2329_interleave_0 = const()[name = tensor("op_2329_interleave_0"), val = tensor(false)]; + tensor var_2329_cast_fp16 = concat(axis = var_95, interleave = var_2329_interleave_0, values = (var_2243_cast_fp16, var_2245_cast_fp16, var_2247_cast_fp16, var_2249_cast_fp16, var_2251_cast_fp16, var_2253_cast_fp16, var_2255_cast_fp16, var_2257_cast_fp16))[name = tensor("op_2329_cast_fp16")]; + tensor var_2331_interleave_0 = const()[name = tensor("op_2331_interleave_0"), val = tensor(false)]; + tensor var_2331_cast_fp16 = concat(axis = var_95, interleave = var_2331_interleave_0, values = (var_2259_cast_fp16, var_2261_cast_fp16, var_2263_cast_fp16, var_2265_cast_fp16, var_2267_cast_fp16, var_2269_cast_fp16, var_2271_cast_fp16, var_2273_cast_fp16))[name = tensor("op_2331_cast_fp16")]; + tensor var_2333_interleave_0 = const()[name = tensor("op_2333_interleave_0"), val = tensor(false)]; + tensor var_2333_cast_fp16 = concat(axis = var_95, interleave = var_2333_interleave_0, values = (var_2275_cast_fp16, var_2277_cast_fp16, var_2279_cast_fp16, var_2281_cast_fp16, var_2283_cast_fp16, var_2285_cast_fp16, var_2287_cast_fp16, var_2289_cast_fp16))[name = tensor("op_2333_cast_fp16")]; + tensor var_2335_interleave_0 = const()[name = tensor("op_2335_interleave_0"), val = tensor(false)]; + tensor var_2335_cast_fp16 = concat(axis = var_95, interleave = var_2335_interleave_0, values = (var_2291_cast_fp16, var_2293_cast_fp16, var_2295_cast_fp16, var_2297_cast_fp16, var_2299_cast_fp16, var_2301_cast_fp16, var_2303_cast_fp16, var_2305_cast_fp16))[name = tensor("op_2335_cast_fp16")]; + tensor var_2337_interleave_0 = const()[name = tensor("op_2337_interleave_0"), val = tensor(false)]; + tensor var_2337_cast_fp16 = concat(axis = var_95, interleave = var_2337_interleave_0, values = (var_2307_cast_fp16, var_2309_cast_fp16, var_2311_cast_fp16, var_2313_cast_fp16, var_2315_cast_fp16, var_2317_cast_fp16, var_2319_cast_fp16, var_2321_cast_fp16))[name = tensor("op_2337_cast_fp16")]; + tensor input_51_interleave_0 = const()[name = tensor("input_51_interleave_0"), val = tensor(false)]; + tensor input_51_cast_fp16 = concat(axis = var_123, interleave = input_51_interleave_0, values = (var_2323_cast_fp16, var_2325_cast_fp16, var_2327_cast_fp16, var_2329_cast_fp16, var_2331_cast_fp16, var_2333_cast_fp16, var_2335_cast_fp16, var_2337_cast_fp16))[name = tensor("input_51_cast_fp16")]; + tensor var_2347_pad_type_0 = const()[name = tensor("op_2347_pad_type_0"), val = tensor("valid")]; + tensor var_2347_strides_0 = const()[name = tensor("op_2347_strides_0"), val = tensor([1, 1])]; + tensor var_2347_pad_0 = const()[name = tensor("op_2347_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2347_dilations_0 = const()[name = tensor("op_2347_dilations_0"), val = tensor([1, 1])]; + tensor var_2347_groups_0 = const()[name = tensor("op_2347_groups_0"), val = tensor(1)]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19065600)))]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19270464)))]; + tensor var_2347_cast_fp16 = conv(bias = down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_2347_dilations_0, groups = var_2347_groups_0, pad = var_2347_pad_0, pad_type = var_2347_pad_type_0, strides = var_2347_strides_0, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_51_cast_fp16)[name = tensor("op_2347_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = var_2347_cast_fp16, y = inputs_7_cast_fp16)[name = tensor("inputs_9_cast_fp16")]; + tensor hidden_states_29_axes_0 = const()[name = tensor("hidden_states_29_axes_0"), val = tensor([1])]; + tensor hidden_states_29_gamma_0_to_fp16 = const()[name = tensor("hidden_states_29_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19271168)))]; + tensor hidden_states_29_beta_0_to_fp16 = const()[name = tensor("hidden_states_29_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19271872)))]; + tensor var_2357_to_fp16 = const()[name = tensor("op_2357_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_29_cast_fp16 = layer_norm(axes = hidden_states_29_axes_0, beta = hidden_states_29_beta_0_to_fp16, epsilon = var_2357_to_fp16, gamma = hidden_states_29_gamma_0_to_fp16, x = inputs_9_cast_fp16)[name = tensor("hidden_states_29_cast_fp16")]; + tensor q_7_pad_type_0 = const()[name = tensor("q_7_pad_type_0"), val = tensor("valid")]; + tensor q_7_strides_0 = const()[name = tensor("q_7_strides_0"), val = tensor([1, 1])]; + tensor q_7_pad_0 = const()[name = tensor("q_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_7_dilations_0 = const()[name = tensor("q_7_dilations_0"), val = tensor([1, 1])]; + tensor q_7_groups_0 = const()[name = tensor("q_7_groups_0"), val = tensor(1)]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19272576)))]; + tensor q_7_cast_fp16 = conv(dilations = q_7_dilations_0, groups = q_7_groups_0, pad = q_7_pad_0, pad_type = q_7_pad_type_0, strides = q_7_strides_0, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_29_cast_fp16)[name = tensor("q_7_cast_fp16")]; + tensor k_13_pad_type_0 = const()[name = tensor("k_13_pad_type_0"), val = tensor("valid")]; + tensor k_13_strides_0 = const()[name = tensor("k_13_strides_0"), val = tensor([1, 1])]; + tensor k_13_pad_0 = const()[name = tensor("k_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_13_dilations_0 = const()[name = tensor("k_13_dilations_0"), val = tensor([1, 1])]; + tensor k_13_groups_0 = const()[name = tensor("k_13_groups_0"), val = tensor(1)]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19477440)))]; + tensor k_13_cast_fp16 = conv(dilations = k_13_dilations_0, groups = k_13_groups_0, pad = k_13_pad_0, pad_type = k_13_pad_type_0, strides = k_13_strides_0, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_13_cast_fp16")]; + tensor v_7_pad_type_0 = const()[name = tensor("v_7_pad_type_0"), val = tensor("valid")]; + tensor v_7_strides_0 = const()[name = tensor("v_7_strides_0"), val = tensor([1, 1])]; + tensor v_7_pad_0 = const()[name = tensor("v_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_7_dilations_0 = const()[name = tensor("v_7_dilations_0"), val = tensor([1, 1])]; + tensor v_7_groups_0 = const()[name = tensor("v_7_groups_0"), val = tensor(1)]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19969024)))]; + tensor v_7_cast_fp16 = conv(dilations = v_7_dilations_0, groups = v_7_groups_0, pad = v_7_pad_0, pad_type = v_7_pad_type_0, strides = v_7_strides_0, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_7_cast_fp16")]; + tensor var_2390_begin_0 = const()[name = tensor("op_2390_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2390_end_0 = const()[name = tensor("op_2390_end_0"), val = tensor([2, 40, 1, 4096])]; + tensor var_2390_end_mask_0 = const()[name = tensor("op_2390_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2390_cast_fp16 = slice_by_index(begin = var_2390_begin_0, end = var_2390_end_0, end_mask = var_2390_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_2390_cast_fp16")]; + tensor var_2394_begin_0 = const()[name = tensor("op_2394_begin_0"), val = tensor([0, 40, 0, 0])]; + tensor var_2394_end_0 = const()[name = tensor("op_2394_end_0"), val = tensor([2, 80, 1, 4096])]; + tensor var_2394_end_mask_0 = const()[name = tensor("op_2394_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2394_cast_fp16 = slice_by_index(begin = var_2394_begin_0, end = var_2394_end_0, end_mask = var_2394_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_2394_cast_fp16")]; + tensor var_2398_begin_0 = const()[name = tensor("op_2398_begin_0"), val = tensor([0, 80, 0, 0])]; + tensor var_2398_end_0 = const()[name = tensor("op_2398_end_0"), val = tensor([2, 120, 1, 4096])]; + tensor var_2398_end_mask_0 = const()[name = tensor("op_2398_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2398_cast_fp16 = slice_by_index(begin = var_2398_begin_0, end = var_2398_end_0, end_mask = var_2398_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_2398_cast_fp16")]; + tensor var_2402_begin_0 = const()[name = tensor("op_2402_begin_0"), val = tensor([0, 120, 0, 0])]; + tensor var_2402_end_0 = const()[name = tensor("op_2402_end_0"), val = tensor([2, 160, 1, 4096])]; + tensor var_2402_end_mask_0 = const()[name = tensor("op_2402_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2402_cast_fp16 = slice_by_index(begin = var_2402_begin_0, end = var_2402_end_0, end_mask = var_2402_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_2402_cast_fp16")]; + tensor var_2406_begin_0 = const()[name = tensor("op_2406_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_2406_end_0 = const()[name = tensor("op_2406_end_0"), val = tensor([2, 200, 1, 4096])]; + tensor var_2406_end_mask_0 = const()[name = tensor("op_2406_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2406_cast_fp16 = slice_by_index(begin = var_2406_begin_0, end = var_2406_end_0, end_mask = var_2406_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_2406_cast_fp16")]; + tensor var_2410_begin_0 = const()[name = tensor("op_2410_begin_0"), val = tensor([0, 200, 0, 0])]; + tensor var_2410_end_0 = const()[name = tensor("op_2410_end_0"), val = tensor([2, 240, 1, 4096])]; + tensor var_2410_end_mask_0 = const()[name = tensor("op_2410_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2410_cast_fp16 = slice_by_index(begin = var_2410_begin_0, end = var_2410_end_0, end_mask = var_2410_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_2410_cast_fp16")]; + tensor var_2414_begin_0 = const()[name = tensor("op_2414_begin_0"), val = tensor([0, 240, 0, 0])]; + tensor var_2414_end_0 = const()[name = tensor("op_2414_end_0"), val = tensor([2, 280, 1, 4096])]; + tensor var_2414_end_mask_0 = const()[name = tensor("op_2414_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2414_cast_fp16 = slice_by_index(begin = var_2414_begin_0, end = var_2414_end_0, end_mask = var_2414_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_2414_cast_fp16")]; + tensor var_2418_begin_0 = const()[name = tensor("op_2418_begin_0"), val = tensor([0, 280, 0, 0])]; + tensor var_2418_end_0 = const()[name = tensor("op_2418_end_0"), val = tensor([2, 1, 1, 4096])]; + tensor var_2418_end_mask_0 = const()[name = tensor("op_2418_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2418_cast_fp16 = slice_by_index(begin = var_2418_begin_0, end = var_2418_end_0, end_mask = var_2418_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_2418_cast_fp16")]; + tensor var_2421_begin_0 = const()[name = tensor("op_2421_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2421_end_0 = const()[name = tensor("op_2421_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_2421_end_mask_0 = const()[name = tensor("op_2421_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2421_cast_fp16 = slice_by_index(begin = var_2421_begin_0, end = var_2421_end_0, end_mask = var_2421_end_mask_0, x = var_2390_cast_fp16)[name = tensor("op_2421_cast_fp16")]; + tensor var_2422_begin_0 = const()[name = tensor("op_2422_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2422_end_0 = const()[name = tensor("op_2422_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_2422_end_mask_0 = const()[name = tensor("op_2422_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2422_cast_fp16 = slice_by_index(begin = var_2422_begin_0, end = var_2422_end_0, end_mask = var_2422_end_mask_0, x = var_2390_cast_fp16)[name = tensor("op_2422_cast_fp16")]; + tensor var_2423_begin_0 = const()[name = tensor("op_2423_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_2423_end_0 = const()[name = tensor("op_2423_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_2423_end_mask_0 = const()[name = tensor("op_2423_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2423_cast_fp16 = slice_by_index(begin = var_2423_begin_0, end = var_2423_end_0, end_mask = var_2423_end_mask_0, x = var_2390_cast_fp16)[name = tensor("op_2423_cast_fp16")]; + tensor var_2424_begin_0 = const()[name = tensor("op_2424_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_2424_end_0 = const()[name = tensor("op_2424_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_2424_end_mask_0 = const()[name = tensor("op_2424_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2424_cast_fp16 = slice_by_index(begin = var_2424_begin_0, end = var_2424_end_0, end_mask = var_2424_end_mask_0, x = var_2390_cast_fp16)[name = tensor("op_2424_cast_fp16")]; + tensor var_2425_begin_0 = const()[name = tensor("op_2425_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_2425_end_0 = const()[name = tensor("op_2425_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_2425_end_mask_0 = const()[name = tensor("op_2425_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2425_cast_fp16 = slice_by_index(begin = var_2425_begin_0, end = var_2425_end_0, end_mask = var_2425_end_mask_0, x = var_2390_cast_fp16)[name = tensor("op_2425_cast_fp16")]; + tensor var_2426_begin_0 = const()[name = tensor("op_2426_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_2426_end_0 = const()[name = tensor("op_2426_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_2426_end_mask_0 = const()[name = tensor("op_2426_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2426_cast_fp16 = slice_by_index(begin = var_2426_begin_0, end = var_2426_end_0, end_mask = var_2426_end_mask_0, x = var_2390_cast_fp16)[name = tensor("op_2426_cast_fp16")]; + tensor var_2427_begin_0 = const()[name = tensor("op_2427_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_2427_end_0 = const()[name = tensor("op_2427_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_2427_end_mask_0 = const()[name = tensor("op_2427_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2427_cast_fp16 = slice_by_index(begin = var_2427_begin_0, end = var_2427_end_0, end_mask = var_2427_end_mask_0, x = var_2390_cast_fp16)[name = tensor("op_2427_cast_fp16")]; + tensor var_2428_begin_0 = const()[name = tensor("op_2428_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_2428_end_0 = const()[name = tensor("op_2428_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_2428_end_mask_0 = const()[name = tensor("op_2428_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2428_cast_fp16 = slice_by_index(begin = var_2428_begin_0, end = var_2428_end_0, end_mask = var_2428_end_mask_0, x = var_2390_cast_fp16)[name = tensor("op_2428_cast_fp16")]; + tensor var_2429_begin_0 = const()[name = tensor("op_2429_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2429_end_0 = const()[name = tensor("op_2429_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_2429_end_mask_0 = const()[name = tensor("op_2429_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2429_cast_fp16 = slice_by_index(begin = var_2429_begin_0, end = var_2429_end_0, end_mask = var_2429_end_mask_0, x = var_2394_cast_fp16)[name = tensor("op_2429_cast_fp16")]; + tensor var_2430_begin_0 = const()[name = tensor("op_2430_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2430_end_0 = const()[name = tensor("op_2430_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_2430_end_mask_0 = const()[name = tensor("op_2430_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2430_cast_fp16 = slice_by_index(begin = var_2430_begin_0, end = var_2430_end_0, end_mask = var_2430_end_mask_0, x = var_2394_cast_fp16)[name = tensor("op_2430_cast_fp16")]; + tensor var_2431_begin_0 = const()[name = tensor("op_2431_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_2431_end_0 = const()[name = tensor("op_2431_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_2431_end_mask_0 = const()[name = tensor("op_2431_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2431_cast_fp16 = slice_by_index(begin = var_2431_begin_0, end = var_2431_end_0, end_mask = var_2431_end_mask_0, x = var_2394_cast_fp16)[name = tensor("op_2431_cast_fp16")]; + tensor var_2432_begin_0 = const()[name = tensor("op_2432_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_2432_end_0 = const()[name = tensor("op_2432_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_2432_end_mask_0 = const()[name = tensor("op_2432_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2432_cast_fp16 = slice_by_index(begin = var_2432_begin_0, end = var_2432_end_0, end_mask = var_2432_end_mask_0, x = var_2394_cast_fp16)[name = tensor("op_2432_cast_fp16")]; + tensor var_2433_begin_0 = const()[name = tensor("op_2433_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_2433_end_0 = const()[name = tensor("op_2433_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_2433_end_mask_0 = const()[name = tensor("op_2433_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2433_cast_fp16 = slice_by_index(begin = var_2433_begin_0, end = var_2433_end_0, end_mask = var_2433_end_mask_0, x = var_2394_cast_fp16)[name = tensor("op_2433_cast_fp16")]; + tensor var_2434_begin_0 = const()[name = tensor("op_2434_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_2434_end_0 = const()[name = tensor("op_2434_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_2434_end_mask_0 = const()[name = tensor("op_2434_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2434_cast_fp16 = slice_by_index(begin = var_2434_begin_0, end = var_2434_end_0, end_mask = var_2434_end_mask_0, x = var_2394_cast_fp16)[name = tensor("op_2434_cast_fp16")]; + tensor var_2435_begin_0 = const()[name = tensor("op_2435_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_2435_end_0 = const()[name = tensor("op_2435_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_2435_end_mask_0 = const()[name = tensor("op_2435_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2435_cast_fp16 = slice_by_index(begin = var_2435_begin_0, end = var_2435_end_0, end_mask = var_2435_end_mask_0, x = var_2394_cast_fp16)[name = tensor("op_2435_cast_fp16")]; + tensor var_2436_begin_0 = const()[name = tensor("op_2436_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_2436_end_0 = const()[name = tensor("op_2436_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_2436_end_mask_0 = const()[name = tensor("op_2436_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2436_cast_fp16 = slice_by_index(begin = var_2436_begin_0, end = var_2436_end_0, end_mask = var_2436_end_mask_0, x = var_2394_cast_fp16)[name = tensor("op_2436_cast_fp16")]; + tensor var_2437_begin_0 = const()[name = tensor("op_2437_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2437_end_0 = const()[name = tensor("op_2437_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_2437_end_mask_0 = const()[name = tensor("op_2437_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2437_cast_fp16 = slice_by_index(begin = var_2437_begin_0, end = var_2437_end_0, end_mask = var_2437_end_mask_0, x = var_2398_cast_fp16)[name = tensor("op_2437_cast_fp16")]; + tensor var_2438_begin_0 = const()[name = tensor("op_2438_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2438_end_0 = const()[name = tensor("op_2438_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_2438_end_mask_0 = const()[name = tensor("op_2438_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2438_cast_fp16 = slice_by_index(begin = var_2438_begin_0, end = var_2438_end_0, end_mask = var_2438_end_mask_0, x = var_2398_cast_fp16)[name = tensor("op_2438_cast_fp16")]; + tensor var_2439_begin_0 = const()[name = tensor("op_2439_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_2439_end_0 = const()[name = tensor("op_2439_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_2439_end_mask_0 = const()[name = tensor("op_2439_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2439_cast_fp16 = slice_by_index(begin = var_2439_begin_0, end = var_2439_end_0, end_mask = var_2439_end_mask_0, x = var_2398_cast_fp16)[name = tensor("op_2439_cast_fp16")]; + tensor var_2440_begin_0 = const()[name = tensor("op_2440_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_2440_end_0 = const()[name = tensor("op_2440_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_2440_end_mask_0 = const()[name = tensor("op_2440_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2440_cast_fp16 = slice_by_index(begin = var_2440_begin_0, end = var_2440_end_0, end_mask = var_2440_end_mask_0, x = var_2398_cast_fp16)[name = tensor("op_2440_cast_fp16")]; + tensor var_2441_begin_0 = const()[name = tensor("op_2441_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_2441_end_0 = const()[name = tensor("op_2441_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_2441_end_mask_0 = const()[name = tensor("op_2441_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2441_cast_fp16 = slice_by_index(begin = var_2441_begin_0, end = var_2441_end_0, end_mask = var_2441_end_mask_0, x = var_2398_cast_fp16)[name = tensor("op_2441_cast_fp16")]; + tensor var_2442_begin_0 = const()[name = tensor("op_2442_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_2442_end_0 = const()[name = tensor("op_2442_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_2442_end_mask_0 = const()[name = tensor("op_2442_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2442_cast_fp16 = slice_by_index(begin = var_2442_begin_0, end = var_2442_end_0, end_mask = var_2442_end_mask_0, x = var_2398_cast_fp16)[name = tensor("op_2442_cast_fp16")]; + tensor var_2443_begin_0 = const()[name = tensor("op_2443_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_2443_end_0 = const()[name = tensor("op_2443_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_2443_end_mask_0 = const()[name = tensor("op_2443_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2443_cast_fp16 = slice_by_index(begin = var_2443_begin_0, end = var_2443_end_0, end_mask = var_2443_end_mask_0, x = var_2398_cast_fp16)[name = tensor("op_2443_cast_fp16")]; + tensor var_2444_begin_0 = const()[name = tensor("op_2444_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_2444_end_0 = const()[name = tensor("op_2444_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_2444_end_mask_0 = const()[name = tensor("op_2444_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2444_cast_fp16 = slice_by_index(begin = var_2444_begin_0, end = var_2444_end_0, end_mask = var_2444_end_mask_0, x = var_2398_cast_fp16)[name = tensor("op_2444_cast_fp16")]; + tensor var_2445_begin_0 = const()[name = tensor("op_2445_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2445_end_0 = const()[name = tensor("op_2445_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_2445_end_mask_0 = const()[name = tensor("op_2445_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2445_cast_fp16 = slice_by_index(begin = var_2445_begin_0, end = var_2445_end_0, end_mask = var_2445_end_mask_0, x = var_2402_cast_fp16)[name = tensor("op_2445_cast_fp16")]; + tensor var_2446_begin_0 = const()[name = tensor("op_2446_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2446_end_0 = const()[name = tensor("op_2446_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_2446_end_mask_0 = const()[name = tensor("op_2446_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2446_cast_fp16 = slice_by_index(begin = var_2446_begin_0, end = var_2446_end_0, end_mask = var_2446_end_mask_0, x = var_2402_cast_fp16)[name = tensor("op_2446_cast_fp16")]; + tensor var_2447_begin_0 = const()[name = tensor("op_2447_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_2447_end_0 = const()[name = tensor("op_2447_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_2447_end_mask_0 = const()[name = tensor("op_2447_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2447_cast_fp16 = slice_by_index(begin = var_2447_begin_0, end = var_2447_end_0, end_mask = var_2447_end_mask_0, x = var_2402_cast_fp16)[name = tensor("op_2447_cast_fp16")]; + tensor var_2448_begin_0 = const()[name = tensor("op_2448_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_2448_end_0 = const()[name = tensor("op_2448_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_2448_end_mask_0 = const()[name = tensor("op_2448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2448_cast_fp16 = slice_by_index(begin = var_2448_begin_0, end = var_2448_end_0, end_mask = var_2448_end_mask_0, x = var_2402_cast_fp16)[name = tensor("op_2448_cast_fp16")]; + tensor var_2449_begin_0 = const()[name = tensor("op_2449_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_2449_end_0 = const()[name = tensor("op_2449_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_2449_end_mask_0 = const()[name = tensor("op_2449_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2449_cast_fp16 = slice_by_index(begin = var_2449_begin_0, end = var_2449_end_0, end_mask = var_2449_end_mask_0, x = var_2402_cast_fp16)[name = tensor("op_2449_cast_fp16")]; + tensor var_2450_begin_0 = const()[name = tensor("op_2450_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_2450_end_0 = const()[name = tensor("op_2450_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_2450_end_mask_0 = const()[name = tensor("op_2450_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2450_cast_fp16 = slice_by_index(begin = var_2450_begin_0, end = var_2450_end_0, end_mask = var_2450_end_mask_0, x = var_2402_cast_fp16)[name = tensor("op_2450_cast_fp16")]; + tensor var_2451_begin_0 = const()[name = tensor("op_2451_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_2451_end_0 = const()[name = tensor("op_2451_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_2451_end_mask_0 = const()[name = tensor("op_2451_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2451_cast_fp16 = slice_by_index(begin = var_2451_begin_0, end = var_2451_end_0, end_mask = var_2451_end_mask_0, x = var_2402_cast_fp16)[name = tensor("op_2451_cast_fp16")]; + tensor var_2452_begin_0 = const()[name = tensor("op_2452_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_2452_end_0 = const()[name = tensor("op_2452_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_2452_end_mask_0 = const()[name = tensor("op_2452_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2452_cast_fp16 = slice_by_index(begin = var_2452_begin_0, end = var_2452_end_0, end_mask = var_2452_end_mask_0, x = var_2402_cast_fp16)[name = tensor("op_2452_cast_fp16")]; + tensor var_2453_begin_0 = const()[name = tensor("op_2453_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2453_end_0 = const()[name = tensor("op_2453_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_2453_end_mask_0 = const()[name = tensor("op_2453_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2453_cast_fp16 = slice_by_index(begin = var_2453_begin_0, end = var_2453_end_0, end_mask = var_2453_end_mask_0, x = var_2406_cast_fp16)[name = tensor("op_2453_cast_fp16")]; + tensor var_2454_begin_0 = const()[name = tensor("op_2454_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2454_end_0 = const()[name = tensor("op_2454_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_2454_end_mask_0 = const()[name = tensor("op_2454_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2454_cast_fp16 = slice_by_index(begin = var_2454_begin_0, end = var_2454_end_0, end_mask = var_2454_end_mask_0, x = var_2406_cast_fp16)[name = tensor("op_2454_cast_fp16")]; + tensor var_2455_begin_0 = const()[name = tensor("op_2455_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_2455_end_0 = const()[name = tensor("op_2455_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_2455_end_mask_0 = const()[name = tensor("op_2455_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2455_cast_fp16 = slice_by_index(begin = var_2455_begin_0, end = var_2455_end_0, end_mask = var_2455_end_mask_0, x = var_2406_cast_fp16)[name = tensor("op_2455_cast_fp16")]; + tensor var_2456_begin_0 = const()[name = tensor("op_2456_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_2456_end_0 = const()[name = tensor("op_2456_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_2456_end_mask_0 = const()[name = tensor("op_2456_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2456_cast_fp16 = slice_by_index(begin = var_2456_begin_0, end = var_2456_end_0, end_mask = var_2456_end_mask_0, x = var_2406_cast_fp16)[name = tensor("op_2456_cast_fp16")]; + tensor var_2457_begin_0 = const()[name = tensor("op_2457_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_2457_end_0 = const()[name = tensor("op_2457_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_2457_end_mask_0 = const()[name = tensor("op_2457_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2457_cast_fp16 = slice_by_index(begin = var_2457_begin_0, end = var_2457_end_0, end_mask = var_2457_end_mask_0, x = var_2406_cast_fp16)[name = tensor("op_2457_cast_fp16")]; + tensor var_2458_begin_0 = const()[name = tensor("op_2458_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_2458_end_0 = const()[name = tensor("op_2458_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_2458_end_mask_0 = const()[name = tensor("op_2458_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2458_cast_fp16 = slice_by_index(begin = var_2458_begin_0, end = var_2458_end_0, end_mask = var_2458_end_mask_0, x = var_2406_cast_fp16)[name = tensor("op_2458_cast_fp16")]; + tensor var_2459_begin_0 = const()[name = tensor("op_2459_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_2459_end_0 = const()[name = tensor("op_2459_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_2459_end_mask_0 = const()[name = tensor("op_2459_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2459_cast_fp16 = slice_by_index(begin = var_2459_begin_0, end = var_2459_end_0, end_mask = var_2459_end_mask_0, x = var_2406_cast_fp16)[name = tensor("op_2459_cast_fp16")]; + tensor var_2460_begin_0 = const()[name = tensor("op_2460_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_2460_end_0 = const()[name = tensor("op_2460_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_2460_end_mask_0 = const()[name = tensor("op_2460_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2460_cast_fp16 = slice_by_index(begin = var_2460_begin_0, end = var_2460_end_0, end_mask = var_2460_end_mask_0, x = var_2406_cast_fp16)[name = tensor("op_2460_cast_fp16")]; + tensor var_2461_begin_0 = const()[name = tensor("op_2461_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2461_end_0 = const()[name = tensor("op_2461_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_2461_end_mask_0 = const()[name = tensor("op_2461_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2461_cast_fp16 = slice_by_index(begin = var_2461_begin_0, end = var_2461_end_0, end_mask = var_2461_end_mask_0, x = var_2410_cast_fp16)[name = tensor("op_2461_cast_fp16")]; + tensor var_2462_begin_0 = const()[name = tensor("op_2462_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2462_end_0 = const()[name = tensor("op_2462_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_2462_end_mask_0 = const()[name = tensor("op_2462_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2462_cast_fp16 = slice_by_index(begin = var_2462_begin_0, end = var_2462_end_0, end_mask = var_2462_end_mask_0, x = var_2410_cast_fp16)[name = tensor("op_2462_cast_fp16")]; + tensor var_2463_begin_0 = const()[name = tensor("op_2463_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_2463_end_0 = const()[name = tensor("op_2463_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_2463_end_mask_0 = const()[name = tensor("op_2463_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2463_cast_fp16 = slice_by_index(begin = var_2463_begin_0, end = var_2463_end_0, end_mask = var_2463_end_mask_0, x = var_2410_cast_fp16)[name = tensor("op_2463_cast_fp16")]; + tensor var_2464_begin_0 = const()[name = tensor("op_2464_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_2464_end_0 = const()[name = tensor("op_2464_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_2464_end_mask_0 = const()[name = tensor("op_2464_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2464_cast_fp16 = slice_by_index(begin = var_2464_begin_0, end = var_2464_end_0, end_mask = var_2464_end_mask_0, x = var_2410_cast_fp16)[name = tensor("op_2464_cast_fp16")]; + tensor var_2465_begin_0 = const()[name = tensor("op_2465_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_2465_end_0 = const()[name = tensor("op_2465_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_2465_end_mask_0 = const()[name = tensor("op_2465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2465_cast_fp16 = slice_by_index(begin = var_2465_begin_0, end = var_2465_end_0, end_mask = var_2465_end_mask_0, x = var_2410_cast_fp16)[name = tensor("op_2465_cast_fp16")]; + tensor var_2466_begin_0 = const()[name = tensor("op_2466_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_2466_end_0 = const()[name = tensor("op_2466_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_2466_end_mask_0 = const()[name = tensor("op_2466_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2466_cast_fp16 = slice_by_index(begin = var_2466_begin_0, end = var_2466_end_0, end_mask = var_2466_end_mask_0, x = var_2410_cast_fp16)[name = tensor("op_2466_cast_fp16")]; + tensor var_2467_begin_0 = const()[name = tensor("op_2467_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_2467_end_0 = const()[name = tensor("op_2467_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_2467_end_mask_0 = const()[name = tensor("op_2467_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2467_cast_fp16 = slice_by_index(begin = var_2467_begin_0, end = var_2467_end_0, end_mask = var_2467_end_mask_0, x = var_2410_cast_fp16)[name = tensor("op_2467_cast_fp16")]; + tensor var_2468_begin_0 = const()[name = tensor("op_2468_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_2468_end_0 = const()[name = tensor("op_2468_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_2468_end_mask_0 = const()[name = tensor("op_2468_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2468_cast_fp16 = slice_by_index(begin = var_2468_begin_0, end = var_2468_end_0, end_mask = var_2468_end_mask_0, x = var_2410_cast_fp16)[name = tensor("op_2468_cast_fp16")]; + tensor var_2469_begin_0 = const()[name = tensor("op_2469_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2469_end_0 = const()[name = tensor("op_2469_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_2469_end_mask_0 = const()[name = tensor("op_2469_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2469_cast_fp16 = slice_by_index(begin = var_2469_begin_0, end = var_2469_end_0, end_mask = var_2469_end_mask_0, x = var_2414_cast_fp16)[name = tensor("op_2469_cast_fp16")]; + tensor var_2470_begin_0 = const()[name = tensor("op_2470_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2470_end_0 = const()[name = tensor("op_2470_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_2470_end_mask_0 = const()[name = tensor("op_2470_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2470_cast_fp16 = slice_by_index(begin = var_2470_begin_0, end = var_2470_end_0, end_mask = var_2470_end_mask_0, x = var_2414_cast_fp16)[name = tensor("op_2470_cast_fp16")]; + tensor var_2471_begin_0 = const()[name = tensor("op_2471_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_2471_end_0 = const()[name = tensor("op_2471_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_2471_end_mask_0 = const()[name = tensor("op_2471_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2471_cast_fp16 = slice_by_index(begin = var_2471_begin_0, end = var_2471_end_0, end_mask = var_2471_end_mask_0, x = var_2414_cast_fp16)[name = tensor("op_2471_cast_fp16")]; + tensor var_2472_begin_0 = const()[name = tensor("op_2472_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_2472_end_0 = const()[name = tensor("op_2472_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_2472_end_mask_0 = const()[name = tensor("op_2472_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2472_cast_fp16 = slice_by_index(begin = var_2472_begin_0, end = var_2472_end_0, end_mask = var_2472_end_mask_0, x = var_2414_cast_fp16)[name = tensor("op_2472_cast_fp16")]; + tensor var_2473_begin_0 = const()[name = tensor("op_2473_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_2473_end_0 = const()[name = tensor("op_2473_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_2473_end_mask_0 = const()[name = tensor("op_2473_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2473_cast_fp16 = slice_by_index(begin = var_2473_begin_0, end = var_2473_end_0, end_mask = var_2473_end_mask_0, x = var_2414_cast_fp16)[name = tensor("op_2473_cast_fp16")]; + tensor var_2474_begin_0 = const()[name = tensor("op_2474_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_2474_end_0 = const()[name = tensor("op_2474_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_2474_end_mask_0 = const()[name = tensor("op_2474_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2474_cast_fp16 = slice_by_index(begin = var_2474_begin_0, end = var_2474_end_0, end_mask = var_2474_end_mask_0, x = var_2414_cast_fp16)[name = tensor("op_2474_cast_fp16")]; + tensor var_2475_begin_0 = const()[name = tensor("op_2475_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_2475_end_0 = const()[name = tensor("op_2475_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_2475_end_mask_0 = const()[name = tensor("op_2475_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2475_cast_fp16 = slice_by_index(begin = var_2475_begin_0, end = var_2475_end_0, end_mask = var_2475_end_mask_0, x = var_2414_cast_fp16)[name = tensor("op_2475_cast_fp16")]; + tensor var_2476_begin_0 = const()[name = tensor("op_2476_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_2476_end_0 = const()[name = tensor("op_2476_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_2476_end_mask_0 = const()[name = tensor("op_2476_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2476_cast_fp16 = slice_by_index(begin = var_2476_begin_0, end = var_2476_end_0, end_mask = var_2476_end_mask_0, x = var_2414_cast_fp16)[name = tensor("op_2476_cast_fp16")]; + tensor var_2477_begin_0 = const()[name = tensor("op_2477_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2477_end_0 = const()[name = tensor("op_2477_end_0"), val = tensor([2, 40, 1, 512])]; + tensor var_2477_end_mask_0 = const()[name = tensor("op_2477_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2477_cast_fp16 = slice_by_index(begin = var_2477_begin_0, end = var_2477_end_0, end_mask = var_2477_end_mask_0, x = var_2418_cast_fp16)[name = tensor("op_2477_cast_fp16")]; + tensor var_2478_begin_0 = const()[name = tensor("op_2478_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2478_end_0 = const()[name = tensor("op_2478_end_0"), val = tensor([2, 40, 1, 1024])]; + tensor var_2478_end_mask_0 = const()[name = tensor("op_2478_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2478_cast_fp16 = slice_by_index(begin = var_2478_begin_0, end = var_2478_end_0, end_mask = var_2478_end_mask_0, x = var_2418_cast_fp16)[name = tensor("op_2478_cast_fp16")]; + tensor var_2479_begin_0 = const()[name = tensor("op_2479_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_2479_end_0 = const()[name = tensor("op_2479_end_0"), val = tensor([2, 40, 1, 1536])]; + tensor var_2479_end_mask_0 = const()[name = tensor("op_2479_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2479_cast_fp16 = slice_by_index(begin = var_2479_begin_0, end = var_2479_end_0, end_mask = var_2479_end_mask_0, x = var_2418_cast_fp16)[name = tensor("op_2479_cast_fp16")]; + tensor var_2480_begin_0 = const()[name = tensor("op_2480_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_2480_end_0 = const()[name = tensor("op_2480_end_0"), val = tensor([2, 40, 1, 2048])]; + tensor var_2480_end_mask_0 = const()[name = tensor("op_2480_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2480_cast_fp16 = slice_by_index(begin = var_2480_begin_0, end = var_2480_end_0, end_mask = var_2480_end_mask_0, x = var_2418_cast_fp16)[name = tensor("op_2480_cast_fp16")]; + tensor var_2481_begin_0 = const()[name = tensor("op_2481_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_2481_end_0 = const()[name = tensor("op_2481_end_0"), val = tensor([2, 40, 1, 2560])]; + tensor var_2481_end_mask_0 = const()[name = tensor("op_2481_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2481_cast_fp16 = slice_by_index(begin = var_2481_begin_0, end = var_2481_end_0, end_mask = var_2481_end_mask_0, x = var_2418_cast_fp16)[name = tensor("op_2481_cast_fp16")]; + tensor var_2482_begin_0 = const()[name = tensor("op_2482_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_2482_end_0 = const()[name = tensor("op_2482_end_0"), val = tensor([2, 40, 1, 3072])]; + tensor var_2482_end_mask_0 = const()[name = tensor("op_2482_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2482_cast_fp16 = slice_by_index(begin = var_2482_begin_0, end = var_2482_end_0, end_mask = var_2482_end_mask_0, x = var_2418_cast_fp16)[name = tensor("op_2482_cast_fp16")]; + tensor var_2483_begin_0 = const()[name = tensor("op_2483_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_2483_end_0 = const()[name = tensor("op_2483_end_0"), val = tensor([2, 40, 1, 3584])]; + tensor var_2483_end_mask_0 = const()[name = tensor("op_2483_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2483_cast_fp16 = slice_by_index(begin = var_2483_begin_0, end = var_2483_end_0, end_mask = var_2483_end_mask_0, x = var_2418_cast_fp16)[name = tensor("op_2483_cast_fp16")]; + tensor var_2484_begin_0 = const()[name = tensor("op_2484_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_2484_end_0 = const()[name = tensor("op_2484_end_0"), val = tensor([2, 40, 1, 1])]; + tensor var_2484_end_mask_0 = const()[name = tensor("op_2484_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2484_cast_fp16 = slice_by_index(begin = var_2484_begin_0, end = var_2484_end_0, end_mask = var_2484_end_mask_0, x = var_2418_cast_fp16)[name = tensor("op_2484_cast_fp16")]; + tensor k_15_perm_0 = const()[name = tensor("k_15_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_2489_begin_0 = const()[name = tensor("op_2489_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2489_end_0 = const()[name = tensor("op_2489_end_0"), val = tensor([2, 77, 1, 40])]; + tensor var_2489_end_mask_0 = const()[name = tensor("op_2489_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_15_cast_fp16 = transpose(perm = k_15_perm_0, x = k_13_cast_fp16)[name = tensor("transpose_10")]; + tensor var_2489_cast_fp16 = slice_by_index(begin = var_2489_begin_0, end = var_2489_end_0, end_mask = var_2489_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_2489_cast_fp16")]; + tensor var_2493_begin_0 = const()[name = tensor("op_2493_begin_0"), val = tensor([0, 0, 0, 40])]; + tensor var_2493_end_0 = const()[name = tensor("op_2493_end_0"), val = tensor([2, 77, 1, 80])]; + tensor var_2493_end_mask_0 = const()[name = tensor("op_2493_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2493_cast_fp16 = slice_by_index(begin = var_2493_begin_0, end = var_2493_end_0, end_mask = var_2493_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_2493_cast_fp16")]; + tensor var_2497_begin_0 = const()[name = tensor("op_2497_begin_0"), val = tensor([0, 0, 0, 80])]; + tensor var_2497_end_0 = const()[name = tensor("op_2497_end_0"), val = tensor([2, 77, 1, 120])]; + tensor var_2497_end_mask_0 = const()[name = tensor("op_2497_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2497_cast_fp16 = slice_by_index(begin = var_2497_begin_0, end = var_2497_end_0, end_mask = var_2497_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_2497_cast_fp16")]; + tensor var_2501_begin_0 = const()[name = tensor("op_2501_begin_0"), val = tensor([0, 0, 0, 120])]; + tensor var_2501_end_0 = const()[name = tensor("op_2501_end_0"), val = tensor([2, 77, 1, 160])]; + tensor var_2501_end_mask_0 = const()[name = tensor("op_2501_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2501_cast_fp16 = slice_by_index(begin = var_2501_begin_0, end = var_2501_end_0, end_mask = var_2501_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_2501_cast_fp16")]; + tensor var_2505_begin_0 = const()[name = tensor("op_2505_begin_0"), val = tensor([0, 0, 0, 160])]; + tensor var_2505_end_0 = const()[name = tensor("op_2505_end_0"), val = tensor([2, 77, 1, 200])]; + tensor var_2505_end_mask_0 = const()[name = tensor("op_2505_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2505_cast_fp16 = slice_by_index(begin = var_2505_begin_0, end = var_2505_end_0, end_mask = var_2505_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_2505_cast_fp16")]; + tensor var_2509_begin_0 = const()[name = tensor("op_2509_begin_0"), val = tensor([0, 0, 0, 200])]; + tensor var_2509_end_0 = const()[name = tensor("op_2509_end_0"), val = tensor([2, 77, 1, 240])]; + tensor var_2509_end_mask_0 = const()[name = tensor("op_2509_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2509_cast_fp16 = slice_by_index(begin = var_2509_begin_0, end = var_2509_end_0, end_mask = var_2509_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_2509_cast_fp16")]; + tensor var_2513_begin_0 = const()[name = tensor("op_2513_begin_0"), val = tensor([0, 0, 0, 240])]; + tensor var_2513_end_0 = const()[name = tensor("op_2513_end_0"), val = tensor([2, 77, 1, 280])]; + tensor var_2513_end_mask_0 = const()[name = tensor("op_2513_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2513_cast_fp16 = slice_by_index(begin = var_2513_begin_0, end = var_2513_end_0, end_mask = var_2513_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_2513_cast_fp16")]; + tensor var_2517_begin_0 = const()[name = tensor("op_2517_begin_0"), val = tensor([0, 0, 0, 280])]; + tensor var_2517_end_0 = const()[name = tensor("op_2517_end_0"), val = tensor([2, 77, 1, 1])]; + tensor var_2517_end_mask_0 = const()[name = tensor("op_2517_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2517_cast_fp16 = slice_by_index(begin = var_2517_begin_0, end = var_2517_end_0, end_mask = var_2517_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_2517_cast_fp16")]; + tensor var_2519_begin_0 = const()[name = tensor("op_2519_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2519_end_0 = const()[name = tensor("op_2519_end_0"), val = tensor([2, 40, 1, 77])]; + tensor var_2519_end_mask_0 = const()[name = tensor("op_2519_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2519_cast_fp16 = slice_by_index(begin = var_2519_begin_0, end = var_2519_end_0, end_mask = var_2519_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_2519_cast_fp16")]; + tensor var_2523_begin_0 = const()[name = tensor("op_2523_begin_0"), val = tensor([0, 40, 0, 0])]; + tensor var_2523_end_0 = const()[name = tensor("op_2523_end_0"), val = tensor([2, 80, 1, 77])]; + tensor var_2523_end_mask_0 = const()[name = tensor("op_2523_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2523_cast_fp16 = slice_by_index(begin = var_2523_begin_0, end = var_2523_end_0, end_mask = var_2523_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_2523_cast_fp16")]; + tensor var_2527_begin_0 = const()[name = tensor("op_2527_begin_0"), val = tensor([0, 80, 0, 0])]; + tensor var_2527_end_0 = const()[name = tensor("op_2527_end_0"), val = tensor([2, 120, 1, 77])]; + tensor var_2527_end_mask_0 = const()[name = tensor("op_2527_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2527_cast_fp16 = slice_by_index(begin = var_2527_begin_0, end = var_2527_end_0, end_mask = var_2527_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_2527_cast_fp16")]; + tensor var_2531_begin_0 = const()[name = tensor("op_2531_begin_0"), val = tensor([0, 120, 0, 0])]; + tensor var_2531_end_0 = const()[name = tensor("op_2531_end_0"), val = tensor([2, 160, 1, 77])]; + tensor var_2531_end_mask_0 = const()[name = tensor("op_2531_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2531_cast_fp16 = slice_by_index(begin = var_2531_begin_0, end = var_2531_end_0, end_mask = var_2531_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_2531_cast_fp16")]; + tensor var_2535_begin_0 = const()[name = tensor("op_2535_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_2535_end_0 = const()[name = tensor("op_2535_end_0"), val = tensor([2, 200, 1, 77])]; + tensor var_2535_end_mask_0 = const()[name = tensor("op_2535_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2535_cast_fp16 = slice_by_index(begin = var_2535_begin_0, end = var_2535_end_0, end_mask = var_2535_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_2535_cast_fp16")]; + tensor var_2539_begin_0 = const()[name = tensor("op_2539_begin_0"), val = tensor([0, 200, 0, 0])]; + tensor var_2539_end_0 = const()[name = tensor("op_2539_end_0"), val = tensor([2, 240, 1, 77])]; + tensor var_2539_end_mask_0 = const()[name = tensor("op_2539_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2539_cast_fp16 = slice_by_index(begin = var_2539_begin_0, end = var_2539_end_0, end_mask = var_2539_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_2539_cast_fp16")]; + tensor var_2543_begin_0 = const()[name = tensor("op_2543_begin_0"), val = tensor([0, 240, 0, 0])]; + tensor var_2543_end_0 = const()[name = tensor("op_2543_end_0"), val = tensor([2, 280, 1, 77])]; + tensor var_2543_end_mask_0 = const()[name = tensor("op_2543_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2543_cast_fp16 = slice_by_index(begin = var_2543_begin_0, end = var_2543_end_0, end_mask = var_2543_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_2543_cast_fp16")]; + tensor var_2547_begin_0 = const()[name = tensor("op_2547_begin_0"), val = tensor([0, 280, 0, 0])]; + tensor var_2547_end_0 = const()[name = tensor("op_2547_end_0"), val = tensor([2, 1, 1, 77])]; + tensor var_2547_end_mask_0 = const()[name = tensor("op_2547_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2547_cast_fp16 = slice_by_index(begin = var_2547_begin_0, end = var_2547_end_0, end_mask = var_2547_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_2547_cast_fp16")]; + tensor var_2551_equation_0 = const()[name = tensor("op_2551_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2551_cast_fp16 = einsum(equation = var_2551_equation_0, values = (var_2489_cast_fp16, var_2421_cast_fp16))[name = tensor("op_2551_cast_fp16")]; + tensor var_2552_to_fp16 = const()[name = tensor("op_2552_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_385_cast_fp16 = mul(x = var_2551_cast_fp16, y = var_2552_to_fp16)[name = tensor("aw_chunk_385_cast_fp16")]; + tensor var_2555_equation_0 = const()[name = tensor("op_2555_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2555_cast_fp16 = einsum(equation = var_2555_equation_0, values = (var_2489_cast_fp16, var_2422_cast_fp16))[name = tensor("op_2555_cast_fp16")]; + tensor var_2556_to_fp16 = const()[name = tensor("op_2556_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_387_cast_fp16 = mul(x = var_2555_cast_fp16, y = var_2556_to_fp16)[name = tensor("aw_chunk_387_cast_fp16")]; + tensor var_2559_equation_0 = const()[name = tensor("op_2559_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2559_cast_fp16 = einsum(equation = var_2559_equation_0, values = (var_2489_cast_fp16, var_2423_cast_fp16))[name = tensor("op_2559_cast_fp16")]; + tensor var_2560_to_fp16 = const()[name = tensor("op_2560_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_389_cast_fp16 = mul(x = var_2559_cast_fp16, y = var_2560_to_fp16)[name = tensor("aw_chunk_389_cast_fp16")]; + tensor var_2563_equation_0 = const()[name = tensor("op_2563_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2563_cast_fp16 = einsum(equation = var_2563_equation_0, values = (var_2489_cast_fp16, var_2424_cast_fp16))[name = tensor("op_2563_cast_fp16")]; + tensor var_2564_to_fp16 = const()[name = tensor("op_2564_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_391_cast_fp16 = mul(x = var_2563_cast_fp16, y = var_2564_to_fp16)[name = tensor("aw_chunk_391_cast_fp16")]; + tensor var_2567_equation_0 = const()[name = tensor("op_2567_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2567_cast_fp16 = einsum(equation = var_2567_equation_0, values = (var_2489_cast_fp16, var_2425_cast_fp16))[name = tensor("op_2567_cast_fp16")]; + tensor var_2568_to_fp16 = const()[name = tensor("op_2568_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_393_cast_fp16 = mul(x = var_2567_cast_fp16, y = var_2568_to_fp16)[name = tensor("aw_chunk_393_cast_fp16")]; + tensor var_2571_equation_0 = const()[name = tensor("op_2571_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2571_cast_fp16 = einsum(equation = var_2571_equation_0, values = (var_2489_cast_fp16, var_2426_cast_fp16))[name = tensor("op_2571_cast_fp16")]; + tensor var_2572_to_fp16 = const()[name = tensor("op_2572_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_395_cast_fp16 = mul(x = var_2571_cast_fp16, y = var_2572_to_fp16)[name = tensor("aw_chunk_395_cast_fp16")]; + tensor var_2575_equation_0 = const()[name = tensor("op_2575_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2575_cast_fp16 = einsum(equation = var_2575_equation_0, values = (var_2489_cast_fp16, var_2427_cast_fp16))[name = tensor("op_2575_cast_fp16")]; + tensor var_2576_to_fp16 = const()[name = tensor("op_2576_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_397_cast_fp16 = mul(x = var_2575_cast_fp16, y = var_2576_to_fp16)[name = tensor("aw_chunk_397_cast_fp16")]; + tensor var_2579_equation_0 = const()[name = tensor("op_2579_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2579_cast_fp16 = einsum(equation = var_2579_equation_0, values = (var_2489_cast_fp16, var_2428_cast_fp16))[name = tensor("op_2579_cast_fp16")]; + tensor var_2580_to_fp16 = const()[name = tensor("op_2580_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_399_cast_fp16 = mul(x = var_2579_cast_fp16, y = var_2580_to_fp16)[name = tensor("aw_chunk_399_cast_fp16")]; + tensor var_2583_equation_0 = const()[name = tensor("op_2583_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2583_cast_fp16 = einsum(equation = var_2583_equation_0, values = (var_2493_cast_fp16, var_2429_cast_fp16))[name = tensor("op_2583_cast_fp16")]; + tensor var_2584_to_fp16 = const()[name = tensor("op_2584_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_401_cast_fp16 = mul(x = var_2583_cast_fp16, y = var_2584_to_fp16)[name = tensor("aw_chunk_401_cast_fp16")]; + tensor var_2587_equation_0 = const()[name = tensor("op_2587_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2587_cast_fp16 = einsum(equation = var_2587_equation_0, values = (var_2493_cast_fp16, var_2430_cast_fp16))[name = tensor("op_2587_cast_fp16")]; + tensor var_2588_to_fp16 = const()[name = tensor("op_2588_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_403_cast_fp16 = mul(x = var_2587_cast_fp16, y = var_2588_to_fp16)[name = tensor("aw_chunk_403_cast_fp16")]; + tensor var_2591_equation_0 = const()[name = tensor("op_2591_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2591_cast_fp16 = einsum(equation = var_2591_equation_0, values = (var_2493_cast_fp16, var_2431_cast_fp16))[name = tensor("op_2591_cast_fp16")]; + tensor var_2592_to_fp16 = const()[name = tensor("op_2592_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_405_cast_fp16 = mul(x = var_2591_cast_fp16, y = var_2592_to_fp16)[name = tensor("aw_chunk_405_cast_fp16")]; + tensor var_2595_equation_0 = const()[name = tensor("op_2595_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2595_cast_fp16 = einsum(equation = var_2595_equation_0, values = (var_2493_cast_fp16, var_2432_cast_fp16))[name = tensor("op_2595_cast_fp16")]; + tensor var_2596_to_fp16 = const()[name = tensor("op_2596_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_407_cast_fp16 = mul(x = var_2595_cast_fp16, y = var_2596_to_fp16)[name = tensor("aw_chunk_407_cast_fp16")]; + tensor var_2599_equation_0 = const()[name = tensor("op_2599_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2599_cast_fp16 = einsum(equation = var_2599_equation_0, values = (var_2493_cast_fp16, var_2433_cast_fp16))[name = tensor("op_2599_cast_fp16")]; + tensor var_2600_to_fp16 = const()[name = tensor("op_2600_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_409_cast_fp16 = mul(x = var_2599_cast_fp16, y = var_2600_to_fp16)[name = tensor("aw_chunk_409_cast_fp16")]; + tensor var_2603_equation_0 = const()[name = tensor("op_2603_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2603_cast_fp16 = einsum(equation = var_2603_equation_0, values = (var_2493_cast_fp16, var_2434_cast_fp16))[name = tensor("op_2603_cast_fp16")]; + tensor var_2604_to_fp16 = const()[name = tensor("op_2604_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_411_cast_fp16 = mul(x = var_2603_cast_fp16, y = var_2604_to_fp16)[name = tensor("aw_chunk_411_cast_fp16")]; + tensor var_2607_equation_0 = const()[name = tensor("op_2607_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2607_cast_fp16 = einsum(equation = var_2607_equation_0, values = (var_2493_cast_fp16, var_2435_cast_fp16))[name = tensor("op_2607_cast_fp16")]; + tensor var_2608_to_fp16 = const()[name = tensor("op_2608_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_413_cast_fp16 = mul(x = var_2607_cast_fp16, y = var_2608_to_fp16)[name = tensor("aw_chunk_413_cast_fp16")]; + tensor var_2611_equation_0 = const()[name = tensor("op_2611_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2611_cast_fp16 = einsum(equation = var_2611_equation_0, values = (var_2493_cast_fp16, var_2436_cast_fp16))[name = tensor("op_2611_cast_fp16")]; + tensor var_2612_to_fp16 = const()[name = tensor("op_2612_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_415_cast_fp16 = mul(x = var_2611_cast_fp16, y = var_2612_to_fp16)[name = tensor("aw_chunk_415_cast_fp16")]; + tensor var_2615_equation_0 = const()[name = tensor("op_2615_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2615_cast_fp16 = einsum(equation = var_2615_equation_0, values = (var_2497_cast_fp16, var_2437_cast_fp16))[name = tensor("op_2615_cast_fp16")]; + tensor var_2616_to_fp16 = const()[name = tensor("op_2616_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_417_cast_fp16 = mul(x = var_2615_cast_fp16, y = var_2616_to_fp16)[name = tensor("aw_chunk_417_cast_fp16")]; + tensor var_2619_equation_0 = const()[name = tensor("op_2619_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2619_cast_fp16 = einsum(equation = var_2619_equation_0, values = (var_2497_cast_fp16, var_2438_cast_fp16))[name = tensor("op_2619_cast_fp16")]; + tensor var_2620_to_fp16 = const()[name = tensor("op_2620_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_419_cast_fp16 = mul(x = var_2619_cast_fp16, y = var_2620_to_fp16)[name = tensor("aw_chunk_419_cast_fp16")]; + tensor var_2623_equation_0 = const()[name = tensor("op_2623_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2623_cast_fp16 = einsum(equation = var_2623_equation_0, values = (var_2497_cast_fp16, var_2439_cast_fp16))[name = tensor("op_2623_cast_fp16")]; + tensor var_2624_to_fp16 = const()[name = tensor("op_2624_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_421_cast_fp16 = mul(x = var_2623_cast_fp16, y = var_2624_to_fp16)[name = tensor("aw_chunk_421_cast_fp16")]; + tensor var_2627_equation_0 = const()[name = tensor("op_2627_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2627_cast_fp16 = einsum(equation = var_2627_equation_0, values = (var_2497_cast_fp16, var_2440_cast_fp16))[name = tensor("op_2627_cast_fp16")]; + tensor var_2628_to_fp16 = const()[name = tensor("op_2628_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_423_cast_fp16 = mul(x = var_2627_cast_fp16, y = var_2628_to_fp16)[name = tensor("aw_chunk_423_cast_fp16")]; + tensor var_2631_equation_0 = const()[name = tensor("op_2631_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2631_cast_fp16 = einsum(equation = var_2631_equation_0, values = (var_2497_cast_fp16, var_2441_cast_fp16))[name = tensor("op_2631_cast_fp16")]; + tensor var_2632_to_fp16 = const()[name = tensor("op_2632_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_425_cast_fp16 = mul(x = var_2631_cast_fp16, y = var_2632_to_fp16)[name = tensor("aw_chunk_425_cast_fp16")]; + tensor var_2635_equation_0 = const()[name = tensor("op_2635_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2635_cast_fp16 = einsum(equation = var_2635_equation_0, values = (var_2497_cast_fp16, var_2442_cast_fp16))[name = tensor("op_2635_cast_fp16")]; + tensor var_2636_to_fp16 = const()[name = tensor("op_2636_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_427_cast_fp16 = mul(x = var_2635_cast_fp16, y = var_2636_to_fp16)[name = tensor("aw_chunk_427_cast_fp16")]; + tensor var_2639_equation_0 = const()[name = tensor("op_2639_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2639_cast_fp16 = einsum(equation = var_2639_equation_0, values = (var_2497_cast_fp16, var_2443_cast_fp16))[name = tensor("op_2639_cast_fp16")]; + tensor var_2640_to_fp16 = const()[name = tensor("op_2640_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_429_cast_fp16 = mul(x = var_2639_cast_fp16, y = var_2640_to_fp16)[name = tensor("aw_chunk_429_cast_fp16")]; + tensor var_2643_equation_0 = const()[name = tensor("op_2643_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2643_cast_fp16 = einsum(equation = var_2643_equation_0, values = (var_2497_cast_fp16, var_2444_cast_fp16))[name = tensor("op_2643_cast_fp16")]; + tensor var_2644_to_fp16 = const()[name = tensor("op_2644_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_431_cast_fp16 = mul(x = var_2643_cast_fp16, y = var_2644_to_fp16)[name = tensor("aw_chunk_431_cast_fp16")]; + tensor var_2647_equation_0 = const()[name = tensor("op_2647_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2647_cast_fp16 = einsum(equation = var_2647_equation_0, values = (var_2501_cast_fp16, var_2445_cast_fp16))[name = tensor("op_2647_cast_fp16")]; + tensor var_2648_to_fp16 = const()[name = tensor("op_2648_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_433_cast_fp16 = mul(x = var_2647_cast_fp16, y = var_2648_to_fp16)[name = tensor("aw_chunk_433_cast_fp16")]; + tensor var_2651_equation_0 = const()[name = tensor("op_2651_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2651_cast_fp16 = einsum(equation = var_2651_equation_0, values = (var_2501_cast_fp16, var_2446_cast_fp16))[name = tensor("op_2651_cast_fp16")]; + tensor var_2652_to_fp16 = const()[name = tensor("op_2652_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_435_cast_fp16 = mul(x = var_2651_cast_fp16, y = var_2652_to_fp16)[name = tensor("aw_chunk_435_cast_fp16")]; + tensor var_2655_equation_0 = const()[name = tensor("op_2655_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2655_cast_fp16 = einsum(equation = var_2655_equation_0, values = (var_2501_cast_fp16, var_2447_cast_fp16))[name = tensor("op_2655_cast_fp16")]; + tensor var_2656_to_fp16 = const()[name = tensor("op_2656_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_437_cast_fp16 = mul(x = var_2655_cast_fp16, y = var_2656_to_fp16)[name = tensor("aw_chunk_437_cast_fp16")]; + tensor var_2659_equation_0 = const()[name = tensor("op_2659_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2659_cast_fp16 = einsum(equation = var_2659_equation_0, values = (var_2501_cast_fp16, var_2448_cast_fp16))[name = tensor("op_2659_cast_fp16")]; + tensor var_2660_to_fp16 = const()[name = tensor("op_2660_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_439_cast_fp16 = mul(x = var_2659_cast_fp16, y = var_2660_to_fp16)[name = tensor("aw_chunk_439_cast_fp16")]; + tensor var_2663_equation_0 = const()[name = tensor("op_2663_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2663_cast_fp16 = einsum(equation = var_2663_equation_0, values = (var_2501_cast_fp16, var_2449_cast_fp16))[name = tensor("op_2663_cast_fp16")]; + tensor var_2664_to_fp16 = const()[name = tensor("op_2664_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_441_cast_fp16 = mul(x = var_2663_cast_fp16, y = var_2664_to_fp16)[name = tensor("aw_chunk_441_cast_fp16")]; + tensor var_2667_equation_0 = const()[name = tensor("op_2667_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2667_cast_fp16 = einsum(equation = var_2667_equation_0, values = (var_2501_cast_fp16, var_2450_cast_fp16))[name = tensor("op_2667_cast_fp16")]; + tensor var_2668_to_fp16 = const()[name = tensor("op_2668_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_443_cast_fp16 = mul(x = var_2667_cast_fp16, y = var_2668_to_fp16)[name = tensor("aw_chunk_443_cast_fp16")]; + tensor var_2671_equation_0 = const()[name = tensor("op_2671_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2671_cast_fp16 = einsum(equation = var_2671_equation_0, values = (var_2501_cast_fp16, var_2451_cast_fp16))[name = tensor("op_2671_cast_fp16")]; + tensor var_2672_to_fp16 = const()[name = tensor("op_2672_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_445_cast_fp16 = mul(x = var_2671_cast_fp16, y = var_2672_to_fp16)[name = tensor("aw_chunk_445_cast_fp16")]; + tensor var_2675_equation_0 = const()[name = tensor("op_2675_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2675_cast_fp16 = einsum(equation = var_2675_equation_0, values = (var_2501_cast_fp16, var_2452_cast_fp16))[name = tensor("op_2675_cast_fp16")]; + tensor var_2676_to_fp16 = const()[name = tensor("op_2676_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_447_cast_fp16 = mul(x = var_2675_cast_fp16, y = var_2676_to_fp16)[name = tensor("aw_chunk_447_cast_fp16")]; + tensor var_2679_equation_0 = const()[name = tensor("op_2679_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2679_cast_fp16 = einsum(equation = var_2679_equation_0, values = (var_2505_cast_fp16, var_2453_cast_fp16))[name = tensor("op_2679_cast_fp16")]; + tensor var_2680_to_fp16 = const()[name = tensor("op_2680_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_449_cast_fp16 = mul(x = var_2679_cast_fp16, y = var_2680_to_fp16)[name = tensor("aw_chunk_449_cast_fp16")]; + tensor var_2683_equation_0 = const()[name = tensor("op_2683_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2683_cast_fp16 = einsum(equation = var_2683_equation_0, values = (var_2505_cast_fp16, var_2454_cast_fp16))[name = tensor("op_2683_cast_fp16")]; + tensor var_2684_to_fp16 = const()[name = tensor("op_2684_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_451_cast_fp16 = mul(x = var_2683_cast_fp16, y = var_2684_to_fp16)[name = tensor("aw_chunk_451_cast_fp16")]; + tensor var_2687_equation_0 = const()[name = tensor("op_2687_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2687_cast_fp16 = einsum(equation = var_2687_equation_0, values = (var_2505_cast_fp16, var_2455_cast_fp16))[name = tensor("op_2687_cast_fp16")]; + tensor var_2688_to_fp16 = const()[name = tensor("op_2688_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_453_cast_fp16 = mul(x = var_2687_cast_fp16, y = var_2688_to_fp16)[name = tensor("aw_chunk_453_cast_fp16")]; + tensor var_2691_equation_0 = const()[name = tensor("op_2691_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2691_cast_fp16 = einsum(equation = var_2691_equation_0, values = (var_2505_cast_fp16, var_2456_cast_fp16))[name = tensor("op_2691_cast_fp16")]; + tensor var_2692_to_fp16 = const()[name = tensor("op_2692_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_455_cast_fp16 = mul(x = var_2691_cast_fp16, y = var_2692_to_fp16)[name = tensor("aw_chunk_455_cast_fp16")]; + tensor var_2695_equation_0 = const()[name = tensor("op_2695_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2695_cast_fp16 = einsum(equation = var_2695_equation_0, values = (var_2505_cast_fp16, var_2457_cast_fp16))[name = tensor("op_2695_cast_fp16")]; + tensor var_2696_to_fp16 = const()[name = tensor("op_2696_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_457_cast_fp16 = mul(x = var_2695_cast_fp16, y = var_2696_to_fp16)[name = tensor("aw_chunk_457_cast_fp16")]; + tensor var_2699_equation_0 = const()[name = tensor("op_2699_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2699_cast_fp16 = einsum(equation = var_2699_equation_0, values = (var_2505_cast_fp16, var_2458_cast_fp16))[name = tensor("op_2699_cast_fp16")]; + tensor var_2700_to_fp16 = const()[name = tensor("op_2700_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_459_cast_fp16 = mul(x = var_2699_cast_fp16, y = var_2700_to_fp16)[name = tensor("aw_chunk_459_cast_fp16")]; + tensor var_2703_equation_0 = const()[name = tensor("op_2703_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2703_cast_fp16 = einsum(equation = var_2703_equation_0, values = (var_2505_cast_fp16, var_2459_cast_fp16))[name = tensor("op_2703_cast_fp16")]; + tensor var_2704_to_fp16 = const()[name = tensor("op_2704_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_461_cast_fp16 = mul(x = var_2703_cast_fp16, y = var_2704_to_fp16)[name = tensor("aw_chunk_461_cast_fp16")]; + tensor var_2707_equation_0 = const()[name = tensor("op_2707_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2707_cast_fp16 = einsum(equation = var_2707_equation_0, values = (var_2505_cast_fp16, var_2460_cast_fp16))[name = tensor("op_2707_cast_fp16")]; + tensor var_2708_to_fp16 = const()[name = tensor("op_2708_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_463_cast_fp16 = mul(x = var_2707_cast_fp16, y = var_2708_to_fp16)[name = tensor("aw_chunk_463_cast_fp16")]; + tensor var_2711_equation_0 = const()[name = tensor("op_2711_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2711_cast_fp16 = einsum(equation = var_2711_equation_0, values = (var_2509_cast_fp16, var_2461_cast_fp16))[name = tensor("op_2711_cast_fp16")]; + tensor var_2712_to_fp16 = const()[name = tensor("op_2712_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_465_cast_fp16 = mul(x = var_2711_cast_fp16, y = var_2712_to_fp16)[name = tensor("aw_chunk_465_cast_fp16")]; + tensor var_2715_equation_0 = const()[name = tensor("op_2715_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2715_cast_fp16 = einsum(equation = var_2715_equation_0, values = (var_2509_cast_fp16, var_2462_cast_fp16))[name = tensor("op_2715_cast_fp16")]; + tensor var_2716_to_fp16 = const()[name = tensor("op_2716_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_467_cast_fp16 = mul(x = var_2715_cast_fp16, y = var_2716_to_fp16)[name = tensor("aw_chunk_467_cast_fp16")]; + tensor var_2719_equation_0 = const()[name = tensor("op_2719_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2719_cast_fp16 = einsum(equation = var_2719_equation_0, values = (var_2509_cast_fp16, var_2463_cast_fp16))[name = tensor("op_2719_cast_fp16")]; + tensor var_2720_to_fp16 = const()[name = tensor("op_2720_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_469_cast_fp16 = mul(x = var_2719_cast_fp16, y = var_2720_to_fp16)[name = tensor("aw_chunk_469_cast_fp16")]; + tensor var_2723_equation_0 = const()[name = tensor("op_2723_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2723_cast_fp16 = einsum(equation = var_2723_equation_0, values = (var_2509_cast_fp16, var_2464_cast_fp16))[name = tensor("op_2723_cast_fp16")]; + tensor var_2724_to_fp16 = const()[name = tensor("op_2724_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_471_cast_fp16 = mul(x = var_2723_cast_fp16, y = var_2724_to_fp16)[name = tensor("aw_chunk_471_cast_fp16")]; + tensor var_2727_equation_0 = const()[name = tensor("op_2727_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2727_cast_fp16 = einsum(equation = var_2727_equation_0, values = (var_2509_cast_fp16, var_2465_cast_fp16))[name = tensor("op_2727_cast_fp16")]; + tensor var_2728_to_fp16 = const()[name = tensor("op_2728_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_473_cast_fp16 = mul(x = var_2727_cast_fp16, y = var_2728_to_fp16)[name = tensor("aw_chunk_473_cast_fp16")]; + tensor var_2731_equation_0 = const()[name = tensor("op_2731_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2731_cast_fp16 = einsum(equation = var_2731_equation_0, values = (var_2509_cast_fp16, var_2466_cast_fp16))[name = tensor("op_2731_cast_fp16")]; + tensor var_2732_to_fp16 = const()[name = tensor("op_2732_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_475_cast_fp16 = mul(x = var_2731_cast_fp16, y = var_2732_to_fp16)[name = tensor("aw_chunk_475_cast_fp16")]; + tensor var_2735_equation_0 = const()[name = tensor("op_2735_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2735_cast_fp16 = einsum(equation = var_2735_equation_0, values = (var_2509_cast_fp16, var_2467_cast_fp16))[name = tensor("op_2735_cast_fp16")]; + tensor var_2736_to_fp16 = const()[name = tensor("op_2736_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_477_cast_fp16 = mul(x = var_2735_cast_fp16, y = var_2736_to_fp16)[name = tensor("aw_chunk_477_cast_fp16")]; + tensor var_2739_equation_0 = const()[name = tensor("op_2739_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2739_cast_fp16 = einsum(equation = var_2739_equation_0, values = (var_2509_cast_fp16, var_2468_cast_fp16))[name = tensor("op_2739_cast_fp16")]; + tensor var_2740_to_fp16 = const()[name = tensor("op_2740_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_479_cast_fp16 = mul(x = var_2739_cast_fp16, y = var_2740_to_fp16)[name = tensor("aw_chunk_479_cast_fp16")]; + tensor var_2743_equation_0 = const()[name = tensor("op_2743_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2743_cast_fp16 = einsum(equation = var_2743_equation_0, values = (var_2513_cast_fp16, var_2469_cast_fp16))[name = tensor("op_2743_cast_fp16")]; + tensor var_2744_to_fp16 = const()[name = tensor("op_2744_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_481_cast_fp16 = mul(x = var_2743_cast_fp16, y = var_2744_to_fp16)[name = tensor("aw_chunk_481_cast_fp16")]; + tensor var_2747_equation_0 = const()[name = tensor("op_2747_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2747_cast_fp16 = einsum(equation = var_2747_equation_0, values = (var_2513_cast_fp16, var_2470_cast_fp16))[name = tensor("op_2747_cast_fp16")]; + tensor var_2748_to_fp16 = const()[name = tensor("op_2748_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_483_cast_fp16 = mul(x = var_2747_cast_fp16, y = var_2748_to_fp16)[name = tensor("aw_chunk_483_cast_fp16")]; + tensor var_2751_equation_0 = const()[name = tensor("op_2751_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2751_cast_fp16 = einsum(equation = var_2751_equation_0, values = (var_2513_cast_fp16, var_2471_cast_fp16))[name = tensor("op_2751_cast_fp16")]; + tensor var_2752_to_fp16 = const()[name = tensor("op_2752_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_485_cast_fp16 = mul(x = var_2751_cast_fp16, y = var_2752_to_fp16)[name = tensor("aw_chunk_485_cast_fp16")]; + tensor var_2755_equation_0 = const()[name = tensor("op_2755_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2755_cast_fp16 = einsum(equation = var_2755_equation_0, values = (var_2513_cast_fp16, var_2472_cast_fp16))[name = tensor("op_2755_cast_fp16")]; + tensor var_2756_to_fp16 = const()[name = tensor("op_2756_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_487_cast_fp16 = mul(x = var_2755_cast_fp16, y = var_2756_to_fp16)[name = tensor("aw_chunk_487_cast_fp16")]; + tensor var_2759_equation_0 = const()[name = tensor("op_2759_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2759_cast_fp16 = einsum(equation = var_2759_equation_0, values = (var_2513_cast_fp16, var_2473_cast_fp16))[name = tensor("op_2759_cast_fp16")]; + tensor var_2760_to_fp16 = const()[name = tensor("op_2760_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_489_cast_fp16 = mul(x = var_2759_cast_fp16, y = var_2760_to_fp16)[name = tensor("aw_chunk_489_cast_fp16")]; + tensor var_2763_equation_0 = const()[name = tensor("op_2763_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2763_cast_fp16 = einsum(equation = var_2763_equation_0, values = (var_2513_cast_fp16, var_2474_cast_fp16))[name = tensor("op_2763_cast_fp16")]; + tensor var_2764_to_fp16 = const()[name = tensor("op_2764_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_491_cast_fp16 = mul(x = var_2763_cast_fp16, y = var_2764_to_fp16)[name = tensor("aw_chunk_491_cast_fp16")]; + tensor var_2767_equation_0 = const()[name = tensor("op_2767_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2767_cast_fp16 = einsum(equation = var_2767_equation_0, values = (var_2513_cast_fp16, var_2475_cast_fp16))[name = tensor("op_2767_cast_fp16")]; + tensor var_2768_to_fp16 = const()[name = tensor("op_2768_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_493_cast_fp16 = mul(x = var_2767_cast_fp16, y = var_2768_to_fp16)[name = tensor("aw_chunk_493_cast_fp16")]; + tensor var_2771_equation_0 = const()[name = tensor("op_2771_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2771_cast_fp16 = einsum(equation = var_2771_equation_0, values = (var_2513_cast_fp16, var_2476_cast_fp16))[name = tensor("op_2771_cast_fp16")]; + tensor var_2772_to_fp16 = const()[name = tensor("op_2772_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_495_cast_fp16 = mul(x = var_2771_cast_fp16, y = var_2772_to_fp16)[name = tensor("aw_chunk_495_cast_fp16")]; + tensor var_2775_equation_0 = const()[name = tensor("op_2775_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2775_cast_fp16 = einsum(equation = var_2775_equation_0, values = (var_2517_cast_fp16, var_2477_cast_fp16))[name = tensor("op_2775_cast_fp16")]; + tensor var_2776_to_fp16 = const()[name = tensor("op_2776_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_497_cast_fp16 = mul(x = var_2775_cast_fp16, y = var_2776_to_fp16)[name = tensor("aw_chunk_497_cast_fp16")]; + tensor var_2779_equation_0 = const()[name = tensor("op_2779_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2779_cast_fp16 = einsum(equation = var_2779_equation_0, values = (var_2517_cast_fp16, var_2478_cast_fp16))[name = tensor("op_2779_cast_fp16")]; + tensor var_2780_to_fp16 = const()[name = tensor("op_2780_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_499_cast_fp16 = mul(x = var_2779_cast_fp16, y = var_2780_to_fp16)[name = tensor("aw_chunk_499_cast_fp16")]; + tensor var_2783_equation_0 = const()[name = tensor("op_2783_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2783_cast_fp16 = einsum(equation = var_2783_equation_0, values = (var_2517_cast_fp16, var_2479_cast_fp16))[name = tensor("op_2783_cast_fp16")]; + tensor var_2784_to_fp16 = const()[name = tensor("op_2784_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_501_cast_fp16 = mul(x = var_2783_cast_fp16, y = var_2784_to_fp16)[name = tensor("aw_chunk_501_cast_fp16")]; + tensor var_2787_equation_0 = const()[name = tensor("op_2787_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2787_cast_fp16 = einsum(equation = var_2787_equation_0, values = (var_2517_cast_fp16, var_2480_cast_fp16))[name = tensor("op_2787_cast_fp16")]; + tensor var_2788_to_fp16 = const()[name = tensor("op_2788_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_503_cast_fp16 = mul(x = var_2787_cast_fp16, y = var_2788_to_fp16)[name = tensor("aw_chunk_503_cast_fp16")]; + tensor var_2791_equation_0 = const()[name = tensor("op_2791_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2791_cast_fp16 = einsum(equation = var_2791_equation_0, values = (var_2517_cast_fp16, var_2481_cast_fp16))[name = tensor("op_2791_cast_fp16")]; + tensor var_2792_to_fp16 = const()[name = tensor("op_2792_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_505_cast_fp16 = mul(x = var_2791_cast_fp16, y = var_2792_to_fp16)[name = tensor("aw_chunk_505_cast_fp16")]; + tensor var_2795_equation_0 = const()[name = tensor("op_2795_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2795_cast_fp16 = einsum(equation = var_2795_equation_0, values = (var_2517_cast_fp16, var_2482_cast_fp16))[name = tensor("op_2795_cast_fp16")]; + tensor var_2796_to_fp16 = const()[name = tensor("op_2796_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_507_cast_fp16 = mul(x = var_2795_cast_fp16, y = var_2796_to_fp16)[name = tensor("aw_chunk_507_cast_fp16")]; + tensor var_2799_equation_0 = const()[name = tensor("op_2799_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2799_cast_fp16 = einsum(equation = var_2799_equation_0, values = (var_2517_cast_fp16, var_2483_cast_fp16))[name = tensor("op_2799_cast_fp16")]; + tensor var_2800_to_fp16 = const()[name = tensor("op_2800_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_509_cast_fp16 = mul(x = var_2799_cast_fp16, y = var_2800_to_fp16)[name = tensor("aw_chunk_509_cast_fp16")]; + tensor var_2803_equation_0 = const()[name = tensor("op_2803_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2803_cast_fp16 = einsum(equation = var_2803_equation_0, values = (var_2517_cast_fp16, var_2484_cast_fp16))[name = tensor("op_2803_cast_fp16")]; + tensor var_2804_to_fp16 = const()[name = tensor("op_2804_to_fp16"), val = tensor(0x1.43cp-3)]; + tensor aw_chunk_511_cast_fp16 = mul(x = var_2803_cast_fp16, y = var_2804_to_fp16)[name = tensor("aw_chunk_511_cast_fp16")]; + tensor var_2806_cast_fp16 = softmax(axis = var_123, x = aw_chunk_385_cast_fp16)[name = tensor("op_2806_cast_fp16")]; + tensor var_2807_cast_fp16 = softmax(axis = var_123, x = aw_chunk_387_cast_fp16)[name = tensor("op_2807_cast_fp16")]; + tensor var_2808_cast_fp16 = softmax(axis = var_123, x = aw_chunk_389_cast_fp16)[name = tensor("op_2808_cast_fp16")]; + tensor var_2809_cast_fp16 = softmax(axis = var_123, x = aw_chunk_391_cast_fp16)[name = tensor("op_2809_cast_fp16")]; + tensor var_2810_cast_fp16 = softmax(axis = var_123, x = aw_chunk_393_cast_fp16)[name = tensor("op_2810_cast_fp16")]; + tensor var_2811_cast_fp16 = softmax(axis = var_123, x = aw_chunk_395_cast_fp16)[name = tensor("op_2811_cast_fp16")]; + tensor var_2812_cast_fp16 = softmax(axis = var_123, x = aw_chunk_397_cast_fp16)[name = tensor("op_2812_cast_fp16")]; + tensor var_2813_cast_fp16 = softmax(axis = var_123, x = aw_chunk_399_cast_fp16)[name = tensor("op_2813_cast_fp16")]; + tensor var_2814_cast_fp16 = softmax(axis = var_123, x = aw_chunk_401_cast_fp16)[name = tensor("op_2814_cast_fp16")]; + tensor var_2815_cast_fp16 = softmax(axis = var_123, x = aw_chunk_403_cast_fp16)[name = tensor("op_2815_cast_fp16")]; + tensor var_2816_cast_fp16 = softmax(axis = var_123, x = aw_chunk_405_cast_fp16)[name = tensor("op_2816_cast_fp16")]; + tensor var_2817_cast_fp16 = softmax(axis = var_123, x = aw_chunk_407_cast_fp16)[name = tensor("op_2817_cast_fp16")]; + tensor var_2818_cast_fp16 = softmax(axis = var_123, x = aw_chunk_409_cast_fp16)[name = tensor("op_2818_cast_fp16")]; + tensor var_2819_cast_fp16 = softmax(axis = var_123, x = aw_chunk_411_cast_fp16)[name = tensor("op_2819_cast_fp16")]; + tensor var_2820_cast_fp16 = softmax(axis = var_123, x = aw_chunk_413_cast_fp16)[name = tensor("op_2820_cast_fp16")]; + tensor var_2821_cast_fp16 = softmax(axis = var_123, x = aw_chunk_415_cast_fp16)[name = tensor("op_2821_cast_fp16")]; + tensor var_2822_cast_fp16 = softmax(axis = var_123, x = aw_chunk_417_cast_fp16)[name = tensor("op_2822_cast_fp16")]; + tensor var_2823_cast_fp16 = softmax(axis = var_123, x = aw_chunk_419_cast_fp16)[name = tensor("op_2823_cast_fp16")]; + tensor var_2824_cast_fp16 = softmax(axis = var_123, x = aw_chunk_421_cast_fp16)[name = tensor("op_2824_cast_fp16")]; + tensor var_2825_cast_fp16 = softmax(axis = var_123, x = aw_chunk_423_cast_fp16)[name = tensor("op_2825_cast_fp16")]; + tensor var_2826_cast_fp16 = softmax(axis = var_123, x = aw_chunk_425_cast_fp16)[name = tensor("op_2826_cast_fp16")]; + tensor var_2827_cast_fp16 = softmax(axis = var_123, x = aw_chunk_427_cast_fp16)[name = tensor("op_2827_cast_fp16")]; + tensor var_2828_cast_fp16 = softmax(axis = var_123, x = aw_chunk_429_cast_fp16)[name = tensor("op_2828_cast_fp16")]; + tensor var_2829_cast_fp16 = softmax(axis = var_123, x = aw_chunk_431_cast_fp16)[name = tensor("op_2829_cast_fp16")]; + tensor var_2830_cast_fp16 = softmax(axis = var_123, x = aw_chunk_433_cast_fp16)[name = tensor("op_2830_cast_fp16")]; + tensor var_2831_cast_fp16 = softmax(axis = var_123, x = aw_chunk_435_cast_fp16)[name = tensor("op_2831_cast_fp16")]; + tensor var_2832_cast_fp16 = softmax(axis = var_123, x = aw_chunk_437_cast_fp16)[name = tensor("op_2832_cast_fp16")]; + tensor var_2833_cast_fp16 = softmax(axis = var_123, x = aw_chunk_439_cast_fp16)[name = tensor("op_2833_cast_fp16")]; + tensor var_2834_cast_fp16 = softmax(axis = var_123, x = aw_chunk_441_cast_fp16)[name = tensor("op_2834_cast_fp16")]; + tensor var_2835_cast_fp16 = softmax(axis = var_123, x = aw_chunk_443_cast_fp16)[name = tensor("op_2835_cast_fp16")]; + tensor var_2836_cast_fp16 = softmax(axis = var_123, x = aw_chunk_445_cast_fp16)[name = tensor("op_2836_cast_fp16")]; + tensor var_2837_cast_fp16 = softmax(axis = var_123, x = aw_chunk_447_cast_fp16)[name = tensor("op_2837_cast_fp16")]; + tensor var_2838_cast_fp16 = softmax(axis = var_123, x = aw_chunk_449_cast_fp16)[name = tensor("op_2838_cast_fp16")]; + tensor var_2839_cast_fp16 = softmax(axis = var_123, x = aw_chunk_451_cast_fp16)[name = tensor("op_2839_cast_fp16")]; + tensor var_2840_cast_fp16 = softmax(axis = var_123, x = aw_chunk_453_cast_fp16)[name = tensor("op_2840_cast_fp16")]; + tensor var_2841_cast_fp16 = softmax(axis = var_123, x = aw_chunk_455_cast_fp16)[name = tensor("op_2841_cast_fp16")]; + tensor var_2842_cast_fp16 = softmax(axis = var_123, x = aw_chunk_457_cast_fp16)[name = tensor("op_2842_cast_fp16")]; + tensor var_2843_cast_fp16 = softmax(axis = var_123, x = aw_chunk_459_cast_fp16)[name = tensor("op_2843_cast_fp16")]; + tensor var_2844_cast_fp16 = softmax(axis = var_123, x = aw_chunk_461_cast_fp16)[name = tensor("op_2844_cast_fp16")]; + tensor var_2845_cast_fp16 = softmax(axis = var_123, x = aw_chunk_463_cast_fp16)[name = tensor("op_2845_cast_fp16")]; + tensor var_2846_cast_fp16 = softmax(axis = var_123, x = aw_chunk_465_cast_fp16)[name = tensor("op_2846_cast_fp16")]; + tensor var_2847_cast_fp16 = softmax(axis = var_123, x = aw_chunk_467_cast_fp16)[name = tensor("op_2847_cast_fp16")]; + tensor var_2848_cast_fp16 = softmax(axis = var_123, x = aw_chunk_469_cast_fp16)[name = tensor("op_2848_cast_fp16")]; + tensor var_2849_cast_fp16 = softmax(axis = var_123, x = aw_chunk_471_cast_fp16)[name = tensor("op_2849_cast_fp16")]; + tensor var_2850_cast_fp16 = softmax(axis = var_123, x = aw_chunk_473_cast_fp16)[name = tensor("op_2850_cast_fp16")]; + tensor var_2851_cast_fp16 = softmax(axis = var_123, x = aw_chunk_475_cast_fp16)[name = tensor("op_2851_cast_fp16")]; + tensor var_2852_cast_fp16 = softmax(axis = var_123, x = aw_chunk_477_cast_fp16)[name = tensor("op_2852_cast_fp16")]; + tensor var_2853_cast_fp16 = softmax(axis = var_123, x = aw_chunk_479_cast_fp16)[name = tensor("op_2853_cast_fp16")]; + tensor var_2854_cast_fp16 = softmax(axis = var_123, x = aw_chunk_481_cast_fp16)[name = tensor("op_2854_cast_fp16")]; + tensor var_2855_cast_fp16 = softmax(axis = var_123, x = aw_chunk_483_cast_fp16)[name = tensor("op_2855_cast_fp16")]; + tensor var_2856_cast_fp16 = softmax(axis = var_123, x = aw_chunk_485_cast_fp16)[name = tensor("op_2856_cast_fp16")]; + tensor var_2857_cast_fp16 = softmax(axis = var_123, x = aw_chunk_487_cast_fp16)[name = tensor("op_2857_cast_fp16")]; + tensor var_2858_cast_fp16 = softmax(axis = var_123, x = aw_chunk_489_cast_fp16)[name = tensor("op_2858_cast_fp16")]; + tensor var_2859_cast_fp16 = softmax(axis = var_123, x = aw_chunk_491_cast_fp16)[name = tensor("op_2859_cast_fp16")]; + tensor var_2860_cast_fp16 = softmax(axis = var_123, x = aw_chunk_493_cast_fp16)[name = tensor("op_2860_cast_fp16")]; + tensor var_2861_cast_fp16 = softmax(axis = var_123, x = aw_chunk_495_cast_fp16)[name = tensor("op_2861_cast_fp16")]; + tensor var_2862_cast_fp16 = softmax(axis = var_123, x = aw_chunk_497_cast_fp16)[name = tensor("op_2862_cast_fp16")]; + tensor var_2863_cast_fp16 = softmax(axis = var_123, x = aw_chunk_499_cast_fp16)[name = tensor("op_2863_cast_fp16")]; + tensor var_2864_cast_fp16 = softmax(axis = var_123, x = aw_chunk_501_cast_fp16)[name = tensor("op_2864_cast_fp16")]; + tensor var_2865_cast_fp16 = softmax(axis = var_123, x = aw_chunk_503_cast_fp16)[name = tensor("op_2865_cast_fp16")]; + tensor var_2866_cast_fp16 = softmax(axis = var_123, x = aw_chunk_505_cast_fp16)[name = tensor("op_2866_cast_fp16")]; + tensor var_2867_cast_fp16 = softmax(axis = var_123, x = aw_chunk_507_cast_fp16)[name = tensor("op_2867_cast_fp16")]; + tensor var_2868_cast_fp16 = softmax(axis = var_123, x = aw_chunk_509_cast_fp16)[name = tensor("op_2868_cast_fp16")]; + tensor var_2869_cast_fp16 = softmax(axis = var_123, x = aw_chunk_511_cast_fp16)[name = tensor("op_2869_cast_fp16")]; + tensor var_2871_equation_0 = const()[name = tensor("op_2871_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2871_cast_fp16 = einsum(equation = var_2871_equation_0, values = (var_2519_cast_fp16, var_2806_cast_fp16))[name = tensor("op_2871_cast_fp16")]; + tensor var_2873_equation_0 = const()[name = tensor("op_2873_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2873_cast_fp16 = einsum(equation = var_2873_equation_0, values = (var_2519_cast_fp16, var_2807_cast_fp16))[name = tensor("op_2873_cast_fp16")]; + tensor var_2875_equation_0 = const()[name = tensor("op_2875_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2875_cast_fp16 = einsum(equation = var_2875_equation_0, values = (var_2519_cast_fp16, var_2808_cast_fp16))[name = tensor("op_2875_cast_fp16")]; + tensor var_2877_equation_0 = const()[name = tensor("op_2877_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2877_cast_fp16 = einsum(equation = var_2877_equation_0, values = (var_2519_cast_fp16, var_2809_cast_fp16))[name = tensor("op_2877_cast_fp16")]; + tensor var_2879_equation_0 = const()[name = tensor("op_2879_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2879_cast_fp16 = einsum(equation = var_2879_equation_0, values = (var_2519_cast_fp16, var_2810_cast_fp16))[name = tensor("op_2879_cast_fp16")]; + tensor var_2881_equation_0 = const()[name = tensor("op_2881_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2881_cast_fp16 = einsum(equation = var_2881_equation_0, values = (var_2519_cast_fp16, var_2811_cast_fp16))[name = tensor("op_2881_cast_fp16")]; + tensor var_2883_equation_0 = const()[name = tensor("op_2883_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2883_cast_fp16 = einsum(equation = var_2883_equation_0, values = (var_2519_cast_fp16, var_2812_cast_fp16))[name = tensor("op_2883_cast_fp16")]; + tensor var_2885_equation_0 = const()[name = tensor("op_2885_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2885_cast_fp16 = einsum(equation = var_2885_equation_0, values = (var_2519_cast_fp16, var_2813_cast_fp16))[name = tensor("op_2885_cast_fp16")]; + tensor var_2887_equation_0 = const()[name = tensor("op_2887_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2887_cast_fp16 = einsum(equation = var_2887_equation_0, values = (var_2523_cast_fp16, var_2814_cast_fp16))[name = tensor("op_2887_cast_fp16")]; + tensor var_2889_equation_0 = const()[name = tensor("op_2889_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2889_cast_fp16 = einsum(equation = var_2889_equation_0, values = (var_2523_cast_fp16, var_2815_cast_fp16))[name = tensor("op_2889_cast_fp16")]; + tensor var_2891_equation_0 = const()[name = tensor("op_2891_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2891_cast_fp16 = einsum(equation = var_2891_equation_0, values = (var_2523_cast_fp16, var_2816_cast_fp16))[name = tensor("op_2891_cast_fp16")]; + tensor var_2893_equation_0 = const()[name = tensor("op_2893_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2893_cast_fp16 = einsum(equation = var_2893_equation_0, values = (var_2523_cast_fp16, var_2817_cast_fp16))[name = tensor("op_2893_cast_fp16")]; + tensor var_2895_equation_0 = const()[name = tensor("op_2895_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2895_cast_fp16 = einsum(equation = var_2895_equation_0, values = (var_2523_cast_fp16, var_2818_cast_fp16))[name = tensor("op_2895_cast_fp16")]; + tensor var_2897_equation_0 = const()[name = tensor("op_2897_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2897_cast_fp16 = einsum(equation = var_2897_equation_0, values = (var_2523_cast_fp16, var_2819_cast_fp16))[name = tensor("op_2897_cast_fp16")]; + tensor var_2899_equation_0 = const()[name = tensor("op_2899_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2899_cast_fp16 = einsum(equation = var_2899_equation_0, values = (var_2523_cast_fp16, var_2820_cast_fp16))[name = tensor("op_2899_cast_fp16")]; + tensor var_2901_equation_0 = const()[name = tensor("op_2901_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2901_cast_fp16 = einsum(equation = var_2901_equation_0, values = (var_2523_cast_fp16, var_2821_cast_fp16))[name = tensor("op_2901_cast_fp16")]; + tensor var_2903_equation_0 = const()[name = tensor("op_2903_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2903_cast_fp16 = einsum(equation = var_2903_equation_0, values = (var_2527_cast_fp16, var_2822_cast_fp16))[name = tensor("op_2903_cast_fp16")]; + tensor var_2905_equation_0 = const()[name = tensor("op_2905_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2905_cast_fp16 = einsum(equation = var_2905_equation_0, values = (var_2527_cast_fp16, var_2823_cast_fp16))[name = tensor("op_2905_cast_fp16")]; + tensor var_2907_equation_0 = const()[name = tensor("op_2907_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2907_cast_fp16 = einsum(equation = var_2907_equation_0, values = (var_2527_cast_fp16, var_2824_cast_fp16))[name = tensor("op_2907_cast_fp16")]; + tensor var_2909_equation_0 = const()[name = tensor("op_2909_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2909_cast_fp16 = einsum(equation = var_2909_equation_0, values = (var_2527_cast_fp16, var_2825_cast_fp16))[name = tensor("op_2909_cast_fp16")]; + tensor var_2911_equation_0 = const()[name = tensor("op_2911_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2911_cast_fp16 = einsum(equation = var_2911_equation_0, values = (var_2527_cast_fp16, var_2826_cast_fp16))[name = tensor("op_2911_cast_fp16")]; + tensor var_2913_equation_0 = const()[name = tensor("op_2913_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2913_cast_fp16 = einsum(equation = var_2913_equation_0, values = (var_2527_cast_fp16, var_2827_cast_fp16))[name = tensor("op_2913_cast_fp16")]; + tensor var_2915_equation_0 = const()[name = tensor("op_2915_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2915_cast_fp16 = einsum(equation = var_2915_equation_0, values = (var_2527_cast_fp16, var_2828_cast_fp16))[name = tensor("op_2915_cast_fp16")]; + tensor var_2917_equation_0 = const()[name = tensor("op_2917_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2917_cast_fp16 = einsum(equation = var_2917_equation_0, values = (var_2527_cast_fp16, var_2829_cast_fp16))[name = tensor("op_2917_cast_fp16")]; + tensor var_2919_equation_0 = const()[name = tensor("op_2919_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2919_cast_fp16 = einsum(equation = var_2919_equation_0, values = (var_2531_cast_fp16, var_2830_cast_fp16))[name = tensor("op_2919_cast_fp16")]; + tensor var_2921_equation_0 = const()[name = tensor("op_2921_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2921_cast_fp16 = einsum(equation = var_2921_equation_0, values = (var_2531_cast_fp16, var_2831_cast_fp16))[name = tensor("op_2921_cast_fp16")]; + tensor var_2923_equation_0 = const()[name = tensor("op_2923_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2923_cast_fp16 = einsum(equation = var_2923_equation_0, values = (var_2531_cast_fp16, var_2832_cast_fp16))[name = tensor("op_2923_cast_fp16")]; + tensor var_2925_equation_0 = const()[name = tensor("op_2925_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2925_cast_fp16 = einsum(equation = var_2925_equation_0, values = (var_2531_cast_fp16, var_2833_cast_fp16))[name = tensor("op_2925_cast_fp16")]; + tensor var_2927_equation_0 = const()[name = tensor("op_2927_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2927_cast_fp16 = einsum(equation = var_2927_equation_0, values = (var_2531_cast_fp16, var_2834_cast_fp16))[name = tensor("op_2927_cast_fp16")]; + tensor var_2929_equation_0 = const()[name = tensor("op_2929_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2929_cast_fp16 = einsum(equation = var_2929_equation_0, values = (var_2531_cast_fp16, var_2835_cast_fp16))[name = tensor("op_2929_cast_fp16")]; + tensor var_2931_equation_0 = const()[name = tensor("op_2931_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2931_cast_fp16 = einsum(equation = var_2931_equation_0, values = (var_2531_cast_fp16, var_2836_cast_fp16))[name = tensor("op_2931_cast_fp16")]; + tensor var_2933_equation_0 = const()[name = tensor("op_2933_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2933_cast_fp16 = einsum(equation = var_2933_equation_0, values = (var_2531_cast_fp16, var_2837_cast_fp16))[name = tensor("op_2933_cast_fp16")]; + tensor var_2935_equation_0 = const()[name = tensor("op_2935_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2935_cast_fp16 = einsum(equation = var_2935_equation_0, values = (var_2535_cast_fp16, var_2838_cast_fp16))[name = tensor("op_2935_cast_fp16")]; + tensor var_2937_equation_0 = const()[name = tensor("op_2937_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2937_cast_fp16 = einsum(equation = var_2937_equation_0, values = (var_2535_cast_fp16, var_2839_cast_fp16))[name = tensor("op_2937_cast_fp16")]; + tensor var_2939_equation_0 = const()[name = tensor("op_2939_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2939_cast_fp16 = einsum(equation = var_2939_equation_0, values = (var_2535_cast_fp16, var_2840_cast_fp16))[name = tensor("op_2939_cast_fp16")]; + tensor var_2941_equation_0 = const()[name = tensor("op_2941_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2941_cast_fp16 = einsum(equation = var_2941_equation_0, values = (var_2535_cast_fp16, var_2841_cast_fp16))[name = tensor("op_2941_cast_fp16")]; + tensor var_2943_equation_0 = const()[name = tensor("op_2943_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2943_cast_fp16 = einsum(equation = var_2943_equation_0, values = (var_2535_cast_fp16, var_2842_cast_fp16))[name = tensor("op_2943_cast_fp16")]; + tensor var_2945_equation_0 = const()[name = tensor("op_2945_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2945_cast_fp16 = einsum(equation = var_2945_equation_0, values = (var_2535_cast_fp16, var_2843_cast_fp16))[name = tensor("op_2945_cast_fp16")]; + tensor var_2947_equation_0 = const()[name = tensor("op_2947_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2947_cast_fp16 = einsum(equation = var_2947_equation_0, values = (var_2535_cast_fp16, var_2844_cast_fp16))[name = tensor("op_2947_cast_fp16")]; + tensor var_2949_equation_0 = const()[name = tensor("op_2949_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2949_cast_fp16 = einsum(equation = var_2949_equation_0, values = (var_2535_cast_fp16, var_2845_cast_fp16))[name = tensor("op_2949_cast_fp16")]; + tensor var_2951_equation_0 = const()[name = tensor("op_2951_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2951_cast_fp16 = einsum(equation = var_2951_equation_0, values = (var_2539_cast_fp16, var_2846_cast_fp16))[name = tensor("op_2951_cast_fp16")]; + tensor var_2953_equation_0 = const()[name = tensor("op_2953_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2953_cast_fp16 = einsum(equation = var_2953_equation_0, values = (var_2539_cast_fp16, var_2847_cast_fp16))[name = tensor("op_2953_cast_fp16")]; + tensor var_2955_equation_0 = const()[name = tensor("op_2955_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2955_cast_fp16 = einsum(equation = var_2955_equation_0, values = (var_2539_cast_fp16, var_2848_cast_fp16))[name = tensor("op_2955_cast_fp16")]; + tensor var_2957_equation_0 = const()[name = tensor("op_2957_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2957_cast_fp16 = einsum(equation = var_2957_equation_0, values = (var_2539_cast_fp16, var_2849_cast_fp16))[name = tensor("op_2957_cast_fp16")]; + tensor var_2959_equation_0 = const()[name = tensor("op_2959_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2959_cast_fp16 = einsum(equation = var_2959_equation_0, values = (var_2539_cast_fp16, var_2850_cast_fp16))[name = tensor("op_2959_cast_fp16")]; + tensor var_2961_equation_0 = const()[name = tensor("op_2961_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2961_cast_fp16 = einsum(equation = var_2961_equation_0, values = (var_2539_cast_fp16, var_2851_cast_fp16))[name = tensor("op_2961_cast_fp16")]; + tensor var_2963_equation_0 = const()[name = tensor("op_2963_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2963_cast_fp16 = einsum(equation = var_2963_equation_0, values = (var_2539_cast_fp16, var_2852_cast_fp16))[name = tensor("op_2963_cast_fp16")]; + tensor var_2965_equation_0 = const()[name = tensor("op_2965_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2965_cast_fp16 = einsum(equation = var_2965_equation_0, values = (var_2539_cast_fp16, var_2853_cast_fp16))[name = tensor("op_2965_cast_fp16")]; + tensor var_2967_equation_0 = const()[name = tensor("op_2967_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2967_cast_fp16 = einsum(equation = var_2967_equation_0, values = (var_2543_cast_fp16, var_2854_cast_fp16))[name = tensor("op_2967_cast_fp16")]; + tensor var_2969_equation_0 = const()[name = tensor("op_2969_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2969_cast_fp16 = einsum(equation = var_2969_equation_0, values = (var_2543_cast_fp16, var_2855_cast_fp16))[name = tensor("op_2969_cast_fp16")]; + tensor var_2971_equation_0 = const()[name = tensor("op_2971_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2971_cast_fp16 = einsum(equation = var_2971_equation_0, values = (var_2543_cast_fp16, var_2856_cast_fp16))[name = tensor("op_2971_cast_fp16")]; + tensor var_2973_equation_0 = const()[name = tensor("op_2973_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2973_cast_fp16 = einsum(equation = var_2973_equation_0, values = (var_2543_cast_fp16, var_2857_cast_fp16))[name = tensor("op_2973_cast_fp16")]; + tensor var_2975_equation_0 = const()[name = tensor("op_2975_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2975_cast_fp16 = einsum(equation = var_2975_equation_0, values = (var_2543_cast_fp16, var_2858_cast_fp16))[name = tensor("op_2975_cast_fp16")]; + tensor var_2977_equation_0 = const()[name = tensor("op_2977_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2977_cast_fp16 = einsum(equation = var_2977_equation_0, values = (var_2543_cast_fp16, var_2859_cast_fp16))[name = tensor("op_2977_cast_fp16")]; + tensor var_2979_equation_0 = const()[name = tensor("op_2979_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2979_cast_fp16 = einsum(equation = var_2979_equation_0, values = (var_2543_cast_fp16, var_2860_cast_fp16))[name = tensor("op_2979_cast_fp16")]; + tensor var_2981_equation_0 = const()[name = tensor("op_2981_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2981_cast_fp16 = einsum(equation = var_2981_equation_0, values = (var_2543_cast_fp16, var_2861_cast_fp16))[name = tensor("op_2981_cast_fp16")]; + tensor var_2983_equation_0 = const()[name = tensor("op_2983_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2983_cast_fp16 = einsum(equation = var_2983_equation_0, values = (var_2547_cast_fp16, var_2862_cast_fp16))[name = tensor("op_2983_cast_fp16")]; + tensor var_2985_equation_0 = const()[name = tensor("op_2985_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2985_cast_fp16 = einsum(equation = var_2985_equation_0, values = (var_2547_cast_fp16, var_2863_cast_fp16))[name = tensor("op_2985_cast_fp16")]; + tensor var_2987_equation_0 = const()[name = tensor("op_2987_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2987_cast_fp16 = einsum(equation = var_2987_equation_0, values = (var_2547_cast_fp16, var_2864_cast_fp16))[name = tensor("op_2987_cast_fp16")]; + tensor var_2989_equation_0 = const()[name = tensor("op_2989_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2989_cast_fp16 = einsum(equation = var_2989_equation_0, values = (var_2547_cast_fp16, var_2865_cast_fp16))[name = tensor("op_2989_cast_fp16")]; + tensor var_2991_equation_0 = const()[name = tensor("op_2991_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2991_cast_fp16 = einsum(equation = var_2991_equation_0, values = (var_2547_cast_fp16, var_2866_cast_fp16))[name = tensor("op_2991_cast_fp16")]; + tensor var_2993_equation_0 = const()[name = tensor("op_2993_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2993_cast_fp16 = einsum(equation = var_2993_equation_0, values = (var_2547_cast_fp16, var_2867_cast_fp16))[name = tensor("op_2993_cast_fp16")]; + tensor var_2995_equation_0 = const()[name = tensor("op_2995_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2995_cast_fp16 = einsum(equation = var_2995_equation_0, values = (var_2547_cast_fp16, var_2868_cast_fp16))[name = tensor("op_2995_cast_fp16")]; + tensor var_2997_equation_0 = const()[name = tensor("op_2997_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2997_cast_fp16 = einsum(equation = var_2997_equation_0, values = (var_2547_cast_fp16, var_2869_cast_fp16))[name = tensor("op_2997_cast_fp16")]; + tensor var_2999_interleave_0 = const()[name = tensor("op_2999_interleave_0"), val = tensor(false)]; + tensor var_2999_cast_fp16 = concat(axis = var_95, interleave = var_2999_interleave_0, values = (var_2871_cast_fp16, var_2873_cast_fp16, var_2875_cast_fp16, var_2877_cast_fp16, var_2879_cast_fp16, var_2881_cast_fp16, var_2883_cast_fp16, var_2885_cast_fp16))[name = tensor("op_2999_cast_fp16")]; + tensor var_3001_interleave_0 = const()[name = tensor("op_3001_interleave_0"), val = tensor(false)]; + tensor var_3001_cast_fp16 = concat(axis = var_95, interleave = var_3001_interleave_0, values = (var_2887_cast_fp16, var_2889_cast_fp16, var_2891_cast_fp16, var_2893_cast_fp16, var_2895_cast_fp16, var_2897_cast_fp16, var_2899_cast_fp16, var_2901_cast_fp16))[name = tensor("op_3001_cast_fp16")]; + tensor var_3003_interleave_0 = const()[name = tensor("op_3003_interleave_0"), val = tensor(false)]; + tensor var_3003_cast_fp16 = concat(axis = var_95, interleave = var_3003_interleave_0, values = (var_2903_cast_fp16, var_2905_cast_fp16, var_2907_cast_fp16, var_2909_cast_fp16, var_2911_cast_fp16, var_2913_cast_fp16, var_2915_cast_fp16, var_2917_cast_fp16))[name = tensor("op_3003_cast_fp16")]; + tensor var_3005_interleave_0 = const()[name = tensor("op_3005_interleave_0"), val = tensor(false)]; + tensor var_3005_cast_fp16 = concat(axis = var_95, interleave = var_3005_interleave_0, values = (var_2919_cast_fp16, var_2921_cast_fp16, var_2923_cast_fp16, var_2925_cast_fp16, var_2927_cast_fp16, var_2929_cast_fp16, var_2931_cast_fp16, var_2933_cast_fp16))[name = tensor("op_3005_cast_fp16")]; + tensor var_3007_interleave_0 = const()[name = tensor("op_3007_interleave_0"), val = tensor(false)]; + tensor var_3007_cast_fp16 = concat(axis = var_95, interleave = var_3007_interleave_0, values = (var_2935_cast_fp16, var_2937_cast_fp16, var_2939_cast_fp16, var_2941_cast_fp16, var_2943_cast_fp16, var_2945_cast_fp16, var_2947_cast_fp16, var_2949_cast_fp16))[name = tensor("op_3007_cast_fp16")]; + tensor var_3009_interleave_0 = const()[name = tensor("op_3009_interleave_0"), val = tensor(false)]; + tensor var_3009_cast_fp16 = concat(axis = var_95, interleave = var_3009_interleave_0, values = (var_2951_cast_fp16, var_2953_cast_fp16, var_2955_cast_fp16, var_2957_cast_fp16, var_2959_cast_fp16, var_2961_cast_fp16, var_2963_cast_fp16, var_2965_cast_fp16))[name = tensor("op_3009_cast_fp16")]; + tensor var_3011_interleave_0 = const()[name = tensor("op_3011_interleave_0"), val = tensor(false)]; + tensor var_3011_cast_fp16 = concat(axis = var_95, interleave = var_3011_interleave_0, values = (var_2967_cast_fp16, var_2969_cast_fp16, var_2971_cast_fp16, var_2973_cast_fp16, var_2975_cast_fp16, var_2977_cast_fp16, var_2979_cast_fp16, var_2981_cast_fp16))[name = tensor("op_3011_cast_fp16")]; + tensor var_3013_interleave_0 = const()[name = tensor("op_3013_interleave_0"), val = tensor(false)]; + tensor var_3013_cast_fp16 = concat(axis = var_95, interleave = var_3013_interleave_0, values = (var_2983_cast_fp16, var_2985_cast_fp16, var_2987_cast_fp16, var_2989_cast_fp16, var_2991_cast_fp16, var_2993_cast_fp16, var_2995_cast_fp16, var_2997_cast_fp16))[name = tensor("op_3013_cast_fp16")]; + tensor input_53_interleave_0 = const()[name = tensor("input_53_interleave_0"), val = tensor(false)]; + tensor input_53_cast_fp16 = concat(axis = var_123, interleave = input_53_interleave_0, values = (var_2999_cast_fp16, var_3001_cast_fp16, var_3003_cast_fp16, var_3005_cast_fp16, var_3007_cast_fp16, var_3009_cast_fp16, var_3011_cast_fp16, var_3013_cast_fp16))[name = tensor("input_53_cast_fp16")]; + tensor var_3023_pad_type_0 = const()[name = tensor("op_3023_pad_type_0"), val = tensor("valid")]; + tensor var_3023_strides_0 = const()[name = tensor("op_3023_strides_0"), val = tensor([1, 1])]; + tensor var_3023_pad_0 = const()[name = tensor("op_3023_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3023_dilations_0 = const()[name = tensor("op_3023_dilations_0"), val = tensor([1, 1])]; + tensor var_3023_groups_0 = const()[name = tensor("op_3023_groups_0"), val = tensor(1)]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20460608)))]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20665472)))]; + tensor var_3023_cast_fp16 = conv(bias = down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_3023_dilations_0, groups = var_3023_groups_0, pad = var_3023_pad_0, pad_type = var_3023_pad_type_0, strides = var_3023_strides_0, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_53_cast_fp16)[name = tensor("op_3023_cast_fp16")]; + tensor inputs_11_cast_fp16 = add(x = var_3023_cast_fp16, y = inputs_9_cast_fp16)[name = tensor("inputs_11_cast_fp16")]; + tensor input_55_axes_0 = const()[name = tensor("input_55_axes_0"), val = tensor([1])]; + tensor input_55_gamma_0_to_fp16 = const()[name = tensor("input_55_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20666176)))]; + tensor input_55_beta_0_to_fp16 = const()[name = tensor("input_55_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20666880)))]; + tensor var_3033_to_fp16 = const()[name = tensor("op_3033_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_55_cast_fp16 = layer_norm(axes = input_55_axes_0, beta = input_55_beta_0_to_fp16, epsilon = var_3033_to_fp16, gamma = input_55_gamma_0_to_fp16, x = inputs_11_cast_fp16)[name = tensor("input_55_cast_fp16")]; + tensor var_3053_pad_type_0 = const()[name = tensor("op_3053_pad_type_0"), val = tensor("valid")]; + tensor var_3053_strides_0 = const()[name = tensor("op_3053_strides_0"), val = tensor([1, 1])]; + tensor var_3053_pad_0 = const()[name = tensor("op_3053_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3053_dilations_0 = const()[name = tensor("op_3053_dilations_0"), val = tensor([1, 1])]; + tensor var_3053_groups_0 = const()[name = tensor("op_3053_groups_0"), val = tensor(1)]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20667584)))]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22306048)))]; + tensor var_3053_cast_fp16 = conv(bias = down_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_3053_dilations_0, groups = var_3053_groups_0, pad = var_3053_pad_0, pad_type = var_3053_pad_type_0, strides = var_3053_strides_0, weight = down_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_55_cast_fp16)[name = tensor("op_3053_cast_fp16")]; + tensor var_3054_split_sizes_0 = const()[name = tensor("op_3054_split_sizes_0"), val = tensor([1280, 1280])]; + tensor var_3054_axis_0 = const()[name = tensor("op_3054_axis_0"), val = tensor(1)]; + tensor var_3054_cast_fp16_0, tensor var_3054_cast_fp16_1 = split(axis = var_3054_axis_0, split_sizes = var_3054_split_sizes_0, x = var_3053_cast_fp16)[name = tensor("op_3054_cast_fp16")]; + tensor var_3056_mode_0 = const()[name = tensor("op_3056_mode_0"), val = tensor("EXACT")]; + tensor var_3056_cast_fp16 = gelu(mode = var_3056_mode_0, x = var_3054_cast_fp16_1)[name = tensor("op_3056_cast_fp16")]; + tensor input_57_cast_fp16 = mul(x = var_3054_cast_fp16_0, y = var_3056_cast_fp16)[name = tensor("input_57_cast_fp16")]; + tensor var_3064_pad_type_0 = const()[name = tensor("op_3064_pad_type_0"), val = tensor("valid")]; + tensor var_3064_strides_0 = const()[name = tensor("op_3064_strides_0"), val = tensor([1, 1])]; + tensor var_3064_pad_0 = const()[name = tensor("op_3064_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3064_dilations_0 = const()[name = tensor("op_3064_dilations_0"), val = tensor([1, 1])]; + tensor var_3064_groups_0 = const()[name = tensor("op_3064_groups_0"), val = tensor(1)]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22311232)))]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23130496)))]; + tensor var_3064_cast_fp16 = conv(bias = down_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_3064_dilations_0, groups = var_3064_groups_0, pad = var_3064_pad_0, pad_type = var_3064_pad_type_0, strides = var_3064_strides_0, weight = down_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_57_cast_fp16)[name = tensor("op_3064_cast_fp16")]; + tensor hidden_states_33_cast_fp16 = add(x = var_3064_cast_fp16, y = inputs_11_cast_fp16)[name = tensor("hidden_states_33_cast_fp16")]; + tensor var_3066 = const()[name = tensor("op_3066"), val = tensor([2, 320, 64, 64])]; + tensor input_59_cast_fp16 = reshape(shape = var_3066, x = hidden_states_33_cast_fp16)[name = tensor("input_59_cast_fp16")]; + tensor hidden_states_35_pad_type_0 = const()[name = tensor("hidden_states_35_pad_type_0"), val = tensor("valid")]; + tensor hidden_states_35_strides_0 = const()[name = tensor("hidden_states_35_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_35_pad_0 = const()[name = tensor("hidden_states_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_35_dilations_0 = const()[name = tensor("hidden_states_35_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_35_groups_0 = const()[name = tensor("hidden_states_35_groups_0"), val = tensor(1)]; + tensor down_blocks_0_attentions_1_proj_out_weight_to_fp16 = const()[name = tensor("down_blocks_0_attentions_1_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23131200)))]; + tensor down_blocks_0_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor("down_blocks_0_attentions_1_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23336064)))]; + tensor hidden_states_35_cast_fp16 = conv(bias = down_blocks_0_attentions_1_proj_out_bias_to_fp16, dilations = hidden_states_35_dilations_0, groups = hidden_states_35_groups_0, pad = hidden_states_35_pad_0, pad_type = hidden_states_35_pad_type_0, strides = hidden_states_35_strides_0, weight = down_blocks_0_attentions_1_proj_out_weight_to_fp16, x = input_59_cast_fp16)[name = tensor("hidden_states_35_cast_fp16")]; + tensor input_61_cast_fp16_1 = add(x = hidden_states_35_cast_fp16, y = hidden_states_23_cast_fp16)[name = tensor("input_61_cast_fp16")]; + tensor input_63_pad_type_0 = const()[name = tensor("input_63_pad_type_0"), val = tensor("custom")]; + tensor input_63_pad_0 = const()[name = tensor("input_63_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor input_63_strides_0 = const()[name = tensor("input_63_strides_0"), val = tensor([2, 2])]; + tensor input_63_dilations_0 = const()[name = tensor("input_63_dilations_0"), val = tensor([1, 1])]; + tensor input_63_groups_0 = const()[name = tensor("input_63_groups_0"), val = tensor(1)]; + tensor down_blocks_0_downsamplers_0_conv_weight_to_fp16 = const()[name = tensor("down_blocks_0_downsamplers_0_conv_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23336768)))]; + tensor down_blocks_0_downsamplers_0_conv_bias_to_fp16 = const()[name = tensor("down_blocks_0_downsamplers_0_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25180032)))]; + tensor input_63_cast_fp16_1 = conv(bias = down_blocks_0_downsamplers_0_conv_bias_to_fp16, dilations = input_63_dilations_0, groups = input_63_groups_0, pad = input_63_pad_0, pad_type = input_63_pad_type_0, strides = input_63_strides_0, weight = down_blocks_0_downsamplers_0_conv_weight_to_fp16, x = input_61_cast_fp16_1)[name = tensor("input_63_cast_fp16")]; + tensor var_3089 = const()[name = tensor("op_3089"), val = tensor(3)]; + tensor var_3111 = const()[name = tensor("op_3111"), val = tensor(1)]; + tensor reshape_24_shape_0 = const()[name = tensor("reshape_24_shape_0"), val = tensor([2, 32, 10, 32, 32])]; + tensor reshape_24_cast_fp16 = reshape(shape = reshape_24_shape_0, x = input_63_cast_fp16_1)[name = tensor("reshape_24_cast_fp16")]; + tensor reduce_mean_18_axes_0 = const()[name = tensor("reduce_mean_18_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_18_keep_dims_0 = const()[name = tensor("reduce_mean_18_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_18_cast_fp16 = reduce_mean(axes = reduce_mean_18_axes_0, keep_dims = reduce_mean_18_keep_dims_0, x = reshape_24_cast_fp16)[name = tensor("reduce_mean_18_cast_fp16")]; + tensor sub_12_cast_fp16 = sub(x = reshape_24_cast_fp16, y = reduce_mean_18_cast_fp16)[name = tensor("sub_12_cast_fp16")]; + tensor square_6_cast_fp16 = square(x = sub_12_cast_fp16)[name = tensor("square_6_cast_fp16")]; + tensor reduce_mean_20_axes_0 = const()[name = tensor("reduce_mean_20_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_20_keep_dims_0 = const()[name = tensor("reduce_mean_20_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_20_cast_fp16 = reduce_mean(axes = reduce_mean_20_axes_0, keep_dims = reduce_mean_20_keep_dims_0, x = square_6_cast_fp16)[name = tensor("reduce_mean_20_cast_fp16")]; + tensor add_12_y_0_to_fp16 = const()[name = tensor("add_12_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_12_cast_fp16 = add(x = reduce_mean_20_cast_fp16, y = add_12_y_0_to_fp16)[name = tensor("add_12_cast_fp16")]; + tensor sqrt_6_cast_fp16 = sqrt(x = add_12_cast_fp16)[name = tensor("sqrt_6_cast_fp16")]; + tensor real_div_6_cast_fp16 = real_div(x = sub_12_cast_fp16, y = sqrt_6_cast_fp16)[name = tensor("real_div_6_cast_fp16")]; + tensor reshape_25_shape_0 = const()[name = tensor("reshape_25_shape_0"), val = tensor([2, 320, 32, 32])]; + tensor reshape_25_cast_fp16 = reshape(shape = reshape_25_shape_0, x = real_div_6_cast_fp16)[name = tensor("reshape_25_cast_fp16")]; + tensor add_13_gamma_0_to_fp16 = const()[name = tensor("add_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25180736)))]; + tensor add_13_beta_0_to_fp16 = const()[name = tensor("add_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25181440)))]; + tensor add_13_epsilon_0_to_fp16 = const()[name = tensor("add_13_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_13_cast_fp16 = batch_norm(beta = add_13_beta_0_to_fp16, epsilon = add_13_epsilon_0_to_fp16, gamma = add_13_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_25_cast_fp16)[name = tensor("add_13_cast_fp16")]; + tensor input_67_cast_fp16 = silu(x = add_13_cast_fp16)[name = tensor("input_67_cast_fp16")]; + tensor hidden_states_37_pad_type_0 = const()[name = tensor("hidden_states_37_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_37_pad_0 = const()[name = tensor("hidden_states_37_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_37_strides_0 = const()[name = tensor("hidden_states_37_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_37_dilations_0 = const()[name = tensor("hidden_states_37_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_37_groups_0 = const()[name = tensor("hidden_states_37_groups_0"), val = tensor(1)]; + tensor down_blocks_1_resnets_0_conv1_weight_to_fp16 = const()[name = tensor("down_blocks_1_resnets_0_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25182144)))]; + tensor down_blocks_1_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28868608)))]; + tensor hidden_states_37_cast_fp16 = conv(bias = down_blocks_1_resnets_0_conv1_bias_to_fp16, dilations = hidden_states_37_dilations_0, groups = hidden_states_37_groups_0, pad = hidden_states_37_pad_0, pad_type = hidden_states_37_pad_type_0, strides = hidden_states_37_strides_0, weight = down_blocks_1_resnets_0_conv1_weight_to_fp16, x = input_67_cast_fp16)[name = tensor("hidden_states_37_cast_fp16")]; + tensor temb_5_pad_type_0 = const()[name = tensor("temb_5_pad_type_0"), val = tensor("valid")]; + tensor temb_5_strides_0 = const()[name = tensor("temb_5_strides_0"), val = tensor([1, 1])]; + tensor temb_5_pad_0 = const()[name = tensor("temb_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor temb_5_dilations_0 = const()[name = tensor("temb_5_dilations_0"), val = tensor([1, 1])]; + tensor temb_5_groups_0 = const()[name = tensor("temb_5_groups_0"), val = tensor(1)]; + tensor down_blocks_1_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor("down_blocks_1_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28869952)))]; + tensor down_blocks_1_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30508416)))]; + tensor temb_5_cast_fp16 = conv(bias = down_blocks_1_resnets_0_time_emb_proj_bias_to_fp16, dilations = temb_5_dilations_0, groups = temb_5_groups_0, pad = temb_5_pad_0, pad_type = temb_5_pad_type_0, strides = temb_5_strides_0, weight = down_blocks_1_resnets_0_time_emb_proj_weight_to_fp16, x = input_15_cast_fp16_1)[name = tensor("temb_5_cast_fp16")]; + tensor input_71_cast_fp16 = add(x = hidden_states_37_cast_fp16, y = temb_5_cast_fp16)[name = tensor("input_71_cast_fp16")]; + tensor reshape_28_shape_0 = const()[name = tensor("reshape_28_shape_0"), val = tensor([2, 32, 20, 32, 32])]; + tensor reshape_28_cast_fp16 = reshape(shape = reshape_28_shape_0, x = input_71_cast_fp16)[name = tensor("reshape_28_cast_fp16")]; + tensor reduce_mean_21_axes_0 = const()[name = tensor("reduce_mean_21_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_21_keep_dims_0 = const()[name = tensor("reduce_mean_21_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_21_cast_fp16 = reduce_mean(axes = reduce_mean_21_axes_0, keep_dims = reduce_mean_21_keep_dims_0, x = reshape_28_cast_fp16)[name = tensor("reduce_mean_21_cast_fp16")]; + tensor sub_14_cast_fp16 = sub(x = reshape_28_cast_fp16, y = reduce_mean_21_cast_fp16)[name = tensor("sub_14_cast_fp16")]; + tensor square_7_cast_fp16 = square(x = sub_14_cast_fp16)[name = tensor("square_7_cast_fp16")]; + tensor reduce_mean_23_axes_0 = const()[name = tensor("reduce_mean_23_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_23_keep_dims_0 = const()[name = tensor("reduce_mean_23_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_23_cast_fp16 = reduce_mean(axes = reduce_mean_23_axes_0, keep_dims = reduce_mean_23_keep_dims_0, x = square_7_cast_fp16)[name = tensor("reduce_mean_23_cast_fp16")]; + tensor add_14_y_0_to_fp16 = const()[name = tensor("add_14_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_14_cast_fp16 = add(x = reduce_mean_23_cast_fp16, y = add_14_y_0_to_fp16)[name = tensor("add_14_cast_fp16")]; + tensor sqrt_7_cast_fp16 = sqrt(x = add_14_cast_fp16)[name = tensor("sqrt_7_cast_fp16")]; + tensor real_div_7_cast_fp16 = real_div(x = sub_14_cast_fp16, y = sqrt_7_cast_fp16)[name = tensor("real_div_7_cast_fp16")]; + tensor reshape_29_shape_0 = const()[name = tensor("reshape_29_shape_0"), val = tensor([2, 640, 32, 32])]; + tensor reshape_29_cast_fp16 = reshape(shape = reshape_29_shape_0, x = real_div_7_cast_fp16)[name = tensor("reshape_29_cast_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(30509760)))]; + 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(30511104)))]; + tensor add_15_gamma_0_to_fp16 = const()[name = tensor("add_15_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30512448)))]; + tensor add_15_beta_0_to_fp16 = const()[name = tensor("add_15_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30513792)))]; + tensor add_15_epsilon_0_to_fp16 = const()[name = tensor("add_15_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_15_cast_fp16 = batch_norm(beta = add_15_beta_0_to_fp16, epsilon = add_15_epsilon_0_to_fp16, gamma = add_15_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_29_cast_fp16)[name = tensor("add_15_cast_fp16")]; + tensor input_75_cast_fp16 = silu(x = add_15_cast_fp16)[name = tensor("input_75_cast_fp16")]; + tensor hidden_states_39_pad_type_0 = const()[name = tensor("hidden_states_39_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_39_pad_0 = const()[name = tensor("hidden_states_39_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_39_strides_0 = const()[name = tensor("hidden_states_39_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_39_dilations_0 = const()[name = tensor("hidden_states_39_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_39_groups_0 = const()[name = tensor("hidden_states_39_groups_0"), val = tensor(1)]; + tensor down_blocks_1_resnets_0_conv2_weight_to_fp16 = const()[name = tensor("down_blocks_1_resnets_0_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30515136)))]; + tensor down_blocks_1_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37888000)))]; + tensor hidden_states_39_cast_fp16 = conv(bias = down_blocks_1_resnets_0_conv2_bias_to_fp16, dilations = hidden_states_39_dilations_0, groups = hidden_states_39_groups_0, pad = hidden_states_39_pad_0, pad_type = hidden_states_39_pad_type_0, strides = hidden_states_39_strides_0, weight = down_blocks_1_resnets_0_conv2_weight_to_fp16, x = input_75_cast_fp16)[name = tensor("hidden_states_39_cast_fp16")]; + tensor x_1_pad_type_0 = const()[name = tensor("x_1_pad_type_0"), val = tensor("valid")]; + tensor x_1_strides_0 = const()[name = tensor("x_1_strides_0"), val = tensor([1, 1])]; + tensor x_1_pad_0 = const()[name = tensor("x_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_1_dilations_0 = const()[name = tensor("x_1_dilations_0"), val = tensor([1, 1])]; + tensor x_1_groups_0 = const()[name = tensor("x_1_groups_0"), val = tensor(1)]; + tensor down_blocks_1_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = tensor("down_blocks_1_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37889344)))]; + tensor down_blocks_1_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38299008)))]; + tensor x_1_cast_fp16 = conv(bias = down_blocks_1_resnets_0_conv_shortcut_bias_to_fp16, dilations = x_1_dilations_0, groups = x_1_groups_0, pad = x_1_pad_0, pad_type = x_1_pad_type_0, strides = x_1_strides_0, weight = down_blocks_1_resnets_0_conv_shortcut_weight_to_fp16, x = input_63_cast_fp16_1)[name = tensor("x_1_cast_fp16")]; + tensor hidden_states_41_cast_fp16 = add(x = x_1_cast_fp16, y = hidden_states_39_cast_fp16)[name = tensor("hidden_states_41_cast_fp16")]; + tensor reshape_32_shape_0 = const()[name = tensor("reshape_32_shape_0"), val = tensor([2, 32, 20, 32, 32])]; + tensor reshape_32_cast_fp16 = reshape(shape = reshape_32_shape_0, x = hidden_states_41_cast_fp16)[name = tensor("reshape_32_cast_fp16")]; + tensor reduce_mean_24_axes_0 = const()[name = tensor("reduce_mean_24_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_24_keep_dims_0 = const()[name = tensor("reduce_mean_24_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_24_cast_fp16 = reduce_mean(axes = reduce_mean_24_axes_0, keep_dims = reduce_mean_24_keep_dims_0, x = reshape_32_cast_fp16)[name = tensor("reduce_mean_24_cast_fp16")]; + tensor sub_16_cast_fp16 = sub(x = reshape_32_cast_fp16, y = reduce_mean_24_cast_fp16)[name = tensor("sub_16_cast_fp16")]; + tensor square_8_cast_fp16 = square(x = sub_16_cast_fp16)[name = tensor("square_8_cast_fp16")]; + tensor reduce_mean_26_axes_0 = const()[name = tensor("reduce_mean_26_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_26_keep_dims_0 = const()[name = tensor("reduce_mean_26_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_26_cast_fp16 = reduce_mean(axes = reduce_mean_26_axes_0, keep_dims = reduce_mean_26_keep_dims_0, x = square_8_cast_fp16)[name = tensor("reduce_mean_26_cast_fp16")]; + tensor add_16_y_0_to_fp16 = const()[name = tensor("add_16_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_16_cast_fp16 = add(x = reduce_mean_26_cast_fp16, y = add_16_y_0_to_fp16)[name = tensor("add_16_cast_fp16")]; + tensor sqrt_8_cast_fp16 = sqrt(x = add_16_cast_fp16)[name = tensor("sqrt_8_cast_fp16")]; + tensor real_div_8_cast_fp16 = real_div(x = sub_16_cast_fp16, y = sqrt_8_cast_fp16)[name = tensor("real_div_8_cast_fp16")]; + tensor reshape_33_shape_0 = const()[name = tensor("reshape_33_shape_0"), val = tensor([2, 640, 32, 32])]; + tensor reshape_33_cast_fp16 = reshape(shape = reshape_33_shape_0, x = real_div_8_cast_fp16)[name = tensor("reshape_33_cast_fp16")]; + tensor add_17_gamma_0_to_fp16 = const()[name = tensor("add_17_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38300352)))]; + tensor add_17_beta_0_to_fp16 = const()[name = tensor("add_17_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38301696)))]; + tensor add_17_epsilon_0_to_fp16 = const()[name = tensor("add_17_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_17_cast_fp16 = batch_norm(beta = add_17_beta_0_to_fp16, epsilon = add_17_epsilon_0_to_fp16, gamma = add_17_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_33_cast_fp16)[name = tensor("add_17_cast_fp16")]; + tensor hidden_states_43_pad_type_0 = const()[name = tensor("hidden_states_43_pad_type_0"), val = tensor("valid")]; + tensor hidden_states_43_strides_0 = const()[name = tensor("hidden_states_43_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_43_pad_0 = const()[name = tensor("hidden_states_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_43_dilations_0 = const()[name = tensor("hidden_states_43_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_43_groups_0 = const()[name = tensor("hidden_states_43_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38303040)))]; + tensor down_blocks_1_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39122304)))]; + tensor hidden_states_43_cast_fp16 = conv(bias = down_blocks_1_attentions_0_proj_in_bias_to_fp16, dilations = hidden_states_43_dilations_0, groups = hidden_states_43_groups_0, pad = hidden_states_43_pad_0, pad_type = hidden_states_43_pad_type_0, strides = hidden_states_43_strides_0, weight = down_blocks_1_attentions_0_proj_in_weight_to_fp16, x = add_17_cast_fp16)[name = tensor("hidden_states_43_cast_fp16")]; + tensor var_3190 = const()[name = tensor("op_3190"), val = tensor([2, 640, 1, 1024])]; + tensor inputs_13_cast_fp16 = reshape(shape = var_3190, x = hidden_states_43_cast_fp16)[name = tensor("inputs_13_cast_fp16")]; + tensor hidden_states_45_axes_0 = const()[name = tensor("hidden_states_45_axes_0"), val = tensor([1])]; + tensor hidden_states_45_gamma_0_to_fp16 = const()[name = tensor("hidden_states_45_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39123648)))]; + tensor hidden_states_45_beta_0_to_fp16 = const()[name = tensor("hidden_states_45_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39124992)))]; + tensor var_3206_to_fp16 = const()[name = tensor("op_3206_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_45_cast_fp16 = layer_norm(axes = hidden_states_45_axes_0, beta = hidden_states_45_beta_0_to_fp16, epsilon = var_3206_to_fp16, gamma = hidden_states_45_gamma_0_to_fp16, x = inputs_13_cast_fp16)[name = tensor("hidden_states_45_cast_fp16")]; + tensor q_9_pad_type_0 = const()[name = tensor("q_9_pad_type_0"), val = tensor("valid")]; + tensor q_9_strides_0 = const()[name = tensor("q_9_strides_0"), val = tensor([1, 1])]; + tensor q_9_pad_0 = const()[name = tensor("q_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_9_dilations_0 = const()[name = tensor("q_9_dilations_0"), val = tensor([1, 1])]; + tensor q_9_groups_0 = const()[name = tensor("q_9_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_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(39126336)))]; + tensor q_9_cast_fp16 = conv(dilations = q_9_dilations_0, groups = q_9_groups_0, pad = q_9_pad_0, pad_type = q_9_pad_type_0, strides = q_9_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_45_cast_fp16)[name = tensor("q_9_cast_fp16")]; + tensor k_17_pad_type_0 = const()[name = tensor("k_17_pad_type_0"), val = tensor("valid")]; + tensor k_17_strides_0 = const()[name = tensor("k_17_strides_0"), val = tensor([1, 1])]; + tensor k_17_pad_0 = const()[name = tensor("k_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_17_dilations_0 = const()[name = tensor("k_17_dilations_0"), val = tensor([1, 1])]; + tensor k_17_groups_0 = const()[name = tensor("k_17_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_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(39945600)))]; + tensor k_17_cast_fp16 = conv(dilations = k_17_dilations_0, groups = k_17_groups_0, pad = k_17_pad_0, pad_type = k_17_pad_type_0, strides = k_17_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_45_cast_fp16)[name = tensor("k_17_cast_fp16")]; + tensor v_9_pad_type_0 = const()[name = tensor("v_9_pad_type_0"), val = tensor("valid")]; + tensor v_9_strides_0 = const()[name = tensor("v_9_strides_0"), val = tensor([1, 1])]; + tensor v_9_pad_0 = const()[name = tensor("v_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_9_dilations_0 = const()[name = tensor("v_9_dilations_0"), val = tensor([1, 1])]; + tensor v_9_groups_0 = const()[name = tensor("v_9_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_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(40764864)))]; + tensor v_9_cast_fp16 = conv(dilations = v_9_dilations_0, groups = v_9_groups_0, pad = v_9_pad_0, pad_type = v_9_pad_type_0, strides = v_9_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_45_cast_fp16)[name = tensor("v_9_cast_fp16")]; + tensor var_3239_begin_0 = const()[name = tensor("op_3239_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3239_end_0 = const()[name = tensor("op_3239_end_0"), val = tensor([2, 80, 1, 1024])]; + tensor var_3239_end_mask_0 = const()[name = tensor("op_3239_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3239_cast_fp16 = slice_by_index(begin = var_3239_begin_0, end = var_3239_end_0, end_mask = var_3239_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_3239_cast_fp16")]; + tensor var_3243_begin_0 = const()[name = tensor("op_3243_begin_0"), val = tensor([0, 80, 0, 0])]; + tensor var_3243_end_0 = const()[name = tensor("op_3243_end_0"), val = tensor([2, 160, 1, 1024])]; + tensor var_3243_end_mask_0 = const()[name = tensor("op_3243_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3243_cast_fp16 = slice_by_index(begin = var_3243_begin_0, end = var_3243_end_0, end_mask = var_3243_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_3243_cast_fp16")]; + tensor var_3247_begin_0 = const()[name = tensor("op_3247_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_3247_end_0 = const()[name = tensor("op_3247_end_0"), val = tensor([2, 240, 1, 1024])]; + tensor var_3247_end_mask_0 = const()[name = tensor("op_3247_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3247_cast_fp16 = slice_by_index(begin = var_3247_begin_0, end = var_3247_end_0, end_mask = var_3247_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_3247_cast_fp16")]; + tensor var_3251_begin_0 = const()[name = tensor("op_3251_begin_0"), val = tensor([0, 240, 0, 0])]; + tensor var_3251_end_0 = const()[name = tensor("op_3251_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_3251_end_mask_0 = const()[name = tensor("op_3251_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3251_cast_fp16 = slice_by_index(begin = var_3251_begin_0, end = var_3251_end_0, end_mask = var_3251_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_3251_cast_fp16")]; + tensor var_3255_begin_0 = const()[name = tensor("op_3255_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3255_end_0 = const()[name = tensor("op_3255_end_0"), val = tensor([2, 400, 1, 1024])]; + tensor var_3255_end_mask_0 = const()[name = tensor("op_3255_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3255_cast_fp16 = slice_by_index(begin = var_3255_begin_0, end = var_3255_end_0, end_mask = var_3255_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_3255_cast_fp16")]; + tensor var_3259_begin_0 = const()[name = tensor("op_3259_begin_0"), val = tensor([0, 400, 0, 0])]; + tensor var_3259_end_0 = const()[name = tensor("op_3259_end_0"), val = tensor([2, 480, 1, 1024])]; + tensor var_3259_end_mask_0 = const()[name = tensor("op_3259_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3259_cast_fp16 = slice_by_index(begin = var_3259_begin_0, end = var_3259_end_0, end_mask = var_3259_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_3259_cast_fp16")]; + tensor var_3263_begin_0 = const()[name = tensor("op_3263_begin_0"), val = tensor([0, 480, 0, 0])]; + tensor var_3263_end_0 = const()[name = tensor("op_3263_end_0"), val = tensor([2, 560, 1, 1024])]; + tensor var_3263_end_mask_0 = const()[name = tensor("op_3263_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3263_cast_fp16 = slice_by_index(begin = var_3263_begin_0, end = var_3263_end_0, end_mask = var_3263_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_3263_cast_fp16")]; + tensor var_3267_begin_0 = const()[name = tensor("op_3267_begin_0"), val = tensor([0, 560, 0, 0])]; + tensor var_3267_end_0 = const()[name = tensor("op_3267_end_0"), val = tensor([2, 1, 1, 1024])]; + tensor var_3267_end_mask_0 = const()[name = tensor("op_3267_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3267_cast_fp16 = slice_by_index(begin = var_3267_begin_0, end = var_3267_end_0, end_mask = var_3267_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_3267_cast_fp16")]; + tensor var_3270_begin_0 = const()[name = tensor("op_3270_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3270_end_0 = const()[name = tensor("op_3270_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_3270_end_mask_0 = const()[name = tensor("op_3270_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3270_cast_fp16 = slice_by_index(begin = var_3270_begin_0, end = var_3270_end_0, end_mask = var_3270_end_mask_0, x = var_3239_cast_fp16)[name = tensor("op_3270_cast_fp16")]; + tensor var_3271_begin_0 = const()[name = tensor("op_3271_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3271_end_0 = const()[name = tensor("op_3271_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_3271_end_mask_0 = const()[name = tensor("op_3271_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3271_cast_fp16 = slice_by_index(begin = var_3271_begin_0, end = var_3271_end_0, end_mask = var_3271_end_mask_0, x = var_3239_cast_fp16)[name = tensor("op_3271_cast_fp16")]; + tensor var_3272_begin_0 = const()[name = tensor("op_3272_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3272_end_0 = const()[name = tensor("op_3272_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_3272_end_mask_0 = const()[name = tensor("op_3272_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3272_cast_fp16 = slice_by_index(begin = var_3272_begin_0, end = var_3272_end_0, end_mask = var_3272_end_mask_0, x = var_3243_cast_fp16)[name = tensor("op_3272_cast_fp16")]; + tensor var_3273_begin_0 = const()[name = tensor("op_3273_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3273_end_0 = const()[name = tensor("op_3273_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_3273_end_mask_0 = const()[name = tensor("op_3273_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3273_cast_fp16 = slice_by_index(begin = var_3273_begin_0, end = var_3273_end_0, end_mask = var_3273_end_mask_0, x = var_3243_cast_fp16)[name = tensor("op_3273_cast_fp16")]; + tensor var_3274_begin_0 = const()[name = tensor("op_3274_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3274_end_0 = const()[name = tensor("op_3274_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_3274_end_mask_0 = const()[name = tensor("op_3274_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3274_cast_fp16 = slice_by_index(begin = var_3274_begin_0, end = var_3274_end_0, end_mask = var_3274_end_mask_0, x = var_3247_cast_fp16)[name = tensor("op_3274_cast_fp16")]; + tensor var_3275_begin_0 = const()[name = tensor("op_3275_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3275_end_0 = const()[name = tensor("op_3275_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_3275_end_mask_0 = const()[name = tensor("op_3275_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3275_cast_fp16 = slice_by_index(begin = var_3275_begin_0, end = var_3275_end_0, end_mask = var_3275_end_mask_0, x = var_3247_cast_fp16)[name = tensor("op_3275_cast_fp16")]; + tensor var_3276_begin_0 = const()[name = tensor("op_3276_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3276_end_0 = const()[name = tensor("op_3276_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_3276_end_mask_0 = const()[name = tensor("op_3276_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3276_cast_fp16 = slice_by_index(begin = var_3276_begin_0, end = var_3276_end_0, end_mask = var_3276_end_mask_0, x = var_3251_cast_fp16)[name = tensor("op_3276_cast_fp16")]; + tensor var_3277_begin_0 = const()[name = tensor("op_3277_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3277_end_0 = const()[name = tensor("op_3277_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_3277_end_mask_0 = const()[name = tensor("op_3277_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3277_cast_fp16 = slice_by_index(begin = var_3277_begin_0, end = var_3277_end_0, end_mask = var_3277_end_mask_0, x = var_3251_cast_fp16)[name = tensor("op_3277_cast_fp16")]; + tensor var_3278_begin_0 = const()[name = tensor("op_3278_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3278_end_0 = const()[name = tensor("op_3278_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_3278_end_mask_0 = const()[name = tensor("op_3278_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3278_cast_fp16 = slice_by_index(begin = var_3278_begin_0, end = var_3278_end_0, end_mask = var_3278_end_mask_0, x = var_3255_cast_fp16)[name = tensor("op_3278_cast_fp16")]; + tensor var_3279_begin_0 = const()[name = tensor("op_3279_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3279_end_0 = const()[name = tensor("op_3279_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_3279_end_mask_0 = const()[name = tensor("op_3279_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3279_cast_fp16 = slice_by_index(begin = var_3279_begin_0, end = var_3279_end_0, end_mask = var_3279_end_mask_0, x = var_3255_cast_fp16)[name = tensor("op_3279_cast_fp16")]; + tensor var_3280_begin_0 = const()[name = tensor("op_3280_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3280_end_0 = const()[name = tensor("op_3280_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_3280_end_mask_0 = const()[name = tensor("op_3280_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3280_cast_fp16 = slice_by_index(begin = var_3280_begin_0, end = var_3280_end_0, end_mask = var_3280_end_mask_0, x = var_3259_cast_fp16)[name = tensor("op_3280_cast_fp16")]; + tensor var_3281_begin_0 = const()[name = tensor("op_3281_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3281_end_0 = const()[name = tensor("op_3281_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_3281_end_mask_0 = const()[name = tensor("op_3281_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3281_cast_fp16 = slice_by_index(begin = var_3281_begin_0, end = var_3281_end_0, end_mask = var_3281_end_mask_0, x = var_3259_cast_fp16)[name = tensor("op_3281_cast_fp16")]; + tensor var_3282_begin_0 = const()[name = tensor("op_3282_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3282_end_0 = const()[name = tensor("op_3282_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_3282_end_mask_0 = const()[name = tensor("op_3282_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3282_cast_fp16 = slice_by_index(begin = var_3282_begin_0, end = var_3282_end_0, end_mask = var_3282_end_mask_0, x = var_3263_cast_fp16)[name = tensor("op_3282_cast_fp16")]; + tensor var_3283_begin_0 = const()[name = tensor("op_3283_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3283_end_0 = const()[name = tensor("op_3283_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_3283_end_mask_0 = const()[name = tensor("op_3283_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3283_cast_fp16 = slice_by_index(begin = var_3283_begin_0, end = var_3283_end_0, end_mask = var_3283_end_mask_0, x = var_3263_cast_fp16)[name = tensor("op_3283_cast_fp16")]; + tensor var_3284_begin_0 = const()[name = tensor("op_3284_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3284_end_0 = const()[name = tensor("op_3284_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_3284_end_mask_0 = const()[name = tensor("op_3284_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3284_cast_fp16 = slice_by_index(begin = var_3284_begin_0, end = var_3284_end_0, end_mask = var_3284_end_mask_0, x = var_3267_cast_fp16)[name = tensor("op_3284_cast_fp16")]; + tensor var_3285_begin_0 = const()[name = tensor("op_3285_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3285_end_0 = const()[name = tensor("op_3285_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_3285_end_mask_0 = const()[name = tensor("op_3285_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3285_cast_fp16 = slice_by_index(begin = var_3285_begin_0, end = var_3285_end_0, end_mask = var_3285_end_mask_0, x = var_3267_cast_fp16)[name = tensor("op_3285_cast_fp16")]; + tensor k_19_perm_0 = const()[name = tensor("k_19_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_3290_begin_0 = const()[name = tensor("op_3290_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3290_end_0 = const()[name = tensor("op_3290_end_0"), val = tensor([2, 1024, 1, 80])]; + tensor var_3290_end_mask_0 = const()[name = tensor("op_3290_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_19_cast_fp16 = transpose(perm = k_19_perm_0, x = k_17_cast_fp16)[name = tensor("transpose_9")]; + tensor var_3290_cast_fp16 = slice_by_index(begin = var_3290_begin_0, end = var_3290_end_0, end_mask = var_3290_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_3290_cast_fp16")]; + tensor var_3294_begin_0 = const()[name = tensor("op_3294_begin_0"), val = tensor([0, 0, 0, 80])]; + tensor var_3294_end_0 = const()[name = tensor("op_3294_end_0"), val = tensor([2, 1024, 1, 160])]; + tensor var_3294_end_mask_0 = const()[name = tensor("op_3294_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3294_cast_fp16 = slice_by_index(begin = var_3294_begin_0, end = var_3294_end_0, end_mask = var_3294_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_3294_cast_fp16")]; + tensor var_3298_begin_0 = const()[name = tensor("op_3298_begin_0"), val = tensor([0, 0, 0, 160])]; + tensor var_3298_end_0 = const()[name = tensor("op_3298_end_0"), val = tensor([2, 1024, 1, 240])]; + tensor var_3298_end_mask_0 = const()[name = tensor("op_3298_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3298_cast_fp16 = slice_by_index(begin = var_3298_begin_0, end = var_3298_end_0, end_mask = var_3298_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_3298_cast_fp16")]; + tensor var_3302_begin_0 = const()[name = tensor("op_3302_begin_0"), val = tensor([0, 0, 0, 240])]; + tensor var_3302_end_0 = const()[name = tensor("op_3302_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_3302_end_mask_0 = const()[name = tensor("op_3302_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3302_cast_fp16 = slice_by_index(begin = var_3302_begin_0, end = var_3302_end_0, end_mask = var_3302_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_3302_cast_fp16")]; + tensor var_3306_begin_0 = const()[name = tensor("op_3306_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_3306_end_0 = const()[name = tensor("op_3306_end_0"), val = tensor([2, 1024, 1, 400])]; + tensor var_3306_end_mask_0 = const()[name = tensor("op_3306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3306_cast_fp16 = slice_by_index(begin = var_3306_begin_0, end = var_3306_end_0, end_mask = var_3306_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_3306_cast_fp16")]; + tensor var_3310_begin_0 = const()[name = tensor("op_3310_begin_0"), val = tensor([0, 0, 0, 400])]; + tensor var_3310_end_0 = const()[name = tensor("op_3310_end_0"), val = tensor([2, 1024, 1, 480])]; + tensor var_3310_end_mask_0 = const()[name = tensor("op_3310_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3310_cast_fp16 = slice_by_index(begin = var_3310_begin_0, end = var_3310_end_0, end_mask = var_3310_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_3310_cast_fp16")]; + tensor var_3314_begin_0 = const()[name = tensor("op_3314_begin_0"), val = tensor([0, 0, 0, 480])]; + tensor var_3314_end_0 = const()[name = tensor("op_3314_end_0"), val = tensor([2, 1024, 1, 560])]; + tensor var_3314_end_mask_0 = const()[name = tensor("op_3314_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3314_cast_fp16 = slice_by_index(begin = var_3314_begin_0, end = var_3314_end_0, end_mask = var_3314_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_3314_cast_fp16")]; + tensor var_3318_begin_0 = const()[name = tensor("op_3318_begin_0"), val = tensor([0, 0, 0, 560])]; + tensor var_3318_end_0 = const()[name = tensor("op_3318_end_0"), val = tensor([2, 1024, 1, 1])]; + tensor var_3318_end_mask_0 = const()[name = tensor("op_3318_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3318_cast_fp16 = slice_by_index(begin = var_3318_begin_0, end = var_3318_end_0, end_mask = var_3318_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_3318_cast_fp16")]; + tensor var_3320_begin_0 = const()[name = tensor("op_3320_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3320_end_0 = const()[name = tensor("op_3320_end_0"), val = tensor([2, 80, 1, 1024])]; + tensor var_3320_end_mask_0 = const()[name = tensor("op_3320_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3320_cast_fp16 = slice_by_index(begin = var_3320_begin_0, end = var_3320_end_0, end_mask = var_3320_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_3320_cast_fp16")]; + tensor var_3324_begin_0 = const()[name = tensor("op_3324_begin_0"), val = tensor([0, 80, 0, 0])]; + tensor var_3324_end_0 = const()[name = tensor("op_3324_end_0"), val = tensor([2, 160, 1, 1024])]; + tensor var_3324_end_mask_0 = const()[name = tensor("op_3324_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3324_cast_fp16 = slice_by_index(begin = var_3324_begin_0, end = var_3324_end_0, end_mask = var_3324_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_3324_cast_fp16")]; + tensor var_3328_begin_0 = const()[name = tensor("op_3328_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_3328_end_0 = const()[name = tensor("op_3328_end_0"), val = tensor([2, 240, 1, 1024])]; + tensor var_3328_end_mask_0 = const()[name = tensor("op_3328_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3328_cast_fp16 = slice_by_index(begin = var_3328_begin_0, end = var_3328_end_0, end_mask = var_3328_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_3328_cast_fp16")]; + tensor var_3332_begin_0 = const()[name = tensor("op_3332_begin_0"), val = tensor([0, 240, 0, 0])]; + tensor var_3332_end_0 = const()[name = tensor("op_3332_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_3332_end_mask_0 = const()[name = tensor("op_3332_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3332_cast_fp16 = slice_by_index(begin = var_3332_begin_0, end = var_3332_end_0, end_mask = var_3332_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_3332_cast_fp16")]; + tensor var_3336_begin_0 = const()[name = tensor("op_3336_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3336_end_0 = const()[name = tensor("op_3336_end_0"), val = tensor([2, 400, 1, 1024])]; + tensor var_3336_end_mask_0 = const()[name = tensor("op_3336_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3336_cast_fp16 = slice_by_index(begin = var_3336_begin_0, end = var_3336_end_0, end_mask = var_3336_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_3336_cast_fp16")]; + tensor var_3340_begin_0 = const()[name = tensor("op_3340_begin_0"), val = tensor([0, 400, 0, 0])]; + tensor var_3340_end_0 = const()[name = tensor("op_3340_end_0"), val = tensor([2, 480, 1, 1024])]; + tensor var_3340_end_mask_0 = const()[name = tensor("op_3340_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3340_cast_fp16 = slice_by_index(begin = var_3340_begin_0, end = var_3340_end_0, end_mask = var_3340_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_3340_cast_fp16")]; + tensor var_3344_begin_0 = const()[name = tensor("op_3344_begin_0"), val = tensor([0, 480, 0, 0])]; + tensor var_3344_end_0 = const()[name = tensor("op_3344_end_0"), val = tensor([2, 560, 1, 1024])]; + tensor var_3344_end_mask_0 = const()[name = tensor("op_3344_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3344_cast_fp16 = slice_by_index(begin = var_3344_begin_0, end = var_3344_end_0, end_mask = var_3344_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_3344_cast_fp16")]; + tensor var_3348_begin_0 = const()[name = tensor("op_3348_begin_0"), val = tensor([0, 560, 0, 0])]; + tensor var_3348_end_0 = const()[name = tensor("op_3348_end_0"), val = tensor([2, 1, 1, 1024])]; + tensor var_3348_end_mask_0 = const()[name = tensor("op_3348_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3348_cast_fp16 = slice_by_index(begin = var_3348_begin_0, end = var_3348_end_0, end_mask = var_3348_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_3348_cast_fp16")]; + tensor var_3352_equation_0 = const()[name = tensor("op_3352_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3352_cast_fp16 = einsum(equation = var_3352_equation_0, values = (var_3290_cast_fp16, var_3270_cast_fp16))[name = tensor("op_3352_cast_fp16")]; + tensor var_3353_to_fp16 = const()[name = tensor("op_3353_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_513_cast_fp16 = mul(x = var_3352_cast_fp16, y = var_3353_to_fp16)[name = tensor("aw_chunk_513_cast_fp16")]; + tensor var_3356_equation_0 = const()[name = tensor("op_3356_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3356_cast_fp16 = einsum(equation = var_3356_equation_0, values = (var_3290_cast_fp16, var_3271_cast_fp16))[name = tensor("op_3356_cast_fp16")]; + tensor var_3357_to_fp16 = const()[name = tensor("op_3357_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_515_cast_fp16 = mul(x = var_3356_cast_fp16, y = var_3357_to_fp16)[name = tensor("aw_chunk_515_cast_fp16")]; + tensor var_3360_equation_0 = const()[name = tensor("op_3360_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3360_cast_fp16 = einsum(equation = var_3360_equation_0, values = (var_3294_cast_fp16, var_3272_cast_fp16))[name = tensor("op_3360_cast_fp16")]; + tensor var_3361_to_fp16 = const()[name = tensor("op_3361_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_517_cast_fp16 = mul(x = var_3360_cast_fp16, y = var_3361_to_fp16)[name = tensor("aw_chunk_517_cast_fp16")]; + tensor var_3364_equation_0 = const()[name = tensor("op_3364_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3364_cast_fp16 = einsum(equation = var_3364_equation_0, values = (var_3294_cast_fp16, var_3273_cast_fp16))[name = tensor("op_3364_cast_fp16")]; + tensor var_3365_to_fp16 = const()[name = tensor("op_3365_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_519_cast_fp16 = mul(x = var_3364_cast_fp16, y = var_3365_to_fp16)[name = tensor("aw_chunk_519_cast_fp16")]; + tensor var_3368_equation_0 = const()[name = tensor("op_3368_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3368_cast_fp16 = einsum(equation = var_3368_equation_0, values = (var_3298_cast_fp16, var_3274_cast_fp16))[name = tensor("op_3368_cast_fp16")]; + tensor var_3369_to_fp16 = const()[name = tensor("op_3369_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_521_cast_fp16 = mul(x = var_3368_cast_fp16, y = var_3369_to_fp16)[name = tensor("aw_chunk_521_cast_fp16")]; + tensor var_3372_equation_0 = const()[name = tensor("op_3372_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3372_cast_fp16 = einsum(equation = var_3372_equation_0, values = (var_3298_cast_fp16, var_3275_cast_fp16))[name = tensor("op_3372_cast_fp16")]; + tensor var_3373_to_fp16 = const()[name = tensor("op_3373_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_523_cast_fp16 = mul(x = var_3372_cast_fp16, y = var_3373_to_fp16)[name = tensor("aw_chunk_523_cast_fp16")]; + tensor var_3376_equation_0 = const()[name = tensor("op_3376_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3376_cast_fp16 = einsum(equation = var_3376_equation_0, values = (var_3302_cast_fp16, var_3276_cast_fp16))[name = tensor("op_3376_cast_fp16")]; + tensor var_3377_to_fp16 = const()[name = tensor("op_3377_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_525_cast_fp16 = mul(x = var_3376_cast_fp16, y = var_3377_to_fp16)[name = tensor("aw_chunk_525_cast_fp16")]; + tensor var_3380_equation_0 = const()[name = tensor("op_3380_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3380_cast_fp16 = einsum(equation = var_3380_equation_0, values = (var_3302_cast_fp16, var_3277_cast_fp16))[name = tensor("op_3380_cast_fp16")]; + tensor var_3381_to_fp16 = const()[name = tensor("op_3381_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_527_cast_fp16 = mul(x = var_3380_cast_fp16, y = var_3381_to_fp16)[name = tensor("aw_chunk_527_cast_fp16")]; + tensor var_3384_equation_0 = const()[name = tensor("op_3384_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3384_cast_fp16 = einsum(equation = var_3384_equation_0, values = (var_3306_cast_fp16, var_3278_cast_fp16))[name = tensor("op_3384_cast_fp16")]; + tensor var_3385_to_fp16 = const()[name = tensor("op_3385_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_529_cast_fp16 = mul(x = var_3384_cast_fp16, y = var_3385_to_fp16)[name = tensor("aw_chunk_529_cast_fp16")]; + tensor var_3388_equation_0 = const()[name = tensor("op_3388_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3388_cast_fp16 = einsum(equation = var_3388_equation_0, values = (var_3306_cast_fp16, var_3279_cast_fp16))[name = tensor("op_3388_cast_fp16")]; + tensor var_3389_to_fp16 = const()[name = tensor("op_3389_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_531_cast_fp16 = mul(x = var_3388_cast_fp16, y = var_3389_to_fp16)[name = tensor("aw_chunk_531_cast_fp16")]; + tensor var_3392_equation_0 = const()[name = tensor("op_3392_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3392_cast_fp16 = einsum(equation = var_3392_equation_0, values = (var_3310_cast_fp16, var_3280_cast_fp16))[name = tensor("op_3392_cast_fp16")]; + tensor var_3393_to_fp16 = const()[name = tensor("op_3393_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_533_cast_fp16 = mul(x = var_3392_cast_fp16, y = var_3393_to_fp16)[name = tensor("aw_chunk_533_cast_fp16")]; + tensor var_3396_equation_0 = const()[name = tensor("op_3396_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3396_cast_fp16 = einsum(equation = var_3396_equation_0, values = (var_3310_cast_fp16, var_3281_cast_fp16))[name = tensor("op_3396_cast_fp16")]; + tensor var_3397_to_fp16 = const()[name = tensor("op_3397_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_535_cast_fp16 = mul(x = var_3396_cast_fp16, y = var_3397_to_fp16)[name = tensor("aw_chunk_535_cast_fp16")]; + tensor var_3400_equation_0 = const()[name = tensor("op_3400_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3400_cast_fp16 = einsum(equation = var_3400_equation_0, values = (var_3314_cast_fp16, var_3282_cast_fp16))[name = tensor("op_3400_cast_fp16")]; + tensor var_3401_to_fp16 = const()[name = tensor("op_3401_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_537_cast_fp16 = mul(x = var_3400_cast_fp16, y = var_3401_to_fp16)[name = tensor("aw_chunk_537_cast_fp16")]; + tensor var_3404_equation_0 = const()[name = tensor("op_3404_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3404_cast_fp16 = einsum(equation = var_3404_equation_0, values = (var_3314_cast_fp16, var_3283_cast_fp16))[name = tensor("op_3404_cast_fp16")]; + tensor var_3405_to_fp16 = const()[name = tensor("op_3405_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_539_cast_fp16 = mul(x = var_3404_cast_fp16, y = var_3405_to_fp16)[name = tensor("aw_chunk_539_cast_fp16")]; + tensor var_3408_equation_0 = const()[name = tensor("op_3408_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3408_cast_fp16 = einsum(equation = var_3408_equation_0, values = (var_3318_cast_fp16, var_3284_cast_fp16))[name = tensor("op_3408_cast_fp16")]; + tensor var_3409_to_fp16 = const()[name = tensor("op_3409_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_541_cast_fp16 = mul(x = var_3408_cast_fp16, y = var_3409_to_fp16)[name = tensor("aw_chunk_541_cast_fp16")]; + tensor var_3412_equation_0 = const()[name = tensor("op_3412_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3412_cast_fp16 = einsum(equation = var_3412_equation_0, values = (var_3318_cast_fp16, var_3285_cast_fp16))[name = tensor("op_3412_cast_fp16")]; + tensor var_3413_to_fp16 = const()[name = tensor("op_3413_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_543_cast_fp16 = mul(x = var_3412_cast_fp16, y = var_3413_to_fp16)[name = tensor("aw_chunk_543_cast_fp16")]; + tensor var_3415_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_513_cast_fp16)[name = tensor("op_3415_cast_fp16")]; + tensor var_3416_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_515_cast_fp16)[name = tensor("op_3416_cast_fp16")]; + tensor var_3417_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_517_cast_fp16)[name = tensor("op_3417_cast_fp16")]; + tensor var_3418_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_519_cast_fp16)[name = tensor("op_3418_cast_fp16")]; + tensor var_3419_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_521_cast_fp16)[name = tensor("op_3419_cast_fp16")]; + tensor var_3420_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_523_cast_fp16)[name = tensor("op_3420_cast_fp16")]; + tensor var_3421_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_525_cast_fp16)[name = tensor("op_3421_cast_fp16")]; + tensor var_3422_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_527_cast_fp16)[name = tensor("op_3422_cast_fp16")]; + tensor var_3423_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_529_cast_fp16)[name = tensor("op_3423_cast_fp16")]; + tensor var_3424_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_531_cast_fp16)[name = tensor("op_3424_cast_fp16")]; + tensor var_3425_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_533_cast_fp16)[name = tensor("op_3425_cast_fp16")]; + tensor var_3426_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_535_cast_fp16)[name = tensor("op_3426_cast_fp16")]; + tensor var_3427_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_537_cast_fp16)[name = tensor("op_3427_cast_fp16")]; + tensor var_3428_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_539_cast_fp16)[name = tensor("op_3428_cast_fp16")]; + tensor var_3429_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_541_cast_fp16)[name = tensor("op_3429_cast_fp16")]; + tensor var_3430_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_543_cast_fp16)[name = tensor("op_3430_cast_fp16")]; + tensor var_3432_equation_0 = const()[name = tensor("op_3432_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3432_cast_fp16 = einsum(equation = var_3432_equation_0, values = (var_3320_cast_fp16, var_3415_cast_fp16))[name = tensor("op_3432_cast_fp16")]; + tensor var_3434_equation_0 = const()[name = tensor("op_3434_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3434_cast_fp16 = einsum(equation = var_3434_equation_0, values = (var_3320_cast_fp16, var_3416_cast_fp16))[name = tensor("op_3434_cast_fp16")]; + tensor var_3436_equation_0 = const()[name = tensor("op_3436_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3436_cast_fp16 = einsum(equation = var_3436_equation_0, values = (var_3324_cast_fp16, var_3417_cast_fp16))[name = tensor("op_3436_cast_fp16")]; + tensor var_3438_equation_0 = const()[name = tensor("op_3438_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3438_cast_fp16 = einsum(equation = var_3438_equation_0, values = (var_3324_cast_fp16, var_3418_cast_fp16))[name = tensor("op_3438_cast_fp16")]; + tensor var_3440_equation_0 = const()[name = tensor("op_3440_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3440_cast_fp16 = einsum(equation = var_3440_equation_0, values = (var_3328_cast_fp16, var_3419_cast_fp16))[name = tensor("op_3440_cast_fp16")]; + tensor var_3442_equation_0 = const()[name = tensor("op_3442_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3442_cast_fp16 = einsum(equation = var_3442_equation_0, values = (var_3328_cast_fp16, var_3420_cast_fp16))[name = tensor("op_3442_cast_fp16")]; + tensor var_3444_equation_0 = const()[name = tensor("op_3444_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3444_cast_fp16 = einsum(equation = var_3444_equation_0, values = (var_3332_cast_fp16, var_3421_cast_fp16))[name = tensor("op_3444_cast_fp16")]; + tensor var_3446_equation_0 = const()[name = tensor("op_3446_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3446_cast_fp16 = einsum(equation = var_3446_equation_0, values = (var_3332_cast_fp16, var_3422_cast_fp16))[name = tensor("op_3446_cast_fp16")]; + tensor var_3448_equation_0 = const()[name = tensor("op_3448_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3448_cast_fp16 = einsum(equation = var_3448_equation_0, values = (var_3336_cast_fp16, var_3423_cast_fp16))[name = tensor("op_3448_cast_fp16")]; + tensor var_3450_equation_0 = const()[name = tensor("op_3450_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3450_cast_fp16 = einsum(equation = var_3450_equation_0, values = (var_3336_cast_fp16, var_3424_cast_fp16))[name = tensor("op_3450_cast_fp16")]; + tensor var_3452_equation_0 = const()[name = tensor("op_3452_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3452_cast_fp16 = einsum(equation = var_3452_equation_0, values = (var_3340_cast_fp16, var_3425_cast_fp16))[name = tensor("op_3452_cast_fp16")]; + tensor var_3454_equation_0 = const()[name = tensor("op_3454_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3454_cast_fp16 = einsum(equation = var_3454_equation_0, values = (var_3340_cast_fp16, var_3426_cast_fp16))[name = tensor("op_3454_cast_fp16")]; + tensor var_3456_equation_0 = const()[name = tensor("op_3456_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3456_cast_fp16 = einsum(equation = var_3456_equation_0, values = (var_3344_cast_fp16, var_3427_cast_fp16))[name = tensor("op_3456_cast_fp16")]; + tensor var_3458_equation_0 = const()[name = tensor("op_3458_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3458_cast_fp16 = einsum(equation = var_3458_equation_0, values = (var_3344_cast_fp16, var_3428_cast_fp16))[name = tensor("op_3458_cast_fp16")]; + tensor var_3460_equation_0 = const()[name = tensor("op_3460_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3460_cast_fp16 = einsum(equation = var_3460_equation_0, values = (var_3348_cast_fp16, var_3429_cast_fp16))[name = tensor("op_3460_cast_fp16")]; + tensor var_3462_equation_0 = const()[name = tensor("op_3462_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3462_cast_fp16 = einsum(equation = var_3462_equation_0, values = (var_3348_cast_fp16, var_3430_cast_fp16))[name = tensor("op_3462_cast_fp16")]; + tensor var_3464_interleave_0 = const()[name = tensor("op_3464_interleave_0"), val = tensor(false)]; + tensor var_3464_cast_fp16 = concat(axis = var_3089, interleave = var_3464_interleave_0, values = (var_3432_cast_fp16, var_3434_cast_fp16))[name = tensor("op_3464_cast_fp16")]; + tensor var_3466_interleave_0 = const()[name = tensor("op_3466_interleave_0"), val = tensor(false)]; + tensor var_3466_cast_fp16 = concat(axis = var_3089, interleave = var_3466_interleave_0, values = (var_3436_cast_fp16, var_3438_cast_fp16))[name = tensor("op_3466_cast_fp16")]; + tensor var_3468_interleave_0 = const()[name = tensor("op_3468_interleave_0"), val = tensor(false)]; + tensor var_3468_cast_fp16 = concat(axis = var_3089, interleave = var_3468_interleave_0, values = (var_3440_cast_fp16, var_3442_cast_fp16))[name = tensor("op_3468_cast_fp16")]; + tensor var_3470_interleave_0 = const()[name = tensor("op_3470_interleave_0"), val = tensor(false)]; + tensor var_3470_cast_fp16 = concat(axis = var_3089, interleave = var_3470_interleave_0, values = (var_3444_cast_fp16, var_3446_cast_fp16))[name = tensor("op_3470_cast_fp16")]; + tensor var_3472_interleave_0 = const()[name = tensor("op_3472_interleave_0"), val = tensor(false)]; + tensor var_3472_cast_fp16 = concat(axis = var_3089, interleave = var_3472_interleave_0, values = (var_3448_cast_fp16, var_3450_cast_fp16))[name = tensor("op_3472_cast_fp16")]; + tensor var_3474_interleave_0 = const()[name = tensor("op_3474_interleave_0"), val = tensor(false)]; + tensor var_3474_cast_fp16 = concat(axis = var_3089, interleave = var_3474_interleave_0, values = (var_3452_cast_fp16, var_3454_cast_fp16))[name = tensor("op_3474_cast_fp16")]; + tensor var_3476_interleave_0 = const()[name = tensor("op_3476_interleave_0"), val = tensor(false)]; + tensor var_3476_cast_fp16 = concat(axis = var_3089, interleave = var_3476_interleave_0, values = (var_3456_cast_fp16, var_3458_cast_fp16))[name = tensor("op_3476_cast_fp16")]; + tensor var_3478_interleave_0 = const()[name = tensor("op_3478_interleave_0"), val = tensor(false)]; + tensor var_3478_cast_fp16 = concat(axis = var_3089, interleave = var_3478_interleave_0, values = (var_3460_cast_fp16, var_3462_cast_fp16))[name = tensor("op_3478_cast_fp16")]; + tensor input_79_interleave_0 = const()[name = tensor("input_79_interleave_0"), val = tensor(false)]; + tensor input_79_cast_fp16 = concat(axis = var_3111, interleave = input_79_interleave_0, values = (var_3464_cast_fp16, var_3466_cast_fp16, var_3468_cast_fp16, var_3470_cast_fp16, var_3472_cast_fp16, var_3474_cast_fp16, var_3476_cast_fp16, var_3478_cast_fp16))[name = tensor("input_79_cast_fp16")]; + tensor var_3488_pad_type_0 = const()[name = tensor("op_3488_pad_type_0"), val = tensor("valid")]; + tensor var_3488_strides_0 = const()[name = tensor("op_3488_strides_0"), val = tensor([1, 1])]; + tensor var_3488_pad_0 = const()[name = tensor("op_3488_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3488_dilations_0 = const()[name = tensor("op_3488_dilations_0"), val = tensor([1, 1])]; + tensor var_3488_groups_0 = const()[name = tensor("op_3488_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_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(41584128)))]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_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(42403392)))]; + tensor var_3488_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_3488_dilations_0, groups = var_3488_groups_0, pad = var_3488_pad_0, pad_type = var_3488_pad_type_0, strides = var_3488_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_79_cast_fp16)[name = tensor("op_3488_cast_fp16")]; + tensor inputs_15_cast_fp16 = add(x = var_3488_cast_fp16, y = inputs_13_cast_fp16)[name = tensor("inputs_15_cast_fp16")]; + tensor hidden_states_47_axes_0 = const()[name = tensor("hidden_states_47_axes_0"), val = tensor([1])]; + tensor hidden_states_47_gamma_0_to_fp16 = const()[name = tensor("hidden_states_47_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42404736)))]; + tensor hidden_states_47_beta_0_to_fp16 = const()[name = tensor("hidden_states_47_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42406080)))]; + tensor var_3498_to_fp16 = const()[name = tensor("op_3498_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_47_cast_fp16 = layer_norm(axes = hidden_states_47_axes_0, beta = hidden_states_47_beta_0_to_fp16, epsilon = var_3498_to_fp16, gamma = hidden_states_47_gamma_0_to_fp16, x = inputs_15_cast_fp16)[name = tensor("hidden_states_47_cast_fp16")]; + tensor q_11_pad_type_0 = const()[name = tensor("q_11_pad_type_0"), val = tensor("valid")]; + tensor q_11_strides_0 = const()[name = tensor("q_11_strides_0"), val = tensor([1, 1])]; + tensor q_11_pad_0 = const()[name = tensor("q_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_11_dilations_0 = const()[name = tensor("q_11_dilations_0"), val = tensor([1, 1])]; + tensor q_11_groups_0 = const()[name = tensor("q_11_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_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(42407424)))]; + tensor q_11_cast_fp16 = conv(dilations = q_11_dilations_0, groups = q_11_groups_0, pad = q_11_pad_0, pad_type = q_11_pad_type_0, strides = q_11_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_47_cast_fp16)[name = tensor("q_11_cast_fp16")]; + tensor k_21_pad_type_0 = const()[name = tensor("k_21_pad_type_0"), val = tensor("valid")]; + tensor k_21_strides_0 = const()[name = tensor("k_21_strides_0"), val = tensor([1, 1])]; + tensor k_21_pad_0 = const()[name = tensor("k_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_21_dilations_0 = const()[name = tensor("k_21_dilations_0"), val = tensor([1, 1])]; + tensor k_21_groups_0 = const()[name = tensor("k_21_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_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(43226688)))]; + tensor k_21_cast_fp16 = conv(dilations = k_21_dilations_0, groups = k_21_groups_0, pad = k_21_pad_0, pad_type = k_21_pad_type_0, strides = k_21_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_21_cast_fp16")]; + tensor v_11_pad_type_0 = const()[name = tensor("v_11_pad_type_0"), val = tensor("valid")]; + tensor v_11_strides_0 = const()[name = tensor("v_11_strides_0"), val = tensor([1, 1])]; + tensor v_11_pad_0 = const()[name = tensor("v_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_11_dilations_0 = const()[name = tensor("v_11_dilations_0"), val = tensor([1, 1])]; + tensor v_11_groups_0 = const()[name = tensor("v_11_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_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(44209792)))]; + tensor v_11_cast_fp16 = conv(dilations = v_11_dilations_0, groups = v_11_groups_0, pad = v_11_pad_0, pad_type = v_11_pad_type_0, strides = v_11_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_11_cast_fp16")]; + tensor var_3531_begin_0 = const()[name = tensor("op_3531_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3531_end_0 = const()[name = tensor("op_3531_end_0"), val = tensor([2, 80, 1, 1024])]; + tensor var_3531_end_mask_0 = const()[name = tensor("op_3531_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3531_cast_fp16 = slice_by_index(begin = var_3531_begin_0, end = var_3531_end_0, end_mask = var_3531_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_3531_cast_fp16")]; + tensor var_3535_begin_0 = const()[name = tensor("op_3535_begin_0"), val = tensor([0, 80, 0, 0])]; + tensor var_3535_end_0 = const()[name = tensor("op_3535_end_0"), val = tensor([2, 160, 1, 1024])]; + tensor var_3535_end_mask_0 = const()[name = tensor("op_3535_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3535_cast_fp16 = slice_by_index(begin = var_3535_begin_0, end = var_3535_end_0, end_mask = var_3535_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_3535_cast_fp16")]; + tensor var_3539_begin_0 = const()[name = tensor("op_3539_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_3539_end_0 = const()[name = tensor("op_3539_end_0"), val = tensor([2, 240, 1, 1024])]; + tensor var_3539_end_mask_0 = const()[name = tensor("op_3539_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3539_cast_fp16 = slice_by_index(begin = var_3539_begin_0, end = var_3539_end_0, end_mask = var_3539_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_3539_cast_fp16")]; + tensor var_3543_begin_0 = const()[name = tensor("op_3543_begin_0"), val = tensor([0, 240, 0, 0])]; + tensor var_3543_end_0 = const()[name = tensor("op_3543_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_3543_end_mask_0 = const()[name = tensor("op_3543_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3543_cast_fp16 = slice_by_index(begin = var_3543_begin_0, end = var_3543_end_0, end_mask = var_3543_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_3543_cast_fp16")]; + tensor var_3547_begin_0 = const()[name = tensor("op_3547_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3547_end_0 = const()[name = tensor("op_3547_end_0"), val = tensor([2, 400, 1, 1024])]; + tensor var_3547_end_mask_0 = const()[name = tensor("op_3547_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3547_cast_fp16 = slice_by_index(begin = var_3547_begin_0, end = var_3547_end_0, end_mask = var_3547_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_3547_cast_fp16")]; + tensor var_3551_begin_0 = const()[name = tensor("op_3551_begin_0"), val = tensor([0, 400, 0, 0])]; + tensor var_3551_end_0 = const()[name = tensor("op_3551_end_0"), val = tensor([2, 480, 1, 1024])]; + tensor var_3551_end_mask_0 = const()[name = tensor("op_3551_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3551_cast_fp16 = slice_by_index(begin = var_3551_begin_0, end = var_3551_end_0, end_mask = var_3551_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_3551_cast_fp16")]; + tensor var_3555_begin_0 = const()[name = tensor("op_3555_begin_0"), val = tensor([0, 480, 0, 0])]; + tensor var_3555_end_0 = const()[name = tensor("op_3555_end_0"), val = tensor([2, 560, 1, 1024])]; + tensor var_3555_end_mask_0 = const()[name = tensor("op_3555_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3555_cast_fp16 = slice_by_index(begin = var_3555_begin_0, end = var_3555_end_0, end_mask = var_3555_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_3555_cast_fp16")]; + tensor var_3559_begin_0 = const()[name = tensor("op_3559_begin_0"), val = tensor([0, 560, 0, 0])]; + tensor var_3559_end_0 = const()[name = tensor("op_3559_end_0"), val = tensor([2, 1, 1, 1024])]; + tensor var_3559_end_mask_0 = const()[name = tensor("op_3559_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3559_cast_fp16 = slice_by_index(begin = var_3559_begin_0, end = var_3559_end_0, end_mask = var_3559_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_3559_cast_fp16")]; + tensor var_3562_begin_0 = const()[name = tensor("op_3562_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3562_end_0 = const()[name = tensor("op_3562_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_3562_end_mask_0 = const()[name = tensor("op_3562_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3562_cast_fp16 = slice_by_index(begin = var_3562_begin_0, end = var_3562_end_0, end_mask = var_3562_end_mask_0, x = var_3531_cast_fp16)[name = tensor("op_3562_cast_fp16")]; + tensor var_3563_begin_0 = const()[name = tensor("op_3563_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3563_end_0 = const()[name = tensor("op_3563_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_3563_end_mask_0 = const()[name = tensor("op_3563_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3563_cast_fp16 = slice_by_index(begin = var_3563_begin_0, end = var_3563_end_0, end_mask = var_3563_end_mask_0, x = var_3531_cast_fp16)[name = tensor("op_3563_cast_fp16")]; + tensor var_3564_begin_0 = const()[name = tensor("op_3564_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3564_end_0 = const()[name = tensor("op_3564_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_3564_end_mask_0 = const()[name = tensor("op_3564_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3564_cast_fp16 = slice_by_index(begin = var_3564_begin_0, end = var_3564_end_0, end_mask = var_3564_end_mask_0, x = var_3535_cast_fp16)[name = tensor("op_3564_cast_fp16")]; + tensor var_3565_begin_0 = const()[name = tensor("op_3565_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3565_end_0 = const()[name = tensor("op_3565_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_3565_end_mask_0 = const()[name = tensor("op_3565_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3565_cast_fp16 = slice_by_index(begin = var_3565_begin_0, end = var_3565_end_0, end_mask = var_3565_end_mask_0, x = var_3535_cast_fp16)[name = tensor("op_3565_cast_fp16")]; + tensor var_3566_begin_0 = const()[name = tensor("op_3566_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3566_end_0 = const()[name = tensor("op_3566_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_3566_end_mask_0 = const()[name = tensor("op_3566_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3566_cast_fp16 = slice_by_index(begin = var_3566_begin_0, end = var_3566_end_0, end_mask = var_3566_end_mask_0, x = var_3539_cast_fp16)[name = tensor("op_3566_cast_fp16")]; + tensor var_3567_begin_0 = const()[name = tensor("op_3567_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3567_end_0 = const()[name = tensor("op_3567_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_3567_end_mask_0 = const()[name = tensor("op_3567_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3567_cast_fp16 = slice_by_index(begin = var_3567_begin_0, end = var_3567_end_0, end_mask = var_3567_end_mask_0, x = var_3539_cast_fp16)[name = tensor("op_3567_cast_fp16")]; + tensor var_3568_begin_0 = const()[name = tensor("op_3568_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3568_end_0 = const()[name = tensor("op_3568_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_3568_end_mask_0 = const()[name = tensor("op_3568_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3568_cast_fp16 = slice_by_index(begin = var_3568_begin_0, end = var_3568_end_0, end_mask = var_3568_end_mask_0, x = var_3543_cast_fp16)[name = tensor("op_3568_cast_fp16")]; + tensor var_3569_begin_0 = const()[name = tensor("op_3569_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3569_end_0 = const()[name = tensor("op_3569_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_3569_end_mask_0 = const()[name = tensor("op_3569_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3569_cast_fp16 = slice_by_index(begin = var_3569_begin_0, end = var_3569_end_0, end_mask = var_3569_end_mask_0, x = var_3543_cast_fp16)[name = tensor("op_3569_cast_fp16")]; + tensor var_3570_begin_0 = const()[name = tensor("op_3570_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3570_end_0 = const()[name = tensor("op_3570_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_3570_end_mask_0 = const()[name = tensor("op_3570_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3570_cast_fp16 = slice_by_index(begin = var_3570_begin_0, end = var_3570_end_0, end_mask = var_3570_end_mask_0, x = var_3547_cast_fp16)[name = tensor("op_3570_cast_fp16")]; + tensor var_3571_begin_0 = const()[name = tensor("op_3571_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3571_end_0 = const()[name = tensor("op_3571_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_3571_end_mask_0 = const()[name = tensor("op_3571_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3571_cast_fp16 = slice_by_index(begin = var_3571_begin_0, end = var_3571_end_0, end_mask = var_3571_end_mask_0, x = var_3547_cast_fp16)[name = tensor("op_3571_cast_fp16")]; + tensor var_3572_begin_0 = const()[name = tensor("op_3572_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3572_end_0 = const()[name = tensor("op_3572_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_3572_end_mask_0 = const()[name = tensor("op_3572_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3572_cast_fp16 = slice_by_index(begin = var_3572_begin_0, end = var_3572_end_0, end_mask = var_3572_end_mask_0, x = var_3551_cast_fp16)[name = tensor("op_3572_cast_fp16")]; + tensor var_3573_begin_0 = const()[name = tensor("op_3573_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3573_end_0 = const()[name = tensor("op_3573_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_3573_end_mask_0 = const()[name = tensor("op_3573_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3573_cast_fp16 = slice_by_index(begin = var_3573_begin_0, end = var_3573_end_0, end_mask = var_3573_end_mask_0, x = var_3551_cast_fp16)[name = tensor("op_3573_cast_fp16")]; + tensor var_3574_begin_0 = const()[name = tensor("op_3574_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3574_end_0 = const()[name = tensor("op_3574_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_3574_end_mask_0 = const()[name = tensor("op_3574_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3574_cast_fp16 = slice_by_index(begin = var_3574_begin_0, end = var_3574_end_0, end_mask = var_3574_end_mask_0, x = var_3555_cast_fp16)[name = tensor("op_3574_cast_fp16")]; + tensor var_3575_begin_0 = const()[name = tensor("op_3575_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3575_end_0 = const()[name = tensor("op_3575_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_3575_end_mask_0 = const()[name = tensor("op_3575_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3575_cast_fp16 = slice_by_index(begin = var_3575_begin_0, end = var_3575_end_0, end_mask = var_3575_end_mask_0, x = var_3555_cast_fp16)[name = tensor("op_3575_cast_fp16")]; + tensor var_3576_begin_0 = const()[name = tensor("op_3576_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3576_end_0 = const()[name = tensor("op_3576_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_3576_end_mask_0 = const()[name = tensor("op_3576_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3576_cast_fp16 = slice_by_index(begin = var_3576_begin_0, end = var_3576_end_0, end_mask = var_3576_end_mask_0, x = var_3559_cast_fp16)[name = tensor("op_3576_cast_fp16")]; + tensor var_3577_begin_0 = const()[name = tensor("op_3577_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3577_end_0 = const()[name = tensor("op_3577_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_3577_end_mask_0 = const()[name = tensor("op_3577_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3577_cast_fp16 = slice_by_index(begin = var_3577_begin_0, end = var_3577_end_0, end_mask = var_3577_end_mask_0, x = var_3559_cast_fp16)[name = tensor("op_3577_cast_fp16")]; + tensor k_23_perm_0 = const()[name = tensor("k_23_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_3582_begin_0 = const()[name = tensor("op_3582_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3582_end_0 = const()[name = tensor("op_3582_end_0"), val = tensor([2, 77, 1, 80])]; + tensor var_3582_end_mask_0 = const()[name = tensor("op_3582_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_23_cast_fp16 = transpose(perm = k_23_perm_0, x = k_21_cast_fp16)[name = tensor("transpose_8")]; + tensor var_3582_cast_fp16 = slice_by_index(begin = var_3582_begin_0, end = var_3582_end_0, end_mask = var_3582_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_3582_cast_fp16")]; + tensor var_3586_begin_0 = const()[name = tensor("op_3586_begin_0"), val = tensor([0, 0, 0, 80])]; + tensor var_3586_end_0 = const()[name = tensor("op_3586_end_0"), val = tensor([2, 77, 1, 160])]; + tensor var_3586_end_mask_0 = const()[name = tensor("op_3586_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3586_cast_fp16 = slice_by_index(begin = var_3586_begin_0, end = var_3586_end_0, end_mask = var_3586_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_3586_cast_fp16")]; + tensor var_3590_begin_0 = const()[name = tensor("op_3590_begin_0"), val = tensor([0, 0, 0, 160])]; + tensor var_3590_end_0 = const()[name = tensor("op_3590_end_0"), val = tensor([2, 77, 1, 240])]; + tensor var_3590_end_mask_0 = const()[name = tensor("op_3590_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3590_cast_fp16 = slice_by_index(begin = var_3590_begin_0, end = var_3590_end_0, end_mask = var_3590_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_3590_cast_fp16")]; + tensor var_3594_begin_0 = const()[name = tensor("op_3594_begin_0"), val = tensor([0, 0, 0, 240])]; + tensor var_3594_end_0 = const()[name = tensor("op_3594_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_3594_end_mask_0 = const()[name = tensor("op_3594_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3594_cast_fp16 = slice_by_index(begin = var_3594_begin_0, end = var_3594_end_0, end_mask = var_3594_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_3594_cast_fp16")]; + tensor var_3598_begin_0 = const()[name = tensor("op_3598_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_3598_end_0 = const()[name = tensor("op_3598_end_0"), val = tensor([2, 77, 1, 400])]; + tensor var_3598_end_mask_0 = const()[name = tensor("op_3598_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3598_cast_fp16 = slice_by_index(begin = var_3598_begin_0, end = var_3598_end_0, end_mask = var_3598_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_3598_cast_fp16")]; + tensor var_3602_begin_0 = const()[name = tensor("op_3602_begin_0"), val = tensor([0, 0, 0, 400])]; + tensor var_3602_end_0 = const()[name = tensor("op_3602_end_0"), val = tensor([2, 77, 1, 480])]; + tensor var_3602_end_mask_0 = const()[name = tensor("op_3602_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3602_cast_fp16 = slice_by_index(begin = var_3602_begin_0, end = var_3602_end_0, end_mask = var_3602_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_3602_cast_fp16")]; + tensor var_3606_begin_0 = const()[name = tensor("op_3606_begin_0"), val = tensor([0, 0, 0, 480])]; + tensor var_3606_end_0 = const()[name = tensor("op_3606_end_0"), val = tensor([2, 77, 1, 560])]; + tensor var_3606_end_mask_0 = const()[name = tensor("op_3606_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3606_cast_fp16 = slice_by_index(begin = var_3606_begin_0, end = var_3606_end_0, end_mask = var_3606_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_3606_cast_fp16")]; + tensor var_3610_begin_0 = const()[name = tensor("op_3610_begin_0"), val = tensor([0, 0, 0, 560])]; + tensor var_3610_end_0 = const()[name = tensor("op_3610_end_0"), val = tensor([2, 77, 1, 1])]; + tensor var_3610_end_mask_0 = const()[name = tensor("op_3610_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3610_cast_fp16 = slice_by_index(begin = var_3610_begin_0, end = var_3610_end_0, end_mask = var_3610_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_3610_cast_fp16")]; + tensor var_3612_begin_0 = const()[name = tensor("op_3612_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3612_end_0 = const()[name = tensor("op_3612_end_0"), val = tensor([2, 80, 1, 77])]; + tensor var_3612_end_mask_0 = const()[name = tensor("op_3612_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3612_cast_fp16 = slice_by_index(begin = var_3612_begin_0, end = var_3612_end_0, end_mask = var_3612_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_3612_cast_fp16")]; + tensor var_3616_begin_0 = const()[name = tensor("op_3616_begin_0"), val = tensor([0, 80, 0, 0])]; + tensor var_3616_end_0 = const()[name = tensor("op_3616_end_0"), val = tensor([2, 160, 1, 77])]; + tensor var_3616_end_mask_0 = const()[name = tensor("op_3616_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3616_cast_fp16 = slice_by_index(begin = var_3616_begin_0, end = var_3616_end_0, end_mask = var_3616_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_3616_cast_fp16")]; + tensor var_3620_begin_0 = const()[name = tensor("op_3620_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_3620_end_0 = const()[name = tensor("op_3620_end_0"), val = tensor([2, 240, 1, 77])]; + tensor var_3620_end_mask_0 = const()[name = tensor("op_3620_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3620_cast_fp16 = slice_by_index(begin = var_3620_begin_0, end = var_3620_end_0, end_mask = var_3620_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_3620_cast_fp16")]; + tensor var_3624_begin_0 = const()[name = tensor("op_3624_begin_0"), val = tensor([0, 240, 0, 0])]; + tensor var_3624_end_0 = const()[name = tensor("op_3624_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_3624_end_mask_0 = const()[name = tensor("op_3624_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3624_cast_fp16 = slice_by_index(begin = var_3624_begin_0, end = var_3624_end_0, end_mask = var_3624_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_3624_cast_fp16")]; + tensor var_3628_begin_0 = const()[name = tensor("op_3628_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3628_end_0 = const()[name = tensor("op_3628_end_0"), val = tensor([2, 400, 1, 77])]; + tensor var_3628_end_mask_0 = const()[name = tensor("op_3628_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3628_cast_fp16 = slice_by_index(begin = var_3628_begin_0, end = var_3628_end_0, end_mask = var_3628_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_3628_cast_fp16")]; + tensor var_3632_begin_0 = const()[name = tensor("op_3632_begin_0"), val = tensor([0, 400, 0, 0])]; + tensor var_3632_end_0 = const()[name = tensor("op_3632_end_0"), val = tensor([2, 480, 1, 77])]; + tensor var_3632_end_mask_0 = const()[name = tensor("op_3632_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3632_cast_fp16 = slice_by_index(begin = var_3632_begin_0, end = var_3632_end_0, end_mask = var_3632_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_3632_cast_fp16")]; + tensor var_3636_begin_0 = const()[name = tensor("op_3636_begin_0"), val = tensor([0, 480, 0, 0])]; + tensor var_3636_end_0 = const()[name = tensor("op_3636_end_0"), val = tensor([2, 560, 1, 77])]; + tensor var_3636_end_mask_0 = const()[name = tensor("op_3636_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3636_cast_fp16 = slice_by_index(begin = var_3636_begin_0, end = var_3636_end_0, end_mask = var_3636_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_3636_cast_fp16")]; + tensor var_3640_begin_0 = const()[name = tensor("op_3640_begin_0"), val = tensor([0, 560, 0, 0])]; + tensor var_3640_end_0 = const()[name = tensor("op_3640_end_0"), val = tensor([2, 1, 1, 77])]; + tensor var_3640_end_mask_0 = const()[name = tensor("op_3640_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3640_cast_fp16 = slice_by_index(begin = var_3640_begin_0, end = var_3640_end_0, end_mask = var_3640_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_3640_cast_fp16")]; + tensor var_3644_equation_0 = const()[name = tensor("op_3644_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3644_cast_fp16 = einsum(equation = var_3644_equation_0, values = (var_3582_cast_fp16, var_3562_cast_fp16))[name = tensor("op_3644_cast_fp16")]; + tensor var_3645_to_fp16 = const()[name = tensor("op_3645_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_545_cast_fp16 = mul(x = var_3644_cast_fp16, y = var_3645_to_fp16)[name = tensor("aw_chunk_545_cast_fp16")]; + tensor var_3648_equation_0 = const()[name = tensor("op_3648_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3648_cast_fp16 = einsum(equation = var_3648_equation_0, values = (var_3582_cast_fp16, var_3563_cast_fp16))[name = tensor("op_3648_cast_fp16")]; + tensor var_3649_to_fp16 = const()[name = tensor("op_3649_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_547_cast_fp16 = mul(x = var_3648_cast_fp16, y = var_3649_to_fp16)[name = tensor("aw_chunk_547_cast_fp16")]; + tensor var_3652_equation_0 = const()[name = tensor("op_3652_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3652_cast_fp16 = einsum(equation = var_3652_equation_0, values = (var_3586_cast_fp16, var_3564_cast_fp16))[name = tensor("op_3652_cast_fp16")]; + tensor var_3653_to_fp16 = const()[name = tensor("op_3653_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_549_cast_fp16 = mul(x = var_3652_cast_fp16, y = var_3653_to_fp16)[name = tensor("aw_chunk_549_cast_fp16")]; + tensor var_3656_equation_0 = const()[name = tensor("op_3656_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3656_cast_fp16 = einsum(equation = var_3656_equation_0, values = (var_3586_cast_fp16, var_3565_cast_fp16))[name = tensor("op_3656_cast_fp16")]; + tensor var_3657_to_fp16 = const()[name = tensor("op_3657_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_551_cast_fp16 = mul(x = var_3656_cast_fp16, y = var_3657_to_fp16)[name = tensor("aw_chunk_551_cast_fp16")]; + tensor var_3660_equation_0 = const()[name = tensor("op_3660_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3660_cast_fp16 = einsum(equation = var_3660_equation_0, values = (var_3590_cast_fp16, var_3566_cast_fp16))[name = tensor("op_3660_cast_fp16")]; + tensor var_3661_to_fp16 = const()[name = tensor("op_3661_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_553_cast_fp16 = mul(x = var_3660_cast_fp16, y = var_3661_to_fp16)[name = tensor("aw_chunk_553_cast_fp16")]; + tensor var_3664_equation_0 = const()[name = tensor("op_3664_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3664_cast_fp16 = einsum(equation = var_3664_equation_0, values = (var_3590_cast_fp16, var_3567_cast_fp16))[name = tensor("op_3664_cast_fp16")]; + tensor var_3665_to_fp16 = const()[name = tensor("op_3665_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_555_cast_fp16 = mul(x = var_3664_cast_fp16, y = var_3665_to_fp16)[name = tensor("aw_chunk_555_cast_fp16")]; + tensor var_3668_equation_0 = const()[name = tensor("op_3668_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3668_cast_fp16 = einsum(equation = var_3668_equation_0, values = (var_3594_cast_fp16, var_3568_cast_fp16))[name = tensor("op_3668_cast_fp16")]; + tensor var_3669_to_fp16 = const()[name = tensor("op_3669_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_557_cast_fp16 = mul(x = var_3668_cast_fp16, y = var_3669_to_fp16)[name = tensor("aw_chunk_557_cast_fp16")]; + tensor var_3672_equation_0 = const()[name = tensor("op_3672_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3672_cast_fp16 = einsum(equation = var_3672_equation_0, values = (var_3594_cast_fp16, var_3569_cast_fp16))[name = tensor("op_3672_cast_fp16")]; + tensor var_3673_to_fp16 = const()[name = tensor("op_3673_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_559_cast_fp16 = mul(x = var_3672_cast_fp16, y = var_3673_to_fp16)[name = tensor("aw_chunk_559_cast_fp16")]; + tensor var_3676_equation_0 = const()[name = tensor("op_3676_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3676_cast_fp16 = einsum(equation = var_3676_equation_0, values = (var_3598_cast_fp16, var_3570_cast_fp16))[name = tensor("op_3676_cast_fp16")]; + tensor var_3677_to_fp16 = const()[name = tensor("op_3677_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_561_cast_fp16 = mul(x = var_3676_cast_fp16, y = var_3677_to_fp16)[name = tensor("aw_chunk_561_cast_fp16")]; + tensor var_3680_equation_0 = const()[name = tensor("op_3680_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3680_cast_fp16 = einsum(equation = var_3680_equation_0, values = (var_3598_cast_fp16, var_3571_cast_fp16))[name = tensor("op_3680_cast_fp16")]; + tensor var_3681_to_fp16 = const()[name = tensor("op_3681_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_563_cast_fp16 = mul(x = var_3680_cast_fp16, y = var_3681_to_fp16)[name = tensor("aw_chunk_563_cast_fp16")]; + tensor var_3684_equation_0 = const()[name = tensor("op_3684_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3684_cast_fp16 = einsum(equation = var_3684_equation_0, values = (var_3602_cast_fp16, var_3572_cast_fp16))[name = tensor("op_3684_cast_fp16")]; + tensor var_3685_to_fp16 = const()[name = tensor("op_3685_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_565_cast_fp16 = mul(x = var_3684_cast_fp16, y = var_3685_to_fp16)[name = tensor("aw_chunk_565_cast_fp16")]; + tensor var_3688_equation_0 = const()[name = tensor("op_3688_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3688_cast_fp16 = einsum(equation = var_3688_equation_0, values = (var_3602_cast_fp16, var_3573_cast_fp16))[name = tensor("op_3688_cast_fp16")]; + tensor var_3689_to_fp16 = const()[name = tensor("op_3689_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_567_cast_fp16 = mul(x = var_3688_cast_fp16, y = var_3689_to_fp16)[name = tensor("aw_chunk_567_cast_fp16")]; + tensor var_3692_equation_0 = const()[name = tensor("op_3692_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3692_cast_fp16 = einsum(equation = var_3692_equation_0, values = (var_3606_cast_fp16, var_3574_cast_fp16))[name = tensor("op_3692_cast_fp16")]; + tensor var_3693_to_fp16 = const()[name = tensor("op_3693_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_569_cast_fp16 = mul(x = var_3692_cast_fp16, y = var_3693_to_fp16)[name = tensor("aw_chunk_569_cast_fp16")]; + tensor var_3696_equation_0 = const()[name = tensor("op_3696_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3696_cast_fp16 = einsum(equation = var_3696_equation_0, values = (var_3606_cast_fp16, var_3575_cast_fp16))[name = tensor("op_3696_cast_fp16")]; + tensor var_3697_to_fp16 = const()[name = tensor("op_3697_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_571_cast_fp16 = mul(x = var_3696_cast_fp16, y = var_3697_to_fp16)[name = tensor("aw_chunk_571_cast_fp16")]; + tensor var_3700_equation_0 = const()[name = tensor("op_3700_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3700_cast_fp16 = einsum(equation = var_3700_equation_0, values = (var_3610_cast_fp16, var_3576_cast_fp16))[name = tensor("op_3700_cast_fp16")]; + tensor var_3701_to_fp16 = const()[name = tensor("op_3701_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_573_cast_fp16 = mul(x = var_3700_cast_fp16, y = var_3701_to_fp16)[name = tensor("aw_chunk_573_cast_fp16")]; + tensor var_3704_equation_0 = const()[name = tensor("op_3704_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3704_cast_fp16 = einsum(equation = var_3704_equation_0, values = (var_3610_cast_fp16, var_3577_cast_fp16))[name = tensor("op_3704_cast_fp16")]; + tensor var_3705_to_fp16 = const()[name = tensor("op_3705_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_575_cast_fp16 = mul(x = var_3704_cast_fp16, y = var_3705_to_fp16)[name = tensor("aw_chunk_575_cast_fp16")]; + tensor var_3707_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_545_cast_fp16)[name = tensor("op_3707_cast_fp16")]; + tensor var_3708_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_547_cast_fp16)[name = tensor("op_3708_cast_fp16")]; + tensor var_3709_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_549_cast_fp16)[name = tensor("op_3709_cast_fp16")]; + tensor var_3710_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_551_cast_fp16)[name = tensor("op_3710_cast_fp16")]; + tensor var_3711_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_553_cast_fp16)[name = tensor("op_3711_cast_fp16")]; + tensor var_3712_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_555_cast_fp16)[name = tensor("op_3712_cast_fp16")]; + tensor var_3713_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_557_cast_fp16)[name = tensor("op_3713_cast_fp16")]; + tensor var_3714_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_559_cast_fp16)[name = tensor("op_3714_cast_fp16")]; + tensor var_3715_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_561_cast_fp16)[name = tensor("op_3715_cast_fp16")]; + tensor var_3716_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_563_cast_fp16)[name = tensor("op_3716_cast_fp16")]; + tensor var_3717_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_565_cast_fp16)[name = tensor("op_3717_cast_fp16")]; + tensor var_3718_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_567_cast_fp16)[name = tensor("op_3718_cast_fp16")]; + tensor var_3719_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_569_cast_fp16)[name = tensor("op_3719_cast_fp16")]; + tensor var_3720_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_571_cast_fp16)[name = tensor("op_3720_cast_fp16")]; + tensor var_3721_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_573_cast_fp16)[name = tensor("op_3721_cast_fp16")]; + tensor var_3722_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_575_cast_fp16)[name = tensor("op_3722_cast_fp16")]; + tensor var_3724_equation_0 = const()[name = tensor("op_3724_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3724_cast_fp16 = einsum(equation = var_3724_equation_0, values = (var_3612_cast_fp16, var_3707_cast_fp16))[name = tensor("op_3724_cast_fp16")]; + tensor var_3726_equation_0 = const()[name = tensor("op_3726_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3726_cast_fp16 = einsum(equation = var_3726_equation_0, values = (var_3612_cast_fp16, var_3708_cast_fp16))[name = tensor("op_3726_cast_fp16")]; + tensor var_3728_equation_0 = const()[name = tensor("op_3728_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3728_cast_fp16 = einsum(equation = var_3728_equation_0, values = (var_3616_cast_fp16, var_3709_cast_fp16))[name = tensor("op_3728_cast_fp16")]; + tensor var_3730_equation_0 = const()[name = tensor("op_3730_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3730_cast_fp16 = einsum(equation = var_3730_equation_0, values = (var_3616_cast_fp16, var_3710_cast_fp16))[name = tensor("op_3730_cast_fp16")]; + tensor var_3732_equation_0 = const()[name = tensor("op_3732_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3732_cast_fp16 = einsum(equation = var_3732_equation_0, values = (var_3620_cast_fp16, var_3711_cast_fp16))[name = tensor("op_3732_cast_fp16")]; + tensor var_3734_equation_0 = const()[name = tensor("op_3734_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3734_cast_fp16 = einsum(equation = var_3734_equation_0, values = (var_3620_cast_fp16, var_3712_cast_fp16))[name = tensor("op_3734_cast_fp16")]; + tensor var_3736_equation_0 = const()[name = tensor("op_3736_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3736_cast_fp16 = einsum(equation = var_3736_equation_0, values = (var_3624_cast_fp16, var_3713_cast_fp16))[name = tensor("op_3736_cast_fp16")]; + tensor var_3738_equation_0 = const()[name = tensor("op_3738_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3738_cast_fp16 = einsum(equation = var_3738_equation_0, values = (var_3624_cast_fp16, var_3714_cast_fp16))[name = tensor("op_3738_cast_fp16")]; + tensor var_3740_equation_0 = const()[name = tensor("op_3740_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3740_cast_fp16 = einsum(equation = var_3740_equation_0, values = (var_3628_cast_fp16, var_3715_cast_fp16))[name = tensor("op_3740_cast_fp16")]; + tensor var_3742_equation_0 = const()[name = tensor("op_3742_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3742_cast_fp16 = einsum(equation = var_3742_equation_0, values = (var_3628_cast_fp16, var_3716_cast_fp16))[name = tensor("op_3742_cast_fp16")]; + tensor var_3744_equation_0 = const()[name = tensor("op_3744_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3744_cast_fp16 = einsum(equation = var_3744_equation_0, values = (var_3632_cast_fp16, var_3717_cast_fp16))[name = tensor("op_3744_cast_fp16")]; + tensor var_3746_equation_0 = const()[name = tensor("op_3746_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3746_cast_fp16 = einsum(equation = var_3746_equation_0, values = (var_3632_cast_fp16, var_3718_cast_fp16))[name = tensor("op_3746_cast_fp16")]; + tensor var_3748_equation_0 = const()[name = tensor("op_3748_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3748_cast_fp16 = einsum(equation = var_3748_equation_0, values = (var_3636_cast_fp16, var_3719_cast_fp16))[name = tensor("op_3748_cast_fp16")]; + tensor var_3750_equation_0 = const()[name = tensor("op_3750_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3750_cast_fp16 = einsum(equation = var_3750_equation_0, values = (var_3636_cast_fp16, var_3720_cast_fp16))[name = tensor("op_3750_cast_fp16")]; + tensor var_3752_equation_0 = const()[name = tensor("op_3752_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3752_cast_fp16 = einsum(equation = var_3752_equation_0, values = (var_3640_cast_fp16, var_3721_cast_fp16))[name = tensor("op_3752_cast_fp16")]; + tensor var_3754_equation_0 = const()[name = tensor("op_3754_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3754_cast_fp16 = einsum(equation = var_3754_equation_0, values = (var_3640_cast_fp16, var_3722_cast_fp16))[name = tensor("op_3754_cast_fp16")]; + tensor var_3756_interleave_0 = const()[name = tensor("op_3756_interleave_0"), val = tensor(false)]; + tensor var_3756_cast_fp16 = concat(axis = var_3089, interleave = var_3756_interleave_0, values = (var_3724_cast_fp16, var_3726_cast_fp16))[name = tensor("op_3756_cast_fp16")]; + tensor var_3758_interleave_0 = const()[name = tensor("op_3758_interleave_0"), val = tensor(false)]; + tensor var_3758_cast_fp16 = concat(axis = var_3089, interleave = var_3758_interleave_0, values = (var_3728_cast_fp16, var_3730_cast_fp16))[name = tensor("op_3758_cast_fp16")]; + tensor var_3760_interleave_0 = const()[name = tensor("op_3760_interleave_0"), val = tensor(false)]; + tensor var_3760_cast_fp16 = concat(axis = var_3089, interleave = var_3760_interleave_0, values = (var_3732_cast_fp16, var_3734_cast_fp16))[name = tensor("op_3760_cast_fp16")]; + tensor var_3762_interleave_0 = const()[name = tensor("op_3762_interleave_0"), val = tensor(false)]; + tensor var_3762_cast_fp16 = concat(axis = var_3089, interleave = var_3762_interleave_0, values = (var_3736_cast_fp16, var_3738_cast_fp16))[name = tensor("op_3762_cast_fp16")]; + tensor var_3764_interleave_0 = const()[name = tensor("op_3764_interleave_0"), val = tensor(false)]; + tensor var_3764_cast_fp16 = concat(axis = var_3089, interleave = var_3764_interleave_0, values = (var_3740_cast_fp16, var_3742_cast_fp16))[name = tensor("op_3764_cast_fp16")]; + tensor var_3766_interleave_0 = const()[name = tensor("op_3766_interleave_0"), val = tensor(false)]; + tensor var_3766_cast_fp16 = concat(axis = var_3089, interleave = var_3766_interleave_0, values = (var_3744_cast_fp16, var_3746_cast_fp16))[name = tensor("op_3766_cast_fp16")]; + tensor var_3768_interleave_0 = const()[name = tensor("op_3768_interleave_0"), val = tensor(false)]; + tensor var_3768_cast_fp16 = concat(axis = var_3089, interleave = var_3768_interleave_0, values = (var_3748_cast_fp16, var_3750_cast_fp16))[name = tensor("op_3768_cast_fp16")]; + tensor var_3770_interleave_0 = const()[name = tensor("op_3770_interleave_0"), val = tensor(false)]; + tensor var_3770_cast_fp16 = concat(axis = var_3089, interleave = var_3770_interleave_0, values = (var_3752_cast_fp16, var_3754_cast_fp16))[name = tensor("op_3770_cast_fp16")]; + tensor input_81_interleave_0 = const()[name = tensor("input_81_interleave_0"), val = tensor(false)]; + tensor input_81_cast_fp16 = concat(axis = var_3111, interleave = input_81_interleave_0, values = (var_3756_cast_fp16, var_3758_cast_fp16, var_3760_cast_fp16, var_3762_cast_fp16, var_3764_cast_fp16, var_3766_cast_fp16, var_3768_cast_fp16, var_3770_cast_fp16))[name = tensor("input_81_cast_fp16")]; + tensor var_3780_pad_type_0 = const()[name = tensor("op_3780_pad_type_0"), val = tensor("valid")]; + tensor var_3780_strides_0 = const()[name = tensor("op_3780_strides_0"), val = tensor([1, 1])]; + tensor var_3780_pad_0 = const()[name = tensor("op_3780_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3780_dilations_0 = const()[name = tensor("op_3780_dilations_0"), val = tensor([1, 1])]; + tensor var_3780_groups_0 = const()[name = tensor("op_3780_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_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(45192896)))]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_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(46012160)))]; + tensor var_3780_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_3780_dilations_0, groups = var_3780_groups_0, pad = var_3780_pad_0, pad_type = var_3780_pad_type_0, strides = var_3780_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_81_cast_fp16)[name = tensor("op_3780_cast_fp16")]; + tensor inputs_17_cast_fp16 = add(x = var_3780_cast_fp16, y = inputs_15_cast_fp16)[name = tensor("inputs_17_cast_fp16")]; + tensor input_83_axes_0 = const()[name = tensor("input_83_axes_0"), val = tensor([1])]; + tensor input_83_gamma_0_to_fp16 = const()[name = tensor("input_83_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46013504)))]; + tensor input_83_beta_0_to_fp16 = const()[name = tensor("input_83_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46014848)))]; + tensor var_3790_to_fp16 = const()[name = tensor("op_3790_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_83_cast_fp16 = layer_norm(axes = input_83_axes_0, beta = input_83_beta_0_to_fp16, epsilon = var_3790_to_fp16, gamma = input_83_gamma_0_to_fp16, x = inputs_17_cast_fp16)[name = tensor("input_83_cast_fp16")]; + tensor var_3810_pad_type_0 = const()[name = tensor("op_3810_pad_type_0"), val = tensor("valid")]; + tensor var_3810_strides_0 = const()[name = tensor("op_3810_strides_0"), val = tensor([1, 1])]; + tensor var_3810_pad_0 = const()[name = tensor("op_3810_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3810_dilations_0 = const()[name = tensor("op_3810_dilations_0"), val = tensor([1, 1])]; + tensor var_3810_groups_0 = const()[name = tensor("op_3810_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_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(46016192)))]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_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(52569856)))]; + tensor var_3810_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_3810_dilations_0, groups = var_3810_groups_0, pad = var_3810_pad_0, pad_type = var_3810_pad_type_0, strides = var_3810_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_83_cast_fp16)[name = tensor("op_3810_cast_fp16")]; + tensor var_3811_split_sizes_0 = const()[name = tensor("op_3811_split_sizes_0"), val = tensor([2560, 2560])]; + tensor var_3811_axis_0 = const()[name = tensor("op_3811_axis_0"), val = tensor(1)]; + tensor var_3811_cast_fp16_0, tensor var_3811_cast_fp16_1 = split(axis = var_3811_axis_0, split_sizes = var_3811_split_sizes_0, x = var_3810_cast_fp16)[name = tensor("op_3811_cast_fp16")]; + tensor var_3813_mode_0 = const()[name = tensor("op_3813_mode_0"), val = tensor("EXACT")]; + tensor var_3813_cast_fp16 = gelu(mode = var_3813_mode_0, x = var_3811_cast_fp16_1)[name = tensor("op_3813_cast_fp16")]; + tensor input_85_cast_fp16 = mul(x = var_3811_cast_fp16_0, y = var_3813_cast_fp16)[name = tensor("input_85_cast_fp16")]; + tensor var_3821_pad_type_0 = const()[name = tensor("op_3821_pad_type_0"), val = tensor("valid")]; + tensor var_3821_strides_0 = const()[name = tensor("op_3821_strides_0"), val = tensor([1, 1])]; + tensor var_3821_pad_0 = const()[name = tensor("op_3821_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3821_dilations_0 = const()[name = tensor("op_3821_dilations_0"), val = tensor([1, 1])]; + tensor var_3821_groups_0 = const()[name = tensor("op_3821_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("down_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(52580160)))]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("down_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(55857024)))]; + tensor var_3821_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_3821_dilations_0, groups = var_3821_groups_0, pad = var_3821_pad_0, pad_type = var_3821_pad_type_0, strides = var_3821_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_85_cast_fp16)[name = tensor("op_3821_cast_fp16")]; + tensor hidden_states_51_cast_fp16 = add(x = var_3821_cast_fp16, y = inputs_17_cast_fp16)[name = tensor("hidden_states_51_cast_fp16")]; + tensor var_3823 = const()[name = tensor("op_3823"), val = tensor([2, 640, 32, 32])]; + tensor input_87_cast_fp16 = reshape(shape = var_3823, x = hidden_states_51_cast_fp16)[name = tensor("input_87_cast_fp16")]; + tensor hidden_states_53_pad_type_0 = const()[name = tensor("hidden_states_53_pad_type_0"), val = tensor("valid")]; + tensor hidden_states_53_strides_0 = const()[name = tensor("hidden_states_53_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_53_pad_0 = const()[name = tensor("hidden_states_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_53_dilations_0 = const()[name = tensor("hidden_states_53_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_53_groups_0 = const()[name = tensor("hidden_states_53_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(55858368)))]; + tensor down_blocks_1_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56677632)))]; + tensor hidden_states_53_cast_fp16 = conv(bias = down_blocks_1_attentions_0_proj_out_bias_to_fp16, dilations = hidden_states_53_dilations_0, groups = hidden_states_53_groups_0, pad = hidden_states_53_pad_0, pad_type = hidden_states_53_pad_type_0, strides = hidden_states_53_strides_0, weight = down_blocks_1_attentions_0_proj_out_weight_to_fp16, x = input_87_cast_fp16)[name = tensor("hidden_states_53_cast_fp16")]; + tensor input_89_cast_fp16_1 = add(x = hidden_states_53_cast_fp16, y = hidden_states_41_cast_fp16)[name = tensor("input_89_cast_fp16")]; + tensor reshape_36_shape_0 = const()[name = tensor("reshape_36_shape_0"), val = tensor([2, 32, 20, 32, 32])]; + tensor reshape_36_cast_fp16 = reshape(shape = reshape_36_shape_0, x = input_89_cast_fp16_1)[name = tensor("reshape_36_cast_fp16")]; + tensor reduce_mean_27_axes_0 = const()[name = tensor("reduce_mean_27_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_27_keep_dims_0 = const()[name = tensor("reduce_mean_27_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_27_cast_fp16 = reduce_mean(axes = reduce_mean_27_axes_0, keep_dims = reduce_mean_27_keep_dims_0, x = reshape_36_cast_fp16)[name = tensor("reduce_mean_27_cast_fp16")]; + tensor sub_18_cast_fp16 = sub(x = reshape_36_cast_fp16, y = reduce_mean_27_cast_fp16)[name = tensor("sub_18_cast_fp16")]; + tensor square_9_cast_fp16 = square(x = sub_18_cast_fp16)[name = tensor("square_9_cast_fp16")]; + tensor reduce_mean_29_axes_0 = const()[name = tensor("reduce_mean_29_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_29_keep_dims_0 = const()[name = tensor("reduce_mean_29_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_29_cast_fp16 = reduce_mean(axes = reduce_mean_29_axes_0, keep_dims = reduce_mean_29_keep_dims_0, x = square_9_cast_fp16)[name = tensor("reduce_mean_29_cast_fp16")]; + tensor add_18_y_0_to_fp16 = const()[name = tensor("add_18_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_18_cast_fp16 = add(x = reduce_mean_29_cast_fp16, y = add_18_y_0_to_fp16)[name = tensor("add_18_cast_fp16")]; + tensor sqrt_9_cast_fp16 = sqrt(x = add_18_cast_fp16)[name = tensor("sqrt_9_cast_fp16")]; + tensor real_div_9_cast_fp16 = real_div(x = sub_18_cast_fp16, y = sqrt_9_cast_fp16)[name = tensor("real_div_9_cast_fp16")]; + tensor reshape_37_shape_0 = const()[name = tensor("reshape_37_shape_0"), val = tensor([2, 640, 32, 32])]; + tensor reshape_37_cast_fp16 = reshape(shape = reshape_37_shape_0, x = real_div_9_cast_fp16)[name = tensor("reshape_37_cast_fp16")]; + tensor add_19_gamma_0_to_fp16 = const()[name = tensor("add_19_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56678976)))]; + tensor add_19_beta_0_to_fp16 = const()[name = tensor("add_19_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56680320)))]; + tensor add_19_epsilon_0_to_fp16 = const()[name = tensor("add_19_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_19_cast_fp16 = batch_norm(beta = add_19_beta_0_to_fp16, epsilon = add_19_epsilon_0_to_fp16, gamma = add_19_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_37_cast_fp16)[name = tensor("add_19_cast_fp16")]; + tensor input_93_cast_fp16 = silu(x = add_19_cast_fp16)[name = tensor("input_93_cast_fp16")]; + tensor hidden_states_55_pad_type_0 = const()[name = tensor("hidden_states_55_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_55_pad_0 = const()[name = tensor("hidden_states_55_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_55_strides_0 = const()[name = tensor("hidden_states_55_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_55_dilations_0 = const()[name = tensor("hidden_states_55_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_55_groups_0 = const()[name = tensor("hidden_states_55_groups_0"), val = tensor(1)]; + tensor down_blocks_1_resnets_1_conv1_weight_to_fp16 = const()[name = tensor("down_blocks_1_resnets_1_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56681664)))]; + tensor down_blocks_1_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64054528)))]; + tensor hidden_states_55_cast_fp16 = conv(bias = down_blocks_1_resnets_1_conv1_bias_to_fp16, dilations = hidden_states_55_dilations_0, groups = hidden_states_55_groups_0, pad = hidden_states_55_pad_0, pad_type = hidden_states_55_pad_type_0, strides = hidden_states_55_strides_0, weight = down_blocks_1_resnets_1_conv1_weight_to_fp16, x = input_93_cast_fp16)[name = tensor("hidden_states_55_cast_fp16")]; + tensor temb_7_pad_type_0 = const()[name = tensor("temb_7_pad_type_0"), val = tensor("valid")]; + tensor temb_7_strides_0 = const()[name = tensor("temb_7_strides_0"), val = tensor([1, 1])]; + tensor temb_7_pad_0 = const()[name = tensor("temb_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor temb_7_dilations_0 = const()[name = tensor("temb_7_dilations_0"), val = tensor([1, 1])]; + tensor temb_7_groups_0 = const()[name = tensor("temb_7_groups_0"), val = tensor(1)]; + tensor down_blocks_1_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor("down_blocks_1_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64055872)))]; + tensor down_blocks_1_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65694336)))]; + tensor temb_7_cast_fp16 = conv(bias = down_blocks_1_resnets_1_time_emb_proj_bias_to_fp16, dilations = temb_7_dilations_0, groups = temb_7_groups_0, pad = temb_7_pad_0, pad_type = temb_7_pad_type_0, strides = temb_7_strides_0, weight = down_blocks_1_resnets_1_time_emb_proj_weight_to_fp16, x = input_15_cast_fp16_1)[name = tensor("temb_7_cast_fp16")]; + tensor input_97_cast_fp16 = add(x = hidden_states_55_cast_fp16, y = temb_7_cast_fp16)[name = tensor("input_97_cast_fp16")]; + tensor reshape_40_shape_0 = const()[name = tensor("reshape_40_shape_0"), val = tensor([2, 32, 20, 32, 32])]; + tensor reshape_40_cast_fp16 = reshape(shape = reshape_40_shape_0, x = input_97_cast_fp16)[name = tensor("reshape_40_cast_fp16")]; + tensor reduce_mean_30_axes_0 = const()[name = tensor("reduce_mean_30_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_30_keep_dims_0 = const()[name = tensor("reduce_mean_30_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_30_cast_fp16 = reduce_mean(axes = reduce_mean_30_axes_0, keep_dims = reduce_mean_30_keep_dims_0, x = reshape_40_cast_fp16)[name = tensor("reduce_mean_30_cast_fp16")]; + tensor sub_20_cast_fp16 = sub(x = reshape_40_cast_fp16, y = reduce_mean_30_cast_fp16)[name = tensor("sub_20_cast_fp16")]; + tensor square_10_cast_fp16 = square(x = sub_20_cast_fp16)[name = tensor("square_10_cast_fp16")]; + tensor reduce_mean_32_axes_0 = const()[name = tensor("reduce_mean_32_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_32_keep_dims_0 = const()[name = tensor("reduce_mean_32_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_32_cast_fp16 = reduce_mean(axes = reduce_mean_32_axes_0, keep_dims = reduce_mean_32_keep_dims_0, x = square_10_cast_fp16)[name = tensor("reduce_mean_32_cast_fp16")]; + tensor add_20_y_0_to_fp16 = const()[name = tensor("add_20_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_20_cast_fp16 = add(x = reduce_mean_32_cast_fp16, y = add_20_y_0_to_fp16)[name = tensor("add_20_cast_fp16")]; + tensor sqrt_10_cast_fp16 = sqrt(x = add_20_cast_fp16)[name = tensor("sqrt_10_cast_fp16")]; + tensor real_div_10_cast_fp16 = real_div(x = sub_20_cast_fp16, y = sqrt_10_cast_fp16)[name = tensor("real_div_10_cast_fp16")]; + tensor reshape_41_shape_0 = const()[name = tensor("reshape_41_shape_0"), val = tensor([2, 640, 32, 32])]; + tensor reshape_41_cast_fp16 = reshape(shape = reshape_41_shape_0, x = real_div_10_cast_fp16)[name = tensor("reshape_41_cast_fp16")]; + tensor add_21_gamma_0_to_fp16 = const()[name = tensor("add_21_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65695680)))]; + tensor add_21_beta_0_to_fp16 = const()[name = tensor("add_21_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65697024)))]; + tensor add_21_epsilon_0_to_fp16 = const()[name = tensor("add_21_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_21_cast_fp16 = batch_norm(beta = add_21_beta_0_to_fp16, epsilon = add_21_epsilon_0_to_fp16, gamma = add_21_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_41_cast_fp16)[name = tensor("add_21_cast_fp16")]; + tensor input_101_cast_fp16 = silu(x = add_21_cast_fp16)[name = tensor("input_101_cast_fp16")]; + tensor hidden_states_57_pad_type_0 = const()[name = tensor("hidden_states_57_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_57_pad_0 = const()[name = tensor("hidden_states_57_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_57_strides_0 = const()[name = tensor("hidden_states_57_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_57_dilations_0 = const()[name = tensor("hidden_states_57_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_57_groups_0 = const()[name = tensor("hidden_states_57_groups_0"), val = tensor(1)]; + tensor down_blocks_1_resnets_1_conv2_weight_to_fp16 = const()[name = tensor("down_blocks_1_resnets_1_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65698368)))]; + tensor down_blocks_1_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73071232)))]; + tensor hidden_states_57_cast_fp16 = conv(bias = down_blocks_1_resnets_1_conv2_bias_to_fp16, dilations = hidden_states_57_dilations_0, groups = hidden_states_57_groups_0, pad = hidden_states_57_pad_0, pad_type = hidden_states_57_pad_type_0, strides = hidden_states_57_strides_0, weight = down_blocks_1_resnets_1_conv2_weight_to_fp16, x = input_101_cast_fp16)[name = tensor("hidden_states_57_cast_fp16")]; + tensor hidden_states_59_cast_fp16 = add(x = input_89_cast_fp16_1, y = hidden_states_57_cast_fp16)[name = tensor("hidden_states_59_cast_fp16")]; + tensor reshape_44_shape_0 = const()[name = tensor("reshape_44_shape_0"), val = tensor([2, 32, 20, 32, 32])]; + tensor reshape_44_cast_fp16 = reshape(shape = reshape_44_shape_0, x = hidden_states_59_cast_fp16)[name = tensor("reshape_44_cast_fp16")]; + tensor reduce_mean_33_axes_0 = const()[name = tensor("reduce_mean_33_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_33_keep_dims_0 = const()[name = tensor("reduce_mean_33_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_33_cast_fp16 = reduce_mean(axes = reduce_mean_33_axes_0, keep_dims = reduce_mean_33_keep_dims_0, x = reshape_44_cast_fp16)[name = tensor("reduce_mean_33_cast_fp16")]; + tensor sub_22_cast_fp16 = sub(x = reshape_44_cast_fp16, y = reduce_mean_33_cast_fp16)[name = tensor("sub_22_cast_fp16")]; + tensor square_11_cast_fp16 = square(x = sub_22_cast_fp16)[name = tensor("square_11_cast_fp16")]; + tensor reduce_mean_35_axes_0 = const()[name = tensor("reduce_mean_35_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_35_keep_dims_0 = const()[name = tensor("reduce_mean_35_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_35_cast_fp16 = reduce_mean(axes = reduce_mean_35_axes_0, keep_dims = reduce_mean_35_keep_dims_0, x = square_11_cast_fp16)[name = tensor("reduce_mean_35_cast_fp16")]; + tensor add_22_y_0_to_fp16 = const()[name = tensor("add_22_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_22_cast_fp16 = add(x = reduce_mean_35_cast_fp16, y = add_22_y_0_to_fp16)[name = tensor("add_22_cast_fp16")]; + tensor sqrt_11_cast_fp16 = sqrt(x = add_22_cast_fp16)[name = tensor("sqrt_11_cast_fp16")]; + tensor real_div_11_cast_fp16 = real_div(x = sub_22_cast_fp16, y = sqrt_11_cast_fp16)[name = tensor("real_div_11_cast_fp16")]; + tensor reshape_45_shape_0 = const()[name = tensor("reshape_45_shape_0"), val = tensor([2, 640, 32, 32])]; + tensor reshape_45_cast_fp16 = reshape(shape = reshape_45_shape_0, x = real_div_11_cast_fp16)[name = tensor("reshape_45_cast_fp16")]; + tensor add_23_gamma_0_to_fp16 = const()[name = tensor("add_23_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73072576)))]; + tensor add_23_beta_0_to_fp16 = const()[name = tensor("add_23_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73073920)))]; + tensor add_23_epsilon_0_to_fp16 = const()[name = tensor("add_23_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_23_cast_fp16 = batch_norm(beta = add_23_beta_0_to_fp16, epsilon = add_23_epsilon_0_to_fp16, gamma = add_23_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_45_cast_fp16)[name = tensor("add_23_cast_fp16")]; + tensor hidden_states_61_pad_type_0 = const()[name = tensor("hidden_states_61_pad_type_0"), val = tensor("valid")]; + tensor hidden_states_61_strides_0 = const()[name = tensor("hidden_states_61_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_61_pad_0 = const()[name = tensor("hidden_states_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_61_dilations_0 = const()[name = tensor("hidden_states_61_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_61_groups_0 = const()[name = tensor("hidden_states_61_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_proj_in_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73075264)))]; + tensor down_blocks_1_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73894528)))]; + tensor hidden_states_61_cast_fp16 = conv(bias = down_blocks_1_attentions_1_proj_in_bias_to_fp16, dilations = hidden_states_61_dilations_0, groups = hidden_states_61_groups_0, pad = hidden_states_61_pad_0, pad_type = hidden_states_61_pad_type_0, strides = hidden_states_61_strides_0, weight = down_blocks_1_attentions_1_proj_in_weight_to_fp16, x = add_23_cast_fp16)[name = tensor("hidden_states_61_cast_fp16")]; + tensor var_3893 = const()[name = tensor("op_3893"), val = tensor([2, 640, 1, 1024])]; + tensor inputs_19_cast_fp16 = reshape(shape = var_3893, x = hidden_states_61_cast_fp16)[name = tensor("inputs_19_cast_fp16")]; + tensor hidden_states_63_axes_0 = const()[name = tensor("hidden_states_63_axes_0"), val = tensor([1])]; + tensor hidden_states_63_gamma_0_to_fp16 = const()[name = tensor("hidden_states_63_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73895872)))]; + tensor hidden_states_63_beta_0_to_fp16 = const()[name = tensor("hidden_states_63_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73897216)))]; + tensor var_3909_to_fp16 = const()[name = tensor("op_3909_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_63_cast_fp16 = layer_norm(axes = hidden_states_63_axes_0, beta = hidden_states_63_beta_0_to_fp16, epsilon = var_3909_to_fp16, gamma = hidden_states_63_gamma_0_to_fp16, x = inputs_19_cast_fp16)[name = tensor("hidden_states_63_cast_fp16")]; + tensor q_13_pad_type_0 = const()[name = tensor("q_13_pad_type_0"), val = tensor("valid")]; + tensor q_13_strides_0 = const()[name = tensor("q_13_strides_0"), val = tensor([1, 1])]; + tensor q_13_pad_0 = const()[name = tensor("q_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_13_dilations_0 = const()[name = tensor("q_13_dilations_0"), val = tensor([1, 1])]; + tensor q_13_groups_0 = const()[name = tensor("q_13_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_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(73898560)))]; + tensor q_13_cast_fp16 = conv(dilations = q_13_dilations_0, groups = q_13_groups_0, pad = q_13_pad_0, pad_type = q_13_pad_type_0, strides = q_13_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_63_cast_fp16)[name = tensor("q_13_cast_fp16")]; + tensor k_25_pad_type_0 = const()[name = tensor("k_25_pad_type_0"), val = tensor("valid")]; + tensor k_25_strides_0 = const()[name = tensor("k_25_strides_0"), val = tensor([1, 1])]; + tensor k_25_pad_0 = const()[name = tensor("k_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_25_dilations_0 = const()[name = tensor("k_25_dilations_0"), val = tensor([1, 1])]; + tensor k_25_groups_0 = const()[name = tensor("k_25_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_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(74717824)))]; + tensor k_25_cast_fp16 = conv(dilations = k_25_dilations_0, groups = k_25_groups_0, pad = k_25_pad_0, pad_type = k_25_pad_type_0, strides = k_25_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_63_cast_fp16)[name = tensor("k_25_cast_fp16")]; + tensor v_13_pad_type_0 = const()[name = tensor("v_13_pad_type_0"), val = tensor("valid")]; + tensor v_13_strides_0 = const()[name = tensor("v_13_strides_0"), val = tensor([1, 1])]; + tensor v_13_pad_0 = const()[name = tensor("v_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_13_dilations_0 = const()[name = tensor("v_13_dilations_0"), val = tensor([1, 1])]; + tensor v_13_groups_0 = const()[name = tensor("v_13_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_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(75537088)))]; + tensor v_13_cast_fp16 = conv(dilations = v_13_dilations_0, groups = v_13_groups_0, pad = v_13_pad_0, pad_type = v_13_pad_type_0, strides = v_13_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_63_cast_fp16)[name = tensor("v_13_cast_fp16")]; + tensor var_3942_begin_0 = const()[name = tensor("op_3942_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3942_end_0 = const()[name = tensor("op_3942_end_0"), val = tensor([2, 80, 1, 1024])]; + tensor var_3942_end_mask_0 = const()[name = tensor("op_3942_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3942_cast_fp16 = slice_by_index(begin = var_3942_begin_0, end = var_3942_end_0, end_mask = var_3942_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_3942_cast_fp16")]; + tensor var_3946_begin_0 = const()[name = tensor("op_3946_begin_0"), val = tensor([0, 80, 0, 0])]; + tensor var_3946_end_0 = const()[name = tensor("op_3946_end_0"), val = tensor([2, 160, 1, 1024])]; + tensor var_3946_end_mask_0 = const()[name = tensor("op_3946_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3946_cast_fp16 = slice_by_index(begin = var_3946_begin_0, end = var_3946_end_0, end_mask = var_3946_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_3946_cast_fp16")]; + tensor var_3950_begin_0 = const()[name = tensor("op_3950_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_3950_end_0 = const()[name = tensor("op_3950_end_0"), val = tensor([2, 240, 1, 1024])]; + tensor var_3950_end_mask_0 = const()[name = tensor("op_3950_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3950_cast_fp16 = slice_by_index(begin = var_3950_begin_0, end = var_3950_end_0, end_mask = var_3950_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_3950_cast_fp16")]; + tensor var_3954_begin_0 = const()[name = tensor("op_3954_begin_0"), val = tensor([0, 240, 0, 0])]; + tensor var_3954_end_0 = const()[name = tensor("op_3954_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_3954_end_mask_0 = const()[name = tensor("op_3954_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3954_cast_fp16 = slice_by_index(begin = var_3954_begin_0, end = var_3954_end_0, end_mask = var_3954_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_3954_cast_fp16")]; + tensor var_3958_begin_0 = const()[name = tensor("op_3958_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3958_end_0 = const()[name = tensor("op_3958_end_0"), val = tensor([2, 400, 1, 1024])]; + tensor var_3958_end_mask_0 = const()[name = tensor("op_3958_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3958_cast_fp16 = slice_by_index(begin = var_3958_begin_0, end = var_3958_end_0, end_mask = var_3958_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_3958_cast_fp16")]; + tensor var_3962_begin_0 = const()[name = tensor("op_3962_begin_0"), val = tensor([0, 400, 0, 0])]; + tensor var_3962_end_0 = const()[name = tensor("op_3962_end_0"), val = tensor([2, 480, 1, 1024])]; + tensor var_3962_end_mask_0 = const()[name = tensor("op_3962_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3962_cast_fp16 = slice_by_index(begin = var_3962_begin_0, end = var_3962_end_0, end_mask = var_3962_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_3962_cast_fp16")]; + tensor var_3966_begin_0 = const()[name = tensor("op_3966_begin_0"), val = tensor([0, 480, 0, 0])]; + tensor var_3966_end_0 = const()[name = tensor("op_3966_end_0"), val = tensor([2, 560, 1, 1024])]; + tensor var_3966_end_mask_0 = const()[name = tensor("op_3966_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3966_cast_fp16 = slice_by_index(begin = var_3966_begin_0, end = var_3966_end_0, end_mask = var_3966_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_3966_cast_fp16")]; + tensor var_3970_begin_0 = const()[name = tensor("op_3970_begin_0"), val = tensor([0, 560, 0, 0])]; + tensor var_3970_end_0 = const()[name = tensor("op_3970_end_0"), val = tensor([2, 1, 1, 1024])]; + tensor var_3970_end_mask_0 = const()[name = tensor("op_3970_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3970_cast_fp16 = slice_by_index(begin = var_3970_begin_0, end = var_3970_end_0, end_mask = var_3970_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_3970_cast_fp16")]; + tensor var_3973_begin_0 = const()[name = tensor("op_3973_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3973_end_0 = const()[name = tensor("op_3973_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_3973_end_mask_0 = const()[name = tensor("op_3973_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3973_cast_fp16 = slice_by_index(begin = var_3973_begin_0, end = var_3973_end_0, end_mask = var_3973_end_mask_0, x = var_3942_cast_fp16)[name = tensor("op_3973_cast_fp16")]; + tensor var_3974_begin_0 = const()[name = tensor("op_3974_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3974_end_0 = const()[name = tensor("op_3974_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_3974_end_mask_0 = const()[name = tensor("op_3974_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3974_cast_fp16 = slice_by_index(begin = var_3974_begin_0, end = var_3974_end_0, end_mask = var_3974_end_mask_0, x = var_3942_cast_fp16)[name = tensor("op_3974_cast_fp16")]; + tensor var_3975_begin_0 = const()[name = tensor("op_3975_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3975_end_0 = const()[name = tensor("op_3975_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_3975_end_mask_0 = const()[name = tensor("op_3975_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3975_cast_fp16 = slice_by_index(begin = var_3975_begin_0, end = var_3975_end_0, end_mask = var_3975_end_mask_0, x = var_3946_cast_fp16)[name = tensor("op_3975_cast_fp16")]; + tensor var_3976_begin_0 = const()[name = tensor("op_3976_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3976_end_0 = const()[name = tensor("op_3976_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_3976_end_mask_0 = const()[name = tensor("op_3976_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3976_cast_fp16 = slice_by_index(begin = var_3976_begin_0, end = var_3976_end_0, end_mask = var_3976_end_mask_0, x = var_3946_cast_fp16)[name = tensor("op_3976_cast_fp16")]; + tensor var_3977_begin_0 = const()[name = tensor("op_3977_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3977_end_0 = const()[name = tensor("op_3977_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_3977_end_mask_0 = const()[name = tensor("op_3977_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3977_cast_fp16 = slice_by_index(begin = var_3977_begin_0, end = var_3977_end_0, end_mask = var_3977_end_mask_0, x = var_3950_cast_fp16)[name = tensor("op_3977_cast_fp16")]; + tensor var_3978_begin_0 = const()[name = tensor("op_3978_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3978_end_0 = const()[name = tensor("op_3978_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_3978_end_mask_0 = const()[name = tensor("op_3978_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3978_cast_fp16 = slice_by_index(begin = var_3978_begin_0, end = var_3978_end_0, end_mask = var_3978_end_mask_0, x = var_3950_cast_fp16)[name = tensor("op_3978_cast_fp16")]; + tensor var_3979_begin_0 = const()[name = tensor("op_3979_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3979_end_0 = const()[name = tensor("op_3979_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_3979_end_mask_0 = const()[name = tensor("op_3979_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3979_cast_fp16 = slice_by_index(begin = var_3979_begin_0, end = var_3979_end_0, end_mask = var_3979_end_mask_0, x = var_3954_cast_fp16)[name = tensor("op_3979_cast_fp16")]; + tensor var_3980_begin_0 = const()[name = tensor("op_3980_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3980_end_0 = const()[name = tensor("op_3980_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_3980_end_mask_0 = const()[name = tensor("op_3980_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3980_cast_fp16 = slice_by_index(begin = var_3980_begin_0, end = var_3980_end_0, end_mask = var_3980_end_mask_0, x = var_3954_cast_fp16)[name = tensor("op_3980_cast_fp16")]; + tensor var_3981_begin_0 = const()[name = tensor("op_3981_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3981_end_0 = const()[name = tensor("op_3981_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_3981_end_mask_0 = const()[name = tensor("op_3981_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3981_cast_fp16 = slice_by_index(begin = var_3981_begin_0, end = var_3981_end_0, end_mask = var_3981_end_mask_0, x = var_3958_cast_fp16)[name = tensor("op_3981_cast_fp16")]; + tensor var_3982_begin_0 = const()[name = tensor("op_3982_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3982_end_0 = const()[name = tensor("op_3982_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_3982_end_mask_0 = const()[name = tensor("op_3982_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3982_cast_fp16 = slice_by_index(begin = var_3982_begin_0, end = var_3982_end_0, end_mask = var_3982_end_mask_0, x = var_3958_cast_fp16)[name = tensor("op_3982_cast_fp16")]; + tensor var_3983_begin_0 = const()[name = tensor("op_3983_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3983_end_0 = const()[name = tensor("op_3983_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_3983_end_mask_0 = const()[name = tensor("op_3983_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3983_cast_fp16 = slice_by_index(begin = var_3983_begin_0, end = var_3983_end_0, end_mask = var_3983_end_mask_0, x = var_3962_cast_fp16)[name = tensor("op_3983_cast_fp16")]; + tensor var_3984_begin_0 = const()[name = tensor("op_3984_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3984_end_0 = const()[name = tensor("op_3984_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_3984_end_mask_0 = const()[name = tensor("op_3984_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3984_cast_fp16 = slice_by_index(begin = var_3984_begin_0, end = var_3984_end_0, end_mask = var_3984_end_mask_0, x = var_3962_cast_fp16)[name = tensor("op_3984_cast_fp16")]; + tensor var_3985_begin_0 = const()[name = tensor("op_3985_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3985_end_0 = const()[name = tensor("op_3985_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_3985_end_mask_0 = const()[name = tensor("op_3985_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3985_cast_fp16 = slice_by_index(begin = var_3985_begin_0, end = var_3985_end_0, end_mask = var_3985_end_mask_0, x = var_3966_cast_fp16)[name = tensor("op_3985_cast_fp16")]; + tensor var_3986_begin_0 = const()[name = tensor("op_3986_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3986_end_0 = const()[name = tensor("op_3986_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_3986_end_mask_0 = const()[name = tensor("op_3986_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3986_cast_fp16 = slice_by_index(begin = var_3986_begin_0, end = var_3986_end_0, end_mask = var_3986_end_mask_0, x = var_3966_cast_fp16)[name = tensor("op_3986_cast_fp16")]; + tensor var_3987_begin_0 = const()[name = tensor("op_3987_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3987_end_0 = const()[name = tensor("op_3987_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_3987_end_mask_0 = const()[name = tensor("op_3987_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3987_cast_fp16 = slice_by_index(begin = var_3987_begin_0, end = var_3987_end_0, end_mask = var_3987_end_mask_0, x = var_3970_cast_fp16)[name = tensor("op_3987_cast_fp16")]; + tensor var_3988_begin_0 = const()[name = tensor("op_3988_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3988_end_0 = const()[name = tensor("op_3988_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_3988_end_mask_0 = const()[name = tensor("op_3988_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3988_cast_fp16 = slice_by_index(begin = var_3988_begin_0, end = var_3988_end_0, end_mask = var_3988_end_mask_0, x = var_3970_cast_fp16)[name = tensor("op_3988_cast_fp16")]; + tensor k_27_perm_0 = const()[name = tensor("k_27_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_3993_begin_0 = const()[name = tensor("op_3993_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3993_end_0 = const()[name = tensor("op_3993_end_0"), val = tensor([2, 1024, 1, 80])]; + tensor var_3993_end_mask_0 = const()[name = tensor("op_3993_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_27_cast_fp16 = transpose(perm = k_27_perm_0, x = k_25_cast_fp16)[name = tensor("transpose_7")]; + tensor var_3993_cast_fp16 = slice_by_index(begin = var_3993_begin_0, end = var_3993_end_0, end_mask = var_3993_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_3993_cast_fp16")]; + tensor var_3997_begin_0 = const()[name = tensor("op_3997_begin_0"), val = tensor([0, 0, 0, 80])]; + tensor var_3997_end_0 = const()[name = tensor("op_3997_end_0"), val = tensor([2, 1024, 1, 160])]; + tensor var_3997_end_mask_0 = const()[name = tensor("op_3997_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3997_cast_fp16 = slice_by_index(begin = var_3997_begin_0, end = var_3997_end_0, end_mask = var_3997_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_3997_cast_fp16")]; + tensor var_4001_begin_0 = const()[name = tensor("op_4001_begin_0"), val = tensor([0, 0, 0, 160])]; + tensor var_4001_end_0 = const()[name = tensor("op_4001_end_0"), val = tensor([2, 1024, 1, 240])]; + tensor var_4001_end_mask_0 = const()[name = tensor("op_4001_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4001_cast_fp16 = slice_by_index(begin = var_4001_begin_0, end = var_4001_end_0, end_mask = var_4001_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_4001_cast_fp16")]; + tensor var_4005_begin_0 = const()[name = tensor("op_4005_begin_0"), val = tensor([0, 0, 0, 240])]; + tensor var_4005_end_0 = const()[name = tensor("op_4005_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_4005_end_mask_0 = const()[name = tensor("op_4005_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4005_cast_fp16 = slice_by_index(begin = var_4005_begin_0, end = var_4005_end_0, end_mask = var_4005_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_4005_cast_fp16")]; + tensor var_4009_begin_0 = const()[name = tensor("op_4009_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_4009_end_0 = const()[name = tensor("op_4009_end_0"), val = tensor([2, 1024, 1, 400])]; + tensor var_4009_end_mask_0 = const()[name = tensor("op_4009_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4009_cast_fp16 = slice_by_index(begin = var_4009_begin_0, end = var_4009_end_0, end_mask = var_4009_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_4009_cast_fp16")]; + tensor var_4013_begin_0 = const()[name = tensor("op_4013_begin_0"), val = tensor([0, 0, 0, 400])]; + tensor var_4013_end_0 = const()[name = tensor("op_4013_end_0"), val = tensor([2, 1024, 1, 480])]; + tensor var_4013_end_mask_0 = const()[name = tensor("op_4013_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4013_cast_fp16 = slice_by_index(begin = var_4013_begin_0, end = var_4013_end_0, end_mask = var_4013_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_4013_cast_fp16")]; + tensor var_4017_begin_0 = const()[name = tensor("op_4017_begin_0"), val = tensor([0, 0, 0, 480])]; + tensor var_4017_end_0 = const()[name = tensor("op_4017_end_0"), val = tensor([2, 1024, 1, 560])]; + tensor var_4017_end_mask_0 = const()[name = tensor("op_4017_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4017_cast_fp16 = slice_by_index(begin = var_4017_begin_0, end = var_4017_end_0, end_mask = var_4017_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_4017_cast_fp16")]; + tensor var_4021_begin_0 = const()[name = tensor("op_4021_begin_0"), val = tensor([0, 0, 0, 560])]; + tensor var_4021_end_0 = const()[name = tensor("op_4021_end_0"), val = tensor([2, 1024, 1, 1])]; + tensor var_4021_end_mask_0 = const()[name = tensor("op_4021_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4021_cast_fp16 = slice_by_index(begin = var_4021_begin_0, end = var_4021_end_0, end_mask = var_4021_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_4021_cast_fp16")]; + tensor var_4023_begin_0 = const()[name = tensor("op_4023_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4023_end_0 = const()[name = tensor("op_4023_end_0"), val = tensor([2, 80, 1, 1024])]; + tensor var_4023_end_mask_0 = const()[name = tensor("op_4023_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4023_cast_fp16 = slice_by_index(begin = var_4023_begin_0, end = var_4023_end_0, end_mask = var_4023_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_4023_cast_fp16")]; + tensor var_4027_begin_0 = const()[name = tensor("op_4027_begin_0"), val = tensor([0, 80, 0, 0])]; + tensor var_4027_end_0 = const()[name = tensor("op_4027_end_0"), val = tensor([2, 160, 1, 1024])]; + tensor var_4027_end_mask_0 = const()[name = tensor("op_4027_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4027_cast_fp16 = slice_by_index(begin = var_4027_begin_0, end = var_4027_end_0, end_mask = var_4027_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_4027_cast_fp16")]; + tensor var_4031_begin_0 = const()[name = tensor("op_4031_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_4031_end_0 = const()[name = tensor("op_4031_end_0"), val = tensor([2, 240, 1, 1024])]; + tensor var_4031_end_mask_0 = const()[name = tensor("op_4031_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4031_cast_fp16 = slice_by_index(begin = var_4031_begin_0, end = var_4031_end_0, end_mask = var_4031_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_4031_cast_fp16")]; + tensor var_4035_begin_0 = const()[name = tensor("op_4035_begin_0"), val = tensor([0, 240, 0, 0])]; + tensor var_4035_end_0 = const()[name = tensor("op_4035_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_4035_end_mask_0 = const()[name = tensor("op_4035_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4035_cast_fp16 = slice_by_index(begin = var_4035_begin_0, end = var_4035_end_0, end_mask = var_4035_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_4035_cast_fp16")]; + tensor var_4039_begin_0 = const()[name = tensor("op_4039_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4039_end_0 = const()[name = tensor("op_4039_end_0"), val = tensor([2, 400, 1, 1024])]; + tensor var_4039_end_mask_0 = const()[name = tensor("op_4039_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4039_cast_fp16 = slice_by_index(begin = var_4039_begin_0, end = var_4039_end_0, end_mask = var_4039_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_4039_cast_fp16")]; + tensor var_4043_begin_0 = const()[name = tensor("op_4043_begin_0"), val = tensor([0, 400, 0, 0])]; + tensor var_4043_end_0 = const()[name = tensor("op_4043_end_0"), val = tensor([2, 480, 1, 1024])]; + tensor var_4043_end_mask_0 = const()[name = tensor("op_4043_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4043_cast_fp16 = slice_by_index(begin = var_4043_begin_0, end = var_4043_end_0, end_mask = var_4043_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_4043_cast_fp16")]; + tensor var_4047_begin_0 = const()[name = tensor("op_4047_begin_0"), val = tensor([0, 480, 0, 0])]; + tensor var_4047_end_0 = const()[name = tensor("op_4047_end_0"), val = tensor([2, 560, 1, 1024])]; + tensor var_4047_end_mask_0 = const()[name = tensor("op_4047_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4047_cast_fp16 = slice_by_index(begin = var_4047_begin_0, end = var_4047_end_0, end_mask = var_4047_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_4047_cast_fp16")]; + tensor var_4051_begin_0 = const()[name = tensor("op_4051_begin_0"), val = tensor([0, 560, 0, 0])]; + tensor var_4051_end_0 = const()[name = tensor("op_4051_end_0"), val = tensor([2, 1, 1, 1024])]; + tensor var_4051_end_mask_0 = const()[name = tensor("op_4051_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4051_cast_fp16 = slice_by_index(begin = var_4051_begin_0, end = var_4051_end_0, end_mask = var_4051_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_4051_cast_fp16")]; + tensor var_4055_equation_0 = const()[name = tensor("op_4055_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4055_cast_fp16 = einsum(equation = var_4055_equation_0, values = (var_3993_cast_fp16, var_3973_cast_fp16))[name = tensor("op_4055_cast_fp16")]; + tensor var_4056_to_fp16 = const()[name = tensor("op_4056_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_577_cast_fp16 = mul(x = var_4055_cast_fp16, y = var_4056_to_fp16)[name = tensor("aw_chunk_577_cast_fp16")]; + tensor var_4059_equation_0 = const()[name = tensor("op_4059_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4059_cast_fp16 = einsum(equation = var_4059_equation_0, values = (var_3993_cast_fp16, var_3974_cast_fp16))[name = tensor("op_4059_cast_fp16")]; + tensor var_4060_to_fp16 = const()[name = tensor("op_4060_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_579_cast_fp16 = mul(x = var_4059_cast_fp16, y = var_4060_to_fp16)[name = tensor("aw_chunk_579_cast_fp16")]; + tensor var_4063_equation_0 = const()[name = tensor("op_4063_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4063_cast_fp16 = einsum(equation = var_4063_equation_0, values = (var_3997_cast_fp16, var_3975_cast_fp16))[name = tensor("op_4063_cast_fp16")]; + tensor var_4064_to_fp16 = const()[name = tensor("op_4064_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_581_cast_fp16 = mul(x = var_4063_cast_fp16, y = var_4064_to_fp16)[name = tensor("aw_chunk_581_cast_fp16")]; + tensor var_4067_equation_0 = const()[name = tensor("op_4067_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4067_cast_fp16 = einsum(equation = var_4067_equation_0, values = (var_3997_cast_fp16, var_3976_cast_fp16))[name = tensor("op_4067_cast_fp16")]; + tensor var_4068_to_fp16 = const()[name = tensor("op_4068_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_583_cast_fp16 = mul(x = var_4067_cast_fp16, y = var_4068_to_fp16)[name = tensor("aw_chunk_583_cast_fp16")]; + tensor var_4071_equation_0 = const()[name = tensor("op_4071_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4071_cast_fp16 = einsum(equation = var_4071_equation_0, values = (var_4001_cast_fp16, var_3977_cast_fp16))[name = tensor("op_4071_cast_fp16")]; + tensor var_4072_to_fp16 = const()[name = tensor("op_4072_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_585_cast_fp16 = mul(x = var_4071_cast_fp16, y = var_4072_to_fp16)[name = tensor("aw_chunk_585_cast_fp16")]; + tensor var_4075_equation_0 = const()[name = tensor("op_4075_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4075_cast_fp16 = einsum(equation = var_4075_equation_0, values = (var_4001_cast_fp16, var_3978_cast_fp16))[name = tensor("op_4075_cast_fp16")]; + tensor var_4076_to_fp16 = const()[name = tensor("op_4076_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_587_cast_fp16 = mul(x = var_4075_cast_fp16, y = var_4076_to_fp16)[name = tensor("aw_chunk_587_cast_fp16")]; + tensor var_4079_equation_0 = const()[name = tensor("op_4079_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4079_cast_fp16 = einsum(equation = var_4079_equation_0, values = (var_4005_cast_fp16, var_3979_cast_fp16))[name = tensor("op_4079_cast_fp16")]; + tensor var_4080_to_fp16 = const()[name = tensor("op_4080_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_589_cast_fp16 = mul(x = var_4079_cast_fp16, y = var_4080_to_fp16)[name = tensor("aw_chunk_589_cast_fp16")]; + tensor var_4083_equation_0 = const()[name = tensor("op_4083_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4083_cast_fp16 = einsum(equation = var_4083_equation_0, values = (var_4005_cast_fp16, var_3980_cast_fp16))[name = tensor("op_4083_cast_fp16")]; + tensor var_4084_to_fp16 = const()[name = tensor("op_4084_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_591_cast_fp16 = mul(x = var_4083_cast_fp16, y = var_4084_to_fp16)[name = tensor("aw_chunk_591_cast_fp16")]; + tensor var_4087_equation_0 = const()[name = tensor("op_4087_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4087_cast_fp16 = einsum(equation = var_4087_equation_0, values = (var_4009_cast_fp16, var_3981_cast_fp16))[name = tensor("op_4087_cast_fp16")]; + tensor var_4088_to_fp16 = const()[name = tensor("op_4088_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_593_cast_fp16 = mul(x = var_4087_cast_fp16, y = var_4088_to_fp16)[name = tensor("aw_chunk_593_cast_fp16")]; + tensor var_4091_equation_0 = const()[name = tensor("op_4091_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4091_cast_fp16 = einsum(equation = var_4091_equation_0, values = (var_4009_cast_fp16, var_3982_cast_fp16))[name = tensor("op_4091_cast_fp16")]; + tensor var_4092_to_fp16 = const()[name = tensor("op_4092_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_595_cast_fp16 = mul(x = var_4091_cast_fp16, y = var_4092_to_fp16)[name = tensor("aw_chunk_595_cast_fp16")]; + tensor var_4095_equation_0 = const()[name = tensor("op_4095_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4095_cast_fp16 = einsum(equation = var_4095_equation_0, values = (var_4013_cast_fp16, var_3983_cast_fp16))[name = tensor("op_4095_cast_fp16")]; + tensor var_4096_to_fp16 = const()[name = tensor("op_4096_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_597_cast_fp16 = mul(x = var_4095_cast_fp16, y = var_4096_to_fp16)[name = tensor("aw_chunk_597_cast_fp16")]; + tensor var_4099_equation_0 = const()[name = tensor("op_4099_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4099_cast_fp16 = einsum(equation = var_4099_equation_0, values = (var_4013_cast_fp16, var_3984_cast_fp16))[name = tensor("op_4099_cast_fp16")]; + tensor var_4100_to_fp16 = const()[name = tensor("op_4100_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_599_cast_fp16 = mul(x = var_4099_cast_fp16, y = var_4100_to_fp16)[name = tensor("aw_chunk_599_cast_fp16")]; + tensor var_4103_equation_0 = const()[name = tensor("op_4103_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4103_cast_fp16 = einsum(equation = var_4103_equation_0, values = (var_4017_cast_fp16, var_3985_cast_fp16))[name = tensor("op_4103_cast_fp16")]; + tensor var_4104_to_fp16 = const()[name = tensor("op_4104_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_601_cast_fp16 = mul(x = var_4103_cast_fp16, y = var_4104_to_fp16)[name = tensor("aw_chunk_601_cast_fp16")]; + tensor var_4107_equation_0 = const()[name = tensor("op_4107_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4107_cast_fp16 = einsum(equation = var_4107_equation_0, values = (var_4017_cast_fp16, var_3986_cast_fp16))[name = tensor("op_4107_cast_fp16")]; + tensor var_4108_to_fp16 = const()[name = tensor("op_4108_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_603_cast_fp16 = mul(x = var_4107_cast_fp16, y = var_4108_to_fp16)[name = tensor("aw_chunk_603_cast_fp16")]; + tensor var_4111_equation_0 = const()[name = tensor("op_4111_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4111_cast_fp16 = einsum(equation = var_4111_equation_0, values = (var_4021_cast_fp16, var_3987_cast_fp16))[name = tensor("op_4111_cast_fp16")]; + tensor var_4112_to_fp16 = const()[name = tensor("op_4112_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_605_cast_fp16 = mul(x = var_4111_cast_fp16, y = var_4112_to_fp16)[name = tensor("aw_chunk_605_cast_fp16")]; + tensor var_4115_equation_0 = const()[name = tensor("op_4115_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4115_cast_fp16 = einsum(equation = var_4115_equation_0, values = (var_4021_cast_fp16, var_3988_cast_fp16))[name = tensor("op_4115_cast_fp16")]; + tensor var_4116_to_fp16 = const()[name = tensor("op_4116_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_607_cast_fp16 = mul(x = var_4115_cast_fp16, y = var_4116_to_fp16)[name = tensor("aw_chunk_607_cast_fp16")]; + tensor var_4118_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_577_cast_fp16)[name = tensor("op_4118_cast_fp16")]; + tensor var_4119_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_579_cast_fp16)[name = tensor("op_4119_cast_fp16")]; + tensor var_4120_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_581_cast_fp16)[name = tensor("op_4120_cast_fp16")]; + tensor var_4121_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_583_cast_fp16)[name = tensor("op_4121_cast_fp16")]; + tensor var_4122_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_585_cast_fp16)[name = tensor("op_4122_cast_fp16")]; + tensor var_4123_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_587_cast_fp16)[name = tensor("op_4123_cast_fp16")]; + tensor var_4124_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_589_cast_fp16)[name = tensor("op_4124_cast_fp16")]; + tensor var_4125_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_591_cast_fp16)[name = tensor("op_4125_cast_fp16")]; + tensor var_4126_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_593_cast_fp16)[name = tensor("op_4126_cast_fp16")]; + tensor var_4127_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_595_cast_fp16)[name = tensor("op_4127_cast_fp16")]; + tensor var_4128_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_597_cast_fp16)[name = tensor("op_4128_cast_fp16")]; + tensor var_4129_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_599_cast_fp16)[name = tensor("op_4129_cast_fp16")]; + tensor var_4130_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_601_cast_fp16)[name = tensor("op_4130_cast_fp16")]; + tensor var_4131_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_603_cast_fp16)[name = tensor("op_4131_cast_fp16")]; + tensor var_4132_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_605_cast_fp16)[name = tensor("op_4132_cast_fp16")]; + tensor var_4133_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_607_cast_fp16)[name = tensor("op_4133_cast_fp16")]; + tensor var_4135_equation_0 = const()[name = tensor("op_4135_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4135_cast_fp16 = einsum(equation = var_4135_equation_0, values = (var_4023_cast_fp16, var_4118_cast_fp16))[name = tensor("op_4135_cast_fp16")]; + tensor var_4137_equation_0 = const()[name = tensor("op_4137_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4137_cast_fp16 = einsum(equation = var_4137_equation_0, values = (var_4023_cast_fp16, var_4119_cast_fp16))[name = tensor("op_4137_cast_fp16")]; + tensor var_4139_equation_0 = const()[name = tensor("op_4139_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4139_cast_fp16 = einsum(equation = var_4139_equation_0, values = (var_4027_cast_fp16, var_4120_cast_fp16))[name = tensor("op_4139_cast_fp16")]; + tensor var_4141_equation_0 = const()[name = tensor("op_4141_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4141_cast_fp16 = einsum(equation = var_4141_equation_0, values = (var_4027_cast_fp16, var_4121_cast_fp16))[name = tensor("op_4141_cast_fp16")]; + tensor var_4143_equation_0 = const()[name = tensor("op_4143_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4143_cast_fp16 = einsum(equation = var_4143_equation_0, values = (var_4031_cast_fp16, var_4122_cast_fp16))[name = tensor("op_4143_cast_fp16")]; + tensor var_4145_equation_0 = const()[name = tensor("op_4145_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4145_cast_fp16 = einsum(equation = var_4145_equation_0, values = (var_4031_cast_fp16, var_4123_cast_fp16))[name = tensor("op_4145_cast_fp16")]; + tensor var_4147_equation_0 = const()[name = tensor("op_4147_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4147_cast_fp16 = einsum(equation = var_4147_equation_0, values = (var_4035_cast_fp16, var_4124_cast_fp16))[name = tensor("op_4147_cast_fp16")]; + tensor var_4149_equation_0 = const()[name = tensor("op_4149_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4149_cast_fp16 = einsum(equation = var_4149_equation_0, values = (var_4035_cast_fp16, var_4125_cast_fp16))[name = tensor("op_4149_cast_fp16")]; + tensor var_4151_equation_0 = const()[name = tensor("op_4151_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4151_cast_fp16 = einsum(equation = var_4151_equation_0, values = (var_4039_cast_fp16, var_4126_cast_fp16))[name = tensor("op_4151_cast_fp16")]; + tensor var_4153_equation_0 = const()[name = tensor("op_4153_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4153_cast_fp16 = einsum(equation = var_4153_equation_0, values = (var_4039_cast_fp16, var_4127_cast_fp16))[name = tensor("op_4153_cast_fp16")]; + tensor var_4155_equation_0 = const()[name = tensor("op_4155_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4155_cast_fp16 = einsum(equation = var_4155_equation_0, values = (var_4043_cast_fp16, var_4128_cast_fp16))[name = tensor("op_4155_cast_fp16")]; + tensor var_4157_equation_0 = const()[name = tensor("op_4157_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4157_cast_fp16 = einsum(equation = var_4157_equation_0, values = (var_4043_cast_fp16, var_4129_cast_fp16))[name = tensor("op_4157_cast_fp16")]; + tensor var_4159_equation_0 = const()[name = tensor("op_4159_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4159_cast_fp16 = einsum(equation = var_4159_equation_0, values = (var_4047_cast_fp16, var_4130_cast_fp16))[name = tensor("op_4159_cast_fp16")]; + tensor var_4161_equation_0 = const()[name = tensor("op_4161_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4161_cast_fp16 = einsum(equation = var_4161_equation_0, values = (var_4047_cast_fp16, var_4131_cast_fp16))[name = tensor("op_4161_cast_fp16")]; + tensor var_4163_equation_0 = const()[name = tensor("op_4163_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4163_cast_fp16 = einsum(equation = var_4163_equation_0, values = (var_4051_cast_fp16, var_4132_cast_fp16))[name = tensor("op_4163_cast_fp16")]; + tensor var_4165_equation_0 = const()[name = tensor("op_4165_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4165_cast_fp16 = einsum(equation = var_4165_equation_0, values = (var_4051_cast_fp16, var_4133_cast_fp16))[name = tensor("op_4165_cast_fp16")]; + tensor var_4167_interleave_0 = const()[name = tensor("op_4167_interleave_0"), val = tensor(false)]; + tensor var_4167_cast_fp16 = concat(axis = var_3089, interleave = var_4167_interleave_0, values = (var_4135_cast_fp16, var_4137_cast_fp16))[name = tensor("op_4167_cast_fp16")]; + tensor var_4169_interleave_0 = const()[name = tensor("op_4169_interleave_0"), val = tensor(false)]; + tensor var_4169_cast_fp16 = concat(axis = var_3089, interleave = var_4169_interleave_0, values = (var_4139_cast_fp16, var_4141_cast_fp16))[name = tensor("op_4169_cast_fp16")]; + tensor var_4171_interleave_0 = const()[name = tensor("op_4171_interleave_0"), val = tensor(false)]; + tensor var_4171_cast_fp16 = concat(axis = var_3089, interleave = var_4171_interleave_0, values = (var_4143_cast_fp16, var_4145_cast_fp16))[name = tensor("op_4171_cast_fp16")]; + tensor var_4173_interleave_0 = const()[name = tensor("op_4173_interleave_0"), val = tensor(false)]; + tensor var_4173_cast_fp16 = concat(axis = var_3089, interleave = var_4173_interleave_0, values = (var_4147_cast_fp16, var_4149_cast_fp16))[name = tensor("op_4173_cast_fp16")]; + tensor var_4175_interleave_0 = const()[name = tensor("op_4175_interleave_0"), val = tensor(false)]; + tensor var_4175_cast_fp16 = concat(axis = var_3089, interleave = var_4175_interleave_0, values = (var_4151_cast_fp16, var_4153_cast_fp16))[name = tensor("op_4175_cast_fp16")]; + tensor var_4177_interleave_0 = const()[name = tensor("op_4177_interleave_0"), val = tensor(false)]; + tensor var_4177_cast_fp16 = concat(axis = var_3089, interleave = var_4177_interleave_0, values = (var_4155_cast_fp16, var_4157_cast_fp16))[name = tensor("op_4177_cast_fp16")]; + tensor var_4179_interleave_0 = const()[name = tensor("op_4179_interleave_0"), val = tensor(false)]; + tensor var_4179_cast_fp16 = concat(axis = var_3089, interleave = var_4179_interleave_0, values = (var_4159_cast_fp16, var_4161_cast_fp16))[name = tensor("op_4179_cast_fp16")]; + tensor var_4181_interleave_0 = const()[name = tensor("op_4181_interleave_0"), val = tensor(false)]; + tensor var_4181_cast_fp16 = concat(axis = var_3089, interleave = var_4181_interleave_0, values = (var_4163_cast_fp16, var_4165_cast_fp16))[name = tensor("op_4181_cast_fp16")]; + tensor input_105_interleave_0 = const()[name = tensor("input_105_interleave_0"), val = tensor(false)]; + tensor input_105_cast_fp16 = concat(axis = var_3111, interleave = input_105_interleave_0, values = (var_4167_cast_fp16, var_4169_cast_fp16, var_4171_cast_fp16, var_4173_cast_fp16, var_4175_cast_fp16, var_4177_cast_fp16, var_4179_cast_fp16, var_4181_cast_fp16))[name = tensor("input_105_cast_fp16")]; + tensor var_4191_pad_type_0 = const()[name = tensor("op_4191_pad_type_0"), val = tensor("valid")]; + tensor var_4191_strides_0 = const()[name = tensor("op_4191_strides_0"), val = tensor([1, 1])]; + tensor var_4191_pad_0 = const()[name = tensor("op_4191_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4191_dilations_0 = const()[name = tensor("op_4191_dilations_0"), val = tensor([1, 1])]; + tensor var_4191_groups_0 = const()[name = tensor("op_4191_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_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(76356352)))]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_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(77175616)))]; + tensor var_4191_cast_fp16 = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_4191_dilations_0, groups = var_4191_groups_0, pad = var_4191_pad_0, pad_type = var_4191_pad_type_0, strides = var_4191_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_105_cast_fp16)[name = tensor("op_4191_cast_fp16")]; + tensor inputs_21_cast_fp16 = add(x = var_4191_cast_fp16, y = inputs_19_cast_fp16)[name = tensor("inputs_21_cast_fp16")]; + tensor hidden_states_65_axes_0 = const()[name = tensor("hidden_states_65_axes_0"), val = tensor([1])]; + tensor hidden_states_65_gamma_0_to_fp16 = const()[name = tensor("hidden_states_65_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77176960)))]; + tensor hidden_states_65_beta_0_to_fp16 = const()[name = tensor("hidden_states_65_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77178304)))]; + tensor var_4201_to_fp16 = const()[name = tensor("op_4201_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_65_cast_fp16 = layer_norm(axes = hidden_states_65_axes_0, beta = hidden_states_65_beta_0_to_fp16, epsilon = var_4201_to_fp16, gamma = hidden_states_65_gamma_0_to_fp16, x = inputs_21_cast_fp16)[name = tensor("hidden_states_65_cast_fp16")]; + tensor q_15_pad_type_0 = const()[name = tensor("q_15_pad_type_0"), val = tensor("valid")]; + tensor q_15_strides_0 = const()[name = tensor("q_15_strides_0"), val = tensor([1, 1])]; + tensor q_15_pad_0 = const()[name = tensor("q_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_15_dilations_0 = const()[name = tensor("q_15_dilations_0"), val = tensor([1, 1])]; + tensor q_15_groups_0 = const()[name = tensor("q_15_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_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(77179648)))]; + tensor q_15_cast_fp16 = conv(dilations = q_15_dilations_0, groups = q_15_groups_0, pad = q_15_pad_0, pad_type = q_15_pad_type_0, strides = q_15_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_65_cast_fp16)[name = tensor("q_15_cast_fp16")]; + tensor k_29_pad_type_0 = const()[name = tensor("k_29_pad_type_0"), val = tensor("valid")]; + tensor k_29_strides_0 = const()[name = tensor("k_29_strides_0"), val = tensor([1, 1])]; + tensor k_29_pad_0 = const()[name = tensor("k_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_29_dilations_0 = const()[name = tensor("k_29_dilations_0"), val = tensor([1, 1])]; + tensor k_29_groups_0 = const()[name = tensor("k_29_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_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(77998912)))]; + tensor k_29_cast_fp16 = conv(dilations = k_29_dilations_0, groups = k_29_groups_0, pad = k_29_pad_0, pad_type = k_29_pad_type_0, strides = k_29_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_29_cast_fp16")]; + tensor v_15_pad_type_0 = const()[name = tensor("v_15_pad_type_0"), val = tensor("valid")]; + tensor v_15_strides_0 = const()[name = tensor("v_15_strides_0"), val = tensor([1, 1])]; + tensor v_15_pad_0 = const()[name = tensor("v_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_15_dilations_0 = const()[name = tensor("v_15_dilations_0"), val = tensor([1, 1])]; + tensor v_15_groups_0 = const()[name = tensor("v_15_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_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(78982016)))]; + tensor v_15_cast_fp16 = conv(dilations = v_15_dilations_0, groups = v_15_groups_0, pad = v_15_pad_0, pad_type = v_15_pad_type_0, strides = v_15_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_15_cast_fp16")]; + tensor var_4234_begin_0 = const()[name = tensor("op_4234_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4234_end_0 = const()[name = tensor("op_4234_end_0"), val = tensor([2, 80, 1, 1024])]; + tensor var_4234_end_mask_0 = const()[name = tensor("op_4234_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4234_cast_fp16 = slice_by_index(begin = var_4234_begin_0, end = var_4234_end_0, end_mask = var_4234_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_4234_cast_fp16")]; + tensor var_4238_begin_0 = const()[name = tensor("op_4238_begin_0"), val = tensor([0, 80, 0, 0])]; + tensor var_4238_end_0 = const()[name = tensor("op_4238_end_0"), val = tensor([2, 160, 1, 1024])]; + tensor var_4238_end_mask_0 = const()[name = tensor("op_4238_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4238_cast_fp16 = slice_by_index(begin = var_4238_begin_0, end = var_4238_end_0, end_mask = var_4238_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_4238_cast_fp16")]; + tensor var_4242_begin_0 = const()[name = tensor("op_4242_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_4242_end_0 = const()[name = tensor("op_4242_end_0"), val = tensor([2, 240, 1, 1024])]; + tensor var_4242_end_mask_0 = const()[name = tensor("op_4242_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4242_cast_fp16 = slice_by_index(begin = var_4242_begin_0, end = var_4242_end_0, end_mask = var_4242_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_4242_cast_fp16")]; + tensor var_4246_begin_0 = const()[name = tensor("op_4246_begin_0"), val = tensor([0, 240, 0, 0])]; + tensor var_4246_end_0 = const()[name = tensor("op_4246_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_4246_end_mask_0 = const()[name = tensor("op_4246_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4246_cast_fp16 = slice_by_index(begin = var_4246_begin_0, end = var_4246_end_0, end_mask = var_4246_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_4246_cast_fp16")]; + tensor var_4250_begin_0 = const()[name = tensor("op_4250_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4250_end_0 = const()[name = tensor("op_4250_end_0"), val = tensor([2, 400, 1, 1024])]; + tensor var_4250_end_mask_0 = const()[name = tensor("op_4250_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4250_cast_fp16 = slice_by_index(begin = var_4250_begin_0, end = var_4250_end_0, end_mask = var_4250_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_4250_cast_fp16")]; + tensor var_4254_begin_0 = const()[name = tensor("op_4254_begin_0"), val = tensor([0, 400, 0, 0])]; + tensor var_4254_end_0 = const()[name = tensor("op_4254_end_0"), val = tensor([2, 480, 1, 1024])]; + tensor var_4254_end_mask_0 = const()[name = tensor("op_4254_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4254_cast_fp16 = slice_by_index(begin = var_4254_begin_0, end = var_4254_end_0, end_mask = var_4254_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_4254_cast_fp16")]; + tensor var_4258_begin_0 = const()[name = tensor("op_4258_begin_0"), val = tensor([0, 480, 0, 0])]; + tensor var_4258_end_0 = const()[name = tensor("op_4258_end_0"), val = tensor([2, 560, 1, 1024])]; + tensor var_4258_end_mask_0 = const()[name = tensor("op_4258_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4258_cast_fp16 = slice_by_index(begin = var_4258_begin_0, end = var_4258_end_0, end_mask = var_4258_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_4258_cast_fp16")]; + tensor var_4262_begin_0 = const()[name = tensor("op_4262_begin_0"), val = tensor([0, 560, 0, 0])]; + tensor var_4262_end_0 = const()[name = tensor("op_4262_end_0"), val = tensor([2, 1, 1, 1024])]; + tensor var_4262_end_mask_0 = const()[name = tensor("op_4262_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4262_cast_fp16 = slice_by_index(begin = var_4262_begin_0, end = var_4262_end_0, end_mask = var_4262_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_4262_cast_fp16")]; + tensor var_4265_begin_0 = const()[name = tensor("op_4265_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4265_end_0 = const()[name = tensor("op_4265_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_4265_end_mask_0 = const()[name = tensor("op_4265_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4265_cast_fp16 = slice_by_index(begin = var_4265_begin_0, end = var_4265_end_0, end_mask = var_4265_end_mask_0, x = var_4234_cast_fp16)[name = tensor("op_4265_cast_fp16")]; + tensor var_4266_begin_0 = const()[name = tensor("op_4266_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_4266_end_0 = const()[name = tensor("op_4266_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_4266_end_mask_0 = const()[name = tensor("op_4266_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4266_cast_fp16 = slice_by_index(begin = var_4266_begin_0, end = var_4266_end_0, end_mask = var_4266_end_mask_0, x = var_4234_cast_fp16)[name = tensor("op_4266_cast_fp16")]; + tensor var_4267_begin_0 = const()[name = tensor("op_4267_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4267_end_0 = const()[name = tensor("op_4267_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_4267_end_mask_0 = const()[name = tensor("op_4267_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4267_cast_fp16 = slice_by_index(begin = var_4267_begin_0, end = var_4267_end_0, end_mask = var_4267_end_mask_0, x = var_4238_cast_fp16)[name = tensor("op_4267_cast_fp16")]; + tensor var_4268_begin_0 = const()[name = tensor("op_4268_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_4268_end_0 = const()[name = tensor("op_4268_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_4268_end_mask_0 = const()[name = tensor("op_4268_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4268_cast_fp16 = slice_by_index(begin = var_4268_begin_0, end = var_4268_end_0, end_mask = var_4268_end_mask_0, x = var_4238_cast_fp16)[name = tensor("op_4268_cast_fp16")]; + tensor var_4269_begin_0 = const()[name = tensor("op_4269_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4269_end_0 = const()[name = tensor("op_4269_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_4269_end_mask_0 = const()[name = tensor("op_4269_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4269_cast_fp16 = slice_by_index(begin = var_4269_begin_0, end = var_4269_end_0, end_mask = var_4269_end_mask_0, x = var_4242_cast_fp16)[name = tensor("op_4269_cast_fp16")]; + tensor var_4270_begin_0 = const()[name = tensor("op_4270_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_4270_end_0 = const()[name = tensor("op_4270_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_4270_end_mask_0 = const()[name = tensor("op_4270_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4270_cast_fp16 = slice_by_index(begin = var_4270_begin_0, end = var_4270_end_0, end_mask = var_4270_end_mask_0, x = var_4242_cast_fp16)[name = tensor("op_4270_cast_fp16")]; + tensor var_4271_begin_0 = const()[name = tensor("op_4271_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4271_end_0 = const()[name = tensor("op_4271_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_4271_end_mask_0 = const()[name = tensor("op_4271_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4271_cast_fp16 = slice_by_index(begin = var_4271_begin_0, end = var_4271_end_0, end_mask = var_4271_end_mask_0, x = var_4246_cast_fp16)[name = tensor("op_4271_cast_fp16")]; + tensor var_4272_begin_0 = const()[name = tensor("op_4272_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_4272_end_0 = const()[name = tensor("op_4272_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_4272_end_mask_0 = const()[name = tensor("op_4272_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4272_cast_fp16 = slice_by_index(begin = var_4272_begin_0, end = var_4272_end_0, end_mask = var_4272_end_mask_0, x = var_4246_cast_fp16)[name = tensor("op_4272_cast_fp16")]; + tensor var_4273_begin_0 = const()[name = tensor("op_4273_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4273_end_0 = const()[name = tensor("op_4273_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_4273_end_mask_0 = const()[name = tensor("op_4273_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4273_cast_fp16 = slice_by_index(begin = var_4273_begin_0, end = var_4273_end_0, end_mask = var_4273_end_mask_0, x = var_4250_cast_fp16)[name = tensor("op_4273_cast_fp16")]; + tensor var_4274_begin_0 = const()[name = tensor("op_4274_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_4274_end_0 = const()[name = tensor("op_4274_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_4274_end_mask_0 = const()[name = tensor("op_4274_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4274_cast_fp16 = slice_by_index(begin = var_4274_begin_0, end = var_4274_end_0, end_mask = var_4274_end_mask_0, x = var_4250_cast_fp16)[name = tensor("op_4274_cast_fp16")]; + tensor var_4275_begin_0 = const()[name = tensor("op_4275_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4275_end_0 = const()[name = tensor("op_4275_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_4275_end_mask_0 = const()[name = tensor("op_4275_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4275_cast_fp16 = slice_by_index(begin = var_4275_begin_0, end = var_4275_end_0, end_mask = var_4275_end_mask_0, x = var_4254_cast_fp16)[name = tensor("op_4275_cast_fp16")]; + tensor var_4276_begin_0 = const()[name = tensor("op_4276_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_4276_end_0 = const()[name = tensor("op_4276_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_4276_end_mask_0 = const()[name = tensor("op_4276_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4276_cast_fp16 = slice_by_index(begin = var_4276_begin_0, end = var_4276_end_0, end_mask = var_4276_end_mask_0, x = var_4254_cast_fp16)[name = tensor("op_4276_cast_fp16")]; + tensor var_4277_begin_0 = const()[name = tensor("op_4277_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4277_end_0 = const()[name = tensor("op_4277_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_4277_end_mask_0 = const()[name = tensor("op_4277_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4277_cast_fp16 = slice_by_index(begin = var_4277_begin_0, end = var_4277_end_0, end_mask = var_4277_end_mask_0, x = var_4258_cast_fp16)[name = tensor("op_4277_cast_fp16")]; + tensor var_4278_begin_0 = const()[name = tensor("op_4278_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_4278_end_0 = const()[name = tensor("op_4278_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_4278_end_mask_0 = const()[name = tensor("op_4278_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4278_cast_fp16 = slice_by_index(begin = var_4278_begin_0, end = var_4278_end_0, end_mask = var_4278_end_mask_0, x = var_4258_cast_fp16)[name = tensor("op_4278_cast_fp16")]; + tensor var_4279_begin_0 = const()[name = tensor("op_4279_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4279_end_0 = const()[name = tensor("op_4279_end_0"), val = tensor([2, 80, 1, 512])]; + tensor var_4279_end_mask_0 = const()[name = tensor("op_4279_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4279_cast_fp16 = slice_by_index(begin = var_4279_begin_0, end = var_4279_end_0, end_mask = var_4279_end_mask_0, x = var_4262_cast_fp16)[name = tensor("op_4279_cast_fp16")]; + tensor var_4280_begin_0 = const()[name = tensor("op_4280_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_4280_end_0 = const()[name = tensor("op_4280_end_0"), val = tensor([2, 80, 1, 1])]; + tensor var_4280_end_mask_0 = const()[name = tensor("op_4280_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4280_cast_fp16 = slice_by_index(begin = var_4280_begin_0, end = var_4280_end_0, end_mask = var_4280_end_mask_0, x = var_4262_cast_fp16)[name = tensor("op_4280_cast_fp16")]; + tensor k_31_perm_0 = const()[name = tensor("k_31_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_4285_begin_0 = const()[name = tensor("op_4285_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4285_end_0 = const()[name = tensor("op_4285_end_0"), val = tensor([2, 77, 1, 80])]; + tensor var_4285_end_mask_0 = const()[name = tensor("op_4285_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_31_cast_fp16 = transpose(perm = k_31_perm_0, x = k_29_cast_fp16)[name = tensor("transpose_6")]; + tensor var_4285_cast_fp16 = slice_by_index(begin = var_4285_begin_0, end = var_4285_end_0, end_mask = var_4285_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_4285_cast_fp16")]; + tensor var_4289_begin_0 = const()[name = tensor("op_4289_begin_0"), val = tensor([0, 0, 0, 80])]; + tensor var_4289_end_0 = const()[name = tensor("op_4289_end_0"), val = tensor([2, 77, 1, 160])]; + tensor var_4289_end_mask_0 = const()[name = tensor("op_4289_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4289_cast_fp16 = slice_by_index(begin = var_4289_begin_0, end = var_4289_end_0, end_mask = var_4289_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_4289_cast_fp16")]; + tensor var_4293_begin_0 = const()[name = tensor("op_4293_begin_0"), val = tensor([0, 0, 0, 160])]; + tensor var_4293_end_0 = const()[name = tensor("op_4293_end_0"), val = tensor([2, 77, 1, 240])]; + tensor var_4293_end_mask_0 = const()[name = tensor("op_4293_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4293_cast_fp16 = slice_by_index(begin = var_4293_begin_0, end = var_4293_end_0, end_mask = var_4293_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_4293_cast_fp16")]; + tensor var_4297_begin_0 = const()[name = tensor("op_4297_begin_0"), val = tensor([0, 0, 0, 240])]; + tensor var_4297_end_0 = const()[name = tensor("op_4297_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_4297_end_mask_0 = const()[name = tensor("op_4297_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4297_cast_fp16 = slice_by_index(begin = var_4297_begin_0, end = var_4297_end_0, end_mask = var_4297_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_4297_cast_fp16")]; + tensor var_4301_begin_0 = const()[name = tensor("op_4301_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_4301_end_0 = const()[name = tensor("op_4301_end_0"), val = tensor([2, 77, 1, 400])]; + tensor var_4301_end_mask_0 = const()[name = tensor("op_4301_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4301_cast_fp16 = slice_by_index(begin = var_4301_begin_0, end = var_4301_end_0, end_mask = var_4301_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_4301_cast_fp16")]; + tensor var_4305_begin_0 = const()[name = tensor("op_4305_begin_0"), val = tensor([0, 0, 0, 400])]; + tensor var_4305_end_0 = const()[name = tensor("op_4305_end_0"), val = tensor([2, 77, 1, 480])]; + tensor var_4305_end_mask_0 = const()[name = tensor("op_4305_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4305_cast_fp16 = slice_by_index(begin = var_4305_begin_0, end = var_4305_end_0, end_mask = var_4305_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_4305_cast_fp16")]; + tensor var_4309_begin_0 = const()[name = tensor("op_4309_begin_0"), val = tensor([0, 0, 0, 480])]; + tensor var_4309_end_0 = const()[name = tensor("op_4309_end_0"), val = tensor([2, 77, 1, 560])]; + tensor var_4309_end_mask_0 = const()[name = tensor("op_4309_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4309_cast_fp16 = slice_by_index(begin = var_4309_begin_0, end = var_4309_end_0, end_mask = var_4309_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_4309_cast_fp16")]; + tensor var_4313_begin_0 = const()[name = tensor("op_4313_begin_0"), val = tensor([0, 0, 0, 560])]; + tensor var_4313_end_0 = const()[name = tensor("op_4313_end_0"), val = tensor([2, 77, 1, 1])]; + tensor var_4313_end_mask_0 = const()[name = tensor("op_4313_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4313_cast_fp16 = slice_by_index(begin = var_4313_begin_0, end = var_4313_end_0, end_mask = var_4313_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_4313_cast_fp16")]; + tensor var_4315_begin_0 = const()[name = tensor("op_4315_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4315_end_0 = const()[name = tensor("op_4315_end_0"), val = tensor([2, 80, 1, 77])]; + tensor var_4315_end_mask_0 = const()[name = tensor("op_4315_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4315_cast_fp16 = slice_by_index(begin = var_4315_begin_0, end = var_4315_end_0, end_mask = var_4315_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_4315_cast_fp16")]; + tensor var_4319_begin_0 = const()[name = tensor("op_4319_begin_0"), val = tensor([0, 80, 0, 0])]; + tensor var_4319_end_0 = const()[name = tensor("op_4319_end_0"), val = tensor([2, 160, 1, 77])]; + tensor var_4319_end_mask_0 = const()[name = tensor("op_4319_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4319_cast_fp16 = slice_by_index(begin = var_4319_begin_0, end = var_4319_end_0, end_mask = var_4319_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_4319_cast_fp16")]; + tensor var_4323_begin_0 = const()[name = tensor("op_4323_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_4323_end_0 = const()[name = tensor("op_4323_end_0"), val = tensor([2, 240, 1, 77])]; + tensor var_4323_end_mask_0 = const()[name = tensor("op_4323_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4323_cast_fp16 = slice_by_index(begin = var_4323_begin_0, end = var_4323_end_0, end_mask = var_4323_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_4323_cast_fp16")]; + tensor var_4327_begin_0 = const()[name = tensor("op_4327_begin_0"), val = tensor([0, 240, 0, 0])]; + tensor var_4327_end_0 = const()[name = tensor("op_4327_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_4327_end_mask_0 = const()[name = tensor("op_4327_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4327_cast_fp16 = slice_by_index(begin = var_4327_begin_0, end = var_4327_end_0, end_mask = var_4327_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_4327_cast_fp16")]; + tensor var_4331_begin_0 = const()[name = tensor("op_4331_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4331_end_0 = const()[name = tensor("op_4331_end_0"), val = tensor([2, 400, 1, 77])]; + tensor var_4331_end_mask_0 = const()[name = tensor("op_4331_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4331_cast_fp16 = slice_by_index(begin = var_4331_begin_0, end = var_4331_end_0, end_mask = var_4331_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_4331_cast_fp16")]; + tensor var_4335_begin_0 = const()[name = tensor("op_4335_begin_0"), val = tensor([0, 400, 0, 0])]; + tensor var_4335_end_0 = const()[name = tensor("op_4335_end_0"), val = tensor([2, 480, 1, 77])]; + tensor var_4335_end_mask_0 = const()[name = tensor("op_4335_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4335_cast_fp16 = slice_by_index(begin = var_4335_begin_0, end = var_4335_end_0, end_mask = var_4335_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_4335_cast_fp16")]; + tensor var_4339_begin_0 = const()[name = tensor("op_4339_begin_0"), val = tensor([0, 480, 0, 0])]; + tensor var_4339_end_0 = const()[name = tensor("op_4339_end_0"), val = tensor([2, 560, 1, 77])]; + tensor var_4339_end_mask_0 = const()[name = tensor("op_4339_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4339_cast_fp16 = slice_by_index(begin = var_4339_begin_0, end = var_4339_end_0, end_mask = var_4339_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_4339_cast_fp16")]; + tensor var_4343_begin_0 = const()[name = tensor("op_4343_begin_0"), val = tensor([0, 560, 0, 0])]; + tensor var_4343_end_0 = const()[name = tensor("op_4343_end_0"), val = tensor([2, 1, 1, 77])]; + tensor var_4343_end_mask_0 = const()[name = tensor("op_4343_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4343_cast_fp16 = slice_by_index(begin = var_4343_begin_0, end = var_4343_end_0, end_mask = var_4343_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_4343_cast_fp16")]; + tensor var_4347_equation_0 = const()[name = tensor("op_4347_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4347_cast_fp16 = einsum(equation = var_4347_equation_0, values = (var_4285_cast_fp16, var_4265_cast_fp16))[name = tensor("op_4347_cast_fp16")]; + tensor var_4348_to_fp16 = const()[name = tensor("op_4348_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_609_cast_fp16 = mul(x = var_4347_cast_fp16, y = var_4348_to_fp16)[name = tensor("aw_chunk_609_cast_fp16")]; + tensor var_4351_equation_0 = const()[name = tensor("op_4351_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4351_cast_fp16 = einsum(equation = var_4351_equation_0, values = (var_4285_cast_fp16, var_4266_cast_fp16))[name = tensor("op_4351_cast_fp16")]; + tensor var_4352_to_fp16 = const()[name = tensor("op_4352_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_611_cast_fp16 = mul(x = var_4351_cast_fp16, y = var_4352_to_fp16)[name = tensor("aw_chunk_611_cast_fp16")]; + tensor var_4355_equation_0 = const()[name = tensor("op_4355_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4355_cast_fp16 = einsum(equation = var_4355_equation_0, values = (var_4289_cast_fp16, var_4267_cast_fp16))[name = tensor("op_4355_cast_fp16")]; + tensor var_4356_to_fp16 = const()[name = tensor("op_4356_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_613_cast_fp16 = mul(x = var_4355_cast_fp16, y = var_4356_to_fp16)[name = tensor("aw_chunk_613_cast_fp16")]; + tensor var_4359_equation_0 = const()[name = tensor("op_4359_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4359_cast_fp16 = einsum(equation = var_4359_equation_0, values = (var_4289_cast_fp16, var_4268_cast_fp16))[name = tensor("op_4359_cast_fp16")]; + tensor var_4360_to_fp16 = const()[name = tensor("op_4360_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_615_cast_fp16 = mul(x = var_4359_cast_fp16, y = var_4360_to_fp16)[name = tensor("aw_chunk_615_cast_fp16")]; + tensor var_4363_equation_0 = const()[name = tensor("op_4363_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4363_cast_fp16 = einsum(equation = var_4363_equation_0, values = (var_4293_cast_fp16, var_4269_cast_fp16))[name = tensor("op_4363_cast_fp16")]; + tensor var_4364_to_fp16 = const()[name = tensor("op_4364_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_617_cast_fp16 = mul(x = var_4363_cast_fp16, y = var_4364_to_fp16)[name = tensor("aw_chunk_617_cast_fp16")]; + tensor var_4367_equation_0 = const()[name = tensor("op_4367_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4367_cast_fp16 = einsum(equation = var_4367_equation_0, values = (var_4293_cast_fp16, var_4270_cast_fp16))[name = tensor("op_4367_cast_fp16")]; + tensor var_4368_to_fp16 = const()[name = tensor("op_4368_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_619_cast_fp16 = mul(x = var_4367_cast_fp16, y = var_4368_to_fp16)[name = tensor("aw_chunk_619_cast_fp16")]; + tensor var_4371_equation_0 = const()[name = tensor("op_4371_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4371_cast_fp16 = einsum(equation = var_4371_equation_0, values = (var_4297_cast_fp16, var_4271_cast_fp16))[name = tensor("op_4371_cast_fp16")]; + tensor var_4372_to_fp16 = const()[name = tensor("op_4372_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_621_cast_fp16 = mul(x = var_4371_cast_fp16, y = var_4372_to_fp16)[name = tensor("aw_chunk_621_cast_fp16")]; + tensor var_4375_equation_0 = const()[name = tensor("op_4375_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4375_cast_fp16 = einsum(equation = var_4375_equation_0, values = (var_4297_cast_fp16, var_4272_cast_fp16))[name = tensor("op_4375_cast_fp16")]; + tensor var_4376_to_fp16 = const()[name = tensor("op_4376_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_623_cast_fp16 = mul(x = var_4375_cast_fp16, y = var_4376_to_fp16)[name = tensor("aw_chunk_623_cast_fp16")]; + tensor var_4379_equation_0 = const()[name = tensor("op_4379_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4379_cast_fp16 = einsum(equation = var_4379_equation_0, values = (var_4301_cast_fp16, var_4273_cast_fp16))[name = tensor("op_4379_cast_fp16")]; + tensor var_4380_to_fp16 = const()[name = tensor("op_4380_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_625_cast_fp16 = mul(x = var_4379_cast_fp16, y = var_4380_to_fp16)[name = tensor("aw_chunk_625_cast_fp16")]; + tensor var_4383_equation_0 = const()[name = tensor("op_4383_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4383_cast_fp16 = einsum(equation = var_4383_equation_0, values = (var_4301_cast_fp16, var_4274_cast_fp16))[name = tensor("op_4383_cast_fp16")]; + tensor var_4384_to_fp16 = const()[name = tensor("op_4384_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_627_cast_fp16 = mul(x = var_4383_cast_fp16, y = var_4384_to_fp16)[name = tensor("aw_chunk_627_cast_fp16")]; + tensor var_4387_equation_0 = const()[name = tensor("op_4387_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4387_cast_fp16 = einsum(equation = var_4387_equation_0, values = (var_4305_cast_fp16, var_4275_cast_fp16))[name = tensor("op_4387_cast_fp16")]; + tensor var_4388_to_fp16 = const()[name = tensor("op_4388_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_629_cast_fp16 = mul(x = var_4387_cast_fp16, y = var_4388_to_fp16)[name = tensor("aw_chunk_629_cast_fp16")]; + tensor var_4391_equation_0 = const()[name = tensor("op_4391_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4391_cast_fp16 = einsum(equation = var_4391_equation_0, values = (var_4305_cast_fp16, var_4276_cast_fp16))[name = tensor("op_4391_cast_fp16")]; + tensor var_4392_to_fp16 = const()[name = tensor("op_4392_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_631_cast_fp16 = mul(x = var_4391_cast_fp16, y = var_4392_to_fp16)[name = tensor("aw_chunk_631_cast_fp16")]; + tensor var_4395_equation_0 = const()[name = tensor("op_4395_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4395_cast_fp16 = einsum(equation = var_4395_equation_0, values = (var_4309_cast_fp16, var_4277_cast_fp16))[name = tensor("op_4395_cast_fp16")]; + tensor var_4396_to_fp16 = const()[name = tensor("op_4396_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_633_cast_fp16 = mul(x = var_4395_cast_fp16, y = var_4396_to_fp16)[name = tensor("aw_chunk_633_cast_fp16")]; + tensor var_4399_equation_0 = const()[name = tensor("op_4399_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4399_cast_fp16 = einsum(equation = var_4399_equation_0, values = (var_4309_cast_fp16, var_4278_cast_fp16))[name = tensor("op_4399_cast_fp16")]; + tensor var_4400_to_fp16 = const()[name = tensor("op_4400_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_635_cast_fp16 = mul(x = var_4399_cast_fp16, y = var_4400_to_fp16)[name = tensor("aw_chunk_635_cast_fp16")]; + tensor var_4403_equation_0 = const()[name = tensor("op_4403_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4403_cast_fp16 = einsum(equation = var_4403_equation_0, values = (var_4313_cast_fp16, var_4279_cast_fp16))[name = tensor("op_4403_cast_fp16")]; + tensor var_4404_to_fp16 = const()[name = tensor("op_4404_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_637_cast_fp16 = mul(x = var_4403_cast_fp16, y = var_4404_to_fp16)[name = tensor("aw_chunk_637_cast_fp16")]; + tensor var_4407_equation_0 = const()[name = tensor("op_4407_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4407_cast_fp16 = einsum(equation = var_4407_equation_0, values = (var_4313_cast_fp16, var_4280_cast_fp16))[name = tensor("op_4407_cast_fp16")]; + tensor var_4408_to_fp16 = const()[name = tensor("op_4408_to_fp16"), val = tensor(0x1.cap-4)]; + tensor aw_chunk_639_cast_fp16 = mul(x = var_4407_cast_fp16, y = var_4408_to_fp16)[name = tensor("aw_chunk_639_cast_fp16")]; + tensor var_4410_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_609_cast_fp16)[name = tensor("op_4410_cast_fp16")]; + tensor var_4411_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_611_cast_fp16)[name = tensor("op_4411_cast_fp16")]; + tensor var_4412_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_613_cast_fp16)[name = tensor("op_4412_cast_fp16")]; + tensor var_4413_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_615_cast_fp16)[name = tensor("op_4413_cast_fp16")]; + tensor var_4414_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_617_cast_fp16)[name = tensor("op_4414_cast_fp16")]; + tensor var_4415_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_619_cast_fp16)[name = tensor("op_4415_cast_fp16")]; + tensor var_4416_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_621_cast_fp16)[name = tensor("op_4416_cast_fp16")]; + tensor var_4417_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_623_cast_fp16)[name = tensor("op_4417_cast_fp16")]; + tensor var_4418_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_625_cast_fp16)[name = tensor("op_4418_cast_fp16")]; + tensor var_4419_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_627_cast_fp16)[name = tensor("op_4419_cast_fp16")]; + tensor var_4420_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_629_cast_fp16)[name = tensor("op_4420_cast_fp16")]; + tensor var_4421_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_631_cast_fp16)[name = tensor("op_4421_cast_fp16")]; + tensor var_4422_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_633_cast_fp16)[name = tensor("op_4422_cast_fp16")]; + tensor var_4423_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_635_cast_fp16)[name = tensor("op_4423_cast_fp16")]; + tensor var_4424_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_637_cast_fp16)[name = tensor("op_4424_cast_fp16")]; + tensor var_4425_cast_fp16 = softmax(axis = var_3111, x = aw_chunk_639_cast_fp16)[name = tensor("op_4425_cast_fp16")]; + tensor var_4427_equation_0 = const()[name = tensor("op_4427_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4427_cast_fp16 = einsum(equation = var_4427_equation_0, values = (var_4315_cast_fp16, var_4410_cast_fp16))[name = tensor("op_4427_cast_fp16")]; + tensor var_4429_equation_0 = const()[name = tensor("op_4429_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4429_cast_fp16 = einsum(equation = var_4429_equation_0, values = (var_4315_cast_fp16, var_4411_cast_fp16))[name = tensor("op_4429_cast_fp16")]; + tensor var_4431_equation_0 = const()[name = tensor("op_4431_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4431_cast_fp16 = einsum(equation = var_4431_equation_0, values = (var_4319_cast_fp16, var_4412_cast_fp16))[name = tensor("op_4431_cast_fp16")]; + tensor var_4433_equation_0 = const()[name = tensor("op_4433_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4433_cast_fp16 = einsum(equation = var_4433_equation_0, values = (var_4319_cast_fp16, var_4413_cast_fp16))[name = tensor("op_4433_cast_fp16")]; + tensor var_4435_equation_0 = const()[name = tensor("op_4435_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4435_cast_fp16 = einsum(equation = var_4435_equation_0, values = (var_4323_cast_fp16, var_4414_cast_fp16))[name = tensor("op_4435_cast_fp16")]; + tensor var_4437_equation_0 = const()[name = tensor("op_4437_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4437_cast_fp16 = einsum(equation = var_4437_equation_0, values = (var_4323_cast_fp16, var_4415_cast_fp16))[name = tensor("op_4437_cast_fp16")]; + tensor var_4439_equation_0 = const()[name = tensor("op_4439_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4439_cast_fp16 = einsum(equation = var_4439_equation_0, values = (var_4327_cast_fp16, var_4416_cast_fp16))[name = tensor("op_4439_cast_fp16")]; + tensor var_4441_equation_0 = const()[name = tensor("op_4441_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4441_cast_fp16 = einsum(equation = var_4441_equation_0, values = (var_4327_cast_fp16, var_4417_cast_fp16))[name = tensor("op_4441_cast_fp16")]; + tensor var_4443_equation_0 = const()[name = tensor("op_4443_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4443_cast_fp16 = einsum(equation = var_4443_equation_0, values = (var_4331_cast_fp16, var_4418_cast_fp16))[name = tensor("op_4443_cast_fp16")]; + tensor var_4445_equation_0 = const()[name = tensor("op_4445_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4445_cast_fp16 = einsum(equation = var_4445_equation_0, values = (var_4331_cast_fp16, var_4419_cast_fp16))[name = tensor("op_4445_cast_fp16")]; + tensor var_4447_equation_0 = const()[name = tensor("op_4447_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4447_cast_fp16 = einsum(equation = var_4447_equation_0, values = (var_4335_cast_fp16, var_4420_cast_fp16))[name = tensor("op_4447_cast_fp16")]; + tensor var_4449_equation_0 = const()[name = tensor("op_4449_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4449_cast_fp16 = einsum(equation = var_4449_equation_0, values = (var_4335_cast_fp16, var_4421_cast_fp16))[name = tensor("op_4449_cast_fp16")]; + tensor var_4451_equation_0 = const()[name = tensor("op_4451_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4451_cast_fp16 = einsum(equation = var_4451_equation_0, values = (var_4339_cast_fp16, var_4422_cast_fp16))[name = tensor("op_4451_cast_fp16")]; + tensor var_4453_equation_0 = const()[name = tensor("op_4453_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4453_cast_fp16 = einsum(equation = var_4453_equation_0, values = (var_4339_cast_fp16, var_4423_cast_fp16))[name = tensor("op_4453_cast_fp16")]; + tensor var_4455_equation_0 = const()[name = tensor("op_4455_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4455_cast_fp16 = einsum(equation = var_4455_equation_0, values = (var_4343_cast_fp16, var_4424_cast_fp16))[name = tensor("op_4455_cast_fp16")]; + tensor var_4457_equation_0 = const()[name = tensor("op_4457_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4457_cast_fp16 = einsum(equation = var_4457_equation_0, values = (var_4343_cast_fp16, var_4425_cast_fp16))[name = tensor("op_4457_cast_fp16")]; + tensor var_4459_interleave_0 = const()[name = tensor("op_4459_interleave_0"), val = tensor(false)]; + tensor var_4459_cast_fp16 = concat(axis = var_3089, interleave = var_4459_interleave_0, values = (var_4427_cast_fp16, var_4429_cast_fp16))[name = tensor("op_4459_cast_fp16")]; + tensor var_4461_interleave_0 = const()[name = tensor("op_4461_interleave_0"), val = tensor(false)]; + tensor var_4461_cast_fp16 = concat(axis = var_3089, interleave = var_4461_interleave_0, values = (var_4431_cast_fp16, var_4433_cast_fp16))[name = tensor("op_4461_cast_fp16")]; + tensor var_4463_interleave_0 = const()[name = tensor("op_4463_interleave_0"), val = tensor(false)]; + tensor var_4463_cast_fp16 = concat(axis = var_3089, interleave = var_4463_interleave_0, values = (var_4435_cast_fp16, var_4437_cast_fp16))[name = tensor("op_4463_cast_fp16")]; + tensor var_4465_interleave_0 = const()[name = tensor("op_4465_interleave_0"), val = tensor(false)]; + tensor var_4465_cast_fp16 = concat(axis = var_3089, interleave = var_4465_interleave_0, values = (var_4439_cast_fp16, var_4441_cast_fp16))[name = tensor("op_4465_cast_fp16")]; + tensor var_4467_interleave_0 = const()[name = tensor("op_4467_interleave_0"), val = tensor(false)]; + tensor var_4467_cast_fp16 = concat(axis = var_3089, interleave = var_4467_interleave_0, values = (var_4443_cast_fp16, var_4445_cast_fp16))[name = tensor("op_4467_cast_fp16")]; + tensor var_4469_interleave_0 = const()[name = tensor("op_4469_interleave_0"), val = tensor(false)]; + tensor var_4469_cast_fp16 = concat(axis = var_3089, interleave = var_4469_interleave_0, values = (var_4447_cast_fp16, var_4449_cast_fp16))[name = tensor("op_4469_cast_fp16")]; + tensor var_4471_interleave_0 = const()[name = tensor("op_4471_interleave_0"), val = tensor(false)]; + tensor var_4471_cast_fp16 = concat(axis = var_3089, interleave = var_4471_interleave_0, values = (var_4451_cast_fp16, var_4453_cast_fp16))[name = tensor("op_4471_cast_fp16")]; + tensor var_4473_interleave_0 = const()[name = tensor("op_4473_interleave_0"), val = tensor(false)]; + tensor var_4473_cast_fp16 = concat(axis = var_3089, interleave = var_4473_interleave_0, values = (var_4455_cast_fp16, var_4457_cast_fp16))[name = tensor("op_4473_cast_fp16")]; + tensor input_107_interleave_0 = const()[name = tensor("input_107_interleave_0"), val = tensor(false)]; + tensor input_107_cast_fp16 = concat(axis = var_3111, interleave = input_107_interleave_0, values = (var_4459_cast_fp16, var_4461_cast_fp16, var_4463_cast_fp16, var_4465_cast_fp16, var_4467_cast_fp16, var_4469_cast_fp16, var_4471_cast_fp16, var_4473_cast_fp16))[name = tensor("input_107_cast_fp16")]; + tensor var_4483_pad_type_0 = const()[name = tensor("op_4483_pad_type_0"), val = tensor("valid")]; + tensor var_4483_strides_0 = const()[name = tensor("op_4483_strides_0"), val = tensor([1, 1])]; + tensor var_4483_pad_0 = const()[name = tensor("op_4483_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4483_dilations_0 = const()[name = tensor("op_4483_dilations_0"), val = tensor([1, 1])]; + tensor var_4483_groups_0 = const()[name = tensor("op_4483_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_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(79965120)))]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_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(80784384)))]; + tensor var_4483_cast_fp16 = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_4483_dilations_0, groups = var_4483_groups_0, pad = var_4483_pad_0, pad_type = var_4483_pad_type_0, strides = var_4483_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_107_cast_fp16)[name = tensor("op_4483_cast_fp16")]; + tensor inputs_23_cast_fp16 = add(x = var_4483_cast_fp16, y = inputs_21_cast_fp16)[name = tensor("inputs_23_cast_fp16")]; + tensor input_109_axes_0 = const()[name = tensor("input_109_axes_0"), val = tensor([1])]; + tensor input_109_gamma_0_to_fp16 = const()[name = tensor("input_109_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80785728)))]; + tensor input_109_beta_0_to_fp16 = const()[name = tensor("input_109_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80787072)))]; + tensor var_4493_to_fp16 = const()[name = tensor("op_4493_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_109_cast_fp16 = layer_norm(axes = input_109_axes_0, beta = input_109_beta_0_to_fp16, epsilon = var_4493_to_fp16, gamma = input_109_gamma_0_to_fp16, x = inputs_23_cast_fp16)[name = tensor("input_109_cast_fp16")]; + tensor var_4513_pad_type_0 = const()[name = tensor("op_4513_pad_type_0"), val = tensor("valid")]; + tensor var_4513_strides_0 = const()[name = tensor("op_4513_strides_0"), val = tensor([1, 1])]; + tensor var_4513_pad_0 = const()[name = tensor("op_4513_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4513_dilations_0 = const()[name = tensor("op_4513_dilations_0"), val = tensor([1, 1])]; + tensor var_4513_groups_0 = const()[name = tensor("op_4513_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_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(80788416)))]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_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(87342080)))]; + tensor var_4513_cast_fp16 = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_4513_dilations_0, groups = var_4513_groups_0, pad = var_4513_pad_0, pad_type = var_4513_pad_type_0, strides = var_4513_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_109_cast_fp16)[name = tensor("op_4513_cast_fp16")]; + tensor var_4514_split_sizes_0 = const()[name = tensor("op_4514_split_sizes_0"), val = tensor([2560, 2560])]; + tensor var_4514_axis_0 = const()[name = tensor("op_4514_axis_0"), val = tensor(1)]; + tensor var_4514_cast_fp16_0, tensor var_4514_cast_fp16_1 = split(axis = var_4514_axis_0, split_sizes = var_4514_split_sizes_0, x = var_4513_cast_fp16)[name = tensor("op_4514_cast_fp16")]; + tensor var_4516_mode_0 = const()[name = tensor("op_4516_mode_0"), val = tensor("EXACT")]; + tensor var_4516_cast_fp16 = gelu(mode = var_4516_mode_0, x = var_4514_cast_fp16_1)[name = tensor("op_4516_cast_fp16")]; + tensor input_111_cast_fp16 = mul(x = var_4514_cast_fp16_0, y = var_4516_cast_fp16)[name = tensor("input_111_cast_fp16")]; + tensor var_4524_pad_type_0 = const()[name = tensor("op_4524_pad_type_0"), val = tensor("valid")]; + tensor var_4524_strides_0 = const()[name = tensor("op_4524_strides_0"), val = tensor([1, 1])]; + tensor var_4524_pad_0 = const()[name = tensor("op_4524_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4524_dilations_0 = const()[name = tensor("op_4524_dilations_0"), val = tensor([1, 1])]; + tensor var_4524_groups_0 = const()[name = tensor("op_4524_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("down_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(87352384)))]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("down_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(90629248)))]; + tensor var_4524_cast_fp16 = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_4524_dilations_0, groups = var_4524_groups_0, pad = var_4524_pad_0, pad_type = var_4524_pad_type_0, strides = var_4524_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_111_cast_fp16)[name = tensor("op_4524_cast_fp16")]; + tensor hidden_states_69_cast_fp16 = add(x = var_4524_cast_fp16, y = inputs_23_cast_fp16)[name = tensor("hidden_states_69_cast_fp16")]; + tensor var_4526 = const()[name = tensor("op_4526"), val = tensor([2, 640, 32, 32])]; + tensor input_113_cast_fp16 = reshape(shape = var_4526, x = hidden_states_69_cast_fp16)[name = tensor("input_113_cast_fp16")]; + tensor hidden_states_71_pad_type_0 = const()[name = tensor("hidden_states_71_pad_type_0"), val = tensor("valid")]; + tensor hidden_states_71_strides_0 = const()[name = tensor("hidden_states_71_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_71_pad_0 = const()[name = tensor("hidden_states_71_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_71_dilations_0 = const()[name = tensor("hidden_states_71_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_71_groups_0 = const()[name = tensor("hidden_states_71_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_proj_out_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90630592)))]; + tensor down_blocks_1_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91449856)))]; + tensor hidden_states_71_cast_fp16 = conv(bias = down_blocks_1_attentions_1_proj_out_bias_to_fp16, dilations = hidden_states_71_dilations_0, groups = hidden_states_71_groups_0, pad = hidden_states_71_pad_0, pad_type = hidden_states_71_pad_type_0, strides = hidden_states_71_strides_0, weight = down_blocks_1_attentions_1_proj_out_weight_to_fp16, x = input_113_cast_fp16)[name = tensor("hidden_states_71_cast_fp16")]; + tensor input_115_cast_fp16_1 = add(x = hidden_states_71_cast_fp16, y = hidden_states_59_cast_fp16)[name = tensor("input_115_cast_fp16")]; + tensor input_117_pad_type_0 = const()[name = tensor("input_117_pad_type_0"), val = tensor("custom")]; + tensor input_117_pad_0 = const()[name = tensor("input_117_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor input_117_strides_0 = const()[name = tensor("input_117_strides_0"), val = tensor([2, 2])]; + tensor input_117_dilations_0 = const()[name = tensor("input_117_dilations_0"), val = tensor([1, 1])]; + tensor input_117_groups_0 = const()[name = tensor("input_117_groups_0"), val = tensor(1)]; + tensor down_blocks_1_downsamplers_0_conv_weight_to_fp16 = const()[name = tensor("down_blocks_1_downsamplers_0_conv_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91451200)))]; + tensor down_blocks_1_downsamplers_0_conv_bias_to_fp16 = const()[name = tensor("down_blocks_1_downsamplers_0_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98824064)))]; + tensor input_117_cast_fp16_1 = conv(bias = down_blocks_1_downsamplers_0_conv_bias_to_fp16, dilations = input_117_dilations_0, groups = input_117_groups_0, pad = input_117_pad_0, pad_type = input_117_pad_type_0, strides = input_117_strides_0, weight = down_blocks_1_downsamplers_0_conv_weight_to_fp16, x = input_115_cast_fp16_1)[name = tensor("input_117_cast_fp16")]; + tensor var_4569 = const()[name = tensor("op_4569"), val = tensor(1)]; + tensor reshape_48_shape_0 = const()[name = tensor("reshape_48_shape_0"), val = tensor([2, 32, 20, 16, 16])]; + tensor reshape_48_cast_fp16 = reshape(shape = reshape_48_shape_0, x = input_117_cast_fp16_1)[name = tensor("reshape_48_cast_fp16")]; + tensor reduce_mean_36_axes_0 = const()[name = tensor("reduce_mean_36_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_36_keep_dims_0 = const()[name = tensor("reduce_mean_36_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_36_cast_fp16 = reduce_mean(axes = reduce_mean_36_axes_0, keep_dims = reduce_mean_36_keep_dims_0, x = reshape_48_cast_fp16)[name = tensor("reduce_mean_36_cast_fp16")]; + tensor sub_24_cast_fp16 = sub(x = reshape_48_cast_fp16, y = reduce_mean_36_cast_fp16)[name = tensor("sub_24_cast_fp16")]; + tensor square_12_cast_fp16 = square(x = sub_24_cast_fp16)[name = tensor("square_12_cast_fp16")]; + tensor reduce_mean_38_axes_0 = const()[name = tensor("reduce_mean_38_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_38_keep_dims_0 = const()[name = tensor("reduce_mean_38_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_38_cast_fp16 = reduce_mean(axes = reduce_mean_38_axes_0, keep_dims = reduce_mean_38_keep_dims_0, x = square_12_cast_fp16)[name = tensor("reduce_mean_38_cast_fp16")]; + tensor add_24_y_0_to_fp16 = const()[name = tensor("add_24_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_24_cast_fp16 = add(x = reduce_mean_38_cast_fp16, y = add_24_y_0_to_fp16)[name = tensor("add_24_cast_fp16")]; + tensor sqrt_12_cast_fp16 = sqrt(x = add_24_cast_fp16)[name = tensor("sqrt_12_cast_fp16")]; + tensor real_div_12_cast_fp16 = real_div(x = sub_24_cast_fp16, y = sqrt_12_cast_fp16)[name = tensor("real_div_12_cast_fp16")]; + tensor reshape_49_shape_0 = const()[name = tensor("reshape_49_shape_0"), val = tensor([2, 640, 16, 16])]; + tensor reshape_49_cast_fp16 = reshape(shape = reshape_49_shape_0, x = real_div_12_cast_fp16)[name = tensor("reshape_49_cast_fp16")]; + tensor add_25_gamma_0_to_fp16 = const()[name = tensor("add_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98825408)))]; + tensor add_25_beta_0_to_fp16 = const()[name = tensor("add_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98826752)))]; + tensor add_25_epsilon_0_to_fp16 = const()[name = tensor("add_25_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_25_cast_fp16 = batch_norm(beta = add_25_beta_0_to_fp16, epsilon = add_25_epsilon_0_to_fp16, gamma = add_25_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_49_cast_fp16)[name = tensor("add_25_cast_fp16")]; + tensor input_121_cast_fp16 = silu(x = add_25_cast_fp16)[name = tensor("input_121_cast_fp16")]; + tensor hidden_states_73_pad_type_0 = const()[name = tensor("hidden_states_73_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_73_pad_0 = const()[name = tensor("hidden_states_73_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_73_strides_0 = const()[name = tensor("hidden_states_73_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_73_dilations_0 = const()[name = tensor("hidden_states_73_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_73_groups_0 = const()[name = tensor("hidden_states_73_groups_0"), val = tensor(1)]; + tensor down_blocks_2_resnets_0_conv1_weight_to_fp16 = const()[name = tensor("down_blocks_2_resnets_0_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98828096)))]; + tensor down_blocks_2_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113573760)))]; + tensor hidden_states_73_cast_fp16 = conv(bias = down_blocks_2_resnets_0_conv1_bias_to_fp16, dilations = hidden_states_73_dilations_0, groups = hidden_states_73_groups_0, pad = hidden_states_73_pad_0, pad_type = hidden_states_73_pad_type_0, strides = hidden_states_73_strides_0, weight = down_blocks_2_resnets_0_conv1_weight_to_fp16, x = input_121_cast_fp16)[name = tensor("hidden_states_73_cast_fp16")]; + tensor temb_9_pad_type_0 = const()[name = tensor("temb_9_pad_type_0"), val = tensor("valid")]; + tensor temb_9_strides_0 = const()[name = tensor("temb_9_strides_0"), val = tensor([1, 1])]; + tensor temb_9_pad_0 = const()[name = tensor("temb_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor temb_9_dilations_0 = const()[name = tensor("temb_9_dilations_0"), val = tensor([1, 1])]; + tensor temb_9_groups_0 = const()[name = tensor("temb_9_groups_0"), val = tensor(1)]; + tensor down_blocks_2_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor("down_blocks_2_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113576384)))]; + tensor down_blocks_2_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116853248)))]; + tensor temb_9_cast_fp16 = conv(bias = down_blocks_2_resnets_0_time_emb_proj_bias_to_fp16, dilations = temb_9_dilations_0, groups = temb_9_groups_0, pad = temb_9_pad_0, pad_type = temb_9_pad_type_0, strides = temb_9_strides_0, weight = down_blocks_2_resnets_0_time_emb_proj_weight_to_fp16, x = input_15_cast_fp16_1)[name = tensor("temb_9_cast_fp16")]; + tensor input_125_cast_fp16 = add(x = hidden_states_73_cast_fp16, y = temb_9_cast_fp16)[name = tensor("input_125_cast_fp16")]; + tensor reshape_52_shape_0 = const()[name = tensor("reshape_52_shape_0"), val = tensor([2, 32, 40, 16, 16])]; + tensor reshape_52_cast_fp16 = reshape(shape = reshape_52_shape_0, x = input_125_cast_fp16)[name = tensor("reshape_52_cast_fp16")]; + tensor reduce_mean_39_axes_0 = const()[name = tensor("reduce_mean_39_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_39_keep_dims_0 = const()[name = tensor("reduce_mean_39_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_39_cast_fp16 = reduce_mean(axes = reduce_mean_39_axes_0, keep_dims = reduce_mean_39_keep_dims_0, x = reshape_52_cast_fp16)[name = tensor("reduce_mean_39_cast_fp16")]; + tensor sub_26_cast_fp16 = sub(x = reshape_52_cast_fp16, y = reduce_mean_39_cast_fp16)[name = tensor("sub_26_cast_fp16")]; + tensor square_13_cast_fp16 = square(x = sub_26_cast_fp16)[name = tensor("square_13_cast_fp16")]; + tensor reduce_mean_41_axes_0 = const()[name = tensor("reduce_mean_41_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_41_keep_dims_0 = const()[name = tensor("reduce_mean_41_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_41_cast_fp16 = reduce_mean(axes = reduce_mean_41_axes_0, keep_dims = reduce_mean_41_keep_dims_0, x = square_13_cast_fp16)[name = tensor("reduce_mean_41_cast_fp16")]; + tensor add_26_y_0_to_fp16 = const()[name = tensor("add_26_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_26_cast_fp16 = add(x = reduce_mean_41_cast_fp16, y = add_26_y_0_to_fp16)[name = tensor("add_26_cast_fp16")]; + tensor sqrt_13_cast_fp16 = sqrt(x = add_26_cast_fp16)[name = tensor("sqrt_13_cast_fp16")]; + tensor real_div_13_cast_fp16 = real_div(x = sub_26_cast_fp16, y = sqrt_13_cast_fp16)[name = tensor("real_div_13_cast_fp16")]; + tensor reshape_53_shape_0 = const()[name = tensor("reshape_53_shape_0"), val = tensor([2, 1280, 16, 16])]; + tensor reshape_53_cast_fp16 = reshape(shape = reshape_53_shape_0, x = real_div_13_cast_fp16)[name = tensor("reshape_53_cast_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(116855872)))]; + 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(116858496)))]; + tensor add_27_gamma_0_to_fp16 = const()[name = tensor("add_27_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116861120)))]; + tensor add_27_beta_0_to_fp16 = const()[name = tensor("add_27_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116863744)))]; + tensor add_27_epsilon_0_to_fp16 = const()[name = tensor("add_27_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_27_cast_fp16 = batch_norm(beta = add_27_beta_0_to_fp16, epsilon = add_27_epsilon_0_to_fp16, gamma = add_27_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_53_cast_fp16)[name = tensor("add_27_cast_fp16")]; + tensor input_129_cast_fp16 = silu(x = add_27_cast_fp16)[name = tensor("input_129_cast_fp16")]; + tensor hidden_states_75_pad_type_0 = const()[name = tensor("hidden_states_75_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_75_pad_0 = const()[name = tensor("hidden_states_75_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_75_strides_0 = const()[name = tensor("hidden_states_75_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_75_dilations_0 = const()[name = tensor("hidden_states_75_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_75_groups_0 = const()[name = tensor("hidden_states_75_groups_0"), val = tensor(1)]; + tensor down_blocks_2_resnets_0_conv2_weight_to_fp16 = const()[name = tensor("down_blocks_2_resnets_0_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116866368)))]; + tensor down_blocks_2_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(146357632)))]; + tensor hidden_states_75_cast_fp16 = conv(bias = down_blocks_2_resnets_0_conv2_bias_to_fp16, dilations = hidden_states_75_dilations_0, groups = hidden_states_75_groups_0, pad = hidden_states_75_pad_0, pad_type = hidden_states_75_pad_type_0, strides = hidden_states_75_strides_0, weight = down_blocks_2_resnets_0_conv2_weight_to_fp16, x = input_129_cast_fp16)[name = tensor("hidden_states_75_cast_fp16")]; + tensor x_3_pad_type_0 = const()[name = tensor("x_3_pad_type_0"), val = tensor("valid")]; + tensor x_3_strides_0 = const()[name = tensor("x_3_strides_0"), val = tensor([1, 1])]; + tensor x_3_pad_0 = const()[name = tensor("x_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_3_dilations_0 = const()[name = tensor("x_3_dilations_0"), val = tensor([1, 1])]; + tensor x_3_groups_0 = const()[name = tensor("x_3_groups_0"), val = tensor(1)]; + tensor down_blocks_2_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = tensor("down_blocks_2_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(146360256)))]; + tensor down_blocks_2_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(147998720)))]; + tensor x_3_cast_fp16 = conv(bias = down_blocks_2_resnets_0_conv_shortcut_bias_to_fp16, dilations = x_3_dilations_0, groups = x_3_groups_0, pad = x_3_pad_0, pad_type = x_3_pad_type_0, strides = x_3_strides_0, weight = down_blocks_2_resnets_0_conv_shortcut_weight_to_fp16, x = input_117_cast_fp16_1)[name = tensor("x_3_cast_fp16")]; + tensor hidden_states_77_cast_fp16 = add(x = x_3_cast_fp16, y = hidden_states_75_cast_fp16)[name = tensor("hidden_states_77_cast_fp16")]; + tensor reshape_56_shape_0 = const()[name = tensor("reshape_56_shape_0"), val = tensor([2, 32, 40, 16, 16])]; + tensor reshape_56_cast_fp16 = reshape(shape = reshape_56_shape_0, x = hidden_states_77_cast_fp16)[name = tensor("reshape_56_cast_fp16")]; + tensor reduce_mean_42_axes_0 = const()[name = tensor("reduce_mean_42_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_42_keep_dims_0 = const()[name = tensor("reduce_mean_42_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_42_cast_fp16 = reduce_mean(axes = reduce_mean_42_axes_0, keep_dims = reduce_mean_42_keep_dims_0, x = reshape_56_cast_fp16)[name = tensor("reduce_mean_42_cast_fp16")]; + tensor sub_28_cast_fp16 = sub(x = reshape_56_cast_fp16, y = reduce_mean_42_cast_fp16)[name = tensor("sub_28_cast_fp16")]; + tensor square_14_cast_fp16 = square(x = sub_28_cast_fp16)[name = tensor("square_14_cast_fp16")]; + tensor reduce_mean_44_axes_0 = const()[name = tensor("reduce_mean_44_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_44_keep_dims_0 = const()[name = tensor("reduce_mean_44_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_44_cast_fp16 = reduce_mean(axes = reduce_mean_44_axes_0, keep_dims = reduce_mean_44_keep_dims_0, x = square_14_cast_fp16)[name = tensor("reduce_mean_44_cast_fp16")]; + tensor add_28_y_0_to_fp16 = const()[name = tensor("add_28_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_28_cast_fp16 = add(x = reduce_mean_44_cast_fp16, y = add_28_y_0_to_fp16)[name = tensor("add_28_cast_fp16")]; + tensor sqrt_14_cast_fp16 = sqrt(x = add_28_cast_fp16)[name = tensor("sqrt_14_cast_fp16")]; + tensor real_div_14_cast_fp16 = real_div(x = sub_28_cast_fp16, y = sqrt_14_cast_fp16)[name = tensor("real_div_14_cast_fp16")]; + tensor reshape_57_shape_0 = const()[name = tensor("reshape_57_shape_0"), val = tensor([2, 1280, 16, 16])]; + tensor reshape_57_cast_fp16 = reshape(shape = reshape_57_shape_0, x = real_div_14_cast_fp16)[name = tensor("reshape_57_cast_fp16")]; + tensor add_29_gamma_0_to_fp16 = const()[name = tensor("add_29_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(148001344)))]; + tensor add_29_beta_0_to_fp16 = const()[name = tensor("add_29_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(148003968)))]; + tensor add_29_epsilon_0_to_fp16 = const()[name = tensor("add_29_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_29_cast_fp16 = batch_norm(beta = add_29_beta_0_to_fp16, epsilon = add_29_epsilon_0_to_fp16, gamma = add_29_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_57_cast_fp16)[name = tensor("add_29_cast_fp16")]; + tensor hidden_states_79_pad_type_0 = const()[name = tensor("hidden_states_79_pad_type_0"), val = tensor("valid")]; + tensor hidden_states_79_strides_0 = const()[name = tensor("hidden_states_79_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_79_pad_0 = const()[name = tensor("hidden_states_79_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_79_dilations_0 = const()[name = tensor("hidden_states_79_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_79_groups_0 = const()[name = tensor("hidden_states_79_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(148006592)))]; + tensor down_blocks_2_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151283456)))]; + tensor hidden_states_79_cast_fp16 = conv(bias = down_blocks_2_attentions_0_proj_in_bias_to_fp16, dilations = hidden_states_79_dilations_0, groups = hidden_states_79_groups_0, pad = hidden_states_79_pad_0, pad_type = hidden_states_79_pad_type_0, strides = hidden_states_79_strides_0, weight = down_blocks_2_attentions_0_proj_in_weight_to_fp16, x = add_29_cast_fp16)[name = tensor("hidden_states_79_cast_fp16")]; + tensor var_4648 = const()[name = tensor("op_4648"), val = tensor([2, 1280, 1, 256])]; + tensor inputs_25_cast_fp16 = reshape(shape = var_4648, x = hidden_states_79_cast_fp16)[name = tensor("inputs_25_cast_fp16")]; + tensor hidden_states_81_axes_0 = const()[name = tensor("hidden_states_81_axes_0"), val = tensor([1])]; + tensor hidden_states_81_gamma_0_to_fp16 = const()[name = tensor("hidden_states_81_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151286080)))]; + tensor hidden_states_81_beta_0_to_fp16 = const()[name = tensor("hidden_states_81_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151288704)))]; + tensor var_4664_to_fp16 = const()[name = tensor("op_4664_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_81_cast_fp16 = layer_norm(axes = hidden_states_81_axes_0, beta = hidden_states_81_beta_0_to_fp16, epsilon = var_4664_to_fp16, gamma = hidden_states_81_gamma_0_to_fp16, x = inputs_25_cast_fp16)[name = tensor("hidden_states_81_cast_fp16")]; + tensor q_17_pad_type_0 = const()[name = tensor("q_17_pad_type_0"), val = tensor("valid")]; + tensor q_17_strides_0 = const()[name = tensor("q_17_strides_0"), val = tensor([1, 1])]; + tensor q_17_pad_0 = const()[name = tensor("q_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_17_dilations_0 = const()[name = tensor("q_17_dilations_0"), val = tensor([1, 1])]; + tensor q_17_groups_0 = const()[name = tensor("q_17_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_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(151291328)))]; + tensor q_17_cast_fp16 = conv(dilations = q_17_dilations_0, groups = q_17_groups_0, pad = q_17_pad_0, pad_type = q_17_pad_type_0, strides = q_17_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_81_cast_fp16)[name = tensor("q_17_cast_fp16")]; + tensor k_33_pad_type_0 = const()[name = tensor("k_33_pad_type_0"), val = tensor("valid")]; + tensor k_33_strides_0 = const()[name = tensor("k_33_strides_0"), val = tensor([1, 1])]; + tensor k_33_pad_0 = const()[name = tensor("k_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_33_dilations_0 = const()[name = tensor("k_33_dilations_0"), val = tensor([1, 1])]; + tensor k_33_groups_0 = const()[name = tensor("k_33_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_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(154568192)))]; + tensor k_33_cast_fp16 = conv(dilations = k_33_dilations_0, groups = k_33_groups_0, pad = k_33_pad_0, pad_type = k_33_pad_type_0, strides = k_33_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_81_cast_fp16)[name = tensor("k_33_cast_fp16")]; + tensor v_17_pad_type_0 = const()[name = tensor("v_17_pad_type_0"), val = tensor("valid")]; + tensor v_17_strides_0 = const()[name = tensor("v_17_strides_0"), val = tensor([1, 1])]; + tensor v_17_pad_0 = const()[name = tensor("v_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_17_dilations_0 = const()[name = tensor("v_17_dilations_0"), val = tensor([1, 1])]; + tensor v_17_groups_0 = const()[name = tensor("v_17_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_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(157845056)))]; + tensor v_17_cast_fp16 = conv(dilations = v_17_dilations_0, groups = v_17_groups_0, pad = v_17_pad_0, pad_type = v_17_pad_type_0, strides = v_17_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_81_cast_fp16)[name = tensor("v_17_cast_fp16")]; + tensor var_4697_begin_0 = const()[name = tensor("op_4697_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4697_end_0 = const()[name = tensor("op_4697_end_0"), val = tensor([2, 160, 1, 256])]; + tensor var_4697_end_mask_0 = const()[name = tensor("op_4697_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4697_cast_fp16 = slice_by_index(begin = var_4697_begin_0, end = var_4697_end_0, end_mask = var_4697_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_4697_cast_fp16")]; + tensor var_4701_begin_0 = const()[name = tensor("op_4701_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_4701_end_0 = const()[name = tensor("op_4701_end_0"), val = tensor([2, 320, 1, 256])]; + tensor var_4701_end_mask_0 = const()[name = tensor("op_4701_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4701_cast_fp16 = slice_by_index(begin = var_4701_begin_0, end = var_4701_end_0, end_mask = var_4701_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_4701_cast_fp16")]; + tensor var_4705_begin_0 = const()[name = tensor("op_4705_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4705_end_0 = const()[name = tensor("op_4705_end_0"), val = tensor([2, 480, 1, 256])]; + tensor var_4705_end_mask_0 = const()[name = tensor("op_4705_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4705_cast_fp16 = slice_by_index(begin = var_4705_begin_0, end = var_4705_end_0, end_mask = var_4705_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_4705_cast_fp16")]; + tensor var_4709_begin_0 = const()[name = tensor("op_4709_begin_0"), val = tensor([0, 480, 0, 0])]; + tensor var_4709_end_0 = const()[name = tensor("op_4709_end_0"), val = tensor([2, 640, 1, 256])]; + tensor var_4709_end_mask_0 = const()[name = tensor("op_4709_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4709_cast_fp16 = slice_by_index(begin = var_4709_begin_0, end = var_4709_end_0, end_mask = var_4709_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_4709_cast_fp16")]; + tensor var_4713_begin_0 = const()[name = tensor("op_4713_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_4713_end_0 = const()[name = tensor("op_4713_end_0"), val = tensor([2, 800, 1, 256])]; + tensor var_4713_end_mask_0 = const()[name = tensor("op_4713_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4713_cast_fp16 = slice_by_index(begin = var_4713_begin_0, end = var_4713_end_0, end_mask = var_4713_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_4713_cast_fp16")]; + tensor var_4717_begin_0 = const()[name = tensor("op_4717_begin_0"), val = tensor([0, 800, 0, 0])]; + tensor var_4717_end_0 = const()[name = tensor("op_4717_end_0"), val = tensor([2, 960, 1, 256])]; + tensor var_4717_end_mask_0 = const()[name = tensor("op_4717_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4717_cast_fp16 = slice_by_index(begin = var_4717_begin_0, end = var_4717_end_0, end_mask = var_4717_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_4717_cast_fp16")]; + tensor var_4721_begin_0 = const()[name = tensor("op_4721_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_4721_end_0 = const()[name = tensor("op_4721_end_0"), val = tensor([2, 1120, 1, 256])]; + tensor var_4721_end_mask_0 = const()[name = tensor("op_4721_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4721_cast_fp16 = slice_by_index(begin = var_4721_begin_0, end = var_4721_end_0, end_mask = var_4721_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_4721_cast_fp16")]; + tensor var_4725_begin_0 = const()[name = tensor("op_4725_begin_0"), val = tensor([0, 1120, 0, 0])]; + tensor var_4725_end_0 = const()[name = tensor("op_4725_end_0"), val = tensor([2, 1, 1, 256])]; + tensor var_4725_end_mask_0 = const()[name = tensor("op_4725_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4725_cast_fp16 = slice_by_index(begin = var_4725_begin_0, end = var_4725_end_0, end_mask = var_4725_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_4725_cast_fp16")]; + tensor k_35_perm_0 = const()[name = tensor("k_35_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_4732_begin_0 = const()[name = tensor("op_4732_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4732_end_0 = const()[name = tensor("op_4732_end_0"), val = tensor([2, 256, 1, 160])]; + tensor var_4732_end_mask_0 = const()[name = tensor("op_4732_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_35_cast_fp16 = transpose(perm = k_35_perm_0, x = k_33_cast_fp16)[name = tensor("transpose_5")]; + tensor var_4732_cast_fp16 = slice_by_index(begin = var_4732_begin_0, end = var_4732_end_0, end_mask = var_4732_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_4732_cast_fp16")]; + tensor var_4736_begin_0 = const()[name = tensor("op_4736_begin_0"), val = tensor([0, 0, 0, 160])]; + tensor var_4736_end_0 = const()[name = tensor("op_4736_end_0"), val = tensor([2, 256, 1, 320])]; + tensor var_4736_end_mask_0 = const()[name = tensor("op_4736_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4736_cast_fp16 = slice_by_index(begin = var_4736_begin_0, end = var_4736_end_0, end_mask = var_4736_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_4736_cast_fp16")]; + tensor var_4740_begin_0 = const()[name = tensor("op_4740_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_4740_end_0 = const()[name = tensor("op_4740_end_0"), val = tensor([2, 256, 1, 480])]; + tensor var_4740_end_mask_0 = const()[name = tensor("op_4740_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4740_cast_fp16 = slice_by_index(begin = var_4740_begin_0, end = var_4740_end_0, end_mask = var_4740_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_4740_cast_fp16")]; + tensor var_4744_begin_0 = const()[name = tensor("op_4744_begin_0"), val = tensor([0, 0, 0, 480])]; + tensor var_4744_end_0 = const()[name = tensor("op_4744_end_0"), val = tensor([2, 256, 1, 640])]; + tensor var_4744_end_mask_0 = const()[name = tensor("op_4744_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4744_cast_fp16 = slice_by_index(begin = var_4744_begin_0, end = var_4744_end_0, end_mask = var_4744_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_4744_cast_fp16")]; + tensor var_4748_begin_0 = const()[name = tensor("op_4748_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_4748_end_0 = const()[name = tensor("op_4748_end_0"), val = tensor([2, 256, 1, 800])]; + tensor var_4748_end_mask_0 = const()[name = tensor("op_4748_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4748_cast_fp16 = slice_by_index(begin = var_4748_begin_0, end = var_4748_end_0, end_mask = var_4748_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_4748_cast_fp16")]; + tensor var_4752_begin_0 = const()[name = tensor("op_4752_begin_0"), val = tensor([0, 0, 0, 800])]; + tensor var_4752_end_0 = const()[name = tensor("op_4752_end_0"), val = tensor([2, 256, 1, 960])]; + tensor var_4752_end_mask_0 = const()[name = tensor("op_4752_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4752_cast_fp16 = slice_by_index(begin = var_4752_begin_0, end = var_4752_end_0, end_mask = var_4752_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_4752_cast_fp16")]; + tensor var_4756_begin_0 = const()[name = tensor("op_4756_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_4756_end_0 = const()[name = tensor("op_4756_end_0"), val = tensor([2, 256, 1, 1120])]; + tensor var_4756_end_mask_0 = const()[name = tensor("op_4756_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4756_cast_fp16 = slice_by_index(begin = var_4756_begin_0, end = var_4756_end_0, end_mask = var_4756_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_4756_cast_fp16")]; + tensor var_4760_begin_0 = const()[name = tensor("op_4760_begin_0"), val = tensor([0, 0, 0, 1120])]; + tensor var_4760_end_0 = const()[name = tensor("op_4760_end_0"), val = tensor([2, 256, 1, 1])]; + tensor var_4760_end_mask_0 = const()[name = tensor("op_4760_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4760_cast_fp16 = slice_by_index(begin = var_4760_begin_0, end = var_4760_end_0, end_mask = var_4760_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_4760_cast_fp16")]; + tensor var_4762_begin_0 = const()[name = tensor("op_4762_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4762_end_0 = const()[name = tensor("op_4762_end_0"), val = tensor([2, 160, 1, 256])]; + tensor var_4762_end_mask_0 = const()[name = tensor("op_4762_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4762_cast_fp16 = slice_by_index(begin = var_4762_begin_0, end = var_4762_end_0, end_mask = var_4762_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_4762_cast_fp16")]; + tensor var_4766_begin_0 = const()[name = tensor("op_4766_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_4766_end_0 = const()[name = tensor("op_4766_end_0"), val = tensor([2, 320, 1, 256])]; + tensor var_4766_end_mask_0 = const()[name = tensor("op_4766_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4766_cast_fp16 = slice_by_index(begin = var_4766_begin_0, end = var_4766_end_0, end_mask = var_4766_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_4766_cast_fp16")]; + tensor var_4770_begin_0 = const()[name = tensor("op_4770_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4770_end_0 = const()[name = tensor("op_4770_end_0"), val = tensor([2, 480, 1, 256])]; + tensor var_4770_end_mask_0 = const()[name = tensor("op_4770_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4770_cast_fp16 = slice_by_index(begin = var_4770_begin_0, end = var_4770_end_0, end_mask = var_4770_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_4770_cast_fp16")]; + tensor var_4774_begin_0 = const()[name = tensor("op_4774_begin_0"), val = tensor([0, 480, 0, 0])]; + tensor var_4774_end_0 = const()[name = tensor("op_4774_end_0"), val = tensor([2, 640, 1, 256])]; + tensor var_4774_end_mask_0 = const()[name = tensor("op_4774_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4774_cast_fp16 = slice_by_index(begin = var_4774_begin_0, end = var_4774_end_0, end_mask = var_4774_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_4774_cast_fp16")]; + tensor var_4778_begin_0 = const()[name = tensor("op_4778_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_4778_end_0 = const()[name = tensor("op_4778_end_0"), val = tensor([2, 800, 1, 256])]; + tensor var_4778_end_mask_0 = const()[name = tensor("op_4778_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4778_cast_fp16 = slice_by_index(begin = var_4778_begin_0, end = var_4778_end_0, end_mask = var_4778_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_4778_cast_fp16")]; + tensor var_4782_begin_0 = const()[name = tensor("op_4782_begin_0"), val = tensor([0, 800, 0, 0])]; + tensor var_4782_end_0 = const()[name = tensor("op_4782_end_0"), val = tensor([2, 960, 1, 256])]; + tensor var_4782_end_mask_0 = const()[name = tensor("op_4782_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4782_cast_fp16 = slice_by_index(begin = var_4782_begin_0, end = var_4782_end_0, end_mask = var_4782_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_4782_cast_fp16")]; + tensor var_4786_begin_0 = const()[name = tensor("op_4786_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_4786_end_0 = const()[name = tensor("op_4786_end_0"), val = tensor([2, 1120, 1, 256])]; + tensor var_4786_end_mask_0 = const()[name = tensor("op_4786_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4786_cast_fp16 = slice_by_index(begin = var_4786_begin_0, end = var_4786_end_0, end_mask = var_4786_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_4786_cast_fp16")]; + tensor var_4790_begin_0 = const()[name = tensor("op_4790_begin_0"), val = tensor([0, 1120, 0, 0])]; + tensor var_4790_end_0 = const()[name = tensor("op_4790_end_0"), val = tensor([2, 1, 1, 256])]; + tensor var_4790_end_mask_0 = const()[name = tensor("op_4790_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4790_cast_fp16 = slice_by_index(begin = var_4790_begin_0, end = var_4790_end_0, end_mask = var_4790_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_4790_cast_fp16")]; + tensor var_4794_equation_0 = const()[name = tensor("op_4794_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4794_cast_fp16 = einsum(equation = var_4794_equation_0, values = (var_4732_cast_fp16, var_4697_cast_fp16))[name = tensor("op_4794_cast_fp16")]; + tensor var_4795_to_fp16 = const()[name = tensor("op_4795_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_1_cast_fp16 = mul(x = var_4794_cast_fp16, y = var_4795_to_fp16)[name = tensor("aw_1_cast_fp16")]; + tensor var_4798_equation_0 = const()[name = tensor("op_4798_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4798_cast_fp16 = einsum(equation = var_4798_equation_0, values = (var_4736_cast_fp16, var_4701_cast_fp16))[name = tensor("op_4798_cast_fp16")]; + tensor var_4799_to_fp16 = const()[name = tensor("op_4799_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_3_cast_fp16 = mul(x = var_4798_cast_fp16, y = var_4799_to_fp16)[name = tensor("aw_3_cast_fp16")]; + tensor var_4802_equation_0 = const()[name = tensor("op_4802_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4802_cast_fp16 = einsum(equation = var_4802_equation_0, values = (var_4740_cast_fp16, var_4705_cast_fp16))[name = tensor("op_4802_cast_fp16")]; + tensor var_4803_to_fp16 = const()[name = tensor("op_4803_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_5_cast_fp16 = mul(x = var_4802_cast_fp16, y = var_4803_to_fp16)[name = tensor("aw_5_cast_fp16")]; + tensor var_4806_equation_0 = const()[name = tensor("op_4806_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4806_cast_fp16 = einsum(equation = var_4806_equation_0, values = (var_4744_cast_fp16, var_4709_cast_fp16))[name = tensor("op_4806_cast_fp16")]; + tensor var_4807_to_fp16 = const()[name = tensor("op_4807_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_7_cast_fp16 = mul(x = var_4806_cast_fp16, y = var_4807_to_fp16)[name = tensor("aw_7_cast_fp16")]; + tensor var_4810_equation_0 = const()[name = tensor("op_4810_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4810_cast_fp16 = einsum(equation = var_4810_equation_0, values = (var_4748_cast_fp16, var_4713_cast_fp16))[name = tensor("op_4810_cast_fp16")]; + tensor var_4811_to_fp16 = const()[name = tensor("op_4811_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_9_cast_fp16 = mul(x = var_4810_cast_fp16, y = var_4811_to_fp16)[name = tensor("aw_9_cast_fp16")]; + tensor var_4814_equation_0 = const()[name = tensor("op_4814_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4814_cast_fp16 = einsum(equation = var_4814_equation_0, values = (var_4752_cast_fp16, var_4717_cast_fp16))[name = tensor("op_4814_cast_fp16")]; + tensor var_4815_to_fp16 = const()[name = tensor("op_4815_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_11_cast_fp16 = mul(x = var_4814_cast_fp16, y = var_4815_to_fp16)[name = tensor("aw_11_cast_fp16")]; + tensor var_4818_equation_0 = const()[name = tensor("op_4818_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4818_cast_fp16 = einsum(equation = var_4818_equation_0, values = (var_4756_cast_fp16, var_4721_cast_fp16))[name = tensor("op_4818_cast_fp16")]; + tensor var_4819_to_fp16 = const()[name = tensor("op_4819_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_13_cast_fp16 = mul(x = var_4818_cast_fp16, y = var_4819_to_fp16)[name = tensor("aw_13_cast_fp16")]; + tensor var_4822_equation_0 = const()[name = tensor("op_4822_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4822_cast_fp16 = einsum(equation = var_4822_equation_0, values = (var_4760_cast_fp16, var_4725_cast_fp16))[name = tensor("op_4822_cast_fp16")]; + tensor var_4823_to_fp16 = const()[name = tensor("op_4823_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_15_cast_fp16 = mul(x = var_4822_cast_fp16, y = var_4823_to_fp16)[name = tensor("aw_15_cast_fp16")]; + tensor var_4825_cast_fp16 = softmax(axis = var_4569, x = aw_1_cast_fp16)[name = tensor("op_4825_cast_fp16")]; + tensor var_4826_cast_fp16 = softmax(axis = var_4569, x = aw_3_cast_fp16)[name = tensor("op_4826_cast_fp16")]; + tensor var_4827_cast_fp16 = softmax(axis = var_4569, x = aw_5_cast_fp16)[name = tensor("op_4827_cast_fp16")]; + tensor var_4828_cast_fp16 = softmax(axis = var_4569, x = aw_7_cast_fp16)[name = tensor("op_4828_cast_fp16")]; + tensor var_4829_cast_fp16 = softmax(axis = var_4569, x = aw_9_cast_fp16)[name = tensor("op_4829_cast_fp16")]; + tensor var_4830_cast_fp16 = softmax(axis = var_4569, x = aw_11_cast_fp16)[name = tensor("op_4830_cast_fp16")]; + tensor var_4831_cast_fp16 = softmax(axis = var_4569, x = aw_13_cast_fp16)[name = tensor("op_4831_cast_fp16")]; + tensor var_4832_cast_fp16 = softmax(axis = var_4569, x = aw_15_cast_fp16)[name = tensor("op_4832_cast_fp16")]; + tensor var_4834_equation_0 = const()[name = tensor("op_4834_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4834_cast_fp16 = einsum(equation = var_4834_equation_0, values = (var_4762_cast_fp16, var_4825_cast_fp16))[name = tensor("op_4834_cast_fp16")]; + tensor var_4836_equation_0 = const()[name = tensor("op_4836_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4836_cast_fp16 = einsum(equation = var_4836_equation_0, values = (var_4766_cast_fp16, var_4826_cast_fp16))[name = tensor("op_4836_cast_fp16")]; + tensor var_4838_equation_0 = const()[name = tensor("op_4838_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4838_cast_fp16 = einsum(equation = var_4838_equation_0, values = (var_4770_cast_fp16, var_4827_cast_fp16))[name = tensor("op_4838_cast_fp16")]; + tensor var_4840_equation_0 = const()[name = tensor("op_4840_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4840_cast_fp16 = einsum(equation = var_4840_equation_0, values = (var_4774_cast_fp16, var_4828_cast_fp16))[name = tensor("op_4840_cast_fp16")]; + tensor var_4842_equation_0 = const()[name = tensor("op_4842_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4842_cast_fp16 = einsum(equation = var_4842_equation_0, values = (var_4778_cast_fp16, var_4829_cast_fp16))[name = tensor("op_4842_cast_fp16")]; + tensor var_4844_equation_0 = const()[name = tensor("op_4844_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4844_cast_fp16 = einsum(equation = var_4844_equation_0, values = (var_4782_cast_fp16, var_4830_cast_fp16))[name = tensor("op_4844_cast_fp16")]; + tensor var_4846_equation_0 = const()[name = tensor("op_4846_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4846_cast_fp16 = einsum(equation = var_4846_equation_0, values = (var_4786_cast_fp16, var_4831_cast_fp16))[name = tensor("op_4846_cast_fp16")]; + tensor var_4848_equation_0 = const()[name = tensor("op_4848_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4848_cast_fp16 = einsum(equation = var_4848_equation_0, values = (var_4790_cast_fp16, var_4832_cast_fp16))[name = tensor("op_4848_cast_fp16")]; + tensor input_133_interleave_0 = const()[name = tensor("input_133_interleave_0"), val = tensor(false)]; + tensor input_133_cast_fp16 = concat(axis = var_4569, interleave = input_133_interleave_0, values = (var_4834_cast_fp16, var_4836_cast_fp16, var_4838_cast_fp16, var_4840_cast_fp16, var_4842_cast_fp16, var_4844_cast_fp16, var_4846_cast_fp16, var_4848_cast_fp16))[name = tensor("input_133_cast_fp16")]; + tensor var_4858_pad_type_0 = const()[name = tensor("op_4858_pad_type_0"), val = tensor("valid")]; + tensor var_4858_strides_0 = const()[name = tensor("op_4858_strides_0"), val = tensor([1, 1])]; + tensor var_4858_pad_0 = const()[name = tensor("op_4858_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4858_dilations_0 = const()[name = tensor("op_4858_dilations_0"), val = tensor([1, 1])]; + tensor var_4858_groups_0 = const()[name = tensor("op_4858_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_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(161121920)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_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(164398784)))]; + tensor var_4858_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_4858_dilations_0, groups = var_4858_groups_0, pad = var_4858_pad_0, pad_type = var_4858_pad_type_0, strides = var_4858_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_133_cast_fp16)[name = tensor("op_4858_cast_fp16")]; + tensor inputs_27_cast_fp16 = add(x = var_4858_cast_fp16, y = inputs_25_cast_fp16)[name = tensor("inputs_27_cast_fp16")]; + tensor hidden_states_83_axes_0 = const()[name = tensor("hidden_states_83_axes_0"), val = tensor([1])]; + tensor hidden_states_83_gamma_0_to_fp16 = const()[name = tensor("hidden_states_83_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(164401408)))]; + tensor hidden_states_83_beta_0_to_fp16 = const()[name = tensor("hidden_states_83_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(164404032)))]; + tensor var_4868_to_fp16 = const()[name = tensor("op_4868_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_83_cast_fp16 = layer_norm(axes = hidden_states_83_axes_0, beta = hidden_states_83_beta_0_to_fp16, epsilon = var_4868_to_fp16, gamma = hidden_states_83_gamma_0_to_fp16, x = inputs_27_cast_fp16)[name = tensor("hidden_states_83_cast_fp16")]; + tensor q_19_pad_type_0 = const()[name = tensor("q_19_pad_type_0"), val = tensor("valid")]; + tensor q_19_strides_0 = const()[name = tensor("q_19_strides_0"), val = tensor([1, 1])]; + tensor q_19_pad_0 = const()[name = tensor("q_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_19_dilations_0 = const()[name = tensor("q_19_dilations_0"), val = tensor([1, 1])]; + tensor q_19_groups_0 = const()[name = tensor("q_19_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_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(164406656)))]; + tensor q_19_cast_fp16 = conv(dilations = q_19_dilations_0, groups = q_19_groups_0, pad = q_19_pad_0, pad_type = q_19_pad_type_0, strides = q_19_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_83_cast_fp16)[name = tensor("q_19_cast_fp16")]; + tensor k_37_pad_type_0 = const()[name = tensor("k_37_pad_type_0"), val = tensor("valid")]; + tensor k_37_strides_0 = const()[name = tensor("k_37_strides_0"), val = tensor([1, 1])]; + tensor k_37_pad_0 = const()[name = tensor("k_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_37_dilations_0 = const()[name = tensor("k_37_dilations_0"), val = tensor([1, 1])]; + tensor k_37_groups_0 = const()[name = tensor("k_37_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_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(167683520)))]; + tensor k_37_cast_fp16 = conv(dilations = k_37_dilations_0, groups = k_37_groups_0, pad = k_37_pad_0, pad_type = k_37_pad_type_0, strides = k_37_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_37_cast_fp16")]; + tensor v_19_pad_type_0 = const()[name = tensor("v_19_pad_type_0"), val = tensor("valid")]; + tensor v_19_strides_0 = const()[name = tensor("v_19_strides_0"), val = tensor([1, 1])]; + tensor v_19_pad_0 = const()[name = tensor("v_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_19_dilations_0 = const()[name = tensor("v_19_dilations_0"), val = tensor([1, 1])]; + tensor v_19_groups_0 = const()[name = tensor("v_19_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_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(169649664)))]; + tensor v_19_cast_fp16 = conv(dilations = v_19_dilations_0, groups = v_19_groups_0, pad = v_19_pad_0, pad_type = v_19_pad_type_0, strides = v_19_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_19_cast_fp16")]; + tensor var_4901_begin_0 = const()[name = tensor("op_4901_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4901_end_0 = const()[name = tensor("op_4901_end_0"), val = tensor([2, 160, 1, 256])]; + tensor var_4901_end_mask_0 = const()[name = tensor("op_4901_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4901_cast_fp16 = slice_by_index(begin = var_4901_begin_0, end = var_4901_end_0, end_mask = var_4901_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_4901_cast_fp16")]; + tensor var_4905_begin_0 = const()[name = tensor("op_4905_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_4905_end_0 = const()[name = tensor("op_4905_end_0"), val = tensor([2, 320, 1, 256])]; + tensor var_4905_end_mask_0 = const()[name = tensor("op_4905_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4905_cast_fp16 = slice_by_index(begin = var_4905_begin_0, end = var_4905_end_0, end_mask = var_4905_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_4905_cast_fp16")]; + tensor var_4909_begin_0 = const()[name = tensor("op_4909_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4909_end_0 = const()[name = tensor("op_4909_end_0"), val = tensor([2, 480, 1, 256])]; + tensor var_4909_end_mask_0 = const()[name = tensor("op_4909_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4909_cast_fp16 = slice_by_index(begin = var_4909_begin_0, end = var_4909_end_0, end_mask = var_4909_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_4909_cast_fp16")]; + tensor var_4913_begin_0 = const()[name = tensor("op_4913_begin_0"), val = tensor([0, 480, 0, 0])]; + tensor var_4913_end_0 = const()[name = tensor("op_4913_end_0"), val = tensor([2, 640, 1, 256])]; + tensor var_4913_end_mask_0 = const()[name = tensor("op_4913_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4913_cast_fp16 = slice_by_index(begin = var_4913_begin_0, end = var_4913_end_0, end_mask = var_4913_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_4913_cast_fp16")]; + tensor var_4917_begin_0 = const()[name = tensor("op_4917_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_4917_end_0 = const()[name = tensor("op_4917_end_0"), val = tensor([2, 800, 1, 256])]; + tensor var_4917_end_mask_0 = const()[name = tensor("op_4917_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4917_cast_fp16 = slice_by_index(begin = var_4917_begin_0, end = var_4917_end_0, end_mask = var_4917_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_4917_cast_fp16")]; + tensor var_4921_begin_0 = const()[name = tensor("op_4921_begin_0"), val = tensor([0, 800, 0, 0])]; + tensor var_4921_end_0 = const()[name = tensor("op_4921_end_0"), val = tensor([2, 960, 1, 256])]; + tensor var_4921_end_mask_0 = const()[name = tensor("op_4921_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4921_cast_fp16 = slice_by_index(begin = var_4921_begin_0, end = var_4921_end_0, end_mask = var_4921_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_4921_cast_fp16")]; + tensor var_4925_begin_0 = const()[name = tensor("op_4925_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_4925_end_0 = const()[name = tensor("op_4925_end_0"), val = tensor([2, 1120, 1, 256])]; + tensor var_4925_end_mask_0 = const()[name = tensor("op_4925_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4925_cast_fp16 = slice_by_index(begin = var_4925_begin_0, end = var_4925_end_0, end_mask = var_4925_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_4925_cast_fp16")]; + tensor var_4929_begin_0 = const()[name = tensor("op_4929_begin_0"), val = tensor([0, 1120, 0, 0])]; + tensor var_4929_end_0 = const()[name = tensor("op_4929_end_0"), val = tensor([2, 1, 1, 256])]; + tensor var_4929_end_mask_0 = const()[name = tensor("op_4929_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4929_cast_fp16 = slice_by_index(begin = var_4929_begin_0, end = var_4929_end_0, end_mask = var_4929_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_4929_cast_fp16")]; + tensor k_39_perm_0 = const()[name = tensor("k_39_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_4936_begin_0 = const()[name = tensor("op_4936_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4936_end_0 = const()[name = tensor("op_4936_end_0"), val = tensor([2, 77, 1, 160])]; + tensor var_4936_end_mask_0 = const()[name = tensor("op_4936_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_39_cast_fp16 = transpose(perm = k_39_perm_0, x = k_37_cast_fp16)[name = tensor("transpose_4")]; + tensor var_4936_cast_fp16 = slice_by_index(begin = var_4936_begin_0, end = var_4936_end_0, end_mask = var_4936_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_4936_cast_fp16")]; + tensor var_4940_begin_0 = const()[name = tensor("op_4940_begin_0"), val = tensor([0, 0, 0, 160])]; + tensor var_4940_end_0 = const()[name = tensor("op_4940_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_4940_end_mask_0 = const()[name = tensor("op_4940_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4940_cast_fp16 = slice_by_index(begin = var_4940_begin_0, end = var_4940_end_0, end_mask = var_4940_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_4940_cast_fp16")]; + tensor var_4944_begin_0 = const()[name = tensor("op_4944_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_4944_end_0 = const()[name = tensor("op_4944_end_0"), val = tensor([2, 77, 1, 480])]; + tensor var_4944_end_mask_0 = const()[name = tensor("op_4944_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4944_cast_fp16 = slice_by_index(begin = var_4944_begin_0, end = var_4944_end_0, end_mask = var_4944_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_4944_cast_fp16")]; + tensor var_4948_begin_0 = const()[name = tensor("op_4948_begin_0"), val = tensor([0, 0, 0, 480])]; + tensor var_4948_end_0 = const()[name = tensor("op_4948_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_4948_end_mask_0 = const()[name = tensor("op_4948_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4948_cast_fp16 = slice_by_index(begin = var_4948_begin_0, end = var_4948_end_0, end_mask = var_4948_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_4948_cast_fp16")]; + tensor var_4952_begin_0 = const()[name = tensor("op_4952_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_4952_end_0 = const()[name = tensor("op_4952_end_0"), val = tensor([2, 77, 1, 800])]; + tensor var_4952_end_mask_0 = const()[name = tensor("op_4952_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4952_cast_fp16 = slice_by_index(begin = var_4952_begin_0, end = var_4952_end_0, end_mask = var_4952_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_4952_cast_fp16")]; + tensor var_4956_begin_0 = const()[name = tensor("op_4956_begin_0"), val = tensor([0, 0, 0, 800])]; + tensor var_4956_end_0 = const()[name = tensor("op_4956_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_4956_end_mask_0 = const()[name = tensor("op_4956_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4956_cast_fp16 = slice_by_index(begin = var_4956_begin_0, end = var_4956_end_0, end_mask = var_4956_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_4956_cast_fp16")]; + tensor var_4960_begin_0 = const()[name = tensor("op_4960_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_4960_end_0 = const()[name = tensor("op_4960_end_0"), val = tensor([2, 77, 1, 1120])]; + tensor var_4960_end_mask_0 = const()[name = tensor("op_4960_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4960_cast_fp16 = slice_by_index(begin = var_4960_begin_0, end = var_4960_end_0, end_mask = var_4960_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_4960_cast_fp16")]; + tensor var_4964_begin_0 = const()[name = tensor("op_4964_begin_0"), val = tensor([0, 0, 0, 1120])]; + tensor var_4964_end_0 = const()[name = tensor("op_4964_end_0"), val = tensor([2, 77, 1, 1])]; + tensor var_4964_end_mask_0 = const()[name = tensor("op_4964_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4964_cast_fp16 = slice_by_index(begin = var_4964_begin_0, end = var_4964_end_0, end_mask = var_4964_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_4964_cast_fp16")]; + tensor var_4966_begin_0 = const()[name = tensor("op_4966_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4966_end_0 = const()[name = tensor("op_4966_end_0"), val = tensor([2, 160, 1, 77])]; + tensor var_4966_end_mask_0 = const()[name = tensor("op_4966_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4966_cast_fp16 = slice_by_index(begin = var_4966_begin_0, end = var_4966_end_0, end_mask = var_4966_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_4966_cast_fp16")]; + tensor var_4970_begin_0 = const()[name = tensor("op_4970_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_4970_end_0 = const()[name = tensor("op_4970_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_4970_end_mask_0 = const()[name = tensor("op_4970_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4970_cast_fp16 = slice_by_index(begin = var_4970_begin_0, end = var_4970_end_0, end_mask = var_4970_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_4970_cast_fp16")]; + tensor var_4974_begin_0 = const()[name = tensor("op_4974_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4974_end_0 = const()[name = tensor("op_4974_end_0"), val = tensor([2, 480, 1, 77])]; + tensor var_4974_end_mask_0 = const()[name = tensor("op_4974_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4974_cast_fp16 = slice_by_index(begin = var_4974_begin_0, end = var_4974_end_0, end_mask = var_4974_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_4974_cast_fp16")]; + tensor var_4978_begin_0 = const()[name = tensor("op_4978_begin_0"), val = tensor([0, 480, 0, 0])]; + tensor var_4978_end_0 = const()[name = tensor("op_4978_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_4978_end_mask_0 = const()[name = tensor("op_4978_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4978_cast_fp16 = slice_by_index(begin = var_4978_begin_0, end = var_4978_end_0, end_mask = var_4978_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_4978_cast_fp16")]; + tensor var_4982_begin_0 = const()[name = tensor("op_4982_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_4982_end_0 = const()[name = tensor("op_4982_end_0"), val = tensor([2, 800, 1, 77])]; + tensor var_4982_end_mask_0 = const()[name = tensor("op_4982_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4982_cast_fp16 = slice_by_index(begin = var_4982_begin_0, end = var_4982_end_0, end_mask = var_4982_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_4982_cast_fp16")]; + tensor var_4986_begin_0 = const()[name = tensor("op_4986_begin_0"), val = tensor([0, 800, 0, 0])]; + tensor var_4986_end_0 = const()[name = tensor("op_4986_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_4986_end_mask_0 = const()[name = tensor("op_4986_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4986_cast_fp16 = slice_by_index(begin = var_4986_begin_0, end = var_4986_end_0, end_mask = var_4986_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_4986_cast_fp16")]; + tensor var_4990_begin_0 = const()[name = tensor("op_4990_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_4990_end_0 = const()[name = tensor("op_4990_end_0"), val = tensor([2, 1120, 1, 77])]; + tensor var_4990_end_mask_0 = const()[name = tensor("op_4990_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4990_cast_fp16 = slice_by_index(begin = var_4990_begin_0, end = var_4990_end_0, end_mask = var_4990_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_4990_cast_fp16")]; + tensor var_4994_begin_0 = const()[name = tensor("op_4994_begin_0"), val = tensor([0, 1120, 0, 0])]; + tensor var_4994_end_0 = const()[name = tensor("op_4994_end_0"), val = tensor([2, 1, 1, 77])]; + tensor var_4994_end_mask_0 = const()[name = tensor("op_4994_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4994_cast_fp16 = slice_by_index(begin = var_4994_begin_0, end = var_4994_end_0, end_mask = var_4994_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_4994_cast_fp16")]; + tensor var_4998_equation_0 = const()[name = tensor("op_4998_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4998_cast_fp16 = einsum(equation = var_4998_equation_0, values = (var_4936_cast_fp16, var_4901_cast_fp16))[name = tensor("op_4998_cast_fp16")]; + tensor var_4999_to_fp16 = const()[name = tensor("op_4999_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_17_cast_fp16 = mul(x = var_4998_cast_fp16, y = var_4999_to_fp16)[name = tensor("aw_17_cast_fp16")]; + tensor var_5002_equation_0 = const()[name = tensor("op_5002_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5002_cast_fp16 = einsum(equation = var_5002_equation_0, values = (var_4940_cast_fp16, var_4905_cast_fp16))[name = tensor("op_5002_cast_fp16")]; + tensor var_5003_to_fp16 = const()[name = tensor("op_5003_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_19_cast_fp16 = mul(x = var_5002_cast_fp16, y = var_5003_to_fp16)[name = tensor("aw_19_cast_fp16")]; + tensor var_5006_equation_0 = const()[name = tensor("op_5006_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5006_cast_fp16 = einsum(equation = var_5006_equation_0, values = (var_4944_cast_fp16, var_4909_cast_fp16))[name = tensor("op_5006_cast_fp16")]; + tensor var_5007_to_fp16 = const()[name = tensor("op_5007_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_21_cast_fp16 = mul(x = var_5006_cast_fp16, y = var_5007_to_fp16)[name = tensor("aw_21_cast_fp16")]; + tensor var_5010_equation_0 = const()[name = tensor("op_5010_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5010_cast_fp16 = einsum(equation = var_5010_equation_0, values = (var_4948_cast_fp16, var_4913_cast_fp16))[name = tensor("op_5010_cast_fp16")]; + tensor var_5011_to_fp16 = const()[name = tensor("op_5011_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_23_cast_fp16 = mul(x = var_5010_cast_fp16, y = var_5011_to_fp16)[name = tensor("aw_23_cast_fp16")]; + tensor var_5014_equation_0 = const()[name = tensor("op_5014_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5014_cast_fp16 = einsum(equation = var_5014_equation_0, values = (var_4952_cast_fp16, var_4917_cast_fp16))[name = tensor("op_5014_cast_fp16")]; + tensor var_5015_to_fp16 = const()[name = tensor("op_5015_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_25_cast_fp16 = mul(x = var_5014_cast_fp16, y = var_5015_to_fp16)[name = tensor("aw_25_cast_fp16")]; + tensor var_5018_equation_0 = const()[name = tensor("op_5018_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5018_cast_fp16 = einsum(equation = var_5018_equation_0, values = (var_4956_cast_fp16, var_4921_cast_fp16))[name = tensor("op_5018_cast_fp16")]; + tensor var_5019_to_fp16 = const()[name = tensor("op_5019_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_27_cast_fp16 = mul(x = var_5018_cast_fp16, y = var_5019_to_fp16)[name = tensor("aw_27_cast_fp16")]; + tensor var_5022_equation_0 = const()[name = tensor("op_5022_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5022_cast_fp16 = einsum(equation = var_5022_equation_0, values = (var_4960_cast_fp16, var_4925_cast_fp16))[name = tensor("op_5022_cast_fp16")]; + tensor var_5023_to_fp16 = const()[name = tensor("op_5023_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_29_cast_fp16 = mul(x = var_5022_cast_fp16, y = var_5023_to_fp16)[name = tensor("aw_29_cast_fp16")]; + tensor var_5026_equation_0 = const()[name = tensor("op_5026_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5026_cast_fp16 = einsum(equation = var_5026_equation_0, values = (var_4964_cast_fp16, var_4929_cast_fp16))[name = tensor("op_5026_cast_fp16")]; + tensor var_5027_to_fp16 = const()[name = tensor("op_5027_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_31_cast_fp16 = mul(x = var_5026_cast_fp16, y = var_5027_to_fp16)[name = tensor("aw_31_cast_fp16")]; + tensor var_5029_cast_fp16 = softmax(axis = var_4569, x = aw_17_cast_fp16)[name = tensor("op_5029_cast_fp16")]; + tensor var_5030_cast_fp16 = softmax(axis = var_4569, x = aw_19_cast_fp16)[name = tensor("op_5030_cast_fp16")]; + tensor var_5031_cast_fp16 = softmax(axis = var_4569, x = aw_21_cast_fp16)[name = tensor("op_5031_cast_fp16")]; + tensor var_5032_cast_fp16 = softmax(axis = var_4569, x = aw_23_cast_fp16)[name = tensor("op_5032_cast_fp16")]; + tensor var_5033_cast_fp16 = softmax(axis = var_4569, x = aw_25_cast_fp16)[name = tensor("op_5033_cast_fp16")]; + tensor var_5034_cast_fp16 = softmax(axis = var_4569, x = aw_27_cast_fp16)[name = tensor("op_5034_cast_fp16")]; + tensor var_5035_cast_fp16 = softmax(axis = var_4569, x = aw_29_cast_fp16)[name = tensor("op_5035_cast_fp16")]; + tensor var_5036_cast_fp16 = softmax(axis = var_4569, x = aw_31_cast_fp16)[name = tensor("op_5036_cast_fp16")]; + tensor var_5038_equation_0 = const()[name = tensor("op_5038_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5038_cast_fp16 = einsum(equation = var_5038_equation_0, values = (var_4966_cast_fp16, var_5029_cast_fp16))[name = tensor("op_5038_cast_fp16")]; + tensor var_5040_equation_0 = const()[name = tensor("op_5040_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5040_cast_fp16 = einsum(equation = var_5040_equation_0, values = (var_4970_cast_fp16, var_5030_cast_fp16))[name = tensor("op_5040_cast_fp16")]; + tensor var_5042_equation_0 = const()[name = tensor("op_5042_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5042_cast_fp16 = einsum(equation = var_5042_equation_0, values = (var_4974_cast_fp16, var_5031_cast_fp16))[name = tensor("op_5042_cast_fp16")]; + tensor var_5044_equation_0 = const()[name = tensor("op_5044_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5044_cast_fp16 = einsum(equation = var_5044_equation_0, values = (var_4978_cast_fp16, var_5032_cast_fp16))[name = tensor("op_5044_cast_fp16")]; + tensor var_5046_equation_0 = const()[name = tensor("op_5046_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5046_cast_fp16 = einsum(equation = var_5046_equation_0, values = (var_4982_cast_fp16, var_5033_cast_fp16))[name = tensor("op_5046_cast_fp16")]; + tensor var_5048_equation_0 = const()[name = tensor("op_5048_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5048_cast_fp16 = einsum(equation = var_5048_equation_0, values = (var_4986_cast_fp16, var_5034_cast_fp16))[name = tensor("op_5048_cast_fp16")]; + tensor var_5050_equation_0 = const()[name = tensor("op_5050_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5050_cast_fp16 = einsum(equation = var_5050_equation_0, values = (var_4990_cast_fp16, var_5035_cast_fp16))[name = tensor("op_5050_cast_fp16")]; + tensor var_5052_equation_0 = const()[name = tensor("op_5052_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5052_cast_fp16 = einsum(equation = var_5052_equation_0, values = (var_4994_cast_fp16, var_5036_cast_fp16))[name = tensor("op_5052_cast_fp16")]; + tensor input_135_interleave_0 = const()[name = tensor("input_135_interleave_0"), val = tensor(false)]; + tensor input_135_cast_fp16 = concat(axis = var_4569, interleave = input_135_interleave_0, values = (var_5038_cast_fp16, var_5040_cast_fp16, var_5042_cast_fp16, var_5044_cast_fp16, var_5046_cast_fp16, var_5048_cast_fp16, var_5050_cast_fp16, var_5052_cast_fp16))[name = tensor("input_135_cast_fp16")]; + tensor var_5062_pad_type_0 = const()[name = tensor("op_5062_pad_type_0"), val = tensor("valid")]; + tensor var_5062_strides_0 = const()[name = tensor("op_5062_strides_0"), val = tensor([1, 1])]; + tensor var_5062_pad_0 = const()[name = tensor("op_5062_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5062_dilations_0 = const()[name = tensor("op_5062_dilations_0"), val = tensor([1, 1])]; + tensor var_5062_groups_0 = const()[name = tensor("op_5062_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_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(171615808)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_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(174892672)))]; + tensor var_5062_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_5062_dilations_0, groups = var_5062_groups_0, pad = var_5062_pad_0, pad_type = var_5062_pad_type_0, strides = var_5062_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_135_cast_fp16)[name = tensor("op_5062_cast_fp16")]; + tensor inputs_29_cast_fp16 = add(x = var_5062_cast_fp16, y = inputs_27_cast_fp16)[name = tensor("inputs_29_cast_fp16")]; + tensor input_137_axes_0 = const()[name = tensor("input_137_axes_0"), val = tensor([1])]; + tensor input_137_gamma_0_to_fp16 = const()[name = tensor("input_137_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(174895296)))]; + tensor input_137_beta_0_to_fp16 = const()[name = tensor("input_137_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(174897920)))]; + tensor var_5072_to_fp16 = const()[name = tensor("op_5072_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_137_cast_fp16 = layer_norm(axes = input_137_axes_0, beta = input_137_beta_0_to_fp16, epsilon = var_5072_to_fp16, gamma = input_137_gamma_0_to_fp16, x = inputs_29_cast_fp16)[name = tensor("input_137_cast_fp16")]; + tensor var_5092_pad_type_0 = const()[name = tensor("op_5092_pad_type_0"), val = tensor("valid")]; + tensor var_5092_strides_0 = const()[name = tensor("op_5092_strides_0"), val = tensor([1, 1])]; + tensor var_5092_pad_0 = const()[name = tensor("op_5092_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5092_dilations_0 = const()[name = tensor("op_5092_dilations_0"), val = tensor([1, 1])]; + tensor var_5092_groups_0 = const()[name = tensor("op_5092_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_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(174900544)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_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(201115008)))]; + tensor var_5092_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_5092_dilations_0, groups = var_5092_groups_0, pad = var_5092_pad_0, pad_type = var_5092_pad_type_0, strides = var_5092_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_137_cast_fp16)[name = tensor("op_5092_cast_fp16")]; + tensor var_5093_split_sizes_0 = const()[name = tensor("op_5093_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_5093_axis_0 = const()[name = tensor("op_5093_axis_0"), val = tensor(1)]; + tensor var_5093_cast_fp16_0, tensor var_5093_cast_fp16_1 = split(axis = var_5093_axis_0, split_sizes = var_5093_split_sizes_0, x = var_5092_cast_fp16)[name = tensor("op_5093_cast_fp16")]; + tensor var_5095_mode_0 = const()[name = tensor("op_5095_mode_0"), val = tensor("EXACT")]; + tensor var_5095_cast_fp16 = gelu(mode = var_5095_mode_0, x = var_5093_cast_fp16_1)[name = tensor("op_5095_cast_fp16")]; + tensor input_139_cast_fp16 = mul(x = var_5093_cast_fp16_0, y = var_5095_cast_fp16)[name = tensor("input_139_cast_fp16")]; + tensor var_5103_pad_type_0 = const()[name = tensor("op_5103_pad_type_0"), val = tensor("valid")]; + tensor var_5103_strides_0 = const()[name = tensor("op_5103_strides_0"), val = tensor([1, 1])]; + tensor var_5103_pad_0 = const()[name = tensor("op_5103_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5103_dilations_0 = const()[name = tensor("op_5103_dilations_0"), val = tensor([1, 1])]; + tensor var_5103_groups_0 = const()[name = tensor("op_5103_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("down_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(201135552)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("down_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(214242816)))]; + tensor var_5103_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_5103_dilations_0, groups = var_5103_groups_0, pad = var_5103_pad_0, pad_type = var_5103_pad_type_0, strides = var_5103_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_139_cast_fp16)[name = tensor("op_5103_cast_fp16")]; + tensor hidden_states_87_cast_fp16 = add(x = var_5103_cast_fp16, y = inputs_29_cast_fp16)[name = tensor("hidden_states_87_cast_fp16")]; + tensor var_5105 = const()[name = tensor("op_5105"), val = tensor([2, 1280, 16, 16])]; + tensor input_141_cast_fp16 = reshape(shape = var_5105, x = hidden_states_87_cast_fp16)[name = tensor("input_141_cast_fp16")]; + tensor hidden_states_89_pad_type_0 = const()[name = tensor("hidden_states_89_pad_type_0"), val = tensor("valid")]; + tensor hidden_states_89_strides_0 = const()[name = tensor("hidden_states_89_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_89_pad_0 = const()[name = tensor("hidden_states_89_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_89_dilations_0 = const()[name = tensor("hidden_states_89_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_89_groups_0 = const()[name = tensor("hidden_states_89_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(214245440)))]; + tensor down_blocks_2_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217522304)))]; + tensor hidden_states_89_cast_fp16 = conv(bias = down_blocks_2_attentions_0_proj_out_bias_to_fp16, dilations = hidden_states_89_dilations_0, groups = hidden_states_89_groups_0, pad = hidden_states_89_pad_0, pad_type = hidden_states_89_pad_type_0, strides = hidden_states_89_strides_0, weight = down_blocks_2_attentions_0_proj_out_weight_to_fp16, x = input_141_cast_fp16)[name = tensor("hidden_states_89_cast_fp16")]; + tensor input_143_cast_fp16_1 = add(x = hidden_states_89_cast_fp16, y = hidden_states_77_cast_fp16)[name = tensor("input_143_cast_fp16")]; + tensor reshape_60_shape_0 = const()[name = tensor("reshape_60_shape_0"), val = tensor([2, 32, 40, 16, 16])]; + tensor reshape_60_cast_fp16 = reshape(shape = reshape_60_shape_0, x = input_143_cast_fp16_1)[name = tensor("reshape_60_cast_fp16")]; + tensor reduce_mean_45_axes_0 = const()[name = tensor("reduce_mean_45_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_45_keep_dims_0 = const()[name = tensor("reduce_mean_45_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_45_cast_fp16 = reduce_mean(axes = reduce_mean_45_axes_0, keep_dims = reduce_mean_45_keep_dims_0, x = reshape_60_cast_fp16)[name = tensor("reduce_mean_45_cast_fp16")]; + tensor sub_30_cast_fp16 = sub(x = reshape_60_cast_fp16, y = reduce_mean_45_cast_fp16)[name = tensor("sub_30_cast_fp16")]; + tensor square_15_cast_fp16 = square(x = sub_30_cast_fp16)[name = tensor("square_15_cast_fp16")]; + tensor reduce_mean_47_axes_0 = const()[name = tensor("reduce_mean_47_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_47_keep_dims_0 = const()[name = tensor("reduce_mean_47_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_47_cast_fp16 = reduce_mean(axes = reduce_mean_47_axes_0, keep_dims = reduce_mean_47_keep_dims_0, x = square_15_cast_fp16)[name = tensor("reduce_mean_47_cast_fp16")]; + tensor add_30_y_0_to_fp16 = const()[name = tensor("add_30_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_30_cast_fp16 = add(x = reduce_mean_47_cast_fp16, y = add_30_y_0_to_fp16)[name = tensor("add_30_cast_fp16")]; + tensor sqrt_15_cast_fp16 = sqrt(x = add_30_cast_fp16)[name = tensor("sqrt_15_cast_fp16")]; + tensor real_div_15_cast_fp16 = real_div(x = sub_30_cast_fp16, y = sqrt_15_cast_fp16)[name = tensor("real_div_15_cast_fp16")]; + tensor reshape_61_shape_0 = const()[name = tensor("reshape_61_shape_0"), val = tensor([2, 1280, 16, 16])]; + tensor reshape_61_cast_fp16 = reshape(shape = reshape_61_shape_0, x = real_div_15_cast_fp16)[name = tensor("reshape_61_cast_fp16")]; + tensor add_31_gamma_0_to_fp16 = const()[name = tensor("add_31_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217524928)))]; + tensor add_31_beta_0_to_fp16 = const()[name = tensor("add_31_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217527552)))]; + tensor add_31_epsilon_0_to_fp16 = const()[name = tensor("add_31_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_31_cast_fp16 = batch_norm(beta = add_31_beta_0_to_fp16, epsilon = add_31_epsilon_0_to_fp16, gamma = add_31_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_61_cast_fp16)[name = tensor("add_31_cast_fp16")]; + tensor input_147_cast_fp16 = silu(x = add_31_cast_fp16)[name = tensor("input_147_cast_fp16")]; + tensor hidden_states_91_pad_type_0 = const()[name = tensor("hidden_states_91_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_91_pad_0 = const()[name = tensor("hidden_states_91_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_91_strides_0 = const()[name = tensor("hidden_states_91_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_91_dilations_0 = const()[name = tensor("hidden_states_91_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_91_groups_0 = const()[name = tensor("hidden_states_91_groups_0"), val = tensor(1)]; + tensor down_blocks_2_resnets_1_conv1_weight_to_fp16 = const()[name = tensor("down_blocks_2_resnets_1_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217530176)))]; + tensor down_blocks_2_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(247021440)))]; + tensor hidden_states_91_cast_fp16 = conv(bias = down_blocks_2_resnets_1_conv1_bias_to_fp16, dilations = hidden_states_91_dilations_0, groups = hidden_states_91_groups_0, pad = hidden_states_91_pad_0, pad_type = hidden_states_91_pad_type_0, strides = hidden_states_91_strides_0, weight = down_blocks_2_resnets_1_conv1_weight_to_fp16, x = input_147_cast_fp16)[name = tensor("hidden_states_91_cast_fp16")]; + tensor temb_11_pad_type_0 = const()[name = tensor("temb_11_pad_type_0"), val = tensor("valid")]; + tensor temb_11_strides_0 = const()[name = tensor("temb_11_strides_0"), val = tensor([1, 1])]; + tensor temb_11_pad_0 = const()[name = tensor("temb_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor temb_11_dilations_0 = const()[name = tensor("temb_11_dilations_0"), val = tensor([1, 1])]; + tensor temb_11_groups_0 = const()[name = tensor("temb_11_groups_0"), val = tensor(1)]; + tensor down_blocks_2_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor("down_blocks_2_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(247024064)))]; + tensor down_blocks_2_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250300928)))]; + tensor temb_11_cast_fp16 = conv(bias = down_blocks_2_resnets_1_time_emb_proj_bias_to_fp16, dilations = temb_11_dilations_0, groups = temb_11_groups_0, pad = temb_11_pad_0, pad_type = temb_11_pad_type_0, strides = temb_11_strides_0, weight = down_blocks_2_resnets_1_time_emb_proj_weight_to_fp16, x = input_15_cast_fp16_1)[name = tensor("temb_11_cast_fp16")]; + tensor input_151_cast_fp16 = add(x = hidden_states_91_cast_fp16, y = temb_11_cast_fp16)[name = tensor("input_151_cast_fp16")]; + tensor reshape_64_shape_0 = const()[name = tensor("reshape_64_shape_0"), val = tensor([2, 32, 40, 16, 16])]; + tensor reshape_64_cast_fp16 = reshape(shape = reshape_64_shape_0, x = input_151_cast_fp16)[name = tensor("reshape_64_cast_fp16")]; + tensor reduce_mean_48_axes_0 = const()[name = tensor("reduce_mean_48_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_48_keep_dims_0 = const()[name = tensor("reduce_mean_48_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_48_cast_fp16 = reduce_mean(axes = reduce_mean_48_axes_0, keep_dims = reduce_mean_48_keep_dims_0, x = reshape_64_cast_fp16)[name = tensor("reduce_mean_48_cast_fp16")]; + tensor sub_32_cast_fp16 = sub(x = reshape_64_cast_fp16, y = reduce_mean_48_cast_fp16)[name = tensor("sub_32_cast_fp16")]; + tensor square_16_cast_fp16 = square(x = sub_32_cast_fp16)[name = tensor("square_16_cast_fp16")]; + tensor reduce_mean_50_axes_0 = const()[name = tensor("reduce_mean_50_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_50_keep_dims_0 = const()[name = tensor("reduce_mean_50_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_50_cast_fp16 = reduce_mean(axes = reduce_mean_50_axes_0, keep_dims = reduce_mean_50_keep_dims_0, x = square_16_cast_fp16)[name = tensor("reduce_mean_50_cast_fp16")]; + tensor add_32_y_0_to_fp16 = const()[name = tensor("add_32_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_32_cast_fp16 = add(x = reduce_mean_50_cast_fp16, y = add_32_y_0_to_fp16)[name = tensor("add_32_cast_fp16")]; + tensor sqrt_16_cast_fp16 = sqrt(x = add_32_cast_fp16)[name = tensor("sqrt_16_cast_fp16")]; + tensor real_div_16_cast_fp16 = real_div(x = sub_32_cast_fp16, y = sqrt_16_cast_fp16)[name = tensor("real_div_16_cast_fp16")]; + tensor reshape_65_shape_0 = const()[name = tensor("reshape_65_shape_0"), val = tensor([2, 1280, 16, 16])]; + tensor reshape_65_cast_fp16 = reshape(shape = reshape_65_shape_0, x = real_div_16_cast_fp16)[name = tensor("reshape_65_cast_fp16")]; + tensor add_33_gamma_0_to_fp16 = const()[name = tensor("add_33_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250303552)))]; + tensor add_33_beta_0_to_fp16 = const()[name = tensor("add_33_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250306176)))]; + tensor add_33_epsilon_0_to_fp16 = const()[name = tensor("add_33_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_33_cast_fp16 = batch_norm(beta = add_33_beta_0_to_fp16, epsilon = add_33_epsilon_0_to_fp16, gamma = add_33_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_65_cast_fp16)[name = tensor("add_33_cast_fp16")]; + tensor input_155_cast_fp16 = silu(x = add_33_cast_fp16)[name = tensor("input_155_cast_fp16")]; + tensor hidden_states_93_pad_type_0 = const()[name = tensor("hidden_states_93_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_93_pad_0 = const()[name = tensor("hidden_states_93_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_93_strides_0 = const()[name = tensor("hidden_states_93_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_93_dilations_0 = const()[name = tensor("hidden_states_93_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_93_groups_0 = const()[name = tensor("hidden_states_93_groups_0"), val = tensor(1)]; + tensor down_blocks_2_resnets_1_conv2_weight_to_fp16 = const()[name = tensor("down_blocks_2_resnets_1_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250308800)))]; + tensor down_blocks_2_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(279800064)))]; + tensor hidden_states_93_cast_fp16 = conv(bias = down_blocks_2_resnets_1_conv2_bias_to_fp16, dilations = hidden_states_93_dilations_0, groups = hidden_states_93_groups_0, pad = hidden_states_93_pad_0, pad_type = hidden_states_93_pad_type_0, strides = hidden_states_93_strides_0, weight = down_blocks_2_resnets_1_conv2_weight_to_fp16, x = input_155_cast_fp16)[name = tensor("hidden_states_93_cast_fp16")]; + tensor hidden_states_95_cast_fp16 = add(x = input_143_cast_fp16_1, y = hidden_states_93_cast_fp16)[name = tensor("hidden_states_95_cast_fp16")]; + tensor reshape_68_shape_0 = const()[name = tensor("reshape_68_shape_0"), val = tensor([2, 32, 40, 16, 16])]; + tensor reshape_68_cast_fp16 = reshape(shape = reshape_68_shape_0, x = hidden_states_95_cast_fp16)[name = tensor("reshape_68_cast_fp16")]; + tensor reduce_mean_51_axes_0 = const()[name = tensor("reduce_mean_51_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_51_keep_dims_0 = const()[name = tensor("reduce_mean_51_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_51_cast_fp16 = reduce_mean(axes = reduce_mean_51_axes_0, keep_dims = reduce_mean_51_keep_dims_0, x = reshape_68_cast_fp16)[name = tensor("reduce_mean_51_cast_fp16")]; + tensor sub_34_cast_fp16 = sub(x = reshape_68_cast_fp16, y = reduce_mean_51_cast_fp16)[name = tensor("sub_34_cast_fp16")]; + tensor square_17_cast_fp16 = square(x = sub_34_cast_fp16)[name = tensor("square_17_cast_fp16")]; + tensor reduce_mean_53_axes_0 = const()[name = tensor("reduce_mean_53_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_53_keep_dims_0 = const()[name = tensor("reduce_mean_53_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_53_cast_fp16 = reduce_mean(axes = reduce_mean_53_axes_0, keep_dims = reduce_mean_53_keep_dims_0, x = square_17_cast_fp16)[name = tensor("reduce_mean_53_cast_fp16")]; + tensor add_34_y_0_to_fp16 = const()[name = tensor("add_34_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_34_cast_fp16 = add(x = reduce_mean_53_cast_fp16, y = add_34_y_0_to_fp16)[name = tensor("add_34_cast_fp16")]; + tensor sqrt_17_cast_fp16 = sqrt(x = add_34_cast_fp16)[name = tensor("sqrt_17_cast_fp16")]; + tensor real_div_17_cast_fp16 = real_div(x = sub_34_cast_fp16, y = sqrt_17_cast_fp16)[name = tensor("real_div_17_cast_fp16")]; + tensor reshape_69_shape_0 = const()[name = tensor("reshape_69_shape_0"), val = tensor([2, 1280, 16, 16])]; + tensor reshape_69_cast_fp16 = reshape(shape = reshape_69_shape_0, x = real_div_17_cast_fp16)[name = tensor("reshape_69_cast_fp16")]; + tensor add_35_gamma_0_to_fp16 = const()[name = tensor("add_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(279802688)))]; + tensor add_35_beta_0_to_fp16 = const()[name = tensor("add_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(279805312)))]; + tensor add_35_epsilon_0_to_fp16 = const()[name = tensor("add_35_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_35_cast_fp16 = batch_norm(beta = add_35_beta_0_to_fp16, epsilon = add_35_epsilon_0_to_fp16, gamma = add_35_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_69_cast_fp16)[name = tensor("add_35_cast_fp16")]; + tensor hidden_states_97_pad_type_0 = const()[name = tensor("hidden_states_97_pad_type_0"), val = tensor("valid")]; + tensor hidden_states_97_strides_0 = const()[name = tensor("hidden_states_97_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_97_pad_0 = const()[name = tensor("hidden_states_97_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_97_dilations_0 = const()[name = tensor("hidden_states_97_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_97_groups_0 = const()[name = tensor("hidden_states_97_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_proj_in_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(279807936)))]; + tensor down_blocks_2_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(283084800)))]; + tensor hidden_states_97_cast_fp16 = conv(bias = down_blocks_2_attentions_1_proj_in_bias_to_fp16, dilations = hidden_states_97_dilations_0, groups = hidden_states_97_groups_0, pad = hidden_states_97_pad_0, pad_type = hidden_states_97_pad_type_0, strides = hidden_states_97_strides_0, weight = down_blocks_2_attentions_1_proj_in_weight_to_fp16, x = add_35_cast_fp16)[name = tensor("hidden_states_97_cast_fp16")]; + tensor var_5175 = const()[name = tensor("op_5175"), val = tensor([2, 1280, 1, 256])]; + tensor inputs_31_cast_fp16 = reshape(shape = var_5175, x = hidden_states_97_cast_fp16)[name = tensor("inputs_31_cast_fp16")]; + tensor hidden_states_99_axes_0 = const()[name = tensor("hidden_states_99_axes_0"), val = tensor([1])]; + tensor hidden_states_99_gamma_0_to_fp16 = const()[name = tensor("hidden_states_99_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(283087424)))]; + tensor hidden_states_99_beta_0_to_fp16 = const()[name = tensor("hidden_states_99_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(283090048)))]; + tensor var_5191_to_fp16 = const()[name = tensor("op_5191_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_99_cast_fp16 = layer_norm(axes = hidden_states_99_axes_0, beta = hidden_states_99_beta_0_to_fp16, epsilon = var_5191_to_fp16, gamma = hidden_states_99_gamma_0_to_fp16, x = inputs_31_cast_fp16)[name = tensor("hidden_states_99_cast_fp16")]; + tensor q_21_pad_type_0 = const()[name = tensor("q_21_pad_type_0"), val = tensor("valid")]; + tensor q_21_strides_0 = const()[name = tensor("q_21_strides_0"), val = tensor([1, 1])]; + tensor q_21_pad_0 = const()[name = tensor("q_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_21_dilations_0 = const()[name = tensor("q_21_dilations_0"), val = tensor([1, 1])]; + tensor q_21_groups_0 = const()[name = tensor("q_21_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_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(283092672)))]; + tensor q_21_cast_fp16 = conv(dilations = q_21_dilations_0, groups = q_21_groups_0, pad = q_21_pad_0, pad_type = q_21_pad_type_0, strides = q_21_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_99_cast_fp16)[name = tensor("q_21_cast_fp16")]; + tensor k_41_pad_type_0 = const()[name = tensor("k_41_pad_type_0"), val = tensor("valid")]; + tensor k_41_strides_0 = const()[name = tensor("k_41_strides_0"), val = tensor([1, 1])]; + tensor k_41_pad_0 = const()[name = tensor("k_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_41_dilations_0 = const()[name = tensor("k_41_dilations_0"), val = tensor([1, 1])]; + tensor k_41_groups_0 = const()[name = tensor("k_41_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_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(286369536)))]; + tensor k_41_cast_fp16 = conv(dilations = k_41_dilations_0, groups = k_41_groups_0, pad = k_41_pad_0, pad_type = k_41_pad_type_0, strides = k_41_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_99_cast_fp16)[name = tensor("k_41_cast_fp16")]; + tensor v_21_pad_type_0 = const()[name = tensor("v_21_pad_type_0"), val = tensor("valid")]; + tensor v_21_strides_0 = const()[name = tensor("v_21_strides_0"), val = tensor([1, 1])]; + tensor v_21_pad_0 = const()[name = tensor("v_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_21_dilations_0 = const()[name = tensor("v_21_dilations_0"), val = tensor([1, 1])]; + tensor v_21_groups_0 = const()[name = tensor("v_21_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_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(289646400)))]; + tensor v_21_cast_fp16 = conv(dilations = v_21_dilations_0, groups = v_21_groups_0, pad = v_21_pad_0, pad_type = v_21_pad_type_0, strides = v_21_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_99_cast_fp16)[name = tensor("v_21_cast_fp16")]; + tensor var_5224_begin_0 = const()[name = tensor("op_5224_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5224_end_0 = const()[name = tensor("op_5224_end_0"), val = tensor([2, 160, 1, 256])]; + tensor var_5224_end_mask_0 = const()[name = tensor("op_5224_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5224_cast_fp16 = slice_by_index(begin = var_5224_begin_0, end = var_5224_end_0, end_mask = var_5224_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_5224_cast_fp16")]; + tensor var_5228_begin_0 = const()[name = tensor("op_5228_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_5228_end_0 = const()[name = tensor("op_5228_end_0"), val = tensor([2, 320, 1, 256])]; + tensor var_5228_end_mask_0 = const()[name = tensor("op_5228_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5228_cast_fp16 = slice_by_index(begin = var_5228_begin_0, end = var_5228_end_0, end_mask = var_5228_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_5228_cast_fp16")]; + tensor var_5232_begin_0 = const()[name = tensor("op_5232_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5232_end_0 = const()[name = tensor("op_5232_end_0"), val = tensor([2, 480, 1, 256])]; + tensor var_5232_end_mask_0 = const()[name = tensor("op_5232_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5232_cast_fp16 = slice_by_index(begin = var_5232_begin_0, end = var_5232_end_0, end_mask = var_5232_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_5232_cast_fp16")]; + tensor var_5236_begin_0 = const()[name = tensor("op_5236_begin_0"), val = tensor([0, 480, 0, 0])]; + tensor var_5236_end_0 = const()[name = tensor("op_5236_end_0"), val = tensor([2, 640, 1, 256])]; + tensor var_5236_end_mask_0 = const()[name = tensor("op_5236_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5236_cast_fp16 = slice_by_index(begin = var_5236_begin_0, end = var_5236_end_0, end_mask = var_5236_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_5236_cast_fp16")]; + tensor var_5240_begin_0 = const()[name = tensor("op_5240_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5240_end_0 = const()[name = tensor("op_5240_end_0"), val = tensor([2, 800, 1, 256])]; + tensor var_5240_end_mask_0 = const()[name = tensor("op_5240_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5240_cast_fp16 = slice_by_index(begin = var_5240_begin_0, end = var_5240_end_0, end_mask = var_5240_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_5240_cast_fp16")]; + tensor var_5244_begin_0 = const()[name = tensor("op_5244_begin_0"), val = tensor([0, 800, 0, 0])]; + tensor var_5244_end_0 = const()[name = tensor("op_5244_end_0"), val = tensor([2, 960, 1, 256])]; + tensor var_5244_end_mask_0 = const()[name = tensor("op_5244_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5244_cast_fp16 = slice_by_index(begin = var_5244_begin_0, end = var_5244_end_0, end_mask = var_5244_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_5244_cast_fp16")]; + tensor var_5248_begin_0 = const()[name = tensor("op_5248_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_5248_end_0 = const()[name = tensor("op_5248_end_0"), val = tensor([2, 1120, 1, 256])]; + tensor var_5248_end_mask_0 = const()[name = tensor("op_5248_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5248_cast_fp16 = slice_by_index(begin = var_5248_begin_0, end = var_5248_end_0, end_mask = var_5248_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_5248_cast_fp16")]; + tensor var_5252_begin_0 = const()[name = tensor("op_5252_begin_0"), val = tensor([0, 1120, 0, 0])]; + tensor var_5252_end_0 = const()[name = tensor("op_5252_end_0"), val = tensor([2, 1, 1, 256])]; + tensor var_5252_end_mask_0 = const()[name = tensor("op_5252_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_5252_cast_fp16 = slice_by_index(begin = var_5252_begin_0, end = var_5252_end_0, end_mask = var_5252_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_5252_cast_fp16")]; + tensor k_43_perm_0 = const()[name = tensor("k_43_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_5259_begin_0 = const()[name = tensor("op_5259_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5259_end_0 = const()[name = tensor("op_5259_end_0"), val = tensor([2, 256, 1, 160])]; + tensor var_5259_end_mask_0 = const()[name = tensor("op_5259_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_43_cast_fp16 = transpose(perm = k_43_perm_0, x = k_41_cast_fp16)[name = tensor("transpose_3")]; + tensor var_5259_cast_fp16 = slice_by_index(begin = var_5259_begin_0, end = var_5259_end_0, end_mask = var_5259_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_5259_cast_fp16")]; + tensor var_5263_begin_0 = const()[name = tensor("op_5263_begin_0"), val = tensor([0, 0, 0, 160])]; + tensor var_5263_end_0 = const()[name = tensor("op_5263_end_0"), val = tensor([2, 256, 1, 320])]; + tensor var_5263_end_mask_0 = const()[name = tensor("op_5263_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5263_cast_fp16 = slice_by_index(begin = var_5263_begin_0, end = var_5263_end_0, end_mask = var_5263_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_5263_cast_fp16")]; + tensor var_5267_begin_0 = const()[name = tensor("op_5267_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_5267_end_0 = const()[name = tensor("op_5267_end_0"), val = tensor([2, 256, 1, 480])]; + tensor var_5267_end_mask_0 = const()[name = tensor("op_5267_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5267_cast_fp16 = slice_by_index(begin = var_5267_begin_0, end = var_5267_end_0, end_mask = var_5267_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_5267_cast_fp16")]; + tensor var_5271_begin_0 = const()[name = tensor("op_5271_begin_0"), val = tensor([0, 0, 0, 480])]; + tensor var_5271_end_0 = const()[name = tensor("op_5271_end_0"), val = tensor([2, 256, 1, 640])]; + tensor var_5271_end_mask_0 = const()[name = tensor("op_5271_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5271_cast_fp16 = slice_by_index(begin = var_5271_begin_0, end = var_5271_end_0, end_mask = var_5271_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_5271_cast_fp16")]; + tensor var_5275_begin_0 = const()[name = tensor("op_5275_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_5275_end_0 = const()[name = tensor("op_5275_end_0"), val = tensor([2, 256, 1, 800])]; + tensor var_5275_end_mask_0 = const()[name = tensor("op_5275_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5275_cast_fp16 = slice_by_index(begin = var_5275_begin_0, end = var_5275_end_0, end_mask = var_5275_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_5275_cast_fp16")]; + tensor var_5279_begin_0 = const()[name = tensor("op_5279_begin_0"), val = tensor([0, 0, 0, 800])]; + tensor var_5279_end_0 = const()[name = tensor("op_5279_end_0"), val = tensor([2, 256, 1, 960])]; + tensor var_5279_end_mask_0 = const()[name = tensor("op_5279_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5279_cast_fp16 = slice_by_index(begin = var_5279_begin_0, end = var_5279_end_0, end_mask = var_5279_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_5279_cast_fp16")]; + tensor var_5283_begin_0 = const()[name = tensor("op_5283_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_5283_end_0 = const()[name = tensor("op_5283_end_0"), val = tensor([2, 256, 1, 1120])]; + tensor var_5283_end_mask_0 = const()[name = tensor("op_5283_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5283_cast_fp16 = slice_by_index(begin = var_5283_begin_0, end = var_5283_end_0, end_mask = var_5283_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_5283_cast_fp16")]; + tensor var_5287_begin_0 = const()[name = tensor("op_5287_begin_0"), val = tensor([0, 0, 0, 1120])]; + tensor var_5287_end_0 = const()[name = tensor("op_5287_end_0"), val = tensor([2, 256, 1, 1])]; + tensor var_5287_end_mask_0 = const()[name = tensor("op_5287_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_5287_cast_fp16 = slice_by_index(begin = var_5287_begin_0, end = var_5287_end_0, end_mask = var_5287_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_5287_cast_fp16")]; + tensor var_5289_begin_0 = const()[name = tensor("op_5289_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5289_end_0 = const()[name = tensor("op_5289_end_0"), val = tensor([2, 160, 1, 256])]; + tensor var_5289_end_mask_0 = const()[name = tensor("op_5289_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5289_cast_fp16 = slice_by_index(begin = var_5289_begin_0, end = var_5289_end_0, end_mask = var_5289_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_5289_cast_fp16")]; + tensor var_5293_begin_0 = const()[name = tensor("op_5293_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_5293_end_0 = const()[name = tensor("op_5293_end_0"), val = tensor([2, 320, 1, 256])]; + tensor var_5293_end_mask_0 = const()[name = tensor("op_5293_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5293_cast_fp16 = slice_by_index(begin = var_5293_begin_0, end = var_5293_end_0, end_mask = var_5293_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_5293_cast_fp16")]; + tensor var_5297_begin_0 = const()[name = tensor("op_5297_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5297_end_0 = const()[name = tensor("op_5297_end_0"), val = tensor([2, 480, 1, 256])]; + tensor var_5297_end_mask_0 = const()[name = tensor("op_5297_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5297_cast_fp16 = slice_by_index(begin = var_5297_begin_0, end = var_5297_end_0, end_mask = var_5297_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_5297_cast_fp16")]; + tensor var_5301_begin_0 = const()[name = tensor("op_5301_begin_0"), val = tensor([0, 480, 0, 0])]; + tensor var_5301_end_0 = const()[name = tensor("op_5301_end_0"), val = tensor([2, 640, 1, 256])]; + tensor var_5301_end_mask_0 = const()[name = tensor("op_5301_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5301_cast_fp16 = slice_by_index(begin = var_5301_begin_0, end = var_5301_end_0, end_mask = var_5301_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_5301_cast_fp16")]; + tensor var_5305_begin_0 = const()[name = tensor("op_5305_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5305_end_0 = const()[name = tensor("op_5305_end_0"), val = tensor([2, 800, 1, 256])]; + tensor var_5305_end_mask_0 = const()[name = tensor("op_5305_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5305_cast_fp16 = slice_by_index(begin = var_5305_begin_0, end = var_5305_end_0, end_mask = var_5305_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_5305_cast_fp16")]; + tensor var_5309_begin_0 = const()[name = tensor("op_5309_begin_0"), val = tensor([0, 800, 0, 0])]; + tensor var_5309_end_0 = const()[name = tensor("op_5309_end_0"), val = tensor([2, 960, 1, 256])]; + tensor var_5309_end_mask_0 = const()[name = tensor("op_5309_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5309_cast_fp16 = slice_by_index(begin = var_5309_begin_0, end = var_5309_end_0, end_mask = var_5309_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_5309_cast_fp16")]; + tensor var_5313_begin_0 = const()[name = tensor("op_5313_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_5313_end_0 = const()[name = tensor("op_5313_end_0"), val = tensor([2, 1120, 1, 256])]; + tensor var_5313_end_mask_0 = const()[name = tensor("op_5313_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5313_cast_fp16 = slice_by_index(begin = var_5313_begin_0, end = var_5313_end_0, end_mask = var_5313_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_5313_cast_fp16")]; + tensor var_5317_begin_0 = const()[name = tensor("op_5317_begin_0"), val = tensor([0, 1120, 0, 0])]; + tensor var_5317_end_0 = const()[name = tensor("op_5317_end_0"), val = tensor([2, 1, 1, 256])]; + tensor var_5317_end_mask_0 = const()[name = tensor("op_5317_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_5317_cast_fp16 = slice_by_index(begin = var_5317_begin_0, end = var_5317_end_0, end_mask = var_5317_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_5317_cast_fp16")]; + tensor var_5321_equation_0 = const()[name = tensor("op_5321_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5321_cast_fp16 = einsum(equation = var_5321_equation_0, values = (var_5259_cast_fp16, var_5224_cast_fp16))[name = tensor("op_5321_cast_fp16")]; + tensor var_5322_to_fp16 = const()[name = tensor("op_5322_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_33_cast_fp16 = mul(x = var_5321_cast_fp16, y = var_5322_to_fp16)[name = tensor("aw_33_cast_fp16")]; + tensor var_5325_equation_0 = const()[name = tensor("op_5325_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5325_cast_fp16 = einsum(equation = var_5325_equation_0, values = (var_5263_cast_fp16, var_5228_cast_fp16))[name = tensor("op_5325_cast_fp16")]; + tensor var_5326_to_fp16 = const()[name = tensor("op_5326_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_35_cast_fp16 = mul(x = var_5325_cast_fp16, y = var_5326_to_fp16)[name = tensor("aw_35_cast_fp16")]; + tensor var_5329_equation_0 = const()[name = tensor("op_5329_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5329_cast_fp16 = einsum(equation = var_5329_equation_0, values = (var_5267_cast_fp16, var_5232_cast_fp16))[name = tensor("op_5329_cast_fp16")]; + tensor var_5330_to_fp16 = const()[name = tensor("op_5330_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_37_cast_fp16 = mul(x = var_5329_cast_fp16, y = var_5330_to_fp16)[name = tensor("aw_37_cast_fp16")]; + tensor var_5333_equation_0 = const()[name = tensor("op_5333_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5333_cast_fp16 = einsum(equation = var_5333_equation_0, values = (var_5271_cast_fp16, var_5236_cast_fp16))[name = tensor("op_5333_cast_fp16")]; + tensor var_5334_to_fp16 = const()[name = tensor("op_5334_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_39_cast_fp16 = mul(x = var_5333_cast_fp16, y = var_5334_to_fp16)[name = tensor("aw_39_cast_fp16")]; + tensor var_5337_equation_0 = const()[name = tensor("op_5337_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5337_cast_fp16 = einsum(equation = var_5337_equation_0, values = (var_5275_cast_fp16, var_5240_cast_fp16))[name = tensor("op_5337_cast_fp16")]; + tensor var_5338_to_fp16 = const()[name = tensor("op_5338_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_41_cast_fp16 = mul(x = var_5337_cast_fp16, y = var_5338_to_fp16)[name = tensor("aw_41_cast_fp16")]; + tensor var_5341_equation_0 = const()[name = tensor("op_5341_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5341_cast_fp16 = einsum(equation = var_5341_equation_0, values = (var_5279_cast_fp16, var_5244_cast_fp16))[name = tensor("op_5341_cast_fp16")]; + tensor var_5342_to_fp16 = const()[name = tensor("op_5342_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_43_cast_fp16 = mul(x = var_5341_cast_fp16, y = var_5342_to_fp16)[name = tensor("aw_43_cast_fp16")]; + tensor var_5345_equation_0 = const()[name = tensor("op_5345_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5345_cast_fp16 = einsum(equation = var_5345_equation_0, values = (var_5283_cast_fp16, var_5248_cast_fp16))[name = tensor("op_5345_cast_fp16")]; + tensor var_5346_to_fp16 = const()[name = tensor("op_5346_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_45_cast_fp16 = mul(x = var_5345_cast_fp16, y = var_5346_to_fp16)[name = tensor("aw_45_cast_fp16")]; + tensor var_5349_equation_0 = const()[name = tensor("op_5349_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5349_cast_fp16 = einsum(equation = var_5349_equation_0, values = (var_5287_cast_fp16, var_5252_cast_fp16))[name = tensor("op_5349_cast_fp16")]; + tensor var_5350_to_fp16 = const()[name = tensor("op_5350_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_47_cast_fp16 = mul(x = var_5349_cast_fp16, y = var_5350_to_fp16)[name = tensor("aw_47_cast_fp16")]; + tensor var_5352_cast_fp16 = softmax(axis = var_4569, x = aw_33_cast_fp16)[name = tensor("op_5352_cast_fp16")]; + tensor var_5353_cast_fp16 = softmax(axis = var_4569, x = aw_35_cast_fp16)[name = tensor("op_5353_cast_fp16")]; + tensor var_5354_cast_fp16 = softmax(axis = var_4569, x = aw_37_cast_fp16)[name = tensor("op_5354_cast_fp16")]; + tensor var_5355_cast_fp16 = softmax(axis = var_4569, x = aw_39_cast_fp16)[name = tensor("op_5355_cast_fp16")]; + tensor var_5356_cast_fp16 = softmax(axis = var_4569, x = aw_41_cast_fp16)[name = tensor("op_5356_cast_fp16")]; + tensor var_5357_cast_fp16 = softmax(axis = var_4569, x = aw_43_cast_fp16)[name = tensor("op_5357_cast_fp16")]; + tensor var_5358_cast_fp16 = softmax(axis = var_4569, x = aw_45_cast_fp16)[name = tensor("op_5358_cast_fp16")]; + tensor var_5359_cast_fp16 = softmax(axis = var_4569, x = aw_47_cast_fp16)[name = tensor("op_5359_cast_fp16")]; + tensor var_5361_equation_0 = const()[name = tensor("op_5361_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5361_cast_fp16 = einsum(equation = var_5361_equation_0, values = (var_5289_cast_fp16, var_5352_cast_fp16))[name = tensor("op_5361_cast_fp16")]; + tensor var_5363_equation_0 = const()[name = tensor("op_5363_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5363_cast_fp16 = einsum(equation = var_5363_equation_0, values = (var_5293_cast_fp16, var_5353_cast_fp16))[name = tensor("op_5363_cast_fp16")]; + tensor var_5365_equation_0 = const()[name = tensor("op_5365_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5365_cast_fp16 = einsum(equation = var_5365_equation_0, values = (var_5297_cast_fp16, var_5354_cast_fp16))[name = tensor("op_5365_cast_fp16")]; + tensor var_5367_equation_0 = const()[name = tensor("op_5367_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5367_cast_fp16 = einsum(equation = var_5367_equation_0, values = (var_5301_cast_fp16, var_5355_cast_fp16))[name = tensor("op_5367_cast_fp16")]; + tensor var_5369_equation_0 = const()[name = tensor("op_5369_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5369_cast_fp16 = einsum(equation = var_5369_equation_0, values = (var_5305_cast_fp16, var_5356_cast_fp16))[name = tensor("op_5369_cast_fp16")]; + tensor var_5371_equation_0 = const()[name = tensor("op_5371_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5371_cast_fp16 = einsum(equation = var_5371_equation_0, values = (var_5309_cast_fp16, var_5357_cast_fp16))[name = tensor("op_5371_cast_fp16")]; + tensor var_5373_equation_0 = const()[name = tensor("op_5373_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5373_cast_fp16 = einsum(equation = var_5373_equation_0, values = (var_5313_cast_fp16, var_5358_cast_fp16))[name = tensor("op_5373_cast_fp16")]; + tensor var_5375_equation_0 = const()[name = tensor("op_5375_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5375_cast_fp16 = einsum(equation = var_5375_equation_0, values = (var_5317_cast_fp16, var_5359_cast_fp16))[name = tensor("op_5375_cast_fp16")]; + tensor input_159_interleave_0 = const()[name = tensor("input_159_interleave_0"), val = tensor(false)]; + tensor input_159_cast_fp16 = concat(axis = var_4569, interleave = input_159_interleave_0, values = (var_5361_cast_fp16, var_5363_cast_fp16, var_5365_cast_fp16, var_5367_cast_fp16, var_5369_cast_fp16, var_5371_cast_fp16, var_5373_cast_fp16, var_5375_cast_fp16))[name = tensor("input_159_cast_fp16")]; + tensor var_5385_pad_type_0 = const()[name = tensor("op_5385_pad_type_0"), val = tensor("valid")]; + tensor var_5385_strides_0 = const()[name = tensor("op_5385_strides_0"), val = tensor([1, 1])]; + tensor var_5385_pad_0 = const()[name = tensor("op_5385_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5385_dilations_0 = const()[name = tensor("op_5385_dilations_0"), val = tensor([1, 1])]; + tensor var_5385_groups_0 = const()[name = tensor("op_5385_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_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(292923264)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_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(296200128)))]; + tensor var_5385_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_5385_dilations_0, groups = var_5385_groups_0, pad = var_5385_pad_0, pad_type = var_5385_pad_type_0, strides = var_5385_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_159_cast_fp16)[name = tensor("op_5385_cast_fp16")]; + tensor inputs_33_cast_fp16 = add(x = var_5385_cast_fp16, y = inputs_31_cast_fp16)[name = tensor("inputs_33_cast_fp16")]; + tensor hidden_states_101_axes_0 = const()[name = tensor("hidden_states_101_axes_0"), val = tensor([1])]; + tensor hidden_states_101_gamma_0_to_fp16 = const()[name = tensor("hidden_states_101_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(296202752)))]; + tensor hidden_states_101_beta_0_to_fp16 = const()[name = tensor("hidden_states_101_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(296205376)))]; + tensor var_5395_to_fp16 = const()[name = tensor("op_5395_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_101_cast_fp16 = layer_norm(axes = hidden_states_101_axes_0, beta = hidden_states_101_beta_0_to_fp16, epsilon = var_5395_to_fp16, gamma = hidden_states_101_gamma_0_to_fp16, x = inputs_33_cast_fp16)[name = tensor("hidden_states_101_cast_fp16")]; + tensor q_23_pad_type_0 = const()[name = tensor("q_23_pad_type_0"), val = tensor("valid")]; + tensor q_23_strides_0 = const()[name = tensor("q_23_strides_0"), val = tensor([1, 1])]; + tensor q_23_pad_0 = const()[name = tensor("q_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_23_dilations_0 = const()[name = tensor("q_23_dilations_0"), val = tensor([1, 1])]; + tensor q_23_groups_0 = const()[name = tensor("q_23_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_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(296208000)))]; + tensor q_23_cast_fp16 = conv(dilations = q_23_dilations_0, groups = q_23_groups_0, pad = q_23_pad_0, pad_type = q_23_pad_type_0, strides = q_23_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_101_cast_fp16)[name = tensor("q_23_cast_fp16")]; + tensor k_45_pad_type_0 = const()[name = tensor("k_45_pad_type_0"), val = tensor("valid")]; + tensor k_45_strides_0 = const()[name = tensor("k_45_strides_0"), val = tensor([1, 1])]; + tensor k_45_pad_0 = const()[name = tensor("k_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_45_dilations_0 = const()[name = tensor("k_45_dilations_0"), val = tensor([1, 1])]; + tensor k_45_groups_0 = const()[name = tensor("k_45_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_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(299484864)))]; + tensor k_45_cast_fp16 = conv(dilations = k_45_dilations_0, groups = k_45_groups_0, pad = k_45_pad_0, pad_type = k_45_pad_type_0, strides = k_45_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_45_cast_fp16")]; + tensor v_23_pad_type_0 = const()[name = tensor("v_23_pad_type_0"), val = tensor("valid")]; + tensor v_23_strides_0 = const()[name = tensor("v_23_strides_0"), val = tensor([1, 1])]; + tensor v_23_pad_0 = const()[name = tensor("v_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_23_dilations_0 = const()[name = tensor("v_23_dilations_0"), val = tensor([1, 1])]; + tensor v_23_groups_0 = const()[name = tensor("v_23_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_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(301451008)))]; + tensor v_23_cast_fp16 = conv(dilations = v_23_dilations_0, groups = v_23_groups_0, pad = v_23_pad_0, pad_type = v_23_pad_type_0, strides = v_23_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_23_cast_fp16")]; + tensor var_5428_begin_0 = const()[name = tensor("op_5428_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5428_end_0 = const()[name = tensor("op_5428_end_0"), val = tensor([2, 160, 1, 256])]; + tensor var_5428_end_mask_0 = const()[name = tensor("op_5428_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5428_cast_fp16 = slice_by_index(begin = var_5428_begin_0, end = var_5428_end_0, end_mask = var_5428_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_5428_cast_fp16")]; + tensor var_5432_begin_0 = const()[name = tensor("op_5432_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_5432_end_0 = const()[name = tensor("op_5432_end_0"), val = tensor([2, 320, 1, 256])]; + tensor var_5432_end_mask_0 = const()[name = tensor("op_5432_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5432_cast_fp16 = slice_by_index(begin = var_5432_begin_0, end = var_5432_end_0, end_mask = var_5432_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_5432_cast_fp16")]; + tensor var_5436_begin_0 = const()[name = tensor("op_5436_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5436_end_0 = const()[name = tensor("op_5436_end_0"), val = tensor([2, 480, 1, 256])]; + tensor var_5436_end_mask_0 = const()[name = tensor("op_5436_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5436_cast_fp16 = slice_by_index(begin = var_5436_begin_0, end = var_5436_end_0, end_mask = var_5436_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_5436_cast_fp16")]; + tensor var_5440_begin_0 = const()[name = tensor("op_5440_begin_0"), val = tensor([0, 480, 0, 0])]; + tensor var_5440_end_0 = const()[name = tensor("op_5440_end_0"), val = tensor([2, 640, 1, 256])]; + tensor var_5440_end_mask_0 = const()[name = tensor("op_5440_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5440_cast_fp16 = slice_by_index(begin = var_5440_begin_0, end = var_5440_end_0, end_mask = var_5440_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_5440_cast_fp16")]; + tensor var_5444_begin_0 = const()[name = tensor("op_5444_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5444_end_0 = const()[name = tensor("op_5444_end_0"), val = tensor([2, 800, 1, 256])]; + tensor var_5444_end_mask_0 = const()[name = tensor("op_5444_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5444_cast_fp16 = slice_by_index(begin = var_5444_begin_0, end = var_5444_end_0, end_mask = var_5444_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_5444_cast_fp16")]; + tensor var_5448_begin_0 = const()[name = tensor("op_5448_begin_0"), val = tensor([0, 800, 0, 0])]; + tensor var_5448_end_0 = const()[name = tensor("op_5448_end_0"), val = tensor([2, 960, 1, 256])]; + tensor var_5448_end_mask_0 = const()[name = tensor("op_5448_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5448_cast_fp16 = slice_by_index(begin = var_5448_begin_0, end = var_5448_end_0, end_mask = var_5448_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_5448_cast_fp16")]; + tensor var_5452_begin_0 = const()[name = tensor("op_5452_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_5452_end_0 = const()[name = tensor("op_5452_end_0"), val = tensor([2, 1120, 1, 256])]; + tensor var_5452_end_mask_0 = const()[name = tensor("op_5452_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5452_cast_fp16 = slice_by_index(begin = var_5452_begin_0, end = var_5452_end_0, end_mask = var_5452_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_5452_cast_fp16")]; + tensor var_5456_begin_0 = const()[name = tensor("op_5456_begin_0"), val = tensor([0, 1120, 0, 0])]; + tensor var_5456_end_0 = const()[name = tensor("op_5456_end_0"), val = tensor([2, 1, 1, 256])]; + tensor var_5456_end_mask_0 = const()[name = tensor("op_5456_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_5456_cast_fp16 = slice_by_index(begin = var_5456_begin_0, end = var_5456_end_0, end_mask = var_5456_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_5456_cast_fp16")]; + tensor k_47_perm_0 = const()[name = tensor("k_47_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_5463_begin_0 = const()[name = tensor("op_5463_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5463_end_0 = const()[name = tensor("op_5463_end_0"), val = tensor([2, 77, 1, 160])]; + tensor var_5463_end_mask_0 = const()[name = tensor("op_5463_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_47_cast_fp16 = transpose(perm = k_47_perm_0, x = k_45_cast_fp16)[name = tensor("transpose_2")]; + tensor var_5463_cast_fp16 = slice_by_index(begin = var_5463_begin_0, end = var_5463_end_0, end_mask = var_5463_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_5463_cast_fp16")]; + tensor var_5467_begin_0 = const()[name = tensor("op_5467_begin_0"), val = tensor([0, 0, 0, 160])]; + tensor var_5467_end_0 = const()[name = tensor("op_5467_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_5467_end_mask_0 = const()[name = tensor("op_5467_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5467_cast_fp16 = slice_by_index(begin = var_5467_begin_0, end = var_5467_end_0, end_mask = var_5467_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_5467_cast_fp16")]; + tensor var_5471_begin_0 = const()[name = tensor("op_5471_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_5471_end_0 = const()[name = tensor("op_5471_end_0"), val = tensor([2, 77, 1, 480])]; + tensor var_5471_end_mask_0 = const()[name = tensor("op_5471_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5471_cast_fp16 = slice_by_index(begin = var_5471_begin_0, end = var_5471_end_0, end_mask = var_5471_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_5471_cast_fp16")]; + tensor var_5475_begin_0 = const()[name = tensor("op_5475_begin_0"), val = tensor([0, 0, 0, 480])]; + tensor var_5475_end_0 = const()[name = tensor("op_5475_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_5475_end_mask_0 = const()[name = tensor("op_5475_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5475_cast_fp16 = slice_by_index(begin = var_5475_begin_0, end = var_5475_end_0, end_mask = var_5475_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_5475_cast_fp16")]; + tensor var_5479_begin_0 = const()[name = tensor("op_5479_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_5479_end_0 = const()[name = tensor("op_5479_end_0"), val = tensor([2, 77, 1, 800])]; + tensor var_5479_end_mask_0 = const()[name = tensor("op_5479_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5479_cast_fp16 = slice_by_index(begin = var_5479_begin_0, end = var_5479_end_0, end_mask = var_5479_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_5479_cast_fp16")]; + tensor var_5483_begin_0 = const()[name = tensor("op_5483_begin_0"), val = tensor([0, 0, 0, 800])]; + tensor var_5483_end_0 = const()[name = tensor("op_5483_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_5483_end_mask_0 = const()[name = tensor("op_5483_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5483_cast_fp16 = slice_by_index(begin = var_5483_begin_0, end = var_5483_end_0, end_mask = var_5483_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_5483_cast_fp16")]; + tensor var_5487_begin_0 = const()[name = tensor("op_5487_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_5487_end_0 = const()[name = tensor("op_5487_end_0"), val = tensor([2, 77, 1, 1120])]; + tensor var_5487_end_mask_0 = const()[name = tensor("op_5487_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5487_cast_fp16 = slice_by_index(begin = var_5487_begin_0, end = var_5487_end_0, end_mask = var_5487_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_5487_cast_fp16")]; + tensor var_5491_begin_0 = const()[name = tensor("op_5491_begin_0"), val = tensor([0, 0, 0, 1120])]; + tensor var_5491_end_0 = const()[name = tensor("op_5491_end_0"), val = tensor([2, 77, 1, 1])]; + tensor var_5491_end_mask_0 = const()[name = tensor("op_5491_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_5491_cast_fp16 = slice_by_index(begin = var_5491_begin_0, end = var_5491_end_0, end_mask = var_5491_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_5491_cast_fp16")]; + tensor var_5493_begin_0 = const()[name = tensor("op_5493_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5493_end_0 = const()[name = tensor("op_5493_end_0"), val = tensor([2, 160, 1, 77])]; + tensor var_5493_end_mask_0 = const()[name = tensor("op_5493_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5493_cast_fp16 = slice_by_index(begin = var_5493_begin_0, end = var_5493_end_0, end_mask = var_5493_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_5493_cast_fp16")]; + tensor var_5497_begin_0 = const()[name = tensor("op_5497_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_5497_end_0 = const()[name = tensor("op_5497_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_5497_end_mask_0 = const()[name = tensor("op_5497_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5497_cast_fp16 = slice_by_index(begin = var_5497_begin_0, end = var_5497_end_0, end_mask = var_5497_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_5497_cast_fp16")]; + tensor var_5501_begin_0 = const()[name = tensor("op_5501_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5501_end_0 = const()[name = tensor("op_5501_end_0"), val = tensor([2, 480, 1, 77])]; + tensor var_5501_end_mask_0 = const()[name = tensor("op_5501_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5501_cast_fp16 = slice_by_index(begin = var_5501_begin_0, end = var_5501_end_0, end_mask = var_5501_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_5501_cast_fp16")]; + tensor var_5505_begin_0 = const()[name = tensor("op_5505_begin_0"), val = tensor([0, 480, 0, 0])]; + tensor var_5505_end_0 = const()[name = tensor("op_5505_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_5505_end_mask_0 = const()[name = tensor("op_5505_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5505_cast_fp16 = slice_by_index(begin = var_5505_begin_0, end = var_5505_end_0, end_mask = var_5505_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_5505_cast_fp16")]; + tensor var_5509_begin_0 = const()[name = tensor("op_5509_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5509_end_0 = const()[name = tensor("op_5509_end_0"), val = tensor([2, 800, 1, 77])]; + tensor var_5509_end_mask_0 = const()[name = tensor("op_5509_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5509_cast_fp16 = slice_by_index(begin = var_5509_begin_0, end = var_5509_end_0, end_mask = var_5509_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_5509_cast_fp16")]; + tensor var_5513_begin_0 = const()[name = tensor("op_5513_begin_0"), val = tensor([0, 800, 0, 0])]; + tensor var_5513_end_0 = const()[name = tensor("op_5513_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_5513_end_mask_0 = const()[name = tensor("op_5513_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5513_cast_fp16 = slice_by_index(begin = var_5513_begin_0, end = var_5513_end_0, end_mask = var_5513_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_5513_cast_fp16")]; + tensor var_5517_begin_0 = const()[name = tensor("op_5517_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_5517_end_0 = const()[name = tensor("op_5517_end_0"), val = tensor([2, 1120, 1, 77])]; + tensor var_5517_end_mask_0 = const()[name = tensor("op_5517_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5517_cast_fp16 = slice_by_index(begin = var_5517_begin_0, end = var_5517_end_0, end_mask = var_5517_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_5517_cast_fp16")]; + tensor var_5521_begin_0 = const()[name = tensor("op_5521_begin_0"), val = tensor([0, 1120, 0, 0])]; + tensor var_5521_end_0 = const()[name = tensor("op_5521_end_0"), val = tensor([2, 1, 1, 77])]; + tensor var_5521_end_mask_0 = const()[name = tensor("op_5521_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_5521_cast_fp16 = slice_by_index(begin = var_5521_begin_0, end = var_5521_end_0, end_mask = var_5521_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_5521_cast_fp16")]; + tensor var_5525_equation_0 = const()[name = tensor("op_5525_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5525_cast_fp16 = einsum(equation = var_5525_equation_0, values = (var_5463_cast_fp16, var_5428_cast_fp16))[name = tensor("op_5525_cast_fp16")]; + tensor var_5526_to_fp16 = const()[name = tensor("op_5526_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_49_cast_fp16 = mul(x = var_5525_cast_fp16, y = var_5526_to_fp16)[name = tensor("aw_49_cast_fp16")]; + tensor var_5529_equation_0 = const()[name = tensor("op_5529_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5529_cast_fp16 = einsum(equation = var_5529_equation_0, values = (var_5467_cast_fp16, var_5432_cast_fp16))[name = tensor("op_5529_cast_fp16")]; + tensor var_5530_to_fp16 = const()[name = tensor("op_5530_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_51_cast_fp16 = mul(x = var_5529_cast_fp16, y = var_5530_to_fp16)[name = tensor("aw_51_cast_fp16")]; + tensor var_5533_equation_0 = const()[name = tensor("op_5533_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5533_cast_fp16 = einsum(equation = var_5533_equation_0, values = (var_5471_cast_fp16, var_5436_cast_fp16))[name = tensor("op_5533_cast_fp16")]; + tensor var_5534_to_fp16 = const()[name = tensor("op_5534_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_53_cast_fp16 = mul(x = var_5533_cast_fp16, y = var_5534_to_fp16)[name = tensor("aw_53_cast_fp16")]; + tensor var_5537_equation_0 = const()[name = tensor("op_5537_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5537_cast_fp16 = einsum(equation = var_5537_equation_0, values = (var_5475_cast_fp16, var_5440_cast_fp16))[name = tensor("op_5537_cast_fp16")]; + tensor var_5538_to_fp16 = const()[name = tensor("op_5538_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_55_cast_fp16 = mul(x = var_5537_cast_fp16, y = var_5538_to_fp16)[name = tensor("aw_55_cast_fp16")]; + tensor var_5541_equation_0 = const()[name = tensor("op_5541_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5541_cast_fp16 = einsum(equation = var_5541_equation_0, values = (var_5479_cast_fp16, var_5444_cast_fp16))[name = tensor("op_5541_cast_fp16")]; + tensor var_5542_to_fp16 = const()[name = tensor("op_5542_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_57_cast_fp16 = mul(x = var_5541_cast_fp16, y = var_5542_to_fp16)[name = tensor("aw_57_cast_fp16")]; + tensor var_5545_equation_0 = const()[name = tensor("op_5545_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5545_cast_fp16 = einsum(equation = var_5545_equation_0, values = (var_5483_cast_fp16, var_5448_cast_fp16))[name = tensor("op_5545_cast_fp16")]; + tensor var_5546_to_fp16 = const()[name = tensor("op_5546_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_59_cast_fp16 = mul(x = var_5545_cast_fp16, y = var_5546_to_fp16)[name = tensor("aw_59_cast_fp16")]; + tensor var_5549_equation_0 = const()[name = tensor("op_5549_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5549_cast_fp16 = einsum(equation = var_5549_equation_0, values = (var_5487_cast_fp16, var_5452_cast_fp16))[name = tensor("op_5549_cast_fp16")]; + tensor var_5550_to_fp16 = const()[name = tensor("op_5550_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_61_cast_fp16 = mul(x = var_5549_cast_fp16, y = var_5550_to_fp16)[name = tensor("aw_61_cast_fp16")]; + tensor var_5553_equation_0 = const()[name = tensor("op_5553_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5553_cast_fp16 = einsum(equation = var_5553_equation_0, values = (var_5491_cast_fp16, var_5456_cast_fp16))[name = tensor("op_5553_cast_fp16")]; + tensor var_5554_to_fp16 = const()[name = tensor("op_5554_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_63_cast_fp16 = mul(x = var_5553_cast_fp16, y = var_5554_to_fp16)[name = tensor("aw_63_cast_fp16")]; + tensor var_5556_cast_fp16 = softmax(axis = var_4569, x = aw_49_cast_fp16)[name = tensor("op_5556_cast_fp16")]; + tensor var_5557_cast_fp16 = softmax(axis = var_4569, x = aw_51_cast_fp16)[name = tensor("op_5557_cast_fp16")]; + tensor var_5558_cast_fp16 = softmax(axis = var_4569, x = aw_53_cast_fp16)[name = tensor("op_5558_cast_fp16")]; + tensor var_5559_cast_fp16 = softmax(axis = var_4569, x = aw_55_cast_fp16)[name = tensor("op_5559_cast_fp16")]; + tensor var_5560_cast_fp16 = softmax(axis = var_4569, x = aw_57_cast_fp16)[name = tensor("op_5560_cast_fp16")]; + tensor var_5561_cast_fp16 = softmax(axis = var_4569, x = aw_59_cast_fp16)[name = tensor("op_5561_cast_fp16")]; + tensor var_5562_cast_fp16 = softmax(axis = var_4569, x = aw_61_cast_fp16)[name = tensor("op_5562_cast_fp16")]; + tensor var_5563_cast_fp16 = softmax(axis = var_4569, x = aw_63_cast_fp16)[name = tensor("op_5563_cast_fp16")]; + tensor var_5565_equation_0 = const()[name = tensor("op_5565_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5565_cast_fp16 = einsum(equation = var_5565_equation_0, values = (var_5493_cast_fp16, var_5556_cast_fp16))[name = tensor("op_5565_cast_fp16")]; + tensor var_5567_equation_0 = const()[name = tensor("op_5567_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5567_cast_fp16 = einsum(equation = var_5567_equation_0, values = (var_5497_cast_fp16, var_5557_cast_fp16))[name = tensor("op_5567_cast_fp16")]; + tensor var_5569_equation_0 = const()[name = tensor("op_5569_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5569_cast_fp16 = einsum(equation = var_5569_equation_0, values = (var_5501_cast_fp16, var_5558_cast_fp16))[name = tensor("op_5569_cast_fp16")]; + tensor var_5571_equation_0 = const()[name = tensor("op_5571_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5571_cast_fp16 = einsum(equation = var_5571_equation_0, values = (var_5505_cast_fp16, var_5559_cast_fp16))[name = tensor("op_5571_cast_fp16")]; + tensor var_5573_equation_0 = const()[name = tensor("op_5573_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5573_cast_fp16 = einsum(equation = var_5573_equation_0, values = (var_5509_cast_fp16, var_5560_cast_fp16))[name = tensor("op_5573_cast_fp16")]; + tensor var_5575_equation_0 = const()[name = tensor("op_5575_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5575_cast_fp16 = einsum(equation = var_5575_equation_0, values = (var_5513_cast_fp16, var_5561_cast_fp16))[name = tensor("op_5575_cast_fp16")]; + tensor var_5577_equation_0 = const()[name = tensor("op_5577_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5577_cast_fp16 = einsum(equation = var_5577_equation_0, values = (var_5517_cast_fp16, var_5562_cast_fp16))[name = tensor("op_5577_cast_fp16")]; + tensor var_5579_equation_0 = const()[name = tensor("op_5579_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5579_cast_fp16 = einsum(equation = var_5579_equation_0, values = (var_5521_cast_fp16, var_5563_cast_fp16))[name = tensor("op_5579_cast_fp16")]; + tensor input_161_interleave_0 = const()[name = tensor("input_161_interleave_0"), val = tensor(false)]; + tensor input_161_cast_fp16 = concat(axis = var_4569, interleave = input_161_interleave_0, values = (var_5565_cast_fp16, var_5567_cast_fp16, var_5569_cast_fp16, var_5571_cast_fp16, var_5573_cast_fp16, var_5575_cast_fp16, var_5577_cast_fp16, var_5579_cast_fp16))[name = tensor("input_161_cast_fp16")]; + tensor var_5589_pad_type_0 = const()[name = tensor("op_5589_pad_type_0"), val = tensor("valid")]; + tensor var_5589_strides_0 = const()[name = tensor("op_5589_strides_0"), val = tensor([1, 1])]; + tensor var_5589_pad_0 = const()[name = tensor("op_5589_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5589_dilations_0 = const()[name = tensor("op_5589_dilations_0"), val = tensor([1, 1])]; + tensor var_5589_groups_0 = const()[name = tensor("op_5589_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_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(303417152)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_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(306694016)))]; + tensor var_5589_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_5589_dilations_0, groups = var_5589_groups_0, pad = var_5589_pad_0, pad_type = var_5589_pad_type_0, strides = var_5589_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_161_cast_fp16)[name = tensor("op_5589_cast_fp16")]; + tensor inputs_35_cast_fp16 = add(x = var_5589_cast_fp16, y = inputs_33_cast_fp16)[name = tensor("inputs_35_cast_fp16")]; + tensor input_163_axes_0 = const()[name = tensor("input_163_axes_0"), val = tensor([1])]; + tensor input_163_gamma_0_to_fp16 = const()[name = tensor("input_163_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(306696640)))]; + tensor input_163_beta_0_to_fp16 = const()[name = tensor("input_163_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(306699264)))]; + tensor var_5599_to_fp16 = const()[name = tensor("op_5599_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_163_cast_fp16 = layer_norm(axes = input_163_axes_0, beta = input_163_beta_0_to_fp16, epsilon = var_5599_to_fp16, gamma = input_163_gamma_0_to_fp16, x = inputs_35_cast_fp16)[name = tensor("input_163_cast_fp16")]; + tensor var_5619_pad_type_0 = const()[name = tensor("op_5619_pad_type_0"), val = tensor("valid")]; + tensor var_5619_strides_0 = const()[name = tensor("op_5619_strides_0"), val = tensor([1, 1])]; + tensor var_5619_pad_0 = const()[name = tensor("op_5619_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5619_dilations_0 = const()[name = tensor("op_5619_dilations_0"), val = tensor([1, 1])]; + tensor var_5619_groups_0 = const()[name = tensor("op_5619_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_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(306701888)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_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(332916352)))]; + tensor var_5619_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_5619_dilations_0, groups = var_5619_groups_0, pad = var_5619_pad_0, pad_type = var_5619_pad_type_0, strides = var_5619_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_163_cast_fp16)[name = tensor("op_5619_cast_fp16")]; + tensor var_5620_split_sizes_0 = const()[name = tensor("op_5620_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_5620_axis_0 = const()[name = tensor("op_5620_axis_0"), val = tensor(1)]; + tensor var_5620_cast_fp16_0, tensor var_5620_cast_fp16_1 = split(axis = var_5620_axis_0, split_sizes = var_5620_split_sizes_0, x = var_5619_cast_fp16)[name = tensor("op_5620_cast_fp16")]; + tensor var_5622_mode_0 = const()[name = tensor("op_5622_mode_0"), val = tensor("EXACT")]; + tensor var_5622_cast_fp16 = gelu(mode = var_5622_mode_0, x = var_5620_cast_fp16_1)[name = tensor("op_5622_cast_fp16")]; + tensor input_165_cast_fp16 = mul(x = var_5620_cast_fp16_0, y = var_5622_cast_fp16)[name = tensor("input_165_cast_fp16")]; + tensor var_5630_pad_type_0 = const()[name = tensor("op_5630_pad_type_0"), val = tensor("valid")]; + tensor var_5630_strides_0 = const()[name = tensor("op_5630_strides_0"), val = tensor([1, 1])]; + tensor var_5630_pad_0 = const()[name = tensor("op_5630_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5630_dilations_0 = const()[name = tensor("op_5630_dilations_0"), val = tensor([1, 1])]; + tensor var_5630_groups_0 = const()[name = tensor("op_5630_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("down_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(332936896)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("down_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(346044160)))]; + tensor var_5630_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_5630_dilations_0, groups = var_5630_groups_0, pad = var_5630_pad_0, pad_type = var_5630_pad_type_0, strides = var_5630_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_165_cast_fp16)[name = tensor("op_5630_cast_fp16")]; + tensor hidden_states_105_cast_fp16 = add(x = var_5630_cast_fp16, y = inputs_35_cast_fp16)[name = tensor("hidden_states_105_cast_fp16")]; + tensor var_5632 = const()[name = tensor("op_5632"), val = tensor([2, 1280, 16, 16])]; + tensor input_167_cast_fp16 = reshape(shape = var_5632, x = hidden_states_105_cast_fp16)[name = tensor("input_167_cast_fp16")]; + tensor hidden_states_107_pad_type_0 = const()[name = tensor("hidden_states_107_pad_type_0"), val = tensor("valid")]; + tensor hidden_states_107_strides_0 = const()[name = tensor("hidden_states_107_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_107_pad_0 = const()[name = tensor("hidden_states_107_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_107_dilations_0 = const()[name = tensor("hidden_states_107_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_107_groups_0 = const()[name = tensor("hidden_states_107_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_proj_out_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(346046784)))]; + tensor down_blocks_2_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(349323648)))]; + tensor hidden_states_107_cast_fp16 = conv(bias = down_blocks_2_attentions_1_proj_out_bias_to_fp16, dilations = hidden_states_107_dilations_0, groups = hidden_states_107_groups_0, pad = hidden_states_107_pad_0, pad_type = hidden_states_107_pad_type_0, strides = hidden_states_107_strides_0, weight = down_blocks_2_attentions_1_proj_out_weight_to_fp16, x = input_167_cast_fp16)[name = tensor("hidden_states_107_cast_fp16")]; + tensor input_169_cast_fp16_1 = add(x = hidden_states_107_cast_fp16, y = hidden_states_95_cast_fp16)[name = tensor("input_169_cast_fp16")]; + tensor input_171_pad_type_0 = const()[name = tensor("input_171_pad_type_0"), val = tensor("custom")]; + tensor input_171_pad_0 = const()[name = tensor("input_171_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor input_171_strides_0 = const()[name = tensor("input_171_strides_0"), val = tensor([2, 2])]; + tensor input_171_dilations_0 = const()[name = tensor("input_171_dilations_0"), val = tensor([1, 1])]; + tensor input_171_groups_0 = const()[name = tensor("input_171_groups_0"), val = tensor(1)]; + tensor down_blocks_2_downsamplers_0_conv_weight_to_fp16 = const()[name = tensor("down_blocks_2_downsamplers_0_conv_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(349326272)))]; + tensor down_blocks_2_downsamplers_0_conv_bias_to_fp16 = const()[name = tensor("down_blocks_2_downsamplers_0_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(378817536)))]; + tensor input_171_cast_fp16_1 = conv(bias = down_blocks_2_downsamplers_0_conv_bias_to_fp16, dilations = input_171_dilations_0, groups = input_171_groups_0, pad = input_171_pad_0, pad_type = input_171_pad_type_0, strides = input_171_strides_0, weight = down_blocks_2_downsamplers_0_conv_weight_to_fp16, x = input_169_cast_fp16_1)[name = tensor("input_171_cast_fp16")]; + tensor reshape_72_shape_0 = const()[name = tensor("reshape_72_shape_0"), val = tensor([2, 32, 40, 8, 8])]; + tensor reshape_72_cast_fp16 = reshape(shape = reshape_72_shape_0, x = input_171_cast_fp16_1)[name = tensor("reshape_72_cast_fp16")]; + tensor reduce_mean_54_axes_0 = const()[name = tensor("reduce_mean_54_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_54_keep_dims_0 = const()[name = tensor("reduce_mean_54_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_54_cast_fp16 = reduce_mean(axes = reduce_mean_54_axes_0, keep_dims = reduce_mean_54_keep_dims_0, x = reshape_72_cast_fp16)[name = tensor("reduce_mean_54_cast_fp16")]; + tensor sub_36_cast_fp16 = sub(x = reshape_72_cast_fp16, y = reduce_mean_54_cast_fp16)[name = tensor("sub_36_cast_fp16")]; + tensor square_18_cast_fp16 = square(x = sub_36_cast_fp16)[name = tensor("square_18_cast_fp16")]; + tensor reduce_mean_56_axes_0 = const()[name = tensor("reduce_mean_56_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_56_keep_dims_0 = const()[name = tensor("reduce_mean_56_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_56_cast_fp16 = reduce_mean(axes = reduce_mean_56_axes_0, keep_dims = reduce_mean_56_keep_dims_0, x = square_18_cast_fp16)[name = tensor("reduce_mean_56_cast_fp16")]; + tensor add_36_y_0_to_fp16 = const()[name = tensor("add_36_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_36_cast_fp16 = add(x = reduce_mean_56_cast_fp16, y = add_36_y_0_to_fp16)[name = tensor("add_36_cast_fp16")]; + tensor sqrt_18_cast_fp16 = sqrt(x = add_36_cast_fp16)[name = tensor("sqrt_18_cast_fp16")]; + tensor real_div_18_cast_fp16 = real_div(x = sub_36_cast_fp16, y = sqrt_18_cast_fp16)[name = tensor("real_div_18_cast_fp16")]; + tensor reshape_73_shape_0 = const()[name = tensor("reshape_73_shape_0"), val = tensor([2, 1280, 8, 8])]; + tensor reshape_73_cast_fp16 = reshape(shape = reshape_73_shape_0, x = real_div_18_cast_fp16)[name = tensor("reshape_73_cast_fp16")]; + tensor add_37_gamma_0_to_fp16 = const()[name = tensor("add_37_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(378820160)))]; + tensor add_37_beta_0_to_fp16 = const()[name = tensor("add_37_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(378822784)))]; + tensor add_37_epsilon_0_to_fp16 = const()[name = tensor("add_37_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_37_cast_fp16 = batch_norm(beta = add_37_beta_0_to_fp16, epsilon = add_37_epsilon_0_to_fp16, gamma = add_37_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_73_cast_fp16)[name = tensor("add_37_cast_fp16")]; + tensor input_175_cast_fp16 = silu(x = add_37_cast_fp16)[name = tensor("input_175_cast_fp16")]; + tensor hidden_states_109_pad_type_0 = const()[name = tensor("hidden_states_109_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_109_pad_0 = const()[name = tensor("hidden_states_109_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_109_strides_0 = const()[name = tensor("hidden_states_109_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_109_dilations_0 = const()[name = tensor("hidden_states_109_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_109_groups_0 = const()[name = tensor("hidden_states_109_groups_0"), val = tensor(1)]; + tensor down_blocks_3_resnets_0_conv1_weight_to_fp16 = const()[name = tensor("down_blocks_3_resnets_0_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(378825408)))]; + tensor down_blocks_3_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_3_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(408316672)))]; + tensor hidden_states_109_cast_fp16 = conv(bias = down_blocks_3_resnets_0_conv1_bias_to_fp16, dilations = hidden_states_109_dilations_0, groups = hidden_states_109_groups_0, pad = hidden_states_109_pad_0, pad_type = hidden_states_109_pad_type_0, strides = hidden_states_109_strides_0, weight = down_blocks_3_resnets_0_conv1_weight_to_fp16, x = input_175_cast_fp16)[name = tensor("hidden_states_109_cast_fp16")]; + tensor temb_13_pad_type_0 = const()[name = tensor("temb_13_pad_type_0"), val = tensor("valid")]; + tensor temb_13_strides_0 = const()[name = tensor("temb_13_strides_0"), val = tensor([1, 1])]; + tensor temb_13_pad_0 = const()[name = tensor("temb_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor temb_13_dilations_0 = const()[name = tensor("temb_13_dilations_0"), val = tensor([1, 1])]; + tensor temb_13_groups_0 = const()[name = tensor("temb_13_groups_0"), val = tensor(1)]; + tensor down_blocks_3_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor("down_blocks_3_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(408319296)))]; + tensor down_blocks_3_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_3_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411596160)))]; + tensor temb_13_cast_fp16 = conv(bias = down_blocks_3_resnets_0_time_emb_proj_bias_to_fp16, dilations = temb_13_dilations_0, groups = temb_13_groups_0, pad = temb_13_pad_0, pad_type = temb_13_pad_type_0, strides = temb_13_strides_0, weight = down_blocks_3_resnets_0_time_emb_proj_weight_to_fp16, x = input_15_cast_fp16_1)[name = tensor("temb_13_cast_fp16")]; + tensor input_179_cast_fp16 = add(x = hidden_states_109_cast_fp16, y = temb_13_cast_fp16)[name = tensor("input_179_cast_fp16")]; + tensor reshape_76_shape_0 = const()[name = tensor("reshape_76_shape_0"), val = tensor([2, 32, 40, 8, 8])]; + tensor reshape_76_cast_fp16 = reshape(shape = reshape_76_shape_0, x = input_179_cast_fp16)[name = tensor("reshape_76_cast_fp16")]; + tensor reduce_mean_57_axes_0 = const()[name = tensor("reduce_mean_57_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_57_keep_dims_0 = const()[name = tensor("reduce_mean_57_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_57_cast_fp16 = reduce_mean(axes = reduce_mean_57_axes_0, keep_dims = reduce_mean_57_keep_dims_0, x = reshape_76_cast_fp16)[name = tensor("reduce_mean_57_cast_fp16")]; + tensor sub_38_cast_fp16 = sub(x = reshape_76_cast_fp16, y = reduce_mean_57_cast_fp16)[name = tensor("sub_38_cast_fp16")]; + tensor square_19_cast_fp16 = square(x = sub_38_cast_fp16)[name = tensor("square_19_cast_fp16")]; + tensor reduce_mean_59_axes_0 = const()[name = tensor("reduce_mean_59_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_59_keep_dims_0 = const()[name = tensor("reduce_mean_59_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_59_cast_fp16 = reduce_mean(axes = reduce_mean_59_axes_0, keep_dims = reduce_mean_59_keep_dims_0, x = square_19_cast_fp16)[name = tensor("reduce_mean_59_cast_fp16")]; + tensor add_38_y_0_to_fp16 = const()[name = tensor("add_38_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_38_cast_fp16 = add(x = reduce_mean_59_cast_fp16, y = add_38_y_0_to_fp16)[name = tensor("add_38_cast_fp16")]; + tensor sqrt_19_cast_fp16 = sqrt(x = add_38_cast_fp16)[name = tensor("sqrt_19_cast_fp16")]; + tensor real_div_19_cast_fp16 = real_div(x = sub_38_cast_fp16, y = sqrt_19_cast_fp16)[name = tensor("real_div_19_cast_fp16")]; + tensor reshape_77_shape_0 = const()[name = tensor("reshape_77_shape_0"), val = tensor([2, 1280, 8, 8])]; + tensor reshape_77_cast_fp16 = reshape(shape = reshape_77_shape_0, x = real_div_19_cast_fp16)[name = tensor("reshape_77_cast_fp16")]; + tensor add_39_gamma_0_to_fp16 = const()[name = tensor("add_39_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411598784)))]; + tensor add_39_beta_0_to_fp16 = const()[name = tensor("add_39_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411601408)))]; + tensor add_39_epsilon_0_to_fp16 = const()[name = tensor("add_39_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_39_cast_fp16 = batch_norm(beta = add_39_beta_0_to_fp16, epsilon = add_39_epsilon_0_to_fp16, gamma = add_39_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_77_cast_fp16)[name = tensor("add_39_cast_fp16")]; + tensor input_183_cast_fp16 = silu(x = add_39_cast_fp16)[name = tensor("input_183_cast_fp16")]; + tensor hidden_states_111_pad_type_0 = const()[name = tensor("hidden_states_111_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_111_pad_0 = const()[name = tensor("hidden_states_111_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_111_strides_0 = const()[name = tensor("hidden_states_111_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_111_dilations_0 = const()[name = tensor("hidden_states_111_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_111_groups_0 = const()[name = tensor("hidden_states_111_groups_0"), val = tensor(1)]; + tensor down_blocks_3_resnets_0_conv2_weight_to_fp16 = const()[name = tensor("down_blocks_3_resnets_0_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411604032)))]; + tensor down_blocks_3_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_3_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(441095296)))]; + tensor hidden_states_111_cast_fp16 = conv(bias = down_blocks_3_resnets_0_conv2_bias_to_fp16, dilations = hidden_states_111_dilations_0, groups = hidden_states_111_groups_0, pad = hidden_states_111_pad_0, pad_type = hidden_states_111_pad_type_0, strides = hidden_states_111_strides_0, weight = down_blocks_3_resnets_0_conv2_weight_to_fp16, x = input_183_cast_fp16)[name = tensor("hidden_states_111_cast_fp16")]; + tensor input_185_cast_fp16 = add(x = input_171_cast_fp16_1, y = hidden_states_111_cast_fp16)[name = tensor("input_185_cast_fp16")]; + tensor reshape_80_shape_0 = const()[name = tensor("reshape_80_shape_0"), val = tensor([2, 32, 40, 8, 8])]; + tensor reshape_80_cast_fp16 = reshape(shape = reshape_80_shape_0, x = input_185_cast_fp16)[name = tensor("reshape_80_cast_fp16")]; + tensor reduce_mean_60_axes_0 = const()[name = tensor("reduce_mean_60_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_60_keep_dims_0 = const()[name = tensor("reduce_mean_60_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_60_cast_fp16 = reduce_mean(axes = reduce_mean_60_axes_0, keep_dims = reduce_mean_60_keep_dims_0, x = reshape_80_cast_fp16)[name = tensor("reduce_mean_60_cast_fp16")]; + tensor sub_40_cast_fp16 = sub(x = reshape_80_cast_fp16, y = reduce_mean_60_cast_fp16)[name = tensor("sub_40_cast_fp16")]; + tensor square_20_cast_fp16 = square(x = sub_40_cast_fp16)[name = tensor("square_20_cast_fp16")]; + tensor reduce_mean_62_axes_0 = const()[name = tensor("reduce_mean_62_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_62_keep_dims_0 = const()[name = tensor("reduce_mean_62_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_62_cast_fp16 = reduce_mean(axes = reduce_mean_62_axes_0, keep_dims = reduce_mean_62_keep_dims_0, x = square_20_cast_fp16)[name = tensor("reduce_mean_62_cast_fp16")]; + tensor add_40_y_0_to_fp16 = const()[name = tensor("add_40_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_40_cast_fp16 = add(x = reduce_mean_62_cast_fp16, y = add_40_y_0_to_fp16)[name = tensor("add_40_cast_fp16")]; + tensor sqrt_20_cast_fp16 = sqrt(x = add_40_cast_fp16)[name = tensor("sqrt_20_cast_fp16")]; + tensor real_div_20_cast_fp16 = real_div(x = sub_40_cast_fp16, y = sqrt_20_cast_fp16)[name = tensor("real_div_20_cast_fp16")]; + tensor reshape_81_shape_0 = const()[name = tensor("reshape_81_shape_0"), val = tensor([2, 1280, 8, 8])]; + tensor reshape_81_cast_fp16 = reshape(shape = reshape_81_shape_0, x = real_div_20_cast_fp16)[name = tensor("reshape_81_cast_fp16")]; + tensor add_41_gamma_0_to_fp16 = const()[name = tensor("add_41_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(441097920)))]; + tensor add_41_beta_0_to_fp16 = const()[name = tensor("add_41_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(441100544)))]; + tensor add_41_epsilon_0_to_fp16 = const()[name = tensor("add_41_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_41_cast_fp16 = batch_norm(beta = add_41_beta_0_to_fp16, epsilon = add_41_epsilon_0_to_fp16, gamma = add_41_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_81_cast_fp16)[name = tensor("add_41_cast_fp16")]; + tensor input_189_cast_fp16 = silu(x = add_41_cast_fp16)[name = tensor("input_189_cast_fp16")]; + tensor hidden_states_113_pad_type_0 = const()[name = tensor("hidden_states_113_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_113_pad_0 = const()[name = tensor("hidden_states_113_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_113_strides_0 = const()[name = tensor("hidden_states_113_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_113_dilations_0 = const()[name = tensor("hidden_states_113_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_113_groups_0 = const()[name = tensor("hidden_states_113_groups_0"), val = tensor(1)]; + tensor down_blocks_3_resnets_1_conv1_weight_to_fp16 = const()[name = tensor("down_blocks_3_resnets_1_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(441103168)))]; + tensor down_blocks_3_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_3_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(470594432)))]; + tensor hidden_states_113_cast_fp16 = conv(bias = down_blocks_3_resnets_1_conv1_bias_to_fp16, dilations = hidden_states_113_dilations_0, groups = hidden_states_113_groups_0, pad = hidden_states_113_pad_0, pad_type = hidden_states_113_pad_type_0, strides = hidden_states_113_strides_0, weight = down_blocks_3_resnets_1_conv1_weight_to_fp16, x = input_189_cast_fp16)[name = tensor("hidden_states_113_cast_fp16")]; + tensor temb_15_pad_type_0 = const()[name = tensor("temb_15_pad_type_0"), val = tensor("valid")]; + tensor temb_15_strides_0 = const()[name = tensor("temb_15_strides_0"), val = tensor([1, 1])]; + tensor temb_15_pad_0 = const()[name = tensor("temb_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor temb_15_dilations_0 = const()[name = tensor("temb_15_dilations_0"), val = tensor([1, 1])]; + tensor temb_15_groups_0 = const()[name = tensor("temb_15_groups_0"), val = tensor(1)]; + tensor down_blocks_3_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor("down_blocks_3_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(470597056)))]; + tensor down_blocks_3_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_3_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(473873920)))]; + tensor temb_15_cast_fp16 = conv(bias = down_blocks_3_resnets_1_time_emb_proj_bias_to_fp16, dilations = temb_15_dilations_0, groups = temb_15_groups_0, pad = temb_15_pad_0, pad_type = temb_15_pad_type_0, strides = temb_15_strides_0, weight = down_blocks_3_resnets_1_time_emb_proj_weight_to_fp16, x = input_15_cast_fp16_1)[name = tensor("temb_15_cast_fp16")]; + tensor input_193_cast_fp16 = add(x = hidden_states_113_cast_fp16, y = temb_15_cast_fp16)[name = tensor("input_193_cast_fp16")]; + tensor reshape_84_shape_0 = const()[name = tensor("reshape_84_shape_0"), val = tensor([2, 32, 40, 8, 8])]; + tensor reshape_84_cast_fp16 = reshape(shape = reshape_84_shape_0, x = input_193_cast_fp16)[name = tensor("reshape_84_cast_fp16")]; + tensor reduce_mean_63_axes_0 = const()[name = tensor("reduce_mean_63_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_63_keep_dims_0 = const()[name = tensor("reduce_mean_63_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_63_cast_fp16 = reduce_mean(axes = reduce_mean_63_axes_0, keep_dims = reduce_mean_63_keep_dims_0, x = reshape_84_cast_fp16)[name = tensor("reduce_mean_63_cast_fp16")]; + tensor sub_42_cast_fp16 = sub(x = reshape_84_cast_fp16, y = reduce_mean_63_cast_fp16)[name = tensor("sub_42_cast_fp16")]; + tensor square_21_cast_fp16 = square(x = sub_42_cast_fp16)[name = tensor("square_21_cast_fp16")]; + tensor reduce_mean_65_axes_0 = const()[name = tensor("reduce_mean_65_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_65_keep_dims_0 = const()[name = tensor("reduce_mean_65_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_65_cast_fp16 = reduce_mean(axes = reduce_mean_65_axes_0, keep_dims = reduce_mean_65_keep_dims_0, x = square_21_cast_fp16)[name = tensor("reduce_mean_65_cast_fp16")]; + tensor add_42_y_0_to_fp16 = const()[name = tensor("add_42_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_42_cast_fp16 = add(x = reduce_mean_65_cast_fp16, y = add_42_y_0_to_fp16)[name = tensor("add_42_cast_fp16")]; + tensor sqrt_21_cast_fp16 = sqrt(x = add_42_cast_fp16)[name = tensor("sqrt_21_cast_fp16")]; + tensor real_div_21_cast_fp16 = real_div(x = sub_42_cast_fp16, y = sqrt_21_cast_fp16)[name = tensor("real_div_21_cast_fp16")]; + tensor reshape_85_shape_0 = const()[name = tensor("reshape_85_shape_0"), val = tensor([2, 1280, 8, 8])]; + tensor reshape_85_cast_fp16 = reshape(shape = reshape_85_shape_0, x = real_div_21_cast_fp16)[name = tensor("reshape_85_cast_fp16")]; + tensor add_43_gamma_0_to_fp16 = const()[name = tensor("add_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(473876544)))]; + tensor add_43_beta_0_to_fp16 = const()[name = tensor("add_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(473879168)))]; + tensor add_43_epsilon_0_to_fp16 = const()[name = tensor("add_43_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_43_cast_fp16 = batch_norm(beta = add_43_beta_0_to_fp16, epsilon = add_43_epsilon_0_to_fp16, gamma = add_43_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_85_cast_fp16)[name = tensor("add_43_cast_fp16")]; + tensor input_197_cast_fp16 = silu(x = add_43_cast_fp16)[name = tensor("input_197_cast_fp16")]; + tensor hidden_states_115_pad_type_0 = const()[name = tensor("hidden_states_115_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_115_pad_0 = const()[name = tensor("hidden_states_115_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_115_strides_0 = const()[name = tensor("hidden_states_115_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_115_dilations_0 = const()[name = tensor("hidden_states_115_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_115_groups_0 = const()[name = tensor("hidden_states_115_groups_0"), val = tensor(1)]; + tensor down_blocks_3_resnets_1_conv2_weight_to_fp16 = const()[name = tensor("down_blocks_3_resnets_1_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(473881792)))]; + tensor down_blocks_3_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_3_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(503373056)))]; + tensor hidden_states_115_cast_fp16 = conv(bias = down_blocks_3_resnets_1_conv2_bias_to_fp16, dilations = hidden_states_115_dilations_0, groups = hidden_states_115_groups_0, pad = hidden_states_115_pad_0, pad_type = hidden_states_115_pad_type_0, strides = hidden_states_115_strides_0, weight = down_blocks_3_resnets_1_conv2_weight_to_fp16, x = input_197_cast_fp16)[name = tensor("hidden_states_115_cast_fp16")]; + tensor input_199_cast_fp16 = add(x = input_185_cast_fp16, y = hidden_states_115_cast_fp16)[name = tensor("input_199_cast_fp16")]; + tensor var_5762 = const()[name = tensor("op_5762"), val = tensor(1)]; + tensor reshape_88_shape_0 = const()[name = tensor("reshape_88_shape_0"), val = tensor([2, 32, 40, 8, 8])]; + tensor reshape_88_cast_fp16 = reshape(shape = reshape_88_shape_0, x = input_199_cast_fp16)[name = tensor("reshape_88_cast_fp16")]; + tensor reduce_mean_66_axes_0 = const()[name = tensor("reduce_mean_66_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_66_keep_dims_0 = const()[name = tensor("reduce_mean_66_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_66_cast_fp16 = reduce_mean(axes = reduce_mean_66_axes_0, keep_dims = reduce_mean_66_keep_dims_0, x = reshape_88_cast_fp16)[name = tensor("reduce_mean_66_cast_fp16")]; + tensor sub_44_cast_fp16 = sub(x = reshape_88_cast_fp16, y = reduce_mean_66_cast_fp16)[name = tensor("sub_44_cast_fp16")]; + tensor square_22_cast_fp16 = square(x = sub_44_cast_fp16)[name = tensor("square_22_cast_fp16")]; + tensor reduce_mean_68_axes_0 = const()[name = tensor("reduce_mean_68_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_68_keep_dims_0 = const()[name = tensor("reduce_mean_68_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_68_cast_fp16 = reduce_mean(axes = reduce_mean_68_axes_0, keep_dims = reduce_mean_68_keep_dims_0, x = square_22_cast_fp16)[name = tensor("reduce_mean_68_cast_fp16")]; + tensor add_44_y_0_to_fp16 = const()[name = tensor("add_44_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_44_cast_fp16 = add(x = reduce_mean_68_cast_fp16, y = add_44_y_0_to_fp16)[name = tensor("add_44_cast_fp16")]; + tensor sqrt_22_cast_fp16 = sqrt(x = add_44_cast_fp16)[name = tensor("sqrt_22_cast_fp16")]; + tensor real_div_22_cast_fp16 = real_div(x = sub_44_cast_fp16, y = sqrt_22_cast_fp16)[name = tensor("real_div_22_cast_fp16")]; + tensor reshape_89_shape_0 = const()[name = tensor("reshape_89_shape_0"), val = tensor([2, 1280, 8, 8])]; + tensor reshape_89_cast_fp16 = reshape(shape = reshape_89_shape_0, x = real_div_22_cast_fp16)[name = tensor("reshape_89_cast_fp16")]; + tensor add_45_gamma_0_to_fp16 = const()[name = tensor("add_45_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(503375680)))]; + tensor add_45_beta_0_to_fp16 = const()[name = tensor("add_45_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(503378304)))]; + tensor add_45_epsilon_0_to_fp16 = const()[name = tensor("add_45_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_45_cast_fp16 = batch_norm(beta = add_45_beta_0_to_fp16, epsilon = add_45_epsilon_0_to_fp16, gamma = add_45_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_89_cast_fp16)[name = tensor("add_45_cast_fp16")]; + tensor input_203_cast_fp16 = silu(x = add_45_cast_fp16)[name = tensor("input_203_cast_fp16")]; + tensor hidden_states_117_pad_type_0 = const()[name = tensor("hidden_states_117_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_117_pad_0 = const()[name = tensor("hidden_states_117_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_117_strides_0 = const()[name = tensor("hidden_states_117_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_117_dilations_0 = const()[name = tensor("hidden_states_117_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_117_groups_0 = const()[name = tensor("hidden_states_117_groups_0"), val = tensor(1)]; + tensor mid_block_resnets_0_conv1_weight_to_fp16 = const()[name = tensor("mid_block_resnets_0_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(503380928)))]; + tensor mid_block_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("mid_block_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(532872192)))]; + tensor hidden_states_117_cast_fp16 = conv(bias = mid_block_resnets_0_conv1_bias_to_fp16, dilations = hidden_states_117_dilations_0, groups = hidden_states_117_groups_0, pad = hidden_states_117_pad_0, pad_type = hidden_states_117_pad_type_0, strides = hidden_states_117_strides_0, weight = mid_block_resnets_0_conv1_weight_to_fp16, x = input_203_cast_fp16)[name = tensor("hidden_states_117_cast_fp16")]; + tensor temb_17_pad_type_0 = const()[name = tensor("temb_17_pad_type_0"), val = tensor("valid")]; + tensor temb_17_strides_0 = const()[name = tensor("temb_17_strides_0"), val = tensor([1, 1])]; + tensor temb_17_pad_0 = const()[name = tensor("temb_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor temb_17_dilations_0 = const()[name = tensor("temb_17_dilations_0"), val = tensor([1, 1])]; + tensor temb_17_groups_0 = const()[name = tensor("temb_17_groups_0"), val = tensor(1)]; + tensor mid_block_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor("mid_block_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(532874816)))]; + tensor mid_block_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("mid_block_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(536151680)))]; + tensor temb_17_cast_fp16 = conv(bias = mid_block_resnets_0_time_emb_proj_bias_to_fp16, dilations = temb_17_dilations_0, groups = temb_17_groups_0, pad = temb_17_pad_0, pad_type = temb_17_pad_type_0, strides = temb_17_strides_0, weight = mid_block_resnets_0_time_emb_proj_weight_to_fp16, x = input_15_cast_fp16_1)[name = tensor("temb_17_cast_fp16")]; + tensor input_207_cast_fp16 = add(x = hidden_states_117_cast_fp16, y = temb_17_cast_fp16)[name = tensor("input_207_cast_fp16")]; + tensor reshape_92_shape_0 = const()[name = tensor("reshape_92_shape_0"), val = tensor([2, 32, 40, 8, 8])]; + tensor reshape_92_cast_fp16 = reshape(shape = reshape_92_shape_0, x = input_207_cast_fp16)[name = tensor("reshape_92_cast_fp16")]; + tensor reduce_mean_69_axes_0 = const()[name = tensor("reduce_mean_69_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_69_keep_dims_0 = const()[name = tensor("reduce_mean_69_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_69_cast_fp16 = reduce_mean(axes = reduce_mean_69_axes_0, keep_dims = reduce_mean_69_keep_dims_0, x = reshape_92_cast_fp16)[name = tensor("reduce_mean_69_cast_fp16")]; + tensor sub_46_cast_fp16 = sub(x = reshape_92_cast_fp16, y = reduce_mean_69_cast_fp16)[name = tensor("sub_46_cast_fp16")]; + tensor square_23_cast_fp16 = square(x = sub_46_cast_fp16)[name = tensor("square_23_cast_fp16")]; + tensor reduce_mean_71_axes_0 = const()[name = tensor("reduce_mean_71_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_71_keep_dims_0 = const()[name = tensor("reduce_mean_71_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_71_cast_fp16 = reduce_mean(axes = reduce_mean_71_axes_0, keep_dims = reduce_mean_71_keep_dims_0, x = square_23_cast_fp16)[name = tensor("reduce_mean_71_cast_fp16")]; + tensor add_46_y_0_to_fp16 = const()[name = tensor("add_46_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_46_cast_fp16 = add(x = reduce_mean_71_cast_fp16, y = add_46_y_0_to_fp16)[name = tensor("add_46_cast_fp16")]; + tensor sqrt_23_cast_fp16 = sqrt(x = add_46_cast_fp16)[name = tensor("sqrt_23_cast_fp16")]; + tensor real_div_23_cast_fp16 = real_div(x = sub_46_cast_fp16, y = sqrt_23_cast_fp16)[name = tensor("real_div_23_cast_fp16")]; + tensor reshape_93_shape_0 = const()[name = tensor("reshape_93_shape_0"), val = tensor([2, 1280, 8, 8])]; + tensor reshape_93_cast_fp16 = reshape(shape = reshape_93_shape_0, x = real_div_23_cast_fp16)[name = tensor("reshape_93_cast_fp16")]; + tensor add_47_gamma_0_to_fp16 = const()[name = tensor("add_47_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(536154304)))]; + tensor add_47_beta_0_to_fp16 = const()[name = tensor("add_47_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(536156928)))]; + tensor add_47_epsilon_0_to_fp16 = const()[name = tensor("add_47_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_47_cast_fp16 = batch_norm(beta = add_47_beta_0_to_fp16, epsilon = add_47_epsilon_0_to_fp16, gamma = add_47_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_93_cast_fp16)[name = tensor("add_47_cast_fp16")]; + tensor input_211_cast_fp16 = silu(x = add_47_cast_fp16)[name = tensor("input_211_cast_fp16")]; + tensor hidden_states_119_pad_type_0 = const()[name = tensor("hidden_states_119_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_119_pad_0 = const()[name = tensor("hidden_states_119_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_119_strides_0 = const()[name = tensor("hidden_states_119_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_119_dilations_0 = const()[name = tensor("hidden_states_119_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_119_groups_0 = const()[name = tensor("hidden_states_119_groups_0"), val = tensor(1)]; + tensor mid_block_resnets_0_conv2_weight_to_fp16 = const()[name = tensor("mid_block_resnets_0_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(536159552)))]; + tensor mid_block_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("mid_block_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565650816)))]; + tensor hidden_states_119_cast_fp16 = conv(bias = mid_block_resnets_0_conv2_bias_to_fp16, dilations = hidden_states_119_dilations_0, groups = hidden_states_119_groups_0, pad = hidden_states_119_pad_0, pad_type = hidden_states_119_pad_type_0, strides = hidden_states_119_strides_0, weight = mid_block_resnets_0_conv2_weight_to_fp16, x = input_211_cast_fp16)[name = tensor("hidden_states_119_cast_fp16")]; + tensor hidden_states_121_cast_fp16 = add(x = input_199_cast_fp16, y = hidden_states_119_cast_fp16)[name = tensor("hidden_states_121_cast_fp16")]; + tensor reshape_96_shape_0 = const()[name = tensor("reshape_96_shape_0"), val = tensor([2, 32, 40, 8, 8])]; + tensor reshape_96_cast_fp16 = reshape(shape = reshape_96_shape_0, x = hidden_states_121_cast_fp16)[name = tensor("reshape_96_cast_fp16")]; + tensor reduce_mean_72_axes_0 = const()[name = tensor("reduce_mean_72_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_72_keep_dims_0 = const()[name = tensor("reduce_mean_72_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_72_cast_fp16 = reduce_mean(axes = reduce_mean_72_axes_0, keep_dims = reduce_mean_72_keep_dims_0, x = reshape_96_cast_fp16)[name = tensor("reduce_mean_72_cast_fp16")]; + tensor sub_48_cast_fp16 = sub(x = reshape_96_cast_fp16, y = reduce_mean_72_cast_fp16)[name = tensor("sub_48_cast_fp16")]; + tensor square_24_cast_fp16 = square(x = sub_48_cast_fp16)[name = tensor("square_24_cast_fp16")]; + tensor reduce_mean_74_axes_0 = const()[name = tensor("reduce_mean_74_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_74_keep_dims_0 = const()[name = tensor("reduce_mean_74_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_74_cast_fp16 = reduce_mean(axes = reduce_mean_74_axes_0, keep_dims = reduce_mean_74_keep_dims_0, x = square_24_cast_fp16)[name = tensor("reduce_mean_74_cast_fp16")]; + tensor add_48_y_0_to_fp16 = const()[name = tensor("add_48_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_48_cast_fp16 = add(x = reduce_mean_74_cast_fp16, y = add_48_y_0_to_fp16)[name = tensor("add_48_cast_fp16")]; + tensor sqrt_24_cast_fp16 = sqrt(x = add_48_cast_fp16)[name = tensor("sqrt_24_cast_fp16")]; + tensor real_div_24_cast_fp16 = real_div(x = sub_48_cast_fp16, y = sqrt_24_cast_fp16)[name = tensor("real_div_24_cast_fp16")]; + tensor reshape_97_shape_0 = const()[name = tensor("reshape_97_shape_0"), val = tensor([2, 1280, 8, 8])]; + tensor reshape_97_cast_fp16 = reshape(shape = reshape_97_shape_0, x = real_div_24_cast_fp16)[name = tensor("reshape_97_cast_fp16")]; + tensor add_49_gamma_0_to_fp16 = const()[name = tensor("add_49_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565653440)))]; + tensor add_49_beta_0_to_fp16 = const()[name = tensor("add_49_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565656064)))]; + tensor add_49_epsilon_0_to_fp16 = const()[name = tensor("add_49_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_49_cast_fp16 = batch_norm(beta = add_49_beta_0_to_fp16, epsilon = add_49_epsilon_0_to_fp16, gamma = add_49_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_97_cast_fp16)[name = tensor("add_49_cast_fp16")]; + tensor hidden_states_123_pad_type_0 = const()[name = tensor("hidden_states_123_pad_type_0"), val = tensor("valid")]; + tensor hidden_states_123_strides_0 = const()[name = tensor("hidden_states_123_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_123_pad_0 = const()[name = tensor("hidden_states_123_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_123_dilations_0 = const()[name = tensor("hidden_states_123_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_123_groups_0 = const()[name = tensor("hidden_states_123_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565658688)))]; + tensor mid_block_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568935552)))]; + tensor hidden_states_123_cast_fp16 = conv(bias = mid_block_attentions_0_proj_in_bias_to_fp16, dilations = hidden_states_123_dilations_0, groups = hidden_states_123_groups_0, pad = hidden_states_123_pad_0, pad_type = hidden_states_123_pad_type_0, strides = hidden_states_123_strides_0, weight = mid_block_attentions_0_proj_in_weight_to_fp16, x = add_49_cast_fp16)[name = tensor("hidden_states_123_cast_fp16")]; + tensor var_5829 = const()[name = tensor("op_5829"), val = tensor([2, 1280, 1, 64])]; + tensor inputs_37_cast_fp16 = reshape(shape = var_5829, x = hidden_states_123_cast_fp16)[name = tensor("inputs_37_cast_fp16")]; + tensor hidden_states_125_axes_0 = const()[name = tensor("hidden_states_125_axes_0"), val = tensor([1])]; + tensor hidden_states_125_gamma_0_to_fp16 = const()[name = tensor("hidden_states_125_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568938176)))]; + tensor hidden_states_125_beta_0_to_fp16 = const()[name = tensor("hidden_states_125_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568940800)))]; + tensor var_5845_to_fp16 = const()[name = tensor("op_5845_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_125_cast_fp16 = layer_norm(axes = hidden_states_125_axes_0, beta = hidden_states_125_beta_0_to_fp16, epsilon = var_5845_to_fp16, gamma = hidden_states_125_gamma_0_to_fp16, x = inputs_37_cast_fp16)[name = tensor("hidden_states_125_cast_fp16")]; + tensor q_25_pad_type_0 = const()[name = tensor("q_25_pad_type_0"), val = tensor("valid")]; + tensor q_25_strides_0 = const()[name = tensor("q_25_strides_0"), val = tensor([1, 1])]; + tensor q_25_pad_0 = const()[name = tensor("q_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_25_dilations_0 = const()[name = tensor("q_25_dilations_0"), val = tensor([1, 1])]; + tensor q_25_groups_0 = const()[name = tensor("q_25_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568943424)))]; + tensor q_25_cast_fp16 = conv(dilations = q_25_dilations_0, groups = q_25_groups_0, pad = q_25_pad_0, pad_type = q_25_pad_type_0, strides = q_25_strides_0, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_125_cast_fp16)[name = tensor("q_25_cast_fp16")]; + tensor k_49_pad_type_0 = const()[name = tensor("k_49_pad_type_0"), val = tensor("valid")]; + tensor k_49_strides_0 = const()[name = tensor("k_49_strides_0"), val = tensor([1, 1])]; + tensor k_49_pad_0 = const()[name = tensor("k_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_49_dilations_0 = const()[name = tensor("k_49_dilations_0"), val = tensor([1, 1])]; + tensor k_49_groups_0 = const()[name = tensor("k_49_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(572220288)))]; + tensor k_49_cast_fp16 = conv(dilations = k_49_dilations_0, groups = k_49_groups_0, pad = k_49_pad_0, pad_type = k_49_pad_type_0, strides = k_49_strides_0, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_125_cast_fp16)[name = tensor("k_49_cast_fp16")]; + tensor v_25_pad_type_0 = const()[name = tensor("v_25_pad_type_0"), val = tensor("valid")]; + tensor v_25_strides_0 = const()[name = tensor("v_25_strides_0"), val = tensor([1, 1])]; + tensor v_25_pad_0 = const()[name = tensor("v_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_25_dilations_0 = const()[name = tensor("v_25_dilations_0"), val = tensor([1, 1])]; + tensor v_25_groups_0 = const()[name = tensor("v_25_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(575497152)))]; + tensor v_25_cast_fp16 = conv(dilations = v_25_dilations_0, groups = v_25_groups_0, pad = v_25_pad_0, pad_type = v_25_pad_type_0, strides = v_25_strides_0, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_125_cast_fp16)[name = tensor("v_25_cast_fp16")]; + tensor var_5878_begin_0 = const()[name = tensor("op_5878_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5878_end_0 = const()[name = tensor("op_5878_end_0"), val = tensor([2, 160, 1, 64])]; + tensor var_5878_end_mask_0 = const()[name = tensor("op_5878_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5878_cast_fp16 = slice_by_index(begin = var_5878_begin_0, end = var_5878_end_0, end_mask = var_5878_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_5878_cast_fp16")]; + tensor var_5882_begin_0 = const()[name = tensor("op_5882_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_5882_end_0 = const()[name = tensor("op_5882_end_0"), val = tensor([2, 320, 1, 64])]; + tensor var_5882_end_mask_0 = const()[name = tensor("op_5882_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5882_cast_fp16 = slice_by_index(begin = var_5882_begin_0, end = var_5882_end_0, end_mask = var_5882_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_5882_cast_fp16")]; + tensor var_5886_begin_0 = const()[name = tensor("op_5886_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5886_end_0 = const()[name = tensor("op_5886_end_0"), val = tensor([2, 480, 1, 64])]; + tensor var_5886_end_mask_0 = const()[name = tensor("op_5886_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5886_cast_fp16 = slice_by_index(begin = var_5886_begin_0, end = var_5886_end_0, end_mask = var_5886_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_5886_cast_fp16")]; + tensor var_5890_begin_0 = const()[name = tensor("op_5890_begin_0"), val = tensor([0, 480, 0, 0])]; + tensor var_5890_end_0 = const()[name = tensor("op_5890_end_0"), val = tensor([2, 640, 1, 64])]; + tensor var_5890_end_mask_0 = const()[name = tensor("op_5890_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5890_cast_fp16 = slice_by_index(begin = var_5890_begin_0, end = var_5890_end_0, end_mask = var_5890_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_5890_cast_fp16")]; + tensor var_5894_begin_0 = const()[name = tensor("op_5894_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5894_end_0 = const()[name = tensor("op_5894_end_0"), val = tensor([2, 800, 1, 64])]; + tensor var_5894_end_mask_0 = const()[name = tensor("op_5894_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5894_cast_fp16 = slice_by_index(begin = var_5894_begin_0, end = var_5894_end_0, end_mask = var_5894_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_5894_cast_fp16")]; + tensor var_5898_begin_0 = const()[name = tensor("op_5898_begin_0"), val = tensor([0, 800, 0, 0])]; + tensor var_5898_end_0 = const()[name = tensor("op_5898_end_0"), val = tensor([2, 960, 1, 64])]; + tensor var_5898_end_mask_0 = const()[name = tensor("op_5898_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5898_cast_fp16 = slice_by_index(begin = var_5898_begin_0, end = var_5898_end_0, end_mask = var_5898_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_5898_cast_fp16")]; + tensor var_5902_begin_0 = const()[name = tensor("op_5902_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_5902_end_0 = const()[name = tensor("op_5902_end_0"), val = tensor([2, 1120, 1, 64])]; + tensor var_5902_end_mask_0 = const()[name = tensor("op_5902_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5902_cast_fp16 = slice_by_index(begin = var_5902_begin_0, end = var_5902_end_0, end_mask = var_5902_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_5902_cast_fp16")]; + tensor var_5906_begin_0 = const()[name = tensor("op_5906_begin_0"), val = tensor([0, 1120, 0, 0])]; + tensor var_5906_end_0 = const()[name = tensor("op_5906_end_0"), val = tensor([2, 1, 1, 64])]; + tensor var_5906_end_mask_0 = const()[name = tensor("op_5906_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_5906_cast_fp16 = slice_by_index(begin = var_5906_begin_0, end = var_5906_end_0, end_mask = var_5906_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_5906_cast_fp16")]; + tensor k_51_perm_0 = const()[name = tensor("k_51_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_5913_begin_0 = const()[name = tensor("op_5913_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5913_end_0 = const()[name = tensor("op_5913_end_0"), val = tensor([2, 64, 1, 160])]; + tensor var_5913_end_mask_0 = const()[name = tensor("op_5913_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_51_cast_fp16 = transpose(perm = k_51_perm_0, x = k_49_cast_fp16)[name = tensor("transpose_1")]; + tensor var_5913_cast_fp16 = slice_by_index(begin = var_5913_begin_0, end = var_5913_end_0, end_mask = var_5913_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_5913_cast_fp16")]; + tensor var_5917_begin_0 = const()[name = tensor("op_5917_begin_0"), val = tensor([0, 0, 0, 160])]; + tensor var_5917_end_0 = const()[name = tensor("op_5917_end_0"), val = tensor([2, 64, 1, 320])]; + tensor var_5917_end_mask_0 = const()[name = tensor("op_5917_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5917_cast_fp16 = slice_by_index(begin = var_5917_begin_0, end = var_5917_end_0, end_mask = var_5917_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_5917_cast_fp16")]; + tensor var_5921_begin_0 = const()[name = tensor("op_5921_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_5921_end_0 = const()[name = tensor("op_5921_end_0"), val = tensor([2, 64, 1, 480])]; + tensor var_5921_end_mask_0 = const()[name = tensor("op_5921_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5921_cast_fp16 = slice_by_index(begin = var_5921_begin_0, end = var_5921_end_0, end_mask = var_5921_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_5921_cast_fp16")]; + tensor var_5925_begin_0 = const()[name = tensor("op_5925_begin_0"), val = tensor([0, 0, 0, 480])]; + tensor var_5925_end_0 = const()[name = tensor("op_5925_end_0"), val = tensor([2, 64, 1, 640])]; + tensor var_5925_end_mask_0 = const()[name = tensor("op_5925_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5925_cast_fp16 = slice_by_index(begin = var_5925_begin_0, end = var_5925_end_0, end_mask = var_5925_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_5925_cast_fp16")]; + tensor var_5929_begin_0 = const()[name = tensor("op_5929_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_5929_end_0 = const()[name = tensor("op_5929_end_0"), val = tensor([2, 64, 1, 800])]; + tensor var_5929_end_mask_0 = const()[name = tensor("op_5929_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5929_cast_fp16 = slice_by_index(begin = var_5929_begin_0, end = var_5929_end_0, end_mask = var_5929_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_5929_cast_fp16")]; + tensor var_5933_begin_0 = const()[name = tensor("op_5933_begin_0"), val = tensor([0, 0, 0, 800])]; + tensor var_5933_end_0 = const()[name = tensor("op_5933_end_0"), val = tensor([2, 64, 1, 960])]; + tensor var_5933_end_mask_0 = const()[name = tensor("op_5933_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5933_cast_fp16 = slice_by_index(begin = var_5933_begin_0, end = var_5933_end_0, end_mask = var_5933_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_5933_cast_fp16")]; + tensor var_5937_begin_0 = const()[name = tensor("op_5937_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_5937_end_0 = const()[name = tensor("op_5937_end_0"), val = tensor([2, 64, 1, 1120])]; + tensor var_5937_end_mask_0 = const()[name = tensor("op_5937_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5937_cast_fp16 = slice_by_index(begin = var_5937_begin_0, end = var_5937_end_0, end_mask = var_5937_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_5937_cast_fp16")]; + tensor var_5941_begin_0 = const()[name = tensor("op_5941_begin_0"), val = tensor([0, 0, 0, 1120])]; + tensor var_5941_end_0 = const()[name = tensor("op_5941_end_0"), val = tensor([2, 64, 1, 1])]; + tensor var_5941_end_mask_0 = const()[name = tensor("op_5941_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_5941_cast_fp16 = slice_by_index(begin = var_5941_begin_0, end = var_5941_end_0, end_mask = var_5941_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_5941_cast_fp16")]; + tensor var_5943_begin_0 = const()[name = tensor("op_5943_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5943_end_0 = const()[name = tensor("op_5943_end_0"), val = tensor([2, 160, 1, 64])]; + tensor var_5943_end_mask_0 = const()[name = tensor("op_5943_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5943_cast_fp16 = slice_by_index(begin = var_5943_begin_0, end = var_5943_end_0, end_mask = var_5943_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_5943_cast_fp16")]; + tensor var_5947_begin_0 = const()[name = tensor("op_5947_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_5947_end_0 = const()[name = tensor("op_5947_end_0"), val = tensor([2, 320, 1, 64])]; + tensor var_5947_end_mask_0 = const()[name = tensor("op_5947_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5947_cast_fp16 = slice_by_index(begin = var_5947_begin_0, end = var_5947_end_0, end_mask = var_5947_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_5947_cast_fp16")]; + tensor var_5951_begin_0 = const()[name = tensor("op_5951_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5951_end_0 = const()[name = tensor("op_5951_end_0"), val = tensor([2, 480, 1, 64])]; + tensor var_5951_end_mask_0 = const()[name = tensor("op_5951_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5951_cast_fp16 = slice_by_index(begin = var_5951_begin_0, end = var_5951_end_0, end_mask = var_5951_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_5951_cast_fp16")]; + tensor var_5955_begin_0 = const()[name = tensor("op_5955_begin_0"), val = tensor([0, 480, 0, 0])]; + tensor var_5955_end_0 = const()[name = tensor("op_5955_end_0"), val = tensor([2, 640, 1, 64])]; + tensor var_5955_end_mask_0 = const()[name = tensor("op_5955_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5955_cast_fp16 = slice_by_index(begin = var_5955_begin_0, end = var_5955_end_0, end_mask = var_5955_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_5955_cast_fp16")]; + tensor var_5959_begin_0 = const()[name = tensor("op_5959_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5959_end_0 = const()[name = tensor("op_5959_end_0"), val = tensor([2, 800, 1, 64])]; + tensor var_5959_end_mask_0 = const()[name = tensor("op_5959_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5959_cast_fp16 = slice_by_index(begin = var_5959_begin_0, end = var_5959_end_0, end_mask = var_5959_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_5959_cast_fp16")]; + tensor var_5963_begin_0 = const()[name = tensor("op_5963_begin_0"), val = tensor([0, 800, 0, 0])]; + tensor var_5963_end_0 = const()[name = tensor("op_5963_end_0"), val = tensor([2, 960, 1, 64])]; + tensor var_5963_end_mask_0 = const()[name = tensor("op_5963_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5963_cast_fp16 = slice_by_index(begin = var_5963_begin_0, end = var_5963_end_0, end_mask = var_5963_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_5963_cast_fp16")]; + tensor var_5967_begin_0 = const()[name = tensor("op_5967_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_5967_end_0 = const()[name = tensor("op_5967_end_0"), val = tensor([2, 1120, 1, 64])]; + tensor var_5967_end_mask_0 = const()[name = tensor("op_5967_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5967_cast_fp16 = slice_by_index(begin = var_5967_begin_0, end = var_5967_end_0, end_mask = var_5967_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_5967_cast_fp16")]; + tensor var_5971_begin_0 = const()[name = tensor("op_5971_begin_0"), val = tensor([0, 1120, 0, 0])]; + tensor var_5971_end_0 = const()[name = tensor("op_5971_end_0"), val = tensor([2, 1, 1, 64])]; + tensor var_5971_end_mask_0 = const()[name = tensor("op_5971_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_5971_cast_fp16 = slice_by_index(begin = var_5971_begin_0, end = var_5971_end_0, end_mask = var_5971_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_5971_cast_fp16")]; + tensor var_5975_equation_0 = const()[name = tensor("op_5975_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5975_cast_fp16 = einsum(equation = var_5975_equation_0, values = (var_5913_cast_fp16, var_5878_cast_fp16))[name = tensor("op_5975_cast_fp16")]; + tensor var_5976_to_fp16 = const()[name = tensor("op_5976_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_65_cast_fp16 = mul(x = var_5975_cast_fp16, y = var_5976_to_fp16)[name = tensor("aw_65_cast_fp16")]; + tensor var_5979_equation_0 = const()[name = tensor("op_5979_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5979_cast_fp16 = einsum(equation = var_5979_equation_0, values = (var_5917_cast_fp16, var_5882_cast_fp16))[name = tensor("op_5979_cast_fp16")]; + tensor var_5980_to_fp16 = const()[name = tensor("op_5980_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_67_cast_fp16 = mul(x = var_5979_cast_fp16, y = var_5980_to_fp16)[name = tensor("aw_67_cast_fp16")]; + tensor var_5983_equation_0 = const()[name = tensor("op_5983_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5983_cast_fp16 = einsum(equation = var_5983_equation_0, values = (var_5921_cast_fp16, var_5886_cast_fp16))[name = tensor("op_5983_cast_fp16")]; + tensor var_5984_to_fp16 = const()[name = tensor("op_5984_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_69_cast_fp16 = mul(x = var_5983_cast_fp16, y = var_5984_to_fp16)[name = tensor("aw_69_cast_fp16")]; + tensor var_5987_equation_0 = const()[name = tensor("op_5987_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5987_cast_fp16 = einsum(equation = var_5987_equation_0, values = (var_5925_cast_fp16, var_5890_cast_fp16))[name = tensor("op_5987_cast_fp16")]; + tensor var_5988_to_fp16 = const()[name = tensor("op_5988_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_71_cast_fp16 = mul(x = var_5987_cast_fp16, y = var_5988_to_fp16)[name = tensor("aw_71_cast_fp16")]; + tensor var_5991_equation_0 = const()[name = tensor("op_5991_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5991_cast_fp16 = einsum(equation = var_5991_equation_0, values = (var_5929_cast_fp16, var_5894_cast_fp16))[name = tensor("op_5991_cast_fp16")]; + tensor var_5992_to_fp16 = const()[name = tensor("op_5992_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_73_cast_fp16 = mul(x = var_5991_cast_fp16, y = var_5992_to_fp16)[name = tensor("aw_73_cast_fp16")]; + tensor var_5995_equation_0 = const()[name = tensor("op_5995_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5995_cast_fp16 = einsum(equation = var_5995_equation_0, values = (var_5933_cast_fp16, var_5898_cast_fp16))[name = tensor("op_5995_cast_fp16")]; + tensor var_5996_to_fp16 = const()[name = tensor("op_5996_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_75_cast_fp16 = mul(x = var_5995_cast_fp16, y = var_5996_to_fp16)[name = tensor("aw_75_cast_fp16")]; + tensor var_5999_equation_0 = const()[name = tensor("op_5999_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5999_cast_fp16 = einsum(equation = var_5999_equation_0, values = (var_5937_cast_fp16, var_5902_cast_fp16))[name = tensor("op_5999_cast_fp16")]; + tensor var_6000_to_fp16 = const()[name = tensor("op_6000_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_77_cast_fp16 = mul(x = var_5999_cast_fp16, y = var_6000_to_fp16)[name = tensor("aw_77_cast_fp16")]; + tensor var_6003_equation_0 = const()[name = tensor("op_6003_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6003_cast_fp16 = einsum(equation = var_6003_equation_0, values = (var_5941_cast_fp16, var_5906_cast_fp16))[name = tensor("op_6003_cast_fp16")]; + tensor var_6004_to_fp16 = const()[name = tensor("op_6004_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_79_cast_fp16 = mul(x = var_6003_cast_fp16, y = var_6004_to_fp16)[name = tensor("aw_79_cast_fp16")]; + tensor var_6006_cast_fp16 = softmax(axis = var_5762, x = aw_65_cast_fp16)[name = tensor("op_6006_cast_fp16")]; + tensor var_6007_cast_fp16 = softmax(axis = var_5762, x = aw_67_cast_fp16)[name = tensor("op_6007_cast_fp16")]; + tensor var_6008_cast_fp16 = softmax(axis = var_5762, x = aw_69_cast_fp16)[name = tensor("op_6008_cast_fp16")]; + tensor var_6009_cast_fp16 = softmax(axis = var_5762, x = aw_71_cast_fp16)[name = tensor("op_6009_cast_fp16")]; + tensor var_6010_cast_fp16 = softmax(axis = var_5762, x = aw_73_cast_fp16)[name = tensor("op_6010_cast_fp16")]; + tensor var_6011_cast_fp16 = softmax(axis = var_5762, x = aw_75_cast_fp16)[name = tensor("op_6011_cast_fp16")]; + tensor var_6012_cast_fp16 = softmax(axis = var_5762, x = aw_77_cast_fp16)[name = tensor("op_6012_cast_fp16")]; + tensor var_6013_cast_fp16 = softmax(axis = var_5762, x = aw_79_cast_fp16)[name = tensor("op_6013_cast_fp16")]; + tensor var_6015_equation_0 = const()[name = tensor("op_6015_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6015_cast_fp16 = einsum(equation = var_6015_equation_0, values = (var_5943_cast_fp16, var_6006_cast_fp16))[name = tensor("op_6015_cast_fp16")]; + tensor var_6017_equation_0 = const()[name = tensor("op_6017_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6017_cast_fp16 = einsum(equation = var_6017_equation_0, values = (var_5947_cast_fp16, var_6007_cast_fp16))[name = tensor("op_6017_cast_fp16")]; + tensor var_6019_equation_0 = const()[name = tensor("op_6019_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6019_cast_fp16 = einsum(equation = var_6019_equation_0, values = (var_5951_cast_fp16, var_6008_cast_fp16))[name = tensor("op_6019_cast_fp16")]; + tensor var_6021_equation_0 = const()[name = tensor("op_6021_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6021_cast_fp16 = einsum(equation = var_6021_equation_0, values = (var_5955_cast_fp16, var_6009_cast_fp16))[name = tensor("op_6021_cast_fp16")]; + tensor var_6023_equation_0 = const()[name = tensor("op_6023_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6023_cast_fp16 = einsum(equation = var_6023_equation_0, values = (var_5959_cast_fp16, var_6010_cast_fp16))[name = tensor("op_6023_cast_fp16")]; + tensor var_6025_equation_0 = const()[name = tensor("op_6025_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6025_cast_fp16 = einsum(equation = var_6025_equation_0, values = (var_5963_cast_fp16, var_6011_cast_fp16))[name = tensor("op_6025_cast_fp16")]; + tensor var_6027_equation_0 = const()[name = tensor("op_6027_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6027_cast_fp16 = einsum(equation = var_6027_equation_0, values = (var_5967_cast_fp16, var_6012_cast_fp16))[name = tensor("op_6027_cast_fp16")]; + tensor var_6029_equation_0 = const()[name = tensor("op_6029_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6029_cast_fp16 = einsum(equation = var_6029_equation_0, values = (var_5971_cast_fp16, var_6013_cast_fp16))[name = tensor("op_6029_cast_fp16")]; + tensor input_215_interleave_0 = const()[name = tensor("input_215_interleave_0"), val = tensor(false)]; + tensor input_215_cast_fp16 = concat(axis = var_5762, interleave = input_215_interleave_0, values = (var_6015_cast_fp16, var_6017_cast_fp16, var_6019_cast_fp16, var_6021_cast_fp16, var_6023_cast_fp16, var_6025_cast_fp16, var_6027_cast_fp16, var_6029_cast_fp16))[name = tensor("input_215_cast_fp16")]; + tensor var_6039_pad_type_0 = const()[name = tensor("op_6039_pad_type_0"), val = tensor("valid")]; + tensor var_6039_strides_0 = const()[name = tensor("op_6039_strides_0"), val = tensor([1, 1])]; + tensor var_6039_pad_0 = const()[name = tensor("op_6039_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6039_dilations_0 = const()[name = tensor("op_6039_dilations_0"), val = tensor([1, 1])]; + tensor var_6039_groups_0 = const()[name = tensor("op_6039_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578774016)))]; + tensor mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(582050880)))]; + tensor var_6039_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_6039_dilations_0, groups = var_6039_groups_0, pad = var_6039_pad_0, pad_type = var_6039_pad_type_0, strides = var_6039_strides_0, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_215_cast_fp16)[name = tensor("op_6039_cast_fp16")]; + tensor inputs_39_cast_fp16 = add(x = var_6039_cast_fp16, y = inputs_37_cast_fp16)[name = tensor("inputs_39_cast_fp16")]; + tensor hidden_states_127_axes_0 = const()[name = tensor("hidden_states_127_axes_0"), val = tensor([1])]; + tensor hidden_states_127_gamma_0_to_fp16 = const()[name = tensor("hidden_states_127_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(582053504)))]; + tensor hidden_states_127_beta_0_to_fp16 = const()[name = tensor("hidden_states_127_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(582056128)))]; + tensor var_6049_to_fp16 = const()[name = tensor("op_6049_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_127_cast_fp16 = layer_norm(axes = hidden_states_127_axes_0, beta = hidden_states_127_beta_0_to_fp16, epsilon = var_6049_to_fp16, gamma = hidden_states_127_gamma_0_to_fp16, x = inputs_39_cast_fp16)[name = tensor("hidden_states_127_cast_fp16")]; + tensor q_27_pad_type_0 = const()[name = tensor("q_27_pad_type_0"), val = tensor("valid")]; + tensor q_27_strides_0 = const()[name = tensor("q_27_strides_0"), val = tensor([1, 1])]; + tensor q_27_pad_0 = const()[name = tensor("q_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_27_dilations_0 = const()[name = tensor("q_27_dilations_0"), val = tensor([1, 1])]; + tensor q_27_groups_0 = const()[name = tensor("q_27_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(582058752)))]; + tensor q_27_cast_fp16 = conv(dilations = q_27_dilations_0, groups = q_27_groups_0, pad = q_27_pad_0, pad_type = q_27_pad_type_0, strides = q_27_strides_0, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_127_cast_fp16)[name = tensor("q_27_cast_fp16")]; + tensor k_53_pad_type_0 = const()[name = tensor("k_53_pad_type_0"), val = tensor("valid")]; + tensor k_53_strides_0 = const()[name = tensor("k_53_strides_0"), val = tensor([1, 1])]; + tensor k_53_pad_0 = const()[name = tensor("k_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_53_dilations_0 = const()[name = tensor("k_53_dilations_0"), val = tensor([1, 1])]; + tensor k_53_groups_0 = const()[name = tensor("k_53_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(585335616)))]; + tensor k_53_cast_fp16 = conv(dilations = k_53_dilations_0, groups = k_53_groups_0, pad = k_53_pad_0, pad_type = k_53_pad_type_0, strides = k_53_strides_0, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_53_cast_fp16")]; + tensor v_27_pad_type_0 = const()[name = tensor("v_27_pad_type_0"), val = tensor("valid")]; + tensor v_27_strides_0 = const()[name = tensor("v_27_strides_0"), val = tensor([1, 1])]; + tensor v_27_pad_0 = const()[name = tensor("v_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_27_dilations_0 = const()[name = tensor("v_27_dilations_0"), val = tensor([1, 1])]; + tensor v_27_groups_0 = const()[name = tensor("v_27_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(587301760)))]; + tensor v_27_cast_fp16 = conv(dilations = v_27_dilations_0, groups = v_27_groups_0, pad = v_27_pad_0, pad_type = v_27_pad_type_0, strides = v_27_strides_0, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_27_cast_fp16")]; + tensor var_6082_begin_0 = const()[name = tensor("op_6082_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6082_end_0 = const()[name = tensor("op_6082_end_0"), val = tensor([2, 160, 1, 64])]; + tensor var_6082_end_mask_0 = const()[name = tensor("op_6082_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6082_cast_fp16 = slice_by_index(begin = var_6082_begin_0, end = var_6082_end_0, end_mask = var_6082_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_6082_cast_fp16")]; + tensor var_6086_begin_0 = const()[name = tensor("op_6086_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_6086_end_0 = const()[name = tensor("op_6086_end_0"), val = tensor([2, 320, 1, 64])]; + tensor var_6086_end_mask_0 = const()[name = tensor("op_6086_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6086_cast_fp16 = slice_by_index(begin = var_6086_begin_0, end = var_6086_end_0, end_mask = var_6086_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_6086_cast_fp16")]; + tensor var_6090_begin_0 = const()[name = tensor("op_6090_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_6090_end_0 = const()[name = tensor("op_6090_end_0"), val = tensor([2, 480, 1, 64])]; + tensor var_6090_end_mask_0 = const()[name = tensor("op_6090_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6090_cast_fp16 = slice_by_index(begin = var_6090_begin_0, end = var_6090_end_0, end_mask = var_6090_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_6090_cast_fp16")]; + tensor var_6094_begin_0 = const()[name = tensor("op_6094_begin_0"), val = tensor([0, 480, 0, 0])]; + tensor var_6094_end_0 = const()[name = tensor("op_6094_end_0"), val = tensor([2, 640, 1, 64])]; + tensor var_6094_end_mask_0 = const()[name = tensor("op_6094_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6094_cast_fp16 = slice_by_index(begin = var_6094_begin_0, end = var_6094_end_0, end_mask = var_6094_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_6094_cast_fp16")]; + tensor var_6098_begin_0 = const()[name = tensor("op_6098_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_6098_end_0 = const()[name = tensor("op_6098_end_0"), val = tensor([2, 800, 1, 64])]; + tensor var_6098_end_mask_0 = const()[name = tensor("op_6098_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6098_cast_fp16 = slice_by_index(begin = var_6098_begin_0, end = var_6098_end_0, end_mask = var_6098_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_6098_cast_fp16")]; + tensor var_6102_begin_0 = const()[name = tensor("op_6102_begin_0"), val = tensor([0, 800, 0, 0])]; + tensor var_6102_end_0 = const()[name = tensor("op_6102_end_0"), val = tensor([2, 960, 1, 64])]; + tensor var_6102_end_mask_0 = const()[name = tensor("op_6102_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6102_cast_fp16 = slice_by_index(begin = var_6102_begin_0, end = var_6102_end_0, end_mask = var_6102_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_6102_cast_fp16")]; + tensor var_6106_begin_0 = const()[name = tensor("op_6106_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_6106_end_0 = const()[name = tensor("op_6106_end_0"), val = tensor([2, 1120, 1, 64])]; + tensor var_6106_end_mask_0 = const()[name = tensor("op_6106_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6106_cast_fp16 = slice_by_index(begin = var_6106_begin_0, end = var_6106_end_0, end_mask = var_6106_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_6106_cast_fp16")]; + tensor var_6110_begin_0 = const()[name = tensor("op_6110_begin_0"), val = tensor([0, 1120, 0, 0])]; + tensor var_6110_end_0 = const()[name = tensor("op_6110_end_0"), val = tensor([2, 1, 1, 64])]; + tensor var_6110_end_mask_0 = const()[name = tensor("op_6110_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_6110_cast_fp16 = slice_by_index(begin = var_6110_begin_0, end = var_6110_end_0, end_mask = var_6110_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_6110_cast_fp16")]; + tensor k_55_perm_0 = const()[name = tensor("k_55_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_6117_begin_0 = const()[name = tensor("op_6117_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6117_end_0 = const()[name = tensor("op_6117_end_0"), val = tensor([2, 77, 1, 160])]; + tensor var_6117_end_mask_0 = const()[name = tensor("op_6117_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_55_cast_fp16 = transpose(perm = k_55_perm_0, x = k_53_cast_fp16)[name = tensor("transpose_0")]; + tensor var_6117_cast_fp16 = slice_by_index(begin = var_6117_begin_0, end = var_6117_end_0, end_mask = var_6117_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_6117_cast_fp16")]; + tensor var_6121_begin_0 = const()[name = tensor("op_6121_begin_0"), val = tensor([0, 0, 0, 160])]; + tensor var_6121_end_0 = const()[name = tensor("op_6121_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_6121_end_mask_0 = const()[name = tensor("op_6121_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6121_cast_fp16 = slice_by_index(begin = var_6121_begin_0, end = var_6121_end_0, end_mask = var_6121_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_6121_cast_fp16")]; + tensor var_6125_begin_0 = const()[name = tensor("op_6125_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_6125_end_0 = const()[name = tensor("op_6125_end_0"), val = tensor([2, 77, 1, 480])]; + tensor var_6125_end_mask_0 = const()[name = tensor("op_6125_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6125_cast_fp16 = slice_by_index(begin = var_6125_begin_0, end = var_6125_end_0, end_mask = var_6125_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_6125_cast_fp16")]; + tensor var_6129_begin_0 = const()[name = tensor("op_6129_begin_0"), val = tensor([0, 0, 0, 480])]; + tensor var_6129_end_0 = const()[name = tensor("op_6129_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_6129_end_mask_0 = const()[name = tensor("op_6129_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6129_cast_fp16 = slice_by_index(begin = var_6129_begin_0, end = var_6129_end_0, end_mask = var_6129_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_6129_cast_fp16")]; + tensor var_6133_begin_0 = const()[name = tensor("op_6133_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_6133_end_0 = const()[name = tensor("op_6133_end_0"), val = tensor([2, 77, 1, 800])]; + tensor var_6133_end_mask_0 = const()[name = tensor("op_6133_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6133_cast_fp16 = slice_by_index(begin = var_6133_begin_0, end = var_6133_end_0, end_mask = var_6133_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_6133_cast_fp16")]; + tensor var_6137_begin_0 = const()[name = tensor("op_6137_begin_0"), val = tensor([0, 0, 0, 800])]; + tensor var_6137_end_0 = const()[name = tensor("op_6137_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_6137_end_mask_0 = const()[name = tensor("op_6137_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6137_cast_fp16 = slice_by_index(begin = var_6137_begin_0, end = var_6137_end_0, end_mask = var_6137_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_6137_cast_fp16")]; + tensor var_6141_begin_0 = const()[name = tensor("op_6141_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_6141_end_0 = const()[name = tensor("op_6141_end_0"), val = tensor([2, 77, 1, 1120])]; + tensor var_6141_end_mask_0 = const()[name = tensor("op_6141_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6141_cast_fp16 = slice_by_index(begin = var_6141_begin_0, end = var_6141_end_0, end_mask = var_6141_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_6141_cast_fp16")]; + tensor var_6145_begin_0 = const()[name = tensor("op_6145_begin_0"), val = tensor([0, 0, 0, 1120])]; + tensor var_6145_end_0 = const()[name = tensor("op_6145_end_0"), val = tensor([2, 77, 1, 1])]; + tensor var_6145_end_mask_0 = const()[name = tensor("op_6145_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_6145_cast_fp16 = slice_by_index(begin = var_6145_begin_0, end = var_6145_end_0, end_mask = var_6145_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_6145_cast_fp16")]; + tensor var_6147_begin_0 = const()[name = tensor("op_6147_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6147_end_0 = const()[name = tensor("op_6147_end_0"), val = tensor([2, 160, 1, 77])]; + tensor var_6147_end_mask_0 = const()[name = tensor("op_6147_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6147_cast_fp16 = slice_by_index(begin = var_6147_begin_0, end = var_6147_end_0, end_mask = var_6147_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_6147_cast_fp16")]; + tensor var_6151_begin_0 = const()[name = tensor("op_6151_begin_0"), val = tensor([0, 160, 0, 0])]; + tensor var_6151_end_0 = const()[name = tensor("op_6151_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_6151_end_mask_0 = const()[name = tensor("op_6151_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6151_cast_fp16 = slice_by_index(begin = var_6151_begin_0, end = var_6151_end_0, end_mask = var_6151_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_6151_cast_fp16")]; + tensor var_6155_begin_0 = const()[name = tensor("op_6155_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_6155_end_0 = const()[name = tensor("op_6155_end_0"), val = tensor([2, 480, 1, 77])]; + tensor var_6155_end_mask_0 = const()[name = tensor("op_6155_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6155_cast_fp16 = slice_by_index(begin = var_6155_begin_0, end = var_6155_end_0, end_mask = var_6155_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_6155_cast_fp16")]; + tensor var_6159_begin_0 = const()[name = tensor("op_6159_begin_0"), val = tensor([0, 480, 0, 0])]; + tensor var_6159_end_0 = const()[name = tensor("op_6159_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_6159_end_mask_0 = const()[name = tensor("op_6159_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6159_cast_fp16 = slice_by_index(begin = var_6159_begin_0, end = var_6159_end_0, end_mask = var_6159_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_6159_cast_fp16")]; + tensor var_6163_begin_0 = const()[name = tensor("op_6163_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_6163_end_0 = const()[name = tensor("op_6163_end_0"), val = tensor([2, 800, 1, 77])]; + tensor var_6163_end_mask_0 = const()[name = tensor("op_6163_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6163_cast_fp16 = slice_by_index(begin = var_6163_begin_0, end = var_6163_end_0, end_mask = var_6163_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_6163_cast_fp16")]; + tensor var_6167_begin_0 = const()[name = tensor("op_6167_begin_0"), val = tensor([0, 800, 0, 0])]; + tensor var_6167_end_0 = const()[name = tensor("op_6167_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_6167_end_mask_0 = const()[name = tensor("op_6167_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6167_cast_fp16 = slice_by_index(begin = var_6167_begin_0, end = var_6167_end_0, end_mask = var_6167_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_6167_cast_fp16")]; + tensor var_6171_begin_0 = const()[name = tensor("op_6171_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_6171_end_0 = const()[name = tensor("op_6171_end_0"), val = tensor([2, 1120, 1, 77])]; + tensor var_6171_end_mask_0 = const()[name = tensor("op_6171_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6171_cast_fp16 = slice_by_index(begin = var_6171_begin_0, end = var_6171_end_0, end_mask = var_6171_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_6171_cast_fp16")]; + tensor var_6175_begin_0 = const()[name = tensor("op_6175_begin_0"), val = tensor([0, 1120, 0, 0])]; + tensor var_6175_end_0 = const()[name = tensor("op_6175_end_0"), val = tensor([2, 1, 1, 77])]; + tensor var_6175_end_mask_0 = const()[name = tensor("op_6175_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_6175_cast_fp16 = slice_by_index(begin = var_6175_begin_0, end = var_6175_end_0, end_mask = var_6175_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_6175_cast_fp16")]; + tensor var_6179_equation_0 = const()[name = tensor("op_6179_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6179_cast_fp16 = einsum(equation = var_6179_equation_0, values = (var_6117_cast_fp16, var_6082_cast_fp16))[name = tensor("op_6179_cast_fp16")]; + tensor var_6180_to_fp16 = const()[name = tensor("op_6180_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_81_cast_fp16 = mul(x = var_6179_cast_fp16, y = var_6180_to_fp16)[name = tensor("aw_81_cast_fp16")]; + tensor var_6183_equation_0 = const()[name = tensor("op_6183_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6183_cast_fp16 = einsum(equation = var_6183_equation_0, values = (var_6121_cast_fp16, var_6086_cast_fp16))[name = tensor("op_6183_cast_fp16")]; + tensor var_6184_to_fp16 = const()[name = tensor("op_6184_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_83_cast_fp16 = mul(x = var_6183_cast_fp16, y = var_6184_to_fp16)[name = tensor("aw_83_cast_fp16")]; + tensor var_6187_equation_0 = const()[name = tensor("op_6187_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6187_cast_fp16 = einsum(equation = var_6187_equation_0, values = (var_6125_cast_fp16, var_6090_cast_fp16))[name = tensor("op_6187_cast_fp16")]; + tensor var_6188_to_fp16 = const()[name = tensor("op_6188_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_85_cast_fp16 = mul(x = var_6187_cast_fp16, y = var_6188_to_fp16)[name = tensor("aw_85_cast_fp16")]; + tensor var_6191_equation_0 = const()[name = tensor("op_6191_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6191_cast_fp16 = einsum(equation = var_6191_equation_0, values = (var_6129_cast_fp16, var_6094_cast_fp16))[name = tensor("op_6191_cast_fp16")]; + tensor var_6192_to_fp16 = const()[name = tensor("op_6192_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_87_cast_fp16 = mul(x = var_6191_cast_fp16, y = var_6192_to_fp16)[name = tensor("aw_87_cast_fp16")]; + tensor var_6195_equation_0 = const()[name = tensor("op_6195_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6195_cast_fp16 = einsum(equation = var_6195_equation_0, values = (var_6133_cast_fp16, var_6098_cast_fp16))[name = tensor("op_6195_cast_fp16")]; + tensor var_6196_to_fp16 = const()[name = tensor("op_6196_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_89_cast_fp16 = mul(x = var_6195_cast_fp16, y = var_6196_to_fp16)[name = tensor("aw_89_cast_fp16")]; + tensor var_6199_equation_0 = const()[name = tensor("op_6199_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6199_cast_fp16 = einsum(equation = var_6199_equation_0, values = (var_6137_cast_fp16, var_6102_cast_fp16))[name = tensor("op_6199_cast_fp16")]; + tensor var_6200_to_fp16 = const()[name = tensor("op_6200_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_91_cast_fp16 = mul(x = var_6199_cast_fp16, y = var_6200_to_fp16)[name = tensor("aw_91_cast_fp16")]; + tensor var_6203_equation_0 = const()[name = tensor("op_6203_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6203_cast_fp16 = einsum(equation = var_6203_equation_0, values = (var_6141_cast_fp16, var_6106_cast_fp16))[name = tensor("op_6203_cast_fp16")]; + tensor var_6204_to_fp16 = const()[name = tensor("op_6204_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_93_cast_fp16 = mul(x = var_6203_cast_fp16, y = var_6204_to_fp16)[name = tensor("aw_93_cast_fp16")]; + tensor var_6207_equation_0 = const()[name = tensor("op_6207_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6207_cast_fp16 = einsum(equation = var_6207_equation_0, values = (var_6145_cast_fp16, var_6110_cast_fp16))[name = tensor("op_6207_cast_fp16")]; + tensor var_6208_to_fp16 = const()[name = tensor("op_6208_to_fp16"), val = tensor(0x1.43cp-4)]; + tensor aw_95_cast_fp16 = mul(x = var_6207_cast_fp16, y = var_6208_to_fp16)[name = tensor("aw_95_cast_fp16")]; + tensor var_6210_cast_fp16 = softmax(axis = var_5762, x = aw_81_cast_fp16)[name = tensor("op_6210_cast_fp16")]; + tensor var_6211_cast_fp16 = softmax(axis = var_5762, x = aw_83_cast_fp16)[name = tensor("op_6211_cast_fp16")]; + tensor var_6212_cast_fp16 = softmax(axis = var_5762, x = aw_85_cast_fp16)[name = tensor("op_6212_cast_fp16")]; + tensor var_6213_cast_fp16 = softmax(axis = var_5762, x = aw_87_cast_fp16)[name = tensor("op_6213_cast_fp16")]; + tensor var_6214_cast_fp16 = softmax(axis = var_5762, x = aw_89_cast_fp16)[name = tensor("op_6214_cast_fp16")]; + tensor var_6215_cast_fp16 = softmax(axis = var_5762, x = aw_91_cast_fp16)[name = tensor("op_6215_cast_fp16")]; + tensor var_6216_cast_fp16 = softmax(axis = var_5762, x = aw_93_cast_fp16)[name = tensor("op_6216_cast_fp16")]; + tensor var_6217_cast_fp16 = softmax(axis = var_5762, x = aw_95_cast_fp16)[name = tensor("op_6217_cast_fp16")]; + tensor var_6219_equation_0 = const()[name = tensor("op_6219_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6219_cast_fp16 = einsum(equation = var_6219_equation_0, values = (var_6147_cast_fp16, var_6210_cast_fp16))[name = tensor("op_6219_cast_fp16")]; + tensor var_6221_equation_0 = const()[name = tensor("op_6221_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6221_cast_fp16 = einsum(equation = var_6221_equation_0, values = (var_6151_cast_fp16, var_6211_cast_fp16))[name = tensor("op_6221_cast_fp16")]; + tensor var_6223_equation_0 = const()[name = tensor("op_6223_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6223_cast_fp16 = einsum(equation = var_6223_equation_0, values = (var_6155_cast_fp16, var_6212_cast_fp16))[name = tensor("op_6223_cast_fp16")]; + tensor var_6225_equation_0 = const()[name = tensor("op_6225_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6225_cast_fp16 = einsum(equation = var_6225_equation_0, values = (var_6159_cast_fp16, var_6213_cast_fp16))[name = tensor("op_6225_cast_fp16")]; + tensor var_6227_equation_0 = const()[name = tensor("op_6227_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6227_cast_fp16 = einsum(equation = var_6227_equation_0, values = (var_6163_cast_fp16, var_6214_cast_fp16))[name = tensor("op_6227_cast_fp16")]; + tensor var_6229_equation_0 = const()[name = tensor("op_6229_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6229_cast_fp16 = einsum(equation = var_6229_equation_0, values = (var_6167_cast_fp16, var_6215_cast_fp16))[name = tensor("op_6229_cast_fp16")]; + tensor var_6231_equation_0 = const()[name = tensor("op_6231_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6231_cast_fp16 = einsum(equation = var_6231_equation_0, values = (var_6171_cast_fp16, var_6216_cast_fp16))[name = tensor("op_6231_cast_fp16")]; + tensor var_6233_equation_0 = const()[name = tensor("op_6233_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6233_cast_fp16 = einsum(equation = var_6233_equation_0, values = (var_6175_cast_fp16, var_6217_cast_fp16))[name = tensor("op_6233_cast_fp16")]; + tensor input_217_interleave_0 = const()[name = tensor("input_217_interleave_0"), val = tensor(false)]; + tensor input_217_cast_fp16 = concat(axis = var_5762, interleave = input_217_interleave_0, values = (var_6219_cast_fp16, var_6221_cast_fp16, var_6223_cast_fp16, var_6225_cast_fp16, var_6227_cast_fp16, var_6229_cast_fp16, var_6231_cast_fp16, var_6233_cast_fp16))[name = tensor("input_217_cast_fp16")]; + tensor var_6243_pad_type_0 = const()[name = tensor("op_6243_pad_type_0"), val = tensor("valid")]; + tensor var_6243_strides_0 = const()[name = tensor("op_6243_strides_0"), val = tensor([1, 1])]; + tensor var_6243_pad_0 = const()[name = tensor("op_6243_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6243_dilations_0 = const()[name = tensor("op_6243_dilations_0"), val = tensor([1, 1])]; + tensor var_6243_groups_0 = const()[name = tensor("op_6243_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(589267904)))]; + tensor mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(592544768)))]; + tensor var_6243_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_6243_dilations_0, groups = var_6243_groups_0, pad = var_6243_pad_0, pad_type = var_6243_pad_type_0, strides = var_6243_strides_0, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_217_cast_fp16)[name = tensor("op_6243_cast_fp16")]; + tensor inputs_41_cast_fp16 = add(x = var_6243_cast_fp16, y = inputs_39_cast_fp16)[name = tensor("inputs_41_cast_fp16")]; + tensor input_219_axes_0 = const()[name = tensor("input_219_axes_0"), val = tensor([1])]; + tensor input_219_gamma_0_to_fp16 = const()[name = tensor("input_219_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(592547392)))]; + tensor input_219_beta_0_to_fp16 = const()[name = tensor("input_219_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(592550016)))]; + tensor var_6253_to_fp16 = const()[name = tensor("op_6253_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_219_cast_fp16 = layer_norm(axes = input_219_axes_0, beta = input_219_beta_0_to_fp16, epsilon = var_6253_to_fp16, gamma = input_219_gamma_0_to_fp16, x = inputs_41_cast_fp16)[name = tensor("input_219_cast_fp16")]; + tensor var_6273_pad_type_0 = const()[name = tensor("op_6273_pad_type_0"), val = tensor("valid")]; + tensor var_6273_strides_0 = const()[name = tensor("op_6273_strides_0"), val = tensor([1, 1])]; + tensor var_6273_pad_0 = const()[name = tensor("op_6273_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6273_dilations_0 = const()[name = tensor("op_6273_dilations_0"), val = tensor([1, 1])]; + tensor var_6273_groups_0 = const()[name = tensor("op_6273_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(592552640)))]; + tensor mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(618767104)))]; + tensor var_6273_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_6273_dilations_0, groups = var_6273_groups_0, pad = var_6273_pad_0, pad_type = var_6273_pad_type_0, strides = var_6273_strides_0, weight = mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_219_cast_fp16)[name = tensor("op_6273_cast_fp16")]; + tensor var_6274_split_sizes_0 = const()[name = tensor("op_6274_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_6274_axis_0 = const()[name = tensor("op_6274_axis_0"), val = tensor(1)]; + tensor var_6274_cast_fp16_0, tensor var_6274_cast_fp16_1 = split(axis = var_6274_axis_0, split_sizes = var_6274_split_sizes_0, x = var_6273_cast_fp16)[name = tensor("op_6274_cast_fp16")]; + tensor var_6276_mode_0 = const()[name = tensor("op_6276_mode_0"), val = tensor("EXACT")]; + tensor var_6276_cast_fp16 = gelu(mode = var_6276_mode_0, x = var_6274_cast_fp16_1)[name = tensor("op_6276_cast_fp16")]; + tensor input_221_cast_fp16 = mul(x = var_6274_cast_fp16_0, y = var_6276_cast_fp16)[name = tensor("input_221_cast_fp16")]; + tensor var_6284_pad_type_0 = const()[name = tensor("op_6284_pad_type_0"), val = tensor("valid")]; + tensor var_6284_strides_0 = const()[name = tensor("op_6284_strides_0"), val = tensor([1, 1])]; + tensor var_6284_pad_0 = const()[name = tensor("op_6284_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6284_dilations_0 = const()[name = tensor("op_6284_dilations_0"), val = tensor([1, 1])]; + tensor var_6284_groups_0 = const()[name = tensor("op_6284_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(618787648)))]; + tensor mid_block_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(631894912)))]; + tensor var_6284_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_6284_dilations_0, groups = var_6284_groups_0, pad = var_6284_pad_0, pad_type = var_6284_pad_type_0, strides = var_6284_strides_0, weight = mid_block_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_221_cast_fp16)[name = tensor("op_6284_cast_fp16")]; + tensor hidden_states_131_cast_fp16 = add(x = var_6284_cast_fp16, y = inputs_41_cast_fp16)[name = tensor("hidden_states_131_cast_fp16")]; + tensor var_6286 = const()[name = tensor("op_6286"), val = tensor([2, 1280, 8, 8])]; + tensor input_223_cast_fp16 = reshape(shape = var_6286, x = hidden_states_131_cast_fp16)[name = tensor("input_223_cast_fp16")]; + tensor hidden_states_133_pad_type_0 = const()[name = tensor("hidden_states_133_pad_type_0"), val = tensor("valid")]; + tensor hidden_states_133_strides_0 = const()[name = tensor("hidden_states_133_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_133_pad_0 = const()[name = tensor("hidden_states_133_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_133_dilations_0 = const()[name = tensor("hidden_states_133_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_133_groups_0 = const()[name = tensor("hidden_states_133_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(631897536)))]; + tensor mid_block_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(635174400)))]; + tensor hidden_states_133_cast_fp16 = conv(bias = mid_block_attentions_0_proj_out_bias_to_fp16, dilations = hidden_states_133_dilations_0, groups = hidden_states_133_groups_0, pad = hidden_states_133_pad_0, pad_type = hidden_states_133_pad_type_0, strides = hidden_states_133_strides_0, weight = mid_block_attentions_0_proj_out_weight_to_fp16, x = input_223_cast_fp16)[name = tensor("hidden_states_133_cast_fp16")]; + tensor input_225_cast_fp16 = add(x = hidden_states_133_cast_fp16, y = hidden_states_121_cast_fp16)[name = tensor("input_225_cast_fp16")]; + tensor reshape_100_shape_0 = const()[name = tensor("reshape_100_shape_0"), val = tensor([2, 32, 40, 8, 8])]; + tensor reshape_100_cast_fp16 = reshape(shape = reshape_100_shape_0, x = input_225_cast_fp16)[name = tensor("reshape_100_cast_fp16")]; + tensor reduce_mean_75_axes_0 = const()[name = tensor("reduce_mean_75_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_75_keep_dims_0 = const()[name = tensor("reduce_mean_75_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_75_cast_fp16 = reduce_mean(axes = reduce_mean_75_axes_0, keep_dims = reduce_mean_75_keep_dims_0, x = reshape_100_cast_fp16)[name = tensor("reduce_mean_75_cast_fp16")]; + tensor sub_50_cast_fp16 = sub(x = reshape_100_cast_fp16, y = reduce_mean_75_cast_fp16)[name = tensor("sub_50_cast_fp16")]; + tensor square_25_cast_fp16 = square(x = sub_50_cast_fp16)[name = tensor("square_25_cast_fp16")]; + tensor reduce_mean_77_axes_0 = const()[name = tensor("reduce_mean_77_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_77_keep_dims_0 = const()[name = tensor("reduce_mean_77_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_77_cast_fp16 = reduce_mean(axes = reduce_mean_77_axes_0, keep_dims = reduce_mean_77_keep_dims_0, x = square_25_cast_fp16)[name = tensor("reduce_mean_77_cast_fp16")]; + tensor add_50_y_0_to_fp16 = const()[name = tensor("add_50_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_50_cast_fp16 = add(x = reduce_mean_77_cast_fp16, y = add_50_y_0_to_fp16)[name = tensor("add_50_cast_fp16")]; + tensor sqrt_25_cast_fp16 = sqrt(x = add_50_cast_fp16)[name = tensor("sqrt_25_cast_fp16")]; + tensor real_div_25_cast_fp16 = real_div(x = sub_50_cast_fp16, y = sqrt_25_cast_fp16)[name = tensor("real_div_25_cast_fp16")]; + tensor reshape_101_shape_0 = const()[name = tensor("reshape_101_shape_0"), val = tensor([2, 1280, 8, 8])]; + tensor reshape_101_cast_fp16 = reshape(shape = reshape_101_shape_0, x = real_div_25_cast_fp16)[name = tensor("reshape_101_cast_fp16")]; + tensor add_51_gamma_0_to_fp16 = const()[name = tensor("add_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(635177024)))]; + tensor add_51_beta_0_to_fp16 = const()[name = tensor("add_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(635179648)))]; + tensor add_51_epsilon_0_to_fp16 = const()[name = tensor("add_51_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_51_cast_fp16 = batch_norm(beta = add_51_beta_0_to_fp16, epsilon = add_51_epsilon_0_to_fp16, gamma = add_51_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_101_cast_fp16)[name = tensor("add_51_cast_fp16")]; + tensor input_229_cast_fp16 = silu(x = add_51_cast_fp16)[name = tensor("input_229_cast_fp16")]; + tensor hidden_states_135_pad_type_0 = const()[name = tensor("hidden_states_135_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_135_pad_0 = const()[name = tensor("hidden_states_135_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_135_strides_0 = const()[name = tensor("hidden_states_135_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_135_dilations_0 = const()[name = tensor("hidden_states_135_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_135_groups_0 = const()[name = tensor("hidden_states_135_groups_0"), val = tensor(1)]; + tensor mid_block_resnets_1_conv1_weight_to_fp16 = const()[name = tensor("mid_block_resnets_1_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(635182272)))]; + tensor mid_block_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("mid_block_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(664673536)))]; + tensor hidden_states_135_cast_fp16 = conv(bias = mid_block_resnets_1_conv1_bias_to_fp16, dilations = hidden_states_135_dilations_0, groups = hidden_states_135_groups_0, pad = hidden_states_135_pad_0, pad_type = hidden_states_135_pad_type_0, strides = hidden_states_135_strides_0, weight = mid_block_resnets_1_conv1_weight_to_fp16, x = input_229_cast_fp16)[name = tensor("hidden_states_135_cast_fp16")]; + tensor temb_19_pad_type_0 = const()[name = tensor("temb_19_pad_type_0"), val = tensor("valid")]; + tensor temb_19_strides_0 = const()[name = tensor("temb_19_strides_0"), val = tensor([1, 1])]; + tensor temb_19_pad_0 = const()[name = tensor("temb_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor temb_19_dilations_0 = const()[name = tensor("temb_19_dilations_0"), val = tensor([1, 1])]; + tensor temb_19_groups_0 = const()[name = tensor("temb_19_groups_0"), val = tensor(1)]; + tensor mid_block_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor("mid_block_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(664676160)))]; + tensor mid_block_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("mid_block_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(667953024)))]; + tensor temb_19_cast_fp16 = conv(bias = mid_block_resnets_1_time_emb_proj_bias_to_fp16, dilations = temb_19_dilations_0, groups = temb_19_groups_0, pad = temb_19_pad_0, pad_type = temb_19_pad_type_0, strides = temb_19_strides_0, weight = mid_block_resnets_1_time_emb_proj_weight_to_fp16, x = input_15_cast_fp16_1)[name = tensor("temb_19_cast_fp16")]; + tensor input_233_cast_fp16 = add(x = hidden_states_135_cast_fp16, y = temb_19_cast_fp16)[name = tensor("input_233_cast_fp16")]; + tensor reshape_104_shape_0 = const()[name = tensor("reshape_104_shape_0"), val = tensor([2, 32, 40, 8, 8])]; + tensor reshape_104_cast_fp16 = reshape(shape = reshape_104_shape_0, x = input_233_cast_fp16)[name = tensor("reshape_104_cast_fp16")]; + tensor reduce_mean_78_axes_0 = const()[name = tensor("reduce_mean_78_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_78_keep_dims_0 = const()[name = tensor("reduce_mean_78_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_78_cast_fp16 = reduce_mean(axes = reduce_mean_78_axes_0, keep_dims = reduce_mean_78_keep_dims_0, x = reshape_104_cast_fp16)[name = tensor("reduce_mean_78_cast_fp16")]; + tensor sub_52_cast_fp16 = sub(x = reshape_104_cast_fp16, y = reduce_mean_78_cast_fp16)[name = tensor("sub_52_cast_fp16")]; + tensor square_26_cast_fp16 = square(x = sub_52_cast_fp16)[name = tensor("square_26_cast_fp16")]; + tensor reduce_mean_80_axes_0 = const()[name = tensor("reduce_mean_80_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_80_keep_dims_0 = const()[name = tensor("reduce_mean_80_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_80_cast_fp16 = reduce_mean(axes = reduce_mean_80_axes_0, keep_dims = reduce_mean_80_keep_dims_0, x = square_26_cast_fp16)[name = tensor("reduce_mean_80_cast_fp16")]; + tensor add_52_y_0_to_fp16 = const()[name = tensor("add_52_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_52_cast_fp16 = add(x = reduce_mean_80_cast_fp16, y = add_52_y_0_to_fp16)[name = tensor("add_52_cast_fp16")]; + tensor sqrt_26_cast_fp16 = sqrt(x = add_52_cast_fp16)[name = tensor("sqrt_26_cast_fp16")]; + tensor real_div_26_cast_fp16 = real_div(x = sub_52_cast_fp16, y = sqrt_26_cast_fp16)[name = tensor("real_div_26_cast_fp16")]; + tensor reshape_105_shape_0 = const()[name = tensor("reshape_105_shape_0"), val = tensor([2, 1280, 8, 8])]; + tensor reshape_105_cast_fp16 = reshape(shape = reshape_105_shape_0, x = real_div_26_cast_fp16)[name = tensor("reshape_105_cast_fp16")]; + tensor add_53_gamma_0_to_fp16 = const()[name = tensor("add_53_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(667955648)))]; + tensor add_53_beta_0_to_fp16 = const()[name = tensor("add_53_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(667958272)))]; + tensor add_53_epsilon_0_to_fp16 = const()[name = tensor("add_53_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_53_cast_fp16 = batch_norm(beta = add_53_beta_0_to_fp16, epsilon = add_53_epsilon_0_to_fp16, gamma = add_53_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_105_cast_fp16)[name = tensor("add_53_cast_fp16")]; + tensor input_237_cast_fp16 = silu(x = add_53_cast_fp16)[name = tensor("input_237_cast_fp16")]; + tensor hidden_states_137_pad_type_0 = const()[name = tensor("hidden_states_137_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_137_pad_0 = const()[name = tensor("hidden_states_137_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_137_strides_0 = const()[name = tensor("hidden_states_137_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_137_dilations_0 = const()[name = tensor("hidden_states_137_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_137_groups_0 = const()[name = tensor("hidden_states_137_groups_0"), val = tensor(1)]; + tensor mid_block_resnets_1_conv2_weight_to_fp16 = const()[name = tensor("mid_block_resnets_1_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(667960896)))]; + tensor mid_block_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("mid_block_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(697452160)))]; + tensor hidden_states_137_cast_fp16 = conv(bias = mid_block_resnets_1_conv2_bias_to_fp16, dilations = hidden_states_137_dilations_0, groups = hidden_states_137_groups_0, pad = hidden_states_137_pad_0, pad_type = hidden_states_137_pad_type_0, strides = hidden_states_137_strides_0, weight = mid_block_resnets_1_conv2_weight_to_fp16, x = input_237_cast_fp16)[name = tensor("hidden_states_137_cast_fp16")]; + tensor hidden_states_139_cast_fp16 = add(x = input_225_cast_fp16, y = hidden_states_137_cast_fp16)[name = tensor("hidden_states_139_cast_fp16")]; + tensor var_6340 = const()[name = tensor("op_6340"), val = tensor(1)]; + tensor input_239_interleave_0 = const()[name = tensor("input_239_interleave_0"), val = tensor(false)]; + tensor input_239_cast_fp16 = concat(axis = var_6340, interleave = input_239_interleave_0, values = (hidden_states_139_cast_fp16, input_199_cast_fp16))[name = tensor("input_239_cast_fp16")]; + tensor reshape_108_shape_0 = const()[name = tensor("reshape_108_shape_0"), val = tensor([2, 32, 80, 8, 8])]; + tensor reshape_108_cast_fp16 = reshape(shape = reshape_108_shape_0, x = input_239_cast_fp16)[name = tensor("reshape_108_cast_fp16")]; + tensor reduce_mean_81_axes_0 = const()[name = tensor("reduce_mean_81_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_81_keep_dims_0 = const()[name = tensor("reduce_mean_81_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_81_cast_fp16 = reduce_mean(axes = reduce_mean_81_axes_0, keep_dims = reduce_mean_81_keep_dims_0, x = reshape_108_cast_fp16)[name = tensor("reduce_mean_81_cast_fp16")]; + tensor sub_54_cast_fp16 = sub(x = reshape_108_cast_fp16, y = reduce_mean_81_cast_fp16)[name = tensor("sub_54_cast_fp16")]; + tensor square_27_cast_fp16 = square(x = sub_54_cast_fp16)[name = tensor("square_27_cast_fp16")]; + tensor reduce_mean_83_axes_0 = const()[name = tensor("reduce_mean_83_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_83_keep_dims_0 = const()[name = tensor("reduce_mean_83_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_83_cast_fp16 = reduce_mean(axes = reduce_mean_83_axes_0, keep_dims = reduce_mean_83_keep_dims_0, x = square_27_cast_fp16)[name = tensor("reduce_mean_83_cast_fp16")]; + tensor add_54_y_0_to_fp16 = const()[name = tensor("add_54_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_54_cast_fp16 = add(x = reduce_mean_83_cast_fp16, y = add_54_y_0_to_fp16)[name = tensor("add_54_cast_fp16")]; + tensor sqrt_27_cast_fp16 = sqrt(x = add_54_cast_fp16)[name = tensor("sqrt_27_cast_fp16")]; + tensor real_div_27_cast_fp16 = real_div(x = sub_54_cast_fp16, y = sqrt_27_cast_fp16)[name = tensor("real_div_27_cast_fp16")]; + tensor reshape_109_shape_0 = const()[name = tensor("reshape_109_shape_0"), val = tensor([2, 2560, 8, 8])]; + tensor reshape_109_cast_fp16 = reshape(shape = reshape_109_shape_0, x = real_div_27_cast_fp16)[name = tensor("reshape_109_cast_fp16")]; + 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(697454784)))]; + 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(697459968)))]; + tensor add_55_gamma_0_to_fp16 = const()[name = tensor("add_55_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(697465152)))]; + tensor add_55_beta_0_to_fp16 = const()[name = tensor("add_55_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(697470336)))]; + tensor add_55_epsilon_0_to_fp16 = const()[name = tensor("add_55_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_55_cast_fp16 = batch_norm(beta = add_55_beta_0_to_fp16, epsilon = add_55_epsilon_0_to_fp16, gamma = add_55_gamma_0_to_fp16, mean = add_55_mean_0_to_fp16, variance = add_55_variance_0_to_fp16, x = reshape_109_cast_fp16)[name = tensor("add_55_cast_fp16")]; + tensor input_243_cast_fp16 = silu(x = add_55_cast_fp16)[name = tensor("input_243_cast_fp16")]; + tensor hidden_states_141_pad_type_0 = const()[name = tensor("hidden_states_141_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_141_pad_0 = const()[name = tensor("hidden_states_141_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_141_strides_0 = const()[name = tensor("hidden_states_141_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_141_dilations_0 = const()[name = tensor("hidden_states_141_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_141_groups_0 = const()[name = tensor("hidden_states_141_groups_0"), val = tensor(1)]; + tensor up_blocks_0_resnets_0_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_0_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(697475520)))]; + tensor up_blocks_0_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(756457984)))]; + tensor hidden_states_141_cast_fp16 = conv(bias = up_blocks_0_resnets_0_conv1_bias_to_fp16, dilations = hidden_states_141_dilations_0, groups = hidden_states_141_groups_0, pad = hidden_states_141_pad_0, pad_type = hidden_states_141_pad_type_0, strides = hidden_states_141_strides_0, weight = up_blocks_0_resnets_0_conv1_weight_to_fp16, x = input_243_cast_fp16)[name = tensor("hidden_states_141_cast_fp16")]; + tensor temb_21_pad_type_0 = const()[name = tensor("temb_21_pad_type_0"), val = tensor("valid")]; + tensor temb_21_strides_0 = const()[name = tensor("temb_21_strides_0"), val = tensor([1, 1])]; + tensor temb_21_pad_0 = const()[name = tensor("temb_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor temb_21_dilations_0 = const()[name = tensor("temb_21_dilations_0"), val = tensor([1, 1])]; + tensor temb_21_groups_0 = const()[name = tensor("temb_21_groups_0"), val = tensor(1)]; + tensor up_blocks_0_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(756460608)))]; + tensor up_blocks_0_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(759737472)))]; + tensor temb_21_cast_fp16 = conv(bias = up_blocks_0_resnets_0_time_emb_proj_bias_to_fp16, dilations = temb_21_dilations_0, groups = temb_21_groups_0, pad = temb_21_pad_0, pad_type = temb_21_pad_type_0, strides = temb_21_strides_0, weight = up_blocks_0_resnets_0_time_emb_proj_weight_to_fp16, x = input_15_cast_fp16_1)[name = tensor("temb_21_cast_fp16")]; + tensor input_247_cast_fp16 = add(x = hidden_states_141_cast_fp16, y = temb_21_cast_fp16)[name = tensor("input_247_cast_fp16")]; + tensor reshape_112_shape_0 = const()[name = tensor("reshape_112_shape_0"), val = tensor([2, 32, 40, 8, 8])]; + tensor reshape_112_cast_fp16 = reshape(shape = reshape_112_shape_0, x = input_247_cast_fp16)[name = tensor("reshape_112_cast_fp16")]; + tensor reduce_mean_84_axes_0 = const()[name = tensor("reduce_mean_84_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_84_keep_dims_0 = const()[name = tensor("reduce_mean_84_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_84_cast_fp16 = reduce_mean(axes = reduce_mean_84_axes_0, keep_dims = reduce_mean_84_keep_dims_0, x = reshape_112_cast_fp16)[name = tensor("reduce_mean_84_cast_fp16")]; + tensor sub_56_cast_fp16 = sub(x = reshape_112_cast_fp16, y = reduce_mean_84_cast_fp16)[name = tensor("sub_56_cast_fp16")]; + tensor square_28_cast_fp16 = square(x = sub_56_cast_fp16)[name = tensor("square_28_cast_fp16")]; + tensor reduce_mean_86_axes_0 = const()[name = tensor("reduce_mean_86_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_86_keep_dims_0 = const()[name = tensor("reduce_mean_86_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_86_cast_fp16 = reduce_mean(axes = reduce_mean_86_axes_0, keep_dims = reduce_mean_86_keep_dims_0, x = square_28_cast_fp16)[name = tensor("reduce_mean_86_cast_fp16")]; + tensor add_56_y_0_to_fp16 = const()[name = tensor("add_56_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_56_cast_fp16 = add(x = reduce_mean_86_cast_fp16, y = add_56_y_0_to_fp16)[name = tensor("add_56_cast_fp16")]; + tensor sqrt_28_cast_fp16 = sqrt(x = add_56_cast_fp16)[name = tensor("sqrt_28_cast_fp16")]; + tensor real_div_28_cast_fp16 = real_div(x = sub_56_cast_fp16, y = sqrt_28_cast_fp16)[name = tensor("real_div_28_cast_fp16")]; + tensor reshape_113_shape_0 = const()[name = tensor("reshape_113_shape_0"), val = tensor([2, 1280, 8, 8])]; + tensor reshape_113_cast_fp16 = reshape(shape = reshape_113_shape_0, x = real_div_28_cast_fp16)[name = tensor("reshape_113_cast_fp16")]; + tensor add_57_gamma_0_to_fp16 = const()[name = tensor("add_57_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(759740096)))]; + tensor add_57_beta_0_to_fp16 = const()[name = tensor("add_57_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(759742720)))]; + tensor add_57_epsilon_0_to_fp16 = const()[name = tensor("add_57_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_57_cast_fp16 = batch_norm(beta = add_57_beta_0_to_fp16, epsilon = add_57_epsilon_0_to_fp16, gamma = add_57_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_113_cast_fp16)[name = tensor("add_57_cast_fp16")]; + tensor input_251_cast_fp16 = silu(x = add_57_cast_fp16)[name = tensor("input_251_cast_fp16")]; + tensor hidden_states_143_pad_type_0 = const()[name = tensor("hidden_states_143_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_143_pad_0 = const()[name = tensor("hidden_states_143_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_143_strides_0 = const()[name = tensor("hidden_states_143_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_143_dilations_0 = const()[name = tensor("hidden_states_143_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_143_groups_0 = const()[name = tensor("hidden_states_143_groups_0"), val = tensor(1)]; + tensor up_blocks_0_resnets_0_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_0_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(759745344)))]; + tensor up_blocks_0_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(789236608)))]; + tensor hidden_states_143_cast_fp16 = conv(bias = up_blocks_0_resnets_0_conv2_bias_to_fp16, dilations = hidden_states_143_dilations_0, groups = hidden_states_143_groups_0, pad = hidden_states_143_pad_0, pad_type = hidden_states_143_pad_type_0, strides = hidden_states_143_strides_0, weight = up_blocks_0_resnets_0_conv2_weight_to_fp16, x = input_251_cast_fp16)[name = tensor("hidden_states_143_cast_fp16")]; + tensor x_5_pad_type_0 = const()[name = tensor("x_5_pad_type_0"), val = tensor("valid")]; + tensor x_5_strides_0 = const()[name = tensor("x_5_strides_0"), val = tensor([1, 1])]; + tensor x_5_pad_0 = const()[name = tensor("x_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_5_dilations_0 = const()[name = tensor("x_5_dilations_0"), val = tensor([1, 1])]; + tensor x_5_groups_0 = const()[name = tensor("x_5_groups_0"), val = tensor(1)]; + tensor up_blocks_0_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(789239232)))]; + tensor up_blocks_0_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(795792896)))]; + tensor x_5_cast_fp16 = conv(bias = up_blocks_0_resnets_0_conv_shortcut_bias_to_fp16, dilations = x_5_dilations_0, groups = x_5_groups_0, pad = x_5_pad_0, pad_type = x_5_pad_type_0, strides = x_5_strides_0, weight = up_blocks_0_resnets_0_conv_shortcut_weight_to_fp16, x = input_239_cast_fp16)[name = tensor("x_5_cast_fp16")]; + tensor hidden_states_145_cast_fp16 = add(x = x_5_cast_fp16, y = hidden_states_143_cast_fp16)[name = tensor("hidden_states_145_cast_fp16")]; + tensor input_253_interleave_0 = const()[name = tensor("input_253_interleave_0"), val = tensor(false)]; + tensor input_253_cast_fp16_1 = concat(axis = var_6340, interleave = input_253_interleave_0, values = (hidden_states_145_cast_fp16, input_185_cast_fp16))[name = tensor("input_253_cast_fp16")]; + tensor reshape_116_shape_0 = const()[name = tensor("reshape_116_shape_0"), val = tensor([2, 32, 80, 8, 8])]; + tensor reshape_116_cast_fp16 = reshape(shape = reshape_116_shape_0, x = input_253_cast_fp16_1)[name = tensor("reshape_116_cast_fp16")]; + tensor reduce_mean_87_axes_0 = const()[name = tensor("reduce_mean_87_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_87_keep_dims_0 = const()[name = tensor("reduce_mean_87_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_87_cast_fp16 = reduce_mean(axes = reduce_mean_87_axes_0, keep_dims = reduce_mean_87_keep_dims_0, x = reshape_116_cast_fp16)[name = tensor("reduce_mean_87_cast_fp16")]; + tensor sub_58_cast_fp16 = sub(x = reshape_116_cast_fp16, y = reduce_mean_87_cast_fp16)[name = tensor("sub_58_cast_fp16")]; + tensor square_29_cast_fp16 = square(x = sub_58_cast_fp16)[name = tensor("square_29_cast_fp16")]; + tensor reduce_mean_89_axes_0 = const()[name = tensor("reduce_mean_89_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_89_keep_dims_0 = const()[name = tensor("reduce_mean_89_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_89_cast_fp16 = reduce_mean(axes = reduce_mean_89_axes_0, keep_dims = reduce_mean_89_keep_dims_0, x = square_29_cast_fp16)[name = tensor("reduce_mean_89_cast_fp16")]; + tensor add_58_y_0_to_fp16 = const()[name = tensor("add_58_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_58_cast_fp16 = add(x = reduce_mean_89_cast_fp16, y = add_58_y_0_to_fp16)[name = tensor("add_58_cast_fp16")]; + tensor sqrt_29_cast_fp16 = sqrt(x = add_58_cast_fp16)[name = tensor("sqrt_29_cast_fp16")]; + tensor real_div_29_cast_fp16 = real_div(x = sub_58_cast_fp16, y = sqrt_29_cast_fp16)[name = tensor("real_div_29_cast_fp16")]; + tensor reshape_117_shape_0 = const()[name = tensor("reshape_117_shape_0"), val = tensor([2, 2560, 8, 8])]; + tensor reshape_117_cast_fp16 = reshape(shape = reshape_117_shape_0, x = real_div_29_cast_fp16)[name = tensor("reshape_117_cast_fp16")]; + tensor add_59_gamma_0_to_fp16 = const()[name = tensor("add_59_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(795795520)))]; + tensor add_59_beta_0_to_fp16 = const()[name = tensor("add_59_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(795800704)))]; + tensor add_59_epsilon_0_to_fp16 = const()[name = tensor("add_59_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_59_cast_fp16 = batch_norm(beta = add_59_beta_0_to_fp16, epsilon = add_59_epsilon_0_to_fp16, gamma = add_59_gamma_0_to_fp16, mean = add_55_mean_0_to_fp16, variance = add_55_variance_0_to_fp16, x = reshape_117_cast_fp16)[name = tensor("add_59_cast_fp16")]; + tensor input_257_cast_fp16 = silu(x = add_59_cast_fp16)[name = tensor("input_257_cast_fp16")]; + tensor hidden_states_147_pad_type_0 = const()[name = tensor("hidden_states_147_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_147_pad_0 = const()[name = tensor("hidden_states_147_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_147_strides_0 = const()[name = tensor("hidden_states_147_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_147_dilations_0 = const()[name = tensor("hidden_states_147_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_147_groups_0 = const()[name = tensor("hidden_states_147_groups_0"), val = tensor(1)]; + tensor up_blocks_0_resnets_1_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_1_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(795805888)))]; + tensor up_blocks_0_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(854788352)))]; + tensor hidden_states_147_cast_fp16 = conv(bias = up_blocks_0_resnets_1_conv1_bias_to_fp16, dilations = hidden_states_147_dilations_0, groups = hidden_states_147_groups_0, pad = hidden_states_147_pad_0, pad_type = hidden_states_147_pad_type_0, strides = hidden_states_147_strides_0, weight = up_blocks_0_resnets_1_conv1_weight_to_fp16, x = input_257_cast_fp16)[name = tensor("hidden_states_147_cast_fp16")]; + tensor temb_23_pad_type_0 = const()[name = tensor("temb_23_pad_type_0"), val = tensor("valid")]; + tensor temb_23_strides_0 = const()[name = tensor("temb_23_strides_0"), val = tensor([1, 1])]; + tensor temb_23_pad_0 = const()[name = tensor("temb_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor temb_23_dilations_0 = const()[name = tensor("temb_23_dilations_0"), val = tensor([1, 1])]; + tensor temb_23_groups_0 = const()[name = tensor("temb_23_groups_0"), val = tensor(1)]; + tensor up_blocks_0_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(854790976)))]; + tensor up_blocks_0_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(858067840)))]; + tensor temb_23_cast_fp16 = conv(bias = up_blocks_0_resnets_1_time_emb_proj_bias_to_fp16, dilations = temb_23_dilations_0, groups = temb_23_groups_0, pad = temb_23_pad_0, pad_type = temb_23_pad_type_0, strides = temb_23_strides_0, weight = up_blocks_0_resnets_1_time_emb_proj_weight_to_fp16, x = input_15_cast_fp16_1)[name = tensor("temb_23_cast_fp16")]; + tensor input_261_cast_fp16 = add(x = hidden_states_147_cast_fp16, y = temb_23_cast_fp16)[name = tensor("input_261_cast_fp16")]; + tensor reshape_120_shape_0 = const()[name = tensor("reshape_120_shape_0"), val = tensor([2, 32, 40, 8, 8])]; + tensor reshape_120_cast_fp16 = reshape(shape = reshape_120_shape_0, x = input_261_cast_fp16)[name = tensor("reshape_120_cast_fp16")]; + tensor reduce_mean_90_axes_0 = const()[name = tensor("reduce_mean_90_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_90_keep_dims_0 = const()[name = tensor("reduce_mean_90_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_90_cast_fp16 = reduce_mean(axes = reduce_mean_90_axes_0, keep_dims = reduce_mean_90_keep_dims_0, x = reshape_120_cast_fp16)[name = tensor("reduce_mean_90_cast_fp16")]; + tensor sub_60_cast_fp16 = sub(x = reshape_120_cast_fp16, y = reduce_mean_90_cast_fp16)[name = tensor("sub_60_cast_fp16")]; + tensor square_30_cast_fp16 = square(x = sub_60_cast_fp16)[name = tensor("square_30_cast_fp16")]; + tensor reduce_mean_92_axes_0 = const()[name = tensor("reduce_mean_92_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_92_keep_dims_0 = const()[name = tensor("reduce_mean_92_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_92_cast_fp16 = reduce_mean(axes = reduce_mean_92_axes_0, keep_dims = reduce_mean_92_keep_dims_0, x = square_30_cast_fp16)[name = tensor("reduce_mean_92_cast_fp16")]; + tensor add_60_y_0_to_fp16 = const()[name = tensor("add_60_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_60_cast_fp16 = add(x = reduce_mean_92_cast_fp16, y = add_60_y_0_to_fp16)[name = tensor("add_60_cast_fp16")]; + tensor sqrt_30_cast_fp16 = sqrt(x = add_60_cast_fp16)[name = tensor("sqrt_30_cast_fp16")]; + tensor real_div_30_cast_fp16 = real_div(x = sub_60_cast_fp16, y = sqrt_30_cast_fp16)[name = tensor("real_div_30_cast_fp16")]; + tensor reshape_121_shape_0 = const()[name = tensor("reshape_121_shape_0"), val = tensor([2, 1280, 8, 8])]; + tensor reshape_121_cast_fp16 = reshape(shape = reshape_121_shape_0, x = real_div_30_cast_fp16)[name = tensor("reshape_121_cast_fp16")]; + tensor add_61_gamma_0_to_fp16 = const()[name = tensor("add_61_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(858070464)))]; + tensor add_61_beta_0_to_fp16 = const()[name = tensor("add_61_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(858073088)))]; + tensor add_61_epsilon_0_to_fp16 = const()[name = tensor("add_61_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_61_cast_fp16 = batch_norm(beta = add_61_beta_0_to_fp16, epsilon = add_61_epsilon_0_to_fp16, gamma = add_61_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_121_cast_fp16)[name = tensor("add_61_cast_fp16")]; + tensor input_265_cast_fp16 = silu(x = add_61_cast_fp16)[name = tensor("input_265_cast_fp16")]; + tensor hidden_states_149_pad_type_0 = const()[name = tensor("hidden_states_149_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_149_pad_0 = const()[name = tensor("hidden_states_149_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_149_strides_0 = const()[name = tensor("hidden_states_149_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_149_dilations_0 = const()[name = tensor("hidden_states_149_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_149_groups_0 = const()[name = tensor("hidden_states_149_groups_0"), val = tensor(1)]; + tensor up_blocks_0_resnets_1_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_1_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(858075712)))]; + tensor up_blocks_0_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(887566976)))]; + tensor hidden_states_149_cast_fp16_1 = conv(bias = up_blocks_0_resnets_1_conv2_bias_to_fp16, dilations = hidden_states_149_dilations_0, groups = hidden_states_149_groups_0, pad = hidden_states_149_pad_0, pad_type = hidden_states_149_pad_type_0, strides = hidden_states_149_strides_0, weight = up_blocks_0_resnets_1_conv2_weight_to_fp16, x = input_265_cast_fp16)[name = tensor("hidden_states_149_cast_fp16")]; + tensor input_89_cast_fp16_dtype_0 = const()[name = tensor("input_89_cast_fp16_dtype_0"), val = tensor("fp32")]; + tensor input_7_cast_fp16_dtype_0 = const()[name = tensor("input_7_cast_fp16_dtype_0"), val = tensor("fp32")]; + tensor input_63_cast_fp16_dtype_0 = const()[name = tensor("input_63_cast_fp16_dtype_0"), val = tensor("fp32")]; + tensor input_117_cast_fp16_dtype_0 = const()[name = tensor("input_117_cast_fp16_dtype_0"), val = tensor("fp32")]; + tensor input_171_cast_fp16_dtype_0 = const()[name = tensor("input_171_cast_fp16_dtype_0"), val = tensor("fp32")]; + tensor input_61_cast_fp16_dtype_0 = const()[name = tensor("input_61_cast_fp16_dtype_0"), val = tensor("fp32")]; + tensor input_15_cast_fp16_dtype_0 = const()[name = tensor("input_15_cast_fp16_dtype_0"), val = tensor("fp32")]; + tensor input_115_cast_fp16_dtype_0 = const()[name = tensor("input_115_cast_fp16_dtype_0"), val = tensor("fp32")]; + tensor input_169_cast_fp16_dtype_0 = const()[name = tensor("input_169_cast_fp16_dtype_0"), val = tensor("fp32")]; + tensor input_35_cast_fp16_dtype_0 = const()[name = tensor("input_35_cast_fp16_dtype_0"), val = tensor("fp32")]; + tensor input_143_cast_fp16_dtype_0 = const()[name = tensor("input_143_cast_fp16_dtype_0"), val = tensor("fp32")]; + tensor hidden_states_149_cast_fp16_dtype_0 = const()[name = tensor("hidden_states_149_cast_fp16_dtype_0"), val = tensor("fp32")]; + tensor input_253_cast_fp16_dtype_0 = const()[name = tensor("input_253_cast_fp16_dtype_0"), val = tensor("fp32")]; + tensor input_253_cast_fp16 = cast(dtype = input_253_cast_fp16_dtype_0, x = input_253_cast_fp16_1)[name = tensor("cast_13")]; + tensor hidden_states_149_cast_fp16 = cast(dtype = hidden_states_149_cast_fp16_dtype_0, x = hidden_states_149_cast_fp16_1)[name = tensor("cast_14")]; + tensor input_143_cast_fp16 = cast(dtype = input_143_cast_fp16_dtype_0, x = input_143_cast_fp16_1)[name = tensor("cast_15")]; + tensor input_35_cast_fp16 = cast(dtype = input_35_cast_fp16_dtype_0, x = input_35_cast_fp16_1)[name = tensor("cast_16")]; + tensor input_169_cast_fp16 = cast(dtype = input_169_cast_fp16_dtype_0, x = input_169_cast_fp16_1)[name = tensor("cast_17")]; + tensor input_115_cast_fp16 = cast(dtype = input_115_cast_fp16_dtype_0, x = input_115_cast_fp16_1)[name = tensor("cast_18")]; + tensor input_15_cast_fp16 = cast(dtype = input_15_cast_fp16_dtype_0, x = input_15_cast_fp16_1)[name = tensor("cast_19")]; + tensor input_61_cast_fp16 = cast(dtype = input_61_cast_fp16_dtype_0, x = input_61_cast_fp16_1)[name = tensor("cast_20")]; + tensor input_171_cast_fp16 = cast(dtype = input_171_cast_fp16_dtype_0, x = input_171_cast_fp16_1)[name = tensor("cast_21")]; + tensor input_117_cast_fp16 = cast(dtype = input_117_cast_fp16_dtype_0, x = input_117_cast_fp16_1)[name = tensor("cast_22")]; + tensor input_63_cast_fp16 = cast(dtype = input_63_cast_fp16_dtype_0, x = input_63_cast_fp16_1)[name = tensor("cast_23")]; + tensor input_7_cast_fp16 = cast(dtype = input_7_cast_fp16_dtype_0, x = input_7_cast_fp16_1)[name = tensor("cast_24")]; + tensor input_89_cast_fp16 = cast(dtype = input_89_cast_fp16_dtype_0, x = input_89_cast_fp16_1)[name = tensor("cast_25")]; + } -> (input_89_cast_fp16, input_7_cast_fp16, input_63_cast_fp16, input_117_cast_fp16, input_171_cast_fp16, input_61_cast_fp16, input_15_cast_fp16, input_115_cast_fp16, input_169_cast_fp16, input_35_cast_fp16, input_143_cast_fp16, hidden_states_149_cast_fp16, input_253_cast_fp16); +} \ No newline at end of file