program(1.3) [buildInfo = dict({{"coremlc-component-MIL", "3520.4.1"}, {"coremlc-version", "3520.5.1"}})] { func main(tensor input_ids) { int32 inputs_embeds_batch_dims_0 = const()[name = string("inputs_embeds_batch_dims_0"), val = int32(0)]; bool inputs_embeds_validate_indices_0 = const()[name = string("inputs_embeds_validate_indices_0"), val = bool(false)]; tensor encoder_embeddings_token_embedding_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28459136))))[name = string("encoder_embeddings_token_embedding_weight_to_fp16_palettized")]; int32 greater_equal_0_y_0 = const()[name = string("greater_equal_0_y_0"), val = int32(0)]; tensor greater_equal_0 = greater_equal(x = input_ids, y = greater_equal_0_y_0)[name = string("greater_equal_0")]; int32 slice_by_index_0 = const()[name = string("slice_by_index_0"), val = int32(49408)]; tensor add_12 = add(x = input_ids, y = slice_by_index_0)[name = string("add_12")]; tensor select_0 = select(a = input_ids, b = add_12, cond = greater_equal_0)[name = string("select_0")]; int32 greater_equal_0_y_0_1 = const()[name = string("greater_equal_0_y_0_1"), val = int32(0)]; tensor greater_equal_0_1 = greater_equal(x = select_0, y = greater_equal_0_y_0_1)[name = string("greater_equal_0_1")]; int32 slice_by_index_0_1 = const()[name = string("slice_by_index_0_1"), val = int32(49408)]; tensor add_0 = add(x = select_0, y = slice_by_index_0_1)[name = string("add_0")]; tensor select_0_1 = select(a = select_0, b = add_0, cond = greater_equal_0_1)[name = string("select_0_1")]; int32 inputs_embeds_cast_fp16_axis_0 = const()[name = string("inputs_embeds_cast_fp16_axis_0"), val = int32(0)]; tensor inputs_embeds_cast_fp16 = gather(axis = inputs_embeds_cast_fp16_axis_0, batch_dims = inputs_embeds_batch_dims_0, indices = select_0_1, validate_indices = inputs_embeds_validate_indices_0, x = encoder_embeddings_token_embedding_weight_to_fp16_palettized)[name = string("inputs_embeds_cast_fp16")]; tensor position_embeddings_to_fp16 = const()[name = string("position_embeddings_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28854464)))]; tensor input_3_cast_fp16 = add(x = inputs_embeds_cast_fp16, y = position_embeddings_to_fp16)[name = string("input_3_cast_fp16")]; tensor hidden_states_1_axes_0 = const()[name = string("hidden_states_1_axes_0"), val = tensor([-1])]; tensor encoder_encoder_layers_0_layer_norm1_weight_to_fp16 = const()[name = string("encoder_encoder_layers_0_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28972800)))]; tensor encoder_encoder_layers_0_layer_norm1_bias_to_fp16 = const()[name = string("encoder_encoder_layers_0_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28974400)))]; fp16 var_11_to_fp16 = const()[name = string("op_11_to_fp16"), val = fp16(0x1.5p-17)]; tensor hidden_states_1_cast_fp16 = layer_norm(axes = hidden_states_1_axes_0, beta = encoder_encoder_layers_0_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = encoder_encoder_layers_0_layer_norm1_weight_to_fp16, x = input_3_cast_fp16)[name = string("hidden_states_1_cast_fp16")]; tensor encoder_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28976000))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29418432))))[name = string("encoder_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_encoder_layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29424640)))]; tensor linear_0_cast_fp16 = linear(bias = encoder_encoder_layers_0_self_attn_q_proj_bias_to_fp16, weight = encoder_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_1_cast_fp16)[name = string("linear_0_cast_fp16")]; tensor encoder_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29426240))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29868672))))[name = string("encoder_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_0_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_encoder_layers_0_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29874880)))]; tensor linear_1_cast_fp16 = linear(bias = encoder_encoder_layers_0_self_attn_k_proj_bias_to_fp16, weight = encoder_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_1_cast_fp16)[name = string("linear_1_cast_fp16")]; tensor encoder_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29876480))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30318912))))[name = string("encoder_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_encoder_layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30325120)))]; tensor linear_2_cast_fp16 = linear(bias = encoder_encoder_layers_0_self_attn_v_proj_bias_to_fp16, weight = encoder_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_1_cast_fp16)[name = string("linear_2_cast_fp16")]; tensor var_82 = const()[name = string("op_82"), val = tensor([1, 77, -1, 64])]; tensor var_83_cast_fp16 = reshape(shape = var_82, x = linear_0_cast_fp16)[name = string("op_83_cast_fp16")]; tensor var_85 = const()[name = string("op_85"), val = tensor([1, 77, -1, 64])]; tensor var_86_cast_fp16 = reshape(shape = var_85, x = linear_1_cast_fp16)[name = string("op_86_cast_fp16")]; tensor var_88 = const()[name = string("op_88"), val = tensor([1, 77, -1, 64])]; tensor var_89_cast_fp16 = reshape(shape = var_88, x = linear_2_cast_fp16)[name = string("op_89_cast_fp16")]; tensor value_1_perm_0 = const()[name = string("value_1_perm_0"), val = tensor([0, 2, -3, -1])]; fp16 var_9_to_fp16 = const()[name = string("op_9_to_fp16"), val = fp16(0x1p-3)]; tensor mul_1_cast_fp16 = mul(x = var_83_cast_fp16, y = var_9_to_fp16)[name = string("mul_1_cast_fp16")]; bool matmul_0_transpose_y_0 = const()[name = string("matmul_0_transpose_y_0"), val = bool(true)]; bool matmul_0_transpose_x_0 = const()[name = string("matmul_0_transpose_x_0"), val = bool(false)]; tensor transpose_48_perm_0 = const()[name = string("transpose_48_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_49_perm_0 = const()[name = string("transpose_49_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_49 = transpose(perm = transpose_49_perm_0, x = var_86_cast_fp16)[name = string("transpose_118")]; tensor transpose_48 = transpose(perm = transpose_48_perm_0, x = mul_1_cast_fp16)[name = string("transpose_119")]; tensor matmul_0_cast_fp16 = matmul(transpose_x = matmul_0_transpose_x_0, transpose_y = matmul_0_transpose_y_0, x = transpose_48, y = transpose_49)[name = string("matmul_0_cast_fp16")]; tensor mul_0_to_fp16 = const()[name = string("mul_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30326720)))]; tensor add_0_cast_fp16 = add(x = matmul_0_cast_fp16, y = mul_0_to_fp16)[name = string("add_0_cast_fp16")]; int32 softmax_0_axis_0 = const()[name = string("softmax_0_axis_0"), val = int32(-1)]; tensor softmax_0_cast_fp16 = softmax(axis = softmax_0_axis_0, x = add_0_cast_fp16)[name = string("softmax_0_cast_fp16")]; bool attn_output_1_transpose_x_0 = const()[name = string("attn_output_1_transpose_x_0"), val = bool(false)]; bool attn_output_1_transpose_y_0 = const()[name = string("attn_output_1_transpose_y_0"), val = bool(false)]; tensor value_1_cast_fp16 = transpose(perm = value_1_perm_0, x = var_89_cast_fp16)[name = string("transpose_117")]; tensor attn_output_1_cast_fp16 = matmul(transpose_x = attn_output_1_transpose_x_0, transpose_y = attn_output_1_transpose_y_0, x = softmax_0_cast_fp16, y = value_1_cast_fp16)[name = string("attn_output_1_cast_fp16")]; tensor var_92_perm_0 = const()[name = string("op_92_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_94 = const()[name = string("op_94"), val = tensor([1, 77, -1])]; tensor var_92_cast_fp16 = transpose(perm = var_92_perm_0, x = attn_output_1_cast_fp16)[name = string("transpose_116")]; tensor var_95_cast_fp16 = reshape(shape = var_94, x = var_92_cast_fp16)[name = string("op_95_cast_fp16")]; tensor encoder_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30338688))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30781120))))[name = string("encoder_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_0_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_encoder_layers_0_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30787328)))]; tensor linear_3_cast_fp16 = linear(bias = encoder_encoder_layers_0_self_attn_out_proj_bias_to_fp16, weight = encoder_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized, x = var_95_cast_fp16)[name = string("linear_3_cast_fp16")]; tensor input_7_cast_fp16 = add(x = input_3_cast_fp16, y = linear_3_cast_fp16)[name = string("input_7_cast_fp16")]; tensor input_9_axes_0 = const()[name = string("input_9_axes_0"), val = tensor([-1])]; tensor encoder_encoder_layers_0_layer_norm2_weight_to_fp16 = const()[name = string("encoder_encoder_layers_0_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30788928)))]; tensor encoder_encoder_layers_0_layer_norm2_bias_to_fp16 = const()[name = string("encoder_encoder_layers_0_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30790528)))]; tensor input_9_cast_fp16 = layer_norm(axes = input_9_axes_0, beta = encoder_encoder_layers_0_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = encoder_encoder_layers_0_layer_norm2_weight_to_fp16, x = input_7_cast_fp16)[name = string("input_9_cast_fp16")]; tensor encoder_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30792128))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32561664))))[name = string("encoder_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_0_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_encoder_layers_0_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32586304)))]; tensor linear_4_cast_fp16 = linear(bias = encoder_encoder_layers_0_mlp_fc1_bias_to_fp16, weight = encoder_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized, x = input_9_cast_fp16)[name = string("linear_4_cast_fp16")]; fp16 var_110_to_fp16 = const()[name = string("op_110_to_fp16"), val = fp16(0x1.b3cp+0)]; tensor var_111_cast_fp16 = mul(x = linear_4_cast_fp16, y = var_110_to_fp16)[name = string("op_111_cast_fp16")]; tensor var_112_cast_fp16 = sigmoid(x = var_111_cast_fp16)[name = string("op_112_cast_fp16")]; tensor input_13_cast_fp16 = mul(x = linear_4_cast_fp16, y = var_112_cast_fp16)[name = string("input_13_cast_fp16")]; tensor encoder_encoder_layers_0_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32592512))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34362048))))[name = string("encoder_encoder_layers_0_mlp_fc2_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_0_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_encoder_layers_0_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34368256)))]; tensor linear_5_cast_fp16 = linear(bias = encoder_encoder_layers_0_mlp_fc2_bias_to_fp16, weight = encoder_encoder_layers_0_mlp_fc2_weight_to_fp16_palettized, x = input_13_cast_fp16)[name = string("linear_5_cast_fp16")]; tensor input_15_cast_fp16 = add(x = input_7_cast_fp16, y = linear_5_cast_fp16)[name = string("input_15_cast_fp16")]; tensor hidden_states_7_axes_0 = const()[name = string("hidden_states_7_axes_0"), val = tensor([-1])]; tensor encoder_encoder_layers_1_layer_norm1_weight_to_fp16 = const()[name = string("encoder_encoder_layers_1_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34369856)))]; tensor encoder_encoder_layers_1_layer_norm1_bias_to_fp16 = const()[name = string("encoder_encoder_layers_1_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34371456)))]; tensor hidden_states_7_cast_fp16 = layer_norm(axes = hidden_states_7_axes_0, beta = encoder_encoder_layers_1_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = encoder_encoder_layers_1_layer_norm1_weight_to_fp16, x = input_15_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; tensor encoder_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34373056))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34815488))))[name = string("encoder_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_encoder_layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34821696)))]; tensor linear_6_cast_fp16 = linear(bias = encoder_encoder_layers_1_self_attn_q_proj_bias_to_fp16, weight = encoder_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_7_cast_fp16)[name = string("linear_6_cast_fp16")]; tensor encoder_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34823296))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35265728))))[name = string("encoder_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_1_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_encoder_layers_1_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35271936)))]; tensor linear_7_cast_fp16 = linear(bias = encoder_encoder_layers_1_self_attn_k_proj_bias_to_fp16, weight = encoder_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_7_cast_fp16)[name = string("linear_7_cast_fp16")]; tensor encoder_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35273536))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35715968))))[name = string("encoder_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_encoder_layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35722176)))]; tensor linear_8_cast_fp16 = linear(bias = encoder_encoder_layers_1_self_attn_v_proj_bias_to_fp16, weight = encoder_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_7_cast_fp16)[name = string("linear_8_cast_fp16")]; tensor var_141 = const()[name = string("op_141"), val = tensor([1, 77, -1, 64])]; tensor var_142_cast_fp16 = reshape(shape = var_141, x = linear_6_cast_fp16)[name = string("op_142_cast_fp16")]; tensor var_144 = const()[name = string("op_144"), val = tensor([1, 77, -1, 64])]; tensor var_145_cast_fp16 = reshape(shape = var_144, x = linear_7_cast_fp16)[name = string("op_145_cast_fp16")]; tensor var_147 = const()[name = string("op_147"), val = tensor([1, 77, -1, 64])]; tensor var_148_cast_fp16 = reshape(shape = var_147, x = linear_8_cast_fp16)[name = string("op_148_cast_fp16")]; tensor value_3_perm_0 = const()[name = string("value_3_perm_0"), val = tensor([0, 2, -3, -1])]; tensor mul_3_cast_fp16 = mul(x = var_142_cast_fp16, y = var_9_to_fp16)[name = string("mul_3_cast_fp16")]; bool matmul_1_transpose_y_0 = const()[name = string("matmul_1_transpose_y_0"), val = bool(true)]; bool matmul_1_transpose_x_0 = const()[name = string("matmul_1_transpose_x_0"), val = bool(false)]; tensor transpose_50_perm_0 = const()[name = string("transpose_50_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_51_perm_0 = const()[name = string("transpose_51_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_51 = transpose(perm = transpose_51_perm_0, x = var_145_cast_fp16)[name = string("transpose_114")]; tensor transpose_50 = transpose(perm = transpose_50_perm_0, x = mul_3_cast_fp16)[name = string("transpose_115")]; tensor matmul_1_cast_fp16 = matmul(transpose_x = matmul_1_transpose_x_0, transpose_y = matmul_1_transpose_y_0, x = transpose_50, y = transpose_51)[name = string("matmul_1_cast_fp16")]; tensor add_1_cast_fp16 = add(x = matmul_1_cast_fp16, y = mul_0_to_fp16)[name = string("add_1_cast_fp16")]; int32 softmax_1_axis_0 = const()[name = string("softmax_1_axis_0"), val = int32(-1)]; tensor softmax_1_cast_fp16 = softmax(axis = softmax_1_axis_0, x = add_1_cast_fp16)[name = string("softmax_1_cast_fp16")]; bool attn_output_5_transpose_x_0 = const()[name = string("attn_output_5_transpose_x_0"), val = bool(false)]; bool attn_output_5_transpose_y_0 = const()[name = string("attn_output_5_transpose_y_0"), val = bool(false)]; tensor value_3_cast_fp16 = transpose(perm = value_3_perm_0, x = var_148_cast_fp16)[name = string("transpose_113")]; tensor attn_output_5_cast_fp16 = matmul(transpose_x = attn_output_5_transpose_x_0, transpose_y = attn_output_5_transpose_y_0, x = softmax_1_cast_fp16, y = value_3_cast_fp16)[name = string("attn_output_5_cast_fp16")]; tensor var_151_perm_0 = const()[name = string("op_151_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_153 = const()[name = string("op_153"), val = tensor([1, 77, -1])]; tensor var_151_cast_fp16 = transpose(perm = var_151_perm_0, x = attn_output_5_cast_fp16)[name = string("transpose_112")]; tensor var_154_cast_fp16 = reshape(shape = var_153, x = var_151_cast_fp16)[name = string("op_154_cast_fp16")]; tensor encoder_encoder_layers_1_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35723776))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36166208))))[name = string("encoder_encoder_layers_1_self_attn_out_proj_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_1_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_encoder_layers_1_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36172416)))]; tensor linear_9_cast_fp16 = linear(bias = encoder_encoder_layers_1_self_attn_out_proj_bias_to_fp16, weight = encoder_encoder_layers_1_self_attn_out_proj_weight_to_fp16_palettized, x = var_154_cast_fp16)[name = string("linear_9_cast_fp16")]; tensor input_19_cast_fp16 = add(x = input_15_cast_fp16, y = linear_9_cast_fp16)[name = string("input_19_cast_fp16")]; tensor input_21_axes_0 = const()[name = string("input_21_axes_0"), val = tensor([-1])]; tensor encoder_encoder_layers_1_layer_norm2_weight_to_fp16 = const()[name = string("encoder_encoder_layers_1_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36174016)))]; tensor encoder_encoder_layers_1_layer_norm2_bias_to_fp16 = const()[name = string("encoder_encoder_layers_1_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36175616)))]; tensor input_21_cast_fp16 = layer_norm(axes = input_21_axes_0, beta = encoder_encoder_layers_1_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = encoder_encoder_layers_1_layer_norm2_weight_to_fp16, x = input_19_cast_fp16)[name = string("input_21_cast_fp16")]; tensor encoder_encoder_layers_1_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36177216))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37946752))))[name = string("encoder_encoder_layers_1_mlp_fc1_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_1_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_encoder_layers_1_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37971392)))]; tensor linear_10_cast_fp16 = linear(bias = encoder_encoder_layers_1_mlp_fc1_bias_to_fp16, weight = encoder_encoder_layers_1_mlp_fc1_weight_to_fp16_palettized, x = input_21_cast_fp16)[name = string("linear_10_cast_fp16")]; fp16 var_169_to_fp16 = const()[name = string("op_169_to_fp16"), val = fp16(0x1.b3cp+0)]; tensor var_170_cast_fp16 = mul(x = linear_10_cast_fp16, y = var_169_to_fp16)[name = string("op_170_cast_fp16")]; tensor var_171_cast_fp16 = sigmoid(x = var_170_cast_fp16)[name = string("op_171_cast_fp16")]; tensor input_25_cast_fp16 = mul(x = linear_10_cast_fp16, y = var_171_cast_fp16)[name = string("input_25_cast_fp16")]; tensor encoder_encoder_layers_1_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37977600))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39747136))))[name = string("encoder_encoder_layers_1_mlp_fc2_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_1_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_encoder_layers_1_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39753344)))]; tensor linear_11_cast_fp16 = linear(bias = encoder_encoder_layers_1_mlp_fc2_bias_to_fp16, weight = encoder_encoder_layers_1_mlp_fc2_weight_to_fp16_palettized, x = input_25_cast_fp16)[name = string("linear_11_cast_fp16")]; tensor input_27_cast_fp16 = add(x = input_19_cast_fp16, y = linear_11_cast_fp16)[name = string("input_27_cast_fp16")]; tensor hidden_states_13_axes_0 = const()[name = string("hidden_states_13_axes_0"), val = tensor([-1])]; tensor encoder_encoder_layers_2_layer_norm1_weight_to_fp16 = const()[name = string("encoder_encoder_layers_2_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39754944)))]; tensor encoder_encoder_layers_2_layer_norm1_bias_to_fp16 = const()[name = string("encoder_encoder_layers_2_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39756544)))]; tensor hidden_states_13_cast_fp16 = layer_norm(axes = hidden_states_13_axes_0, beta = encoder_encoder_layers_2_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = encoder_encoder_layers_2_layer_norm1_weight_to_fp16, x = input_27_cast_fp16)[name = string("hidden_states_13_cast_fp16")]; tensor encoder_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39758144))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40200576))))[name = string("encoder_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_encoder_layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40206784)))]; tensor linear_12_cast_fp16 = linear(bias = encoder_encoder_layers_2_self_attn_q_proj_bias_to_fp16, weight = encoder_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_13_cast_fp16)[name = string("linear_12_cast_fp16")]; tensor encoder_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40208384))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40650816))))[name = string("encoder_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_2_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_encoder_layers_2_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40657024)))]; tensor linear_13_cast_fp16 = linear(bias = encoder_encoder_layers_2_self_attn_k_proj_bias_to_fp16, weight = encoder_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_13_cast_fp16)[name = string("linear_13_cast_fp16")]; tensor encoder_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40658624))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41101056))))[name = string("encoder_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_encoder_layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41107264)))]; tensor linear_14_cast_fp16 = linear(bias = encoder_encoder_layers_2_self_attn_v_proj_bias_to_fp16, weight = encoder_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_13_cast_fp16)[name = string("linear_14_cast_fp16")]; tensor var_200 = const()[name = string("op_200"), val = tensor([1, 77, -1, 64])]; tensor var_201_cast_fp16 = reshape(shape = var_200, x = linear_12_cast_fp16)[name = string("op_201_cast_fp16")]; tensor var_203 = const()[name = string("op_203"), val = tensor([1, 77, -1, 64])]; tensor var_204_cast_fp16 = reshape(shape = var_203, x = linear_13_cast_fp16)[name = string("op_204_cast_fp16")]; tensor var_206 = const()[name = string("op_206"), val = tensor([1, 77, -1, 64])]; tensor var_207_cast_fp16 = reshape(shape = var_206, x = linear_14_cast_fp16)[name = string("op_207_cast_fp16")]; tensor value_5_perm_0 = const()[name = string("value_5_perm_0"), val = tensor([0, 2, -3, -1])]; tensor mul_5_cast_fp16 = mul(x = var_201_cast_fp16, y = var_9_to_fp16)[name = string("mul_5_cast_fp16")]; bool matmul_2_transpose_y_0 = const()[name = string("matmul_2_transpose_y_0"), val = bool(true)]; bool matmul_2_transpose_x_0 = const()[name = string("matmul_2_transpose_x_0"), val = bool(false)]; tensor transpose_52_perm_0 = const()[name = string("transpose_52_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_53_perm_0 = const()[name = string("transpose_53_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_53 = transpose(perm = transpose_53_perm_0, x = var_204_cast_fp16)[name = string("transpose_110")]; tensor transpose_52 = transpose(perm = transpose_52_perm_0, x = mul_5_cast_fp16)[name = string("transpose_111")]; tensor matmul_2_cast_fp16 = matmul(transpose_x = matmul_2_transpose_x_0, transpose_y = matmul_2_transpose_y_0, x = transpose_52, y = transpose_53)[name = string("matmul_2_cast_fp16")]; tensor add_2_cast_fp16 = add(x = matmul_2_cast_fp16, y = mul_0_to_fp16)[name = string("add_2_cast_fp16")]; int32 softmax_2_axis_0 = const()[name = string("softmax_2_axis_0"), val = int32(-1)]; tensor softmax_2_cast_fp16 = softmax(axis = softmax_2_axis_0, x = add_2_cast_fp16)[name = string("softmax_2_cast_fp16")]; bool attn_output_9_transpose_x_0 = const()[name = string("attn_output_9_transpose_x_0"), val = bool(false)]; bool attn_output_9_transpose_y_0 = const()[name = string("attn_output_9_transpose_y_0"), val = bool(false)]; tensor value_5_cast_fp16 = transpose(perm = value_5_perm_0, x = var_207_cast_fp16)[name = string("transpose_109")]; tensor attn_output_9_cast_fp16 = matmul(transpose_x = attn_output_9_transpose_x_0, transpose_y = attn_output_9_transpose_y_0, x = softmax_2_cast_fp16, y = value_5_cast_fp16)[name = string("attn_output_9_cast_fp16")]; tensor var_210_perm_0 = const()[name = string("op_210_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_212 = const()[name = string("op_212"), val = tensor([1, 77, -1])]; tensor var_210_cast_fp16 = transpose(perm = var_210_perm_0, x = attn_output_9_cast_fp16)[name = string("transpose_108")]; tensor var_213_cast_fp16 = reshape(shape = var_212, x = var_210_cast_fp16)[name = string("op_213_cast_fp16")]; tensor encoder_encoder_layers_2_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41108864))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41551296))))[name = string("encoder_encoder_layers_2_self_attn_out_proj_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_2_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_encoder_layers_2_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41557504)))]; tensor linear_15_cast_fp16 = linear(bias = encoder_encoder_layers_2_self_attn_out_proj_bias_to_fp16, weight = encoder_encoder_layers_2_self_attn_out_proj_weight_to_fp16_palettized, x = var_213_cast_fp16)[name = string("linear_15_cast_fp16")]; tensor input_31_cast_fp16 = add(x = input_27_cast_fp16, y = linear_15_cast_fp16)[name = string("input_31_cast_fp16")]; tensor input_33_axes_0 = const()[name = string("input_33_axes_0"), val = tensor([-1])]; tensor encoder_encoder_layers_2_layer_norm2_weight_to_fp16 = const()[name = string("encoder_encoder_layers_2_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41559104)))]; tensor encoder_encoder_layers_2_layer_norm2_bias_to_fp16 = const()[name = string("encoder_encoder_layers_2_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41560704)))]; tensor input_33_cast_fp16 = layer_norm(axes = input_33_axes_0, beta = encoder_encoder_layers_2_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = encoder_encoder_layers_2_layer_norm2_weight_to_fp16, x = input_31_cast_fp16)[name = string("input_33_cast_fp16")]; tensor encoder_encoder_layers_2_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41562304))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43331840))))[name = string("encoder_encoder_layers_2_mlp_fc1_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_2_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_encoder_layers_2_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43356480)))]; tensor linear_16_cast_fp16 = linear(bias = encoder_encoder_layers_2_mlp_fc1_bias_to_fp16, weight = encoder_encoder_layers_2_mlp_fc1_weight_to_fp16_palettized, x = input_33_cast_fp16)[name = string("linear_16_cast_fp16")]; fp16 var_228_to_fp16 = const()[name = string("op_228_to_fp16"), val = fp16(0x1.b3cp+0)]; tensor var_229_cast_fp16 = mul(x = linear_16_cast_fp16, y = var_228_to_fp16)[name = string("op_229_cast_fp16")]; tensor var_230_cast_fp16 = sigmoid(x = var_229_cast_fp16)[name = string("op_230_cast_fp16")]; tensor input_37_cast_fp16 = mul(x = linear_16_cast_fp16, y = var_230_cast_fp16)[name = string("input_37_cast_fp16")]; tensor encoder_encoder_layers_2_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43362688))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45132224))))[name = string("encoder_encoder_layers_2_mlp_fc2_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_2_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_encoder_layers_2_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45138432)))]; tensor linear_17_cast_fp16 = linear(bias = encoder_encoder_layers_2_mlp_fc2_bias_to_fp16, weight = encoder_encoder_layers_2_mlp_fc2_weight_to_fp16_palettized, x = input_37_cast_fp16)[name = string("linear_17_cast_fp16")]; tensor input_39_cast_fp16 = add(x = input_31_cast_fp16, y = linear_17_cast_fp16)[name = string("input_39_cast_fp16")]; tensor hidden_states_19_axes_0 = const()[name = string("hidden_states_19_axes_0"), val = tensor([-1])]; tensor encoder_encoder_layers_3_layer_norm1_weight_to_fp16 = const()[name = string("encoder_encoder_layers_3_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45140032)))]; tensor encoder_encoder_layers_3_layer_norm1_bias_to_fp16 = const()[name = string("encoder_encoder_layers_3_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45141632)))]; tensor hidden_states_19_cast_fp16 = layer_norm(axes = hidden_states_19_axes_0, beta = encoder_encoder_layers_3_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = encoder_encoder_layers_3_layer_norm1_weight_to_fp16, x = input_39_cast_fp16)[name = string("hidden_states_19_cast_fp16")]; tensor encoder_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45143232))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45585664))))[name = string("encoder_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_encoder_layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45591872)))]; tensor linear_18_cast_fp16 = linear(bias = encoder_encoder_layers_3_self_attn_q_proj_bias_to_fp16, weight = encoder_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_19_cast_fp16)[name = string("linear_18_cast_fp16")]; tensor encoder_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45593472))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46035904))))[name = string("encoder_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_3_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_encoder_layers_3_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46042112)))]; tensor linear_19_cast_fp16 = linear(bias = encoder_encoder_layers_3_self_attn_k_proj_bias_to_fp16, weight = encoder_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_19_cast_fp16)[name = string("linear_19_cast_fp16")]; tensor encoder_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46043712))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46486144))))[name = string("encoder_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_encoder_layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46492352)))]; tensor linear_20_cast_fp16 = linear(bias = encoder_encoder_layers_3_self_attn_v_proj_bias_to_fp16, weight = encoder_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_19_cast_fp16)[name = string("linear_20_cast_fp16")]; tensor var_259 = const()[name = string("op_259"), val = tensor([1, 77, -1, 64])]; tensor var_260_cast_fp16 = reshape(shape = var_259, x = linear_18_cast_fp16)[name = string("op_260_cast_fp16")]; tensor var_262 = const()[name = string("op_262"), val = tensor([1, 77, -1, 64])]; tensor var_263_cast_fp16 = reshape(shape = var_262, x = linear_19_cast_fp16)[name = string("op_263_cast_fp16")]; tensor var_265 = const()[name = string("op_265"), val = tensor([1, 77, -1, 64])]; tensor var_266_cast_fp16 = reshape(shape = var_265, x = linear_20_cast_fp16)[name = string("op_266_cast_fp16")]; tensor value_7_perm_0 = const()[name = string("value_7_perm_0"), val = tensor([0, 2, -3, -1])]; tensor mul_7_cast_fp16 = mul(x = var_260_cast_fp16, y = var_9_to_fp16)[name = string("mul_7_cast_fp16")]; bool matmul_3_transpose_y_0 = const()[name = string("matmul_3_transpose_y_0"), val = bool(true)]; bool matmul_3_transpose_x_0 = const()[name = string("matmul_3_transpose_x_0"), val = bool(false)]; tensor transpose_54_perm_0 = const()[name = string("transpose_54_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_55_perm_0 = const()[name = string("transpose_55_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_55 = transpose(perm = transpose_55_perm_0, x = var_263_cast_fp16)[name = string("transpose_106")]; tensor transpose_54 = transpose(perm = transpose_54_perm_0, x = mul_7_cast_fp16)[name = string("transpose_107")]; tensor matmul_3_cast_fp16 = matmul(transpose_x = matmul_3_transpose_x_0, transpose_y = matmul_3_transpose_y_0, x = transpose_54, y = transpose_55)[name = string("matmul_3_cast_fp16")]; tensor add_3_cast_fp16 = add(x = matmul_3_cast_fp16, y = mul_0_to_fp16)[name = string("add_3_cast_fp16")]; int32 softmax_3_axis_0 = const()[name = string("softmax_3_axis_0"), val = int32(-1)]; tensor softmax_3_cast_fp16 = softmax(axis = softmax_3_axis_0, x = add_3_cast_fp16)[name = string("softmax_3_cast_fp16")]; bool attn_output_13_transpose_x_0 = const()[name = string("attn_output_13_transpose_x_0"), val = bool(false)]; bool attn_output_13_transpose_y_0 = const()[name = string("attn_output_13_transpose_y_0"), val = bool(false)]; tensor value_7_cast_fp16 = transpose(perm = value_7_perm_0, x = var_266_cast_fp16)[name = string("transpose_105")]; tensor attn_output_13_cast_fp16 = matmul(transpose_x = attn_output_13_transpose_x_0, transpose_y = attn_output_13_transpose_y_0, x = softmax_3_cast_fp16, y = value_7_cast_fp16)[name = string("attn_output_13_cast_fp16")]; tensor var_269_perm_0 = const()[name = string("op_269_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_271 = const()[name = string("op_271"), val = tensor([1, 77, -1])]; tensor var_269_cast_fp16 = transpose(perm = var_269_perm_0, x = attn_output_13_cast_fp16)[name = string("transpose_104")]; tensor var_272_cast_fp16 = reshape(shape = var_271, x = var_269_cast_fp16)[name = string("op_272_cast_fp16")]; tensor encoder_encoder_layers_3_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46493952))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46936384))))[name = string("encoder_encoder_layers_3_self_attn_out_proj_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_3_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_encoder_layers_3_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46942592)))]; tensor linear_21_cast_fp16 = linear(bias = encoder_encoder_layers_3_self_attn_out_proj_bias_to_fp16, weight = encoder_encoder_layers_3_self_attn_out_proj_weight_to_fp16_palettized, x = var_272_cast_fp16)[name = string("linear_21_cast_fp16")]; tensor input_43_cast_fp16 = add(x = input_39_cast_fp16, y = linear_21_cast_fp16)[name = string("input_43_cast_fp16")]; tensor input_45_axes_0 = const()[name = string("input_45_axes_0"), val = tensor([-1])]; tensor encoder_encoder_layers_3_layer_norm2_weight_to_fp16 = const()[name = string("encoder_encoder_layers_3_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46944192)))]; tensor encoder_encoder_layers_3_layer_norm2_bias_to_fp16 = const()[name = string("encoder_encoder_layers_3_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46945792)))]; tensor input_45_cast_fp16 = layer_norm(axes = input_45_axes_0, beta = encoder_encoder_layers_3_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = encoder_encoder_layers_3_layer_norm2_weight_to_fp16, x = input_43_cast_fp16)[name = string("input_45_cast_fp16")]; tensor encoder_encoder_layers_3_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46947392))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48716928))))[name = string("encoder_encoder_layers_3_mlp_fc1_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_3_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_encoder_layers_3_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48741568)))]; tensor linear_22_cast_fp16 = linear(bias = encoder_encoder_layers_3_mlp_fc1_bias_to_fp16, weight = encoder_encoder_layers_3_mlp_fc1_weight_to_fp16_palettized, x = input_45_cast_fp16)[name = string("linear_22_cast_fp16")]; fp16 var_287_to_fp16 = const()[name = string("op_287_to_fp16"), val = fp16(0x1.b3cp+0)]; tensor var_288_cast_fp16 = mul(x = linear_22_cast_fp16, y = var_287_to_fp16)[name = string("op_288_cast_fp16")]; tensor var_289_cast_fp16 = sigmoid(x = var_288_cast_fp16)[name = string("op_289_cast_fp16")]; tensor input_49_cast_fp16 = mul(x = linear_22_cast_fp16, y = var_289_cast_fp16)[name = string("input_49_cast_fp16")]; tensor encoder_encoder_layers_3_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48747776))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50517312))))[name = string("encoder_encoder_layers_3_mlp_fc2_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_3_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_encoder_layers_3_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50523520)))]; tensor linear_23_cast_fp16 = linear(bias = encoder_encoder_layers_3_mlp_fc2_bias_to_fp16, weight = encoder_encoder_layers_3_mlp_fc2_weight_to_fp16_palettized, x = input_49_cast_fp16)[name = string("linear_23_cast_fp16")]; tensor input_51_cast_fp16 = add(x = input_43_cast_fp16, y = linear_23_cast_fp16)[name = string("input_51_cast_fp16")]; tensor hidden_states_25_axes_0 = const()[name = string("hidden_states_25_axes_0"), val = tensor([-1])]; tensor encoder_encoder_layers_4_layer_norm1_weight_to_fp16 = const()[name = string("encoder_encoder_layers_4_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50525120)))]; tensor encoder_encoder_layers_4_layer_norm1_bias_to_fp16 = const()[name = string("encoder_encoder_layers_4_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50526720)))]; tensor hidden_states_25_cast_fp16 = layer_norm(axes = hidden_states_25_axes_0, beta = encoder_encoder_layers_4_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = encoder_encoder_layers_4_layer_norm1_weight_to_fp16, x = input_51_cast_fp16)[name = string("hidden_states_25_cast_fp16")]; tensor encoder_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50528320))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50970752))))[name = string("encoder_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_encoder_layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50976960)))]; tensor linear_24_cast_fp16 = linear(bias = encoder_encoder_layers_4_self_attn_q_proj_bias_to_fp16, weight = encoder_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_25_cast_fp16)[name = string("linear_24_cast_fp16")]; tensor encoder_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50978560))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51420992))))[name = string("encoder_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_4_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_encoder_layers_4_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51427200)))]; tensor linear_25_cast_fp16 = linear(bias = encoder_encoder_layers_4_self_attn_k_proj_bias_to_fp16, weight = encoder_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_25_cast_fp16)[name = string("linear_25_cast_fp16")]; tensor encoder_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51428800))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51871232))))[name = string("encoder_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized")]; tensor encoder_encoder_layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_encoder_layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51877440)))]; tensor linear_26_cast_fp16 = linear(bias = encoder_encoder_layers_4_self_attn_v_proj_bias_to_fp16, weight = encoder_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_25_cast_fp16)[name = string("linear_26_cast_fp16")]; tensor var_318 = const()[name = string("op_318"), val = tensor([1, 77, -1, 64])]; tensor var_319_cast_fp16 = reshape(shape = var_318, x = linear_24_cast_fp16)[name = string("op_319_cast_fp16")]; tensor var_321 = const()[name = string("op_321"), val = tensor([1, 77, -1, 64])]; tensor var_322_cast_fp16 = reshape(shape = var_321, x = linear_25_cast_fp16)[name = string("op_322_cast_fp16")]; tensor var_324 = const()[name = string("op_324"), val = tensor([1, 77, -1, 64])]; tensor var_325_cast_fp16 = reshape(shape = var_324, x = linear_26_cast_fp16)[name = string("op_325_cast_fp16")]; tensor value_9_perm_0 = const()[name = string("value_9_perm_0"), val = tensor([0, 2, -3, -1])]; tensor mul_9_cast_fp16 = mul(x = var_319_cast_fp16, y = var_9_to_fp16)[name = string("mul_9_cast_fp16")]; bool matmul_4_transpose_y_0 = const()[name = string("matmul_4_transpose_y_0"), val = bool(true)]; bool matmul_4_transpose_x_0 = const()[name = string("matmul_4_transpose_x_0"), val = bool(false)]; tensor transpose_56_perm_0 = const()[name = string("transpose_56_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_57_perm_0 = const()[name = string("transpose_57_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_57 = transpose(perm = transpose_57_perm_0, x = var_322_cast_fp16)[name = string("transpose_102")]; tensor transpose_56 = transpose(perm = transpose_56_perm_0, x = mul_9_cast_fp16)[name = string("transpose_103")]; tensor matmul_4_cast_fp16 = matmul(transpose_x = matmul_4_transpose_x_0, transpose_y = matmul_4_transpose_y_0, x = transpose_56, y = transpose_57)[name = string("matmul_4_cast_fp16")]; tensor add_4_cast_fp16 = add(x = matmul_4_cast_fp16, y = mul_0_to_fp16)[name = string("add_4_cast_fp16")]; int32 softmax_4_axis_0 = const()[name = string("softmax_4_axis_0"), val = int32(-1)]; tensor softmax_4_cast_fp16 = softmax(axis = softmax_4_axis_0, x = add_4_cast_fp16)[name = string("softmax_4_cast_fp16")]; bool attn_output_17_transpose_x_0 = const()[name = string("attn_output_17_transpose_x_0"), val = bool(false)]; bool attn_output_17_transpose_y_0 = const()[name = string("attn_output_17_transpose_y_0"), val = bool(false)]; tensor