program(1.3) [buildInfo = dict({{"coremlc-component-MIL", "3510.2.1"}, {"coremlc-version", "3500.32.1"}, {"coremltools-component-torch", "2.7.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})] { func main(tensor all_tokens) [FlexibleShapeInformation = tuple>>, tuple, ?>>>>((("DefaultShapes", {{"all_tokens", [1, 500]}}), ("RangeDims", {{"all_tokens", [[1, 1], [1, 2048]]}})))] { tensor input_embedding_weight = const()[name = string("input_embedding_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; tensor encoder_embed_out_0_bias = const()[name = string("encoder_embed_out_0_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13437056)))]; tensor encoder_embed_out_0_weight = const()[name = string("encoder_embed_out_0_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13439168)))]; tensor encoder_embed_out_1_bias = const()[name = string("encoder_embed_out_1_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14487808)))]; tensor encoder_embed_out_1_weight = const()[name = string("encoder_embed_out_1_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14489920)))]; tensor encoder_pre_lookahead_layer_conv1_bias = const()[name = string("encoder_pre_lookahead_layer_conv1_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14492032)))]; tensor encoder_pre_lookahead_layer_conv1_weight = const()[name = string("encoder_pre_lookahead_layer_conv1_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14494144)))]; tensor encoder_pre_lookahead_layer_conv2_bias = const()[name = string("encoder_pre_lookahead_layer_conv2_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18688512)))]; tensor encoder_pre_lookahead_layer_conv2_weight = const()[name = string("encoder_pre_lookahead_layer_conv2_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18690624)))]; tensor encoder_encoders_0_norm_mha_bias = const()[name = string("encoder_encoders_0_norm_mha_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21836416)))]; tensor encoder_encoders_0_norm_mha_weight = const()[name = string("encoder_encoders_0_norm_mha_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21838528)))]; tensor encoder_encoders_0_self_attn_linear_q_bias = const()[name = string("encoder_encoders_0_self_attn_linear_q_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21840640)))]; tensor encoder_encoders_0_self_attn_linear_q_weight = const()[name = string("encoder_encoders_0_self_attn_linear_q_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21842752)))]; tensor encoder_encoders_0_self_attn_linear_k_bias = const()[name = string("encoder_encoders_0_self_attn_linear_k_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22891392)))]; tensor encoder_encoders_0_self_attn_linear_k_weight = const()[name = string("encoder_encoders_0_self_attn_linear_k_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22893504)))]; tensor encoder_encoders_0_self_attn_linear_v_bias = const()[name = string("encoder_encoders_0_self_attn_linear_v_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23942144)))]; tensor encoder_encoders_0_self_attn_linear_v_weight = const()[name = string("encoder_encoders_0_self_attn_linear_v_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23944256)))]; tensor encoder_encoders_0_self_attn_linear_pos_weight = const()[name = string("encoder_encoders_0_self_attn_linear_pos_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24992896)))]; tensor encoder_encoders_0_self_attn_linear_out_bias = const()[name = string("encoder_encoders_0_self_attn_linear_out_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26041536)))]; tensor encoder_encoders_0_self_attn_linear_out_weight = const()[name = string("encoder_encoders_0_self_attn_linear_out_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26043648)))]; tensor encoder_encoders_0_norm_ff_bias = const()[name = string("encoder_encoders_0_norm_ff_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27092288)))]; tensor encoder_encoders_0_norm_ff_weight = const()[name = string("encoder_encoders_0_norm_ff_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27094400)))]; tensor encoder_encoders_0_feed_forward_w_1_bias = const()[name = string("encoder_encoders_0_feed_forward_w_1_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27096512)))]; tensor encoder_encoders_0_feed_forward_w_1_weight = const()[name = string("encoder_encoders_0_feed_forward_w_1_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27104768)))]; tensor encoder_encoders_0_feed_forward_w_2_bias = const()[name = string("encoder_encoders_0_feed_forward_w_2_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31299136)))]; tensor encoder_encoders_0_feed_forward_w_2_weight = const()[name = string("encoder_encoders_0_feed_forward_w_2_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31301248)))]; tensor encoder_encoders_1_norm_mha_bias = const()[name = string("encoder_encoders_1_norm_mha_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35495616)))]; tensor encoder_encoders_1_norm_mha_weight = const()[name = string("encoder_encoders_1_norm_mha_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35497728)))]; tensor encoder_encoders_1_self_attn_linear_q_bias = const()[name = string("encoder_encoders_1_self_attn_linear_q_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35499840)))]; tensor encoder_encoders_1_self_attn_linear_q_weight = const()[name = string("encoder_encoders_1_self_attn_linear_q_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35501952)))]; tensor encoder_encoders_1_self_attn_linear_k_bias = const()[name = string("encoder_encoders_1_self_attn_linear_k_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36550592)))]; tensor encoder_encoders_1_self_attn_linear_k_weight = const()[name = string("encoder_encoders_1_self_attn_linear_k_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36552704)))]; tensor encoder_encoders_1_self_attn_linear_v_bias = const()[name = string("encoder_encoders_1_self_attn_linear_v_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37601344)))]; tensor encoder_encoders_1_self_attn_linear_v_weight = const()[name = string("encoder_encoders_1_self_attn_linear_v_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37603456)))]; tensor encoder_encoders_1_self_attn_linear_pos_weight = const()[name = string("encoder_encoders_1_self_attn_linear_pos_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38652096)))]; tensor encoder_encoders_1_self_attn_linear_out_bias = const()[name = string("encoder_encoders_1_self_attn_linear_out_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39700736)))]; tensor encoder_encoders_1_self_attn_linear_out_weight = const()[name = string("encoder_encoders_1_self_attn_linear_out_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39702848)))]; tensor encoder_encoders_1_norm_ff_bias = const()[name = string("encoder_encoders_1_norm_ff_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40751488)))]; tensor encoder_encoders_1_norm_ff_weight = const()[name = string("encoder_encoders_1_norm_ff_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40753600)))]; tensor encoder_encoders_1_feed_forward_w_1_bias = const()[name = string("encoder_encoders_1_feed_forward_w_1_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40755712)))]; tensor encoder_encoders_1_feed_forward_w_1_weight = const()[name = string("encoder_encoders_1_feed_forward_w_1_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40763968)))]; tensor encoder_encoders_1_feed_forward_w_2_bias = const()[name = string("encoder_encoders_1_feed_forward_w_2_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44958336)))]; tensor encoder_encoders_1_feed_forward_w_2_weight = const()[name = string("encoder_encoders_1_feed_forward_w_2_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44960448)))]; tensor encoder_encoders_2_norm_mha_bias = const()[name = string("encoder_encoders_2_norm_mha_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49154816)))]; tensor encoder_encoders_2_norm_mha_weight = const()[name = string("encoder_encoders_2_norm_mha_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49156928)))]; tensor encoder_encoders_2_self_attn_linear_q_bias = const()[name = string("encoder_encoders_2_self_attn_linear_q_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49159040)))]; tensor encoder_encoders_2_self_attn_linear_q_weight = const()[name = string("encoder_encoders_2_self_attn_linear_q_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49161152)))]; tensor encoder_encoders_2_self_attn_linear_k_bias = const()[name = string("encoder_encoders_2_self_attn_linear_k_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50209792)))]; tensor encoder_encoders_2_self_attn_linear_k_weight = const()[name = string("encoder_encoders_2_self_attn_linear_k_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50211904)))]; tensor encoder_encoders_2_self_attn_linear_v_bias = const()[name = string("encoder_encoders_2_self_attn_linear_v_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51260544)))]; tensor encoder_encoders_2_self_attn_linear_v_weight = const()[name = string("encoder_encoders_2_self_attn_linear_v_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51262656)))]; tensor encoder_encoders_2_self_attn_linear_pos_weight = const()[name = string("encoder_encoders_2_self_attn_linear_pos_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52311296)))]; tensor encoder_encoders_2_self_attn_linear_out_bias = const()[name = string("encoder_encoders_2_self_attn_linear_out_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53359936)))]; tensor encoder_encoders_2_self_attn_linear_out_weight = const()[name = string("encoder_encoders_2_self_attn_linear_out_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53362048)))]; tensor encoder_encoders_2_norm_ff_bias = const()[name = string("encoder_encoders_2_norm_ff_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54410688)))]; tensor encoder_encoders_2_norm_ff_weight = const()[name = string("encoder_encoders_2_norm_ff_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54412800)))]; tensor encoder_encoders_2_feed_forward_w_1_bias = const()[name = string("encoder_encoders_2_feed_forward_w_1_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54414912)))]; tensor encoder_encoders_2_feed_forward_w_1_weight = const()[name = string("encoder_encoders_2_feed_forward_w_1_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54423168)))]; tensor encoder_encoders_2_feed_forward_w_2_bias = const()[name = string("encoder_encoders_2_feed_forward_w_2_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58617536)))]; tensor encoder_encoders_2_feed_forward_w_2_weight = const()[name = string("encoder_encoders_2_feed_forward_w_2_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58619648)))]; tensor encoder_encoders_3_norm_mha_bias = const()[name = string("encoder_encoders_3_norm_mha_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62814016)))]; tensor encoder_encoders_3_norm_mha_weight = const()[name = string("encoder_encoders_3_norm_mha_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62816128)))]; tensor encoder_encoders_3_self_attn_linear_q_bias = const()[name = string("encoder_encoders_3_self_attn_linear_q_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62818240)))]; tensor encoder_encoders_3_self_attn_linear_q_weight = const()[name = string("encoder_encoders_3_self_attn_linear_q_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62820352)))]; tensor encoder_encoders_3_self_attn_linear_k_bias = const()[name = string("encoder_encoders_3_self_attn_linear_k_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63868992)))]; tensor encoder_encoders_3_self_attn_linear_k_weight = const()[name = string("encoder_encoders_3_self_attn_linear_k_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63871104)))]; tensor encoder_encoders_3_self_attn_linear_v_bias = const()[name = string("encoder_encoders_3_self_attn_linear_v_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64919744)))]; tensor encoder_encoders_3_self_attn_linear_v_weight = const()[name = string("encoder_encoders_3_self_attn_linear_v_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64921856)))]; tensor encoder_encoders_3_self_attn_linear_pos_weight = const()[name = string("encoder_encoders_3_self_attn_linear_pos_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65970496)))]; tensor encoder_encoders_3_self_attn_linear_out_bias = const()[name = string("encoder_encoders_3_self_attn_linear_out_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67019136)))]; tensor encoder_encoders_3_self_attn_linear_out_weight = const()[name = string("encoder_encoders_3_self_attn_linear_out_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67021248)))]; tensor encoder_encoders_3_norm_ff_bias = const()[name = string("encoder_encoders_3_norm_ff_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68069888)))]; tensor encoder_encoders_3_norm_ff_weight = const()[name = string("encoder_encoders_3_norm_ff_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68072000)))]; tensor encoder_encoders_3_feed_forward_w_1_bias = const()[name = string("encoder_encoders_3_feed_forward_w_1_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68074112)))]; tensor encoder_encoders_3_feed_forward_w_1_weight = const()[name = string("encoder_encoders_3_feed_forward_w_1_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68082368)))]; tensor encoder_encoders_3_feed_forward_w_2_bias = const()[name = string("encoder_encoders_3_feed_forward_w_2_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72276736)))]; tensor encoder_encoders_3_feed_forward_w_2_weight = const()[name = string("encoder_encoders_3_feed_forward_w_2_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72278848)))]; tensor encoder_encoders_4_norm_mha_bias = const()[name = string("encoder_encoders_4_norm_mha_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76473216)))]; tensor encoder_encoders_4_norm_mha_weight = const()[name = string("encoder_encoders_4_norm_mha_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76475328)))]; tensor encoder_encoders_4_self_attn_linear_q_bias = const()[name = string("encoder_encoders_4_self_attn_linear_q_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76477440)))]; tensor encoder_encoders_4_self_attn_linear_q_weight = const()[name = string("encoder_encoders_4_self_attn_linear_q_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76479552)))]; tensor encoder_encoders_4_self_attn_linear_k_bias = const()[name = string("encoder_encoders_4_self_attn_linear_k_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77528192)))]; tensor encoder_encoders_4_self_attn_linear_k_weight = const()[name = string("encoder_encoders_4_self_attn_linear_k_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77530304)))]; tensor encoder_encoders_4_self_attn_linear_v_bias = const()[name = string("encoder_encoders_4_self_attn_linear_v_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78578944)))]; tensor encoder_encoders_4_self_attn_linear_v_weight = const()[name = string("encoder_encoders_4_self_attn_linear_v_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78581056)))]; tensor encoder_encoders_4_self_attn_linear_pos_weight = const()[name = string("encoder_encoders_4_self_attn_linear_pos_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79629696)))]; tensor encoder_encoders_4_self_attn_linear_out_bias = const()[name = string("encoder_encoders_4_self_attn_linear_out_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80678336)))]; tensor encoder_encoders_4_self_attn_linear_out_weight = const()[name = string("encoder_encoders_4_self_attn_linear_out_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80680448)))]; tensor encoder_encoders_4_norm_ff_bias = const()[name = string("encoder_encoders_4_norm_ff_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81729088)))]; tensor encoder_encoders_4_norm_ff_weight = const()[name = string("encoder_encoders_4_norm_ff_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81731200)))]; tensor encoder_encoders_4_feed_forward_w_1_bias = const()[name = string("encoder_encoders_4_feed_forward_w_1_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81733312)))]; tensor encoder_encoders_4_feed_forward_w_1_weight = const()[name = string("encoder_encoders_4_feed_forward_w_1_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81741568)))]; tensor encoder_encoders_4_feed_forward_w_2_bias = const()[name = string("encoder_encoders_4_feed_forward_w_2_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85935936)))]; tensor encoder_encoders_4_feed_forward_w_2_weight = const()[name = string("encoder_encoders_4_feed_forward_w_2_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85938048)))]; tensor encoder_encoders_5_norm_mha_bias = const()[name = string("encoder_encoders_5_norm_mha_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90132416)))]; tensor encoder_encoders_5_norm_mha_weight = const()[name = string("encoder_encoders_5_norm_mha_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90134528)))]; tensor encoder_encoders_5_self_attn_linear_q_bias = const()[name = string("encoder_encoders_5_self_attn_linear_q_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90136640)))]; tensor encoder_encoders_5_self_attn_linear_q_weight = const()[name = string("encoder_encoders_5_self_attn_linear_q_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90138752)))]; tensor encoder_encoders_5_self_attn_linear_k_bias = const()[name = string("encoder_encoders_5_self_attn_linear_k_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91187392)))]; tensor encoder_encoders_5_self_attn_linear_k_weight = const()[name = string("encoder_encoders_5_self_attn_linear_k_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91189504)))]; tensor encoder_encoders_5_self_attn_linear_v_bias = const()[name = string("encoder_encoders_5_self_attn_linear_v_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92238144)))]; tensor encoder_encoders_5_self_attn_linear_v_weight = const()[name = string("encoder_encoders_5_self_attn_linear_v_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92240256)))]; tensor encoder_encoders_5_self_attn_linear_pos_weight = const()[name = string("encoder_encoders_5_self_attn_linear_pos_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93288896)))]; tensor encoder_encoders_5_self_attn_linear_out_bias = const()[name = string("encoder_encoders_5_self_attn_linear_out_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94337536)))]; tensor encoder_encoders_5_self_attn_linear_out_weight = const()[name = string("encoder_encoders_5_self_attn_linear_out_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94339648)))]; tensor encoder_encoders_5_norm_ff_bias = const()[name = string("encoder_encoders_5_norm_ff_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95388288)))]; tensor encoder_encoders_5_norm_ff_weight = const()[name = string("encoder_encoders_5_norm_ff_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95390400)))]; tensor encoder_encoders_5_feed_forward_w_1_bias = const()[name = string("encoder_encoders_5_feed_forward_w_1_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95392512)))]; tensor encoder_encoders_5_feed_forward_w_1_weight = const()[name = string("encoder_encoders_5_feed_forward_w_1_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95400768)))]; tensor encoder_encoders_5_feed_forward_w_2_bias = const()[name = string("encoder_encoders_5_feed_forward_w_2_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99595136)))]; tensor encoder_encoders_5_feed_forward_w_2_weight = const()[name = string("encoder_encoders_5_feed_forward_w_2_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99597248)))]; tensor encoder_up_layer_conv_bias = const()[name = string("encoder_up_layer_conv_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103791616)))]; tensor encoder_up_layer_conv_weight = const()[name = string("encoder_up_layer_conv_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103793728)))]; tensor encoder_up_embed_out_0_bias = const()[name = string("encoder_up_embed_out_0_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109036672)))]; tensor encoder_up_embed_out_0_weight = const()[name = string("encoder_up_embed_out_0_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109038784)))]; tensor encoder_up_embed_out_1_bias = const()[name = string("encoder_up_embed_out_1_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110087424)))]; tensor encoder_up_embed_out_1_weight = const()[name = string("encoder_up_embed_out_1_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110089536)))]; tensor encoder_up_encoders_0_norm_mha_bias = const()[name = string("encoder_up_encoders_0_norm_mha_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110091648)))]; tensor encoder_up_encoders_0_norm_mha_weight = const()[name = string("encoder_up_encoders_0_norm_mha_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110093760)))]; tensor encoder_up_encoders_0_self_attn_linear_q_bias = const()[name = string("encoder_up_encoders_0_self_attn_linear_q_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110095872)))]; tensor encoder_up_encoders_0_self_attn_linear_q_weight = const()[name = string("encoder_up_encoders_0_self_attn_linear_q_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110097984)))]; tensor encoder_up_encoders_0_self_attn_linear_k_bias = const()[name = string("encoder_up_encoders_0_self_attn_linear_k_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111146624)))]; tensor encoder_up_encoders_0_self_attn_linear_k_weight = const()[name = string("encoder_up_encoders_0_self_attn_linear_k_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111148736)))]; tensor encoder_up_encoders_0_self_attn_linear_v_bias = const()[name = string("encoder_up_encoders_0_self_attn_linear_v_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112197376)))]; tensor encoder_up_encoders_0_self_attn_linear_v_weight = const()[name = string("encoder_up_encoders_0_self_attn_linear_v_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112199488)))]; tensor encoder_up_encoders_0_self_attn_linear_pos_weight = const()[name = string("encoder_up_encoders_0_self_attn_linear_pos_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113248128)))]; tensor encoder_up_encoders_0_self_attn_linear_out_bias = const()[name = string("encoder_up_encoders_0_self_attn_linear_out_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114296768)))]; tensor encoder_up_encoders_0_self_attn_linear_out_weight = const()[name = string("encoder_up_encoders_0_self_attn_linear_out_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114298880)))]; tensor encoder_up_encoders_0_norm_ff_bias = const()[name = string("encoder_up_encoders_0_norm_ff_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115347520)))]; tensor encoder_up_encoders_0_norm_ff_weight = const()[name = string("encoder_up_encoders_0_norm_ff_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115349632)))]; tensor encoder_up_encoders_0_feed_forward_w_1_bias = const()[name = string("encoder_up_encoders_0_feed_forward_w_1_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115351744)))]; tensor encoder_up_encoders_0_feed_forward_w_1_weight = const()[name = string("encoder_up_encoders_0_feed_forward_w_1_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115360000)))]; tensor encoder_up_encoders_0_feed_forward_w_2_bias = const()[name = string("encoder_up_encoders_0_feed_forward_w_2_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119554368)))]; tensor encoder_up_encoders_0_feed_forward_w_2_weight = const()[name = string("encoder_up_encoders_0_feed_forward_w_2_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119556480)))]; tensor encoder_up_encoders_1_norm_mha_bias = const()[name = string("encoder_up_encoders_1_norm_mha_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123750848)))]; tensor encoder_up_encoders_1_norm_mha_weight = const()[name = string("encoder_up_encoders_1_norm_mha_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123752960)))]; tensor encoder_up_encoders_1_self_attn_linear_q_bias = const()[name = string("encoder_up_encoders_1_self_attn_linear_q_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123755072)))]; tensor encoder_up_encoders_1_self_attn_linear_q_weight = const()[name = string("encoder_up_encoders_1_self_attn_linear_q_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123757184)))]; tensor encoder_up_encoders_1_self_attn_linear_k_bias = const()[name = string("encoder_up_encoders_1_self_attn_linear_k_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124805824)))]; tensor encoder_up_encoders_1_self_attn_linear_k_weight = const()[name = string("encoder_up_encoders_1_self_attn_linear_k_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124807936)))]; tensor encoder_up_encoders_1_self_attn_linear_v_bias = const()[name = string("encoder_up_encoders_1_self_attn_linear_v_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125856576)))]; tensor encoder_up_encoders_1_self_attn_linear_v_weight = const()[name = string("encoder_up_encoders_1_self_attn_linear_v_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125858688)))]; tensor encoder_up_encoders_1_self_attn_linear_pos_weight = const()[name = string("encoder_up_encoders_1_self_attn_linear_pos_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126907328)))]; tensor encoder_up_encoders_1_self_attn_linear_out_bias = const()[name = string("encoder_up_encoders_1_self_attn_linear_out_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127955968)))]; tensor encoder_up_encoders_1_self_attn_linear_out_weight = const()[name = string("encoder_up_encoders_1_self_attn_linear_out_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127958080)))]; tensor encoder_up_encoders_1_norm_ff_bias = const()[name = string("encoder_up_encoders_1_norm_ff_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129006720)))]; tensor encoder_up_encoders_1_norm_ff_weight = const()[name = string("encoder_up_encoders_1_norm_ff_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129008832)))]; tensor encoder_up_encoders_1_feed_forward_w_1_bias = const()[name = string("encoder_up_encoders_1_feed_forward_w_1_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129010944)))]; tensor encoder_up_encoders_1_feed_forward_w_1_weight = const()[name = string("encoder_up_encoders_1_feed_forward_w_1_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129019200)))]; tensor encoder_up_encoders_1_feed_forward_w_2_bias = const()[name = string("encoder_up_encoders_1_feed_forward_w_2_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133213568)))]; tensor encoder_up_encoders_1_feed_forward_w_2_weight = const()[name = string("encoder_up_encoders_1_feed_forward_w_2_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133215680)))]; tensor encoder_up_encoders_2_norm_mha_bias = const()[name = string("encoder_up_encoders_2_norm_mha_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137410048)))]; tensor encoder_up_encoders_2_norm_mha_weight = const()[name = string("encoder_up_encoders_2_norm_mha_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137412160)))]; tensor encoder_up_encoders_2_self_attn_linear_q_bias = const()[name = string("encoder_up_encoders_2_self_attn_linear_q_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137414272)))]; tensor encoder_up_encoders_2_self_attn_linear_q_weight = const()[name = string("encoder_up_encoders_2_self_attn_linear_q_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137416384)))]; tensor encoder_up_encoders_2_self_attn_linear_k_bias = const()[name = string("encoder_up_encoders_2_self_attn_linear_k_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138465024)))]; tensor encoder_up_encoders_2_self_attn_linear_k_weight = const()[name = string("encoder_up_encoders_2_self_attn_linear_k_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138467136)))]; tensor encoder_up_encoders_2_self_attn_linear_v_bias = const()[name = string("encoder_up_encoders_2_self_attn_linear_v_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139515776)))]; tensor encoder_up_encoders_2_self_attn_linear_v_weight = const()[name = string("encoder_up_encoders_2_self_attn_linear_v_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139517888)))]; tensor encoder_up_encoders_2_self_attn_linear_pos_weight = const()[name = string("encoder_up_encoders_2_self_attn_linear_pos_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140566528)))]; tensor encoder_up_encoders_2_self_attn_linear_out_bias = const()[name = string("encoder_up_encoders_2_self_attn_linear_out_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141615168)))]; tensor encoder_up_encoders_2_self_attn_linear_out_weight = const()[name = string("encoder_up_encoders_2_self_attn_linear_out_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141617280)))]; tensor encoder_up_encoders_2_norm_ff_bias = const()[name = string("encoder_up_encoders_2_norm_ff_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142665920)))]; tensor encoder_up_encoders_2_norm_ff_weight = const()[name = string("encoder_up_encoders_2_norm_ff_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142668032)))]; tensor encoder_up_encoders_2_feed_forward_w_1_bias = const()[name = string("encoder_up_encoders_2_feed_forward_w_1_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142670144)))]; tensor encoder_up_encoders_2_feed_forward_w_1_weight = const()[name = string("encoder_up_encoders_2_feed_forward_w_1_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142678400)))]; tensor encoder_up_encoders_2_feed_forward_w_2_bias = const()[name = string("encoder_up_encoders_2_feed_forward_w_2_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146872768)))]; tensor encoder_up_encoders_2_feed_forward_w_2_weight = const()[name = string("encoder_up_encoders_2_feed_forward_w_2_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146874880)))]; tensor encoder_up_encoders_3_norm_mha_bias = const()[name = string("encoder_up_encoders_3_norm_mha_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151069248)))]; tensor encoder_up_encoders_3_norm_mha_weight = const()[name = string("encoder_up_encoders_3_norm_mha_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151071360)))]; tensor encoder_up_encoders_3_self_attn_linear_q_bias = const()[name = string("encoder_up_encoders_3_self_attn_linear_q_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151073472)))]; tensor encoder_up_encoders_3_self_attn_linear_q_weight = const()[name = string("encoder_up_encoders_3_self_attn_linear_q_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151075584)))]; tensor encoder_up_encoders_3_self_attn_linear_k_bias = const()[name = string("encoder_up_encoders_3_self_attn_linear_k_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152124224)))]; tensor encoder_up_encoders_3_self_attn_linear_k_weight = const()[name = string("encoder_up_encoders_3_self_attn_linear_k_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152126336)))]; tensor encoder_up_encoders_3_self_attn_linear_v_bias = const()[name = string("encoder_up_encoders_3_self_attn_linear_v_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153174976)))]; tensor encoder_up_encoders_3_self_attn_linear_v_weight = const()[name = string("encoder_up_encoders_3_self_attn_linear_v_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153177088)))]; tensor encoder_up_encoders_3_self_attn_linear_pos_weight = const()[name = string("encoder_up_encoders_3_self_attn_linear_pos_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154225728)))]; tensor encoder_up_encoders_3_self_attn_linear_out_bias = const()[name = string("encoder_up_encoders_3_self_attn_linear_out_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155274368)))]; tensor encoder_up_encoders_3_self_attn_linear_out_weight = const()[name = string("encoder_up_encoders_3_self_attn_linear_out_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155276480)))]; tensor encoder_up_encoders_3_norm_ff_bias = const()[name = string("encoder_up_encoders_3_norm_ff_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156325120)))]; tensor encoder_up_encoders_3_norm_ff_weight = const()[name = string("encoder_up_encoders_3_norm_ff_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156327232)))]; tensor encoder_up_encoders_3_feed_forward_w_1_bias = const()[name = string("encoder_up_encoders_3_feed_forward_w_1_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156329344)))]; tensor encoder_up_encoders_3_feed_forward_w_1_weight = const()[name = string("encoder_up_encoders_3_feed_forward_w_1_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156337600)))]; tensor encoder_up_encoders_3_feed_forward_w_2_bias = const()[name = string("encoder_up_encoders_3_feed_forward_w_2_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160531968)))]; tensor encoder_up_encoders_3_feed_forward_w_2_weight = const()[name = string("encoder_up_encoders_3_feed_forward_w_2_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160534080)))]; tensor encoder_after_norm_bias = const()[name = string("encoder_after_norm_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164728448)))]; tensor encoder_after_norm_weight = const()[name = string("encoder_after_norm_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164730560)))]; tensor encoder_proj_bias = const()[name = string("encoder_proj_bias"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164732672)))]; tensor encoder_proj_weight = const()[name = string("encoder_proj_weight"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164733056)))]; int32 var_33_batch_dims_0 = const()[name = string("op_33_batch_dims_0"), val = int32(0)]; bool var_33_validate_indices_0 = const()[name = string("op_33_validate_indices_0"), val = bool(false)]; int32 greater_equal_1_y_0 = const()[name = string("greater_equal_1_y_0"), val = int32(0)]; tensor greater_equal_1 = greater_equal(x = all_tokens, y = greater_equal_1_y_0)[name = string("greater_equal_1")]; int32 slice_by_index_1 = const()[name = string("slice_by_index_1"), val = int32(6561)]; tensor add_1 = add(x = all_tokens, y = slice_by_index_1)[name = string("add_1")]; tensor select_3 = select(a = all_tokens, b = add_1, cond = greater_equal_1)[name = string("select_3")]; int32 var_33_axis_1 = const()[name = string("op_33_axis_1"), val = int32(0)]; tensor var_33 = gather(axis = var_33_axis_1, batch_dims = var_33_batch_dims_0, indices = select_3, validate_indices = var_33_validate_indices_0, x = input_embedding_weight)[name = string("op_33")]; fp32 var_36 = const()[name = string("op_36"), val = fp32(0x1.197998p-40)]; fp32 var_37 = const()[name = string("op_37"), val = fp32(-0x1.ff933cp+127)]; fp32 var_46 = const()[name = string("op_46"), val = fp32(0x0p+0)]; fp32 var_47 = const()[name = string("op_47"), val = fp32(0x1.47ae14p-7)]; fp32 var_50 = const()[name = string("op_50"), val = fp32(0x1.4f8b58p-17)]; int32 var_53 = const()[name = string("op_53"), val = int32(2)]; tensor var_54 = const()[name = string("op_54"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164896960)))]; int32 var_55 = const()[name = string("op_55"), val = int32(-1)]; int32 var_57 = const()[name = string("op_57"), val = int32(0)]; int32 var_61 = const()[name = string("op_61"), val = int32(1)]; tensor var_87_shape = shape(x = var_33)[name = string("op_87_shape")]; int32 gather_1_batch_dims_0 = const()[name = string("gather_1_batch_dims_0"), val = int32(0)]; bool gather_1_validate_indices_0 = const()[name = string("gather_1_validate_indices_0"), val = bool(false)]; int32 select_4 = const()[name = string("select_4"), val = int32(1)]; int32 gather_1_axis_1 = const()[name = string("gather_1_axis_1"), val = int32(0)]; int32 gather_1 = gather(axis = gather_1_axis_1, batch_dims = gather_1_batch_dims_0, indices = select_4, validate_indices = gather_1_validate_indices_0, x = var_87_shape)[name = string("gather_1")]; int32 const_2 = const()[name = string("const_2"), val = int32(1)]; int32 const_3 = const()[name = string("const_3"), val = int32(1)]; tensor seq_range_1 = range_1d(end = gather_1, start = var_57, step = const_3)[name = string("seq_range_1")]; tensor var_91_axes_0 = const()[name = string("op_91_axes_0"), val = tensor([0])]; tensor var_91 = expand_dims(axes = var_91_axes_0, x = seq_range_1)[name = string("op_91")]; int32 concat_1_axis_0 = const()[name = string("concat_1_axis_0"), val = int32(0)]; bool concat_1_interleave_0 = const()[name = string("concat_1_interleave_0"), val = bool(false)]; tensor concat_1 = concat(axis = concat_1_axis_0, interleave = concat_1_interleave_0, values = (const_2, gather_1))[name = string("concat_1")]; tensor shape_0 = shape(x = var_91)[name = string("shape_0")]; int32 equal_0_y_0 = const()[name = string("equal_0_y_0"), val = int32(-1)]; tensor equal_0 = equal(x = concat_1, y = equal_0_y_0)[name = string("equal_0")]; tensor select_0 = select(a = shape_0, b = concat_1, cond = equal_0)[name = string("select_0")]; tensor real_div_0 = real_div(x = select_0, y = shape_0)[name = string("real_div_0")]; tensor seq_range_expand_1 = tile(reps = real_div_0, x = var_91)[name = string("seq_range_expand_1")]; tensor seq_length_expand_1 = const()[name = string("seq_length_expand_1"), val = tensor([[500]])]; tensor var_95 = greater_equal(x = seq_range_expand_1, y = seq_length_expand_1)[name = string("op_95")]; tensor var_96_axes_0 = const()[name = string("op_96_axes_0"), val = tensor([1])]; tensor var_96 = expand_dims(axes = var_96_axes_0, x = var_95)[name = string("op_96")]; tensor masks_1 = logical_not(x = var_96)[name = string("masks_1")]; tensor input_3 = linear(bias = encoder_embed_out_0_bias, weight = encoder_embed_out_0_weight, x = var_33)[name = string("linear_0")]; tensor input_5_axes_0 = const()[name = string("input_5_axes_0"), val = tensor([-1])]; tensor input_5 = layer_norm(axes = input_5_axes_0, beta = encoder_embed_out_1_bias, epsilon = var_50, gamma = encoder_embed_out_1_weight, x = input_3)[name = string("input_5")]; fp32 var_109 = const()[name = string("op_109"), val = fp32(0x1.6a09e6p+4)]; tensor x_3 = mul(x = input_5, y = var_109)[name = string("x_3")]; tensor var_111_shape = shape(x = x_3)[name = string("op_111_shape")]; int32 gather_2_batch_dims_0 = const()[name = string("gather_2_batch_dims_0"), val = int32(0)]; bool gather_2_validate_indices_0 = const()[name = string("gather_2_validate_indices_0"), val = bool(false)]; int32 select_5 = const()[name = string("select_5"), val = int32(1)]; int32 gather_2_axis_1 = const()[name = string("gather_2_axis_1"), val = int32(0)]; int32 gather_2 = gather(axis = gather_2_axis_1, batch_dims = gather_2_batch_dims_0, indices = select_5, validate_indices = gather_2_validate_indices_0, x = var_111_shape)[name = string("gather_2")]; tensor var_115 = const()[name = string("op_115"), val = tensor([0x1.387p+12])]; string gather_2_promoted_dtype_0 = const()[name = string("gather_2_promoted_dtype_0"), val = string("fp32")]; fp32 gather_2_promoted = cast(dtype = gather_2_promoted_dtype_0, x = gather_2)[name = string("cast_130")]; tensor var_116 = sub(x = var_115, y = gather_2_promoted)[name = string("op_116")]; fp32 var_117_promoted = const()[name = string("op_117_promoted"), val = fp32(0x1p+0)]; tensor var_118 = add(x = var_116, y = var_117_promoted)[name = string("op_118")]; fp32 var_119_item = squeeze(x = var_118)[name = string("op_119_item")]; string var_119_dtype_0 = const()[name = string("op_119_dtype_0"), val = string("int32")]; tensor var_122 = const()[name = string("op_122"), val = tensor([0x1.387p+12])]; tensor var_123 = add(x = var_122, y = gather_2_promoted)[name = string("op_123")]; fp32 var_124_item = squeeze(x = var_123)[name = string("op_124_item")]; string var_124_dtype_0 = const()[name = string("op_124_dtype_0"), val = string("int32")]; int32 concat_2_values0_0 = const()[name = string("concat_2_values0_0"), val = int32(0)]; int32 concat_2_values2_0 = const()[name = string("concat_2_values2_0"), val = int32(0)]; int32 concat_2_axis_0 = const()[name = string("concat_2_axis_0"), val = int32(0)]; bool concat_2_interleave_0 = const()[name = string("concat_2_interleave_0"), val = bool(false)]; int32 var_119 = cast(dtype = var_119_dtype_0, x = var_119_item)[name = string("cast_129")]; tensor concat_2 = concat(axis = concat_2_axis_0, interleave = concat_2_interleave_0, values = (concat_2_values0_0, var_119, concat_2_values2_0))[name = string("concat_2")]; int32 concat_3_values0_0 = const()[name = string("concat_3_values0_0"), val = int32(1)]; int32 concat_3_values2_0 = const()[name = string("concat_3_values2_0"), val = int32(512)]; int32 concat_3_axis_0 = const()[name = string("concat_3_axis_0"), val = int32(0)]; bool concat_3_interleave_0 = const()[name = string("concat_3_interleave_0"), val = bool(false)]; int32 var_124 = cast(dtype = var_124_dtype_0, x = var_124_item)[name = string("cast_128")]; tensor concat_3 = concat(axis = concat_3_axis_0, interleave = concat_3_interleave_0, values = (concat_3_values0_0, var_124, concat_3_values2_0))[name = string("concat_3")]; tensor input_7_end_mask_0 = const()[name = string("input_7_end_mask_0"), val = tensor([true, false, true])]; tensor input_7 = slice_by_index(begin = concat_2, end = concat_3, end_mask = input_7_end_mask_0, x = var_54)[name = string("input_7")]; tensor var_137_perm_0 = const()[name = string("op_137_perm_0"), val = tensor([0, 2, 1])]; fp32 const_6 = const()[name = string("const_6"), val = fp32(0x0p+0)]; tensor input_11_pad_0 = const()[name = string("input_11_pad_0"), val = tensor([0, 0, 0, 0, 0, 3])]; string input_11_mode_0 = const()[name = string("input_11_mode_0"), val = string("constant")]; tensor var_137 = transpose(perm = var_137_perm_0, x = x_3)[name = string("transpose_164")]; tensor input_11 = pad(constant_val = const_6, mode = input_11_mode_0, pad = input_11_pad_0, x = var_137)[name = string("input_11")]; string input_13_pad_type_0 = const()[name = string("input_13_pad_type_0"), val = string("valid")]; tensor input_13_strides_0 = const()[name = string("input_13_strides_0"), val = tensor([1])]; tensor input_13_pad_0 = const()[name = string("input_13_pad_0"), val = tensor([0, 0])]; tensor input_13_dilations_0 = const()[name = string("input_13_dilations_0"), val = tensor([1])]; int32 input_13_groups_0 = const()[name = string("input_13_groups_0"), val = int32(1)]; tensor input_13 = conv(bias = encoder_pre_lookahead_layer_conv1_bias, 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 = encoder_pre_lookahead_layer_conv1_weight, x = input_11)[name = string("input_13")]; tensor input_15 = leaky_relu(alpha = var_47, x = input_13)[name = string("input_15")]; fp32 const_7 = const()[name = string("const_7"), val = fp32(0x0p+0)]; tensor input_17_pad_0 = const()[name = string("input_17_pad_0"), val = tensor([0, 0, 0, 0, 2, 0])]; string input_17_mode_0 = const()[name = string("input_17_mode_0"), val = string("constant")]; tensor input_17 = pad(constant_val = const_7, mode = input_17_mode_0, pad = input_17_pad_0, x = input_15)[name = string("input_17")]; string outputs_1_pad_type_0 = const()[name = string("outputs_1_pad_type_0"), val = string("valid")]; tensor outputs_1_strides_0 = const()[name = string("outputs_1_strides_0"), val = tensor([1])]; tensor outputs_1_pad_0 = const()[name = string("outputs_1_pad_0"), val = tensor([0, 0])]; tensor outputs_1_dilations_0 = const()[name = string("outputs_1_dilations_0"), val = tensor([1])]; int32 outputs_1_groups_0 = const()[name = string("outputs_1_groups_0"), val = int32(1)]; tensor outputs_1 = conv(bias = encoder_pre_lookahead_layer_conv2_bias, dilations = outputs_1_dilations_0, groups = outputs_1_groups_0, pad = outputs_1_pad_0, pad_type = outputs_1_pad_type_0, strides = outputs_1_strides_0, weight = encoder_pre_lookahead_layer_conv2_weight, x = input_17)[name = string("outputs_1")]; tensor var_158_perm_0 = const()[name = string("op_158_perm_0"), val = tensor([0, 2, 1])]; tensor var_158 = transpose(perm = var_158_perm_0, x = outputs_1)[name = string("transpose_163")]; tensor input_19 = add(x = var_158, y = x_3)[name = string("input_19")]; tensor query_1_axes_0 = const()[name = string("query_1_axes_0"), val = tensor([-1])]; tensor query_1 = layer_norm(axes = query_1_axes_0, beta = encoder_encoders_0_norm_mha_bias, epsilon = var_36, gamma = encoder_encoders_0_norm_mha_weight, x = input_19)[name = string("query_1")]; tensor var_179 = linear(bias = encoder_encoders_0_self_attn_linear_q_bias, weight = encoder_encoders_0_self_attn_linear_q_weight, x = query_1)[name = string("linear_1")]; tensor var_180 = const()[name = string("op_180"), val = tensor([1, -1, 8, 64])]; tensor q_1 = reshape(shape = var_180, x = var_179)[name = string("q_1")]; tensor var_184 = linear(bias = encoder_encoders_0_self_attn_linear_k_bias, weight = encoder_encoders_0_self_attn_linear_k_weight, x = query_1)[name = string("linear_2")]; tensor var_185 = const()[name = string("op_185"), val = tensor([1, -1, 8, 64])]; tensor k_1 = reshape(shape = var_185, x = var_184)[name = string("k_1")]; tensor var_189 = linear(bias = encoder_encoders_0_self_attn_linear_v_bias, weight = encoder_encoders_0_self_attn_linear_v_weight, x = query_1)[name = string("linear_3")]; tensor var_190 = const()[name = string("op_190"), val = tensor([1, -1, 8, 64])]; tensor v_1 = reshape(shape = var_190, x = var_189)[name = string("v_1")]; tensor v_3_perm_0 = const()[name = string("v_3_perm_0"), val = tensor([0, 2, 1, 3])]; tensor linear_4_bias_0 = const()[name = string("linear_4_bias_0"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185374976)))]; tensor var_198 = linear(bias = linear_4_bias_0, weight = encoder_encoders_0_self_attn_linear_pos_weight, x = input_7)[name = string("linear_4")]; tensor var_199 = const()[name = string("op_199"), val = tensor([1, -1, 8, 64])]; tensor p_1 = reshape(shape = var_199, x = var_198)[name = string("p_1")]; tensor const_8 = const()[name = string("const_8"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185377088)))]; tensor var_203 = add(x = q_1, y = const_8)[name = string("op_203")]; tensor const_9 = const()[name = string("const_9"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185379200)))]; tensor var_206 = add(x = q_1, y = const_9)[name = string("op_206")]; bool matrix_ac_1_transpose_x_0 = const()[name = string("matrix_ac_1_transpose_x_0"), val = bool(false)]; bool matrix_ac_1_transpose_y_0 = const()[name = string("matrix_ac_1_transpose_y_0"), val = bool(false)]; tensor transpose_60_perm_0 = const()[name = string("transpose_60_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_61_perm_0 = const()[name = string("transpose_61_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_61 = transpose(perm = transpose_61_perm_0, x = k_1)[name = string("transpose_160")]; tensor transpose_60 = transpose(perm = transpose_60_perm_0, x = var_203)[name = string("transpose_161")]; tensor matrix_ac_1 = matmul(transpose_x = matrix_ac_1_transpose_x_0, transpose_y = matrix_ac_1_transpose_y_0, x = transpose_60, y = transpose_61)[name = string("matrix_ac_1")]; bool matrix_bd_1_transpose_x_0 = const()[name = string("matrix_bd_1_transpose_x_0"), val = bool(false)]; bool matrix_bd_1_transpose_y_0 = const()[name = string("matrix_bd_1_transpose_y_0"), val = bool(false)]; tensor transpose_62_perm_0 = const()[name = string("transpose_62_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_63_perm_0 = const()[name = string("transpose_63_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_63 = transpose(perm = transpose_63_perm_0, x = p_1)[name = string("transpose_158")]; tensor transpose_62 = transpose(perm = transpose_62_perm_0, x = var_206)[name = string("transpose_159")]; tensor matrix_bd_1 = matmul(transpose_x = matrix_bd_1_transpose_x_0, transpose_y = matrix_bd_1_transpose_y_0, x = transpose_62, y = transpose_63)[name = string("matrix_bd_1")]; tensor var_212_shape = shape(x = matrix_bd_1)[name = string("op_212_shape")]; int32 gather_5 = const()[name = string("gather_5"), val = int32(1)]; int32 gather_6 = const()[name = string("gather_6"), val = int32(8)]; int32 gather_7_batch_dims_0 = const()[name = string("gather_7_batch_dims_0"), val = int32(0)]; bool gather_7_validate_indices_0 = const()[name = string("gather_7_validate_indices_0"), val = bool(false)]; int32 select_6 = const()[name = string("select_6"), val = int32(2)]; int32 gather_7_axis_1 = const()[name = string("gather_7_axis_1"), val = int32(0)]; int32 gather_7 = gather(axis = gather_7_axis_1, batch_dims = gather_7_batch_dims_0, indices = select_6, validate_indices = gather_7_validate_indices_0, x = var_212_shape)[name = string("gather_7")]; int32 concat_4_axis_0 = const()[name = string("concat_4_axis_0"), val = int32(0)]; bool concat_4_interleave_0 = const()[name = string("concat_4_interleave_0"), val = bool(false)]; tensor concat_4 = concat(axis = concat_4_axis_0, interleave = concat_4_interleave_0, values = (gather_5, gather_6, gather_7, var_61))[name = string("concat_4")]; fp32 zero_pad_1_value_0 = const()[name = string("zero_pad_1_value_0"), val = fp32(0x0p+0)]; tensor zero_pad_1 = fill(shape = concat_4, value = zero_pad_1_value_0)[name = string("zero_pad_1")]; bool x_padded_1_interleave_0 = const()[name = string("x_padded_1_interleave_0"), val = bool(false)]; tensor x_padded_1 = concat(axis = var_55, interleave = x_padded_1_interleave_0, values = (zero_pad_1, matrix_bd_1))[name = string("x_padded_1")]; int32 gather_8 = const()[name = string("gather_8"), val = int32(1)]; int32 gather_9 = const()[name = string("gather_9"), val = int32(8)]; int32 gather_10_batch_dims_0 = const()[name = string("gather_10_batch_dims_0"), val = int32(0)]; bool gather_10_validate_indices_0 = const()[name = string("gather_10_validate_indices_0"), val = bool(false)]; int32 select_7 = const()[name = string("select_7"), val = int32(3)]; int32 gather_10_axis_1 = const()[name = string("gather_10_axis_1"), val = int32(0)]; int32 gather_10 = gather(axis = gather_10_axis_1, batch_dims = gather_10_batch_dims_0, indices = select_7, validate_indices = gather_10_validate_indices_0, x = var_212_shape)[name = string("gather_10")]; int32 var_223 = const()[name = string("op_223"), val = int32(1)]; int32 var_224 = add(x = gather_10, y = var_223)[name = string("op_224")]; int32 concat_5_axis_0 = const()[name = string("concat_5_axis_0"), val = int32(0)]; bool concat_5_interleave_0 = const()[name = string("concat_5_interleave_0"), val = bool(false)]; tensor concat_5 = concat(axis = concat_5_axis_0, interleave = concat_5_interleave_0, values = (gather_8, gather_9, var_224, gather_7))[name = string("concat_5")]; tensor x_padded_3 = reshape(shape = concat_5, x = x_padded_1)[name = string("x_padded_3")]; tensor var_231_begin_0 = const()[name = string("op_231_begin_0"), val = tensor([0, 0, 1, 0])]; tensor var_231_end_0 = const()[name = string("op_231_end_0"), val = tensor([1, 8, 0, 0])]; tensor var_231_end_mask_0 = const()[name = string("op_231_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_231 = slice_by_index(begin = var_231_begin_0, end = var_231_end_0, end_mask = var_231_end_mask_0, x = x_padded_3)[name = string("op_231")]; int32 gather_12 = const()[name = string("gather_12"), val = int32(1)]; int32 gather_13 = const()[name = string("gather_13"), val = int32(8)]; int32 concat_6_axis_0 = const()[name = string("concat_6_axis_0"), val = int32(0)]; bool concat_6_interleave_0 = const()[name = string("concat_6_interleave_0"), val = bool(false)]; tensor concat_6 = concat(axis = concat_6_axis_0, interleave = concat_6_interleave_0, values = (gather_12, gather_13, gather_7, gather_10))[name = string("concat_6")]; tensor var_237 = reshape(shape = concat_6, x = var_231)[name = string("op_237")]; int32 floor_div_2 = floor_div(x = gather_10, y = var_53)[name = string("floor_div_2")]; string var_240_dtype_0 = const()[name = string("op_240_dtype_0"), val = string("fp32")]; fp32 var_241_promoted = const()[name = string("op_241_promoted"), val = fp32(0x1p+0)]; fp32 var_240 = cast(dtype = var_240_dtype_0, x = floor_div_2)[name = string("cast_127")]; fp32 var_242 = add(x = var_240, y = var_241_promoted)[name = string("op_242")]; string var_243_dtype_0 = const()[name = string("op_243_dtype_0"), val = string("int32")]; int32 concat_7_values0_0 = const()[name = string("concat_7_values0_0"), val = int32(1)]; int32 concat_7_values1_0 = const()[name = string("concat_7_values1_0"), val = int32(8)]; int32 concat_7_values2_0 = const()[name = string("concat_7_values2_0"), val = int32(0)]; int32 concat_7_axis_0 = const()[name = string("concat_7_axis_0"), val = int32(0)]; bool concat_7_interleave_0 = const()[name = string("concat_7_interleave_0"), val = bool(false)]; int32 var_243 = cast(dtype = var_243_dtype_0, x = var_242)[name = string("cast_126")]; tensor concat_7 = concat(axis = concat_7_axis_0, interleave = concat_7_interleave_0, values = (concat_7_values0_0, concat_7_values1_0, concat_7_values2_0, var_243))[name = string("concat_7")]; tensor matrix_bd_3_begin_0 = const()[name = string("matrix_bd_3_begin_0"), val = tensor([0, 0, 0, 0])]; tensor matrix_bd_3_end_mask_0 = const()[name = string("matrix_bd_3_end_mask_0"), val = tensor([true, true, true, false])]; tensor matrix_bd_3 = slice_by_index(begin = matrix_bd_3_begin_0, end = concat_7, end_mask = matrix_bd_3_end_mask_0, x = var_237)[name = string("matrix_bd_3")]; tensor var_248 = add(x = matrix_ac_1, y = matrix_bd_3)[name = string("op_248")]; fp32 _inversed_scores_1_y_0 = const()[name = string("_inversed_scores_1_y_0"), val = fp32(0x1p-3)]; tensor _inversed_scores_1 = mul(x = var_248, y = _inversed_scores_1_y_0)[name = string("_inversed_scores_1")]; tensor var_252_axes_0 = const()[name = string("op_252_axes_0"), val = tensor([1])]; tensor var_252 = expand_dims(axes = var_252_axes_0, x = masks_1)[name = string("op_252")]; string cast_10_dtype_0 = const()[name = string("cast_10_dtype_0"), val = string("int32")]; tensor cast_10 = cast(dtype = cast_10_dtype_0, x = var_252)[name = string("cast_125")]; tensor mask_3 = equal(x = cast_10, y = var_57)[name = string("mask_3")]; tensor var_254_shape = shape(x = _inversed_scores_1)[name = string("op_254_shape")]; int32 gather_18_batch_dims_0 = const()[name = string("gather_18_batch_dims_0"), val = int32(0)]; bool gather_18_validate_indices_0 = const()[name = string("gather_18_validate_indices_0"), val = bool(false)]; int32 select_9 = const()[name = string("select_9"), val = int32(3)]; int32 gather_18_axis_1 = const()[name = string("gather_18_axis_1"), val = int32(0)]; int32 gather_18 = gather(axis = gather_18_axis_1, batch_dims = gather_18_batch_dims_0, indices = select_9, validate_indices = gather_18_validate_indices_0, x = var_254_shape)[name = string("gather_18")]; int32 concat_8_values0_0 = const()[name = string("concat_8_values0_0"), val = int32(0)]; int32 concat_8_values1_0 = const()[name = string("concat_8_values1_0"), val = int32(1)]; int32 concat_8_values2_0 = const()[name = string("concat_8_values2_0"), val = int32(1)]; int32 concat_8_axis_0 = const()[name = string("concat_8_axis_0"), val = int32(0)]; bool concat_8_interleave_0 = const()[name = string("concat_8_interleave_0"), val = bool(false)]; tensor concat_8 = concat(axis = concat_8_axis_0, interleave = concat_8_interleave_0, values = (concat_8_values0_0, concat_8_values1_0, concat_8_values2_0, gather_18))[name = string("concat_8")]; tensor mask_5_begin_0 = const()[name = string("mask_5_begin_0"), val = tensor([0, 0, 0, 0])]; tensor mask_5_end_mask_0 = const()[name = string("mask_5_end_mask_0"), val = tensor([true, true, true, false])]; tensor mask_5 = slice_by_index(begin = mask_5_begin_0, end = concat_8, end_mask = mask_5_end_mask_0, x = mask_3)[name = string("mask_5")]; tensor scores_3 = select(a = var_37, b = _inversed_scores_1, cond = mask_5)[name = string("scores_3")]; tensor var_260 = softmax(axis = var_55, x = scores_3)[name = string("op_260")]; tensor input_21 = select(a = var_46, b = var_260, cond = mask_5)[name = string("input_21")]; bool x_5_transpose_x_0 = const()[name = string("x_5_transpose_x_0"), val = bool(false)]; bool x_5_transpose_y_0 = const()[name = string("x_5_transpose_y_0"), val = bool(false)]; tensor v_3 = transpose(perm = v_3_perm_0, x = v_1)[name = string("transpose_162")]; tensor x_5 = matmul(transpose_x = x_5_transpose_x_0, transpose_y = x_5_transpose_y_0, x = input_21, y = v_3)[name = string("x_5")]; tensor var_264_perm_0 = const()[name = string("op_264_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_266 = const()[name = string("op_266"), val = tensor([1, -1, 512])]; tensor var_264 = transpose(perm = var_264_perm_0, x = x_5)[name = string("transpose_157")]; tensor input_23 = reshape(shape = var_266, x = var_264)[name = string("input_23")]; tensor input_25 = linear(bias = encoder_encoders_0_self_attn_linear_out_bias, weight = encoder_encoders_0_self_attn_linear_out_weight, x = input_23)[name = string("linear_5")]; tensor input_27 = add(x = input_19, y = input_25)[name = string("input_27")]; tensor input_29_axes_0 = const()[name = string("input_29_axes_0"), val = tensor([-1])]; tensor input_29 = layer_norm(axes = input_29_axes_0, beta = encoder_encoders_0_norm_ff_bias, epsilon = var_36, gamma = encoder_encoders_0_norm_ff_weight, x = input_27)[name = string("input_29")]; tensor input_31 = linear(bias = encoder_encoders_0_feed_forward_w_1_bias, weight = encoder_encoders_0_feed_forward_w_1_weight, x = input_29)[name = string("linear_6")]; tensor input_33 = silu(x = input_31)[name = string("input_33")]; tensor input_37 = linear(bias = encoder_encoders_0_feed_forward_w_2_bias, weight = encoder_encoders_0_feed_forward_w_2_weight, x = input_33)[name = string("linear_7")]; tensor input_39 = add(x = input_27, y = input_37)[name = string("input_39")]; tensor query_3_axes_0 = const()[name = string("query_3_axes_0"), val = tensor([-1])]; tensor query_3 = layer_norm(axes = query_3_axes_0, beta = encoder_encoders_1_norm_mha_bias, epsilon = var_36, gamma = encoder_encoders_1_norm_mha_weight, x = input_39)[name = string("query_3")]; tensor var_315 = linear(bias = encoder_encoders_1_self_attn_linear_q_bias, weight = encoder_encoders_1_self_attn_linear_q_weight, x = query_3)[name = string("linear_8")]; tensor var_316 = const()[name = string("op_316"), val = tensor([1, -1, 8, 64])]; tensor q_7 = reshape(shape = var_316, x = var_315)[name = string("q_7")]; tensor var_320 = linear(bias = encoder_encoders_1_self_attn_linear_k_bias, weight = encoder_encoders_1_self_attn_linear_k_weight, x = query_3)[name = string("linear_9")]; tensor var_321 = const()[name = string("op_321"), val = tensor([1, -1, 8, 64])]; tensor k_5 = reshape(shape = var_321, x = var_320)[name = string("k_5")]; tensor var_325 = linear(bias = encoder_encoders_1_self_attn_linear_v_bias, weight = encoder_encoders_1_self_attn_linear_v_weight, x = query_3)[name = string("linear_10")]; tensor var_326 = const()[name = string("op_326"), val = tensor([1, -1, 8, 64])]; tensor v_5 = reshape(shape = var_326, x = var_325)[name = string("v_5")]; tensor v_7_perm_0 = const()[name = string("v_7_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_334 = linear(bias = linear_4_bias_0, weight = encoder_encoders_1_self_attn_linear_pos_weight, x = input_7)[name = string("linear_11")]; tensor var_335 = const()[name = string("op_335"), val = tensor([1, -1, 8, 64])]; tensor p_5 = reshape(shape = var_335, x = var_334)[name = string("p_5")]; tensor const_10 = const()[name = string("const_10"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185381312)))]; tensor var_339 = add(x = q_7, y = const_10)[name = string("op_339")]; tensor const_11 = const()[name = string("const_11"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185383424)))]; tensor var_342 = add(x = q_7, y = const_11)[name = string("op_342")]; bool matrix_ac_3_transpose_x_0 = const()[name = string("matrix_ac_3_transpose_x_0"), val = bool(false)]; bool matrix_ac_3_transpose_y_0 = const()[name = string("matrix_ac_3_transpose_y_0"), val = bool(false)]; tensor transpose_64_perm_0 = const()[name = string("transpose_64_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_65_perm_0 = const()[name = string("transpose_65_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_65 = transpose(perm = transpose_65_perm_0, x = k_5)[name = string("transpose_154")]; tensor transpose_64 = transpose(perm = transpose_64_perm_0, x = var_339)[name = string("transpose_155")]; tensor matrix_ac_3 = matmul(transpose_x = matrix_ac_3_transpose_x_0, transpose_y = matrix_ac_3_transpose_y_0, x = transpose_64, y = transpose_65)[name = string("matrix_ac_3")]; bool matrix_bd_5_transpose_x_0 = const()[name = string("matrix_bd_5_transpose_x_0"), val = bool(false)]; bool matrix_bd_5_transpose_y_0 = const()[name = string("matrix_bd_5_transpose_y_0"), val = bool(false)]; tensor transpose_66_perm_0 = const()[name = string("transpose_66_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_67_perm_0 = const()[name = string("transpose_67_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_67 = transpose(perm = transpose_67_perm_0, x = p_5)[name = string("transpose_152")]; tensor transpose_66 = transpose(perm = transpose_66_perm_0, x = var_342)[name = string("transpose_153")]; tensor matrix_bd_5 = matmul(transpose_x = matrix_bd_5_transpose_x_0, transpose_y = matrix_bd_5_transpose_y_0, x = transpose_66, y = transpose_67)[name = string("matrix_bd_5")]; tensor var_348_shape = shape(x = matrix_bd_5)[name = string("op_348_shape")]; int32 gather_21 = const()[name = string("gather_21"), val = int32(1)]; int32 gather_22 = const()[name = string("gather_22"), val = int32(8)]; int32 gather_23_batch_dims_0 = const()[name = string("gather_23_batch_dims_0"), val = int32(0)]; bool gather_23_validate_indices_0 = const()[name = string("gather_23_validate_indices_0"), val = bool(false)]; int32 select_10 = const()[name = string("select_10"), val = int32(2)]; int32 gather_23_axis_1 = const()[name = string("gather_23_axis_1"), val = int32(0)]; int32 gather_23 = gather(axis = gather_23_axis_1, batch_dims = gather_23_batch_dims_0, indices = select_10, validate_indices = gather_23_validate_indices_0, x = var_348_shape)[name = string("gather_23")]; int32 concat_9_axis_0 = const()[name = string("concat_9_axis_0"), val = int32(0)]; bool concat_9_interleave_0 = const()[name = string("concat_9_interleave_0"), val = bool(false)]; tensor concat_9 = concat(axis = concat_9_axis_0, interleave = concat_9_interleave_0, values = (gather_21, gather_22, gather_23, var_61))[name = string("concat_9")]; fp32 zero_pad_3_value_0 = const()[name = string("zero_pad_3_value_0"), val = fp32(0x0p+0)]; tensor zero_pad_3 = fill(shape = concat_9, value = zero_pad_3_value_0)[name = string("zero_pad_3")]; bool x_padded_5_interleave_0 = const()[name = string("x_padded_5_interleave_0"), val = bool(false)]; tensor x_padded_5 = concat(axis = var_55, interleave = x_padded_5_interleave_0, values = (zero_pad_3, matrix_bd_5))[name = string("x_padded_5")]; int32 gather_24 = const()[name = string("gather_24"), val = int32(1)]; int32 gather_25 = const()[name = string("gather_25"), val = int32(8)]; int32 gather_26_batch_dims_0 = const()[name = string("gather_26_batch_dims_0"), val = int32(0)]; bool gather_26_validate_indices_0 = const()[name = string("gather_26_validate_indices_0"), val = bool(false)]; int32 select_11 = const()[name = string("select_11"), val = int32(3)]; int32 gather_26_axis_1 = const()[name = string("gather_26_axis_1"), val = int32(0)]; int32 gather_26 = gather(axis = gather_26_axis_1, batch_dims = gather_26_batch_dims_0, indices = select_11, validate_indices = gather_26_validate_indices_0, x = var_348_shape)[name = string("gather_26")]; int32 var_359 = const()[name = string("op_359"), val = int32(1)]; int32 var_360 = add(x = gather_26, y = var_359)[name = string("op_360")]; int32 concat_10_axis_0 = const()[name = string("concat_10_axis_0"), val = int32(0)]; bool concat_10_interleave_0 = const()[name = string("concat_10_interleave_0"), val = bool(false)]; tensor concat_10 = concat(axis = concat_10_axis_0, interleave = concat_10_interleave_0, values = (gather_24, gather_25, var_360, gather_23))[name = string("concat_10")]; tensor x_padded_7 = reshape(shape = concat_10, x = x_padded_5)[name = string("x_padded_7")]; tensor var_367_begin_0 = const()[name = string("op_367_begin_0"), val = tensor([0, 0, 1, 0])]; tensor var_367_end_0 = const()[name = string("op_367_end_0"), val = tensor([1, 8, 0, 0])]; tensor var_367_end_mask_0 = const()[name = string("op_367_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_367 = slice_by_index(begin = var_367_begin_0, end = var_367_end_0, end_mask = var_367_end_mask_0, x = x_padded_7)[name = string("op_367")]; int32 gather_28 = const()[name = string("gather_28"), val = int32(1)]; int32 gather_29 = const()[name = string("gather_29"), val = int32(8)]; int32 concat_11_axis_0 = const()[name = string("concat_11_axis_0"), val = int32(0)]; bool concat_11_interleave_0 = const()[name = string("concat_11_interleave_0"), val = bool(false)]; tensor concat_11 = concat(axis = concat_11_axis_0, interleave = concat_11_interleave_0, values = (gather_28, gather_29, gather_23, gather_26))[name = string("concat_11")]; tensor var_373 = reshape(shape = concat_11, x = var_367)[name = string("op_373")]; int32 floor_div_3 = floor_div(x = gather_26, y = var_53)[name = string("floor_div_3")]; string var_376_dtype_0 = const()[name = string("op_376_dtype_0"), val = string("fp32")]; fp32 var_377_promoted = const()[name = string("op_377_promoted"), val = fp32(0x1p+0)]; fp32 var_376 = cast(dtype = var_376_dtype_0, x = floor_div_3)[name = string("cast_124")]; fp32 var_378 = add(x = var_376, y = var_377_promoted)[name = string("op_378")]; string var_379_dtype_0 = const()[name = string("op_379_dtype_0"), val = string("int32")]; int32 concat_12_values0_0 = const()[name = string("concat_12_values0_0"), val = int32(1)]; int32 concat_12_values1_0 = const()[name = string("concat_12_values1_0"), val = int32(8)]; int32 concat_12_values2_0 = const()[name = string("concat_12_values2_0"), val = int32(0)]; int32 concat_12_axis_0 = const()[name = string("concat_12_axis_0"), val = int32(0)]; bool concat_12_interleave_0 = const()[name = string("concat_12_interleave_0"), val = bool(false)]; int32 var_379 = cast(dtype = var_379_dtype_0, x = var_378)[name = string("cast_123")]; tensor concat_12 = concat(axis = concat_12_axis_0, interleave = concat_12_interleave_0, values = (concat_12_values0_0, concat_12_values1_0, concat_12_values2_0, var_379))[name = string("concat_12")]; tensor matrix_bd_7_begin_0 = const()[name = string("matrix_bd_7_begin_0"), val = tensor([0, 0, 0, 0])]; tensor matrix_bd_7_end_mask_0 = const()[name = string("matrix_bd_7_end_mask_0"), val = tensor([true, true, true, false])]; tensor matrix_bd_7 = slice_by_index(begin = matrix_bd_7_begin_0, end = concat_12, end_mask = matrix_bd_7_end_mask_0, x = var_373)[name = string("matrix_bd_7")]; tensor var_384 = add(x = matrix_ac_3, y = matrix_bd_7)[name = string("op_384")]; fp32 _inversed_scores_5_y_0 = const()[name = string("_inversed_scores_5_y_0"), val = fp32(0x1p-3)]; tensor _inversed_scores_5 = mul(x = var_384, y = _inversed_scores_5_y_0)[name = string("_inversed_scores_5")]; tensor var_390_shape = shape(x = _inversed_scores_5)[name = string("op_390_shape")]; int32 gather_34_batch_dims_0 = const()[name = string("gather_34_batch_dims_0"), val = int32(0)]; bool gather_34_validate_indices_0 = const()[name = string("gather_34_validate_indices_0"), val = bool(false)]; int32 select_13 = const()[name = string("select_13"), val = int32(3)]; int32 gather_34_axis_1 = const()[name = string("gather_34_axis_1"), val = int32(0)]; int32 gather_34 = gather(axis = gather_34_axis_1, batch_dims = gather_34_batch_dims_0, indices = select_13, validate_indices = gather_34_validate_indices_0, x = var_390_shape)[name = string("gather_34")]; int32 concat_13_values0_0 = const()[name = string("concat_13_values0_0"), val = int32(0)]; int32 concat_13_values1_0 = const()[name = string("concat_13_values1_0"), val = int32(1)]; int32 concat_13_values2_0 = const()[name = string("concat_13_values2_0"), val = int32(1)]; int32 concat_13_axis_0 = const()[name = string("concat_13_axis_0"), val = int32(0)]; bool concat_13_interleave_0 = const()[name = string("concat_13_interleave_0"), val = bool(false)]; tensor concat_13 = concat(axis = concat_13_axis_0, interleave = concat_13_interleave_0, values = (concat_13_values0_0, concat_13_values1_0, concat_13_values2_0, gather_34))[name = string("concat_13")]; tensor mask_9_begin_0 = const()[name = string("mask_9_begin_0"), val = tensor([0, 0, 0, 0])]; tensor mask_9_end_mask_0 = const()[name = string("mask_9_end_mask_0"), val = tensor([true, true, true, false])]; tensor mask_9 = slice_by_index(begin = mask_9_begin_0, end = concat_13, end_mask = mask_9_end_mask_0, x = mask_3)[name = string("mask_9")]; tensor scores_7 = select(a = var_37, b = _inversed_scores_5, cond = mask_9)[name = string("scores_7")]; tensor var_396 = softmax(axis = var_55, x = scores_7)[name = string("op_396")]; tensor input_41 = select(a = var_46, b = var_396, cond = mask_9)[name = string("input_41")]; bool x_7_transpose_x_0 = const()[name = string("x_7_transpose_x_0"), val = bool(false)]; bool x_7_transpose_y_0 = const()[name = string("x_7_transpose_y_0"), val = bool(false)]; tensor v_7 = transpose(perm = v_7_perm_0, x = v_5)[name = string("transpose_156")]; tensor x_7 = matmul(transpose_x = x_7_transpose_x_0, transpose_y = x_7_transpose_y_0, x = input_41, y = v_7)[name = string("x_7")]; tensor var_400_perm_0 = const()[name = string("op_400_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_402 = const()[name = string("op_402"), val = tensor([1, -1, 512])]; tensor var_400 = transpose(perm = var_400_perm_0, x = x_7)[name = string("transpose_151")]; tensor input_43 = reshape(shape = var_402, x = var_400)[name = string("input_43")]; tensor input_45 = linear(bias = encoder_encoders_1_self_attn_linear_out_bias, weight = encoder_encoders_1_self_attn_linear_out_weight, x = input_43)[name = string("linear_12")]; tensor input_47 = add(x = input_39, y = input_45)[name = string("input_47")]; tensor input_49_axes_0 = const()[name = string("input_49_axes_0"), val = tensor([-1])]; tensor input_49 = layer_norm(axes = input_49_axes_0, beta = encoder_encoders_1_norm_ff_bias, epsilon = var_36, gamma = encoder_encoders_1_norm_ff_weight, x = input_47)[name = string("input_49")]; tensor input_51 = linear(bias = encoder_encoders_1_feed_forward_w_1_bias, weight = encoder_encoders_1_feed_forward_w_1_weight, x = input_49)[name = string("linear_13")]; tensor input_53 = silu(x = input_51)[name = string("input_53")]; tensor input_57 = linear(bias = encoder_encoders_1_feed_forward_w_2_bias, weight = encoder_encoders_1_feed_forward_w_2_weight, x = input_53)[name = string("linear_14")]; tensor input_59 = add(x = input_47, y = input_57)[name = string("input_59")]; tensor query_5_axes_0 = const()[name = string("query_5_axes_0"), val = tensor([-1])]; tensor query_5 = layer_norm(axes = query_5_axes_0, beta = encoder_encoders_2_norm_mha_bias, epsilon = var_36, gamma = encoder_encoders_2_norm_mha_weight, x = input_59)[name = string("query_5")]; tensor var_445 = linear(bias = encoder_encoders_2_self_attn_linear_q_bias, weight = encoder_encoders_2_self_attn_linear_q_weight, x = query_5)[name = string("linear_15")]; tensor var_446 = const()[name = string("op_446"), val = tensor([1, -1, 8, 64])]; tensor q_13 = reshape(shape = var_446, x = var_445)[name = string("q_13")]; tensor var_450 = linear(bias = encoder_encoders_2_self_attn_linear_k_bias, weight = encoder_encoders_2_self_attn_linear_k_weight, x = query_5)[name = string("linear_16")]; tensor var_451 = const()[name = string("op_451"), val = tensor([1, -1, 8, 64])]; tensor k_9 = reshape(shape = var_451, x = var_450)[name = string("k_9")]; tensor var_455 = linear(bias = encoder_encoders_2_self_attn_linear_v_bias, weight = encoder_encoders_2_self_attn_linear_v_weight, x = query_5)[name = string("linear_17")]; tensor var_456 = const()[name = string("op_456"), val = tensor([1, -1, 8, 64])]; tensor v_9 = reshape(shape = var_456, x = var_455)[name = string("v_9")]; tensor v_11_perm_0 = const()[name = string("v_11_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_464 = linear(bias = linear_4_bias_0, weight = encoder_encoders_2_self_attn_linear_pos_weight, x = input_7)[name = string("linear_18")]; tensor var_465 = const()[name = string("op_465"), val = tensor([1, -1, 8, 64])]; tensor p_9 = reshape(shape = var_465, x = var_464)[name = string("p_9")]; tensor const_12 = const()[name = string("const_12"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185385536)))]; tensor var_469 = add(x = q_13, y = const_12)[name = string("op_469")]; tensor const_13 = const()[name = string("const_13"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185387648)))]; tensor var_472 = add(x = q_13, y = const_13)[name = string("op_472")]; bool matrix_ac_5_transpose_x_0 = const()[name = string("matrix_ac_5_transpose_x_0"), val = bool(false)]; bool matrix_ac_5_transpose_y_0 = const()[name = string("matrix_ac_5_transpose_y_0"), val = bool(false)]; tensor transpose_68_perm_0 = const()[name = string("transpose_68_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_69_perm_0 = const()[name = string("transpose_69_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_69 = transpose(perm = transpose_69_perm_0, x = k_9)[name = string("transpose_148")]; tensor transpose_68 = transpose(perm = transpose_68_perm_0, x = var_469)[name = string("transpose_149")]; tensor matrix_ac_5 = matmul(transpose_x = matrix_ac_5_transpose_x_0, transpose_y = matrix_ac_5_transpose_y_0, x = transpose_68, y = transpose_69)[name = string("matrix_ac_5")]; bool matrix_bd_9_transpose_x_0 = const()[name = string("matrix_bd_9_transpose_x_0"), val = bool(false)]; bool matrix_bd_9_transpose_y_0 = const()[name = string("matrix_bd_9_transpose_y_0"), val = bool(false)]; tensor transpose_70_perm_0 = const()[name = string("transpose_70_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_71_perm_0 = const()[name = string("transpose_71_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_71 = transpose(perm = transpose_71_perm_0, x = p_9)[name = string("transpose_146")]; tensor transpose_70 = transpose(perm = transpose_70_perm_0, x = var_472)[name = string("transpose_147")]; tensor matrix_bd_9 = matmul(transpose_x = matrix_bd_9_transpose_x_0, transpose_y = matrix_bd_9_transpose_y_0, x = transpose_70, y = transpose_71)[name = string("matrix_bd_9")]; tensor var_478_shape = shape(x = matrix_bd_9)[name = string("op_478_shape")]; int32 gather_37 = const()[name = string("gather_37"), val = int32(1)]; int32 gather_38 = const()[name = string("gather_38"), val = int32(8)]; int32 gather_39_batch_dims_0 = const()[name = string("gather_39_batch_dims_0"), val = int32(0)]; bool gather_39_validate_indices_0 = const()[name = string("gather_39_validate_indices_0"), val = bool(false)]; int32 select_14 = const()[name = string("select_14"), val = int32(2)]; int32 gather_39_axis_1 = const()[name = string("gather_39_axis_1"), val = int32(0)]; int32 gather_39 = gather(axis = gather_39_axis_1, batch_dims = gather_39_batch_dims_0, indices = select_14, validate_indices = gather_39_validate_indices_0, x = var_478_shape)[name = string("gather_39")]; int32 concat_14_axis_0 = const()[name = string("concat_14_axis_0"), val = int32(0)]; bool concat_14_interleave_0 = const()[name = string("concat_14_interleave_0"), val = bool(false)]; tensor concat_14 = concat(axis = concat_14_axis_0, interleave = concat_14_interleave_0, values = (gather_37, gather_38, gather_39, var_61))[name = string("concat_14")]; fp32 zero_pad_5_value_0 = const()[name = string("zero_pad_5_value_0"), val = fp32(0x0p+0)]; tensor zero_pad_5 = fill(shape = concat_14, value = zero_pad_5_value_0)[name = string("zero_pad_5")]; bool x_padded_9_interleave_0 = const()[name = string("x_padded_9_interleave_0"), val = bool(false)]; tensor x_padded_9 = concat(axis = var_55, interleave = x_padded_9_interleave_0, values = (zero_pad_5, matrix_bd_9))[name = string("x_padded_9")]; int32 gather_40 = const()[name = string("gather_40"), val = int32(1)]; int32 gather_41 = const()[name = string("gather_41"), val = int32(8)]; int32 gather_42_batch_dims_0 = const()[name = string("gather_42_batch_dims_0"), val = int32(0)]; bool gather_42_validate_indices_0 = const()[name = string("gather_42_validate_indices_0"), val = bool(false)]; int32 select_15 = const()[name = string("select_15"), val = int32(3)]; int32 gather_42_axis_1 = const()[name = string("gather_42_axis_1"), val = int32(0)]; int32 gather_42 = gather(axis = gather_42_axis_1, batch_dims = gather_42_batch_dims_0, indices = select_15, validate_indices = gather_42_validate_indices_0, x = var_478_shape)[name = string("gather_42")]; int32 var_489 = const()[name = string("op_489"), val = int32(1)]; int32 var_490 = add(x = gather_42, y = var_489)[name = string("op_490")]; int32 concat_15_axis_0 = const()[name = string("concat_15_axis_0"), val = int32(0)]; bool concat_15_interleave_0 = const()[name = string("concat_15_interleave_0"), val = bool(false)]; tensor concat_15 = concat(axis = concat_15_axis_0, interleave = concat_15_interleave_0, values = (gather_40, gather_41, var_490, gather_39))[name = string("concat_15")]; tensor x_padded_11 = reshape(shape = concat_15, x = x_padded_9)[name = string("x_padded_11")]; tensor var_497_begin_0 = const()[name = string("op_497_begin_0"), val = tensor([0, 0, 1, 0])]; tensor var_497_end_0 = const()[name = string("op_497_end_0"), val = tensor([1, 8, 0, 0])]; tensor var_497_end_mask_0 = const()[name = string("op_497_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_497 = slice_by_index(begin = var_497_begin_0, end = var_497_end_0, end_mask = var_497_end_mask_0, x = x_padded_11)[name = string("op_497")]; int32 gather_44 = const()[name = string("gather_44"), val = int32(1)]; int32 gather_45 = const()[name = string("gather_45"), val = int32(8)]; int32 concat_16_axis_0 = const()[name = string("concat_16_axis_0"), val = int32(0)]; bool concat_16_interleave_0 = const()[name = string("concat_16_interleave_0"), val = bool(false)]; tensor concat_16 = concat(axis = concat_16_axis_0, interleave = concat_16_interleave_0, values = (gather_44, gather_45, gather_39, gather_42))[name = string("concat_16")]; tensor var_503 = reshape(shape = concat_16, x = var_497)[name = string("op_503")]; int32 floor_div_4 = floor_div(x = gather_42, y = var_53)[name = string("floor_div_4")]; string var_506_dtype_0 = const()[name = string("op_506_dtype_0"), val = string("fp32")]; fp32 var_507_promoted = const()[name = string("op_507_promoted"), val = fp32(0x1p+0)]; fp32 var_506 = cast(dtype = var_506_dtype_0, x = floor_div_4)[name = string("cast_122")]; fp32 var_508 = add(x = var_506, y = var_507_promoted)[name = string("op_508")]; string var_509_dtype_0 = const()[name = string("op_509_dtype_0"), val = string("int32")]; int32 concat_17_values0_0 = const()[name = string("concat_17_values0_0"), val = int32(1)]; int32 concat_17_values1_0 = const()[name = string("concat_17_values1_0"), val = int32(8)]; int32 concat_17_values2_0 = const()[name = string("concat_17_values2_0"), val = int32(0)]; int32 concat_17_axis_0 = const()[name = string("concat_17_axis_0"), val = int32(0)]; bool concat_17_interleave_0 = const()[name = string("concat_17_interleave_0"), val = bool(false)]; int32 var_509 = cast(dtype = var_509_dtype_0, x = var_508)[name = string("cast_121")]; tensor concat_17 = concat(axis = concat_17_axis_0, interleave = concat_17_interleave_0, values = (concat_17_values0_0, concat_17_values1_0, concat_17_values2_0, var_509))[name = string("concat_17")]; tensor matrix_bd_11_begin_0 = const()[name = string("matrix_bd_11_begin_0"), val = tensor([0, 0, 0, 0])]; tensor matrix_bd_11_end_mask_0 = const()[name = string("matrix_bd_11_end_mask_0"), val = tensor([true, true, true, false])]; tensor matrix_bd_11 = slice_by_index(begin = matrix_bd_11_begin_0, end = concat_17, end_mask = matrix_bd_11_end_mask_0, x = var_503)[name = string("matrix_bd_11")]; tensor var_514 = add(x = matrix_ac_5, y = matrix_bd_11)[name = string("op_514")]; fp32 _inversed_scores_9_y_0 = const()[name = string("_inversed_scores_9_y_0"), val = fp32(0x1p-3)]; tensor _inversed_scores_9 = mul(x = var_514, y = _inversed_scores_9_y_0)[name = string("_inversed_scores_9")]; tensor var_520_shape = shape(x = _inversed_scores_9)[name = string("op_520_shape")]; int32 gather_50_batch_dims_0 = const()[name = string("gather_50_batch_dims_0"), val = int32(0)]; bool gather_50_validate_indices_0 = const()[name = string("gather_50_validate_indices_0"), val = bool(false)]; int32 select_17 = const()[name = string("select_17"), val = int32(3)]; int32 gather_50_axis_1 = const()[name = string("gather_50_axis_1"), val = int32(0)]; int32 gather_50 = gather(axis = gather_50_axis_1, batch_dims = gather_50_batch_dims_0, indices = select_17, validate_indices = gather_50_validate_indices_0, x = var_520_shape)[name = string("gather_50")]; int32 concat_18_values0_0 = const()[name = string("concat_18_values0_0"), val = int32(0)]; int32 concat_18_values1_0 = const()[name = string("concat_18_values1_0"), val = int32(1)]; int32 concat_18_values2_0 = const()[name = string("concat_18_values2_0"), val = int32(1)]; int32 concat_18_axis_0 = const()[name = string("concat_18_axis_0"), val = int32(0)]; bool concat_18_interleave_0 = const()[name = string("concat_18_interleave_0"), val = bool(false)]; tensor concat_18 = concat(axis = concat_18_axis_0, interleave = concat_18_interleave_0, values = (concat_18_values0_0, concat_18_values1_0, concat_18_values2_0, gather_50))[name = string("concat_18")]; tensor mask_13_begin_0 = const()[name = string("mask_13_begin_0"), val = tensor([0, 0, 0, 0])]; tensor mask_13_end_mask_0 = const()[name = string("mask_13_end_mask_0"), val = tensor([true, true, true, false])]; tensor mask_13 = slice_by_index(begin = mask_13_begin_0, end = concat_18, end_mask = mask_13_end_mask_0, x = mask_3)[name = string("mask_13")]; tensor scores_11 = select(a = var_37, b = _inversed_scores_9, cond = mask_13)[name = string("scores_11")]; tensor var_526 = softmax(axis = var_55, x = scores_11)[name = string("op_526")]; tensor input_61 = select(a = var_46, b = var_526, cond = mask_13)[name = string("input_61")]; bool x_9_transpose_x_0 = const()[name = string("x_9_transpose_x_0"), val = bool(false)]; bool x_9_transpose_y_0 = const()[name = string("x_9_transpose_y_0"), val = bool(false)]; tensor v_11 = transpose(perm = v_11_perm_0, x = v_9)[name = string("transpose_150")]; tensor x_9 = matmul(transpose_x = x_9_transpose_x_0, transpose_y = x_9_transpose_y_0, x = input_61, y = v_11)[name = string("x_9")]; tensor var_530_perm_0 = const()[name = string("op_530_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_532 = const()[name = string("op_532"), val = tensor([1, -1, 512])]; tensor var_530 = transpose(perm = var_530_perm_0, x = x_9)[name = string("transpose_145")]; tensor input_63 = reshape(shape = var_532, x = var_530)[name = string("input_63")]; tensor input_65 = linear(bias = encoder_encoders_2_self_attn_linear_out_bias, weight = encoder_encoders_2_self_attn_linear_out_weight, x = input_63)[name = string("linear_19")]; tensor input_67 = add(x = input_59, y = input_65)[name = string("input_67")]; tensor input_69_axes_0 = const()[name = string("input_69_axes_0"), val = tensor([-1])]; tensor input_69 = layer_norm(axes = input_69_axes_0, beta = encoder_encoders_2_norm_ff_bias, epsilon = var_36, gamma = encoder_encoders_2_norm_ff_weight, x = input_67)[name = string("input_69")]; tensor input_71 = linear(bias = encoder_encoders_2_feed_forward_w_1_bias, weight = encoder_encoders_2_feed_forward_w_1_weight, x = input_69)[name = string("linear_20")]; tensor input_73 = silu(x = input_71)[name = string("input_73")]; tensor input_77 = linear(bias = encoder_encoders_2_feed_forward_w_2_bias, weight = encoder_encoders_2_feed_forward_w_2_weight, x = input_73)[name = string("linear_21")]; tensor input_79 = add(x = input_67, y = input_77)[name = string("input_79")]; tensor query_7_axes_0 = const()[name = string("query_7_axes_0"), val = tensor([-1])]; tensor query_7 = layer_norm(axes = query_7_axes_0, beta = encoder_encoders_3_norm_mha_bias, epsilon = var_36, gamma = encoder_encoders_3_norm_mha_weight, x = input_79)[name = string("query_7")]; tensor var_575 = linear(bias = encoder_encoders_3_self_attn_linear_q_bias, weight = encoder_encoders_3_self_attn_linear_q_weight, x = query_7)[name = string("linear_22")]; tensor var_576 = const()[name = string("op_576"), val = tensor([1, -1, 8, 64])]; tensor q_19 = reshape(shape = var_576, x = var_575)[name = string("q_19")]; tensor var_580 = linear(bias = encoder_encoders_3_self_attn_linear_k_bias, weight = encoder_encoders_3_self_attn_linear_k_weight, x = query_7)[name = string("linear_23")]; tensor var_581 = const()[name = string("op_581"), val = tensor([1, -1, 8, 64])]; tensor k_13 = reshape(shape = var_581, x = var_580)[name = string("k_13")]; tensor var_585 = linear(bias = encoder_encoders_3_self_attn_linear_v_bias, weight = encoder_encoders_3_self_attn_linear_v_weight, x = query_7)[name = string("linear_24")]; tensor var_586 = const()[name = string("op_586"), val = tensor([1, -1, 8, 64])]; tensor v_13 = reshape(shape = var_586, x = var_585)[name = string("v_13")]; tensor v_15_perm_0 = const()[name = string("v_15_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_594 = linear(bias = linear_4_bias_0, weight = encoder_encoders_3_self_attn_linear_pos_weight, x = input_7)[name = string("linear_25")]; tensor var_595 = const()[name = string("op_595"), val = tensor([1, -1, 8, 64])]; tensor p_13 = reshape(shape = var_595, x = var_594)[name = string("p_13")]; tensor const_14 = const()[name = string("const_14"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185389760)))]; tensor var_599 = add(x = q_19, y = const_14)[name = string("op_599")]; tensor const_15 = const()[name = string("const_15"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185391872)))]; tensor var_602 = add(x = q_19, y = const_15)[name = string("op_602")]; bool matrix_ac_7_transpose_x_0 = const()[name = string("matrix_ac_7_transpose_x_0"), val = bool(false)]; bool matrix_ac_7_transpose_y_0 = const()[name = string("matrix_ac_7_transpose_y_0"), val = bool(false)]; tensor transpose_72_perm_0 = const()[name = string("transpose_72_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_73_perm_0 = const()[name = string("transpose_73_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_73 = transpose(perm = transpose_73_perm_0, x = k_13)[name = string("transpose_142")]; tensor transpose_72 = transpose(perm = transpose_72_perm_0, x = var_599)[name = string("transpose_143")]; tensor matrix_ac_7 = matmul(transpose_x = matrix_ac_7_transpose_x_0, transpose_y = matrix_ac_7_transpose_y_0, x = transpose_72, y = transpose_73)[name = string("matrix_ac_7")]; bool matrix_bd_13_transpose_x_0 = const()[name = string("matrix_bd_13_transpose_x_0"), val = bool(false)]; bool matrix_bd_13_transpose_y_0 = const()[name = string("matrix_bd_13_transpose_y_0"), val = bool(false)]; tensor transpose_74_perm_0 = const()[name = string("transpose_74_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_75_perm_0 = const()[name = string("transpose_75_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_75 = transpose(perm = transpose_75_perm_0, x = p_13)[name = string("transpose_140")]; tensor transpose_74 = transpose(perm = transpose_74_perm_0, x = var_602)[name = string("transpose_141")]; tensor matrix_bd_13 = matmul(transpose_x = matrix_bd_13_transpose_x_0, transpose_y = matrix_bd_13_transpose_y_0, x = transpose_74, y = transpose_75)[name = string("matrix_bd_13")]; tensor var_608_shape = shape(x = matrix_bd_13)[name = string("op_608_shape")]; int32 gather_53 = const()[name = string("gather_53"), val = int32(1)]; int32 gather_54 = const()[name = string("gather_54"), val = int32(8)]; int32 gather_55_batch_dims_0 = const()[name = string("gather_55_batch_dims_0"), val = int32(0)]; bool gather_55_validate_indices_0 = const()[name = string("gather_55_validate_indices_0"), val = bool(false)]; int32 select_18 = const()[name = string("select_18"), val = int32(2)]; int32 gather_55_axis_1 = const()[name = string("gather_55_axis_1"), val = int32(0)]; int32 gather_55 = gather(axis = gather_55_axis_1, batch_dims = gather_55_batch_dims_0, indices = select_18, validate_indices = gather_55_validate_indices_0, x = var_608_shape)[name = string("gather_55")]; int32 concat_19_axis_0 = const()[name = string("concat_19_axis_0"), val = int32(0)]; bool concat_19_interleave_0 = const()[name = string("concat_19_interleave_0"), val = bool(false)]; tensor concat_19 = concat(axis = concat_19_axis_0, interleave = concat_19_interleave_0, values = (gather_53, gather_54, gather_55, var_61))[name = string("concat_19")]; fp32 zero_pad_7_value_0 = const()[name = string("zero_pad_7_value_0"), val = fp32(0x0p+0)]; tensor zero_pad_7 = fill(shape = concat_19, value = zero_pad_7_value_0)[name = string("zero_pad_7")]; bool x_padded_13_interleave_0 = const()[name = string("x_padded_13_interleave_0"), val = bool(false)]; tensor x_padded_13 = concat(axis = var_55, interleave = x_padded_13_interleave_0, values = (zero_pad_7, matrix_bd_13))[name = string("x_padded_13")]; int32 gather_56 = const()[name = string("gather_56"), val = int32(1)]; int32 gather_57 = const()[name = string("gather_57"), val = int32(8)]; int32 gather_58_batch_dims_0 = const()[name = string("gather_58_batch_dims_0"), val = int32(0)]; bool gather_58_validate_indices_0 = const()[name = string("gather_58_validate_indices_0"), val = bool(false)]; int32 select_19 = const()[name = string("select_19"), val = int32(3)]; int32 gather_58_axis_1 = const()[name = string("gather_58_axis_1"), val = int32(0)]; int32 gather_58 = gather(axis = gather_58_axis_1, batch_dims = gather_58_batch_dims_0, indices = select_19, validate_indices = gather_58_validate_indices_0, x = var_608_shape)[name = string("gather_58")]; int32 var_619 = const()[name = string("op_619"), val = int32(1)]; int32 var_620 = add(x = gather_58, y = var_619)[name = string("op_620")]; int32 concat_20_axis_0 = const()[name = string("concat_20_axis_0"), val = int32(0)]; bool concat_20_interleave_0 = const()[name = string("concat_20_interleave_0"), val = bool(false)]; tensor concat_20 = concat(axis = concat_20_axis_0, interleave = concat_20_interleave_0, values = (gather_56, gather_57, var_620, gather_55))[name = string("concat_20")]; tensor x_padded_15 = reshape(shape = concat_20, x = x_padded_13)[name = string("x_padded_15")]; tensor var_627_begin_0 = const()[name = string("op_627_begin_0"), val = tensor([0, 0, 1, 0])]; tensor var_627_end_0 = const()[name = string("op_627_end_0"), val = tensor([1, 8, 0, 0])]; tensor var_627_end_mask_0 = const()[name = string("op_627_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_627 = slice_by_index(begin = var_627_begin_0, end = var_627_end_0, end_mask = var_627_end_mask_0, x = x_padded_15)[name = string("op_627")]; int32 gather_60 = const()[name = string("gather_60"), val = int32(1)]; int32 gather_61 = const()[name = string("gather_61"), val = int32(8)]; int32 concat_21_axis_0 = const()[name = string("concat_21_axis_0"), val = int32(0)]; bool concat_21_interleave_0 = const()[name = string("concat_21_interleave_0"), val = bool(false)]; tensor concat_21 = concat(axis = concat_21_axis_0, interleave = concat_21_interleave_0, values = (gather_60, gather_61, gather_55, gather_58))[name = string("concat_21")]; tensor var_633 = reshape(shape = concat_21, x = var_627)[name = string("op_633")]; int32 floor_div_5 = floor_div(x = gather_58, y = var_53)[name = string("floor_div_5")]; string var_636_dtype_0 = const()[name = string("op_636_dtype_0"), val = string("fp32")]; fp32 var_637_promoted = const()[name = string("op_637_promoted"), val = fp32(0x1p+0)]; fp32 var_636 = cast(dtype = var_636_dtype_0, x = floor_div_5)[name = string("cast_120")]; fp32 var_638 = add(x = var_636, y = var_637_promoted)[name = string("op_638")]; string var_639_dtype_0 = const()[name = string("op_639_dtype_0"), val = string("int32")]; int32 concat_22_values0_0 = const()[name = string("concat_22_values0_0"), val = int32(1)]; int32 concat_22_values1_0 = const()[name = string("concat_22_values1_0"), val = int32(8)]; int32 concat_22_values2_0 = const()[name = string("concat_22_values2_0"), val = int32(0)]; int32 concat_22_axis_0 = const()[name = string("concat_22_axis_0"), val = int32(0)]; bool concat_22_interleave_0 = const()[name = string("concat_22_interleave_0"), val = bool(false)]; int32 var_639 = cast(dtype = var_639_dtype_0, x = var_638)[name = string("cast_119")]; tensor concat_22 = concat(axis = concat_22_axis_0, interleave = concat_22_interleave_0, values = (concat_22_values0_0, concat_22_values1_0, concat_22_values2_0, var_639))[name = string("concat_22")]; tensor matrix_bd_15_begin_0 = const()[name = string("matrix_bd_15_begin_0"), val = tensor([0, 0, 0, 0])]; tensor matrix_bd_15_end_mask_0 = const()[name = string("matrix_bd_15_end_mask_0"), val = tensor([true, true, true, false])]; tensor matrix_bd_15 = slice_by_index(begin = matrix_bd_15_begin_0, end = concat_22, end_mask = matrix_bd_15_end_mask_0, x = var_633)[name = string("matrix_bd_15")]; tensor var_644 = add(x = matrix_ac_7, y = matrix_bd_15)[name = string("op_644")]; fp32 _inversed_scores_13_y_0 = const()[name = string("_inversed_scores_13_y_0"), val = fp32(0x1p-3)]; tensor _inversed_scores_13 = mul(x = var_644, y = _inversed_scores_13_y_0)[name = string("_inversed_scores_13")]; tensor var_650_shape = shape(x = _inversed_scores_13)[name = string("op_650_shape")]; int32 gather_66_batch_dims_0 = const()[name = string("gather_66_batch_dims_0"), val = int32(0)]; bool gather_66_validate_indices_0 = const()[name = string("gather_66_validate_indices_0"), val = bool(false)]; int32 select_21 = const()[name = string("select_21"), val = int32(3)]; int32 gather_66_axis_1 = const()[name = string("gather_66_axis_1"), val = int32(0)]; int32 gather_66 = gather(axis = gather_66_axis_1, batch_dims = gather_66_batch_dims_0, indices = select_21, validate_indices = gather_66_validate_indices_0, x = var_650_shape)[name = string("gather_66")]; int32 concat_23_values0_0 = const()[name = string("concat_23_values0_0"), val = int32(0)]; int32 concat_23_values1_0 = const()[name = string("concat_23_values1_0"), val = int32(1)]; int32 concat_23_values2_0 = const()[name = string("concat_23_values2_0"), val = int32(1)]; int32 concat_23_axis_0 = const()[name = string("concat_23_axis_0"), val = int32(0)]; bool concat_23_interleave_0 = const()[name = string("concat_23_interleave_0"), val = bool(false)]; tensor concat_23 = concat(axis = concat_23_axis_0, interleave = concat_23_interleave_0, values = (concat_23_values0_0, concat_23_values1_0, concat_23_values2_0, gather_66))[name = string("concat_23")]; tensor mask_17_begin_0 = const()[name = string("mask_17_begin_0"), val = tensor([0, 0, 0, 0])]; tensor mask_17_end_mask_0 = const()[name = string("mask_17_end_mask_0"), val = tensor([true, true, true, false])]; tensor mask_17 = slice_by_index(begin = mask_17_begin_0, end = concat_23, end_mask = mask_17_end_mask_0, x = mask_3)[name = string("mask_17")]; tensor scores_15 = select(a = var_37, b = _inversed_scores_13, cond = mask_17)[name = string("scores_15")]; tensor var_656 = softmax(axis = var_55, x = scores_15)[name = string("op_656")]; tensor input_81 = select(a = var_46, b = var_656, cond = mask_17)[name = string("input_81")]; bool x_11_transpose_x_0 = const()[name = string("x_11_transpose_x_0"), val = bool(false)]; bool x_11_transpose_y_0 = const()[name = string("x_11_transpose_y_0"), val = bool(false)]; tensor v_15 = transpose(perm = v_15_perm_0, x = v_13)[name = string("transpose_144")]; tensor x_11 = matmul(transpose_x = x_11_transpose_x_0, transpose_y = x_11_transpose_y_0, x = input_81, y = v_15)[name = string("x_11")]; tensor var_660_perm_0 = const()[name = string("op_660_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_662 = const()[name = string("op_662"), val = tensor([1, -1, 512])]; tensor var_660 = transpose(perm = var_660_perm_0, x = x_11)[name = string("transpose_139")]; tensor input_83 = reshape(shape = var_662, x = var_660)[name = string("input_83")]; tensor input_85 = linear(bias = encoder_encoders_3_self_attn_linear_out_bias, weight = encoder_encoders_3_self_attn_linear_out_weight, x = input_83)[name = string("linear_26")]; tensor input_87 = add(x = input_79, y = input_85)[name = string("input_87")]; tensor input_89_axes_0 = const()[name = string("input_89_axes_0"), val = tensor([-1])]; tensor input_89 = layer_norm(axes = input_89_axes_0, beta = encoder_encoders_3_norm_ff_bias, epsilon = var_36, gamma = encoder_encoders_3_norm_ff_weight, x = input_87)[name = string("input_89")]; tensor input_91 = linear(bias = encoder_encoders_3_feed_forward_w_1_bias, weight = encoder_encoders_3_feed_forward_w_1_weight, x = input_89)[name = string("linear_27")]; tensor input_93 = silu(x = input_91)[name = string("input_93")]; tensor input_97 = linear(bias = encoder_encoders_3_feed_forward_w_2_bias, weight = encoder_encoders_3_feed_forward_w_2_weight, x = input_93)[name = string("linear_28")]; tensor input_99 = add(x = input_87, y = input_97)[name = string("input_99")]; tensor query_9_axes_0 = const()[name = string("query_9_axes_0"), val = tensor([-1])]; tensor query_9 = layer_norm(axes = query_9_axes_0, beta = encoder_encoders_4_norm_mha_bias, epsilon = var_36, gamma = encoder_encoders_4_norm_mha_weight, x = input_99)[name = string("query_9")]; tensor var_705 = linear(bias = encoder_encoders_4_self_attn_linear_q_bias, weight = encoder_encoders_4_self_attn_linear_q_weight, x = query_9)[name = string("linear_29")]; tensor var_706 = const()[name = string("op_706"), val = tensor([1, -1, 8, 64])]; tensor q_25 = reshape(shape = var_706, x = var_705)[name = string("q_25")]; tensor var_710 = linear(bias = encoder_encoders_4_self_attn_linear_k_bias, weight = encoder_encoders_4_self_attn_linear_k_weight, x = query_9)[name = string("linear_30")]; tensor var_711 = const()[name = string("op_711"), val = tensor([1, -1, 8, 64])]; tensor k_17 = reshape(shape = var_711, x = var_710)[name = string("k_17")]; tensor var_715 = linear(bias = encoder_encoders_4_self_attn_linear_v_bias, weight = encoder_encoders_4_self_attn_linear_v_weight, x = query_9)[name = string("linear_31")]; tensor var_716 = const()[name = string("op_716"), val = tensor([1, -1, 8, 64])]; tensor v_17 = reshape(shape = var_716, x = var_715)[name = string("v_17")]; tensor v_19_perm_0 = const()[name = string("v_19_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_724 = linear(bias = linear_4_bias_0, weight = encoder_encoders_4_self_attn_linear_pos_weight, x = input_7)[name = string("linear_32")]; tensor var_725 = const()[name = string("op_725"), val = tensor([1, -1, 8, 64])]; tensor p_17 = reshape(shape = var_725, x = var_724)[name = string("p_17")]; tensor const_16 = const()[name = string("const_16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185393984)))]; tensor var_729 = add(x = q_25, y = const_16)[name = string("op_729")]; tensor const_17 = const()[name = string("const_17"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185396096)))]; tensor var_732 = add(x = q_25, y = const_17)[name = string("op_732")]; bool matrix_ac_9_transpose_x_0 = const()[name = string("matrix_ac_9_transpose_x_0"), val = bool(false)]; bool matrix_ac_9_transpose_y_0 = const()[name = string("matrix_ac_9_transpose_y_0"), val = bool(false)]; tensor transpose_76_perm_0 = const()[name = string("transpose_76_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_77_perm_0 = const()[name = string("transpose_77_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_77 = transpose(perm = transpose_77_perm_0, x = k_17)[name = string("transpose_136")]; tensor transpose_76 = transpose(perm = transpose_76_perm_0, x = var_729)[name = string("transpose_137")]; tensor matrix_ac_9 = matmul(transpose_x = matrix_ac_9_transpose_x_0, transpose_y = matrix_ac_9_transpose_y_0, x = transpose_76, y = transpose_77)[name = string("matrix_ac_9")]; bool matrix_bd_17_transpose_x_0 = const()[name = string("matrix_bd_17_transpose_x_0"), val = bool(false)]; bool matrix_bd_17_transpose_y_0 = const()[name = string("matrix_bd_17_transpose_y_0"), val = bool(false)]; tensor transpose_78_perm_0 = const()[name = string("transpose_78_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_79_perm_0 = const()[name = string("transpose_79_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_79 = transpose(perm = transpose_79_perm_0, x = p_17)[name = string("transpose_134")]; tensor transpose_78 = transpose(perm = transpose_78_perm_0, x = var_732)[name = string("transpose_135")]; tensor matrix_bd_17 = matmul(transpose_x = matrix_bd_17_transpose_x_0, transpose_y = matrix_bd_17_transpose_y_0, x = transpose_78, y = transpose_79)[name = string("matrix_bd_17")]; tensor var_738_shape = shape(x = matrix_bd_17)[name = string("op_738_shape")]; int32 gather_69 = const()[name = string("gather_69"), val = int32(1)]; int32 gather_70 = const()[name = string("gather_70"), val = int32(8)]; int32 gather_71_batch_dims_0 = const()[name = string("gather_71_batch_dims_0"), val = int32(0)]; bool gather_71_validate_indices_0 = const()[name = string("gather_71_validate_indices_0"), val = bool(false)]; int32 select_22 = const()[name = string("select_22"), val = int32(2)]; int32 gather_71_axis_1 = const()[name = string("gather_71_axis_1"), val = int32(0)]; int32 gather_71 = gather(axis = gather_71_axis_1, batch_dims = gather_71_batch_dims_0, indices = select_22, validate_indices = gather_71_validate_indices_0, x = var_738_shape)[name = string("gather_71")]; int32 concat_24_axis_0 = const()[name = string("concat_24_axis_0"), val = int32(0)]; bool concat_24_interleave_0 = const()[name = string("concat_24_interleave_0"), val = bool(false)]; tensor concat_24 = concat(axis = concat_24_axis_0, interleave = concat_24_interleave_0, values = (gather_69, gather_70, gather_71, var_61))[name = string("concat_24")]; fp32 zero_pad_9_value_0 = const()[name = string("zero_pad_9_value_0"), val = fp32(0x0p+0)]; tensor zero_pad_9 = fill(shape = concat_24, value = zero_pad_9_value_0)[name = string("zero_pad_9")]; bool x_padded_17_interleave_0 = const()[name = string("x_padded_17_interleave_0"), val = bool(false)]; tensor x_padded_17 = concat(axis = var_55, interleave = x_padded_17_interleave_0, values = (zero_pad_9, matrix_bd_17))[name = string("x_padded_17")]; int32 gather_72 = const()[name = string("gather_72"), val = int32(1)]; int32 gather_73 = const()[name = string("gather_73"), val = int32(8)]; int32 gather_74_batch_dims_0 = const()[name = string("gather_74_batch_dims_0"), val = int32(0)]; bool gather_74_validate_indices_0 = const()[name = string("gather_74_validate_indices_0"), val = bool(false)]; int32 select_23 = const()[name = string("select_23"), val = int32(3)]; int32 gather_74_axis_1 = const()[name = string("gather_74_axis_1"), val = int32(0)]; int32 gather_74 = gather(axis = gather_74_axis_1, batch_dims = gather_74_batch_dims_0, indices = select_23, validate_indices = gather_74_validate_indices_0, x = var_738_shape)[name = string("gather_74")]; int32 var_749 = const()[name = string("op_749"), val = int32(1)]; int32 var_750 = add(x = gather_74, y = var_749)[name = string("op_750")]; int32 concat_25_axis_0 = const()[name = string("concat_25_axis_0"), val = int32(0)]; bool concat_25_interleave_0 = const()[name = string("concat_25_interleave_0"), val = bool(false)]; tensor concat_25 = concat(axis = concat_25_axis_0, interleave = concat_25_interleave_0, values = (gather_72, gather_73, var_750, gather_71))[name = string("concat_25")]; tensor x_padded_19 = reshape(shape = concat_25, x = x_padded_17)[name = string("x_padded_19")]; tensor var_757_begin_0 = const()[name = string("op_757_begin_0"), val = tensor([0, 0, 1, 0])]; tensor var_757_end_0 = const()[name = string("op_757_end_0"), val = tensor([1, 8, 0, 0])]; tensor var_757_end_mask_0 = const()[name = string("op_757_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_757 = slice_by_index(begin = var_757_begin_0, end = var_757_end_0, end_mask = var_757_end_mask_0, x = x_padded_19)[name = string("op_757")]; int32 gather_76 = const()[name = string("gather_76"), val = int32(1)]; int32 gather_77 = const()[name = string("gather_77"), val = int32(8)]; int32 concat_26_axis_0 = const()[name = string("concat_26_axis_0"), val = int32(0)]; bool concat_26_interleave_0 = const()[name = string("concat_26_interleave_0"), val = bool(false)]; tensor concat_26 = concat(axis = concat_26_axis_0, interleave = concat_26_interleave_0, values = (gather_76, gather_77, gather_71, gather_74))[name = string("concat_26")]; tensor var_763 = reshape(shape = concat_26, x = var_757)[name = string("op_763")]; int32 floor_div_6 = floor_div(x = gather_74, y = var_53)[name = string("floor_div_6")]; string var_766_dtype_0 = const()[name = string("op_766_dtype_0"), val = string("fp32")]; fp32 var_767_promoted = const()[name = string("op_767_promoted"), val = fp32(0x1p+0)]; fp32 var_766 = cast(dtype = var_766_dtype_0, x = floor_div_6)[name = string("cast_118")]; fp32 var_768 = add(x = var_766, y = var_767_promoted)[name = string("op_768")]; string var_769_dtype_0 = const()[name = string("op_769_dtype_0"), val = string("int32")]; int32 concat_27_values0_0 = const()[name = string("concat_27_values0_0"), val = int32(1)]; int32 concat_27_values1_0 = const()[name = string("concat_27_values1_0"), val = int32(8)]; int32 concat_27_values2_0 = const()[name = string("concat_27_values2_0"), val = int32(0)]; int32 concat_27_axis_0 = const()[name = string("concat_27_axis_0"), val = int32(0)]; bool concat_27_interleave_0 = const()[name = string("concat_27_interleave_0"), val = bool(false)]; int32 var_769 = cast(dtype = var_769_dtype_0, x = var_768)[name = string("cast_117")]; tensor concat_27 = concat(axis = concat_27_axis_0, interleave = concat_27_interleave_0, values = (concat_27_values0_0, concat_27_values1_0, concat_27_values2_0, var_769))[name = string("concat_27")]; tensor matrix_bd_19_begin_0 = const()[name = string("matrix_bd_19_begin_0"), val = tensor([0, 0, 0, 0])]; tensor matrix_bd_19_end_mask_0 = const()[name = string("matrix_bd_19_end_mask_0"), val = tensor([true, true, true, false])]; tensor matrix_bd_19 = slice_by_index(begin = matrix_bd_19_begin_0, end = concat_27, end_mask = matrix_bd_19_end_mask_0, x = var_763)[name = string("matrix_bd_19")]; tensor var_774 = add(x = matrix_ac_9, y = matrix_bd_19)[name = string("op_774")]; fp32 _inversed_scores_17_y_0 = const()[name = string("_inversed_scores_17_y_0"), val = fp32(0x1p-3)]; tensor _inversed_scores_17 = mul(x = var_774, y = _inversed_scores_17_y_0)[name = string("_inversed_scores_17")]; tensor var_780_shape = shape(x = _inversed_scores_17)[name = string("op_780_shape")]; int32 gather_82_batch_dims_0 = const()[name = string("gather_82_batch_dims_0"), val = int32(0)]; bool gather_82_validate_indices_0 = const()[name = string("gather_82_validate_indices_0"), val = bool(false)]; int32 select_25 = const()[name = string("select_25"), val = int32(3)]; int32 gather_82_axis_1 = const()[name = string("gather_82_axis_1"), val = int32(0)]; int32 gather_82 = gather(axis = gather_82_axis_1, batch_dims = gather_82_batch_dims_0, indices = select_25, validate_indices = gather_82_validate_indices_0, x = var_780_shape)[name = string("gather_82")]; int32 concat_28_values0_0 = const()[name = string("concat_28_values0_0"), val = int32(0)]; int32 concat_28_values1_0 = const()[name = string("concat_28_values1_0"), val = int32(1)]; int32 concat_28_values2_0 = const()[name = string("concat_28_values2_0"), val = int32(1)]; int32 concat_28_axis_0 = const()[name = string("concat_28_axis_0"), val = int32(0)]; bool concat_28_interleave_0 = const()[name = string("concat_28_interleave_0"), val = bool(false)]; tensor concat_28 = concat(axis = concat_28_axis_0, interleave = concat_28_interleave_0, values = (concat_28_values0_0, concat_28_values1_0, concat_28_values2_0, gather_82))[name = string("concat_28")]; tensor mask_21_begin_0 = const()[name = string("mask_21_begin_0"), val = tensor([0, 0, 0, 0])]; tensor mask_21_end_mask_0 = const()[name = string("mask_21_end_mask_0"), val = tensor([true, true, true, false])]; tensor mask_21 = slice_by_index(begin = mask_21_begin_0, end = concat_28, end_mask = mask_21_end_mask_0, x = mask_3)[name = string("mask_21")]; tensor scores_19 = select(a = var_37, b = _inversed_scores_17, cond = mask_21)[name = string("scores_19")]; tensor var_786 = softmax(axis = var_55, x = scores_19)[name = string("op_786")]; tensor input_101 = select(a = var_46, b = var_786, cond = mask_21)[name = string("input_101")]; bool x_13_transpose_x_0 = const()[name = string("x_13_transpose_x_0"), val = bool(false)]; bool x_13_transpose_y_0 = const()[name = string("x_13_transpose_y_0"), val = bool(false)]; tensor v_19 = transpose(perm = v_19_perm_0, x = v_17)[name = string("transpose_138")]; tensor x_13 = matmul(transpose_x = x_13_transpose_x_0, transpose_y = x_13_transpose_y_0, x = input_101, y = v_19)[name = string("x_13")]; tensor var_790_perm_0 = const()[name = string("op_790_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_792 = const()[name = string("op_792"), val = tensor([1, -1, 512])]; tensor var_790 = transpose(perm = var_790_perm_0, x = x_13)[name = string("transpose_133")]; tensor input_103 = reshape(shape = var_792, x = var_790)[name = string("input_103")]; tensor input_105 = linear(bias = encoder_encoders_4_self_attn_linear_out_bias, weight = encoder_encoders_4_self_attn_linear_out_weight, x = input_103)[name = string("linear_33")]; tensor input_107 = add(x = input_99, y = input_105)[name = string("input_107")]; tensor input_109_axes_0 = const()[name = string("input_109_axes_0"), val = tensor([-1])]; tensor input_109 = layer_norm(axes = input_109_axes_0, beta = encoder_encoders_4_norm_ff_bias, epsilon = var_36, gamma = encoder_encoders_4_norm_ff_weight, x = input_107)[name = string("input_109")]; tensor input_111 = linear(bias = encoder_encoders_4_feed_forward_w_1_bias, weight = encoder_encoders_4_feed_forward_w_1_weight, x = input_109)[name = string("linear_34")]; tensor input_113 = silu(x = input_111)[name = string("input_113")]; tensor input_117 = linear(bias = encoder_encoders_4_feed_forward_w_2_bias, weight = encoder_encoders_4_feed_forward_w_2_weight, x = input_113)[name = string("linear_35")]; tensor input_119 = add(x = input_107, y = input_117)[name = string("input_119")]; tensor query_11_axes_0 = const()[name = string("query_11_axes_0"), val = tensor([-1])]; tensor query_11 = layer_norm(axes = query_11_axes_0, beta = encoder_encoders_5_norm_mha_bias, epsilon = var_36, gamma = encoder_encoders_5_norm_mha_weight, x = input_119)[name = string("query_11")]; tensor var_835 = linear(bias = encoder_encoders_5_self_attn_linear_q_bias, weight = encoder_encoders_5_self_attn_linear_q_weight, x = query_11)[name = string("linear_36")]; tensor var_836 = const()[name = string("op_836"), val = tensor([1, -1, 8, 64])]; tensor q_31 = reshape(shape = var_836, x = var_835)[name = string("q_31")]; tensor var_840 = linear(bias = encoder_encoders_5_self_attn_linear_k_bias, weight = encoder_encoders_5_self_attn_linear_k_weight, x = query_11)[name = string("linear_37")]; tensor var_841 = const()[name = string("op_841"), val = tensor([1, -1, 8, 64])]; tensor k_21 = reshape(shape = var_841, x = var_840)[name = string("k_21")]; tensor var_845 = linear(bias = encoder_encoders_5_self_attn_linear_v_bias, weight = encoder_encoders_5_self_attn_linear_v_weight, x = query_11)[name = string("linear_38")]; tensor var_846 = const()[name = string("op_846"), val = tensor([1, -1, 8, 64])]; tensor v_21 = reshape(shape = var_846, x = var_845)[name = string("v_21")]; tensor v_23_perm_0 = const()[name = string("v_23_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_854 = linear(bias = linear_4_bias_0, weight = encoder_encoders_5_self_attn_linear_pos_weight, x = input_7)[name = string("linear_39")]; tensor var_855 = const()[name = string("op_855"), val = tensor([1, -1, 8, 64])]; tensor p_21 = reshape(shape = var_855, x = var_854)[name = string("p_21")]; tensor const_18 = const()[name = string("const_18"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185398208)))]; tensor var_859 = add(x = q_31, y = const_18)[name = string("op_859")]; tensor const_19 = const()[name = string("const_19"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185400320)))]; tensor var_862 = add(x = q_31, y = const_19)[name = string("op_862")]; bool matrix_ac_11_transpose_x_0 = const()[name = string("matrix_ac_11_transpose_x_0"), val = bool(false)]; bool matrix_ac_11_transpose_y_0 = const()[name = string("matrix_ac_11_transpose_y_0"), val = bool(false)]; tensor transpose_80_perm_0 = const()[name = string("transpose_80_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_81_perm_0 = const()[name = string("transpose_81_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_81 = transpose(perm = transpose_81_perm_0, x = k_21)[name = string("transpose_130")]; tensor transpose_80 = transpose(perm = transpose_80_perm_0, x = var_859)[name = string("transpose_131")]; tensor matrix_ac_11 = matmul(transpose_x = matrix_ac_11_transpose_x_0, transpose_y = matrix_ac_11_transpose_y_0, x = transpose_80, y = transpose_81)[name = string("matrix_ac_11")]; bool matrix_bd_21_transpose_x_0 = const()[name = string("matrix_bd_21_transpose_x_0"), val = bool(false)]; bool matrix_bd_21_transpose_y_0 = const()[name = string("matrix_bd_21_transpose_y_0"), val = bool(false)]; tensor transpose_82_perm_0 = const()[name = string("transpose_82_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_83_perm_0 = const()[name = string("transpose_83_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_83 = transpose(perm = transpose_83_perm_0, x = p_21)[name = string("transpose_128")]; tensor transpose_82 = transpose(perm = transpose_82_perm_0, x = var_862)[name = string("transpose_129")]; tensor matrix_bd_21 = matmul(transpose_x = matrix_bd_21_transpose_x_0, transpose_y = matrix_bd_21_transpose_y_0, x = transpose_82, y = transpose_83)[name = string("matrix_bd_21")]; tensor var_868_shape = shape(x = matrix_bd_21)[name = string("op_868_shape")]; int32 gather_85 = const()[name = string("gather_85"), val = int32(1)]; int32 gather_86 = const()[name = string("gather_86"), val = int32(8)]; int32 gather_87_batch_dims_0 = const()[name = string("gather_87_batch_dims_0"), val = int32(0)]; bool gather_87_validate_indices_0 = const()[name = string("gather_87_validate_indices_0"), val = bool(false)]; int32 select_26 = const()[name = string("select_26"), val = int32(2)]; int32 gather_87_axis_1 = const()[name = string("gather_87_axis_1"), val = int32(0)]; int32 gather_87 = gather(axis = gather_87_axis_1, batch_dims = gather_87_batch_dims_0, indices = select_26, validate_indices = gather_87_validate_indices_0, x = var_868_shape)[name = string("gather_87")]; int32 concat_29_axis_0 = const()[name = string("concat_29_axis_0"), val = int32(0)]; bool concat_29_interleave_0 = const()[name = string("concat_29_interleave_0"), val = bool(false)]; tensor concat_29 = concat(axis = concat_29_axis_0, interleave = concat_29_interleave_0, values = (gather_85, gather_86, gather_87, var_61))[name = string("concat_29")]; fp32 zero_pad_11_value_0 = const()[name = string("zero_pad_11_value_0"), val = fp32(0x0p+0)]; tensor zero_pad_11 = fill(shape = concat_29, value = zero_pad_11_value_0)[name = string("zero_pad_11")]; bool x_padded_21_interleave_0 = const()[name = string("x_padded_21_interleave_0"), val = bool(false)]; tensor x_padded_21 = concat(axis = var_55, interleave = x_padded_21_interleave_0, values = (zero_pad_11, matrix_bd_21))[name = string("x_padded_21")]; int32 gather_88 = const()[name = string("gather_88"), val = int32(1)]; int32 gather_89 = const()[name = string("gather_89"), val = int32(8)]; int32 gather_90_batch_dims_0 = const()[name = string("gather_90_batch_dims_0"), val = int32(0)]; bool gather_90_validate_indices_0 = const()[name = string("gather_90_validate_indices_0"), val = bool(false)]; int32 select_27 = const()[name = string("select_27"), val = int32(3)]; int32 gather_90_axis_1 = const()[name = string("gather_90_axis_1"), val = int32(0)]; int32 gather_90 = gather(axis = gather_90_axis_1, batch_dims = gather_90_batch_dims_0, indices = select_27, validate_indices = gather_90_validate_indices_0, x = var_868_shape)[name = string("gather_90")]; int32 var_879 = const()[name = string("op_879"), val = int32(1)]; int32 var_880 = add(x = gather_90, y = var_879)[name = string("op_880")]; int32 concat_30_axis_0 = const()[name = string("concat_30_axis_0"), val = int32(0)]; bool concat_30_interleave_0 = const()[name = string("concat_30_interleave_0"), val = bool(false)]; tensor concat_30 = concat(axis = concat_30_axis_0, interleave = concat_30_interleave_0, values = (gather_88, gather_89, var_880, gather_87))[name = string("concat_30")]; tensor x_padded_23 = reshape(shape = concat_30, x = x_padded_21)[name = string("x_padded_23")]; tensor var_887_begin_0 = const()[name = string("op_887_begin_0"), val = tensor([0, 0, 1, 0])]; tensor var_887_end_0 = const()[name = string("op_887_end_0"), val = tensor([1, 8, 0, 0])]; tensor var_887_end_mask_0 = const()[name = string("op_887_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_887 = slice_by_index(begin = var_887_begin_0, end = var_887_end_0, end_mask = var_887_end_mask_0, x = x_padded_23)[name = string("op_887")]; int32 gather_92 = const()[name = string("gather_92"), val = int32(1)]; int32 gather_93 = const()[name = string("gather_93"), val = int32(8)]; int32 concat_31_axis_0 = const()[name = string("concat_31_axis_0"), val = int32(0)]; bool concat_31_interleave_0 = const()[name = string("concat_31_interleave_0"), val = bool(false)]; tensor concat_31 = concat(axis = concat_31_axis_0, interleave = concat_31_interleave_0, values = (gather_92, gather_93, gather_87, gather_90))[name = string("concat_31")]; tensor var_893 = reshape(shape = concat_31, x = var_887)[name = string("op_893")]; int32 floor_div_7 = floor_div(x = gather_90, y = var_53)[name = string("floor_div_7")]; string var_896_dtype_0 = const()[name = string("op_896_dtype_0"), val = string("fp32")]; fp32 var_897_promoted = const()[name = string("op_897_promoted"), val = fp32(0x1p+0)]; fp32 var_896 = cast(dtype = var_896_dtype_0, x = floor_div_7)[name = string("cast_116")]; fp32 var_898 = add(x = var_896, y = var_897_promoted)[name = string("op_898")]; string var_899_dtype_0 = const()[name = string("op_899_dtype_0"), val = string("int32")]; int32 concat_32_values0_0 = const()[name = string("concat_32_values0_0"), val = int32(1)]; int32 concat_32_values1_0 = const()[name = string("concat_32_values1_0"), val = int32(8)]; int32 concat_32_values2_0 = const()[name = string("concat_32_values2_0"), val = int32(0)]; int32 concat_32_axis_0 = const()[name = string("concat_32_axis_0"), val = int32(0)]; bool concat_32_interleave_0 = const()[name = string("concat_32_interleave_0"), val = bool(false)]; int32 var_899 = cast(dtype = var_899_dtype_0, x = var_898)[name = string("cast_115")]; tensor concat_32 = concat(axis = concat_32_axis_0, interleave = concat_32_interleave_0, values = (concat_32_values0_0, concat_32_values1_0, concat_32_values2_0, var_899))[name = string("concat_32")]; tensor matrix_bd_23_begin_0 = const()[name = string("matrix_bd_23_begin_0"), val = tensor([0, 0, 0, 0])]; tensor matrix_bd_23_end_mask_0 = const()[name = string("matrix_bd_23_end_mask_0"), val = tensor([true, true, true, false])]; tensor matrix_bd_23 = slice_by_index(begin = matrix_bd_23_begin_0, end = concat_32, end_mask = matrix_bd_23_end_mask_0, x = var_893)[name = string("matrix_bd_23")]; tensor var_904 = add(x = matrix_ac_11, y = matrix_bd_23)[name = string("op_904")]; fp32 _inversed_scores_21_y_0 = const()[name = string("_inversed_scores_21_y_0"), val = fp32(0x1p-3)]; tensor _inversed_scores_21 = mul(x = var_904, y = _inversed_scores_21_y_0)[name = string("_inversed_scores_21")]; tensor var_910_shape = shape(x = _inversed_scores_21)[name = string("op_910_shape")]; int32 gather_98_batch_dims_0 = const()[name = string("gather_98_batch_dims_0"), val = int32(0)]; bool gather_98_validate_indices_0 = const()[name = string("gather_98_validate_indices_0"), val = bool(false)]; int32 select_29 = const()[name = string("select_29"), val = int32(3)]; int32 gather_98_axis_1 = const()[name = string("gather_98_axis_1"), val = int32(0)]; int32 gather_98 = gather(axis = gather_98_axis_1, batch_dims = gather_98_batch_dims_0, indices = select_29, validate_indices = gather_98_validate_indices_0, x = var_910_shape)[name = string("gather_98")]; int32 concat_33_values0_0 = const()[name = string("concat_33_values0_0"), val = int32(0)]; int32 concat_33_values1_0 = const()[name = string("concat_33_values1_0"), val = int32(1)]; int32 concat_33_values2_0 = const()[name = string("concat_33_values2_0"), val = int32(1)]; int32 concat_33_axis_0 = const()[name = string("concat_33_axis_0"), val = int32(0)]; bool concat_33_interleave_0 = const()[name = string("concat_33_interleave_0"), val = bool(false)]; tensor concat_33 = concat(axis = concat_33_axis_0, interleave = concat_33_interleave_0, values = (concat_33_values0_0, concat_33_values1_0, concat_33_values2_0, gather_98))[name = string("concat_33")]; tensor mask_25_begin_0 = const()[name = string("mask_25_begin_0"), val = tensor([0, 0, 0, 0])]; tensor mask_25_end_mask_0 = const()[name = string("mask_25_end_mask_0"), val = tensor([true, true, true, false])]; tensor mask_25 = slice_by_index(begin = mask_25_begin_0, end = concat_33, end_mask = mask_25_end_mask_0, x = mask_3)[name = string("mask_25")]; tensor scores_23 = select(a = var_37, b = _inversed_scores_21, cond = mask_25)[name = string("scores_23")]; tensor var_916 = softmax(axis = var_55, x = scores_23)[name = string("op_916")]; tensor input_121 = select(a = var_46, b = var_916, cond = mask_25)[name = string("input_121")]; bool x_15_transpose_x_0 = const()[name = string("x_15_transpose_x_0"), val = bool(false)]; bool x_15_transpose_y_0 = const()[name = string("x_15_transpose_y_0"), val = bool(false)]; tensor v_23 = transpose(perm = v_23_perm_0, x = v_21)[name = string("transpose_132")]; tensor x_15 = matmul(transpose_x = x_15_transpose_x_0, transpose_y = x_15_transpose_y_0, x = input_121, y = v_23)[name = string("x_15")]; tensor var_920_perm_0 = const()[name = string("op_920_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_922 = const()[name = string("op_922"), val = tensor([1, -1, 512])]; tensor var_920 = transpose(perm = var_920_perm_0, x = x_15)[name = string("transpose_127")]; tensor input_123 = reshape(shape = var_922, x = var_920)[name = string("input_123")]; tensor input_125 = linear(bias = encoder_encoders_5_self_attn_linear_out_bias, weight = encoder_encoders_5_self_attn_linear_out_weight, x = input_123)[name = string("linear_40")]; tensor input_127 = add(x = input_119, y = input_125)[name = string("input_127")]; tensor input_129_axes_0 = const()[name = string("input_129_axes_0"), val = tensor([-1])]; tensor input_129 = layer_norm(axes = input_129_axes_0, beta = encoder_encoders_5_norm_ff_bias, epsilon = var_36, gamma = encoder_encoders_5_norm_ff_weight, x = input_127)[name = string("input_129")]; tensor input_131 = linear(bias = encoder_encoders_5_feed_forward_w_1_bias, weight = encoder_encoders_5_feed_forward_w_1_weight, x = input_129)[name = string("linear_41")]; tensor input_133 = silu(x = input_131)[name = string("input_133")]; tensor input_137 = linear(bias = encoder_encoders_5_feed_forward_w_2_bias, weight = encoder_encoders_5_feed_forward_w_2_weight, x = input_133)[name = string("linear_42")]; tensor xs_3 = add(x = input_127, y = input_137)[name = string("xs_3")]; tensor var_947_perm_0 = const()[name = string("op_947_perm_0"), val = tensor([0, 2, 1])]; tensor expand_dims_0_axes_0 = const()[name = string("expand_dims_0_axes_0"), val = tensor([3])]; tensor var_947 = transpose(perm = var_947_perm_0, x = xs_3)[name = string("transpose_126")]; tensor expand_dims_0 = expand_dims(axes = expand_dims_0_axes_0, x = var_947)[name = string("expand_dims_0")]; int32 upsample_nearest_neighbor_0_scale_factor_height_0 = const()[name = string("upsample_nearest_neighbor_0_scale_factor_height_0"), val = int32(2)]; int32 upsample_nearest_neighbor_0_scale_factor_width_0 = const()[name = string("upsample_nearest_neighbor_0_scale_factor_width_0"), val = int32(1)]; tensor upsample_nearest_neighbor_0 = upsample_nearest_neighbor(scale_factor_height = upsample_nearest_neighbor_0_scale_factor_height_0, scale_factor_width = upsample_nearest_neighbor_0_scale_factor_width_0, x = expand_dims_0)[name = string("upsample_nearest_neighbor_0")]; tensor input_141_axes_0 = const()[name = string("input_141_axes_0"), val = tensor([3])]; tensor input_141 = squeeze(axes = input_141_axes_0, x = upsample_nearest_neighbor_0)[name = string("input_141")]; fp32 const_20 = const()[name = string("const_20"), val = fp32(0x0p+0)]; tensor input_143_pad_0 = const()[name = string("input_143_pad_0"), val = tensor([0, 0, 0, 0, 4, 0])]; string input_143_mode_0 = const()[name = string("input_143_mode_0"), val = string("constant")]; tensor input_143 = pad(constant_val = const_20, mode = input_143_mode_0, pad = input_143_pad_0, x = input_141)[name = string("input_143")]; string xs_5_pad_type_0 = const()[name = string("xs_5_pad_type_0"), val = string("valid")]; tensor xs_5_strides_0 = const()[name = string("xs_5_strides_0"), val = tensor([1])]; tensor xs_5_pad_0 = const()[name = string("xs_5_pad_0"), val = tensor([0, 0])]; tensor xs_5_dilations_0 = const()[name = string("xs_5_dilations_0"), val = tensor([1])]; int32 xs_5_groups_0 = const()[name = string("xs_5_groups_0"), val = int32(1)]; tensor xs_5 = conv(bias = encoder_up_layer_conv_bias, dilations = xs_5_dilations_0, groups = xs_5_groups_0, pad = xs_5_pad_0, pad_type = xs_5_pad_type_0, strides = xs_5_strides_0, weight = encoder_up_layer_conv_weight, x = input_143)[name = string("xs_5")]; tensor var_966_perm_0 = const()[name = string("op_966_perm_0"), val = tensor([0, 2, 1])]; tensor var_966 = transpose(perm = var_966_perm_0, x = xs_5)[name = string("transpose_125")]; tensor var_968_shape = shape(x = var_966)[name = string("op_968_shape")]; int32 gather_99_batch_dims_0 = const()[name = string("gather_99_batch_dims_0"), val = int32(0)]; bool gather_99_validate_indices_0 = const()[name = string("gather_99_validate_indices_0"), val = bool(false)]; int32 select_30 = const()[name = string("select_30"), val = int32(1)]; int32 gather_99_axis_1 = const()[name = string("gather_99_axis_1"), val = int32(0)]; int32 gather_99 = gather(axis = gather_99_axis_1, batch_dims = gather_99_batch_dims_0, indices = select_30, validate_indices = gather_99_validate_indices_0, x = var_968_shape)[name = string("gather_99")]; int32 const_22 = const()[name = string("const_22"), val = int32(1)]; int32 const_23 = const()[name = string("const_23"), val = int32(1)]; tensor seq_range = range_1d(end = gather_99, start = var_57, step = const_23)[name = string("seq_range")]; tensor var_972_axes_0 = const()[name = string("op_972_axes_0"), val = tensor([0])]; tensor var_972 = expand_dims(axes = var_972_axes_0, x = seq_range)[name = string("op_972")]; int32 concat_34_axis_0 = const()[name = string("concat_34_axis_0"), val = int32(0)]; bool concat_34_interleave_0 = const()[name = string("concat_34_interleave_0"), val = bool(false)]; tensor concat_34 = concat(axis = concat_34_axis_0, interleave = concat_34_interleave_0, values = (const_22, gather_99))[name = string("concat_34")]; tensor shape_1 = shape(x = var_972)[name = string("shape_1")]; int32 equal_1_y_0 = const()[name = string("equal_1_y_0"), val = int32(-1)]; tensor equal_1 = equal(x = concat_34, y = equal_1_y_0)[name = string("equal_1")]; tensor select_1 = select(a = shape_1, b = concat_34, cond = equal_1)[name = string("select_1")]; tensor real_div_1 = real_div(x = select_1, y = shape_1)[name = string("real_div_1")]; tensor seq_range_expand = tile(reps = real_div_1, x = var_972)[name = string("seq_range_expand")]; tensor seq_length_expand = const()[name = string("seq_length_expand"), val = tensor([[1000]])]; tensor var_976 = greater_equal(x = seq_range_expand, y = seq_length_expand)[name = string("op_976")]; tensor var_977_axes_0 = const()[name = string("op_977_axes_0"), val = tensor([1])]; tensor var_977 = expand_dims(axes = var_977_axes_0, x = var_976)[name = string("op_977")]; tensor masks = logical_not(x = var_977)[name = string("masks")]; tensor input_145 = linear(bias = encoder_up_embed_out_0_bias, weight = encoder_up_embed_out_0_weight, x = var_966)[name = string("linear_43")]; tensor input_147_axes_0 = const()[name = string("input_147_axes_0"), val = tensor([-1])]; tensor input_147 = layer_norm(axes = input_147_axes_0, beta = encoder_up_embed_out_1_bias, epsilon = var_50, gamma = encoder_up_embed_out_1_weight, x = input_145)[name = string("input_147")]; fp32 var_990 = const()[name = string("op_990"), val = fp32(0x1.6a09e6p+4)]; tensor x_19 = mul(x = input_147, y = var_990)[name = string("x_19")]; tensor var_992_shape = shape(x = x_19)[name = string("op_992_shape")]; int32 gather_100_batch_dims_0 = const()[name = string("gather_100_batch_dims_0"), val = int32(0)]; bool gather_100_validate_indices_0 = const()[name = string("gather_100_validate_indices_0"), val = bool(false)]; int32 select_31 = const()[name = string("select_31"), val = int32(1)]; int32 gather_100_axis_1 = const()[name = string("gather_100_axis_1"), val = int32(0)]; int32 gather_100 = gather(axis = gather_100_axis_1, batch_dims = gather_100_batch_dims_0, indices = select_31, validate_indices = gather_100_validate_indices_0, x = var_992_shape)[name = string("gather_100")]; tensor var_996 = const()[name = string("op_996"), val = tensor([0x1.387p+12])]; string gather_100_promoted_dtype_0 = const()[name = string("gather_100_promoted_dtype_0"), val = string("fp32")]; fp32 gather_100_promoted = cast(dtype = gather_100_promoted_dtype_0, x = gather_100)[name = string("cast_114")]; tensor var_997 = sub(x = var_996, y = gather_100_promoted)[name = string("op_997")]; fp32 var_998_promoted = const()[name = string("op_998_promoted"), val = fp32(0x1p+0)]; tensor var_999 = add(x = var_997, y = var_998_promoted)[name = string("op_999")]; fp32 var_1000_item = squeeze(x = var_999)[name = string("op_1000_item")]; string var_1000_dtype_0 = const()[name = string("op_1000_dtype_0"), val = string("int32")]; tensor var_1003 = const()[name = string("op_1003"), val = tensor([0x1.387p+12])]; tensor var_1004 = add(x = var_1003, y = gather_100_promoted)[name = string("op_1004")]; fp32 var_1005_item = squeeze(x = var_1004)[name = string("op_1005_item")]; string var_1005_dtype_0 = const()[name = string("op_1005_dtype_0"), val = string("int32")]; int32 concat_35_values0_0 = const()[name = string("concat_35_values0_0"), val = int32(0)]; int32 concat_35_values2_0 = const()[name = string("concat_35_values2_0"), val = int32(0)]; int32 concat_35_axis_0 = const()[name = string("concat_35_axis_0"), val = int32(0)]; bool concat_35_interleave_0 = const()[name = string("concat_35_interleave_0"), val = bool(false)]; int32 var_1000 = cast(dtype = var_1000_dtype_0, x = var_1000_item)[name = string("cast_113")]; tensor concat_35 = concat(axis = concat_35_axis_0, interleave = concat_35_interleave_0, values = (concat_35_values0_0, var_1000, concat_35_values2_0))[name = string("concat_35")]; int32 concat_36_values0_0 = const()[name = string("concat_36_values0_0"), val = int32(1)]; int32 concat_36_values2_0 = const()[name = string("concat_36_values2_0"), val = int32(512)]; int32 concat_36_axis_0 = const()[name = string("concat_36_axis_0"), val = int32(0)]; bool concat_36_interleave_0 = const()[name = string("concat_36_interleave_0"), val = bool(false)]; int32 var_1005 = cast(dtype = var_1005_dtype_0, x = var_1005_item)[name = string("cast_112")]; tensor concat_36 = concat(axis = concat_36_axis_0, interleave = concat_36_interleave_0, values = (concat_36_values0_0, var_1005, concat_36_values2_0))[name = string("concat_36")]; tensor input_149_end_mask_0 = const()[name = string("input_149_end_mask_0"), val = tensor([true, false, true])]; tensor input_149 = slice_by_index(begin = concat_35, end = concat_36, end_mask = input_149_end_mask_0, x = var_54)[name = string("input_149")]; tensor query_13_axes_0 = const()[name = string("query_13_axes_0"), val = tensor([-1])]; tensor query_13 = layer_norm(axes = query_13_axes_0, beta = encoder_up_encoders_0_norm_mha_bias, epsilon = var_36, gamma = encoder_up_encoders_0_norm_mha_weight, x = x_19)[name = string("query_13")]; tensor var_1034 = linear(bias = encoder_up_encoders_0_self_attn_linear_q_bias, weight = encoder_up_encoders_0_self_attn_linear_q_weight, x = query_13)[name = string("linear_44")]; tensor var_1035 = const()[name = string("op_1035"), val = tensor([1, -1, 8, 64])]; tensor q_37 = reshape(shape = var_1035, x = var_1034)[name = string("q_37")]; tensor var_1039 = linear(bias = encoder_up_encoders_0_self_attn_linear_k_bias, weight = encoder_up_encoders_0_self_attn_linear_k_weight, x = query_13)[name = string("linear_45")]; tensor var_1040 = const()[name = string("op_1040"), val = tensor([1, -1, 8, 64])]; tensor k_25 = reshape(shape = var_1040, x = var_1039)[name = string("k_25")]; tensor var_1044 = linear(bias = encoder_up_encoders_0_self_attn_linear_v_bias, weight = encoder_up_encoders_0_self_attn_linear_v_weight, x = query_13)[name = string("linear_46")]; tensor var_1045 = const()[name = string("op_1045"), val = tensor([1, -1, 8, 64])]; tensor v_25 = reshape(shape = var_1045, x = var_1044)[name = string("v_25")]; tensor v_27_perm_0 = const()[name = string("v_27_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_1053 = linear(bias = linear_4_bias_0, weight = encoder_up_encoders_0_self_attn_linear_pos_weight, x = input_149)[name = string("linear_47")]; tensor var_1054 = const()[name = string("op_1054"), val = tensor([1, -1, 8, 64])]; tensor p_25 = reshape(shape = var_1054, x = var_1053)[name = string("p_25")]; tensor const_26 = const()[name = string("const_26"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185402432)))]; tensor var_1058 = add(x = q_37, y = const_26)[name = string("op_1058")]; tensor const_27 = const()[name = string("const_27"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185404544)))]; tensor var_1061 = add(x = q_37, y = const_27)[name = string("op_1061")]; bool matrix_ac_13_transpose_x_0 = const()[name = string("matrix_ac_13_transpose_x_0"), val = bool(false)]; bool matrix_ac_13_transpose_y_0 = const()[name = string("matrix_ac_13_transpose_y_0"), val = bool(false)]; tensor transpose_84_perm_0 = const()[name = string("transpose_84_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_85_perm_0 = const()[name = string("transpose_85_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_85 = transpose(perm = transpose_85_perm_0, x = k_25)[name = string("transpose_122")]; tensor transpose_84 = transpose(perm = transpose_84_perm_0, x = var_1058)[name = string("transpose_123")]; tensor matrix_ac_13 = matmul(transpose_x = matrix_ac_13_transpose_x_0, transpose_y = matrix_ac_13_transpose_y_0, x = transpose_84, y = transpose_85)[name = string("matrix_ac_13")]; bool matrix_bd_25_transpose_x_0 = const()[name = string("matrix_bd_25_transpose_x_0"), val = bool(false)]; bool matrix_bd_25_transpose_y_0 = const()[name = string("matrix_bd_25_transpose_y_0"), val = bool(false)]; tensor transpose_86_perm_0 = const()[name = string("transpose_86_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_87_perm_0 = const()[name = string("transpose_87_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_87 = transpose(perm = transpose_87_perm_0, x = p_25)[name = string("transpose_120")]; tensor transpose_86 = transpose(perm = transpose_86_perm_0, x = var_1061)[name = string("transpose_121")]; tensor matrix_bd_25 = matmul(transpose_x = matrix_bd_25_transpose_x_0, transpose_y = matrix_bd_25_transpose_y_0, x = transpose_86, y = transpose_87)[name = string("matrix_bd_25")]; tensor var_1067_shape = shape(x = matrix_bd_25)[name = string("op_1067_shape")]; int32 gather_103 = const()[name = string("gather_103"), val = int32(1)]; int32 gather_104 = const()[name = string("gather_104"), val = int32(8)]; int32 gather_105_batch_dims_0 = const()[name = string("gather_105_batch_dims_0"), val = int32(0)]; bool gather_105_validate_indices_0 = const()[name = string("gather_105_validate_indices_0"), val = bool(false)]; int32 select_32 = const()[name = string("select_32"), val = int32(2)]; int32 gather_105_axis_1 = const()[name = string("gather_105_axis_1"), val = int32(0)]; int32 gather_105 = gather(axis = gather_105_axis_1, batch_dims = gather_105_batch_dims_0, indices = select_32, validate_indices = gather_105_validate_indices_0, x = var_1067_shape)[name = string("gather_105")]; int32 concat_37_axis_0 = const()[name = string("concat_37_axis_0"), val = int32(0)]; bool concat_37_interleave_0 = const()[name = string("concat_37_interleave_0"), val = bool(false)]; tensor concat_37 = concat(axis = concat_37_axis_0, interleave = concat_37_interleave_0, values = (gather_103, gather_104, gather_105, var_61))[name = string("concat_37")]; fp32 zero_pad_13_value_0 = const()[name = string("zero_pad_13_value_0"), val = fp32(0x0p+0)]; tensor zero_pad_13 = fill(shape = concat_37, value = zero_pad_13_value_0)[name = string("zero_pad_13")]; bool x_padded_25_interleave_0 = const()[name = string("x_padded_25_interleave_0"), val = bool(false)]; tensor x_padded_25 = concat(axis = var_55, interleave = x_padded_25_interleave_0, values = (zero_pad_13, matrix_bd_25))[name = string("x_padded_25")]; int32 gather_106 = const()[name = string("gather_106"), val = int32(1)]; int32 gather_107 = const()[name = string("gather_107"), val = int32(8)]; int32 gather_108_batch_dims_0 = const()[name = string("gather_108_batch_dims_0"), val = int32(0)]; bool gather_108_validate_indices_0 = const()[name = string("gather_108_validate_indices_0"), val = bool(false)]; int32 select_33 = const()[name = string("select_33"), val = int32(3)]; int32 gather_108_axis_1 = const()[name = string("gather_108_axis_1"), val = int32(0)]; int32 gather_108 = gather(axis = gather_108_axis_1, batch_dims = gather_108_batch_dims_0, indices = select_33, validate_indices = gather_108_validate_indices_0, x = var_1067_shape)[name = string("gather_108")]; int32 var_1078 = const()[name = string("op_1078"), val = int32(1)]; int32 var_1079 = add(x = gather_108, y = var_1078)[name = string("op_1079")]; int32 concat_38_axis_0 = const()[name = string("concat_38_axis_0"), val = int32(0)]; bool concat_38_interleave_0 = const()[name = string("concat_38_interleave_0"), val = bool(false)]; tensor concat_38 = concat(axis = concat_38_axis_0, interleave = concat_38_interleave_0, values = (gather_106, gather_107, var_1079, gather_105))[name = string("concat_38")]; tensor x_padded_27 = reshape(shape = concat_38, x = x_padded_25)[name = string("x_padded_27")]; tensor var_1086_begin_0 = const()[name = string("op_1086_begin_0"), val = tensor([0, 0, 1, 0])]; tensor var_1086_end_0 = const()[name = string("op_1086_end_0"), val = tensor([1, 8, 0, 0])]; tensor var_1086_end_mask_0 = const()[name = string("op_1086_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_1086 = slice_by_index(begin = var_1086_begin_0, end = var_1086_end_0, end_mask = var_1086_end_mask_0, x = x_padded_27)[name = string("op_1086")]; int32 gather_110 = const()[name = string("gather_110"), val = int32(1)]; int32 gather_111 = const()[name = string("gather_111"), val = int32(8)]; int32 concat_39_axis_0 = const()[name = string("concat_39_axis_0"), val = int32(0)]; bool concat_39_interleave_0 = const()[name = string("concat_39_interleave_0"), val = bool(false)]; tensor concat_39 = concat(axis = concat_39_axis_0, interleave = concat_39_interleave_0, values = (gather_110, gather_111, gather_105, gather_108))[name = string("concat_39")]; tensor var_1092 = reshape(shape = concat_39, x = var_1086)[name = string("op_1092")]; int32 floor_div_10 = floor_div(x = gather_108, y = var_53)[name = string("floor_div_10")]; string var_1095_dtype_0 = const()[name = string("op_1095_dtype_0"), val = string("fp32")]; fp32 var_1096_promoted = const()[name = string("op_1096_promoted"), val = fp32(0x1p+0)]; fp32 var_1095 = cast(dtype = var_1095_dtype_0, x = floor_div_10)[name = string("cast_111")]; fp32 var_1097 = add(x = var_1095, y = var_1096_promoted)[name = string("op_1097")]; string var_1098_dtype_0 = const()[name = string("op_1098_dtype_0"), val = string("int32")]; int32 concat_40_values0_0 = const()[name = string("concat_40_values0_0"), val = int32(1)]; int32 concat_40_values1_0 = const()[name = string("concat_40_values1_0"), val = int32(8)]; int32 concat_40_values2_0 = const()[name = string("concat_40_values2_0"), val = int32(0)]; int32 concat_40_axis_0 = const()[name = string("concat_40_axis_0"), val = int32(0)]; bool concat_40_interleave_0 = const()[name = string("concat_40_interleave_0"), val = bool(false)]; int32 var_1098 = cast(dtype = var_1098_dtype_0, x = var_1097)[name = string("cast_110")]; tensor concat_40 = concat(axis = concat_40_axis_0, interleave = concat_40_interleave_0, values = (concat_40_values0_0, concat_40_values1_0, concat_40_values2_0, var_1098))[name = string("concat_40")]; tensor matrix_bd_27_begin_0 = const()[name = string("matrix_bd_27_begin_0"), val = tensor([0, 0, 0, 0])]; tensor matrix_bd_27_end_mask_0 = const()[name = string("matrix_bd_27_end_mask_0"), val = tensor([true, true, true, false])]; tensor matrix_bd_27 = slice_by_index(begin = matrix_bd_27_begin_0, end = concat_40, end_mask = matrix_bd_27_end_mask_0, x = var_1092)[name = string("matrix_bd_27")]; tensor var_1103 = add(x = matrix_ac_13, y = matrix_bd_27)[name = string("op_1103")]; fp32 _inversed_scores_25_y_0 = const()[name = string("_inversed_scores_25_y_0"), val = fp32(0x1p-3)]; tensor _inversed_scores_25 = mul(x = var_1103, y = _inversed_scores_25_y_0)[name = string("_inversed_scores_25")]; tensor var_1107_axes_0 = const()[name = string("op_1107_axes_0"), val = tensor([1])]; tensor var_1107 = expand_dims(axes = var_1107_axes_0, x = masks)[name = string("op_1107")]; string cast_71_dtype_0 = const()[name = string("cast_71_dtype_0"), val = string("int32")]; tensor cast_71 = cast(dtype = cast_71_dtype_0, x = var_1107)[name = string("cast_109")]; tensor mask_27 = equal(x = cast_71, y = var_57)[name = string("mask_27")]; tensor var_1109_shape = shape(x = _inversed_scores_25)[name = string("op_1109_shape")]; int32 gather_116_batch_dims_0 = const()[name = string("gather_116_batch_dims_0"), val = int32(0)]; bool gather_116_validate_indices_0 = const()[name = string("gather_116_validate_indices_0"), val = bool(false)]; int32 select_35 = const()[name = string("select_35"), val = int32(3)]; int32 gather_116_axis_1 = const()[name = string("gather_116_axis_1"), val = int32(0)]; int32 gather_116 = gather(axis = gather_116_axis_1, batch_dims = gather_116_batch_dims_0, indices = select_35, validate_indices = gather_116_validate_indices_0, x = var_1109_shape)[name = string("gather_116")]; int32 concat_41_values0_0 = const()[name = string("concat_41_values0_0"), val = int32(0)]; int32 concat_41_values1_0 = const()[name = string("concat_41_values1_0"), val = int32(1)]; int32 concat_41_values2_0 = const()[name = string("concat_41_values2_0"), val = int32(1)]; int32 concat_41_axis_0 = const()[name = string("concat_41_axis_0"), val = int32(0)]; bool concat_41_interleave_0 = const()[name = string("concat_41_interleave_0"), val = bool(false)]; tensor concat_41 = concat(axis = concat_41_axis_0, interleave = concat_41_interleave_0, values = (concat_41_values0_0, concat_41_values1_0, concat_41_values2_0, gather_116))[name = string("concat_41")]; tensor mask_29_begin_0 = const()[name = string("mask_29_begin_0"), val = tensor([0, 0, 0, 0])]; tensor mask_29_end_mask_0 = const()[name = string("mask_29_end_mask_0"), val = tensor([true, true, true, false])]; tensor mask_29 = slice_by_index(begin = mask_29_begin_0, end = concat_41, end_mask = mask_29_end_mask_0, x = mask_27)[name = string("mask_29")]; tensor scores_27 = select(a = var_37, b = _inversed_scores_25, cond = mask_29)[name = string("scores_27")]; tensor var_1115 = softmax(axis = var_55, x = scores_27)[name = string("op_1115")]; tensor input_153 = select(a = var_46, b = var_1115, cond = mask_29)[name = string("input_153")]; bool x_21_transpose_x_0 = const()[name = string("x_21_transpose_x_0"), val = bool(false)]; bool x_21_transpose_y_0 = const()[name = string("x_21_transpose_y_0"), val = bool(false)]; tensor v_27 = transpose(perm = v_27_perm_0, x = v_25)[name = string("transpose_124")]; tensor x_21 = matmul(transpose_x = x_21_transpose_x_0, transpose_y = x_21_transpose_y_0, x = input_153, y = v_27)[name = string("x_21")]; tensor var_1119_perm_0 = const()[name = string("op_1119_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_1121 = const()[name = string("op_1121"), val = tensor([1, -1, 512])]; tensor var_1119 = transpose(perm = var_1119_perm_0, x = x_21)[name = string("transpose_119")]; tensor input_155 = reshape(shape = var_1121, x = var_1119)[name = string("input_155")]; tensor input_157 = linear(bias = encoder_up_encoders_0_self_attn_linear_out_bias, weight = encoder_up_encoders_0_self_attn_linear_out_weight, x = input_155)[name = string("linear_48")]; tensor input_159 = add(x = x_19, y = input_157)[name = string("input_159")]; tensor input_161_axes_0 = const()[name = string("input_161_axes_0"), val = tensor([-1])]; tensor input_161 = layer_norm(axes = input_161_axes_0, beta = encoder_up_encoders_0_norm_ff_bias, epsilon = var_36, gamma = encoder_up_encoders_0_norm_ff_weight, x = input_159)[name = string("input_161")]; tensor input_163 = linear(bias = encoder_up_encoders_0_feed_forward_w_1_bias, weight = encoder_up_encoders_0_feed_forward_w_1_weight, x = input_161)[name = string("linear_49")]; tensor input_165 = silu(x = input_163)[name = string("input_165")]; tensor input_169 = linear(bias = encoder_up_encoders_0_feed_forward_w_2_bias, weight = encoder_up_encoders_0_feed_forward_w_2_weight, x = input_165)[name = string("linear_50")]; tensor input_171 = add(x = input_159, y = input_169)[name = string("input_171")]; tensor query_15_axes_0 = const()[name = string("query_15_axes_0"), val = tensor([-1])]; tensor query_15 = layer_norm(axes = query_15_axes_0, beta = encoder_up_encoders_1_norm_mha_bias, epsilon = var_36, gamma = encoder_up_encoders_1_norm_mha_weight, x = input_171)[name = string("query_15")]; tensor var_1164 = linear(bias = encoder_up_encoders_1_self_attn_linear_q_bias, weight = encoder_up_encoders_1_self_attn_linear_q_weight, x = query_15)[name = string("linear_51")]; tensor var_1165 = const()[name = string("op_1165"), val = tensor([1, -1, 8, 64])]; tensor q_43 = reshape(shape = var_1165, x = var_1164)[name = string("q_43")]; tensor var_1169 = linear(bias = encoder_up_encoders_1_self_attn_linear_k_bias, weight = encoder_up_encoders_1_self_attn_linear_k_weight, x = query_15)[name = string("linear_52")]; tensor var_1170 = const()[name = string("op_1170"), val = tensor([1, -1, 8, 64])]; tensor k_29 = reshape(shape = var_1170, x = var_1169)[name = string("k_29")]; tensor var_1174 = linear(bias = encoder_up_encoders_1_self_attn_linear_v_bias, weight = encoder_up_encoders_1_self_attn_linear_v_weight, x = query_15)[name = string("linear_53")]; tensor var_1175 = const()[name = string("op_1175"), val = tensor([1, -1, 8, 64])]; tensor v_29 = reshape(shape = var_1175, x = var_1174)[name = string("v_29")]; tensor v_31_perm_0 = const()[name = string("v_31_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_1183 = linear(bias = linear_4_bias_0, weight = encoder_up_encoders_1_self_attn_linear_pos_weight, x = input_149)[name = string("linear_54")]; tensor var_1184 = const()[name = string("op_1184"), val = tensor([1, -1, 8, 64])]; tensor p_29 = reshape(shape = var_1184, x = var_1183)[name = string("p_29")]; tensor const_28 = const()[name = string("const_28"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185406656)))]; tensor var_1188 = add(x = q_43, y = const_28)[name = string("op_1188")]; tensor const_29 = const()[name = string("const_29"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185408768)))]; tensor var_1191 = add(x = q_43, y = const_29)[name = string("op_1191")]; bool matrix_ac_15_transpose_x_0 = const()[name = string("matrix_ac_15_transpose_x_0"), val = bool(false)]; bool matrix_ac_15_transpose_y_0 = const()[name = string("matrix_ac_15_transpose_y_0"), val = bool(false)]; tensor transpose_88_perm_0 = const()[name = string("transpose_88_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_89_perm_0 = const()[name = string("transpose_89_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_89 = transpose(perm = transpose_89_perm_0, x = k_29)[name = string("transpose_116")]; tensor transpose_88 = transpose(perm = transpose_88_perm_0, x = var_1188)[name = string("transpose_117")]; tensor matrix_ac_15 = matmul(transpose_x = matrix_ac_15_transpose_x_0, transpose_y = matrix_ac_15_transpose_y_0, x = transpose_88, y = transpose_89)[name = string("matrix_ac_15")]; bool matrix_bd_29_transpose_x_0 = const()[name = string("matrix_bd_29_transpose_x_0"), val = bool(false)]; bool matrix_bd_29_transpose_y_0 = const()[name = string("matrix_bd_29_transpose_y_0"), val = bool(false)]; tensor transpose_90_perm_0 = const()[name = string("transpose_90_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_91_perm_0 = const()[name = string("transpose_91_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_91 = transpose(perm = transpose_91_perm_0, x = p_29)[name = string("transpose_114")]; tensor transpose_90 = transpose(perm = transpose_90_perm_0, x = var_1191)[name = string("transpose_115")]; tensor matrix_bd_29 = matmul(transpose_x = matrix_bd_29_transpose_x_0, transpose_y = matrix_bd_29_transpose_y_0, x = transpose_90, y = transpose_91)[name = string("matrix_bd_29")]; tensor var_1197_shape = shape(x = matrix_bd_29)[name = string("op_1197_shape")]; int32 gather_119 = const()[name = string("gather_119"), val = int32(1)]; int32 gather_120 = const()[name = string("gather_120"), val = int32(8)]; int32 gather_121_batch_dims_0 = const()[name = string("gather_121_batch_dims_0"), val = int32(0)]; bool gather_121_validate_indices_0 = const()[name = string("gather_121_validate_indices_0"), val = bool(false)]; int32 select_36 = const()[name = string("select_36"), val = int32(2)]; int32 gather_121_axis_1 = const()[name = string("gather_121_axis_1"), val = int32(0)]; int32 gather_121 = gather(axis = gather_121_axis_1, batch_dims = gather_121_batch_dims_0, indices = select_36, validate_indices = gather_121_validate_indices_0, x = var_1197_shape)[name = string("gather_121")]; int32 concat_42_axis_0 = const()[name = string("concat_42_axis_0"), val = int32(0)]; bool concat_42_interleave_0 = const()[name = string("concat_42_interleave_0"), val = bool(false)]; tensor concat_42 = concat(axis = concat_42_axis_0, interleave = concat_42_interleave_0, values = (gather_119, gather_120, gather_121, var_61))[name = string("concat_42")]; fp32 zero_pad_15_value_0 = const()[name = string("zero_pad_15_value_0"), val = fp32(0x0p+0)]; tensor zero_pad_15 = fill(shape = concat_42, value = zero_pad_15_value_0)[name = string("zero_pad_15")]; bool x_padded_29_interleave_0 = const()[name = string("x_padded_29_interleave_0"), val = bool(false)]; tensor x_padded_29 = concat(axis = var_55, interleave = x_padded_29_interleave_0, values = (zero_pad_15, matrix_bd_29))[name = string("x_padded_29")]; int32 gather_122 = const()[name = string("gather_122"), val = int32(1)]; int32 gather_123 = const()[name = string("gather_123"), val = int32(8)]; int32 gather_124_batch_dims_0 = const()[name = string("gather_124_batch_dims_0"), val = int32(0)]; bool gather_124_validate_indices_0 = const()[name = string("gather_124_validate_indices_0"), val = bool(false)]; int32 select_37 = const()[name = string("select_37"), val = int32(3)]; int32 gather_124_axis_1 = const()[name = string("gather_124_axis_1"), val = int32(0)]; int32 gather_124 = gather(axis = gather_124_axis_1, batch_dims = gather_124_batch_dims_0, indices = select_37, validate_indices = gather_124_validate_indices_0, x = var_1197_shape)[name = string("gather_124")]; int32 var_1208 = const()[name = string("op_1208"), val = int32(1)]; int32 var_1209 = add(x = gather_124, y = var_1208)[name = string("op_1209")]; int32 concat_43_axis_0 = const()[name = string("concat_43_axis_0"), val = int32(0)]; bool concat_43_interleave_0 = const()[name = string("concat_43_interleave_0"), val = bool(false)]; tensor concat_43 = concat(axis = concat_43_axis_0, interleave = concat_43_interleave_0, values = (gather_122, gather_123, var_1209, gather_121))[name = string("concat_43")]; tensor x_padded_31 = reshape(shape = concat_43, x = x_padded_29)[name = string("x_padded_31")]; tensor var_1216_begin_0 = const()[name = string("op_1216_begin_0"), val = tensor([0, 0, 1, 0])]; tensor var_1216_end_0 = const()[name = string("op_1216_end_0"), val = tensor([1, 8, 0, 0])]; tensor var_1216_end_mask_0 = const()[name = string("op_1216_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_1216 = slice_by_index(begin = var_1216_begin_0, end = var_1216_end_0, end_mask = var_1216_end_mask_0, x = x_padded_31)[name = string("op_1216")]; int32 gather_126 = const()[name = string("gather_126"), val = int32(1)]; int32 gather_127 = const()[name = string("gather_127"), val = int32(8)]; int32 concat_44_axis_0 = const()[name = string("concat_44_axis_0"), val = int32(0)]; bool concat_44_interleave_0 = const()[name = string("concat_44_interleave_0"), val = bool(false)]; tensor concat_44 = concat(axis = concat_44_axis_0, interleave = concat_44_interleave_0, values = (gather_126, gather_127, gather_121, gather_124))[name = string("concat_44")]; tensor var_1222 = reshape(shape = concat_44, x = var_1216)[name = string("op_1222")]; int32 floor_div_11 = floor_div(x = gather_124, y = var_53)[name = string("floor_div_11")]; string var_1225_dtype_0 = const()[name = string("op_1225_dtype_0"), val = string("fp32")]; fp32 var_1226_promoted = const()[name = string("op_1226_promoted"), val = fp32(0x1p+0)]; fp32 var_1225 = cast(dtype = var_1225_dtype_0, x = floor_div_11)[name = string("cast_108")]; fp32 var_1227 = add(x = var_1225, y = var_1226_promoted)[name = string("op_1227")]; string var_1228_dtype_0 = const()[name = string("op_1228_dtype_0"), val = string("int32")]; int32 concat_45_values0_0 = const()[name = string("concat_45_values0_0"), val = int32(1)]; int32 concat_45_values1_0 = const()[name = string("concat_45_values1_0"), val = int32(8)]; int32 concat_45_values2_0 = const()[name = string("concat_45_values2_0"), val = int32(0)]; int32 concat_45_axis_0 = const()[name = string("concat_45_axis_0"), val = int32(0)]; bool concat_45_interleave_0 = const()[name = string("concat_45_interleave_0"), val = bool(false)]; int32 var_1228 = cast(dtype = var_1228_dtype_0, x = var_1227)[name = string("cast_107")]; tensor concat_45 = concat(axis = concat_45_axis_0, interleave = concat_45_interleave_0, values = (concat_45_values0_0, concat_45_values1_0, concat_45_values2_0, var_1228))[name = string("concat_45")]; tensor matrix_bd_31_begin_0 = const()[name = string("matrix_bd_31_begin_0"), val = tensor([0, 0, 0, 0])]; tensor matrix_bd_31_end_mask_0 = const()[name = string("matrix_bd_31_end_mask_0"), val = tensor([true, true, true, false])]; tensor matrix_bd_31 = slice_by_index(begin = matrix_bd_31_begin_0, end = concat_45, end_mask = matrix_bd_31_end_mask_0, x = var_1222)[name = string("matrix_bd_31")]; tensor var_1233 = add(x = matrix_ac_15, y = matrix_bd_31)[name = string("op_1233")]; fp32 _inversed_scores_29_y_0 = const()[name = string("_inversed_scores_29_y_0"), val = fp32(0x1p-3)]; tensor _inversed_scores_29 = mul(x = var_1233, y = _inversed_scores_29_y_0)[name = string("_inversed_scores_29")]; tensor var_1239_shape = shape(x = _inversed_scores_29)[name = string("op_1239_shape")]; int32 gather_132_batch_dims_0 = const()[name = string("gather_132_batch_dims_0"), val = int32(0)]; bool gather_132_validate_indices_0 = const()[name = string("gather_132_validate_indices_0"), val = bool(false)]; int32 select_39 = const()[name = string("select_39"), val = int32(3)]; int32 gather_132_axis_1 = const()[name = string("gather_132_axis_1"), val = int32(0)]; int32 gather_132 = gather(axis = gather_132_axis_1, batch_dims = gather_132_batch_dims_0, indices = select_39, validate_indices = gather_132_validate_indices_0, x = var_1239_shape)[name = string("gather_132")]; int32 concat_46_values0_0 = const()[name = string("concat_46_values0_0"), val = int32(0)]; int32 concat_46_values1_0 = const()[name = string("concat_46_values1_0"), val = int32(1)]; int32 concat_46_values2_0 = const()[name = string("concat_46_values2_0"), val = int32(1)]; int32 concat_46_axis_0 = const()[name = string("concat_46_axis_0"), val = int32(0)]; bool concat_46_interleave_0 = const()[name = string("concat_46_interleave_0"), val = bool(false)]; tensor concat_46 = concat(axis = concat_46_axis_0, interleave = concat_46_interleave_0, values = (concat_46_values0_0, concat_46_values1_0, concat_46_values2_0, gather_132))[name = string("concat_46")]; tensor mask_33_begin_0 = const()[name = string("mask_33_begin_0"), val = tensor([0, 0, 0, 0])]; tensor mask_33_end_mask_0 = const()[name = string("mask_33_end_mask_0"), val = tensor([true, true, true, false])]; tensor mask_33 = slice_by_index(begin = mask_33_begin_0, end = concat_46, end_mask = mask_33_end_mask_0, x = mask_27)[name = string("mask_33")]; tensor scores_31 = select(a = var_37, b = _inversed_scores_29, cond = mask_33)[name = string("scores_31")]; tensor var_1245 = softmax(axis = var_55, x = scores_31)[name = string("op_1245")]; tensor input_173 = select(a = var_46, b = var_1245, cond = mask_33)[name = string("input_173")]; bool x_23_transpose_x_0 = const()[name = string("x_23_transpose_x_0"), val = bool(false)]; bool x_23_transpose_y_0 = const()[name = string("x_23_transpose_y_0"), val = bool(false)]; tensor v_31 = transpose(perm = v_31_perm_0, x = v_29)[name = string("transpose_118")]; tensor x_23 = matmul(transpose_x = x_23_transpose_x_0, transpose_y = x_23_transpose_y_0, x = input_173, y = v_31)[name = string("x_23")]; tensor var_1249_perm_0 = const()[name = string("op_1249_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_1251 = const()[name = string("op_1251"), val = tensor([1, -1, 512])]; tensor var_1249 = transpose(perm = var_1249_perm_0, x = x_23)[name = string("transpose_113")]; tensor input_175 = reshape(shape = var_1251, x = var_1249)[name = string("input_175")]; tensor input_177 = linear(bias = encoder_up_encoders_1_self_attn_linear_out_bias, weight = encoder_up_encoders_1_self_attn_linear_out_weight, x = input_175)[name = string("linear_55")]; tensor input_179 = add(x = input_171, y = input_177)[name = string("input_179")]; tensor input_181_axes_0 = const()[name = string("input_181_axes_0"), val = tensor([-1])]; tensor input_181 = layer_norm(axes = input_181_axes_0, beta = encoder_up_encoders_1_norm_ff_bias, epsilon = var_36, gamma = encoder_up_encoders_1_norm_ff_weight, x = input_179)[name = string("input_181")]; tensor input_183 = linear(bias = encoder_up_encoders_1_feed_forward_w_1_bias, weight = encoder_up_encoders_1_feed_forward_w_1_weight, x = input_181)[name = string("linear_56")]; tensor input_185 = silu(x = input_183)[name = string("input_185")]; tensor input_189 = linear(bias = encoder_up_encoders_1_feed_forward_w_2_bias, weight = encoder_up_encoders_1_feed_forward_w_2_weight, x = input_185)[name = string("linear_57")]; tensor input_191 = add(x = input_179, y = input_189)[name = string("input_191")]; tensor query_17_axes_0 = const()[name = string("query_17_axes_0"), val = tensor([-1])]; tensor query_17 = layer_norm(axes = query_17_axes_0, beta = encoder_up_encoders_2_norm_mha_bias, epsilon = var_36, gamma = encoder_up_encoders_2_norm_mha_weight, x = input_191)[name = string("query_17")]; tensor var_1294 = linear(bias = encoder_up_encoders_2_self_attn_linear_q_bias, weight = encoder_up_encoders_2_self_attn_linear_q_weight, x = query_17)[name = string("linear_58")]; tensor var_1295 = const()[name = string("op_1295"), val = tensor([1, -1, 8, 64])]; tensor q_49 = reshape(shape = var_1295, x = var_1294)[name = string("q_49")]; tensor var_1299 = linear(bias = encoder_up_encoders_2_self_attn_linear_k_bias, weight = encoder_up_encoders_2_self_attn_linear_k_weight, x = query_17)[name = string("linear_59")]; tensor var_1300 = const()[name = string("op_1300"), val = tensor([1, -1, 8, 64])]; tensor k_33 = reshape(shape = var_1300, x = var_1299)[name = string("k_33")]; tensor var_1304 = linear(bias = encoder_up_encoders_2_self_attn_linear_v_bias, weight = encoder_up_encoders_2_self_attn_linear_v_weight, x = query_17)[name = string("linear_60")]; tensor var_1305 = const()[name = string("op_1305"), val = tensor([1, -1, 8, 64])]; tensor v_33 = reshape(shape = var_1305, x = var_1304)[name = string("v_33")]; tensor v_35_perm_0 = const()[name = string("v_35_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_1313 = linear(bias = linear_4_bias_0, weight = encoder_up_encoders_2_self_attn_linear_pos_weight, x = input_149)[name = string("linear_61")]; tensor var_1314 = const()[name = string("op_1314"), val = tensor([1, -1, 8, 64])]; tensor p_33 = reshape(shape = var_1314, x = var_1313)[name = string("p_33")]; tensor const_30 = const()[name = string("const_30"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185410880)))]; tensor var_1318 = add(x = q_49, y = const_30)[name = string("op_1318")]; tensor const_31 = const()[name = string("const_31"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185412992)))]; tensor var_1321 = add(x = q_49, y = const_31)[name = string("op_1321")]; bool matrix_ac_17_transpose_x_0 = const()[name = string("matrix_ac_17_transpose_x_0"), val = bool(false)]; bool matrix_ac_17_transpose_y_0 = const()[name = string("matrix_ac_17_transpose_y_0"), val = bool(false)]; tensor transpose_92_perm_0 = const()[name = string("transpose_92_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_93_perm_0 = const()[name = string("transpose_93_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_93 = transpose(perm = transpose_93_perm_0, x = k_33)[name = string("transpose_110")]; tensor transpose_92 = transpose(perm = transpose_92_perm_0, x = var_1318)[name = string("transpose_111")]; tensor matrix_ac_17 = matmul(transpose_x = matrix_ac_17_transpose_x_0, transpose_y = matrix_ac_17_transpose_y_0, x = transpose_92, y = transpose_93)[name = string("matrix_ac_17")]; bool matrix_bd_33_transpose_x_0 = const()[name = string("matrix_bd_33_transpose_x_0"), val = bool(false)]; bool matrix_bd_33_transpose_y_0 = const()[name = string("matrix_bd_33_transpose_y_0"), val = bool(false)]; tensor transpose_94_perm_0 = const()[name = string("transpose_94_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_95_perm_0 = const()[name = string("transpose_95_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_95 = transpose(perm = transpose_95_perm_0, x = p_33)[name = string("transpose_108")]; tensor transpose_94 = transpose(perm = transpose_94_perm_0, x = var_1321)[name = string("transpose_109")]; tensor matrix_bd_33 = matmul(transpose_x = matrix_bd_33_transpose_x_0, transpose_y = matrix_bd_33_transpose_y_0, x = transpose_94, y = transpose_95)[name = string("matrix_bd_33")]; tensor var_1327_shape = shape(x = matrix_bd_33)[name = string("op_1327_shape")]; int32 gather_135 = const()[name = string("gather_135"), val = int32(1)]; int32 gather_136 = const()[name = string("gather_136"), val = int32(8)]; int32 gather_137_batch_dims_0 = const()[name = string("gather_137_batch_dims_0"), val = int32(0)]; bool gather_137_validate_indices_0 = const()[name = string("gather_137_validate_indices_0"), val = bool(false)]; int32 select_40 = const()[name = string("select_40"), val = int32(2)]; int32 gather_137_axis_1 = const()[name = string("gather_137_axis_1"), val = int32(0)]; int32 gather_137 = gather(axis = gather_137_axis_1, batch_dims = gather_137_batch_dims_0, indices = select_40, validate_indices = gather_137_validate_indices_0, x = var_1327_shape)[name = string("gather_137")]; int32 concat_47_axis_0 = const()[name = string("concat_47_axis_0"), val = int32(0)]; bool concat_47_interleave_0 = const()[name = string("concat_47_interleave_0"), val = bool(false)]; tensor concat_47 = concat(axis = concat_47_axis_0, interleave = concat_47_interleave_0, values = (gather_135, gather_136, gather_137, var_61))[name = string("concat_47")]; fp32 zero_pad_17_value_0 = const()[name = string("zero_pad_17_value_0"), val = fp32(0x0p+0)]; tensor zero_pad_17 = fill(shape = concat_47, value = zero_pad_17_value_0)[name = string("zero_pad_17")]; bool x_padded_33_interleave_0 = const()[name = string("x_padded_33_interleave_0"), val = bool(false)]; tensor x_padded_33 = concat(axis = var_55, interleave = x_padded_33_interleave_0, values = (zero_pad_17, matrix_bd_33))[name = string("x_padded_33")]; int32 gather_138 = const()[name = string("gather_138"), val = int32(1)]; int32 gather_139 = const()[name = string("gather_139"), val = int32(8)]; int32 gather_140_batch_dims_0 = const()[name = string("gather_140_batch_dims_0"), val = int32(0)]; bool gather_140_validate_indices_0 = const()[name = string("gather_140_validate_indices_0"), val = bool(false)]; int32 select_41 = const()[name = string("select_41"), val = int32(3)]; int32 gather_140_axis_1 = const()[name = string("gather_140_axis_1"), val = int32(0)]; int32 gather_140 = gather(axis = gather_140_axis_1, batch_dims = gather_140_batch_dims_0, indices = select_41, validate_indices = gather_140_validate_indices_0, x = var_1327_shape)[name = string("gather_140")]; int32 var_1338 = const()[name = string("op_1338"), val = int32(1)]; int32 var_1339 = add(x = gather_140, y = var_1338)[name = string("op_1339")]; int32 concat_48_axis_0 = const()[name = string("concat_48_axis_0"), val = int32(0)]; bool concat_48_interleave_0 = const()[name = string("concat_48_interleave_0"), val = bool(false)]; tensor concat_48 = concat(axis = concat_48_axis_0, interleave = concat_48_interleave_0, values = (gather_138, gather_139, var_1339, gather_137))[name = string("concat_48")]; tensor x_padded_35 = reshape(shape = concat_48, x = x_padded_33)[name = string("x_padded_35")]; tensor var_1346_begin_0 = const()[name = string("op_1346_begin_0"), val = tensor([0, 0, 1, 0])]; tensor var_1346_end_0 = const()[name = string("op_1346_end_0"), val = tensor([1, 8, 0, 0])]; tensor var_1346_end_mask_0 = const()[name = string("op_1346_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_1346 = slice_by_index(begin = var_1346_begin_0, end = var_1346_end_0, end_mask = var_1346_end_mask_0, x = x_padded_35)[name = string("op_1346")]; int32 gather_142 = const()[name = string("gather_142"), val = int32(1)]; int32 gather_143 = const()[name = string("gather_143"), val = int32(8)]; int32 concat_49_axis_0 = const()[name = string("concat_49_axis_0"), val = int32(0)]; bool concat_49_interleave_0 = const()[name = string("concat_49_interleave_0"), val = bool(false)]; tensor concat_49 = concat(axis = concat_49_axis_0, interleave = concat_49_interleave_0, values = (gather_142, gather_143, gather_137, gather_140))[name = string("concat_49")]; tensor var_1352 = reshape(shape = concat_49, x = var_1346)[name = string("op_1352")]; int32 floor_div_12 = floor_div(x = gather_140, y = var_53)[name = string("floor_div_12")]; string var_1355_dtype_0 = const()[name = string("op_1355_dtype_0"), val = string("fp32")]; fp32 var_1356_promoted = const()[name = string("op_1356_promoted"), val = fp32(0x1p+0)]; fp32 var_1355 = cast(dtype = var_1355_dtype_0, x = floor_div_12)[name = string("cast_106")]; fp32 var_1357 = add(x = var_1355, y = var_1356_promoted)[name = string("op_1357")]; string var_1358_dtype_0 = const()[name = string("op_1358_dtype_0"), val = string("int32")]; int32 concat_50_values0_0 = const()[name = string("concat_50_values0_0"), val = int32(1)]; int32 concat_50_values1_0 = const()[name = string("concat_50_values1_0"), val = int32(8)]; int32 concat_50_values2_0 = const()[name = string("concat_50_values2_0"), val = int32(0)]; int32 concat_50_axis_0 = const()[name = string("concat_50_axis_0"), val = int32(0)]; bool concat_50_interleave_0 = const()[name = string("concat_50_interleave_0"), val = bool(false)]; int32 var_1358 = cast(dtype = var_1358_dtype_0, x = var_1357)[name = string("cast_105")]; tensor concat_50 = concat(axis = concat_50_axis_0, interleave = concat_50_interleave_0, values = (concat_50_values0_0, concat_50_values1_0, concat_50_values2_0, var_1358))[name = string("concat_50")]; tensor matrix_bd_35_begin_0 = const()[name = string("matrix_bd_35_begin_0"), val = tensor([0, 0, 0, 0])]; tensor matrix_bd_35_end_mask_0 = const()[name = string("matrix_bd_35_end_mask_0"), val = tensor([true, true, true, false])]; tensor matrix_bd_35 = slice_by_index(begin = matrix_bd_35_begin_0, end = concat_50, end_mask = matrix_bd_35_end_mask_0, x = var_1352)[name = string("matrix_bd_35")]; tensor var_1363 = add(x = matrix_ac_17, y = matrix_bd_35)[name = string("op_1363")]; fp32 _inversed_scores_33_y_0 = const()[name = string("_inversed_scores_33_y_0"), val = fp32(0x1p-3)]; tensor _inversed_scores_33 = mul(x = var_1363, y = _inversed_scores_33_y_0)[name = string("_inversed_scores_33")]; tensor var_1369_shape = shape(x = _inversed_scores_33)[name = string("op_1369_shape")]; int32 gather_148_batch_dims_0 = const()[name = string("gather_148_batch_dims_0"), val = int32(0)]; bool gather_148_validate_indices_0 = const()[name = string("gather_148_validate_indices_0"), val = bool(false)]; int32 select_43 = const()[name = string("select_43"), val = int32(3)]; int32 gather_148_axis_1 = const()[name = string("gather_148_axis_1"), val = int32(0)]; int32 gather_148 = gather(axis = gather_148_axis_1, batch_dims = gather_148_batch_dims_0, indices = select_43, validate_indices = gather_148_validate_indices_0, x = var_1369_shape)[name = string("gather_148")]; int32 concat_51_values0_0 = const()[name = string("concat_51_values0_0"), val = int32(0)]; int32 concat_51_values1_0 = const()[name = string("concat_51_values1_0"), val = int32(1)]; int32 concat_51_values2_0 = const()[name = string("concat_51_values2_0"), val = int32(1)]; int32 concat_51_axis_0 = const()[name = string("concat_51_axis_0"), val = int32(0)]; bool concat_51_interleave_0 = const()[name = string("concat_51_interleave_0"), val = bool(false)]; tensor concat_51 = concat(axis = concat_51_axis_0, interleave = concat_51_interleave_0, values = (concat_51_values0_0, concat_51_values1_0, concat_51_values2_0, gather_148))[name = string("concat_51")]; tensor mask_37_begin_0 = const()[name = string("mask_37_begin_0"), val = tensor([0, 0, 0, 0])]; tensor mask_37_end_mask_0 = const()[name = string("mask_37_end_mask_0"), val = tensor([true, true, true, false])]; tensor mask_37 = slice_by_index(begin = mask_37_begin_0, end = concat_51, end_mask = mask_37_end_mask_0, x = mask_27)[name = string("mask_37")]; tensor scores_35 = select(a = var_37, b = _inversed_scores_33, cond = mask_37)[name = string("scores_35")]; tensor var_1375 = softmax(axis = var_55, x = scores_35)[name = string("op_1375")]; tensor input_193 = select(a = var_46, b = var_1375, cond = mask_37)[name = string("input_193")]; bool x_25_transpose_x_0 = const()[name = string("x_25_transpose_x_0"), val = bool(false)]; bool x_25_transpose_y_0 = const()[name = string("x_25_transpose_y_0"), val = bool(false)]; tensor v_35 = transpose(perm = v_35_perm_0, x = v_33)[name = string("transpose_112")]; tensor x_25 = matmul(transpose_x = x_25_transpose_x_0, transpose_y = x_25_transpose_y_0, x = input_193, y = v_35)[name = string("x_25")]; tensor var_1379_perm_0 = const()[name = string("op_1379_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_1381 = const()[name = string("op_1381"), val = tensor([1, -1, 512])]; tensor var_1379 = transpose(perm = var_1379_perm_0, x = x_25)[name = string("transpose_107")]; tensor input_195 = reshape(shape = var_1381, x = var_1379)[name = string("input_195")]; tensor input_197 = linear(bias = encoder_up_encoders_2_self_attn_linear_out_bias, weight = encoder_up_encoders_2_self_attn_linear_out_weight, x = input_195)[name = string("linear_62")]; tensor input_199 = add(x = input_191, y = input_197)[name = string("input_199")]; tensor input_201_axes_0 = const()[name = string("input_201_axes_0"), val = tensor([-1])]; tensor input_201 = layer_norm(axes = input_201_axes_0, beta = encoder_up_encoders_2_norm_ff_bias, epsilon = var_36, gamma = encoder_up_encoders_2_norm_ff_weight, x = input_199)[name = string("input_201")]; tensor input_203 = linear(bias = encoder_up_encoders_2_feed_forward_w_1_bias, weight = encoder_up_encoders_2_feed_forward_w_1_weight, x = input_201)[name = string("linear_63")]; tensor input_205 = silu(x = input_203)[name = string("input_205")]; tensor input_209 = linear(bias = encoder_up_encoders_2_feed_forward_w_2_bias, weight = encoder_up_encoders_2_feed_forward_w_2_weight, x = input_205)[name = string("linear_64")]; tensor input_211 = add(x = input_199, y = input_209)[name = string("input_211")]; tensor query_axes_0 = const()[name = string("query_axes_0"), val = tensor([-1])]; tensor query = layer_norm(axes = query_axes_0, beta = encoder_up_encoders_3_norm_mha_bias, epsilon = var_36, gamma = encoder_up_encoders_3_norm_mha_weight, x = input_211)[name = string("query")]; tensor var_1424 = linear(bias = encoder_up_encoders_3_self_attn_linear_q_bias, weight = encoder_up_encoders_3_self_attn_linear_q_weight, x = query)[name = string("linear_65")]; tensor var_1425 = const()[name = string("op_1425"), val = tensor([1, -1, 8, 64])]; tensor q_55 = reshape(shape = var_1425, x = var_1424)[name = string("q_55")]; tensor var_1429 = linear(bias = encoder_up_encoders_3_self_attn_linear_k_bias, weight = encoder_up_encoders_3_self_attn_linear_k_weight, x = query)[name = string("linear_66")]; tensor var_1430 = const()[name = string("op_1430"), val = tensor([1, -1, 8, 64])]; tensor k_37 = reshape(shape = var_1430, x = var_1429)[name = string("k_37")]; tensor var_1434 = linear(bias = encoder_up_encoders_3_self_attn_linear_v_bias, weight = encoder_up_encoders_3_self_attn_linear_v_weight, x = query)[name = string("linear_67")]; tensor var_1435 = const()[name = string("op_1435"), val = tensor([1, -1, 8, 64])]; tensor v_37 = reshape(shape = var_1435, x = var_1434)[name = string("v_37")]; tensor v_perm_0 = const()[name = string("v_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_1443 = linear(bias = linear_4_bias_0, weight = encoder_up_encoders_3_self_attn_linear_pos_weight, x = input_149)[name = string("linear_68")]; tensor var_1444 = const()[name = string("op_1444"), val = tensor([1, -1, 8, 64])]; tensor p_37 = reshape(shape = var_1444, x = var_1443)[name = string("p_37")]; tensor const_32 = const()[name = string("const_32"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185415104)))]; tensor var_1448 = add(x = q_55, y = const_32)[name = string("op_1448")]; tensor const_33 = const()[name = string("const_33"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185417216)))]; tensor var_1451 = add(x = q_55, y = const_33)[name = string("op_1451")]; bool matrix_ac_transpose_x_0 = const()[name = string("matrix_ac_transpose_x_0"), val = bool(false)]; bool matrix_ac_transpose_y_0 = const()[name = string("matrix_ac_transpose_y_0"), val = bool(false)]; tensor transpose_96_perm_0 = const()[name = string("transpose_96_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_97_perm_0 = const()[name = string("transpose_97_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_97 = transpose(perm = transpose_97_perm_0, x = k_37)[name = string("transpose_104")]; tensor transpose_96 = transpose(perm = transpose_96_perm_0, x = var_1448)[name = string("transpose_105")]; tensor matrix_ac = matmul(transpose_x = matrix_ac_transpose_x_0, transpose_y = matrix_ac_transpose_y_0, x = transpose_96, y = transpose_97)[name = string("matrix_ac")]; bool matrix_bd_37_transpose_x_0 = const()[name = string("matrix_bd_37_transpose_x_0"), val = bool(false)]; bool matrix_bd_37_transpose_y_0 = const()[name = string("matrix_bd_37_transpose_y_0"), val = bool(false)]; tensor transpose_98_perm_0 = const()[name = string("transpose_98_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_99_perm_0 = const()[name = string("transpose_99_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_99 = transpose(perm = transpose_99_perm_0, x = p_37)[name = string("transpose_102")]; tensor transpose_98 = transpose(perm = transpose_98_perm_0, x = var_1451)[name = string("transpose_103")]; tensor matrix_bd_37 = matmul(transpose_x = matrix_bd_37_transpose_x_0, transpose_y = matrix_bd_37_transpose_y_0, x = transpose_98, y = transpose_99)[name = string("matrix_bd_37")]; tensor var_1457_shape = shape(x = matrix_bd_37)[name = string("op_1457_shape")]; int32 gather_151 = const()[name = string("gather_151"), val = int32(1)]; int32 gather_152 = const()[name = string("gather_152"), val = int32(8)]; int32 gather_153_batch_dims_0 = const()[name = string("gather_153_batch_dims_0"), val = int32(0)]; bool gather_153_validate_indices_0 = const()[name = string("gather_153_validate_indices_0"), val = bool(false)]; int32 select_44 = const()[name = string("select_44"), val = int32(2)]; int32 gather_153_axis_1 = const()[name = string("gather_153_axis_1"), val = int32(0)]; int32 gather_153 = gather(axis = gather_153_axis_1, batch_dims = gather_153_batch_dims_0, indices = select_44, validate_indices = gather_153_validate_indices_0, x = var_1457_shape)[name = string("gather_153")]; int32 concat_52_axis_0 = const()[name = string("concat_52_axis_0"), val = int32(0)]; bool concat_52_interleave_0 = const()[name = string("concat_52_interleave_0"), val = bool(false)]; tensor concat_52 = concat(axis = concat_52_axis_0, interleave = concat_52_interleave_0, values = (gather_151, gather_152, gather_153, var_61))[name = string("concat_52")]; fp32 zero_pad_value_0 = const()[name = string("zero_pad_value_0"), val = fp32(0x0p+0)]; tensor zero_pad = fill(shape = concat_52, value = zero_pad_value_0)[name = string("zero_pad")]; bool x_padded_37_interleave_0 = const()[name = string("x_padded_37_interleave_0"), val = bool(false)]; tensor x_padded_37 = concat(axis = var_55, interleave = x_padded_37_interleave_0, values = (zero_pad, matrix_bd_37))[name = string("x_padded_37")]; int32 gather_154 = const()[name = string("gather_154"), val = int32(1)]; int32 gather_155 = const()[name = string("gather_155"), val = int32(8)]; int32 gather_156_batch_dims_0 = const()[name = string("gather_156_batch_dims_0"), val = int32(0)]; bool gather_156_validate_indices_0 = const()[name = string("gather_156_validate_indices_0"), val = bool(false)]; int32 select_45 = const()[name = string("select_45"), val = int32(3)]; int32 gather_156_axis_1 = const()[name = string("gather_156_axis_1"), val = int32(0)]; int32 gather_156 = gather(axis = gather_156_axis_1, batch_dims = gather_156_batch_dims_0, indices = select_45, validate_indices = gather_156_validate_indices_0, x = var_1457_shape)[name = string("gather_156")]; int32 var_1468 = const()[name = string("op_1468"), val = int32(1)]; int32 var_1469 = add(x = gather_156, y = var_1468)[name = string("op_1469")]; int32 concat_53_axis_0 = const()[name = string("concat_53_axis_0"), val = int32(0)]; bool concat_53_interleave_0 = const()[name = string("concat_53_interleave_0"), val = bool(false)]; tensor concat_53 = concat(axis = concat_53_axis_0, interleave = concat_53_interleave_0, values = (gather_154, gather_155, var_1469, gather_153))[name = string("concat_53")]; tensor x_padded = reshape(shape = concat_53, x = x_padded_37)[name = string("x_padded")]; tensor var_1476_begin_0 = const()[name = string("op_1476_begin_0"), val = tensor([0, 0, 1, 0])]; tensor var_1476_end_0 = const()[name = string("op_1476_end_0"), val = tensor([1, 8, 0, 0])]; tensor var_1476_end_mask_0 = const()[name = string("op_1476_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_1476 = slice_by_index(begin = var_1476_begin_0, end = var_1476_end_0, end_mask = var_1476_end_mask_0, x = x_padded)[name = string("op_1476")]; int32 gather_158 = const()[name = string("gather_158"), val = int32(1)]; int32 gather_159 = const()[name = string("gather_159"), val = int32(8)]; int32 concat_54_axis_0 = const()[name = string("concat_54_axis_0"), val = int32(0)]; bool concat_54_interleave_0 = const()[name = string("concat_54_interleave_0"), val = bool(false)]; tensor concat_54 = concat(axis = concat_54_axis_0, interleave = concat_54_interleave_0, values = (gather_158, gather_159, gather_153, gather_156))[name = string("concat_54")]; tensor var_1482 = reshape(shape = concat_54, x = var_1476)[name = string("op_1482")]; int32 floor_div_13 = floor_div(x = gather_156, y = var_53)[name = string("floor_div_13")]; string var_1485_dtype_0 = const()[name = string("op_1485_dtype_0"), val = string("fp32")]; fp32 var_1486_promoted = const()[name = string("op_1486_promoted"), val = fp32(0x1p+0)]; fp32 var_1485 = cast(dtype = var_1485_dtype_0, x = floor_div_13)[name = string("cast_104")]; fp32 var_1487 = add(x = var_1485, y = var_1486_promoted)[name = string("op_1487")]; string var_1488_dtype_0 = const()[name = string("op_1488_dtype_0"), val = string("int32")]; int32 concat_55_values0_0 = const()[name = string("concat_55_values0_0"), val = int32(1)]; int32 concat_55_values1_0 = const()[name = string("concat_55_values1_0"), val = int32(8)]; int32 concat_55_values2_0 = const()[name = string("concat_55_values2_0"), val = int32(0)]; int32 concat_55_axis_0 = const()[name = string("concat_55_axis_0"), val = int32(0)]; bool concat_55_interleave_0 = const()[name = string("concat_55_interleave_0"), val = bool(false)]; int32 var_1488 = cast(dtype = var_1488_dtype_0, x = var_1487)[name = string("cast_103")]; tensor concat_55 = concat(axis = concat_55_axis_0, interleave = concat_55_interleave_0, values = (concat_55_values0_0, concat_55_values1_0, concat_55_values2_0, var_1488))[name = string("concat_55")]; tensor matrix_bd_begin_0 = const()[name = string("matrix_bd_begin_0"), val = tensor([0, 0, 0, 0])]; tensor matrix_bd_end_mask_0 = const()[name = string("matrix_bd_end_mask_0"), val = tensor([true, true, true, false])]; tensor matrix_bd = slice_by_index(begin = matrix_bd_begin_0, end = concat_55, end_mask = matrix_bd_end_mask_0, x = var_1482)[name = string("matrix_bd")]; tensor var_1493 = add(x = matrix_ac, y = matrix_bd)[name = string("op_1493")]; fp32 _inversed_scores_37_y_0 = const()[name = string("_inversed_scores_37_y_0"), val = fp32(0x1p-3)]; tensor _inversed_scores_37 = mul(x = var_1493, y = _inversed_scores_37_y_0)[name = string("_inversed_scores_37")]; tensor var_1499_shape = shape(x = _inversed_scores_37)[name = string("op_1499_shape")]; int32 gather_164_batch_dims_0 = const()[name = string("gather_164_batch_dims_0"), val = int32(0)]; bool gather_164_validate_indices_0 = const()[name = string("gather_164_validate_indices_0"), val = bool(false)]; int32 select_47 = const()[name = string("select_47"), val = int32(3)]; int32 gather_164_axis_1 = const()[name = string("gather_164_axis_1"), val = int32(0)]; int32 gather_164 = gather(axis = gather_164_axis_1, batch_dims = gather_164_batch_dims_0, indices = select_47, validate_indices = gather_164_validate_indices_0, x = var_1499_shape)[name = string("gather_164")]; int32 concat_56_values0_0 = const()[name = string("concat_56_values0_0"), val = int32(0)]; int32 concat_56_values1_0 = const()[name = string("concat_56_values1_0"), val = int32(1)]; int32 concat_56_values2_0 = const()[name = string("concat_56_values2_0"), val = int32(1)]; int32 concat_56_axis_0 = const()[name = string("concat_56_axis_0"), val = int32(0)]; bool concat_56_interleave_0 = const()[name = string("concat_56_interleave_0"), val = bool(false)]; tensor concat_56 = concat(axis = concat_56_axis_0, interleave = concat_56_interleave_0, values = (concat_56_values0_0, concat_56_values1_0, concat_56_values2_0, gather_164))[name = string("concat_56")]; tensor mask_begin_0 = const()[name = string("mask_begin_0"), val = tensor([0, 0, 0, 0])]; tensor mask_end_mask_0 = const()[name = string("mask_end_mask_0"), val = tensor([true, true, true, false])]; tensor mask = slice_by_index(begin = mask_begin_0, end = concat_56, end_mask = mask_end_mask_0, x = mask_27)[name = string("mask")]; tensor scores = select(a = var_37, b = _inversed_scores_37, cond = mask)[name = string("scores")]; tensor var_1505 = softmax(axis = var_55, x = scores)[name = string("op_1505")]; tensor input_213 = select(a = var_46, b = var_1505, cond = mask)[name = string("input_213")]; bool x_transpose_x_0 = const()[name = string("x_transpose_x_0"), val = bool(false)]; bool x_transpose_y_0 = const()[name = string("x_transpose_y_0"), val = bool(false)]; tensor v = transpose(perm = v_perm_0, x = v_37)[name = string("transpose_106")]; tensor x = matmul(transpose_x = x_transpose_x_0, transpose_y = x_transpose_y_0, x = input_213, y = v)[name = string("x")]; tensor var_1509_perm_0 = const()[name = string("op_1509_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_1511 = const()[name = string("op_1511"), val = tensor([1, -1, 512])]; tensor var_1509 = transpose(perm = var_1509_perm_0, x = x)[name = string("transpose_101")]; tensor input_215 = reshape(shape = var_1511, x = var_1509)[name = string("input_215")]; tensor input_217 = linear(bias = encoder_up_encoders_3_self_attn_linear_out_bias, weight = encoder_up_encoders_3_self_attn_linear_out_weight, x = input_215)[name = string("linear_69")]; tensor input_219 = add(x = input_211, y = input_217)[name = string("input_219")]; tensor input_221_axes_0 = const()[name = string("input_221_axes_0"), val = tensor([-1])]; tensor input_221 = layer_norm(axes = input_221_axes_0, beta = encoder_up_encoders_3_norm_ff_bias, epsilon = var_36, gamma = encoder_up_encoders_3_norm_ff_weight, x = input_219)[name = string("input_221")]; tensor input_223 = linear(bias = encoder_up_encoders_3_feed_forward_w_1_bias, weight = encoder_up_encoders_3_feed_forward_w_1_weight, x = input_221)[name = string("linear_70")]; tensor input_225 = silu(x = input_223)[name = string("input_225")]; tensor input_229 = linear(bias = encoder_up_encoders_3_feed_forward_w_2_bias, weight = encoder_up_encoders_3_feed_forward_w_2_weight, x = input_225)[name = string("linear_71")]; tensor input_231 = add(x = input_219, y = input_229)[name = string("input_231")]; tensor input_axes_0 = const()[name = string("input_axes_0"), val = tensor([-1])]; tensor input = layer_norm(axes = input_axes_0, beta = encoder_after_norm_bias, epsilon = var_50, gamma = encoder_after_norm_weight, x = input_231)[name = string("input")]; tensor var_1542 = linear(bias = encoder_proj_bias, weight = encoder_proj_weight, x = input)[name = string("linear_72")]; tensor var_1545_perm_0 = const()[name = string("op_1545_perm_0"), val = tensor([0, 2, 1])]; tensor encoder_proj = transpose(perm = var_1545_perm_0, x = var_1542)[name = string("transpose_100")]; } -> (encoder_proj); }