diff --git "a/original/compiled/TextEncoder.mlmodelc/model.mil" "b/original/compiled/TextEncoder.mlmodelc/model.mil" --- "a/original/compiled/TextEncoder.mlmodelc/model.mil" +++ "b/original/compiled/TextEncoder.mlmodelc/model.mil" @@ -2,882 +2,886 @@ program(1.0) [buildInfo = dict, tensor>({{"coremlc-component-MIL", "5.33.4"}, {"coremlc-version", "1436.100.10"}, {"coremltools-component-torch", "2.1.0.dev20230718"}, {"coremltools-version", "7.0b1"}})] { func main(tensor input_ids) { - tensor text_encoder_text_model_embeddings_token_embedding_weight = const()[name = tensor("text_encoder_text_model_embeddings_token_embedding_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; - tensor text_encoder_text_model_encoder_layers_0_layer_norm1_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_0_layer_norm1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151781504)))]; - tensor text_encoder_text_model_encoder_layers_0_layer_norm1_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_0_layer_norm1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151784640)))]; - tensor text_encoder_text_model_encoder_layers_0_self_attn_q_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_q_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151787776)))]; - tensor text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151790912)))]; - tensor text_encoder_text_model_encoder_layers_0_self_attn_k_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_k_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(154150272)))]; - tensor text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(154153408)))]; - tensor text_encoder_text_model_encoder_layers_0_self_attn_v_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_v_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(156512768)))]; - tensor text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(156515904)))]; - tensor text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158875264)))]; - tensor text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158878400)))]; - tensor text_encoder_text_model_encoder_layers_0_layer_norm2_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_0_layer_norm2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161237760)))]; - tensor text_encoder_text_model_encoder_layers_0_layer_norm2_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_0_layer_norm2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161240896)))]; - tensor text_encoder_text_model_encoder_layers_0_mlp_fc1_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_0_mlp_fc1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161244032)))]; - tensor text_encoder_text_model_encoder_layers_0_mlp_fc1_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_0_mlp_fc1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161256384)))]; - tensor text_encoder_text_model_encoder_layers_0_mlp_fc2_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_0_mlp_fc2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(170693632)))]; - tensor text_encoder_text_model_encoder_layers_0_mlp_fc2_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_0_mlp_fc2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(170696768)))]; - tensor text_encoder_text_model_encoder_layers_1_layer_norm1_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_1_layer_norm1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(180134016)))]; - tensor text_encoder_text_model_encoder_layers_1_layer_norm1_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_1_layer_norm1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(180137152)))]; - tensor text_encoder_text_model_encoder_layers_1_self_attn_q_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_q_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(180140288)))]; - tensor text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(180143424)))]; - tensor text_encoder_text_model_encoder_layers_1_self_attn_k_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_k_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(182502784)))]; - tensor text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(182505920)))]; - tensor text_encoder_text_model_encoder_layers_1_self_attn_v_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_v_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184865280)))]; - tensor text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184868416)))]; - tensor text_encoder_text_model_encoder_layers_1_self_attn_out_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_out_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(187227776)))]; - tensor text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(187230912)))]; - tensor text_encoder_text_model_encoder_layers_1_layer_norm2_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_1_layer_norm2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(189590272)))]; - tensor text_encoder_text_model_encoder_layers_1_layer_norm2_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_1_layer_norm2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(189593408)))]; - tensor text_encoder_text_model_encoder_layers_1_mlp_fc1_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_1_mlp_fc1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(189596544)))]; - tensor text_encoder_text_model_encoder_layers_1_mlp_fc1_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_1_mlp_fc1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(189608896)))]; - tensor text_encoder_text_model_encoder_layers_1_mlp_fc2_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_1_mlp_fc2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(199046144)))]; - tensor text_encoder_text_model_encoder_layers_1_mlp_fc2_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_1_mlp_fc2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(199049280)))]; - tensor text_encoder_text_model_encoder_layers_2_layer_norm1_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_2_layer_norm1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(208486528)))]; - tensor text_encoder_text_model_encoder_layers_2_layer_norm1_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_2_layer_norm1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(208489664)))]; - tensor text_encoder_text_model_encoder_layers_2_self_attn_q_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_q_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(208492800)))]; - tensor text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(208495936)))]; - tensor text_encoder_text_model_encoder_layers_2_self_attn_k_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_k_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(210855296)))]; - tensor text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(210858432)))]; - tensor text_encoder_text_model_encoder_layers_2_self_attn_v_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_v_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(213217792)))]; - tensor text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(213220928)))]; - tensor text_encoder_text_model_encoder_layers_2_self_attn_out_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_out_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215580288)))]; - tensor text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215583424)))]; - tensor text_encoder_text_model_encoder_layers_2_layer_norm2_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_2_layer_norm2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217942784)))]; - tensor text_encoder_text_model_encoder_layers_2_layer_norm2_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_2_layer_norm2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217945920)))]; - tensor text_encoder_text_model_encoder_layers_2_mlp_fc1_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_2_mlp_fc1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217949056)))]; - tensor text_encoder_text_model_encoder_layers_2_mlp_fc1_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_2_mlp_fc1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217961408)))]; - tensor text_encoder_text_model_encoder_layers_2_mlp_fc2_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_2_mlp_fc2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(227398656)))]; - tensor text_encoder_text_model_encoder_layers_2_mlp_fc2_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_2_mlp_fc2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(227401792)))]; - tensor text_encoder_text_model_encoder_layers_3_layer_norm1_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_3_layer_norm1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236839040)))]; - tensor text_encoder_text_model_encoder_layers_3_layer_norm1_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_3_layer_norm1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236842176)))]; - tensor text_encoder_text_model_encoder_layers_3_self_attn_q_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_q_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236845312)))]; - tensor text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236848448)))]; - tensor text_encoder_text_model_encoder_layers_3_self_attn_k_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_k_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(239207808)))]; - tensor text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(239210944)))]; - tensor text_encoder_text_model_encoder_layers_3_self_attn_v_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_v_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(241570304)))]; - tensor text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(241573440)))]; - tensor text_encoder_text_model_encoder_layers_3_self_attn_out_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_out_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(243932800)))]; - tensor text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(243935936)))]; - tensor text_encoder_text_model_encoder_layers_3_layer_norm2_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_3_layer_norm2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(246295296)))]; - tensor text_encoder_text_model_encoder_layers_3_layer_norm2_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_3_layer_norm2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(246298432)))]; - tensor text_encoder_text_model_encoder_layers_3_mlp_fc1_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_3_mlp_fc1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(246301568)))]; - tensor text_encoder_text_model_encoder_layers_3_mlp_fc1_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_3_mlp_fc1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(246313920)))]; - tensor text_encoder_text_model_encoder_layers_3_mlp_fc2_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_3_mlp_fc2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(255751168)))]; - tensor text_encoder_text_model_encoder_layers_3_mlp_fc2_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_3_mlp_fc2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(255754304)))]; - tensor text_encoder_text_model_encoder_layers_4_layer_norm1_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_4_layer_norm1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265191552)))]; - tensor text_encoder_text_model_encoder_layers_4_layer_norm1_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_4_layer_norm1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265194688)))]; - tensor text_encoder_text_model_encoder_layers_4_self_attn_q_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_q_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265197824)))]; - tensor text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265200960)))]; - tensor text_encoder_text_model_encoder_layers_4_self_attn_k_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_k_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(267560320)))]; - tensor text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(267563456)))]; - tensor text_encoder_text_model_encoder_layers_4_self_attn_v_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_v_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(269922816)))]; - tensor text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(269925952)))]; - tensor text_encoder_text_model_encoder_layers_4_self_attn_out_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_out_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(272285312)))]; - tensor text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(272288448)))]; - tensor text_encoder_text_model_encoder_layers_4_layer_norm2_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_4_layer_norm2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(274647808)))]; - tensor text_encoder_text_model_encoder_layers_4_layer_norm2_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_4_layer_norm2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(274650944)))]; - tensor text_encoder_text_model_encoder_layers_4_mlp_fc1_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_4_mlp_fc1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(274654080)))]; - tensor text_encoder_text_model_encoder_layers_4_mlp_fc1_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_4_mlp_fc1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(274666432)))]; - tensor text_encoder_text_model_encoder_layers_4_mlp_fc2_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_4_mlp_fc2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(284103680)))]; - tensor text_encoder_text_model_encoder_layers_4_mlp_fc2_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_4_mlp_fc2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(284106816)))]; - tensor text_encoder_text_model_encoder_layers_5_layer_norm1_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_5_layer_norm1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293544064)))]; - tensor text_encoder_text_model_encoder_layers_5_layer_norm1_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_5_layer_norm1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293547200)))]; - tensor text_encoder_text_model_encoder_layers_5_self_attn_q_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_q_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293550336)))]; - tensor text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293553472)))]; - tensor text_encoder_text_model_encoder_layers_5_self_attn_k_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_k_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(295912832)))]; - tensor text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(295915968)))]; - tensor text_encoder_text_model_encoder_layers_5_self_attn_v_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_v_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(298275328)))]; - tensor text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(298278464)))]; - tensor text_encoder_text_model_encoder_layers_5_self_attn_out_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_out_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300637824)))]; - tensor text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300640960)))]; - tensor text_encoder_text_model_encoder_layers_5_layer_norm2_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_5_layer_norm2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303000320)))]; - tensor text_encoder_text_model_encoder_layers_5_layer_norm2_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_5_layer_norm2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303003456)))]; - tensor text_encoder_text_model_encoder_layers_5_mlp_fc1_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_5_mlp_fc1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303006592)))]; - tensor text_encoder_text_model_encoder_layers_5_mlp_fc1_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_5_mlp_fc1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303018944)))]; - tensor text_encoder_text_model_encoder_layers_5_mlp_fc2_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_5_mlp_fc2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(312456192)))]; - tensor text_encoder_text_model_encoder_layers_5_mlp_fc2_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_5_mlp_fc2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(312459328)))]; - tensor text_encoder_text_model_encoder_layers_6_layer_norm1_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_6_layer_norm1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(321896576)))]; - tensor text_encoder_text_model_encoder_layers_6_layer_norm1_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_6_layer_norm1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(321899712)))]; - tensor text_encoder_text_model_encoder_layers_6_self_attn_q_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_q_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(321902848)))]; - tensor text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(321905984)))]; - tensor text_encoder_text_model_encoder_layers_6_self_attn_k_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_k_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(324265344)))]; - tensor text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(324268480)))]; - tensor text_encoder_text_model_encoder_layers_6_self_attn_v_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_v_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(326627840)))]; - tensor text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(326630976)))]; - tensor text_encoder_text_model_encoder_layers_6_self_attn_out_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_out_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(328990336)))]; - tensor text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(328993472)))]; - tensor text_encoder_text_model_encoder_layers_6_layer_norm2_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_6_layer_norm2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(331352832)))]; - tensor text_encoder_text_model_encoder_layers_6_layer_norm2_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_6_layer_norm2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(331355968)))]; - tensor text_encoder_text_model_encoder_layers_6_mlp_fc1_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_6_mlp_fc1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(331359104)))]; - tensor text_encoder_text_model_encoder_layers_6_mlp_fc1_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_6_mlp_fc1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(331371456)))]; - tensor text_encoder_text_model_encoder_layers_6_mlp_fc2_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_6_mlp_fc2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(340808704)))]; - tensor text_encoder_text_model_encoder_layers_6_mlp_fc2_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_6_mlp_fc2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(340811840)))]; - tensor text_encoder_text_model_encoder_layers_7_layer_norm1_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_7_layer_norm1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(350249088)))]; - tensor text_encoder_text_model_encoder_layers_7_layer_norm1_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_7_layer_norm1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(350252224)))]; - tensor text_encoder_text_model_encoder_layers_7_self_attn_q_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_q_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(350255360)))]; - tensor text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(350258496)))]; - tensor text_encoder_text_model_encoder_layers_7_self_attn_k_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_k_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(352617856)))]; - tensor text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(352620992)))]; - tensor text_encoder_text_model_encoder_layers_7_self_attn_v_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_v_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(354980352)))]; - tensor text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(354983488)))]; - tensor text_encoder_text_model_encoder_layers_7_self_attn_out_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_out_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(357342848)))]; - tensor text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(357345984)))]; - tensor text_encoder_text_model_encoder_layers_7_layer_norm2_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_7_layer_norm2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(359705344)))]; - tensor text_encoder_text_model_encoder_layers_7_layer_norm2_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_7_layer_norm2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(359708480)))]; - tensor text_encoder_text_model_encoder_layers_7_mlp_fc1_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_7_mlp_fc1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(359711616)))]; - tensor text_encoder_text_model_encoder_layers_7_mlp_fc1_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_7_mlp_fc1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(359723968)))]; - tensor text_encoder_text_model_encoder_layers_7_mlp_fc2_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_7_mlp_fc2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(369161216)))]; - tensor text_encoder_text_model_encoder_layers_7_mlp_fc2_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_7_mlp_fc2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(369164352)))]; - tensor text_encoder_text_model_encoder_layers_8_layer_norm1_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_8_layer_norm1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(378601600)))]; - tensor text_encoder_text_model_encoder_layers_8_layer_norm1_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_8_layer_norm1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(378604736)))]; - tensor text_encoder_text_model_encoder_layers_8_self_attn_q_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_q_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(378607872)))]; - tensor text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(378611008)))]; - tensor text_encoder_text_model_encoder_layers_8_self_attn_k_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_k_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(380970368)))]; - tensor text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(380973504)))]; - tensor text_encoder_text_model_encoder_layers_8_self_attn_v_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_v_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(383332864)))]; - tensor text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(383336000)))]; - tensor text_encoder_text_model_encoder_layers_8_self_attn_out_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_out_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(385695360)))]; - tensor text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(385698496)))]; - tensor text_encoder_text_model_encoder_layers_8_layer_norm2_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_8_layer_norm2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(388057856)))]; - tensor text_encoder_text_model_encoder_layers_8_layer_norm2_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_8_layer_norm2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(388060992)))]; - tensor text_encoder_text_model_encoder_layers_8_mlp_fc1_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_8_mlp_fc1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(388064128)))]; - tensor text_encoder_text_model_encoder_layers_8_mlp_fc1_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_8_mlp_fc1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(388076480)))]; - tensor text_encoder_text_model_encoder_layers_8_mlp_fc2_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_8_mlp_fc2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(397513728)))]; - tensor text_encoder_text_model_encoder_layers_8_mlp_fc2_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_8_mlp_fc2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(397516864)))]; - tensor text_encoder_text_model_encoder_layers_9_layer_norm1_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_9_layer_norm1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(406954112)))]; - tensor text_encoder_text_model_encoder_layers_9_layer_norm1_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_9_layer_norm1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(406957248)))]; - tensor text_encoder_text_model_encoder_layers_9_self_attn_q_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_q_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(406960384)))]; - tensor text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(406963520)))]; - tensor text_encoder_text_model_encoder_layers_9_self_attn_k_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_k_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(409322880)))]; - tensor text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(409326016)))]; - tensor text_encoder_text_model_encoder_layers_9_self_attn_v_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_v_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411685376)))]; - tensor text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411688512)))]; - tensor text_encoder_text_model_encoder_layers_9_self_attn_out_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_out_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(414047872)))]; - tensor text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(414051008)))]; - tensor text_encoder_text_model_encoder_layers_9_layer_norm2_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_9_layer_norm2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(416410368)))]; - tensor text_encoder_text_model_encoder_layers_9_layer_norm2_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_9_layer_norm2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(416413504)))]; - tensor text_encoder_text_model_encoder_layers_9_mlp_fc1_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_9_mlp_fc1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(416416640)))]; - tensor text_encoder_text_model_encoder_layers_9_mlp_fc1_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_9_mlp_fc1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(416428992)))]; - tensor text_encoder_text_model_encoder_layers_9_mlp_fc2_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_9_mlp_fc2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(425866240)))]; - tensor text_encoder_text_model_encoder_layers_9_mlp_fc2_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_9_mlp_fc2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(425869376)))]; - tensor text_encoder_text_model_encoder_layers_10_layer_norm1_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_10_layer_norm1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(435306624)))]; - tensor text_encoder_text_model_encoder_layers_10_layer_norm1_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_10_layer_norm1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(435309760)))]; - tensor text_encoder_text_model_encoder_layers_10_self_attn_q_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_q_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(435312896)))]; - tensor text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(435316032)))]; - tensor text_encoder_text_model_encoder_layers_10_self_attn_k_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_k_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(437675392)))]; - tensor text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(437678528)))]; - tensor text_encoder_text_model_encoder_layers_10_self_attn_v_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_v_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(440037888)))]; - tensor text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(440041024)))]; - tensor text_encoder_text_model_encoder_layers_10_self_attn_out_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_out_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(442400384)))]; - tensor text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(442403520)))]; - tensor text_encoder_text_model_encoder_layers_10_layer_norm2_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_10_layer_norm2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(444762880)))]; - tensor text_encoder_text_model_encoder_layers_10_layer_norm2_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_10_layer_norm2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(444766016)))]; - tensor text_encoder_text_model_encoder_layers_10_mlp_fc1_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_10_mlp_fc1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(444769152)))]; - tensor text_encoder_text_model_encoder_layers_10_mlp_fc1_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_10_mlp_fc1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(444781504)))]; - tensor text_encoder_text_model_encoder_layers_10_mlp_fc2_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_10_mlp_fc2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(454218752)))]; - tensor text_encoder_text_model_encoder_layers_10_mlp_fc2_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_10_mlp_fc2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(454221888)))]; - tensor text_encoder_text_model_encoder_layers_11_layer_norm1_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_11_layer_norm1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(463659136)))]; - tensor text_encoder_text_model_encoder_layers_11_layer_norm1_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_11_layer_norm1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(463662272)))]; - tensor text_encoder_text_model_encoder_layers_11_self_attn_q_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_q_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(463665408)))]; - tensor text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(463668544)))]; - tensor text_encoder_text_model_encoder_layers_11_self_attn_k_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_k_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(466027904)))]; - tensor text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(466031040)))]; - tensor text_encoder_text_model_encoder_layers_11_self_attn_v_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_v_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(468390400)))]; - tensor text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(468393536)))]; - tensor text_encoder_text_model_encoder_layers_11_self_attn_out_proj_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_out_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(470752896)))]; - tensor text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(470756032)))]; - tensor text_encoder_text_model_encoder_layers_11_layer_norm2_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_11_layer_norm2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(473115392)))]; - tensor text_encoder_text_model_encoder_layers_11_layer_norm2_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_11_layer_norm2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(473118528)))]; - tensor text_encoder_text_model_encoder_layers_11_mlp_fc1_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_11_mlp_fc1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(473121664)))]; - tensor text_encoder_text_model_encoder_layers_11_mlp_fc1_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_11_mlp_fc1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(473134016)))]; - tensor text_encoder_text_model_encoder_layers_11_mlp_fc2_bias = const()[name = tensor("text_encoder_text_model_encoder_layers_11_mlp_fc2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(482571264)))]; - tensor text_encoder_text_model_encoder_layers_11_mlp_fc2_weight = const()[name = tensor("text_encoder_text_model_encoder_layers_11_mlp_fc2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(482574400)))]; - tensor text_encoder_text_model_final_layer_norm_bias = const()[name = tensor("text_encoder_text_model_final_layer_norm_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(492011648)))]; - tensor text_encoder_text_model_final_layer_norm_weight = const()[name = tensor("text_encoder_text_model_final_layer_norm_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(492014784)))]; tensor var_5 = const()[name = tensor("op_5"), val = tensor(-1)]; - tensor var_12 = const()[name = tensor("op_12"), val = tensor(0x1.4f8b58p-17)]; tensor inputs_embeds_axis_0 = const()[name = tensor("inputs_embeds_axis_0"), val = tensor(0)]; tensor inputs_embeds_batch_dims_0 = const()[name = tensor("inputs_embeds_batch_dims_0"), val = tensor(0)]; - tensor inputs_embeds = gather(axis = inputs_embeds_axis_0, batch_dims = inputs_embeds_batch_dims_0, indices = input_ids, x = text_encoder_text_model_embeddings_token_embedding_weight)[name = tensor("inputs_embeds")]; - tensor position_embeddings = const()[name = tensor("position_embeddings"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(492017920)))]; - tensor input_3 = add(x = inputs_embeds, y = position_embeddings)[name = tensor("input_3")]; - tensor causal_attention_mask = const()[name = tensor("causal_attention_mask"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(492254528)))]; + tensor text_encoder_text_model_embeddings_token_embedding_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_embeddings_token_embedding_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; + tensor inputs_embeds_cast = gather(axis = inputs_embeds_axis_0, batch_dims = inputs_embeds_batch_dims_0, indices = input_ids, x = text_encoder_text_model_embeddings_token_embedding_weight_to_fp16)[name = tensor("inputs_embeds_cast")]; + tensor position_embeddings_to_fp16 = const()[name = tensor("position_embeddings_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(75890816)))]; + tensor input_3_cast = add(x = inputs_embeds_cast, y = position_embeddings_to_fp16)[name = tensor("input_3_cast")]; tensor hidden_states_1_axes_0 = const()[name = tensor("hidden_states_1_axes_0"), val = tensor([-1])]; - tensor hidden_states_1 = layer_norm(axes = hidden_states_1_axes_0, beta = text_encoder_text_model_encoder_layers_0_layer_norm1_bias, epsilon = var_12, gamma = text_encoder_text_model_encoder_layers_0_layer_norm1_weight, x = input_3)[name = tensor("hidden_states_1")]; - tensor var_85 = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_q_proj_bias, weight = text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight, x = hidden_states_1)[name = tensor("op_85")]; - tensor var_86 = const()[name = tensor("op_86"), val = tensor(0x1p-3)]; - tensor tensor_5 = mul(x = var_85, y = var_86)[name = tensor("tensor_5")]; - tensor tensor_1 = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_k_proj_bias, weight = text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight, x = hidden_states_1)[name = tensor("tensor_1")]; - tensor var_91 = const()[name = tensor("op_91"), val = tensor([1, -1, 12, 64])]; - tensor var_92 = reshape(shape = var_91, x = tensor_1)[name = tensor("op_92")]; - tensor var_93_perm_0 = const()[name = tensor("op_93_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor tensor_3 = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_v_proj_bias, weight = text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight, x = hidden_states_1)[name = tensor("tensor_3")]; - tensor var_98 = const()[name = tensor("op_98"), val = tensor([1, -1, 12, 64])]; - tensor var_99 = reshape(shape = var_98, x = tensor_3)[name = tensor("op_99")]; - tensor var_100_perm_0 = const()[name = tensor("op_100_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_107 = const()[name = tensor("op_107"), val = tensor([1, 77, 12, 64])]; - tensor var_108 = reshape(shape = var_107, x = tensor_5)[name = tensor("op_108")]; - tensor var_109_perm_0 = const()[name = tensor("op_109_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_111 = const()[name = tensor("op_111"), val = tensor([12, -1, 64])]; - tensor transpose_57 = transpose(perm = var_109_perm_0, x = var_108)[name = tensor("transpose_57")]; - tensor query_states_1 = reshape(shape = var_111, x = transpose_57)[name = tensor("query_states_1")]; + tensor text_encoder_text_model_encoder_layers_0_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76009152)))]; + tensor text_encoder_text_model_encoder_layers_0_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76010752)))]; + tensor var_13_to_fp16 = const()[name = tensor("op_13_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_1_cast = layer_norm(axes = hidden_states_1_axes_0, beta = text_encoder_text_model_encoder_layers_0_layer_norm1_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_0_layer_norm1_weight_to_fp16, x = input_3_cast)[name = tensor("hidden_states_1_cast")]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76012352)))]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77192064)))]; + tensor var_87_cast = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16, x = hidden_states_1_cast)[name = tensor("op_87_cast")]; + tensor var_88_to_fp16 = const()[name = tensor("op_88_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_5_cast = mul(x = var_87_cast, y = var_88_to_fp16)[name = tensor("tensor_5_cast")]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77193664)))]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(78373376)))]; + tensor tensor_1_cast = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16, x = hidden_states_1_cast)[name = tensor("tensor_1_cast")]; + tensor var_93 = const()[name = tensor("op_93"), val = tensor([1, -1, 12, 64])]; + tensor var_94_cast = reshape(shape = var_93, x = tensor_1_cast)[name = tensor("op_94_cast")]; + tensor var_95_perm_0 = const()[name = tensor("op_95_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(78374976)))]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79554688)))]; + tensor tensor_3_cast = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16, x = hidden_states_1_cast)[name = tensor("tensor_3_cast")]; + tensor var_100 = const()[name = tensor("op_100"), val = tensor([1, -1, 12, 64])]; + tensor var_101_cast = reshape(shape = var_100, x = tensor_3_cast)[name = tensor("op_101_cast")]; + tensor var_102_perm_0 = const()[name = tensor("op_102_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_109 = const()[name = tensor("op_109"), val = tensor([1, 77, 12, 64])]; + tensor var_110_cast = reshape(shape = var_109, x = tensor_5_cast)[name = tensor("op_110_cast")]; + tensor var_111_perm_0 = const()[name = tensor("op_111_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_113 = const()[name = tensor("op_113"), val = tensor([12, -1, 64])]; - tensor transpose_59 = transpose(perm = var_93_perm_0, x = var_92)[name = tensor("transpose_59")]; - tensor key_states_3 = reshape(shape = var_113, x = transpose_59)[name = tensor("key_states_3")]; + tensor transpose_57 = transpose(perm = var_111_perm_0, x = var_110_cast)[name = tensor("transpose_57")]; + tensor query_states_1_cast = reshape(shape = var_113, x = transpose_57)[name = tensor("query_states_1_cast")]; tensor var_115 = const()[name = tensor("op_115"), val = tensor([12, -1, 64])]; - tensor transpose_58 = transpose(perm = var_100_perm_0, x = var_99)[name = tensor("transpose_58")]; - tensor value_states_3 = reshape(shape = var_115, x = transpose_58)[name = tensor("value_states_3")]; - tensor var_118_perm_0 = const()[name = tensor("op_118_perm_0"), val = tensor([0, 2, 1])]; + tensor transpose_59 = transpose(perm = var_95_perm_0, x = var_94_cast)[name = tensor("transpose_59")]; + tensor key_states_3_cast = reshape(shape = var_115, x = transpose_59)[name = tensor("key_states_3_cast")]; + tensor var_117 = const()[name = tensor("op_117"), val = tensor([12, -1, 64])]; + tensor transpose_58 = transpose(perm = var_102_perm_0, x = var_101_cast)[name = tensor("transpose_58")]; + tensor value_states_3_cast = reshape(shape = var_117, x = transpose_58)[name = tensor("value_states_3_cast")]; + tensor var_120_perm_0 = const()[name = tensor("op_120_perm_0"), val = tensor([0, 2, 1])]; tensor attn_weights_1_transpose_x_0 = const()[name = tensor("attn_weights_1_transpose_x_0"), val = tensor(false)]; tensor attn_weights_1_transpose_y_0 = const()[name = tensor("attn_weights_1_transpose_y_0"), val = tensor(false)]; - tensor transpose_56 = transpose(perm = var_118_perm_0, x = key_states_3)[name = tensor("transpose_56")]; - tensor attn_weights_1 = matmul(transpose_x = attn_weights_1_transpose_x_0, transpose_y = attn_weights_1_transpose_y_0, x = query_states_1, y = transpose_56)[name = tensor("attn_weights_1")]; - tensor var_120 = const()[name = tensor("op_120"), val = tensor([1, 12, 77, 77])]; - tensor var_121 = reshape(shape = var_120, x = attn_weights_1)[name = tensor("op_121")]; - tensor attn_weights_3 = add(x = var_121, y = causal_attention_mask)[name = tensor("attn_weights_3")]; - tensor var_126 = const()[name = tensor("op_126"), val = tensor([12, 77, 77])]; - tensor input_5 = reshape(shape = var_126, x = attn_weights_3)[name = tensor("input_5")]; - tensor input_7 = softmax(axis = var_5, x = input_5)[name = tensor("input_7")]; + tensor transpose_56 = transpose(perm = var_120_perm_0, x = key_states_3_cast)[name = tensor("transpose_56")]; + tensor attn_weights_1_cast = matmul(transpose_x = attn_weights_1_transpose_x_0, transpose_y = attn_weights_1_transpose_y_0, x = query_states_1_cast, y = transpose_56)[name = tensor("attn_weights_1_cast")]; + tensor var_122 = const()[name = tensor("op_122"), val = tensor([1, 12, 77, 77])]; + tensor var_123_cast = reshape(shape = var_122, x = attn_weights_1_cast)[name = tensor("op_123_cast")]; + tensor causal_attention_mask_to_fp16 = const()[name = tensor("causal_attention_mask_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79556288)))]; + tensor attn_weights_3_cast = add(x = var_123_cast, y = causal_attention_mask_to_fp16)[name = tensor("attn_weights_3_cast")]; + tensor var_128 = const()[name = tensor("op_128"), val = tensor([12, 77, 77])]; + tensor input_5_cast = reshape(shape = var_128, x = attn_weights_3_cast)[name = tensor("input_5_cast")]; + tensor input_7_cast = softmax(axis = var_5, x = input_5_cast)[name = tensor("input_7_cast")]; tensor attn_output_1_transpose_x_0 = const()[name = tensor("attn_output_1_transpose_x_0"), val = tensor(false)]; tensor attn_output_1_transpose_y_0 = const()[name = tensor("attn_output_1_transpose_y_0"), val = tensor(false)]; - tensor attn_output_1 = matmul(transpose_x = attn_output_1_transpose_x_0, transpose_y = attn_output_1_transpose_y_0, x = input_7, y = value_states_3)[name = tensor("attn_output_1")]; - tensor var_131 = const()[name = tensor("op_131"), val = tensor([1, 12, 77, 64])]; - tensor attn_output_3 = reshape(shape = var_131, x = attn_output_1)[name = tensor("attn_output_3")]; + tensor attn_output_1_cast = matmul(transpose_x = attn_output_1_transpose_x_0, transpose_y = attn_output_1_transpose_y_0, x = input_7_cast, y = value_states_3_cast)[name = tensor("attn_output_1_cast")]; + tensor var_133 = const()[name = tensor("op_133"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_3_cast = reshape(shape = var_133, x = attn_output_1_cast)[name = tensor("attn_output_3_cast")]; tensor attn_output_5_perm_0 = const()[name = tensor("attn_output_5_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_134 = const()[name = tensor("op_134"), val = tensor([1, 77, 768])]; - tensor transpose_55 = transpose(perm = attn_output_5_perm_0, x = attn_output_3)[name = tensor("transpose_55")]; - tensor input_9 = reshape(shape = var_134, x = transpose_55)[name = tensor("input_9")]; - tensor hidden_states_3 = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias, weight = text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight, x = input_9)[name = tensor("hidden_states_3")]; - tensor input_11 = add(x = input_3, y = hidden_states_3)[name = tensor("input_11")]; + tensor var_136 = const()[name = tensor("op_136"), val = tensor([1, 77, 768])]; + tensor transpose_55 = transpose(perm = attn_output_5_perm_0, x = attn_output_3_cast)[name = tensor("transpose_55")]; + tensor input_9_cast = reshape(shape = var_136, x = transpose_55)[name = tensor("input_9_cast")]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79568256)))]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80747968)))]; + tensor hidden_states_3_cast = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16, x = input_9_cast)[name = tensor("hidden_states_3_cast")]; + tensor input_11_cast = add(x = input_3_cast, y = hidden_states_3_cast)[name = tensor("input_11_cast")]; tensor input_13_axes_0 = const()[name = tensor("input_13_axes_0"), val = tensor([-1])]; - tensor input_13 = layer_norm(axes = input_13_axes_0, beta = text_encoder_text_model_encoder_layers_0_layer_norm2_bias, epsilon = var_12, gamma = text_encoder_text_model_encoder_layers_0_layer_norm2_weight, x = input_11)[name = tensor("input_13")]; - tensor input_15 = linear(bias = text_encoder_text_model_encoder_layers_0_mlp_fc1_bias, weight = text_encoder_text_model_encoder_layers_0_mlp_fc1_weight, x = input_13)[name = tensor("input_15")]; - tensor var_149 = const()[name = tensor("op_149"), val = tensor(0x1.b3b646p+0)]; - tensor var_150 = mul(x = input_15, y = var_149)[name = tensor("op_150")]; - tensor var_151 = sigmoid(x = var_150)[name = tensor("op_151")]; - tensor input_17 = mul(x = input_15, y = var_151)[name = tensor("input_17")]; - tensor hidden_states_5 = linear(bias = text_encoder_text_model_encoder_layers_0_mlp_fc2_bias, weight = text_encoder_text_model_encoder_layers_0_mlp_fc2_weight, x = input_17)[name = tensor("hidden_states_5")]; - tensor input_19 = add(x = input_11, y = hidden_states_5)[name = tensor("input_19")]; + tensor text_encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80749568)))]; + tensor text_encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80751168)))]; + tensor input_13_cast = layer_norm(axes = input_13_axes_0, beta = text_encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16, x = input_11_cast)[name = tensor("input_13_cast")]; + tensor text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80752768)))]; + tensor text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85471424)))]; + tensor input_15_cast = linear(bias = text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16, x = input_13_cast)[name = tensor("input_15_cast")]; + tensor var_151_to_fp16 = const()[name = tensor("op_151_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_152_cast = mul(x = input_15_cast, y = var_151_to_fp16)[name = tensor("op_152_cast")]; + tensor var_153_cast = sigmoid(x = var_152_cast)[name = tensor("op_153_cast")]; + tensor input_17_cast = mul(x = input_15_cast, y = var_153_cast)[name = tensor("input_17_cast")]; + tensor text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85477632)))]; + tensor text_encoder_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90196288)))]; + tensor hidden_states_5_cast = linear(bias = text_encoder_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16, x = input_17_cast)[name = tensor("hidden_states_5_cast")]; + tensor input_19_cast = add(x = input_11_cast, y = hidden_states_5_cast)[name = tensor("input_19_cast")]; tensor hidden_states_7_axes_0 = const()[name = tensor("hidden_states_7_axes_0"), val = tensor([-1])]; - tensor hidden_states_7 = layer_norm(axes = hidden_states_7_axes_0, beta = text_encoder_text_model_encoder_layers_1_layer_norm1_bias, epsilon = var_12, gamma = text_encoder_text_model_encoder_layers_1_layer_norm1_weight, x = input_19)[name = tensor("hidden_states_7")]; - tensor var_175 = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_q_proj_bias, weight = text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight, x = hidden_states_7)[name = tensor("op_175")]; - tensor var_176 = const()[name = tensor("op_176"), val = tensor(0x1p-3)]; - tensor tensor_11 = mul(x = var_175, y = var_176)[name = tensor("tensor_11")]; - tensor tensor_7 = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_k_proj_bias, weight = text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight, x = hidden_states_7)[name = tensor("tensor_7")]; - tensor var_181 = const()[name = tensor("op_181"), val = tensor([1, -1, 12, 64])]; - tensor var_182 = reshape(shape = var_181, x = tensor_7)[name = tensor("op_182")]; - tensor var_183_perm_0 = const()[name = tensor("op_183_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor tensor_9 = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_v_proj_bias, weight = text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight, x = hidden_states_7)[name = tensor("tensor_9")]; - tensor var_188 = const()[name = tensor("op_188"), val = tensor([1, -1, 12, 64])]; - tensor var_189 = reshape(shape = var_188, x = tensor_9)[name = tensor("op_189")]; - tensor var_190_perm_0 = const()[name = tensor("op_190_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_197 = const()[name = tensor("op_197"), val = tensor([1, 77, 12, 64])]; - tensor var_198 = reshape(shape = var_197, x = tensor_11)[name = tensor("op_198")]; - tensor var_199_perm_0 = const()[name = tensor("op_199_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_201 = const()[name = tensor("op_201"), val = tensor([12, -1, 64])]; - tensor transpose_52 = transpose(perm = var_199_perm_0, x = var_198)[name = tensor("transpose_52")]; - tensor query_states_3 = reshape(shape = var_201, x = transpose_52)[name = tensor("query_states_3")]; + tensor text_encoder_text_model_encoder_layers_1_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90197888)))]; + tensor text_encoder_text_model_encoder_layers_1_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90199488)))]; + tensor hidden_states_7_cast = layer_norm(axes = hidden_states_7_axes_0, beta = text_encoder_text_model_encoder_layers_1_layer_norm1_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_1_layer_norm1_weight_to_fp16, x = input_19_cast)[name = tensor("hidden_states_7_cast")]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90201088)))]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91380800)))]; + tensor var_177_cast = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16, x = hidden_states_7_cast)[name = tensor("op_177_cast")]; + tensor var_178_to_fp16 = const()[name = tensor("op_178_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_11_cast = mul(x = var_177_cast, y = var_178_to_fp16)[name = tensor("tensor_11_cast")]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91382400)))]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92562112)))]; + tensor tensor_7_cast = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16, x = hidden_states_7_cast)[name = tensor("tensor_7_cast")]; + tensor var_183 = const()[name = tensor("op_183"), val = tensor([1, -1, 12, 64])]; + tensor var_184_cast = reshape(shape = var_183, x = tensor_7_cast)[name = tensor("op_184_cast")]; + tensor var_185_perm_0 = const()[name = tensor("op_185_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92563712)))]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93743424)))]; + tensor tensor_9_cast = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16, x = hidden_states_7_cast)[name = tensor("tensor_9_cast")]; + tensor var_190 = const()[name = tensor("op_190"), val = tensor([1, -1, 12, 64])]; + tensor var_191_cast = reshape(shape = var_190, x = tensor_9_cast)[name = tensor("op_191_cast")]; + tensor var_192_perm_0 = const()[name = tensor("op_192_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_199 = const()[name = tensor("op_199"), val = tensor([1, 77, 12, 64])]; + tensor var_200_cast = reshape(shape = var_199, x = tensor_11_cast)[name = tensor("op_200_cast")]; + tensor var_201_perm_0 = const()[name = tensor("op_201_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_203 = const()[name = tensor("op_203"), val = tensor([12, -1, 64])]; - tensor transpose_54 = transpose(perm = var_183_perm_0, x = var_182)[name = tensor("transpose_54")]; - tensor key_states_7 = reshape(shape = var_203, x = transpose_54)[name = tensor("key_states_7")]; + tensor transpose_52 = transpose(perm = var_201_perm_0, x = var_200_cast)[name = tensor("transpose_52")]; + tensor query_states_3_cast = reshape(shape = var_203, x = transpose_52)[name = tensor("query_states_3_cast")]; tensor var_205 = const()[name = tensor("op_205"), val = tensor([12, -1, 64])]; - tensor transpose_53 = transpose(perm = var_190_perm_0, x = var_189)[name = tensor("transpose_53")]; - tensor value_states_7 = reshape(shape = var_205, x = transpose_53)[name = tensor("value_states_7")]; - tensor var_208_perm_0 = const()[name = tensor("op_208_perm_0"), val = tensor([0, 2, 1])]; + tensor transpose_54 = transpose(perm = var_185_perm_0, x = var_184_cast)[name = tensor("transpose_54")]; + tensor key_states_7_cast = reshape(shape = var_205, x = transpose_54)[name = tensor("key_states_7_cast")]; + tensor var_207 = const()[name = tensor("op_207"), val = tensor([12, -1, 64])]; + tensor transpose_53 = transpose(perm = var_192_perm_0, x = var_191_cast)[name = tensor("transpose_53")]; + tensor value_states_7_cast = reshape(shape = var_207, x = transpose_53)[name = tensor("value_states_7_cast")]; + tensor var_210_perm_0 = const()[name = tensor("op_210_perm_0"), val = tensor([0, 2, 1])]; tensor attn_weights_7_transpose_x_0 = const()[name = tensor("attn_weights_7_transpose_x_0"), val = tensor(false)]; tensor attn_weights_7_transpose_y_0 = const()[name = tensor("attn_weights_7_transpose_y_0"), val = tensor(false)]; - tensor transpose_51 = transpose(perm = var_208_perm_0, x = key_states_7)[name = tensor("transpose_51")]; - tensor attn_weights_7 = matmul(transpose_x = attn_weights_7_transpose_x_0, transpose_y = attn_weights_7_transpose_y_0, x = query_states_3, y = transpose_51)[name = tensor("attn_weights_7")]; - tensor var_210 = const()[name = tensor("op_210"), val = tensor([1, 12, 77, 77])]; - tensor var_211 = reshape(shape = var_210, x = attn_weights_7)[name = tensor("op_211")]; - tensor attn_weights_9 = add(x = var_211, y = causal_attention_mask)[name = tensor("attn_weights_9")]; - tensor var_216 = const()[name = tensor("op_216"), val = tensor([12, 77, 77])]; - tensor input_21 = reshape(shape = var_216, x = attn_weights_9)[name = tensor("input_21")]; - tensor input_23 = softmax(axis = var_5, x = input_21)[name = tensor("input_23")]; + tensor transpose_51 = transpose(perm = var_210_perm_0, x = key_states_7_cast)[name = tensor("transpose_51")]; + tensor attn_weights_7_cast = matmul(transpose_x = attn_weights_7_transpose_x_0, transpose_y = attn_weights_7_transpose_y_0, x = query_states_3_cast, y = transpose_51)[name = tensor("attn_weights_7_cast")]; + tensor var_212 = const()[name = tensor("op_212"), val = tensor([1, 12, 77, 77])]; + tensor var_213_cast = reshape(shape = var_212, x = attn_weights_7_cast)[name = tensor("op_213_cast")]; + tensor attn_weights_9_cast = add(x = var_213_cast, y = causal_attention_mask_to_fp16)[name = tensor("attn_weights_9_cast")]; + tensor var_218 = const()[name = tensor("op_218"), val = tensor([12, 77, 77])]; + tensor input_21_cast = reshape(shape = var_218, x = attn_weights_9_cast)[name = tensor("input_21_cast")]; + tensor input_23_cast = softmax(axis = var_5, x = input_21_cast)[name = tensor("input_23_cast")]; tensor attn_output_7_transpose_x_0 = const()[name = tensor("attn_output_7_transpose_x_0"), val = tensor(false)]; tensor attn_output_7_transpose_y_0 = const()[name = tensor("attn_output_7_transpose_y_0"), val = tensor(false)]; - tensor attn_output_7 = matmul(transpose_x = attn_output_7_transpose_x_0, transpose_y = attn_output_7_transpose_y_0, x = input_23, y = value_states_7)[name = tensor("attn_output_7")]; - tensor var_221 = const()[name = tensor("op_221"), val = tensor([1, 12, 77, 64])]; - tensor attn_output_9 = reshape(shape = var_221, x = attn_output_7)[name = tensor("attn_output_9")]; + tensor attn_output_7_cast = matmul(transpose_x = attn_output_7_transpose_x_0, transpose_y = attn_output_7_transpose_y_0, x = input_23_cast, y = value_states_7_cast)[name = tensor("attn_output_7_cast")]; + tensor var_223 = const()[name = tensor("op_223"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_9_cast = reshape(shape = var_223, x = attn_output_7_cast)[name = tensor("attn_output_9_cast")]; tensor attn_output_11_perm_0 = const()[name = tensor("attn_output_11_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_224 = const()[name = tensor("op_224"), val = tensor([1, 77, 768])]; - tensor transpose_50 = transpose(perm = attn_output_11_perm_0, x = attn_output_9)[name = tensor("transpose_50")]; - tensor input_25 = reshape(shape = var_224, x = transpose_50)[name = tensor("input_25")]; - tensor hidden_states_9 = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_out_proj_bias, weight = text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight, x = input_25)[name = tensor("hidden_states_9")]; - tensor input_27 = add(x = input_19, y = hidden_states_9)[name = tensor("input_27")]; + tensor var_226 = const()[name = tensor("op_226"), val = tensor([1, 77, 768])]; + tensor transpose_50 = transpose(perm = attn_output_11_perm_0, x = attn_output_9_cast)[name = tensor("transpose_50")]; + tensor input_25_cast = reshape(shape = var_226, x = transpose_50)[name = tensor("input_25_cast")]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93745024)))]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(94924736)))]; + tensor hidden_states_9_cast = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16, x = input_25_cast)[name = tensor("hidden_states_9_cast")]; + tensor input_27_cast = add(x = input_19_cast, y = hidden_states_9_cast)[name = tensor("input_27_cast")]; tensor input_29_axes_0 = const()[name = tensor("input_29_axes_0"), val = tensor([-1])]; - tensor input_29 = layer_norm(axes = input_29_axes_0, beta = text_encoder_text_model_encoder_layers_1_layer_norm2_bias, epsilon = var_12, gamma = text_encoder_text_model_encoder_layers_1_layer_norm2_weight, x = input_27)[name = tensor("input_29")]; - tensor input_31 = linear(bias = text_encoder_text_model_encoder_layers_1_mlp_fc1_bias, weight = text_encoder_text_model_encoder_layers_1_mlp_fc1_weight, x = input_29)[name = tensor("input_31")]; - tensor var_239 = const()[name = tensor("op_239"), val = tensor(0x1.b3b646p+0)]; - tensor var_240 = mul(x = input_31, y = var_239)[name = tensor("op_240")]; - tensor var_241 = sigmoid(x = var_240)[name = tensor("op_241")]; - tensor input_33 = mul(x = input_31, y = var_241)[name = tensor("input_33")]; - tensor hidden_states_11 = linear(bias = text_encoder_text_model_encoder_layers_1_mlp_fc2_bias, weight = text_encoder_text_model_encoder_layers_1_mlp_fc2_weight, x = input_33)[name = tensor("hidden_states_11")]; - tensor input_35 = add(x = input_27, y = hidden_states_11)[name = tensor("input_35")]; + tensor text_encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(94926336)))]; + tensor text_encoder_text_model_encoder_layers_1_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(94927936)))]; + tensor input_29_cast = layer_norm(axes = input_29_axes_0, beta = text_encoder_text_model_encoder_layers_1_layer_norm2_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16, x = input_27_cast)[name = tensor("input_29_cast")]; + tensor text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(94929536)))]; + tensor text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99648192)))]; + tensor input_31_cast = linear(bias = text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16, x = input_29_cast)[name = tensor("input_31_cast")]; + tensor var_241_to_fp16 = const()[name = tensor("op_241_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_242_cast = mul(x = input_31_cast, y = var_241_to_fp16)[name = tensor("op_242_cast")]; + tensor var_243_cast = sigmoid(x = var_242_cast)[name = tensor("op_243_cast")]; + tensor input_33_cast = mul(x = input_31_cast, y = var_243_cast)[name = tensor("input_33_cast")]; + tensor text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99654400)))]; + tensor text_encoder_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104373056)))]; + tensor hidden_states_11_cast = linear(bias = text_encoder_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16, x = input_33_cast)[name = tensor("hidden_states_11_cast")]; + tensor input_35_cast = add(x = input_27_cast, y = hidden_states_11_cast)[name = tensor("input_35_cast")]; tensor hidden_states_13_axes_0 = const()[name = tensor("hidden_states_13_axes_0"), val = tensor([-1])]; - tensor hidden_states_13 = layer_norm(axes = hidden_states_13_axes_0, beta = text_encoder_text_model_encoder_layers_2_layer_norm1_bias, epsilon = var_12, gamma = text_encoder_text_model_encoder_layers_2_layer_norm1_weight, x = input_35)[name = tensor("hidden_states_13")]; - tensor var_265 = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_q_proj_bias, weight = text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight, x = hidden_states_13)[name = tensor("op_265")]; - tensor var_266 = const()[name = tensor("op_266"), val = tensor(0x1p-3)]; - tensor tensor_17 = mul(x = var_265, y = var_266)[name = tensor("tensor_17")]; - tensor tensor_13 = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_k_proj_bias, weight = text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight, x = hidden_states_13)[name = tensor("tensor_13")]; - tensor var_271 = const()[name = tensor("op_271"), val = tensor([1, -1, 12, 64])]; - tensor var_272 = reshape(shape = var_271, x = tensor_13)[name = tensor("op_272")]; - tensor var_273_perm_0 = const()[name = tensor("op_273_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor tensor_15 = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_v_proj_bias, weight = text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight, x = hidden_states_13)[name = tensor("tensor_15")]; - tensor var_278 = const()[name = tensor("op_278"), val = tensor([1, -1, 12, 64])]; - tensor var_279 = reshape(shape = var_278, x = tensor_15)[name = tensor("op_279")]; - tensor var_280_perm_0 = const()[name = tensor("op_280_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_287 = const()[name = tensor("op_287"), val = tensor([1, 77, 12, 64])]; - tensor var_288 = reshape(shape = var_287, x = tensor_17)[name = tensor("op_288")]; - tensor var_289_perm_0 = const()[name = tensor("op_289_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_291 = const()[name = tensor("op_291"), val = tensor([12, -1, 64])]; - tensor transpose_47 = transpose(perm = var_289_perm_0, x = var_288)[name = tensor("transpose_47")]; - tensor query_states_5 = reshape(shape = var_291, x = transpose_47)[name = tensor("query_states_5")]; + tensor text_encoder_text_model_encoder_layers_2_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104374656)))]; + tensor text_encoder_text_model_encoder_layers_2_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104376256)))]; + tensor hidden_states_13_cast = layer_norm(axes = hidden_states_13_axes_0, beta = text_encoder_text_model_encoder_layers_2_layer_norm1_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_2_layer_norm1_weight_to_fp16, x = input_35_cast)[name = tensor("hidden_states_13_cast")]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104377856)))]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(105557568)))]; + tensor var_267_cast = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16, x = hidden_states_13_cast)[name = tensor("op_267_cast")]; + tensor var_268_to_fp16 = const()[name = tensor("op_268_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_17_cast = mul(x = var_267_cast, y = var_268_to_fp16)[name = tensor("tensor_17_cast")]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(105559168)))]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106738880)))]; + tensor tensor_13_cast = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16, x = hidden_states_13_cast)[name = tensor("tensor_13_cast")]; + tensor var_273 = const()[name = tensor("op_273"), val = tensor([1, -1, 12, 64])]; + tensor var_274_cast = reshape(shape = var_273, x = tensor_13_cast)[name = tensor("op_274_cast")]; + tensor var_275_perm_0 = const()[name = tensor("op_275_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106740480)))]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107920192)))]; + tensor tensor_15_cast = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16, x = hidden_states_13_cast)[name = tensor("tensor_15_cast")]; + tensor var_280 = const()[name = tensor("op_280"), val = tensor([1, -1, 12, 64])]; + tensor var_281_cast = reshape(shape = var_280, x = tensor_15_cast)[name = tensor("op_281_cast")]; + tensor var_282_perm_0 = const()[name = tensor("op_282_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_289 = const()[name = tensor("op_289"), val = tensor([1, 77, 12, 64])]; + tensor var_290_cast = reshape(shape = var_289, x = tensor_17_cast)[name = tensor("op_290_cast")]; + tensor var_291_perm_0 = const()[name = tensor("op_291_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_293 = const()[name = tensor("op_293"), val = tensor([12, -1, 64])]; - tensor transpose_49 = transpose(perm = var_273_perm_0, x = var_272)[name = tensor("transpose_49")]; - tensor key_states_11 = reshape(shape = var_293, x = transpose_49)[name = tensor("key_states_11")]; + tensor transpose_47 = transpose(perm = var_291_perm_0, x = var_290_cast)[name = tensor("transpose_47")]; + tensor query_states_5_cast = reshape(shape = var_293, x = transpose_47)[name = tensor("query_states_5_cast")]; tensor var_295 = const()[name = tensor("op_295"), val = tensor([12, -1, 64])]; - tensor transpose_48 = transpose(perm = var_280_perm_0, x = var_279)[name = tensor("transpose_48")]; - tensor value_states_11 = reshape(shape = var_295, x = transpose_48)[name = tensor("value_states_11")]; - tensor var_298_perm_0 = const()[name = tensor("op_298_perm_0"), val = tensor([0, 2, 1])]; + tensor transpose_49 = transpose(perm = var_275_perm_0, x = var_274_cast)[name = tensor("transpose_49")]; + tensor key_states_11_cast = reshape(shape = var_295, x = transpose_49)[name = tensor("key_states_11_cast")]; + tensor var_297 = const()[name = tensor("op_297"), val = tensor([12, -1, 64])]; + tensor transpose_48 = transpose(perm = var_282_perm_0, x = var_281_cast)[name = tensor("transpose_48")]; + tensor value_states_11_cast = reshape(shape = var_297, x = transpose_48)[name = tensor("value_states_11_cast")]; + tensor var_300_perm_0 = const()[name = tensor("op_300_perm_0"), val = tensor([0, 2, 1])]; tensor attn_weights_13_transpose_x_0 = const()[name = tensor("attn_weights_13_transpose_x_0"), val = tensor(false)]; tensor attn_weights_13_transpose_y_0 = const()[name = tensor("attn_weights_13_transpose_y_0"), val = tensor(false)]; - tensor transpose_46 = transpose(perm = var_298_perm_0, x = key_states_11)[name = tensor("transpose_46")]; - tensor attn_weights_13 = matmul(transpose_x = attn_weights_13_transpose_x_0, transpose_y = attn_weights_13_transpose_y_0, x = query_states_5, y = transpose_46)[name = tensor("attn_weights_13")]; - tensor var_300 = const()[name = tensor("op_300"), val = tensor([1, 12, 77, 77])]; - tensor var_301 = reshape(shape = var_300, x = attn_weights_13)[name = tensor("op_301")]; - tensor attn_weights_15 = add(x = var_301, y = causal_attention_mask)[name = tensor("attn_weights_15")]; - tensor var_306 = const()[name = tensor("op_306"), val = tensor([12, 77, 77])]; - tensor input_37 = reshape(shape = var_306, x = attn_weights_15)[name = tensor("input_37")]; - tensor input_39 = softmax(axis = var_5, x = input_37)[name = tensor("input_39")]; + tensor transpose_46 = transpose(perm = var_300_perm_0, x = key_states_11_cast)[name = tensor("transpose_46")]; + tensor attn_weights_13_cast = matmul(transpose_x = attn_weights_13_transpose_x_0, transpose_y = attn_weights_13_transpose_y_0, x = query_states_5_cast, y = transpose_46)[name = tensor("attn_weights_13_cast")]; + tensor var_302 = const()[name = tensor("op_302"), val = tensor([1, 12, 77, 77])]; + tensor var_303_cast = reshape(shape = var_302, x = attn_weights_13_cast)[name = tensor("op_303_cast")]; + tensor attn_weights_15_cast = add(x = var_303_cast, y = causal_attention_mask_to_fp16)[name = tensor("attn_weights_15_cast")]; + tensor var_308 = const()[name = tensor("op_308"), val = tensor([12, 77, 77])]; + tensor input_37_cast = reshape(shape = var_308, x = attn_weights_15_cast)[name = tensor("input_37_cast")]; + tensor input_39_cast = softmax(axis = var_5, x = input_37_cast)[name = tensor("input_39_cast")]; tensor attn_output_13_transpose_x_0 = const()[name = tensor("attn_output_13_transpose_x_0"), val = tensor(false)]; tensor attn_output_13_transpose_y_0 = const()[name = tensor("attn_output_13_transpose_y_0"), val = tensor(false)]; - tensor attn_output_13 = matmul(transpose_x = attn_output_13_transpose_x_0, transpose_y = attn_output_13_transpose_y_0, x = input_39, y = value_states_11)[name = tensor("attn_output_13")]; - tensor var_311 = const()[name = tensor("op_311"), val = tensor([1, 12, 77, 64])]; - tensor attn_output_15 = reshape(shape = var_311, x = attn_output_13)[name = tensor("attn_output_15")]; + tensor attn_output_13_cast = matmul(transpose_x = attn_output_13_transpose_x_0, transpose_y = attn_output_13_transpose_y_0, x = input_39_cast, y = value_states_11_cast)[name = tensor("attn_output_13_cast")]; + tensor var_313 = const()[name = tensor("op_313"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_15_cast = reshape(shape = var_313, x = attn_output_13_cast)[name = tensor("attn_output_15_cast")]; tensor attn_output_17_perm_0 = const()[name = tensor("attn_output_17_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_314 = const()[name = tensor("op_314"), val = tensor([1, 77, 768])]; - tensor transpose_45 = transpose(perm = attn_output_17_perm_0, x = attn_output_15)[name = tensor("transpose_45")]; - tensor input_41 = reshape(shape = var_314, x = transpose_45)[name = tensor("input_41")]; - tensor hidden_states_15 = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_out_proj_bias, weight = text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight, x = input_41)[name = tensor("hidden_states_15")]; - tensor input_43 = add(x = input_35, y = hidden_states_15)[name = tensor("input_43")]; + tensor var_316 = const()[name = tensor("op_316"), val = tensor([1, 77, 768])]; + tensor transpose_45 = transpose(perm = attn_output_17_perm_0, x = attn_output_15_cast)[name = tensor("transpose_45")]; + tensor input_41_cast = reshape(shape = var_316, x = transpose_45)[name = tensor("input_41_cast")]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107921792)))]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109101504)))]; + tensor hidden_states_15_cast = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16, x = input_41_cast)[name = tensor("hidden_states_15_cast")]; + tensor input_43_cast = add(x = input_35_cast, y = hidden_states_15_cast)[name = tensor("input_43_cast")]; tensor input_45_axes_0 = const()[name = tensor("input_45_axes_0"), val = tensor([-1])]; - tensor input_45 = layer_norm(axes = input_45_axes_0, beta = text_encoder_text_model_encoder_layers_2_layer_norm2_bias, epsilon = var_12, gamma = text_encoder_text_model_encoder_layers_2_layer_norm2_weight, x = input_43)[name = tensor("input_45")]; - tensor input_47 = linear(bias = text_encoder_text_model_encoder_layers_2_mlp_fc1_bias, weight = text_encoder_text_model_encoder_layers_2_mlp_fc1_weight, x = input_45)[name = tensor("input_47")]; - tensor var_329 = const()[name = tensor("op_329"), val = tensor(0x1.b3b646p+0)]; - tensor var_330 = mul(x = input_47, y = var_329)[name = tensor("op_330")]; - tensor var_331 = sigmoid(x = var_330)[name = tensor("op_331")]; - tensor input_49 = mul(x = input_47, y = var_331)[name = tensor("input_49")]; - tensor hidden_states_17 = linear(bias = text_encoder_text_model_encoder_layers_2_mlp_fc2_bias, weight = text_encoder_text_model_encoder_layers_2_mlp_fc2_weight, x = input_49)[name = tensor("hidden_states_17")]; - tensor input_51 = add(x = input_43, y = hidden_states_17)[name = tensor("input_51")]; + tensor text_encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109103104)))]; + tensor text_encoder_text_model_encoder_layers_2_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109104704)))]; + tensor input_45_cast = layer_norm(axes = input_45_axes_0, beta = text_encoder_text_model_encoder_layers_2_layer_norm2_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16, x = input_43_cast)[name = tensor("input_45_cast")]; + tensor text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109106304)))]; + tensor text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113824960)))]; + tensor input_47_cast = linear(bias = text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16, x = input_45_cast)[name = tensor("input_47_cast")]; + tensor var_331_to_fp16 = const()[name = tensor("op_331_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_332_cast = mul(x = input_47_cast, y = var_331_to_fp16)[name = tensor("op_332_cast")]; + tensor var_333_cast = sigmoid(x = var_332_cast)[name = tensor("op_333_cast")]; + tensor input_49_cast = mul(x = input_47_cast, y = var_333_cast)[name = tensor("input_49_cast")]; + tensor text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113831168)))]; + tensor text_encoder_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(118549824)))]; + tensor hidden_states_17_cast = linear(bias = text_encoder_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16, x = input_49_cast)[name = tensor("hidden_states_17_cast")]; + tensor input_51_cast = add(x = input_43_cast, y = hidden_states_17_cast)[name = tensor("input_51_cast")]; tensor hidden_states_19_axes_0 = const()[name = tensor("hidden_states_19_axes_0"), val = tensor([-1])]; - tensor hidden_states_19 = layer_norm(axes = hidden_states_19_axes_0, beta = text_encoder_text_model_encoder_layers_3_layer_norm1_bias, epsilon = var_12, gamma = text_encoder_text_model_encoder_layers_3_layer_norm1_weight, x = input_51)[name = tensor("hidden_states_19")]; - tensor var_355 = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_q_proj_bias, weight = text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight, x = hidden_states_19)[name = tensor("op_355")]; - tensor var_356 = const()[name = tensor("op_356"), val = tensor(0x1p-3)]; - tensor tensor_23 = mul(x = var_355, y = var_356)[name = tensor("tensor_23")]; - tensor tensor_19 = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_k_proj_bias, weight = text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight, x = hidden_states_19)[name = tensor("tensor_19")]; - tensor var_361 = const()[name = tensor("op_361"), val = tensor([1, -1, 12, 64])]; - tensor var_362 = reshape(shape = var_361, x = tensor_19)[name = tensor("op_362")]; - tensor var_363_perm_0 = const()[name = tensor("op_363_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor tensor_21 = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_v_proj_bias, weight = text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight, x = hidden_states_19)[name = tensor("tensor_21")]; - tensor var_368 = const()[name = tensor("op_368"), val = tensor([1, -1, 12, 64])]; - tensor var_369 = reshape(shape = var_368, x = tensor_21)[name = tensor("op_369")]; - tensor var_370_perm_0 = const()[name = tensor("op_370_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_377 = const()[name = tensor("op_377"), val = tensor([1, 77, 12, 64])]; - tensor var_378 = reshape(shape = var_377, x = tensor_23)[name = tensor("op_378")]; - tensor var_379_perm_0 = const()[name = tensor("op_379_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_381 = const()[name = tensor("op_381"), val = tensor([12, -1, 64])]; - tensor transpose_42 = transpose(perm = var_379_perm_0, x = var_378)[name = tensor("transpose_42")]; - tensor query_states_7 = reshape(shape = var_381, x = transpose_42)[name = tensor("query_states_7")]; + tensor text_encoder_text_model_encoder_layers_3_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(118551424)))]; + tensor text_encoder_text_model_encoder_layers_3_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(118553024)))]; + tensor hidden_states_19_cast = layer_norm(axes = hidden_states_19_axes_0, beta = text_encoder_text_model_encoder_layers_3_layer_norm1_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_3_layer_norm1_weight_to_fp16, x = input_51_cast)[name = tensor("hidden_states_19_cast")]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(118554624)))]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119734336)))]; + tensor var_357_cast = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16, x = hidden_states_19_cast)[name = tensor("op_357_cast")]; + tensor var_358_to_fp16 = const()[name = tensor("op_358_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_23_cast = mul(x = var_357_cast, y = var_358_to_fp16)[name = tensor("tensor_23_cast")]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119735936)))]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120915648)))]; + tensor tensor_19_cast = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16, x = hidden_states_19_cast)[name = tensor("tensor_19_cast")]; + tensor var_363 = const()[name = tensor("op_363"), val = tensor([1, -1, 12, 64])]; + tensor var_364_cast = reshape(shape = var_363, x = tensor_19_cast)[name = tensor("op_364_cast")]; + tensor var_365_perm_0 = const()[name = tensor("op_365_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120917248)))]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(122096960)))]; + tensor tensor_21_cast = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16, x = hidden_states_19_cast)[name = tensor("tensor_21_cast")]; + tensor var_370 = const()[name = tensor("op_370"), val = tensor([1, -1, 12, 64])]; + tensor var_371_cast = reshape(shape = var_370, x = tensor_21_cast)[name = tensor("op_371_cast")]; + tensor var_372_perm_0 = const()[name = tensor("op_372_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_379 = const()[name = tensor("op_379"), val = tensor([1, 77, 12, 64])]; + tensor var_380_cast = reshape(shape = var_379, x = tensor_23_cast)[name = tensor("op_380_cast")]; + tensor var_381_perm_0 = const()[name = tensor("op_381_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_383 = const()[name = tensor("op_383"), val = tensor([12, -1, 64])]; - tensor transpose_44 = transpose(perm = var_363_perm_0, x = var_362)[name = tensor("transpose_44")]; - tensor key_states_15 = reshape(shape = var_383, x = transpose_44)[name = tensor("key_states_15")]; + tensor transpose_42 = transpose(perm = var_381_perm_0, x = var_380_cast)[name = tensor("transpose_42")]; + tensor query_states_7_cast = reshape(shape = var_383, x = transpose_42)[name = tensor("query_states_7_cast")]; tensor var_385 = const()[name = tensor("op_385"), val = tensor([12, -1, 64])]; - tensor transpose_43 = transpose(perm = var_370_perm_0, x = var_369)[name = tensor("transpose_43")]; - tensor value_states_15 = reshape(shape = var_385, x = transpose_43)[name = tensor("value_states_15")]; - tensor var_388_perm_0 = const()[name = tensor("op_388_perm_0"), val = tensor([0, 2, 1])]; + tensor transpose_44 = transpose(perm = var_365_perm_0, x = var_364_cast)[name = tensor("transpose_44")]; + tensor key_states_15_cast = reshape(shape = var_385, x = transpose_44)[name = tensor("key_states_15_cast")]; + tensor var_387 = const()[name = tensor("op_387"), val = tensor([12, -1, 64])]; + tensor transpose_43 = transpose(perm = var_372_perm_0, x = var_371_cast)[name = tensor("transpose_43")]; + tensor value_states_15_cast = reshape(shape = var_387, x = transpose_43)[name = tensor("value_states_15_cast")]; + tensor var_390_perm_0 = const()[name = tensor("op_390_perm_0"), val = tensor([0, 2, 1])]; tensor attn_weights_19_transpose_x_0 = const()[name = tensor("attn_weights_19_transpose_x_0"), val = tensor(false)]; tensor attn_weights_19_transpose_y_0 = const()[name = tensor("attn_weights_19_transpose_y_0"), val = tensor(false)]; - tensor transpose_41 = transpose(perm = var_388_perm_0, x = key_states_15)[name = tensor("transpose_41")]; - tensor attn_weights_19 = matmul(transpose_x = attn_weights_19_transpose_x_0, transpose_y = attn_weights_19_transpose_y_0, x = query_states_7, y = transpose_41)[name = tensor("attn_weights_19")]; - tensor var_390 = const()[name = tensor("op_390"), val = tensor([1, 12, 77, 77])]; - tensor var_391 = reshape(shape = var_390, x = attn_weights_19)[name = tensor("op_391")]; - tensor attn_weights_21 = add(x = var_391, y = causal_attention_mask)[name = tensor("attn_weights_21")]; - tensor var_396 = const()[name = tensor("op_396"), val = tensor([12, 77, 77])]; - tensor input_53 = reshape(shape = var_396, x = attn_weights_21)[name = tensor("input_53")]; - tensor input_55 = softmax(axis = var_5, x = input_53)[name = tensor("input_55")]; + tensor transpose_41 = transpose(perm = var_390_perm_0, x = key_states_15_cast)[name = tensor("transpose_41")]; + tensor attn_weights_19_cast = matmul(transpose_x = attn_weights_19_transpose_x_0, transpose_y = attn_weights_19_transpose_y_0, x = query_states_7_cast, y = transpose_41)[name = tensor("attn_weights_19_cast")]; + tensor var_392 = const()[name = tensor("op_392"), val = tensor([1, 12, 77, 77])]; + tensor var_393_cast = reshape(shape = var_392, x = attn_weights_19_cast)[name = tensor("op_393_cast")]; + tensor attn_weights_21_cast = add(x = var_393_cast, y = causal_attention_mask_to_fp16)[name = tensor("attn_weights_21_cast")]; + tensor var_398 = const()[name = tensor("op_398"), val = tensor([12, 77, 77])]; + tensor input_53_cast = reshape(shape = var_398, x = attn_weights_21_cast)[name = tensor("input_53_cast")]; + tensor input_55_cast = softmax(axis = var_5, x = input_53_cast)[name = tensor("input_55_cast")]; tensor attn_output_19_transpose_x_0 = const()[name = tensor("attn_output_19_transpose_x_0"), val = tensor(false)]; tensor attn_output_19_transpose_y_0 = const()[name = tensor("attn_output_19_transpose_y_0"), val = tensor(false)]; - tensor attn_output_19 = matmul(transpose_x = attn_output_19_transpose_x_0, transpose_y = attn_output_19_transpose_y_0, x = input_55, y = value_states_15)[name = tensor("attn_output_19")]; - tensor var_401 = const()[name = tensor("op_401"), val = tensor([1, 12, 77, 64])]; - tensor attn_output_21 = reshape(shape = var_401, x = attn_output_19)[name = tensor("attn_output_21")]; + tensor attn_output_19_cast = matmul(transpose_x = attn_output_19_transpose_x_0, transpose_y = attn_output_19_transpose_y_0, x = input_55_cast, y = value_states_15_cast)[name = tensor("attn_output_19_cast")]; + tensor var_403 = const()[name = tensor("op_403"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_21_cast = reshape(shape = var_403, x = attn_output_19_cast)[name = tensor("attn_output_21_cast")]; tensor attn_output_23_perm_0 = const()[name = tensor("attn_output_23_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_404 = const()[name = tensor("op_404"), val = tensor([1, 77, 768])]; - tensor transpose_40 = transpose(perm = attn_output_23_perm_0, x = attn_output_21)[name = tensor("transpose_40")]; - tensor input_57 = reshape(shape = var_404, x = transpose_40)[name = tensor("input_57")]; - tensor hidden_states_21 = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_out_proj_bias, weight = text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight, x = input_57)[name = tensor("hidden_states_21")]; - tensor input_59 = add(x = input_51, y = hidden_states_21)[name = tensor("input_59")]; + tensor var_406 = const()[name = tensor("op_406"), val = tensor([1, 77, 768])]; + tensor transpose_40 = transpose(perm = attn_output_23_perm_0, x = attn_output_21_cast)[name = tensor("transpose_40")]; + tensor input_57_cast = reshape(shape = var_406, x = transpose_40)[name = tensor("input_57_cast")]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(122098560)))]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123278272)))]; + tensor hidden_states_21_cast = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16, x = input_57_cast)[name = tensor("hidden_states_21_cast")]; + tensor input_59_cast = add(x = input_51_cast, y = hidden_states_21_cast)[name = tensor("input_59_cast")]; tensor input_61_axes_0 = const()[name = tensor("input_61_axes_0"), val = tensor([-1])]; - tensor input_61 = layer_norm(axes = input_61_axes_0, beta = text_encoder_text_model_encoder_layers_3_layer_norm2_bias, epsilon = var_12, gamma = text_encoder_text_model_encoder_layers_3_layer_norm2_weight, x = input_59)[name = tensor("input_61")]; - tensor input_63 = linear(bias = text_encoder_text_model_encoder_layers_3_mlp_fc1_bias, weight = text_encoder_text_model_encoder_layers_3_mlp_fc1_weight, x = input_61)[name = tensor("input_63")]; - tensor var_419 = const()[name = tensor("op_419"), val = tensor(0x1.b3b646p+0)]; - tensor var_420 = mul(x = input_63, y = var_419)[name = tensor("op_420")]; - tensor var_421 = sigmoid(x = var_420)[name = tensor("op_421")]; - tensor input_65 = mul(x = input_63, y = var_421)[name = tensor("input_65")]; - tensor hidden_states_23 = linear(bias = text_encoder_text_model_encoder_layers_3_mlp_fc2_bias, weight = text_encoder_text_model_encoder_layers_3_mlp_fc2_weight, x = input_65)[name = tensor("hidden_states_23")]; - tensor input_67 = add(x = input_59, y = hidden_states_23)[name = tensor("input_67")]; + tensor text_encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123279872)))]; + tensor text_encoder_text_model_encoder_layers_3_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123281472)))]; + tensor input_61_cast = layer_norm(axes = input_61_axes_0, beta = text_encoder_text_model_encoder_layers_3_layer_norm2_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16, x = input_59_cast)[name = tensor("input_61_cast")]; + tensor text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123283072)))]; + tensor text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(128001728)))]; + tensor input_63_cast = linear(bias = text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16, x = input_61_cast)[name = tensor("input_63_cast")]; + tensor var_421_to_fp16 = const()[name = tensor("op_421_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_422_cast = mul(x = input_63_cast, y = var_421_to_fp16)[name = tensor("op_422_cast")]; + tensor var_423_cast = sigmoid(x = var_422_cast)[name = tensor("op_423_cast")]; + tensor input_65_cast = mul(x = input_63_cast, y = var_423_cast)[name = tensor("input_65_cast")]; + tensor text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(128007936)))]; + tensor text_encoder_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132726592)))]; + tensor hidden_states_23_cast = linear(bias = text_encoder_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16, x = input_65_cast)[name = tensor("hidden_states_23_cast")]; + tensor input_67_cast = add(x = input_59_cast, y = hidden_states_23_cast)[name = tensor("input_67_cast")]; tensor hidden_states_25_axes_0 = const()[name = tensor("hidden_states_25_axes_0"), val = tensor([-1])]; - tensor hidden_states_25 = layer_norm(axes = hidden_states_25_axes_0, beta = text_encoder_text_model_encoder_layers_4_layer_norm1_bias, epsilon = var_12, gamma = text_encoder_text_model_encoder_layers_4_layer_norm1_weight, x = input_67)[name = tensor("hidden_states_25")]; - tensor var_445 = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_q_proj_bias, weight = text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight, x = hidden_states_25)[name = tensor("op_445")]; - tensor var_446 = const()[name = tensor("op_446"), val = tensor(0x1p-3)]; - tensor tensor_29 = mul(x = var_445, y = var_446)[name = tensor("tensor_29")]; - tensor tensor_25 = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_k_proj_bias, weight = text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight, x = hidden_states_25)[name = tensor("tensor_25")]; - tensor var_451 = const()[name = tensor("op_451"), val = tensor([1, -1, 12, 64])]; - tensor var_452 = reshape(shape = var_451, x = tensor_25)[name = tensor("op_452")]; - tensor var_453_perm_0 = const()[name = tensor("op_453_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor tensor_27 = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_v_proj_bias, weight = text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight, x = hidden_states_25)[name = tensor("tensor_27")]; - tensor var_458 = const()[name = tensor("op_458"), val = tensor([1, -1, 12, 64])]; - tensor var_459 = reshape(shape = var_458, x = tensor_27)[name = tensor("op_459")]; - tensor var_460_perm_0 = const()[name = tensor("op_460_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_467 = const()[name = tensor("op_467"), val = tensor([1, 77, 12, 64])]; - tensor var_468 = reshape(shape = var_467, x = tensor_29)[name = tensor("op_468")]; - tensor var_469_perm_0 = const()[name = tensor("op_469_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_471 = const()[name = tensor("op_471"), val = tensor([12, -1, 64])]; - tensor transpose_37 = transpose(perm = var_469_perm_0, x = var_468)[name = tensor("transpose_37")]; - tensor query_states_9 = reshape(shape = var_471, x = transpose_37)[name = tensor("query_states_9")]; + tensor text_encoder_text_model_encoder_layers_4_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132728192)))]; + tensor text_encoder_text_model_encoder_layers_4_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132729792)))]; + tensor hidden_states_25_cast = layer_norm(axes = hidden_states_25_axes_0, beta = text_encoder_text_model_encoder_layers_4_layer_norm1_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_4_layer_norm1_weight_to_fp16, x = input_67_cast)[name = tensor("hidden_states_25_cast")]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132731392)))]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133911104)))]; + tensor var_447_cast = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16, x = hidden_states_25_cast)[name = tensor("op_447_cast")]; + tensor var_448_to_fp16 = const()[name = tensor("op_448_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_29_cast = mul(x = var_447_cast, y = var_448_to_fp16)[name = tensor("tensor_29_cast")]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133912704)))]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135092416)))]; + tensor tensor_25_cast = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16, x = hidden_states_25_cast)[name = tensor("tensor_25_cast")]; + tensor var_453 = const()[name = tensor("op_453"), val = tensor([1, -1, 12, 64])]; + tensor var_454_cast = reshape(shape = var_453, x = tensor_25_cast)[name = tensor("op_454_cast")]; + tensor var_455_perm_0 = const()[name = tensor("op_455_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135094016)))]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136273728)))]; + tensor tensor_27_cast = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16, x = hidden_states_25_cast)[name = tensor("tensor_27_cast")]; + tensor var_460 = const()[name = tensor("op_460"), val = tensor([1, -1, 12, 64])]; + tensor var_461_cast = reshape(shape = var_460, x = tensor_27_cast)[name = tensor("op_461_cast")]; + tensor var_462_perm_0 = const()[name = tensor("op_462_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_469 = const()[name = tensor("op_469"), val = tensor([1, 77, 12, 64])]; + tensor var_470_cast = reshape(shape = var_469, x = tensor_29_cast)[name = tensor("op_470_cast")]; + tensor var_471_perm_0 = const()[name = tensor("op_471_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_473 = const()[name = tensor("op_473"), val = tensor([12, -1, 64])]; - tensor transpose_39 = transpose(perm = var_453_perm_0, x = var_452)[name = tensor("transpose_39")]; - tensor key_states_19 = reshape(shape = var_473, x = transpose_39)[name = tensor("key_states_19")]; + tensor transpose_37 = transpose(perm = var_471_perm_0, x = var_470_cast)[name = tensor("transpose_37")]; + tensor query_states_9_cast = reshape(shape = var_473, x = transpose_37)[name = tensor("query_states_9_cast")]; tensor var_475 = const()[name = tensor("op_475"), val = tensor([12, -1, 64])]; - tensor transpose_38 = transpose(perm = var_460_perm_0, x = var_459)[name = tensor("transpose_38")]; - tensor value_states_19 = reshape(shape = var_475, x = transpose_38)[name = tensor("value_states_19")]; - tensor var_478_perm_0 = const()[name = tensor("op_478_perm_0"), val = tensor([0, 2, 1])]; + tensor transpose_39 = transpose(perm = var_455_perm_0, x = var_454_cast)[name = tensor("transpose_39")]; + tensor key_states_19_cast = reshape(shape = var_475, x = transpose_39)[name = tensor("key_states_19_cast")]; + tensor var_477 = const()[name = tensor("op_477"), val = tensor([12, -1, 64])]; + tensor transpose_38 = transpose(perm = var_462_perm_0, x = var_461_cast)[name = tensor("transpose_38")]; + tensor value_states_19_cast = reshape(shape = var_477, x = transpose_38)[name = tensor("value_states_19_cast")]; + tensor var_480_perm_0 = const()[name = tensor("op_480_perm_0"), val = tensor([0, 2, 1])]; tensor attn_weights_25_transpose_x_0 = const()[name = tensor("attn_weights_25_transpose_x_0"), val = tensor(false)]; tensor attn_weights_25_transpose_y_0 = const()[name = tensor("attn_weights_25_transpose_y_0"), val = tensor(false)]; - tensor transpose_36 = transpose(perm = var_478_perm_0, x = key_states_19)[name = tensor("transpose_36")]; - tensor attn_weights_25 = matmul(transpose_x = attn_weights_25_transpose_x_0, transpose_y = attn_weights_25_transpose_y_0, x = query_states_9, y = transpose_36)[name = tensor("attn_weights_25")]; - tensor var_480 = const()[name = tensor("op_480"), val = tensor([1, 12, 77, 77])]; - tensor var_481 = reshape(shape = var_480, x = attn_weights_25)[name = tensor("op_481")]; - tensor attn_weights_27 = add(x = var_481, y = causal_attention_mask)[name = tensor("attn_weights_27")]; - tensor var_486 = const()[name = tensor("op_486"), val = tensor([12, 77, 77])]; - tensor input_69 = reshape(shape = var_486, x = attn_weights_27)[name = tensor("input_69")]; - tensor input_71 = softmax(axis = var_5, x = input_69)[name = tensor("input_71")]; + tensor transpose_36 = transpose(perm = var_480_perm_0, x = key_states_19_cast)[name = tensor("transpose_36")]; + tensor attn_weights_25_cast = matmul(transpose_x = attn_weights_25_transpose_x_0, transpose_y = attn_weights_25_transpose_y_0, x = query_states_9_cast, y = transpose_36)[name = tensor("attn_weights_25_cast")]; + tensor var_482 = const()[name = tensor("op_482"), val = tensor([1, 12, 77, 77])]; + tensor var_483_cast = reshape(shape = var_482, x = attn_weights_25_cast)[name = tensor("op_483_cast")]; + tensor attn_weights_27_cast = add(x = var_483_cast, y = causal_attention_mask_to_fp16)[name = tensor("attn_weights_27_cast")]; + tensor var_488 = const()[name = tensor("op_488"), val = tensor([12, 77, 77])]; + tensor input_69_cast = reshape(shape = var_488, x = attn_weights_27_cast)[name = tensor("input_69_cast")]; + tensor input_71_cast = softmax(axis = var_5, x = input_69_cast)[name = tensor("input_71_cast")]; tensor attn_output_25_transpose_x_0 = const()[name = tensor("attn_output_25_transpose_x_0"), val = tensor(false)]; tensor attn_output_25_transpose_y_0 = const()[name = tensor("attn_output_25_transpose_y_0"), val = tensor(false)]; - tensor attn_output_25 = matmul(transpose_x = attn_output_25_transpose_x_0, transpose_y = attn_output_25_transpose_y_0, x = input_71, y = value_states_19)[name = tensor("attn_output_25")]; - tensor var_491 = const()[name = tensor("op_491"), val = tensor([1, 12, 77, 64])]; - tensor attn_output_27 = reshape(shape = var_491, x = attn_output_25)[name = tensor("attn_output_27")]; + tensor attn_output_25_cast = matmul(transpose_x = attn_output_25_transpose_x_0, transpose_y = attn_output_25_transpose_y_0, x = input_71_cast, y = value_states_19_cast)[name = tensor("attn_output_25_cast")]; + tensor var_493 = const()[name = tensor("op_493"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_27_cast = reshape(shape = var_493, x = attn_output_25_cast)[name = tensor("attn_output_27_cast")]; tensor attn_output_29_perm_0 = const()[name = tensor("attn_output_29_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_494 = const()[name = tensor("op_494"), val = tensor([1, 77, 768])]; - tensor transpose_35 = transpose(perm = attn_output_29_perm_0, x = attn_output_27)[name = tensor("transpose_35")]; - tensor input_73 = reshape(shape = var_494, x = transpose_35)[name = tensor("input_73")]; - tensor hidden_states_27 = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_out_proj_bias, weight = text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight, x = input_73)[name = tensor("hidden_states_27")]; - tensor input_75 = add(x = input_67, y = hidden_states_27)[name = tensor("input_75")]; + tensor var_496 = const()[name = tensor("op_496"), val = tensor([1, 77, 768])]; + tensor transpose_35 = transpose(perm = attn_output_29_perm_0, x = attn_output_27_cast)[name = tensor("transpose_35")]; + tensor input_73_cast = reshape(shape = var_496, x = transpose_35)[name = tensor("input_73_cast")]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136275328)))]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137455040)))]; + tensor hidden_states_27_cast = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16, x = input_73_cast)[name = tensor("hidden_states_27_cast")]; + tensor input_75_cast = add(x = input_67_cast, y = hidden_states_27_cast)[name = tensor("input_75_cast")]; tensor input_77_axes_0 = const()[name = tensor("input_77_axes_0"), val = tensor([-1])]; - tensor input_77 = layer_norm(axes = input_77_axes_0, beta = text_encoder_text_model_encoder_layers_4_layer_norm2_bias, epsilon = var_12, gamma = text_encoder_text_model_encoder_layers_4_layer_norm2_weight, x = input_75)[name = tensor("input_77")]; - tensor input_79 = linear(bias = text_encoder_text_model_encoder_layers_4_mlp_fc1_bias, weight = text_encoder_text_model_encoder_layers_4_mlp_fc1_weight, x = input_77)[name = tensor("input_79")]; - tensor var_509 = const()[name = tensor("op_509"), val = tensor(0x1.b3b646p+0)]; - tensor var_510 = mul(x = input_79, y = var_509)[name = tensor("op_510")]; - tensor var_511 = sigmoid(x = var_510)[name = tensor("op_511")]; - tensor input_81 = mul(x = input_79, y = var_511)[name = tensor("input_81")]; - tensor hidden_states_29 = linear(bias = text_encoder_text_model_encoder_layers_4_mlp_fc2_bias, weight = text_encoder_text_model_encoder_layers_4_mlp_fc2_weight, x = input_81)[name = tensor("hidden_states_29")]; - tensor input_83 = add(x = input_75, y = hidden_states_29)[name = tensor("input_83")]; + tensor text_encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137456640)))]; + tensor text_encoder_text_model_encoder_layers_4_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137458240)))]; + tensor input_77_cast = layer_norm(axes = input_77_axes_0, beta = text_encoder_text_model_encoder_layers_4_layer_norm2_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16, x = input_75_cast)[name = tensor("input_77_cast")]; + tensor text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137459840)))]; + tensor text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142178496)))]; + tensor input_79_cast = linear(bias = text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16, x = input_77_cast)[name = tensor("input_79_cast")]; + tensor var_511_to_fp16 = const()[name = tensor("op_511_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_512_cast = mul(x = input_79_cast, y = var_511_to_fp16)[name = tensor("op_512_cast")]; + tensor var_513_cast = sigmoid(x = var_512_cast)[name = tensor("op_513_cast")]; + tensor input_81_cast = mul(x = input_79_cast, y = var_513_cast)[name = tensor("input_81_cast")]; + tensor text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142184704)))]; + tensor text_encoder_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(146903360)))]; + tensor hidden_states_29_cast = linear(bias = text_encoder_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16, x = input_81_cast)[name = tensor("hidden_states_29_cast")]; + tensor input_83_cast = add(x = input_75_cast, y = hidden_states_29_cast)[name = tensor("input_83_cast")]; tensor hidden_states_31_axes_0 = const()[name = tensor("hidden_states_31_axes_0"), val = tensor([-1])]; - tensor hidden_states_31 = layer_norm(axes = hidden_states_31_axes_0, beta = text_encoder_text_model_encoder_layers_5_layer_norm1_bias, epsilon = var_12, gamma = text_encoder_text_model_encoder_layers_5_layer_norm1_weight, x = input_83)[name = tensor("hidden_states_31")]; - tensor var_535 = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_q_proj_bias, weight = text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight, x = hidden_states_31)[name = tensor("op_535")]; - tensor var_536 = const()[name = tensor("op_536"), val = tensor(0x1p-3)]; - tensor tensor_35 = mul(x = var_535, y = var_536)[name = tensor("tensor_35")]; - tensor tensor_31 = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_k_proj_bias, weight = text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight, x = hidden_states_31)[name = tensor("tensor_31")]; - tensor var_541 = const()[name = tensor("op_541"), val = tensor([1, -1, 12, 64])]; - tensor var_542 = reshape(shape = var_541, x = tensor_31)[name = tensor("op_542")]; - tensor var_543_perm_0 = const()[name = tensor("op_543_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor tensor_33 = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_v_proj_bias, weight = text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight, x = hidden_states_31)[name = tensor("tensor_33")]; - tensor var_548 = const()[name = tensor("op_548"), val = tensor([1, -1, 12, 64])]; - tensor var_549 = reshape(shape = var_548, x = tensor_33)[name = tensor("op_549")]; - tensor var_550_perm_0 = const()[name = tensor("op_550_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_557 = const()[name = tensor("op_557"), val = tensor([1, 77, 12, 64])]; - tensor var_558 = reshape(shape = var_557, x = tensor_35)[name = tensor("op_558")]; - tensor var_559_perm_0 = const()[name = tensor("op_559_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_561 = const()[name = tensor("op_561"), val = tensor([12, -1, 64])]; - tensor transpose_32 = transpose(perm = var_559_perm_0, x = var_558)[name = tensor("transpose_32")]; - tensor query_states_11 = reshape(shape = var_561, x = transpose_32)[name = tensor("query_states_11")]; + tensor text_encoder_text_model_encoder_layers_5_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(146904960)))]; + tensor text_encoder_text_model_encoder_layers_5_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(146906560)))]; + tensor hidden_states_31_cast = layer_norm(axes = hidden_states_31_axes_0, beta = text_encoder_text_model_encoder_layers_5_layer_norm1_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_5_layer_norm1_weight_to_fp16, x = input_83_cast)[name = tensor("hidden_states_31_cast")]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(146908160)))]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(148087872)))]; + tensor var_537_cast = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16, x = hidden_states_31_cast)[name = tensor("op_537_cast")]; + tensor var_538_to_fp16 = const()[name = tensor("op_538_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_35_cast = mul(x = var_537_cast, y = var_538_to_fp16)[name = tensor("tensor_35_cast")]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(148089472)))]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(149269184)))]; + tensor tensor_31_cast = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16, x = hidden_states_31_cast)[name = tensor("tensor_31_cast")]; + tensor var_543 = const()[name = tensor("op_543"), val = tensor([1, -1, 12, 64])]; + tensor var_544_cast = reshape(shape = var_543, x = tensor_31_cast)[name = tensor("op_544_cast")]; + tensor var_545_perm_0 = const()[name = tensor("op_545_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(149270784)))]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(150450496)))]; + tensor tensor_33_cast = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16, x = hidden_states_31_cast)[name = tensor("tensor_33_cast")]; + tensor var_550 = const()[name = tensor("op_550"), val = tensor([1, -1, 12, 64])]; + tensor var_551_cast = reshape(shape = var_550, x = tensor_33_cast)[name = tensor("op_551_cast")]; + tensor var_552_perm_0 = const()[name = tensor("op_552_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_559 = const()[name = tensor("op_559"), val = tensor([1, 77, 12, 64])]; + tensor var_560_cast = reshape(shape = var_559, x = tensor_35_cast)[name = tensor("op_560_cast")]; + tensor var_561_perm_0 = const()[name = tensor("op_561_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_563 = const()[name = tensor("op_563"), val = tensor([12, -1, 64])]; - tensor transpose_34 = transpose(perm = var_543_perm_0, x = var_542)[name = tensor("transpose_34")]; - tensor key_states_23 = reshape(shape = var_563, x = transpose_34)[name = tensor("key_states_23")]; + tensor transpose_32 = transpose(perm = var_561_perm_0, x = var_560_cast)[name = tensor("transpose_32")]; + tensor query_states_11_cast = reshape(shape = var_563, x = transpose_32)[name = tensor("query_states_11_cast")]; tensor var_565 = const()[name = tensor("op_565"), val = tensor([12, -1, 64])]; - tensor transpose_33 = transpose(perm = var_550_perm_0, x = var_549)[name = tensor("transpose_33")]; - tensor value_states_23 = reshape(shape = var_565, x = transpose_33)[name = tensor("value_states_23")]; - tensor var_568_perm_0 = const()[name = tensor("op_568_perm_0"), val = tensor([0, 2, 1])]; + tensor transpose_34 = transpose(perm = var_545_perm_0, x = var_544_cast)[name = tensor("transpose_34")]; + tensor key_states_23_cast = reshape(shape = var_565, x = transpose_34)[name = tensor("key_states_23_cast")]; + tensor var_567 = const()[name = tensor("op_567"), val = tensor([12, -1, 64])]; + tensor transpose_33 = transpose(perm = var_552_perm_0, x = var_551_cast)[name = tensor("transpose_33")]; + tensor value_states_23_cast = reshape(shape = var_567, x = transpose_33)[name = tensor("value_states_23_cast")]; + tensor var_570_perm_0 = const()[name = tensor("op_570_perm_0"), val = tensor([0, 2, 1])]; tensor attn_weights_31_transpose_x_0 = const()[name = tensor("attn_weights_31_transpose_x_0"), val = tensor(false)]; tensor attn_weights_31_transpose_y_0 = const()[name = tensor("attn_weights_31_transpose_y_0"), val = tensor(false)]; - tensor transpose_31 = transpose(perm = var_568_perm_0, x = key_states_23)[name = tensor("transpose_31")]; - tensor attn_weights_31 = matmul(transpose_x = attn_weights_31_transpose_x_0, transpose_y = attn_weights_31_transpose_y_0, x = query_states_11, y = transpose_31)[name = tensor("attn_weights_31")]; - tensor var_570 = const()[name = tensor("op_570"), val = tensor([1, 12, 77, 77])]; - tensor var_571 = reshape(shape = var_570, x = attn_weights_31)[name = tensor("op_571")]; - tensor attn_weights_33 = add(x = var_571, y = causal_attention_mask)[name = tensor("attn_weights_33")]; - tensor var_576 = const()[name = tensor("op_576"), val = tensor([12, 77, 77])]; - tensor input_85 = reshape(shape = var_576, x = attn_weights_33)[name = tensor("input_85")]; - tensor input_87 = softmax(axis = var_5, x = input_85)[name = tensor("input_87")]; + tensor transpose_31 = transpose(perm = var_570_perm_0, x = key_states_23_cast)[name = tensor("transpose_31")]; + tensor attn_weights_31_cast = matmul(transpose_x = attn_weights_31_transpose_x_0, transpose_y = attn_weights_31_transpose_y_0, x = query_states_11_cast, y = transpose_31)[name = tensor("attn_weights_31_cast")]; + tensor var_572 = const()[name = tensor("op_572"), val = tensor([1, 12, 77, 77])]; + tensor var_573_cast = reshape(shape = var_572, x = attn_weights_31_cast)[name = tensor("op_573_cast")]; + tensor attn_weights_33_cast = add(x = var_573_cast, y = causal_attention_mask_to_fp16)[name = tensor("attn_weights_33_cast")]; + tensor var_578 = const()[name = tensor("op_578"), val = tensor([12, 77, 77])]; + tensor input_85_cast = reshape(shape = var_578, x = attn_weights_33_cast)[name = tensor("input_85_cast")]; + tensor input_87_cast = softmax(axis = var_5, x = input_85_cast)[name = tensor("input_87_cast")]; tensor attn_output_31_transpose_x_0 = const()[name = tensor("attn_output_31_transpose_x_0"), val = tensor(false)]; tensor attn_output_31_transpose_y_0 = const()[name = tensor("attn_output_31_transpose_y_0"), val = tensor(false)]; - tensor attn_output_31 = matmul(transpose_x = attn_output_31_transpose_x_0, transpose_y = attn_output_31_transpose_y_0, x = input_87, y = value_states_23)[name = tensor("attn_output_31")]; - tensor var_581 = const()[name = tensor("op_581"), val = tensor([1, 12, 77, 64])]; - tensor attn_output_33 = reshape(shape = var_581, x = attn_output_31)[name = tensor("attn_output_33")]; + tensor attn_output_31_cast = matmul(transpose_x = attn_output_31_transpose_x_0, transpose_y = attn_output_31_transpose_y_0, x = input_87_cast, y = value_states_23_cast)[name = tensor("attn_output_31_cast")]; + tensor var_583 = const()[name = tensor("op_583"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_33_cast = reshape(shape = var_583, x = attn_output_31_cast)[name = tensor("attn_output_33_cast")]; tensor attn_output_35_perm_0 = const()[name = tensor("attn_output_35_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_584 = const()[name = tensor("op_584"), val = tensor([1, 77, 768])]; - tensor transpose_30 = transpose(perm = attn_output_35_perm_0, x = attn_output_33)[name = tensor("transpose_30")]; - tensor input_89 = reshape(shape = var_584, x = transpose_30)[name = tensor("input_89")]; - tensor hidden_states_33 = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_out_proj_bias, weight = text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight, x = input_89)[name = tensor("hidden_states_33")]; - tensor input_91 = add(x = input_83, y = hidden_states_33)[name = tensor("input_91")]; + tensor var_586 = const()[name = tensor("op_586"), val = tensor([1, 77, 768])]; + tensor transpose_30 = transpose(perm = attn_output_35_perm_0, x = attn_output_33_cast)[name = tensor("transpose_30")]; + tensor input_89_cast = reshape(shape = var_586, x = transpose_30)[name = tensor("input_89_cast")]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(150452096)))]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151631808)))]; + tensor hidden_states_33_cast = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16, x = input_89_cast)[name = tensor("hidden_states_33_cast")]; + tensor input_91_cast = add(x = input_83_cast, y = hidden_states_33_cast)[name = tensor("input_91_cast")]; tensor input_93_axes_0 = const()[name = tensor("input_93_axes_0"), val = tensor([-1])]; - tensor input_93 = layer_norm(axes = input_93_axes_0, beta = text_encoder_text_model_encoder_layers_5_layer_norm2_bias, epsilon = var_12, gamma = text_encoder_text_model_encoder_layers_5_layer_norm2_weight, x = input_91)[name = tensor("input_93")]; - tensor input_95 = linear(bias = text_encoder_text_model_encoder_layers_5_mlp_fc1_bias, weight = text_encoder_text_model_encoder_layers_5_mlp_fc1_weight, x = input_93)[name = tensor("input_95")]; - tensor var_599 = const()[name = tensor("op_599"), val = tensor(0x1.b3b646p+0)]; - tensor var_600 = mul(x = input_95, y = var_599)[name = tensor("op_600")]; - tensor var_601 = sigmoid(x = var_600)[name = tensor("op_601")]; - tensor input_97 = mul(x = input_95, y = var_601)[name = tensor("input_97")]; - tensor hidden_states_35 = linear(bias = text_encoder_text_model_encoder_layers_5_mlp_fc2_bias, weight = text_encoder_text_model_encoder_layers_5_mlp_fc2_weight, x = input_97)[name = tensor("hidden_states_35")]; - tensor input_99 = add(x = input_91, y = hidden_states_35)[name = tensor("input_99")]; + tensor text_encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151633408)))]; + tensor text_encoder_text_model_encoder_layers_5_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151635008)))]; + tensor input_93_cast = layer_norm(axes = input_93_axes_0, beta = text_encoder_text_model_encoder_layers_5_layer_norm2_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16, x = input_91_cast)[name = tensor("input_93_cast")]; + tensor text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151636608)))]; + tensor text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(156355264)))]; + tensor input_95_cast = linear(bias = text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16, x = input_93_cast)[name = tensor("input_95_cast")]; + tensor var_601_to_fp16 = const()[name = tensor("op_601_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_602_cast = mul(x = input_95_cast, y = var_601_to_fp16)[name = tensor("op_602_cast")]; + tensor var_603_cast = sigmoid(x = var_602_cast)[name = tensor("op_603_cast")]; + tensor input_97_cast = mul(x = input_95_cast, y = var_603_cast)[name = tensor("input_97_cast")]; + tensor text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(156361472)))]; + tensor text_encoder_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161080128)))]; + tensor hidden_states_35_cast = linear(bias = text_encoder_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16, x = input_97_cast)[name = tensor("hidden_states_35_cast")]; + tensor input_99_cast = add(x = input_91_cast, y = hidden_states_35_cast)[name = tensor("input_99_cast")]; tensor hidden_states_37_axes_0 = const()[name = tensor("hidden_states_37_axes_0"), val = tensor([-1])]; - tensor hidden_states_37 = layer_norm(axes = hidden_states_37_axes_0, beta = text_encoder_text_model_encoder_layers_6_layer_norm1_bias, epsilon = var_12, gamma = text_encoder_text_model_encoder_layers_6_layer_norm1_weight, x = input_99)[name = tensor("hidden_states_37")]; - tensor var_625 = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_q_proj_bias, weight = text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight, x = hidden_states_37)[name = tensor("op_625")]; - tensor var_626 = const()[name = tensor("op_626"), val = tensor(0x1p-3)]; - tensor tensor_41 = mul(x = var_625, y = var_626)[name = tensor("tensor_41")]; - tensor tensor_37 = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_k_proj_bias, weight = text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight, x = hidden_states_37)[name = tensor("tensor_37")]; - tensor var_631 = const()[name = tensor("op_631"), val = tensor([1, -1, 12, 64])]; - tensor var_632 = reshape(shape = var_631, x = tensor_37)[name = tensor("op_632")]; - tensor var_633_perm_0 = const()[name = tensor("op_633_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor tensor_39 = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_v_proj_bias, weight = text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight, x = hidden_states_37)[name = tensor("tensor_39")]; - tensor var_638 = const()[name = tensor("op_638"), val = tensor([1, -1, 12, 64])]; - tensor var_639 = reshape(shape = var_638, x = tensor_39)[name = tensor("op_639")]; - tensor var_640_perm_0 = const()[name = tensor("op_640_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_647 = const()[name = tensor("op_647"), val = tensor([1, 77, 12, 64])]; - tensor var_648 = reshape(shape = var_647, x = tensor_41)[name = tensor("op_648")]; - tensor var_649_perm_0 = const()[name = tensor("op_649_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_651 = const()[name = tensor("op_651"), val = tensor([12, -1, 64])]; - tensor transpose_27 = transpose(perm = var_649_perm_0, x = var_648)[name = tensor("transpose_27")]; - tensor query_states_13 = reshape(shape = var_651, x = transpose_27)[name = tensor("query_states_13")]; + tensor text_encoder_text_model_encoder_layers_6_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161081728)))]; + tensor text_encoder_text_model_encoder_layers_6_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161083328)))]; + tensor hidden_states_37_cast = layer_norm(axes = hidden_states_37_axes_0, beta = text_encoder_text_model_encoder_layers_6_layer_norm1_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_6_layer_norm1_weight_to_fp16, x = input_99_cast)[name = tensor("hidden_states_37_cast")]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161084928)))]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162264640)))]; + tensor var_627_cast = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16, x = hidden_states_37_cast)[name = tensor("op_627_cast")]; + tensor var_628_to_fp16 = const()[name = tensor("op_628_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_41_cast = mul(x = var_627_cast, y = var_628_to_fp16)[name = tensor("tensor_41_cast")]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162266240)))]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163445952)))]; + tensor tensor_37_cast = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16, x = hidden_states_37_cast)[name = tensor("tensor_37_cast")]; + tensor var_633 = const()[name = tensor("op_633"), val = tensor([1, -1, 12, 64])]; + tensor var_634_cast = reshape(shape = var_633, x = tensor_37_cast)[name = tensor("op_634_cast")]; + tensor var_635_perm_0 = const()[name = tensor("op_635_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163447552)))]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(164627264)))]; + tensor tensor_39_cast = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16, x = hidden_states_37_cast)[name = tensor("tensor_39_cast")]; + tensor var_640 = const()[name = tensor("op_640"), val = tensor([1, -1, 12, 64])]; + tensor var_641_cast = reshape(shape = var_640, x = tensor_39_cast)[name = tensor("op_641_cast")]; + tensor var_642_perm_0 = const()[name = tensor("op_642_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_649 = const()[name = tensor("op_649"), val = tensor([1, 77, 12, 64])]; + tensor var_650_cast = reshape(shape = var_649, x = tensor_41_cast)[name = tensor("op_650_cast")]; + tensor var_651_perm_0 = const()[name = tensor("op_651_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_653 = const()[name = tensor("op_653"), val = tensor([12, -1, 64])]; - tensor transpose_29 = transpose(perm = var_633_perm_0, x = var_632)[name = tensor("transpose_29")]; - tensor key_states_27 = reshape(shape = var_653, x = transpose_29)[name = tensor("key_states_27")]; + tensor transpose_27 = transpose(perm = var_651_perm_0, x = var_650_cast)[name = tensor("transpose_27")]; + tensor query_states_13_cast = reshape(shape = var_653, x = transpose_27)[name = tensor("query_states_13_cast")]; tensor var_655 = const()[name = tensor("op_655"), val = tensor([12, -1, 64])]; - tensor transpose_28 = transpose(perm = var_640_perm_0, x = var_639)[name = tensor("transpose_28")]; - tensor value_states_27 = reshape(shape = var_655, x = transpose_28)[name = tensor("value_states_27")]; - tensor var_658_perm_0 = const()[name = tensor("op_658_perm_0"), val = tensor([0, 2, 1])]; + tensor transpose_29 = transpose(perm = var_635_perm_0, x = var_634_cast)[name = tensor("transpose_29")]; + tensor key_states_27_cast = reshape(shape = var_655, x = transpose_29)[name = tensor("key_states_27_cast")]; + tensor var_657 = const()[name = tensor("op_657"), val = tensor([12, -1, 64])]; + tensor transpose_28 = transpose(perm = var_642_perm_0, x = var_641_cast)[name = tensor("transpose_28")]; + tensor value_states_27_cast = reshape(shape = var_657, x = transpose_28)[name = tensor("value_states_27_cast")]; + tensor var_660_perm_0 = const()[name = tensor("op_660_perm_0"), val = tensor([0, 2, 1])]; tensor attn_weights_37_transpose_x_0 = const()[name = tensor("attn_weights_37_transpose_x_0"), val = tensor(false)]; tensor attn_weights_37_transpose_y_0 = const()[name = tensor("attn_weights_37_transpose_y_0"), val = tensor(false)]; - tensor transpose_26 = transpose(perm = var_658_perm_0, x = key_states_27)[name = tensor("transpose_26")]; - tensor attn_weights_37 = matmul(transpose_x = attn_weights_37_transpose_x_0, transpose_y = attn_weights_37_transpose_y_0, x = query_states_13, y = transpose_26)[name = tensor("attn_weights_37")]; - tensor var_660 = const()[name = tensor("op_660"), val = tensor([1, 12, 77, 77])]; - tensor var_661 = reshape(shape = var_660, x = attn_weights_37)[name = tensor("op_661")]; - tensor attn_weights_39 = add(x = var_661, y = causal_attention_mask)[name = tensor("attn_weights_39")]; - tensor var_666 = const()[name = tensor("op_666"), val = tensor([12, 77, 77])]; - tensor input_101 = reshape(shape = var_666, x = attn_weights_39)[name = tensor("input_101")]; - tensor input_103 = softmax(axis = var_5, x = input_101)[name = tensor("input_103")]; + tensor transpose_26 = transpose(perm = var_660_perm_0, x = key_states_27_cast)[name = tensor("transpose_26")]; + tensor attn_weights_37_cast = matmul(transpose_x = attn_weights_37_transpose_x_0, transpose_y = attn_weights_37_transpose_y_0, x = query_states_13_cast, y = transpose_26)[name = tensor("attn_weights_37_cast")]; + tensor var_662 = const()[name = tensor("op_662"), val = tensor([1, 12, 77, 77])]; + tensor var_663_cast = reshape(shape = var_662, x = attn_weights_37_cast)[name = tensor("op_663_cast")]; + tensor attn_weights_39_cast = add(x = var_663_cast, y = causal_attention_mask_to_fp16)[name = tensor("attn_weights_39_cast")]; + tensor var_668 = const()[name = tensor("op_668"), val = tensor([12, 77, 77])]; + tensor input_101_cast = reshape(shape = var_668, x = attn_weights_39_cast)[name = tensor("input_101_cast")]; + tensor input_103_cast = softmax(axis = var_5, x = input_101_cast)[name = tensor("input_103_cast")]; tensor attn_output_37_transpose_x_0 = const()[name = tensor("attn_output_37_transpose_x_0"), val = tensor(false)]; tensor attn_output_37_transpose_y_0 = const()[name = tensor("attn_output_37_transpose_y_0"), val = tensor(false)]; - tensor attn_output_37 = matmul(transpose_x = attn_output_37_transpose_x_0, transpose_y = attn_output_37_transpose_y_0, x = input_103, y = value_states_27)[name = tensor("attn_output_37")]; - tensor var_671 = const()[name = tensor("op_671"), val = tensor([1, 12, 77, 64])]; - tensor attn_output_39 = reshape(shape = var_671, x = attn_output_37)[name = tensor("attn_output_39")]; + tensor attn_output_37_cast = matmul(transpose_x = attn_output_37_transpose_x_0, transpose_y = attn_output_37_transpose_y_0, x = input_103_cast, y = value_states_27_cast)[name = tensor("attn_output_37_cast")]; + tensor var_673 = const()[name = tensor("op_673"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_39_cast = reshape(shape = var_673, x = attn_output_37_cast)[name = tensor("attn_output_39_cast")]; tensor attn_output_41_perm_0 = const()[name = tensor("attn_output_41_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_674 = const()[name = tensor("op_674"), val = tensor([1, 77, 768])]; - tensor transpose_25 = transpose(perm = attn_output_41_perm_0, x = attn_output_39)[name = tensor("transpose_25")]; - tensor input_105 = reshape(shape = var_674, x = transpose_25)[name = tensor("input_105")]; - tensor hidden_states_39 = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_out_proj_bias, weight = text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight, x = input_105)[name = tensor("hidden_states_39")]; - tensor input_107 = add(x = input_99, y = hidden_states_39)[name = tensor("input_107")]; + tensor var_676 = const()[name = tensor("op_676"), val = tensor([1, 77, 768])]; + tensor transpose_25 = transpose(perm = attn_output_41_perm_0, x = attn_output_39_cast)[name = tensor("transpose_25")]; + tensor input_105_cast = reshape(shape = var_676, x = transpose_25)[name = tensor("input_105_cast")]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(164628864)))]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165808576)))]; + tensor hidden_states_39_cast = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16, x = input_105_cast)[name = tensor("hidden_states_39_cast")]; + tensor input_107_cast = add(x = input_99_cast, y = hidden_states_39_cast)[name = tensor("input_107_cast")]; tensor input_109_axes_0 = const()[name = tensor("input_109_axes_0"), val = tensor([-1])]; - tensor input_109 = layer_norm(axes = input_109_axes_0, beta = text_encoder_text_model_encoder_layers_6_layer_norm2_bias, epsilon = var_12, gamma = text_encoder_text_model_encoder_layers_6_layer_norm2_weight, x = input_107)[name = tensor("input_109")]; - tensor input_111 = linear(bias = text_encoder_text_model_encoder_layers_6_mlp_fc1_bias, weight = text_encoder_text_model_encoder_layers_6_mlp_fc1_weight, x = input_109)[name = tensor("input_111")]; - tensor var_689 = const()[name = tensor("op_689"), val = tensor(0x1.b3b646p+0)]; - tensor var_690 = mul(x = input_111, y = var_689)[name = tensor("op_690")]; - tensor var_691 = sigmoid(x = var_690)[name = tensor("op_691")]; - tensor input_113 = mul(x = input_111, y = var_691)[name = tensor("input_113")]; - tensor hidden_states_41 = linear(bias = text_encoder_text_model_encoder_layers_6_mlp_fc2_bias, weight = text_encoder_text_model_encoder_layers_6_mlp_fc2_weight, x = input_113)[name = tensor("hidden_states_41")]; - tensor input_115 = add(x = input_107, y = hidden_states_41)[name = tensor("input_115")]; + tensor text_encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165810176)))]; + tensor text_encoder_text_model_encoder_layers_6_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165811776)))]; + tensor input_109_cast = layer_norm(axes = input_109_axes_0, beta = text_encoder_text_model_encoder_layers_6_layer_norm2_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16, x = input_107_cast)[name = tensor("input_109_cast")]; + tensor text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165813376)))]; + tensor text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(170532032)))]; + tensor input_111_cast = linear(bias = text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16, x = input_109_cast)[name = tensor("input_111_cast")]; + tensor var_691_to_fp16 = const()[name = tensor("op_691_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_692_cast = mul(x = input_111_cast, y = var_691_to_fp16)[name = tensor("op_692_cast")]; + tensor var_693_cast = sigmoid(x = var_692_cast)[name = tensor("op_693_cast")]; + tensor input_113_cast = mul(x = input_111_cast, y = var_693_cast)[name = tensor("input_113_cast")]; + tensor text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(170538240)))]; + tensor text_encoder_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(175256896)))]; + tensor hidden_states_41_cast = linear(bias = text_encoder_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16, x = input_113_cast)[name = tensor("hidden_states_41_cast")]; + tensor input_115_cast = add(x = input_107_cast, y = hidden_states_41_cast)[name = tensor("input_115_cast")]; tensor hidden_states_43_axes_0 = const()[name = tensor("hidden_states_43_axes_0"), val = tensor([-1])]; - tensor hidden_states_43 = layer_norm(axes = hidden_states_43_axes_0, beta = text_encoder_text_model_encoder_layers_7_layer_norm1_bias, epsilon = var_12, gamma = text_encoder_text_model_encoder_layers_7_layer_norm1_weight, x = input_115)[name = tensor("hidden_states_43")]; - tensor var_715 = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_q_proj_bias, weight = text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight, x = hidden_states_43)[name = tensor("op_715")]; - tensor var_716 = const()[name = tensor("op_716"), val = tensor(0x1p-3)]; - tensor tensor_47 = mul(x = var_715, y = var_716)[name = tensor("tensor_47")]; - tensor tensor_43 = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_k_proj_bias, weight = text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight, x = hidden_states_43)[name = tensor("tensor_43")]; - tensor var_721 = const()[name = tensor("op_721"), val = tensor([1, -1, 12, 64])]; - tensor var_722 = reshape(shape = var_721, x = tensor_43)[name = tensor("op_722")]; - tensor var_723_perm_0 = const()[name = tensor("op_723_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor tensor_45 = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_v_proj_bias, weight = text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight, x = hidden_states_43)[name = tensor("tensor_45")]; - tensor var_728 = const()[name = tensor("op_728"), val = tensor([1, -1, 12, 64])]; - tensor var_729 = reshape(shape = var_728, x = tensor_45)[name = tensor("op_729")]; - tensor var_730_perm_0 = const()[name = tensor("op_730_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_737 = const()[name = tensor("op_737"), val = tensor([1, 77, 12, 64])]; - tensor var_738 = reshape(shape = var_737, x = tensor_47)[name = tensor("op_738")]; - tensor var_739_perm_0 = const()[name = tensor("op_739_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_741 = const()[name = tensor("op_741"), val = tensor([12, -1, 64])]; - tensor transpose_22 = transpose(perm = var_739_perm_0, x = var_738)[name = tensor("transpose_22")]; - tensor query_states_15 = reshape(shape = var_741, x = transpose_22)[name = tensor("query_states_15")]; + tensor text_encoder_text_model_encoder_layers_7_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(175258496)))]; + tensor text_encoder_text_model_encoder_layers_7_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(175260096)))]; + tensor hidden_states_43_cast = layer_norm(axes = hidden_states_43_axes_0, beta = text_encoder_text_model_encoder_layers_7_layer_norm1_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_7_layer_norm1_weight_to_fp16, x = input_115_cast)[name = tensor("hidden_states_43_cast")]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(175261696)))]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176441408)))]; + tensor var_717_cast = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16, x = hidden_states_43_cast)[name = tensor("op_717_cast")]; + tensor var_718_to_fp16 = const()[name = tensor("op_718_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_47_cast = mul(x = var_717_cast, y = var_718_to_fp16)[name = tensor("tensor_47_cast")]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176443008)))]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(177622720)))]; + tensor tensor_43_cast = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16, x = hidden_states_43_cast)[name = tensor("tensor_43_cast")]; + tensor var_723 = const()[name = tensor("op_723"), val = tensor([1, -1, 12, 64])]; + tensor var_724_cast = reshape(shape = var_723, x = tensor_43_cast)[name = tensor("op_724_cast")]; + tensor var_725_perm_0 = const()[name = tensor("op_725_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(177624320)))]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178804032)))]; + tensor tensor_45_cast = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16, x = hidden_states_43_cast)[name = tensor("tensor_45_cast")]; + tensor var_730 = const()[name = tensor("op_730"), val = tensor([1, -1, 12, 64])]; + tensor var_731_cast = reshape(shape = var_730, x = tensor_45_cast)[name = tensor("op_731_cast")]; + tensor var_732_perm_0 = const()[name = tensor("op_732_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_739 = const()[name = tensor("op_739"), val = tensor([1, 77, 12, 64])]; + tensor var_740_cast = reshape(shape = var_739, x = tensor_47_cast)[name = tensor("op_740_cast")]; + tensor var_741_perm_0 = const()[name = tensor("op_741_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_743 = const()[name = tensor("op_743"), val = tensor([12, -1, 64])]; - tensor transpose_24 = transpose(perm = var_723_perm_0, x = var_722)[name = tensor("transpose_24")]; - tensor key_states_31 = reshape(shape = var_743, x = transpose_24)[name = tensor("key_states_31")]; + tensor transpose_22 = transpose(perm = var_741_perm_0, x = var_740_cast)[name = tensor("transpose_22")]; + tensor query_states_15_cast = reshape(shape = var_743, x = transpose_22)[name = tensor("query_states_15_cast")]; tensor var_745 = const()[name = tensor("op_745"), val = tensor([12, -1, 64])]; - tensor transpose_23 = transpose(perm = var_730_perm_0, x = var_729)[name = tensor("transpose_23")]; - tensor value_states_31 = reshape(shape = var_745, x = transpose_23)[name = tensor("value_states_31")]; - tensor var_748_perm_0 = const()[name = tensor("op_748_perm_0"), val = tensor([0, 2, 1])]; + tensor transpose_24 = transpose(perm = var_725_perm_0, x = var_724_cast)[name = tensor("transpose_24")]; + tensor key_states_31_cast = reshape(shape = var_745, x = transpose_24)[name = tensor("key_states_31_cast")]; + tensor var_747 = const()[name = tensor("op_747"), val = tensor([12, -1, 64])]; + tensor transpose_23 = transpose(perm = var_732_perm_0, x = var_731_cast)[name = tensor("transpose_23")]; + tensor value_states_31_cast = reshape(shape = var_747, x = transpose_23)[name = tensor("value_states_31_cast")]; + tensor var_750_perm_0 = const()[name = tensor("op_750_perm_0"), val = tensor([0, 2, 1])]; tensor attn_weights_43_transpose_x_0 = const()[name = tensor("attn_weights_43_transpose_x_0"), val = tensor(false)]; tensor attn_weights_43_transpose_y_0 = const()[name = tensor("attn_weights_43_transpose_y_0"), val = tensor(false)]; - tensor transpose_21 = transpose(perm = var_748_perm_0, x = key_states_31)[name = tensor("transpose_21")]; - tensor attn_weights_43 = matmul(transpose_x = attn_weights_43_transpose_x_0, transpose_y = attn_weights_43_transpose_y_0, x = query_states_15, y = transpose_21)[name = tensor("attn_weights_43")]; - tensor var_750 = const()[name = tensor("op_750"), val = tensor([1, 12, 77, 77])]; - tensor var_751 = reshape(shape = var_750, x = attn_weights_43)[name = tensor("op_751")]; - tensor attn_weights_45 = add(x = var_751, y = causal_attention_mask)[name = tensor("attn_weights_45")]; - tensor var_756 = const()[name = tensor("op_756"), val = tensor([12, 77, 77])]; - tensor input_117 = reshape(shape = var_756, x = attn_weights_45)[name = tensor("input_117")]; - tensor input_119 = softmax(axis = var_5, x = input_117)[name = tensor("input_119")]; + tensor transpose_21 = transpose(perm = var_750_perm_0, x = key_states_31_cast)[name = tensor("transpose_21")]; + tensor attn_weights_43_cast = matmul(transpose_x = attn_weights_43_transpose_x_0, transpose_y = attn_weights_43_transpose_y_0, x = query_states_15_cast, y = transpose_21)[name = tensor("attn_weights_43_cast")]; + tensor var_752 = const()[name = tensor("op_752"), val = tensor([1, 12, 77, 77])]; + tensor var_753_cast = reshape(shape = var_752, x = attn_weights_43_cast)[name = tensor("op_753_cast")]; + tensor attn_weights_45_cast = add(x = var_753_cast, y = causal_attention_mask_to_fp16)[name = tensor("attn_weights_45_cast")]; + tensor var_758 = const()[name = tensor("op_758"), val = tensor([12, 77, 77])]; + tensor input_117_cast = reshape(shape = var_758, x = attn_weights_45_cast)[name = tensor("input_117_cast")]; + tensor input_119_cast = softmax(axis = var_5, x = input_117_cast)[name = tensor("input_119_cast")]; tensor attn_output_43_transpose_x_0 = const()[name = tensor("attn_output_43_transpose_x_0"), val = tensor(false)]; tensor attn_output_43_transpose_y_0 = const()[name = tensor("attn_output_43_transpose_y_0"), val = tensor(false)]; - tensor attn_output_43 = matmul(transpose_x = attn_output_43_transpose_x_0, transpose_y = attn_output_43_transpose_y_0, x = input_119, y = value_states_31)[name = tensor("attn_output_43")]; - tensor var_761 = const()[name = tensor("op_761"), val = tensor([1, 12, 77, 64])]; - tensor attn_output_45 = reshape(shape = var_761, x = attn_output_43)[name = tensor("attn_output_45")]; + tensor attn_output_43_cast = matmul(transpose_x = attn_output_43_transpose_x_0, transpose_y = attn_output_43_transpose_y_0, x = input_119_cast, y = value_states_31_cast)[name = tensor("attn_output_43_cast")]; + tensor var_763 = const()[name = tensor("op_763"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_45_cast = reshape(shape = var_763, x = attn_output_43_cast)[name = tensor("attn_output_45_cast")]; tensor attn_output_47_perm_0 = const()[name = tensor("attn_output_47_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_764 = const()[name = tensor("op_764"), val = tensor([1, 77, 768])]; - tensor transpose_20 = transpose(perm = attn_output_47_perm_0, x = attn_output_45)[name = tensor("transpose_20")]; - tensor input_121 = reshape(shape = var_764, x = transpose_20)[name = tensor("input_121")]; - tensor hidden_states_45 = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_out_proj_bias, weight = text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight, x = input_121)[name = tensor("hidden_states_45")]; - tensor input_123 = add(x = input_115, y = hidden_states_45)[name = tensor("input_123")]; + tensor var_766 = const()[name = tensor("op_766"), val = tensor([1, 77, 768])]; + tensor transpose_20 = transpose(perm = attn_output_47_perm_0, x = attn_output_45_cast)[name = tensor("transpose_20")]; + tensor input_121_cast = reshape(shape = var_766, x = transpose_20)[name = tensor("input_121_cast")]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178805632)))]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(179985344)))]; + tensor hidden_states_45_cast = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16, x = input_121_cast)[name = tensor("hidden_states_45_cast")]; + tensor input_123_cast = add(x = input_115_cast, y = hidden_states_45_cast)[name = tensor("input_123_cast")]; tensor input_125_axes_0 = const()[name = tensor("input_125_axes_0"), val = tensor([-1])]; - tensor input_125 = layer_norm(axes = input_125_axes_0, beta = text_encoder_text_model_encoder_layers_7_layer_norm2_bias, epsilon = var_12, gamma = text_encoder_text_model_encoder_layers_7_layer_norm2_weight, x = input_123)[name = tensor("input_125")]; - tensor input_127 = linear(bias = text_encoder_text_model_encoder_layers_7_mlp_fc1_bias, weight = text_encoder_text_model_encoder_layers_7_mlp_fc1_weight, x = input_125)[name = tensor("input_127")]; - tensor var_779 = const()[name = tensor("op_779"), val = tensor(0x1.b3b646p+0)]; - tensor var_780 = mul(x = input_127, y = var_779)[name = tensor("op_780")]; - tensor var_781 = sigmoid(x = var_780)[name = tensor("op_781")]; - tensor input_129 = mul(x = input_127, y = var_781)[name = tensor("input_129")]; - tensor hidden_states_47 = linear(bias = text_encoder_text_model_encoder_layers_7_mlp_fc2_bias, weight = text_encoder_text_model_encoder_layers_7_mlp_fc2_weight, x = input_129)[name = tensor("hidden_states_47")]; - tensor input_131 = add(x = input_123, y = hidden_states_47)[name = tensor("input_131")]; + tensor text_encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(179986944)))]; + tensor text_encoder_text_model_encoder_layers_7_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(179988544)))]; + tensor input_125_cast = layer_norm(axes = input_125_axes_0, beta = text_encoder_text_model_encoder_layers_7_layer_norm2_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16, x = input_123_cast)[name = tensor("input_125_cast")]; + tensor text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(179990144)))]; + tensor text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184708800)))]; + tensor input_127_cast = linear(bias = text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16, x = input_125_cast)[name = tensor("input_127_cast")]; + tensor var_781_to_fp16 = const()[name = tensor("op_781_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_782_cast = mul(x = input_127_cast, y = var_781_to_fp16)[name = tensor("op_782_cast")]; + tensor var_783_cast = sigmoid(x = var_782_cast)[name = tensor("op_783_cast")]; + tensor input_129_cast = mul(x = input_127_cast, y = var_783_cast)[name = tensor("input_129_cast")]; + tensor text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184715008)))]; + tensor text_encoder_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(189433664)))]; + tensor hidden_states_47_cast = linear(bias = text_encoder_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16, x = input_129_cast)[name = tensor("hidden_states_47_cast")]; + tensor input_131_cast = add(x = input_123_cast, y = hidden_states_47_cast)[name = tensor("input_131_cast")]; tensor hidden_states_49_axes_0 = const()[name = tensor("hidden_states_49_axes_0"), val = tensor([-1])]; - tensor hidden_states_49 = layer_norm(axes = hidden_states_49_axes_0, beta = text_encoder_text_model_encoder_layers_8_layer_norm1_bias, epsilon = var_12, gamma = text_encoder_text_model_encoder_layers_8_layer_norm1_weight, x = input_131)[name = tensor("hidden_states_49")]; - tensor var_805 = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_q_proj_bias, weight = text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight, x = hidden_states_49)[name = tensor("op_805")]; - tensor var_806 = const()[name = tensor("op_806"), val = tensor(0x1p-3)]; - tensor tensor_53 = mul(x = var_805, y = var_806)[name = tensor("tensor_53")]; - tensor tensor_49 = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_k_proj_bias, weight = text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight, x = hidden_states_49)[name = tensor("tensor_49")]; - tensor var_811 = const()[name = tensor("op_811"), val = tensor([1, -1, 12, 64])]; - tensor var_812 = reshape(shape = var_811, x = tensor_49)[name = tensor("op_812")]; - tensor var_813_perm_0 = const()[name = tensor("op_813_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor tensor_51 = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_v_proj_bias, weight = text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight, x = hidden_states_49)[name = tensor("tensor_51")]; - tensor var_818 = const()[name = tensor("op_818"), val = tensor([1, -1, 12, 64])]; - tensor var_819 = reshape(shape = var_818, x = tensor_51)[name = tensor("op_819")]; - tensor var_820_perm_0 = const()[name = tensor("op_820_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_827 = const()[name = tensor("op_827"), val = tensor([1, 77, 12, 64])]; - tensor var_828 = reshape(shape = var_827, x = tensor_53)[name = tensor("op_828")]; - tensor var_829_perm_0 = const()[name = tensor("op_829_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_831 = const()[name = tensor("op_831"), val = tensor([12, -1, 64])]; - tensor transpose_17 = transpose(perm = var_829_perm_0, x = var_828)[name = tensor("transpose_17")]; - tensor query_states_17 = reshape(shape = var_831, x = transpose_17)[name = tensor("query_states_17")]; + tensor text_encoder_text_model_encoder_layers_8_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(189435264)))]; + tensor text_encoder_text_model_encoder_layers_8_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(189436864)))]; + tensor hidden_states_49_cast = layer_norm(axes = hidden_states_49_axes_0, beta = text_encoder_text_model_encoder_layers_8_layer_norm1_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_8_layer_norm1_weight_to_fp16, x = input_131_cast)[name = tensor("hidden_states_49_cast")]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(189438464)))]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190618176)))]; + tensor var_807_cast = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16, x = hidden_states_49_cast)[name = tensor("op_807_cast")]; + tensor var_808_to_fp16 = const()[name = tensor("op_808_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_53_cast = mul(x = var_807_cast, y = var_808_to_fp16)[name = tensor("tensor_53_cast")]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190619776)))]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(191799488)))]; + tensor tensor_49_cast = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16, x = hidden_states_49_cast)[name = tensor("tensor_49_cast")]; + tensor var_813 = const()[name = tensor("op_813"), val = tensor([1, -1, 12, 64])]; + tensor var_814_cast = reshape(shape = var_813, x = tensor_49_cast)[name = tensor("op_814_cast")]; + tensor var_815_perm_0 = const()[name = tensor("op_815_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(191801088)))]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(192980800)))]; + tensor tensor_51_cast = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16, x = hidden_states_49_cast)[name = tensor("tensor_51_cast")]; + tensor var_820 = const()[name = tensor("op_820"), val = tensor([1, -1, 12, 64])]; + tensor var_821_cast = reshape(shape = var_820, x = tensor_51_cast)[name = tensor("op_821_cast")]; + tensor var_822_perm_0 = const()[name = tensor("op_822_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_829 = const()[name = tensor("op_829"), val = tensor([1, 77, 12, 64])]; + tensor var_830_cast = reshape(shape = var_829, x = tensor_53_cast)[name = tensor("op_830_cast")]; + tensor var_831_perm_0 = const()[name = tensor("op_831_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_833 = const()[name = tensor("op_833"), val = tensor([12, -1, 64])]; - tensor transpose_19 = transpose(perm = var_813_perm_0, x = var_812)[name = tensor("transpose_19")]; - tensor key_states_35 = reshape(shape = var_833, x = transpose_19)[name = tensor("key_states_35")]; + tensor transpose_17 = transpose(perm = var_831_perm_0, x = var_830_cast)[name = tensor("transpose_17")]; + tensor query_states_17_cast = reshape(shape = var_833, x = transpose_17)[name = tensor("query_states_17_cast")]; tensor var_835 = const()[name = tensor("op_835"), val = tensor([12, -1, 64])]; - tensor transpose_18 = transpose(perm = var_820_perm_0, x = var_819)[name = tensor("transpose_18")]; - tensor value_states_35 = reshape(shape = var_835, x = transpose_18)[name = tensor("value_states_35")]; - tensor var_838_perm_0 = const()[name = tensor("op_838_perm_0"), val = tensor([0, 2, 1])]; + tensor transpose_19 = transpose(perm = var_815_perm_0, x = var_814_cast)[name = tensor("transpose_19")]; + tensor key_states_35_cast = reshape(shape = var_835, x = transpose_19)[name = tensor("key_states_35_cast")]; + tensor var_837 = const()[name = tensor("op_837"), val = tensor([12, -1, 64])]; + tensor transpose_18 = transpose(perm = var_822_perm_0, x = var_821_cast)[name = tensor("transpose_18")]; + tensor value_states_35_cast = reshape(shape = var_837, x = transpose_18)[name = tensor("value_states_35_cast")]; + tensor var_840_perm_0 = const()[name = tensor("op_840_perm_0"), val = tensor([0, 2, 1])]; tensor attn_weights_49_transpose_x_0 = const()[name = tensor("attn_weights_49_transpose_x_0"), val = tensor(false)]; tensor attn_weights_49_transpose_y_0 = const()[name = tensor("attn_weights_49_transpose_y_0"), val = tensor(false)]; - tensor transpose_16 = transpose(perm = var_838_perm_0, x = key_states_35)[name = tensor("transpose_16")]; - tensor attn_weights_49 = matmul(transpose_x = attn_weights_49_transpose_x_0, transpose_y = attn_weights_49_transpose_y_0, x = query_states_17, y = transpose_16)[name = tensor("attn_weights_49")]; - tensor var_840 = const()[name = tensor("op_840"), val = tensor([1, 12, 77, 77])]; - tensor var_841 = reshape(shape = var_840, x = attn_weights_49)[name = tensor("op_841")]; - tensor attn_weights_51 = add(x = var_841, y = causal_attention_mask)[name = tensor("attn_weights_51")]; - tensor var_846 = const()[name = tensor("op_846"), val = tensor([12, 77, 77])]; - tensor input_133 = reshape(shape = var_846, x = attn_weights_51)[name = tensor("input_133")]; - tensor input_135 = softmax(axis = var_5, x = input_133)[name = tensor("input_135")]; + tensor transpose_16 = transpose(perm = var_840_perm_0, x = key_states_35_cast)[name = tensor("transpose_16")]; + tensor attn_weights_49_cast = matmul(transpose_x = attn_weights_49_transpose_x_0, transpose_y = attn_weights_49_transpose_y_0, x = query_states_17_cast, y = transpose_16)[name = tensor("attn_weights_49_cast")]; + tensor var_842 = const()[name = tensor("op_842"), val = tensor([1, 12, 77, 77])]; + tensor var_843_cast = reshape(shape = var_842, x = attn_weights_49_cast)[name = tensor("op_843_cast")]; + tensor attn_weights_51_cast = add(x = var_843_cast, y = causal_attention_mask_to_fp16)[name = tensor("attn_weights_51_cast")]; + tensor var_848 = const()[name = tensor("op_848"), val = tensor([12, 77, 77])]; + tensor input_133_cast = reshape(shape = var_848, x = attn_weights_51_cast)[name = tensor("input_133_cast")]; + tensor input_135_cast = softmax(axis = var_5, x = input_133_cast)[name = tensor("input_135_cast")]; tensor attn_output_49_transpose_x_0 = const()[name = tensor("attn_output_49_transpose_x_0"), val = tensor(false)]; tensor attn_output_49_transpose_y_0 = const()[name = tensor("attn_output_49_transpose_y_0"), val = tensor(false)]; - tensor attn_output_49 = matmul(transpose_x = attn_output_49_transpose_x_0, transpose_y = attn_output_49_transpose_y_0, x = input_135, y = value_states_35)[name = tensor("attn_output_49")]; - tensor var_851 = const()[name = tensor("op_851"), val = tensor([1, 12, 77, 64])]; - tensor attn_output_51 = reshape(shape = var_851, x = attn_output_49)[name = tensor("attn_output_51")]; + tensor attn_output_49_cast = matmul(transpose_x = attn_output_49_transpose_x_0, transpose_y = attn_output_49_transpose_y_0, x = input_135_cast, y = value_states_35_cast)[name = tensor("attn_output_49_cast")]; + tensor var_853 = const()[name = tensor("op_853"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_51_cast = reshape(shape = var_853, x = attn_output_49_cast)[name = tensor("attn_output_51_cast")]; tensor attn_output_53_perm_0 = const()[name = tensor("attn_output_53_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_854 = const()[name = tensor("op_854"), val = tensor([1, 77, 768])]; - tensor transpose_15 = transpose(perm = attn_output_53_perm_0, x = attn_output_51)[name = tensor("transpose_15")]; - tensor input_137 = reshape(shape = var_854, x = transpose_15)[name = tensor("input_137")]; - tensor hidden_states_51 = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_out_proj_bias, weight = text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight, x = input_137)[name = tensor("hidden_states_51")]; - tensor input_139 = add(x = input_131, y = hidden_states_51)[name = tensor("input_139")]; + tensor var_856 = const()[name = tensor("op_856"), val = tensor([1, 77, 768])]; + tensor transpose_15 = transpose(perm = attn_output_53_perm_0, x = attn_output_51_cast)[name = tensor("transpose_15")]; + tensor input_137_cast = reshape(shape = var_856, x = transpose_15)[name = tensor("input_137_cast")]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(192982400)))]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194162112)))]; + tensor hidden_states_51_cast = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16, x = input_137_cast)[name = tensor("hidden_states_51_cast")]; + tensor input_139_cast = add(x = input_131_cast, y = hidden_states_51_cast)[name = tensor("input_139_cast")]; tensor input_141_axes_0 = const()[name = tensor("input_141_axes_0"), val = tensor([-1])]; - tensor input_141 = layer_norm(axes = input_141_axes_0, beta = text_encoder_text_model_encoder_layers_8_layer_norm2_bias, epsilon = var_12, gamma = text_encoder_text_model_encoder_layers_8_layer_norm2_weight, x = input_139)[name = tensor("input_141")]; - tensor input_143 = linear(bias = text_encoder_text_model_encoder_layers_8_mlp_fc1_bias, weight = text_encoder_text_model_encoder_layers_8_mlp_fc1_weight, x = input_141)[name = tensor("input_143")]; - tensor var_869 = const()[name = tensor("op_869"), val = tensor(0x1.b3b646p+0)]; - tensor var_870 = mul(x = input_143, y = var_869)[name = tensor("op_870")]; - tensor var_871 = sigmoid(x = var_870)[name = tensor("op_871")]; - tensor input_145 = mul(x = input_143, y = var_871)[name = tensor("input_145")]; - tensor hidden_states_53 = linear(bias = text_encoder_text_model_encoder_layers_8_mlp_fc2_bias, weight = text_encoder_text_model_encoder_layers_8_mlp_fc2_weight, x = input_145)[name = tensor("hidden_states_53")]; - tensor input_147 = add(x = input_139, y = hidden_states_53)[name = tensor("input_147")]; + tensor text_encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194163712)))]; + tensor text_encoder_text_model_encoder_layers_8_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194165312)))]; + tensor input_141_cast = layer_norm(axes = input_141_axes_0, beta = text_encoder_text_model_encoder_layers_8_layer_norm2_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16, x = input_139_cast)[name = tensor("input_141_cast")]; + tensor text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194166912)))]; + tensor text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(198885568)))]; + tensor input_143_cast = linear(bias = text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16, x = input_141_cast)[name = tensor("input_143_cast")]; + tensor var_871_to_fp16 = const()[name = tensor("op_871_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_872_cast = mul(x = input_143_cast, y = var_871_to_fp16)[name = tensor("op_872_cast")]; + tensor var_873_cast = sigmoid(x = var_872_cast)[name = tensor("op_873_cast")]; + tensor input_145_cast = mul(x = input_143_cast, y = var_873_cast)[name = tensor("input_145_cast")]; + tensor text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(198891776)))]; + tensor text_encoder_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(203610432)))]; + tensor hidden_states_53_cast = linear(bias = text_encoder_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16, x = input_145_cast)[name = tensor("hidden_states_53_cast")]; + tensor input_147_cast = add(x = input_139_cast, y = hidden_states_53_cast)[name = tensor("input_147_cast")]; tensor hidden_states_55_axes_0 = const()[name = tensor("hidden_states_55_axes_0"), val = tensor([-1])]; - tensor hidden_states_55 = layer_norm(axes = hidden_states_55_axes_0, beta = text_encoder_text_model_encoder_layers_9_layer_norm1_bias, epsilon = var_12, gamma = text_encoder_text_model_encoder_layers_9_layer_norm1_weight, x = input_147)[name = tensor("hidden_states_55")]; - tensor var_895 = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_q_proj_bias, weight = text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight, x = hidden_states_55)[name = tensor("op_895")]; - tensor var_896 = const()[name = tensor("op_896"), val = tensor(0x1p-3)]; - tensor tensor_59 = mul(x = var_895, y = var_896)[name = tensor("tensor_59")]; - tensor tensor_55 = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_k_proj_bias, weight = text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight, x = hidden_states_55)[name = tensor("tensor_55")]; - tensor var_901 = const()[name = tensor("op_901"), val = tensor([1, -1, 12, 64])]; - tensor var_902 = reshape(shape = var_901, x = tensor_55)[name = tensor("op_902")]; - tensor var_903_perm_0 = const()[name = tensor("op_903_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor tensor_57 = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_v_proj_bias, weight = text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight, x = hidden_states_55)[name = tensor("tensor_57")]; - tensor var_908 = const()[name = tensor("op_908"), val = tensor([1, -1, 12, 64])]; - tensor var_909 = reshape(shape = var_908, x = tensor_57)[name = tensor("op_909")]; - tensor var_910_perm_0 = const()[name = tensor("op_910_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_917 = const()[name = tensor("op_917"), val = tensor([1, 77, 12, 64])]; - tensor var_918 = reshape(shape = var_917, x = tensor_59)[name = tensor("op_918")]; - tensor var_919_perm_0 = const()[name = tensor("op_919_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_921 = const()[name = tensor("op_921"), val = tensor([12, -1, 64])]; - tensor transpose_12 = transpose(perm = var_919_perm_0, x = var_918)[name = tensor("transpose_12")]; - tensor query_states_19 = reshape(shape = var_921, x = transpose_12)[name = tensor("query_states_19")]; + tensor text_encoder_text_model_encoder_layers_9_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(203612032)))]; + tensor text_encoder_text_model_encoder_layers_9_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(203613632)))]; + tensor hidden_states_55_cast = layer_norm(axes = hidden_states_55_axes_0, beta = text_encoder_text_model_encoder_layers_9_layer_norm1_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_9_layer_norm1_weight_to_fp16, x = input_147_cast)[name = tensor("hidden_states_55_cast")]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(203615232)))]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(204794944)))]; + tensor var_897_cast = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16, x = hidden_states_55_cast)[name = tensor("op_897_cast")]; + tensor var_898_to_fp16 = const()[name = tensor("op_898_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_59_cast = mul(x = var_897_cast, y = var_898_to_fp16)[name = tensor("tensor_59_cast")]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(204796544)))]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(205976256)))]; + tensor tensor_55_cast = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16, x = hidden_states_55_cast)[name = tensor("tensor_55_cast")]; + tensor var_903 = const()[name = tensor("op_903"), val = tensor([1, -1, 12, 64])]; + tensor var_904_cast = reshape(shape = var_903, x = tensor_55_cast)[name = tensor("op_904_cast")]; + tensor var_905_perm_0 = const()[name = tensor("op_905_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(205977856)))]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(207157568)))]; + tensor tensor_57_cast = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16, x = hidden_states_55_cast)[name = tensor("tensor_57_cast")]; + tensor var_910 = const()[name = tensor("op_910"), val = tensor([1, -1, 12, 64])]; + tensor var_911_cast = reshape(shape = var_910, x = tensor_57_cast)[name = tensor("op_911_cast")]; + tensor var_912_perm_0 = const()[name = tensor("op_912_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_919 = const()[name = tensor("op_919"), val = tensor([1, 77, 12, 64])]; + tensor var_920_cast = reshape(shape = var_919, x = tensor_59_cast)[name = tensor("op_920_cast")]; + tensor var_921_perm_0 = const()[name = tensor("op_921_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_923 = const()[name = tensor("op_923"), val = tensor([12, -1, 64])]; - tensor transpose_14 = transpose(perm = var_903_perm_0, x = var_902)[name = tensor("transpose_14")]; - tensor key_states_39 = reshape(shape = var_923, x = transpose_14)[name = tensor("key_states_39")]; + tensor transpose_12 = transpose(perm = var_921_perm_0, x = var_920_cast)[name = tensor("transpose_12")]; + tensor query_states_19_cast = reshape(shape = var_923, x = transpose_12)[name = tensor("query_states_19_cast")]; tensor var_925 = const()[name = tensor("op_925"), val = tensor([12, -1, 64])]; - tensor transpose_13 = transpose(perm = var_910_perm_0, x = var_909)[name = tensor("transpose_13")]; - tensor value_states_39 = reshape(shape = var_925, x = transpose_13)[name = tensor("value_states_39")]; - tensor var_928_perm_0 = const()[name = tensor("op_928_perm_0"), val = tensor([0, 2, 1])]; + tensor transpose_14 = transpose(perm = var_905_perm_0, x = var_904_cast)[name = tensor("transpose_14")]; + tensor key_states_39_cast = reshape(shape = var_925, x = transpose_14)[name = tensor("key_states_39_cast")]; + tensor var_927 = const()[name = tensor("op_927"), val = tensor([12, -1, 64])]; + tensor transpose_13 = transpose(perm = var_912_perm_0, x = var_911_cast)[name = tensor("transpose_13")]; + tensor value_states_39_cast = reshape(shape = var_927, x = transpose_13)[name = tensor("value_states_39_cast")]; + tensor var_930_perm_0 = const()[name = tensor("op_930_perm_0"), val = tensor([0, 2, 1])]; tensor attn_weights_55_transpose_x_0 = const()[name = tensor("attn_weights_55_transpose_x_0"), val = tensor(false)]; tensor attn_weights_55_transpose_y_0 = const()[name = tensor("attn_weights_55_transpose_y_0"), val = tensor(false)]; - tensor transpose_11 = transpose(perm = var_928_perm_0, x = key_states_39)[name = tensor("transpose_11")]; - tensor attn_weights_55 = matmul(transpose_x = attn_weights_55_transpose_x_0, transpose_y = attn_weights_55_transpose_y_0, x = query_states_19, y = transpose_11)[name = tensor("attn_weights_55")]; - tensor var_930 = const()[name = tensor("op_930"), val = tensor([1, 12, 77, 77])]; - tensor var_931 = reshape(shape = var_930, x = attn_weights_55)[name = tensor("op_931")]; - tensor attn_weights_57 = add(x = var_931, y = causal_attention_mask)[name = tensor("attn_weights_57")]; - tensor var_936 = const()[name = tensor("op_936"), val = tensor([12, 77, 77])]; - tensor input_149 = reshape(shape = var_936, x = attn_weights_57)[name = tensor("input_149")]; - tensor input_151 = softmax(axis = var_5, x = input_149)[name = tensor("input_151")]; + tensor transpose_11 = transpose(perm = var_930_perm_0, x = key_states_39_cast)[name = tensor("transpose_11")]; + tensor attn_weights_55_cast = matmul(transpose_x = attn_weights_55_transpose_x_0, transpose_y = attn_weights_55_transpose_y_0, x = query_states_19_cast, y = transpose_11)[name = tensor("attn_weights_55_cast")]; + tensor var_932 = const()[name = tensor("op_932"), val = tensor([1, 12, 77, 77])]; + tensor var_933_cast = reshape(shape = var_932, x = attn_weights_55_cast)[name = tensor("op_933_cast")]; + tensor attn_weights_57_cast = add(x = var_933_cast, y = causal_attention_mask_to_fp16)[name = tensor("attn_weights_57_cast")]; + tensor var_938 = const()[name = tensor("op_938"), val = tensor([12, 77, 77])]; + tensor input_149_cast = reshape(shape = var_938, x = attn_weights_57_cast)[name = tensor("input_149_cast")]; + tensor input_151_cast = softmax(axis = var_5, x = input_149_cast)[name = tensor("input_151_cast")]; tensor attn_output_55_transpose_x_0 = const()[name = tensor("attn_output_55_transpose_x_0"), val = tensor(false)]; tensor attn_output_55_transpose_y_0 = const()[name = tensor("attn_output_55_transpose_y_0"), val = tensor(false)]; - tensor attn_output_55 = matmul(transpose_x = attn_output_55_transpose_x_0, transpose_y = attn_output_55_transpose_y_0, x = input_151, y = value_states_39)[name = tensor("attn_output_55")]; - tensor var_941 = const()[name = tensor("op_941"), val = tensor([1, 12, 77, 64])]; - tensor attn_output_57 = reshape(shape = var_941, x = attn_output_55)[name = tensor("attn_output_57")]; + tensor attn_output_55_cast = matmul(transpose_x = attn_output_55_transpose_x_0, transpose_y = attn_output_55_transpose_y_0, x = input_151_cast, y = value_states_39_cast)[name = tensor("attn_output_55_cast")]; + tensor var_943 = const()[name = tensor("op_943"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_57_cast = reshape(shape = var_943, x = attn_output_55_cast)[name = tensor("attn_output_57_cast")]; tensor attn_output_59_perm_0 = const()[name = tensor("attn_output_59_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_944 = const()[name = tensor("op_944"), val = tensor([1, 77, 768])]; - tensor transpose_10 = transpose(perm = attn_output_59_perm_0, x = attn_output_57)[name = tensor("transpose_10")]; - tensor input_153 = reshape(shape = var_944, x = transpose_10)[name = tensor("input_153")]; - tensor hidden_states_57 = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_out_proj_bias, weight = text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight, x = input_153)[name = tensor("hidden_states_57")]; - tensor input_155 = add(x = input_147, y = hidden_states_57)[name = tensor("input_155")]; + tensor var_946 = const()[name = tensor("op_946"), val = tensor([1, 77, 768])]; + tensor transpose_10 = transpose(perm = attn_output_59_perm_0, x = attn_output_57_cast)[name = tensor("transpose_10")]; + tensor input_153_cast = reshape(shape = var_946, x = transpose_10)[name = tensor("input_153_cast")]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(207159168)))]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(208338880)))]; + tensor hidden_states_57_cast = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16, x = input_153_cast)[name = tensor("hidden_states_57_cast")]; + tensor input_155_cast = add(x = input_147_cast, y = hidden_states_57_cast)[name = tensor("input_155_cast")]; tensor input_157_axes_0 = const()[name = tensor("input_157_axes_0"), val = tensor([-1])]; - tensor input_157 = layer_norm(axes = input_157_axes_0, beta = text_encoder_text_model_encoder_layers_9_layer_norm2_bias, epsilon = var_12, gamma = text_encoder_text_model_encoder_layers_9_layer_norm2_weight, x = input_155)[name = tensor("input_157")]; - tensor input_159 = linear(bias = text_encoder_text_model_encoder_layers_9_mlp_fc1_bias, weight = text_encoder_text_model_encoder_layers_9_mlp_fc1_weight, x = input_157)[name = tensor("input_159")]; - tensor var_959 = const()[name = tensor("op_959"), val = tensor(0x1.b3b646p+0)]; - tensor var_960 = mul(x = input_159, y = var_959)[name = tensor("op_960")]; - tensor var_961 = sigmoid(x = var_960)[name = tensor("op_961")]; - tensor input_161 = mul(x = input_159, y = var_961)[name = tensor("input_161")]; - tensor hidden_states_59 = linear(bias = text_encoder_text_model_encoder_layers_9_mlp_fc2_bias, weight = text_encoder_text_model_encoder_layers_9_mlp_fc2_weight, x = input_161)[name = tensor("hidden_states_59")]; - tensor input_163 = add(x = input_155, y = hidden_states_59)[name = tensor("input_163")]; + tensor text_encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(208340480)))]; + tensor text_encoder_text_model_encoder_layers_9_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(208342080)))]; + tensor input_157_cast = layer_norm(axes = input_157_axes_0, beta = text_encoder_text_model_encoder_layers_9_layer_norm2_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16, x = input_155_cast)[name = tensor("input_157_cast")]; + tensor text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(208343680)))]; + tensor text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(213062336)))]; + tensor input_159_cast = linear(bias = text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16, x = input_157_cast)[name = tensor("input_159_cast")]; + tensor var_961_to_fp16 = const()[name = tensor("op_961_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_962_cast = mul(x = input_159_cast, y = var_961_to_fp16)[name = tensor("op_962_cast")]; + tensor var_963_cast = sigmoid(x = var_962_cast)[name = tensor("op_963_cast")]; + tensor input_161_cast = mul(x = input_159_cast, y = var_963_cast)[name = tensor("input_161_cast")]; + tensor text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(213068544)))]; + tensor text_encoder_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217787200)))]; + tensor hidden_states_59_cast = linear(bias = text_encoder_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16, x = input_161_cast)[name = tensor("hidden_states_59_cast")]; + tensor input_163_cast = add(x = input_155_cast, y = hidden_states_59_cast)[name = tensor("input_163_cast")]; tensor hidden_states_61_axes_0 = const()[name = tensor("hidden_states_61_axes_0"), val = tensor([-1])]; - tensor hidden_states_61 = layer_norm(axes = hidden_states_61_axes_0, beta = text_encoder_text_model_encoder_layers_10_layer_norm1_bias, epsilon = var_12, gamma = text_encoder_text_model_encoder_layers_10_layer_norm1_weight, x = input_163)[name = tensor("hidden_states_61")]; - tensor var_985 = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_q_proj_bias, weight = text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight, x = hidden_states_61)[name = tensor("op_985")]; - tensor var_986 = const()[name = tensor("op_986"), val = tensor(0x1p-3)]; - tensor tensor_65 = mul(x = var_985, y = var_986)[name = tensor("tensor_65")]; - tensor tensor_61 = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_k_proj_bias, weight = text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight, x = hidden_states_61)[name = tensor("tensor_61")]; - tensor var_991 = const()[name = tensor("op_991"), val = tensor([1, -1, 12, 64])]; - tensor var_992 = reshape(shape = var_991, x = tensor_61)[name = tensor("op_992")]; - tensor var_993_perm_0 = const()[name = tensor("op_993_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor tensor_63 = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_v_proj_bias, weight = text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight, x = hidden_states_61)[name = tensor("tensor_63")]; - tensor var_998 = const()[name = tensor("op_998"), val = tensor([1, -1, 12, 64])]; - tensor var_999 = reshape(shape = var_998, x = tensor_63)[name = tensor("op_999")]; - tensor var_1000_perm_0 = const()[name = tensor("op_1000_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_1007 = const()[name = tensor("op_1007"), val = tensor([1, 77, 12, 64])]; - tensor var_1008 = reshape(shape = var_1007, x = tensor_65)[name = tensor("op_1008")]; - tensor var_1009_perm_0 = const()[name = tensor("op_1009_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_1011 = const()[name = tensor("op_1011"), val = tensor([12, -1, 64])]; - tensor transpose_7 = transpose(perm = var_1009_perm_0, x = var_1008)[name = tensor("transpose_7")]; - tensor query_states_21 = reshape(shape = var_1011, x = transpose_7)[name = tensor("query_states_21")]; + tensor text_encoder_text_model_encoder_layers_10_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217788800)))]; + tensor text_encoder_text_model_encoder_layers_10_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217790400)))]; + tensor hidden_states_61_cast = layer_norm(axes = hidden_states_61_axes_0, beta = text_encoder_text_model_encoder_layers_10_layer_norm1_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_10_layer_norm1_weight_to_fp16, x = input_163_cast)[name = tensor("hidden_states_61_cast")]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217792000)))]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(218971712)))]; + tensor var_987_cast = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16, x = hidden_states_61_cast)[name = tensor("op_987_cast")]; + tensor var_988_to_fp16 = const()[name = tensor("op_988_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_65_cast = mul(x = var_987_cast, y = var_988_to_fp16)[name = tensor("tensor_65_cast")]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(218973312)))]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(220153024)))]; + tensor tensor_61_cast = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16, x = hidden_states_61_cast)[name = tensor("tensor_61_cast")]; + tensor var_993 = const()[name = tensor("op_993"), val = tensor([1, -1, 12, 64])]; + tensor var_994_cast = reshape(shape = var_993, x = tensor_61_cast)[name = tensor("op_994_cast")]; + tensor var_995_perm_0 = const()[name = tensor("op_995_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(220154624)))]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(221334336)))]; + tensor tensor_63_cast = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16, x = hidden_states_61_cast)[name = tensor("tensor_63_cast")]; + tensor var_1000 = const()[name = tensor("op_1000"), val = tensor([1, -1, 12, 64])]; + tensor var_1001_cast = reshape(shape = var_1000, x = tensor_63_cast)[name = tensor("op_1001_cast")]; + tensor var_1002_perm_0 = const()[name = tensor("op_1002_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_1009 = const()[name = tensor("op_1009"), val = tensor([1, 77, 12, 64])]; + tensor var_1010_cast = reshape(shape = var_1009, x = tensor_65_cast)[name = tensor("op_1010_cast")]; + tensor var_1011_perm_0 = const()[name = tensor("op_1011_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_1013 = const()[name = tensor("op_1013"), val = tensor([12, -1, 64])]; - tensor transpose_9 = transpose(perm = var_993_perm_0, x = var_992)[name = tensor("transpose_9")]; - tensor key_states_43 = reshape(shape = var_1013, x = transpose_9)[name = tensor("key_states_43")]; + tensor transpose_7 = transpose(perm = var_1011_perm_0, x = var_1010_cast)[name = tensor("transpose_7")]; + tensor query_states_21_cast = reshape(shape = var_1013, x = transpose_7)[name = tensor("query_states_21_cast")]; tensor var_1015 = const()[name = tensor("op_1015"), val = tensor([12, -1, 64])]; - tensor transpose_8 = transpose(perm = var_1000_perm_0, x = var_999)[name = tensor("transpose_8")]; - tensor value_states_43 = reshape(shape = var_1015, x = transpose_8)[name = tensor("value_states_43")]; - tensor var_1018_perm_0 = const()[name = tensor("op_1018_perm_0"), val = tensor([0, 2, 1])]; + tensor transpose_9 = transpose(perm = var_995_perm_0, x = var_994_cast)[name = tensor("transpose_9")]; + tensor key_states_43_cast = reshape(shape = var_1015, x = transpose_9)[name = tensor("key_states_43_cast")]; + tensor var_1017 = const()[name = tensor("op_1017"), val = tensor([12, -1, 64])]; + tensor transpose_8 = transpose(perm = var_1002_perm_0, x = var_1001_cast)[name = tensor("transpose_8")]; + tensor value_states_43_cast = reshape(shape = var_1017, x = transpose_8)[name = tensor("value_states_43_cast")]; + tensor var_1020_perm_0 = const()[name = tensor("op_1020_perm_0"), val = tensor([0, 2, 1])]; tensor attn_weights_61_transpose_x_0 = const()[name = tensor("attn_weights_61_transpose_x_0"), val = tensor(false)]; tensor attn_weights_61_transpose_y_0 = const()[name = tensor("attn_weights_61_transpose_y_0"), val = tensor(false)]; - tensor transpose_6 = transpose(perm = var_1018_perm_0, x = key_states_43)[name = tensor("transpose_6")]; - tensor attn_weights_61 = matmul(transpose_x = attn_weights_61_transpose_x_0, transpose_y = attn_weights_61_transpose_y_0, x = query_states_21, y = transpose_6)[name = tensor("attn_weights_61")]; - tensor var_1020 = const()[name = tensor("op_1020"), val = tensor([1, 12, 77, 77])]; - tensor var_1021 = reshape(shape = var_1020, x = attn_weights_61)[name = tensor("op_1021")]; - tensor attn_weights_63 = add(x = var_1021, y = causal_attention_mask)[name = tensor("attn_weights_63")]; - tensor var_1026 = const()[name = tensor("op_1026"), val = tensor([12, 77, 77])]; - tensor input_165 = reshape(shape = var_1026, x = attn_weights_63)[name = tensor("input_165")]; - tensor input_167 = softmax(axis = var_5, x = input_165)[name = tensor("input_167")]; + tensor transpose_6 = transpose(perm = var_1020_perm_0, x = key_states_43_cast)[name = tensor("transpose_6")]; + tensor attn_weights_61_cast = matmul(transpose_x = attn_weights_61_transpose_x_0, transpose_y = attn_weights_61_transpose_y_0, x = query_states_21_cast, y = transpose_6)[name = tensor("attn_weights_61_cast")]; + tensor var_1022 = const()[name = tensor("op_1022"), val = tensor([1, 12, 77, 77])]; + tensor var_1023_cast = reshape(shape = var_1022, x = attn_weights_61_cast)[name = tensor("op_1023_cast")]; + tensor attn_weights_63_cast = add(x = var_1023_cast, y = causal_attention_mask_to_fp16)[name = tensor("attn_weights_63_cast")]; + tensor var_1028 = const()[name = tensor("op_1028"), val = tensor([12, 77, 77])]; + tensor input_165_cast = reshape(shape = var_1028, x = attn_weights_63_cast)[name = tensor("input_165_cast")]; + tensor input_167_cast = softmax(axis = var_5, x = input_165_cast)[name = tensor("input_167_cast")]; tensor attn_output_61_transpose_x_0 = const()[name = tensor("attn_output_61_transpose_x_0"), val = tensor(false)]; tensor attn_output_61_transpose_y_0 = const()[name = tensor("attn_output_61_transpose_y_0"), val = tensor(false)]; - tensor attn_output_61 = matmul(transpose_x = attn_output_61_transpose_x_0, transpose_y = attn_output_61_transpose_y_0, x = input_167, y = value_states_43)[name = tensor("attn_output_61")]; - tensor var_1031 = const()[name = tensor("op_1031"), val = tensor([1, 12, 77, 64])]; - tensor attn_output_63 = reshape(shape = var_1031, x = attn_output_61)[name = tensor("attn_output_63")]; + tensor attn_output_61_cast = matmul(transpose_x = attn_output_61_transpose_x_0, transpose_y = attn_output_61_transpose_y_0, x = input_167_cast, y = value_states_43_cast)[name = tensor("attn_output_61_cast")]; + tensor var_1033 = const()[name = tensor("op_1033"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_63_cast = reshape(shape = var_1033, x = attn_output_61_cast)[name = tensor("attn_output_63_cast")]; tensor attn_output_65_perm_0 = const()[name = tensor("attn_output_65_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_1034 = const()[name = tensor("op_1034"), val = tensor([1, 77, 768])]; - tensor transpose_5 = transpose(perm = attn_output_65_perm_0, x = attn_output_63)[name = tensor("transpose_5")]; - tensor input_169 = reshape(shape = var_1034, x = transpose_5)[name = tensor("input_169")]; - tensor hidden_states_63 = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_out_proj_bias, weight = text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight, x = input_169)[name = tensor("hidden_states_63")]; - tensor input_171 = add(x = input_163, y = hidden_states_63)[name = tensor("input_171")]; + tensor var_1036 = const()[name = tensor("op_1036"), val = tensor([1, 77, 768])]; + tensor transpose_5 = transpose(perm = attn_output_65_perm_0, x = attn_output_63_cast)[name = tensor("transpose_5")]; + tensor input_169_cast = reshape(shape = var_1036, x = transpose_5)[name = tensor("input_169_cast")]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(221335936)))]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222515648)))]; + tensor hidden_states_63_cast = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16, x = input_169_cast)[name = tensor("hidden_states_63_cast")]; + tensor input_171_cast = add(x = input_163_cast, y = hidden_states_63_cast)[name = tensor("input_171_cast")]; tensor input_173_axes_0 = const()[name = tensor("input_173_axes_0"), val = tensor([-1])]; - tensor input_173 = layer_norm(axes = input_173_axes_0, beta = text_encoder_text_model_encoder_layers_10_layer_norm2_bias, epsilon = var_12, gamma = text_encoder_text_model_encoder_layers_10_layer_norm2_weight, x = input_171)[name = tensor("input_173")]; - tensor input_175 = linear(bias = text_encoder_text_model_encoder_layers_10_mlp_fc1_bias, weight = text_encoder_text_model_encoder_layers_10_mlp_fc1_weight, x = input_173)[name = tensor("input_175")]; - tensor var_1049 = const()[name = tensor("op_1049"), val = tensor(0x1.b3b646p+0)]; - tensor var_1050 = mul(x = input_175, y = var_1049)[name = tensor("op_1050")]; - tensor var_1051 = sigmoid(x = var_1050)[name = tensor("op_1051")]; - tensor input_177 = mul(x = input_175, y = var_1051)[name = tensor("input_177")]; - tensor hidden_states_65 = linear(bias = text_encoder_text_model_encoder_layers_10_mlp_fc2_bias, weight = text_encoder_text_model_encoder_layers_10_mlp_fc2_weight, x = input_177)[name = tensor("hidden_states_65")]; - tensor last_hidden_state = add(x = input_171, y = hidden_states_65)[name = tensor("input_179")]; + tensor text_encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222517248)))]; + tensor text_encoder_text_model_encoder_layers_10_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222518848)))]; + tensor input_173_cast = layer_norm(axes = input_173_axes_0, beta = text_encoder_text_model_encoder_layers_10_layer_norm2_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16, x = input_171_cast)[name = tensor("input_173_cast")]; + tensor text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222520448)))]; + tensor text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(227239104)))]; + tensor input_175_cast = linear(bias = text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16, x = input_173_cast)[name = tensor("input_175_cast")]; + tensor var_1051_to_fp16 = const()[name = tensor("op_1051_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_1052_cast = mul(x = input_175_cast, y = var_1051_to_fp16)[name = tensor("op_1052_cast")]; + tensor var_1053_cast = sigmoid(x = var_1052_cast)[name = tensor("op_1053_cast")]; + tensor input_177_cast = mul(x = input_175_cast, y = var_1053_cast)[name = tensor("input_177_cast")]; + tensor text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(227245312)))]; + tensor text_encoder_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(231963968)))]; + tensor hidden_states_65_cast = linear(bias = text_encoder_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16, x = input_177_cast)[name = tensor("hidden_states_65_cast")]; + tensor input_179_cast = add(x = input_171_cast, y = hidden_states_65_cast)[name = tensor("input_179_cast")]; + tensor input_179_cast_to_fp32_dtype_0 = const()[name = tensor("input_179_cast_to_fp32_dtype_0"), val = tensor("fp32")]; tensor hidden_states_67_axes_0 = const()[name = tensor("hidden_states_67_axes_0"), val = tensor([-1])]; - tensor hidden_states_67 = layer_norm(axes = hidden_states_67_axes_0, beta = text_encoder_text_model_encoder_layers_11_layer_norm1_bias, epsilon = var_12, gamma = text_encoder_text_model_encoder_layers_11_layer_norm1_weight, x = last_hidden_state)[name = tensor("hidden_states_67")]; - tensor var_1075 = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_q_proj_bias, weight = text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight, x = hidden_states_67)[name = tensor("op_1075")]; - tensor var_1076 = const()[name = tensor("op_1076"), val = tensor(0x1p-3)]; - tensor tensor_workaround = mul(x = var_1075, y = var_1076)[name = tensor("tensor_workaround")]; - tensor tensor_67 = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_k_proj_bias, weight = text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight, x = hidden_states_67)[name = tensor("tensor_67")]; - tensor var_1081 = const()[name = tensor("op_1081"), val = tensor([1, -1, 12, 64])]; - tensor var_1082 = reshape(shape = var_1081, x = tensor_67)[name = tensor("op_1082")]; - tensor var_1083_perm_0 = const()[name = tensor("op_1083_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor tensor_69 = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_v_proj_bias, weight = text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight, x = hidden_states_67)[name = tensor("tensor_69")]; - tensor var_1088 = const()[name = tensor("op_1088"), val = tensor([1, -1, 12, 64])]; - tensor var_1089 = reshape(shape = var_1088, x = tensor_69)[name = tensor("op_1089")]; - tensor var_1090_perm_0 = const()[name = tensor("op_1090_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_1097 = const()[name = tensor("op_1097"), val = tensor([1, 77, 12, 64])]; - tensor var_1098 = reshape(shape = var_1097, x = tensor_workaround)[name = tensor("op_1098")]; - tensor var_1099_perm_0 = const()[name = tensor("op_1099_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_1101 = const()[name = tensor("op_1101"), val = tensor([12, -1, 64])]; - tensor transpose_2 = transpose(perm = var_1099_perm_0, x = var_1098)[name = tensor("transpose_2")]; - tensor query_states = reshape(shape = var_1101, x = transpose_2)[name = tensor("query_states")]; + tensor text_encoder_text_model_encoder_layers_11_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(231965568)))]; + tensor text_encoder_text_model_encoder_layers_11_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(231967168)))]; + tensor hidden_states_67_cast = layer_norm(axes = hidden_states_67_axes_0, beta = text_encoder_text_model_encoder_layers_11_layer_norm1_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_11_layer_norm1_weight_to_fp16, x = input_179_cast)[name = tensor("hidden_states_67_cast")]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(231968768)))]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(233148480)))]; + tensor var_1077_cast = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16, x = hidden_states_67_cast)[name = tensor("op_1077_cast")]; + tensor var_1078_to_fp16 = const()[name = tensor("op_1078_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_cast = mul(x = var_1077_cast, y = var_1078_to_fp16)[name = tensor("tensor_cast")]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(233150080)))]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(234329792)))]; + tensor tensor_67_cast = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16, x = hidden_states_67_cast)[name = tensor("tensor_67_cast")]; + tensor var_1083 = const()[name = tensor("op_1083"), val = tensor([1, -1, 12, 64])]; + tensor var_1084_cast = reshape(shape = var_1083, x = tensor_67_cast)[name = tensor("op_1084_cast")]; + tensor var_1085_perm_0 = const()[name = tensor("op_1085_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(234331392)))]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(235511104)))]; + tensor tensor_69_cast = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16, x = hidden_states_67_cast)[name = tensor("tensor_69_cast")]; + tensor var_1090 = const()[name = tensor("op_1090"), val = tensor([1, -1, 12, 64])]; + tensor var_1091_cast = reshape(shape = var_1090, x = tensor_69_cast)[name = tensor("op_1091_cast")]; + tensor var_1092_perm_0 = const()[name = tensor("op_1092_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_1099 = const()[name = tensor("op_1099"), val = tensor([1, 77, 12, 64])]; + tensor var_1100_cast = reshape(shape = var_1099, x = tensor_cast)[name = tensor("op_1100_cast")]; + tensor var_1101_perm_0 = const()[name = tensor("op_1101_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_1103 = const()[name = tensor("op_1103"), val = tensor([12, -1, 64])]; - tensor transpose_4 = transpose(perm = var_1083_perm_0, x = var_1082)[name = tensor("transpose_4")]; - tensor key_states = reshape(shape = var_1103, x = transpose_4)[name = tensor("key_states")]; + tensor transpose_2 = transpose(perm = var_1101_perm_0, x = var_1100_cast)[name = tensor("transpose_2")]; + tensor query_states_cast = reshape(shape = var_1103, x = transpose_2)[name = tensor("query_states_cast")]; tensor var_1105 = const()[name = tensor("op_1105"), val = tensor([12, -1, 64])]; - tensor transpose_3 = transpose(perm = var_1090_perm_0, x = var_1089)[name = tensor("transpose_3")]; - tensor value_states = reshape(shape = var_1105, x = transpose_3)[name = tensor("value_states")]; - tensor var_1108_perm_0 = const()[name = tensor("op_1108_perm_0"), val = tensor([0, 2, 1])]; + tensor transpose_4 = transpose(perm = var_1085_perm_0, x = var_1084_cast)[name = tensor("transpose_4")]; + tensor key_states_cast = reshape(shape = var_1105, x = transpose_4)[name = tensor("key_states_cast")]; + tensor var_1107 = const()[name = tensor("op_1107"), val = tensor([12, -1, 64])]; + tensor transpose_3 = transpose(perm = var_1092_perm_0, x = var_1091_cast)[name = tensor("transpose_3")]; + tensor value_states_cast = reshape(shape = var_1107, x = transpose_3)[name = tensor("value_states_cast")]; + tensor var_1110_perm_0 = const()[name = tensor("op_1110_perm_0"), val = tensor([0, 2, 1])]; tensor attn_weights_67_transpose_x_0 = const()[name = tensor("attn_weights_67_transpose_x_0"), val = tensor(false)]; tensor attn_weights_67_transpose_y_0 = const()[name = tensor("attn_weights_67_transpose_y_0"), val = tensor(false)]; - tensor transpose_1 = transpose(perm = var_1108_perm_0, x = key_states)[name = tensor("transpose_1")]; - tensor attn_weights_67 = matmul(transpose_x = attn_weights_67_transpose_x_0, transpose_y = attn_weights_67_transpose_y_0, x = query_states, y = transpose_1)[name = tensor("attn_weights_67")]; - tensor var_1110 = const()[name = tensor("op_1110"), val = tensor([1, 12, 77, 77])]; - tensor var_1111 = reshape(shape = var_1110, x = attn_weights_67)[name = tensor("op_1111")]; - tensor attn_weights_69 = add(x = var_1111, y = causal_attention_mask)[name = tensor("attn_weights_69")]; - tensor var_1116 = const()[name = tensor("op_1116"), val = tensor([12, 77, 77])]; - tensor input_181 = reshape(shape = var_1116, x = attn_weights_69)[name = tensor("input_181")]; - tensor input_183 = softmax(axis = var_5, x = input_181)[name = tensor("input_183")]; + tensor transpose_1 = transpose(perm = var_1110_perm_0, x = key_states_cast)[name = tensor("transpose_1")]; + tensor attn_weights_67_cast = matmul(transpose_x = attn_weights_67_transpose_x_0, transpose_y = attn_weights_67_transpose_y_0, x = query_states_cast, y = transpose_1)[name = tensor("attn_weights_67_cast")]; + tensor var_1112 = const()[name = tensor("op_1112"), val = tensor([1, 12, 77, 77])]; + tensor var_1113_cast = reshape(shape = var_1112, x = attn_weights_67_cast)[name = tensor("op_1113_cast")]; + tensor attn_weights_69_cast = add(x = var_1113_cast, y = causal_attention_mask_to_fp16)[name = tensor("attn_weights_69_cast")]; + tensor var_1118 = const()[name = tensor("op_1118"), val = tensor([12, 77, 77])]; + tensor input_181_cast = reshape(shape = var_1118, x = attn_weights_69_cast)[name = tensor("input_181_cast")]; + tensor input_183_cast = softmax(axis = var_5, x = input_181_cast)[name = tensor("input_183_cast")]; tensor attn_output_67_transpose_x_0 = const()[name = tensor("attn_output_67_transpose_x_0"), val = tensor(false)]; tensor attn_output_67_transpose_y_0 = const()[name = tensor("attn_output_67_transpose_y_0"), val = tensor(false)]; - tensor attn_output_67 = matmul(transpose_x = attn_output_67_transpose_x_0, transpose_y = attn_output_67_transpose_y_0, x = input_183, y = value_states)[name = tensor("attn_output_67")]; - tensor var_1121 = const()[name = tensor("op_1121"), val = tensor([1, 12, 77, 64])]; - tensor attn_output_69 = reshape(shape = var_1121, x = attn_output_67)[name = tensor("attn_output_69")]; + tensor attn_output_67_cast = matmul(transpose_x = attn_output_67_transpose_x_0, transpose_y = attn_output_67_transpose_y_0, x = input_183_cast, y = value_states_cast)[name = tensor("attn_output_67_cast")]; + tensor var_1123 = const()[name = tensor("op_1123"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_69_cast = reshape(shape = var_1123, x = attn_output_67_cast)[name = tensor("attn_output_69_cast")]; tensor attn_output_perm_0 = const()[name = tensor("attn_output_perm_0"), val = tensor([0, 2, 1, 3])]; - tensor var_1124 = const()[name = tensor("op_1124"), val = tensor([1, 77, 768])]; - tensor transpose_0 = transpose(perm = attn_output_perm_0, x = attn_output_69)[name = tensor("transpose_0")]; - tensor input_185 = reshape(shape = var_1124, x = transpose_0)[name = tensor("input_185")]; - tensor hidden_states_69 = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_out_proj_bias, weight = text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight, x = input_185)[name = tensor("hidden_states_69")]; - tensor input_187 = add(x = last_hidden_state, y = hidden_states_69)[name = tensor("input_187")]; + tensor var_1126 = const()[name = tensor("op_1126"), val = tensor([1, 77, 768])]; + tensor transpose_0 = transpose(perm = attn_output_perm_0, x = attn_output_69_cast)[name = tensor("transpose_0")]; + tensor input_185_cast = reshape(shape = var_1126, x = transpose_0)[name = tensor("input_185_cast")]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(235512704)))]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236692416)))]; + tensor hidden_states_69_cast = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16, x = input_185_cast)[name = tensor("hidden_states_69_cast")]; + tensor input_187_cast = add(x = input_179_cast, y = hidden_states_69_cast)[name = tensor("input_187_cast")]; tensor input_189_axes_0 = const()[name = tensor("input_189_axes_0"), val = tensor([-1])]; - tensor input_189 = layer_norm(axes = input_189_axes_0, beta = text_encoder_text_model_encoder_layers_11_layer_norm2_bias, epsilon = var_12, gamma = text_encoder_text_model_encoder_layers_11_layer_norm2_weight, x = input_187)[name = tensor("input_189")]; - tensor input_191 = linear(bias = text_encoder_text_model_encoder_layers_11_mlp_fc1_bias, weight = text_encoder_text_model_encoder_layers_11_mlp_fc1_weight, x = input_189)[name = tensor("input_191")]; - tensor var_1139 = const()[name = tensor("op_1139"), val = tensor(0x1.b3b646p+0)]; - tensor var_1140 = mul(x = input_191, y = var_1139)[name = tensor("op_1140")]; - tensor var_1141 = sigmoid(x = var_1140)[name = tensor("op_1141")]; - tensor input_193 = mul(x = input_191, y = var_1141)[name = tensor("input_193")]; - tensor hidden_states = linear(bias = text_encoder_text_model_encoder_layers_11_mlp_fc2_bias, weight = text_encoder_text_model_encoder_layers_11_mlp_fc2_weight, x = input_193)[name = tensor("hidden_states")]; - tensor input = add(x = input_187, y = hidden_states)[name = tensor("input")]; + tensor text_encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236694016)))]; + tensor text_encoder_text_model_encoder_layers_11_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236695616)))]; + tensor input_189_cast = layer_norm(axes = input_189_axes_0, beta = text_encoder_text_model_encoder_layers_11_layer_norm2_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16, x = input_187_cast)[name = tensor("input_189_cast")]; + tensor text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236697216)))]; + tensor text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(241415872)))]; + tensor input_191_cast = linear(bias = text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16, x = input_189_cast)[name = tensor("input_191_cast")]; + tensor var_1141_to_fp16 = const()[name = tensor("op_1141_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_1142_cast = mul(x = input_191_cast, y = var_1141_to_fp16)[name = tensor("op_1142_cast")]; + tensor var_1143_cast = sigmoid(x = var_1142_cast)[name = tensor("op_1143_cast")]; + tensor input_193_cast = mul(x = input_191_cast, y = var_1143_cast)[name = tensor("input_193_cast")]; + tensor text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(241422080)))]; + tensor text_encoder_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(246140736)))]; + tensor hidden_states_cast = linear(bias = text_encoder_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16, x = input_193_cast)[name = tensor("hidden_states_cast")]; + tensor input_cast = add(x = input_187_cast, y = hidden_states_cast)[name = tensor("input_cast")]; tensor last_hidden_state_axes_0 = const()[name = tensor("last_hidden_state_axes_0"), val = tensor([-1])]; - tensor pooled_outputs = layer_norm(axes = last_hidden_state_axes_0, beta = text_encoder_text_model_final_layer_norm_bias, epsilon = var_12, gamma = text_encoder_text_model_final_layer_norm_weight, x = input)[name = tensor("last_hidden_state")]; + tensor text_encoder_text_model_final_layer_norm_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_final_layer_norm_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(246142336)))]; + tensor text_encoder_text_model_final_layer_norm_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_final_layer_norm_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(246143936)))]; + tensor last_hidden_state_cast = layer_norm(axes = last_hidden_state_axes_0, beta = text_encoder_text_model_final_layer_norm_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_final_layer_norm_weight_to_fp16, x = input_cast)[name = tensor("last_hidden_state_cast")]; + tensor last_hidden_state_cast_to_fp32_dtype_0 = const()[name = tensor("last_hidden_state_cast_to_fp32_dtype_0"), val = tensor("fp32")]; + tensor pooled_outputs = cast(dtype = last_hidden_state_cast_to_fp32_dtype_0, x = last_hidden_state_cast)[name = tensor("cast_123")]; + tensor last_hidden_state = cast(dtype = input_179_cast_to_fp32_dtype_0, x = input_179_cast)[name = tensor("cast_157")]; } -> (last_hidden_state, pooled_outputs); } \ No newline at end of file