program(1.0) [buildInfo = dict, tensor>({{"coremlc-component-MIL", "3520.4.1"}, {"coremlc-version", "3520.5.1"}})] { func main(tensor cache_length, tensor decoder_key_padding_mask, tensor encoder_output_embeds, tensor input_ids, tensor key_cache, tensor kv_cache_update_mask, tensor value_cache) { tensor var_64_axis_0 = const()[name = tensor("op_64_axis_0"), val = tensor(0)]; tensor var_64_batch_dims_0 = const()[name = tensor("op_64_batch_dims_0"), val = tensor(0)]; tensor embed_tokens_weight_to_fp16 = const()[name = tensor("embed_tokens_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; tensor var_64_cast_fp16 = gather(axis = var_64_axis_0, batch_dims = var_64_batch_dims_0, indices = input_ids, x = embed_tokens_weight_to_fp16)[name = tensor("op_64_cast_fp16")]; tensor var_68_axis_0 = const()[name = tensor("op_68_axis_0"), val = tensor(0)]; tensor var_68_batch_dims_0 = const()[name = tensor("op_68_batch_dims_0"), val = tensor(0)]; tensor embed_positions_weight_to_fp16 = const()[name = tensor("embed_positions_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106219648)))]; tensor var_68_cast_fp16 = gather(axis = var_68_axis_0, batch_dims = var_68_batch_dims_0, indices = cache_length, x = embed_positions_weight_to_fp16)[name = tensor("op_68_cast_fp16")]; tensor hidden_states_1_cast_fp16 = add(x = var_64_cast_fp16, y = var_68_cast_fp16)[name = tensor("hidden_states_1_cast_fp16")]; tensor var_82_axes_0 = const()[name = tensor("op_82_axes_0"), val = tensor([2])]; tensor var_82_cast_fp16 = expand_dims(axes = var_82_axes_0, x = hidden_states_1_cast_fp16)[name = tensor("op_82_cast_fp16")]; tensor inputs_1_axes_0 = const()[name = tensor("inputs_1_axes_0"), val = tensor([3])]; tensor inputs_1_cast_fp16 = expand_dims(axes = inputs_1_axes_0, x = var_82_cast_fp16)[name = tensor("inputs_1_cast_fp16")]; tensor tile_0 = const()[name = tensor("tile_0"), val = tensor([1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024])]; tensor var_87_axis_0 = const()[name = tensor("op_87_axis_0"), val = tensor(1)]; tensor var_87_cast_fp16_0, tensor var_87_cast_fp16_1, tensor var_87_cast_fp16_2, tensor var_87_cast_fp16_3, tensor var_87_cast_fp16_4, tensor var_87_cast_fp16_5, tensor var_87_cast_fp16_6, tensor var_87_cast_fp16_7, tensor var_87_cast_fp16_8, tensor var_87_cast_fp16_9, tensor var_87_cast_fp16_10, tensor var_87_cast_fp16_11, tensor var_87_cast_fp16_12, tensor var_87_cast_fp16_13, tensor var_87_cast_fp16_14, tensor var_87_cast_fp16_15, tensor var_87_cast_fp16_16, tensor var_87_cast_fp16_17, tensor var_87_cast_fp16_18, tensor var_87_cast_fp16_19, tensor var_87_cast_fp16_20, tensor var_87_cast_fp16_21, tensor var_87_cast_fp16_22, tensor var_87_cast_fp16_23 = split(axis = var_87_axis_0, split_sizes = tile_0, x = key_cache)[name = tensor("op_87_cast_fp16")]; tensor tile_1 = const()[name = tensor("tile_1"), val = tensor([1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024])]; tensor var_114_axis_0 = const()[name = tensor("op_114_axis_0"), val = tensor(1)]; tensor var_114_cast_fp16_0, tensor var_114_cast_fp16_1, tensor var_114_cast_fp16_2, tensor var_114_cast_fp16_3, tensor var_114_cast_fp16_4, tensor var_114_cast_fp16_5, tensor var_114_cast_fp16_6, tensor var_114_cast_fp16_7, tensor var_114_cast_fp16_8, tensor var_114_cast_fp16_9, tensor var_114_cast_fp16_10, tensor var_114_cast_fp16_11, tensor var_114_cast_fp16_12, tensor var_114_cast_fp16_13, tensor var_114_cast_fp16_14, tensor var_114_cast_fp16_15, tensor var_114_cast_fp16_16, tensor var_114_cast_fp16_17, tensor var_114_cast_fp16_18, tensor var_114_cast_fp16_19, tensor var_114_cast_fp16_20, tensor var_114_cast_fp16_21, tensor var_114_cast_fp16_22, tensor var_114_cast_fp16_23 = split(axis = var_114_axis_0, split_sizes = tile_1, x = value_cache)[name = tensor("op_114_cast_fp16")]; tensor var_144 = const()[name = tensor("op_144"), val = tensor(3)]; tensor out_1_axes_0 = const()[name = tensor("out_1_axes_0"), val = tensor([1])]; tensor var_169_to_fp16 = const()[name = tensor("op_169_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_169_to_fp16, x = inputs_1_cast_fp16)[name = tensor("out_1_cast_fp16")]; tensor obj_1_mean_0_to_fp16 = const()[name = tensor("obj_1_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107137216)))]; tensor obj_1_variance_0_to_fp16 = const()[name = tensor("obj_1_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107139328)))]; tensor obj_1_gamma_0_to_fp16 = const()[name = tensor("obj_1_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107141440)))]; tensor obj_1_beta_0_to_fp16 = const()[name = tensor("obj_1_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107143552)))]; tensor obj_1_epsilon_0_to_fp16 = const()[name = tensor("obj_1_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_1_cast_fp16 = batch_norm(beta = obj_1_beta_0_to_fp16, epsilon = obj_1_epsilon_0_to_fp16, gamma = obj_1_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_1_cast_fp16)[name = tensor("obj_1_cast_fp16")]; tensor query_1_pad_type_0 = const()[name = tensor("query_1_pad_type_0"), val = tensor("valid")]; tensor query_1_strides_0 = const()[name = tensor("query_1_strides_0"), val = tensor([1, 1])]; tensor query_1_pad_0 = const()[name = tensor("query_1_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_1_dilations_0 = const()[name = tensor("query_1_dilations_0"), val = tensor([1, 1])]; tensor query_1_groups_0 = const()[name = tensor("query_1_groups_0"), val = tensor(1)]; tensor layers_0_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107145664))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107932160))), name = tensor("layers_0_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107932352)))]; tensor query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16_palettized, x = obj_1_cast_fp16)[name = tensor("query_1_cast_fp16")]; tensor current_key_1_pad_type_0 = const()[name = tensor("current_key_1_pad_type_0"), val = tensor("valid")]; tensor current_key_1_strides_0 = const()[name = tensor("current_key_1_strides_0"), val = tensor([1, 1])]; tensor current_key_1_pad_0 = const()[name = tensor("current_key_1_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_1_dilations_0 = const()[name = tensor("current_key_1_dilations_0"), val = tensor([1, 1])]; tensor current_key_1_groups_0 = const()[name = tensor("current_key_1_groups_0"), val = tensor(1)]; tensor layers_0_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107934464))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108720960))), name = tensor("layers_0_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor current_key_1_cast_fp16 = conv(dilations = current_key_1_dilations_0, groups = current_key_1_groups_0, pad = current_key_1_pad_0, pad_type = current_key_1_pad_type_0, strides = current_key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16_palettized, x = obj_1_cast_fp16)[name = tensor("current_key_1_cast_fp16")]; tensor current_value_1_pad_type_0 = const()[name = tensor("current_value_1_pad_type_0"), val = tensor("valid")]; tensor current_value_1_strides_0 = const()[name = tensor("current_value_1_strides_0"), val = tensor([1, 1])]; tensor current_value_1_pad_0 = const()[name = tensor("current_value_1_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_1_dilations_0 = const()[name = tensor("current_value_1_dilations_0"), val = tensor([1, 1])]; tensor current_value_1_groups_0 = const()[name = tensor("current_value_1_groups_0"), val = tensor(1)]; tensor layers_0_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108721152))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109507648))), name = tensor("layers_0_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109507840)))]; tensor current_value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = current_value_1_dilations_0, groups = current_value_1_groups_0, pad = current_value_1_pad_0, pad_type = current_value_1_pad_type_0, strides = current_value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16_palettized, x = obj_1_cast_fp16)[name = tensor("current_value_1_cast_fp16")]; tensor var_204_axes_0 = const()[name = tensor("op_204_axes_0"), val = tensor([1])]; tensor var_204_cast_fp16 = expand_dims(axes = var_204_axes_0, x = kv_cache_update_mask)[name = tensor("op_204_cast_fp16")]; tensor var_205_axes_0 = const()[name = tensor("op_205_axes_0"), val = tensor([2])]; tensor var_205_cast_fp16 = expand_dims(axes = var_205_axes_0, x = var_204_cast_fp16)[name = tensor("op_205_cast_fp16")]; tensor var_145_to_fp16 = const()[name = tensor("op_145_to_fp16"), val = tensor(0x1p+0)]; tensor var_207_cast_fp16 = sub(x = var_145_to_fp16, y = var_205_cast_fp16)[name = tensor("op_207_cast_fp16")]; tensor var_208_cast_fp16 = mul(x = var_87_cast_fp16_0, y = var_207_cast_fp16)[name = tensor("op_208_cast_fp16")]; tensor var_209_cast_fp16 = mul(x = current_key_1_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_209_cast_fp16")]; tensor key_1_cast_fp16 = add(x = var_208_cast_fp16, y = var_209_cast_fp16)[name = tensor("key_1_cast_fp16")]; tensor var_212_cast_fp16 = mul(x = var_114_cast_fp16_0, y = var_207_cast_fp16)[name = tensor("op_212_cast_fp16")]; tensor var_213_cast_fp16 = mul(x = current_value_1_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_213_cast_fp16")]; tensor value_1_cast_fp16 = add(x = var_212_cast_fp16, y = var_213_cast_fp16)[name = tensor("value_1_cast_fp16")]; tensor var_217 = const()[name = tensor("op_217"), val = tensor([1, 16, 64, 1])]; tensor mh_q_1_cast_fp16 = reshape(shape = var_217, x = query_1_cast_fp16)[name = tensor("mh_q_1_cast_fp16")]; tensor var_219_to_fp16 = const()[name = tensor("op_219_to_fp16"), val = tensor(0x1p-3)]; tensor var_220_cast_fp16 = mul(x = mh_q_1_cast_fp16, y = var_219_to_fp16)[name = tensor("op_220_cast_fp16")]; tensor var_223 = const()[name = tensor("op_223"), val = tensor([1, 16, 64, 448])]; tensor var_224_cast_fp16 = reshape(shape = var_223, x = key_1_cast_fp16)[name = tensor("op_224_cast_fp16")]; tensor mh_w_1_transpose_x_0 = const()[name = tensor("mh_w_1_transpose_x_0"), val = tensor(true)]; tensor mh_w_1_transpose_y_0 = const()[name = tensor("mh_w_1_transpose_y_0"), val = tensor(false)]; tensor mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_220_cast_fp16, y = var_224_cast_fp16)[name = tensor("mh_w_1_cast_fp16")]; tensor var_228_axes_0 = const()[name = tensor("op_228_axes_0"), val = tensor([1])]; tensor var_228_cast_fp16 = expand_dims(axes = var_228_axes_0, x = decoder_key_padding_mask)[name = tensor("op_228_cast_fp16")]; tensor var_229_axes_0 = const()[name = tensor("op_229_axes_0"), val = tensor([2])]; tensor var_229_cast_fp16 = expand_dims(axes = var_229_axes_0, x = var_228_cast_fp16)[name = tensor("op_229_cast_fp16")]; tensor mh_w_3_cast_fp16 = add(x = mh_w_1_cast_fp16, y = var_229_cast_fp16)[name = tensor("mh_w_3_cast_fp16")]; tensor var_232_cast_fp16 = softmax(axis = var_144, x = mh_w_3_cast_fp16)[name = tensor("op_232_cast_fp16")]; tensor var_233 = const()[name = tensor("op_233"), val = tensor([1, 16, 64, 448])]; tensor var_234_cast_fp16 = reshape(shape = var_233, x = value_1_cast_fp16)[name = tensor("op_234_cast_fp16")]; tensor attn_1_transpose_x_0 = const()[name = tensor("attn_1_transpose_x_0"), val = tensor(false)]; tensor attn_1_transpose_y_0 = const()[name = tensor("attn_1_transpose_y_0"), val = tensor(true)]; tensor attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_234_cast_fp16, y = var_232_cast_fp16)[name = tensor("attn_1_cast_fp16")]; tensor var_237 = const()[name = tensor("op_237"), val = tensor([1, 1024, 1, 1])]; tensor input_1_cast_fp16 = reshape(shape = var_237, x = attn_1_cast_fp16)[name = tensor("input_1_cast_fp16")]; tensor obj_7_pad_type_0 = const()[name = tensor("obj_7_pad_type_0"), val = tensor("valid")]; tensor obj_7_strides_0 = const()[name = tensor("obj_7_strides_0"), val = tensor([1, 1])]; tensor obj_7_pad_0 = const()[name = tensor("obj_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_7_dilations_0 = const()[name = tensor("obj_7_dilations_0"), val = tensor([1, 1])]; tensor obj_7_groups_0 = const()[name = tensor("obj_7_groups_0"), val = tensor(1)]; tensor layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109509952)))]; tensor layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111607168)))]; tensor obj_7_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_7_dilations_0, groups = obj_7_groups_0, pad = obj_7_pad_0, pad_type = obj_7_pad_type_0, strides = obj_7_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = tensor("obj_7_cast_fp16")]; tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_7_cast_fp16)[name = tensor("inputs_3_cast_fp16")]; tensor out_3_axes_0 = const()[name = tensor("out_3_axes_0"), val = tensor([1])]; tensor var_259_to_fp16 = const()[name = tensor("op_259_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_259_to_fp16, x = inputs_3_cast_fp16)[name = tensor("out_3_cast_fp16")]; tensor obj_9_gamma_0_to_fp16 = const()[name = tensor("obj_9_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111609280)))]; tensor obj_9_beta_0_to_fp16 = const()[name = tensor("obj_9_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111611392)))]; tensor obj_9_epsilon_0_to_fp16 = const()[name = tensor("obj_9_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_9_cast_fp16 = batch_norm(beta = obj_9_beta_0_to_fp16, epsilon = obj_9_epsilon_0_to_fp16, gamma = obj_9_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_3_cast_fp16)[name = tensor("obj_9_cast_fp16")]; tensor query_3_pad_type_0 = const()[name = tensor("query_3_pad_type_0"), val = tensor("valid")]; tensor query_3_strides_0 = const()[name = tensor("query_3_strides_0"), val = tensor([1, 1])]; tensor query_3_pad_0 = const()[name = tensor("query_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_3_dilations_0 = const()[name = tensor("query_3_dilations_0"), val = tensor([1, 1])]; tensor query_3_groups_0 = const()[name = tensor("query_3_groups_0"), val = tensor(1)]; tensor layers_0_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111613504))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112400000))), name = tensor("layers_0_encoder_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_0_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_0_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112400192)))]; tensor query_3_cast_fp16 = conv(bias = layers_0_encoder_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_0_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_9_cast_fp16)[name = tensor("query_3_cast_fp16")]; tensor key_3_pad_type_0 = const()[name = tensor("key_3_pad_type_0"), val = tensor("valid")]; tensor key_3_strides_0 = const()[name = tensor("key_3_strides_0"), val = tensor([1, 1])]; tensor key_3_pad_0 = const()[name = tensor("key_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_3_dilations_0 = const()[name = tensor("key_3_dilations_0"), val = tensor([1, 1])]; tensor key_3_groups_0 = const()[name = tensor("key_3_groups_0"), val = tensor(1)]; tensor layers_0_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112402304))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113188800))), name = tensor("layers_0_encoder_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor key_3_cast_fp16 = conv(dilations = key_3_dilations_0, groups = key_3_groups_0, pad = key_3_pad_0, pad_type = key_3_pad_type_0, strides = key_3_strides_0, weight = layers_0_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("key_3_cast_fp16")]; tensor value_3_pad_type_0 = const()[name = tensor("value_3_pad_type_0"), val = tensor("valid")]; tensor value_3_strides_0 = const()[name = tensor("value_3_strides_0"), val = tensor([1, 1])]; tensor value_3_pad_0 = const()[name = tensor("value_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_3_dilations_0 = const()[name = tensor("value_3_dilations_0"), val = tensor([1, 1])]; tensor value_3_groups_0 = const()[name = tensor("value_3_groups_0"), val = tensor(1)]; tensor layers_0_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113188992))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113975488))), name = tensor("layers_0_encoder_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_0_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_0_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113975680)))]; tensor value_3_cast_fp16 = conv(bias = layers_0_encoder_attn_v_proj_bias_to_fp16, dilations = value_3_dilations_0, groups = value_3_groups_0, pad = value_3_pad_0, pad_type = value_3_pad_type_0, strides = value_3_strides_0, weight = layers_0_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("value_3_cast_fp16")]; tensor var_295 = const()[name = tensor("op_295"), val = tensor([1, 16, 64, 1])]; tensor mh_q_3_cast_fp16 = reshape(shape = var_295, x = query_3_cast_fp16)[name = tensor("mh_q_3_cast_fp16")]; tensor var_297_to_fp16 = const()[name = tensor("op_297_to_fp16"), val = tensor(0x1p-3)]; tensor var_298_cast_fp16 = mul(x = mh_q_3_cast_fp16, y = var_297_to_fp16)[name = tensor("op_298_cast_fp16")]; tensor var_301 = const()[name = tensor("op_301"), val = tensor([1, 16, 64, 1500])]; tensor var_302_cast_fp16 = reshape(shape = var_301, x = key_3_cast_fp16)[name = tensor("op_302_cast_fp16")]; tensor mh_w_5_transpose_x_0 = const()[name = tensor("mh_w_5_transpose_x_0"), val = tensor(true)]; tensor mh_w_5_transpose_y_0 = const()[name = tensor("mh_w_5_transpose_y_0"), val = tensor(false)]; tensor mh_w_5_cast_fp16 = matmul(transpose_x = mh_w_5_transpose_x_0, transpose_y = mh_w_5_transpose_y_0, x = var_298_cast_fp16, y = var_302_cast_fp16)[name = tensor("mh_w_5_cast_fp16")]; tensor obj_13_cast_fp16 = softmax(axis = var_144, x = mh_w_5_cast_fp16)[name = tensor("obj_13_cast_fp16")]; tensor var_306 = const()[name = tensor("op_306"), val = tensor([1, 16, 64, 1500])]; tensor var_307_cast_fp16 = reshape(shape = var_306, x = value_3_cast_fp16)[name = tensor("op_307_cast_fp16")]; tensor attn_3_transpose_x_0 = const()[name = tensor("attn_3_transpose_x_0"), val = tensor(false)]; tensor attn_3_transpose_y_0 = const()[name = tensor("attn_3_transpose_y_0"), val = tensor(true)]; tensor attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_307_cast_fp16, y = obj_13_cast_fp16)[name = tensor("attn_3_cast_fp16")]; tensor var_310 = const()[name = tensor("op_310"), val = tensor([1, 1024, 1, 1])]; tensor input_3_cast_fp16 = reshape(shape = var_310, x = attn_3_cast_fp16)[name = tensor("input_3_cast_fp16")]; tensor obj_11_pad_type_0 = const()[name = tensor("obj_11_pad_type_0"), val = tensor("valid")]; tensor obj_11_strides_0 = const()[name = tensor("obj_11_strides_0"), val = tensor([1, 1])]; tensor obj_11_pad_0 = const()[name = tensor("obj_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_11_dilations_0 = const()[name = tensor("obj_11_dilations_0"), val = tensor([1, 1])]; tensor obj_11_groups_0 = const()[name = tensor("obj_11_groups_0"), val = tensor(1)]; tensor layers_0_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113977792))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114764288))), name = tensor("layers_0_encoder_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_0_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_0_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114764480)))]; tensor obj_11_cast_fp16 = conv(bias = layers_0_encoder_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_0_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_3_cast_fp16)[name = tensor("obj_11_cast_fp16")]; tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = obj_11_cast_fp16)[name = tensor("inputs_5_cast_fp16")]; tensor out_5_axes_0 = const()[name = tensor("out_5_axes_0"), val = tensor([1])]; tensor var_328_to_fp16 = const()[name = tensor("op_328_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_328_to_fp16, x = inputs_5_cast_fp16)[name = tensor("out_5_cast_fp16")]; tensor input_5_gamma_0_to_fp16 = const()[name = tensor("input_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114766592)))]; tensor input_5_beta_0_to_fp16 = const()[name = tensor("input_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114768704)))]; tensor input_5_epsilon_0_to_fp16 = const()[name = tensor("input_5_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_5_cast_fp16 = batch_norm(beta = input_5_beta_0_to_fp16, epsilon = input_5_epsilon_0_to_fp16, gamma = input_5_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_5_cast_fp16)[name = tensor("input_5_cast_fp16")]; tensor input_7_pad_type_0 = const()[name = tensor("input_7_pad_type_0"), val = tensor("valid")]; tensor input_7_strides_0 = const()[name = tensor("input_7_strides_0"), val = tensor([1, 1])]; tensor input_7_pad_0 = const()[name = tensor("input_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_7_dilations_0 = const()[name = tensor("input_7_dilations_0"), val = tensor([1, 1])]; tensor input_7_groups_0 = const()[name = tensor("input_7_groups_0"), val = tensor(1)]; tensor layers_0_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114770816))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(117916608))), name = tensor("layers_0_fc1_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; tensor layers_0_fc1_bias_to_fp16 = const()[name = tensor("layers_0_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(117916800)))]; tensor input_7_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = layers_0_fc1_weight_to_fp16_palettized, x = input_5_cast_fp16)[name = tensor("input_7_cast_fp16")]; tensor input_9_mode_0 = const()[name = tensor("input_9_mode_0"), val = tensor("EXACT")]; tensor input_9_cast_fp16 = gelu(mode = input_9_mode_0, x = input_7_cast_fp16)[name = tensor("input_9_cast_fp16")]; tensor hidden_states_3_pad_type_0 = const()[name = tensor("hidden_states_3_pad_type_0"), val = tensor("valid")]; tensor hidden_states_3_strides_0 = const()[name = tensor("hidden_states_3_strides_0"), val = tensor([1, 1])]; tensor hidden_states_3_pad_0 = const()[name = tensor("hidden_states_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_3_dilations_0 = const()[name = tensor("hidden_states_3_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_3_groups_0 = const()[name = tensor("hidden_states_3_groups_0"), val = tensor(1)]; tensor layers_0_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(117925056))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121070848))), name = tensor("layers_0_fc2_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; tensor layers_0_fc2_bias_to_fp16 = const()[name = tensor("layers_0_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121071040)))]; tensor hidden_states_3_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_3_dilations_0, groups = hidden_states_3_groups_0, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = hidden_states_3_strides_0, weight = layers_0_fc2_weight_to_fp16_palettized, x = input_9_cast_fp16)[name = tensor("hidden_states_3_cast_fp16")]; tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = hidden_states_3_cast_fp16)[name = tensor("inputs_7_cast_fp16")]; tensor var_363 = const()[name = tensor("op_363"), val = tensor(3)]; tensor out_7_axes_0 = const()[name = tensor("out_7_axes_0"), val = tensor([1])]; tensor var_388_to_fp16 = const()[name = tensor("op_388_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_388_to_fp16, x = inputs_7_cast_fp16)[name = tensor("out_7_cast_fp16")]; tensor obj_15_gamma_0_to_fp16 = const()[name = tensor("obj_15_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121073152)))]; tensor obj_15_beta_0_to_fp16 = const()[name = tensor("obj_15_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121075264)))]; tensor obj_15_epsilon_0_to_fp16 = const()[name = tensor("obj_15_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_15_cast_fp16 = batch_norm(beta = obj_15_beta_0_to_fp16, epsilon = obj_15_epsilon_0_to_fp16, gamma = obj_15_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_7_cast_fp16)[name = tensor("obj_15_cast_fp16")]; tensor query_5_pad_type_0 = const()[name = tensor("query_5_pad_type_0"), val = tensor("valid")]; tensor query_5_strides_0 = const()[name = tensor("query_5_strides_0"), val = tensor([1, 1])]; tensor query_5_pad_0 = const()[name = tensor("query_5_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_5_dilations_0 = const()[name = tensor("query_5_dilations_0"), val = tensor([1, 1])]; tensor query_5_groups_0 = const()[name = tensor("query_5_groups_0"), val = tensor(1)]; tensor layers_1_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121077376))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121863872))), name = tensor("layers_1_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121864064)))]; tensor query_5_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16_palettized, x = obj_15_cast_fp16)[name = tensor("query_5_cast_fp16")]; tensor current_key_3_pad_type_0 = const()[name = tensor("current_key_3_pad_type_0"), val = tensor("valid")]; tensor current_key_3_strides_0 = const()[name = tensor("current_key_3_strides_0"), val = tensor([1, 1])]; tensor current_key_3_pad_0 = const()[name = tensor("current_key_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_3_dilations_0 = const()[name = tensor("current_key_3_dilations_0"), val = tensor([1, 1])]; tensor current_key_3_groups_0 = const()[name = tensor("current_key_3_groups_0"), val = tensor(1)]; tensor layers_1_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121866176))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(122652672))), name = tensor("layers_1_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor current_key_3_cast_fp16 = conv(dilations = current_key_3_dilations_0, groups = current_key_3_groups_0, pad = current_key_3_pad_0, pad_type = current_key_3_pad_type_0, strides = current_key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16_palettized, x = obj_15_cast_fp16)[name = tensor("current_key_3_cast_fp16")]; tensor current_value_3_pad_type_0 = const()[name = tensor("current_value_3_pad_type_0"), val = tensor("valid")]; tensor current_value_3_strides_0 = const()[name = tensor("current_value_3_strides_0"), val = tensor([1, 1])]; tensor current_value_3_pad_0 = const()[name = tensor("current_value_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_3_dilations_0 = const()[name = tensor("current_value_3_dilations_0"), val = tensor([1, 1])]; tensor current_value_3_groups_0 = const()[name = tensor("current_value_3_groups_0"), val = tensor(1)]; tensor layers_1_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(122652864))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123439360))), name = tensor("layers_1_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123439552)))]; tensor current_value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = current_value_3_dilations_0, groups = current_value_3_groups_0, pad = current_value_3_pad_0, pad_type = current_value_3_pad_type_0, strides = current_value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16_palettized, x = obj_15_cast_fp16)[name = tensor("current_value_3_cast_fp16")]; tensor var_427_cast_fp16 = mul(x = var_87_cast_fp16_1, y = var_207_cast_fp16)[name = tensor("op_427_cast_fp16")]; tensor var_428_cast_fp16 = mul(x = current_key_3_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_428_cast_fp16")]; tensor key_5_cast_fp16 = add(x = var_427_cast_fp16, y = var_428_cast_fp16)[name = tensor("key_5_cast_fp16")]; tensor var_431_cast_fp16 = mul(x = var_114_cast_fp16_1, y = var_207_cast_fp16)[name = tensor("op_431_cast_fp16")]; tensor var_432_cast_fp16 = mul(x = current_value_3_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_432_cast_fp16")]; tensor value_5_cast_fp16 = add(x = var_431_cast_fp16, y = var_432_cast_fp16)[name = tensor("value_5_cast_fp16")]; tensor var_436 = const()[name = tensor("op_436"), val = tensor([1, 16, 64, 1])]; tensor mh_q_5_cast_fp16 = reshape(shape = var_436, x = query_5_cast_fp16)[name = tensor("mh_q_5_cast_fp16")]; tensor var_438_to_fp16 = const()[name = tensor("op_438_to_fp16"), val = tensor(0x1p-3)]; tensor var_439_cast_fp16 = mul(x = mh_q_5_cast_fp16, y = var_438_to_fp16)[name = tensor("op_439_cast_fp16")]; tensor var_442 = const()[name = tensor("op_442"), val = tensor([1, 16, 64, 448])]; tensor var_443_cast_fp16 = reshape(shape = var_442, x = key_5_cast_fp16)[name = tensor("op_443_cast_fp16")]; tensor mh_w_7_transpose_x_0 = const()[name = tensor("mh_w_7_transpose_x_0"), val = tensor(true)]; tensor mh_w_7_transpose_y_0 = const()[name = tensor("mh_w_7_transpose_y_0"), val = tensor(false)]; tensor mh_w_7_cast_fp16 = matmul(transpose_x = mh_w_7_transpose_x_0, transpose_y = mh_w_7_transpose_y_0, x = var_439_cast_fp16, y = var_443_cast_fp16)[name = tensor("mh_w_7_cast_fp16")]; tensor mh_w_9_cast_fp16 = add(x = mh_w_7_cast_fp16, y = var_229_cast_fp16)[name = tensor("mh_w_9_cast_fp16")]; tensor var_451_cast_fp16 = softmax(axis = var_363, x = mh_w_9_cast_fp16)[name = tensor("op_451_cast_fp16")]; tensor var_452 = const()[name = tensor("op_452"), val = tensor([1, 16, 64, 448])]; tensor var_453_cast_fp16 = reshape(shape = var_452, x = value_5_cast_fp16)[name = tensor("op_453_cast_fp16")]; tensor attn_5_transpose_x_0 = const()[name = tensor("attn_5_transpose_x_0"), val = tensor(false)]; tensor attn_5_transpose_y_0 = const()[name = tensor("attn_5_transpose_y_0"), val = tensor(true)]; tensor attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_453_cast_fp16, y = var_451_cast_fp16)[name = tensor("attn_5_cast_fp16")]; tensor var_456 = const()[name = tensor("op_456"), val = tensor([1, 1024, 1, 1])]; tensor input_11_cast_fp16 = reshape(shape = var_456, x = attn_5_cast_fp16)[name = tensor("input_11_cast_fp16")]; tensor obj_21_pad_type_0 = const()[name = tensor("obj_21_pad_type_0"), val = tensor("valid")]; tensor obj_21_strides_0 = const()[name = tensor("obj_21_strides_0"), val = tensor([1, 1])]; tensor obj_21_pad_0 = const()[name = tensor("obj_21_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_21_dilations_0 = const()[name = tensor("obj_21_dilations_0"), val = tensor([1, 1])]; tensor obj_21_groups_0 = const()[name = tensor("obj_21_groups_0"), val = tensor(1)]; tensor layers_1_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123441664))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124228160))), name = tensor("layers_1_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124228352)))]; tensor obj_21_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_21_dilations_0, groups = obj_21_groups_0, pad = obj_21_pad_0, pad_type = obj_21_pad_type_0, strides = obj_21_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16_palettized, x = input_11_cast_fp16)[name = tensor("obj_21_cast_fp16")]; tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = obj_21_cast_fp16)[name = tensor("inputs_9_cast_fp16")]; tensor out_9_axes_0 = const()[name = tensor("out_9_axes_0"), val = tensor([1])]; tensor var_478_to_fp16 = const()[name = tensor("op_478_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_478_to_fp16, x = inputs_9_cast_fp16)[name = tensor("out_9_cast_fp16")]; tensor obj_23_gamma_0_to_fp16 = const()[name = tensor("obj_23_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124230464)))]; tensor obj_23_beta_0_to_fp16 = const()[name = tensor("obj_23_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124232576)))]; tensor obj_23_epsilon_0_to_fp16 = const()[name = tensor("obj_23_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_23_cast_fp16 = batch_norm(beta = obj_23_beta_0_to_fp16, epsilon = obj_23_epsilon_0_to_fp16, gamma = obj_23_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_9_cast_fp16)[name = tensor("obj_23_cast_fp16")]; tensor query_7_pad_type_0 = const()[name = tensor("query_7_pad_type_0"), val = tensor("valid")]; tensor query_7_strides_0 = const()[name = tensor("query_7_strides_0"), val = tensor([1, 1])]; tensor query_7_pad_0 = const()[name = tensor("query_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_7_dilations_0 = const()[name = tensor("query_7_dilations_0"), val = tensor([1, 1])]; tensor query_7_groups_0 = const()[name = tensor("query_7_groups_0"), val = tensor(1)]; tensor layers_1_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124234688))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(125021184))), name = tensor("layers_1_encoder_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_1_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_1_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(125021376)))]; tensor query_7_cast_fp16 = conv(bias = layers_1_encoder_attn_q_proj_bias_to_fp16, dilations = query_7_dilations_0, groups = query_7_groups_0, pad = query_7_pad_0, pad_type = query_7_pad_type_0, strides = query_7_strides_0, weight = layers_1_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_23_cast_fp16)[name = tensor("query_7_cast_fp16")]; tensor key_7_pad_type_0 = const()[name = tensor("key_7_pad_type_0"), val = tensor("valid")]; tensor key_7_strides_0 = const()[name = tensor("key_7_strides_0"), val = tensor([1, 1])]; tensor key_7_pad_0 = const()[name = tensor("key_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_7_dilations_0 = const()[name = tensor("key_7_dilations_0"), val = tensor([1, 1])]; tensor key_7_groups_0 = const()[name = tensor("key_7_groups_0"), val = tensor(1)]; tensor layers_1_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(125023488))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(125809984))), name = tensor("layers_1_encoder_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor key_7_cast_fp16 = conv(dilations = key_7_dilations_0, groups = key_7_groups_0, pad = key_7_pad_0, pad_type = key_7_pad_type_0, strides = key_7_strides_0, weight = layers_1_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("key_7_cast_fp16")]; tensor value_7_pad_type_0 = const()[name = tensor("value_7_pad_type_0"), val = tensor("valid")]; tensor value_7_strides_0 = const()[name = tensor("value_7_strides_0"), val = tensor([1, 1])]; tensor value_7_pad_0 = const()[name = tensor("value_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_7_dilations_0 = const()[name = tensor("value_7_dilations_0"), val = tensor([1, 1])]; tensor value_7_groups_0 = const()[name = tensor("value_7_groups_0"), val = tensor(1)]; tensor layers_1_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(125810176))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(126596672))), name = tensor("layers_1_encoder_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_1_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_1_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(126596864)))]; tensor value_7_cast_fp16 = conv(bias = layers_1_encoder_attn_v_proj_bias_to_fp16, dilations = value_7_dilations_0, groups = value_7_groups_0, pad = value_7_pad_0, pad_type = value_7_pad_type_0, strides = value_7_strides_0, weight = layers_1_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("value_7_cast_fp16")]; tensor var_514 = const()[name = tensor("op_514"), val = tensor([1, 16, 64, 1])]; tensor mh_q_7_cast_fp16 = reshape(shape = var_514, x = query_7_cast_fp16)[name = tensor("mh_q_7_cast_fp16")]; tensor var_516_to_fp16 = const()[name = tensor("op_516_to_fp16"), val = tensor(0x1p-3)]; tensor var_517_cast_fp16 = mul(x = mh_q_7_cast_fp16, y = var_516_to_fp16)[name = tensor("op_517_cast_fp16")]; tensor var_520 = const()[name = tensor("op_520"), val = tensor([1, 16, 64, 1500])]; tensor var_521_cast_fp16 = reshape(shape = var_520, x = key_7_cast_fp16)[name = tensor("op_521_cast_fp16")]; tensor mh_w_11_transpose_x_0 = const()[name = tensor("mh_w_11_transpose_x_0"), val = tensor(true)]; tensor mh_w_11_transpose_y_0 = const()[name = tensor("mh_w_11_transpose_y_0"), val = tensor(false)]; tensor mh_w_11_cast_fp16 = matmul(transpose_x = mh_w_11_transpose_x_0, transpose_y = mh_w_11_transpose_y_0, x = var_517_cast_fp16, y = var_521_cast_fp16)[name = tensor("mh_w_11_cast_fp16")]; tensor obj_27_cast_fp16 = softmax(axis = var_363, x = mh_w_11_cast_fp16)[name = tensor("obj_27_cast_fp16")]; tensor var_525 = const()[name = tensor("op_525"), val = tensor([1, 16, 64, 1500])]; tensor var_526_cast_fp16 = reshape(shape = var_525, x = value_7_cast_fp16)[name = tensor("op_526_cast_fp16")]; tensor attn_7_transpose_x_0 = const()[name = tensor("attn_7_transpose_x_0"), val = tensor(false)]; tensor attn_7_transpose_y_0 = const()[name = tensor("attn_7_transpose_y_0"), val = tensor(true)]; tensor attn_7_cast_fp16 = matmul(transpose_x = attn_7_transpose_x_0, transpose_y = attn_7_transpose_y_0, x = var_526_cast_fp16, y = obj_27_cast_fp16)[name = tensor("attn_7_cast_fp16")]; tensor var_529 = const()[name = tensor("op_529"), val = tensor([1, 1024, 1, 1])]; tensor input_13_cast_fp16 = reshape(shape = var_529, x = attn_7_cast_fp16)[name = tensor("input_13_cast_fp16")]; tensor obj_25_pad_type_0 = const()[name = tensor("obj_25_pad_type_0"), val = tensor("valid")]; tensor obj_25_strides_0 = const()[name = tensor("obj_25_strides_0"), val = tensor([1, 1])]; tensor obj_25_pad_0 = const()[name = tensor("obj_25_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_25_dilations_0 = const()[name = tensor("obj_25_dilations_0"), val = tensor([1, 1])]; tensor obj_25_groups_0 = const()[name = tensor("obj_25_groups_0"), val = tensor(1)]; tensor layers_1_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(126598976))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127385472))), name = tensor("layers_1_encoder_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_1_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_1_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127385664)))]; tensor obj_25_cast_fp16 = conv(bias = layers_1_encoder_attn_o_proj_bias_to_fp16, dilations = obj_25_dilations_0, groups = obj_25_groups_0, pad = obj_25_pad_0, pad_type = obj_25_pad_type_0, strides = obj_25_strides_0, weight = layers_1_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_13_cast_fp16)[name = tensor("obj_25_cast_fp16")]; tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_25_cast_fp16)[name = tensor("inputs_11_cast_fp16")]; tensor out_11_axes_0 = const()[name = tensor("out_11_axes_0"), val = tensor([1])]; tensor var_547_to_fp16 = const()[name = tensor("op_547_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_547_to_fp16, x = inputs_11_cast_fp16)[name = tensor("out_11_cast_fp16")]; tensor input_15_gamma_0_to_fp16 = const()[name = tensor("input_15_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127387776)))]; tensor input_15_beta_0_to_fp16 = const()[name = tensor("input_15_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127389888)))]; tensor input_15_epsilon_0_to_fp16 = const()[name = tensor("input_15_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_15_cast_fp16 = batch_norm(beta = input_15_beta_0_to_fp16, epsilon = input_15_epsilon_0_to_fp16, gamma = input_15_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_11_cast_fp16)[name = tensor("input_15_cast_fp16")]; tensor input_17_pad_type_0 = const()[name = tensor("input_17_pad_type_0"), val = tensor("valid")]; tensor input_17_strides_0 = const()[name = tensor("input_17_strides_0"), val = tensor([1, 1])]; tensor input_17_pad_0 = const()[name = tensor("input_17_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_17_dilations_0 = const()[name = tensor("input_17_dilations_0"), val = tensor([1, 1])]; tensor input_17_groups_0 = const()[name = tensor("input_17_groups_0"), val = tensor(1)]; tensor layers_1_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127392000))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130537792))), name = tensor("layers_1_fc1_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; tensor layers_1_fc1_bias_to_fp16 = const()[name = tensor("layers_1_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130537984)))]; tensor input_17_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_17_dilations_0, groups = input_17_groups_0, pad = input_17_pad_0, pad_type = input_17_pad_type_0, strides = input_17_strides_0, weight = layers_1_fc1_weight_to_fp16_palettized, x = input_15_cast_fp16)[name = tensor("input_17_cast_fp16")]; tensor input_19_mode_0 = const()[name = tensor("input_19_mode_0"), val = tensor("EXACT")]; tensor input_19_cast_fp16 = gelu(mode = input_19_mode_0, x = input_17_cast_fp16)[name = tensor("input_19_cast_fp16")]; tensor hidden_states_5_pad_type_0 = const()[name = tensor("hidden_states_5_pad_type_0"), val = tensor("valid")]; tensor hidden_states_5_strides_0 = const()[name = tensor("hidden_states_5_strides_0"), val = tensor([1, 1])]; tensor hidden_states_5_pad_0 = const()[name = tensor("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_5_dilations_0 = const()[name = tensor("hidden_states_5_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_5_groups_0 = const()[name = tensor("hidden_states_5_groups_0"), val = tensor(1)]; tensor layers_1_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130546240))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133692032))), name = tensor("layers_1_fc2_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; tensor layers_1_fc2_bias_to_fp16 = const()[name = tensor("layers_1_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133692224)))]; tensor hidden_states_5_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_1_fc2_weight_to_fp16_palettized, x = input_19_cast_fp16)[name = tensor("hidden_states_5_cast_fp16")]; tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_5_cast_fp16)[name = tensor("inputs_13_cast_fp16")]; tensor var_582 = const()[name = tensor("op_582"), val = tensor(3)]; tensor out_13_axes_0 = const()[name = tensor("out_13_axes_0"), val = tensor([1])]; tensor var_607_to_fp16 = const()[name = tensor("op_607_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_607_to_fp16, x = inputs_13_cast_fp16)[name = tensor("out_13_cast_fp16")]; tensor obj_29_gamma_0_to_fp16 = const()[name = tensor("obj_29_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133694336)))]; tensor obj_29_beta_0_to_fp16 = const()[name = tensor("obj_29_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133696448)))]; tensor obj_29_epsilon_0_to_fp16 = const()[name = tensor("obj_29_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_29_cast_fp16 = batch_norm(beta = obj_29_beta_0_to_fp16, epsilon = obj_29_epsilon_0_to_fp16, gamma = obj_29_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_13_cast_fp16)[name = tensor("obj_29_cast_fp16")]; tensor query_9_pad_type_0 = const()[name = tensor("query_9_pad_type_0"), val = tensor("valid")]; tensor query_9_strides_0 = const()[name = tensor("query_9_strides_0"), val = tensor([1, 1])]; tensor query_9_pad_0 = const()[name = tensor("query_9_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_9_dilations_0 = const()[name = tensor("query_9_dilations_0"), val = tensor([1, 1])]; tensor query_9_groups_0 = const()[name = tensor("query_9_groups_0"), val = tensor(1)]; tensor layers_2_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133698560))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134485056))), name = tensor("layers_2_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134485248)))]; tensor query_9_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = query_9_dilations_0, groups = query_9_groups_0, pad = query_9_pad_0, pad_type = query_9_pad_type_0, strides = query_9_strides_0, weight = layers_2_self_attn_q_proj_weight_to_fp16_palettized, x = obj_29_cast_fp16)[name = tensor("query_9_cast_fp16")]; tensor current_key_5_pad_type_0 = const()[name = tensor("current_key_5_pad_type_0"), val = tensor("valid")]; tensor current_key_5_strides_0 = const()[name = tensor("current_key_5_strides_0"), val = tensor([1, 1])]; tensor current_key_5_pad_0 = const()[name = tensor("current_key_5_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_5_dilations_0 = const()[name = tensor("current_key_5_dilations_0"), val = tensor([1, 1])]; tensor current_key_5_groups_0 = const()[name = tensor("current_key_5_groups_0"), val = tensor(1)]; tensor layers_2_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134487360))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135273856))), name = tensor("layers_2_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor current_key_5_cast_fp16 = conv(dilations = current_key_5_dilations_0, groups = current_key_5_groups_0, pad = current_key_5_pad_0, pad_type = current_key_5_pad_type_0, strides = current_key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16_palettized, x = obj_29_cast_fp16)[name = tensor("current_key_5_cast_fp16")]; tensor current_value_5_pad_type_0 = const()[name = tensor("current_value_5_pad_type_0"), val = tensor("valid")]; tensor current_value_5_strides_0 = const()[name = tensor("current_value_5_strides_0"), val = tensor([1, 1])]; tensor current_value_5_pad_0 = const()[name = tensor("current_value_5_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_5_dilations_0 = const()[name = tensor("current_value_5_dilations_0"), val = tensor([1, 1])]; tensor current_value_5_groups_0 = const()[name = tensor("current_value_5_groups_0"), val = tensor(1)]; tensor layers_2_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135274048))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136060544))), name = tensor("layers_2_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136060736)))]; tensor current_value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = current_value_5_dilations_0, groups = current_value_5_groups_0, pad = current_value_5_pad_0, pad_type = current_value_5_pad_type_0, strides = current_value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16_palettized, x = obj_29_cast_fp16)[name = tensor("current_value_5_cast_fp16")]; tensor var_646_cast_fp16 = mul(x = var_87_cast_fp16_2, y = var_207_cast_fp16)[name = tensor("op_646_cast_fp16")]; tensor var_647_cast_fp16 = mul(x = current_key_5_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_647_cast_fp16")]; tensor key_9_cast_fp16 = add(x = var_646_cast_fp16, y = var_647_cast_fp16)[name = tensor("key_9_cast_fp16")]; tensor var_650_cast_fp16 = mul(x = var_114_cast_fp16_2, y = var_207_cast_fp16)[name = tensor("op_650_cast_fp16")]; tensor var_651_cast_fp16 = mul(x = current_value_5_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_651_cast_fp16")]; tensor value_9_cast_fp16 = add(x = var_650_cast_fp16, y = var_651_cast_fp16)[name = tensor("value_9_cast_fp16")]; tensor var_655 = const()[name = tensor("op_655"), val = tensor([1, 16, 64, 1])]; tensor mh_q_9_cast_fp16 = reshape(shape = var_655, x = query_9_cast_fp16)[name = tensor("mh_q_9_cast_fp16")]; tensor var_657_to_fp16 = const()[name = tensor("op_657_to_fp16"), val = tensor(0x1p-3)]; tensor var_658_cast_fp16 = mul(x = mh_q_9_cast_fp16, y = var_657_to_fp16)[name = tensor("op_658_cast_fp16")]; tensor var_661 = const()[name = tensor("op_661"), val = tensor([1, 16, 64, 448])]; tensor var_662_cast_fp16 = reshape(shape = var_661, x = key_9_cast_fp16)[name = tensor("op_662_cast_fp16")]; tensor mh_w_13_transpose_x_0 = const()[name = tensor("mh_w_13_transpose_x_0"), val = tensor(true)]; tensor mh_w_13_transpose_y_0 = const()[name = tensor("mh_w_13_transpose_y_0"), val = tensor(false)]; tensor mh_w_13_cast_fp16 = matmul(transpose_x = mh_w_13_transpose_x_0, transpose_y = mh_w_13_transpose_y_0, x = var_658_cast_fp16, y = var_662_cast_fp16)[name = tensor("mh_w_13_cast_fp16")]; tensor mh_w_15_cast_fp16 = add(x = mh_w_13_cast_fp16, y = var_229_cast_fp16)[name = tensor("mh_w_15_cast_fp16")]; tensor var_670_cast_fp16 = softmax(axis = var_582, x = mh_w_15_cast_fp16)[name = tensor("op_670_cast_fp16")]; tensor var_671 = const()[name = tensor("op_671"), val = tensor([1, 16, 64, 448])]; tensor var_672_cast_fp16 = reshape(shape = var_671, x = value_9_cast_fp16)[name = tensor("op_672_cast_fp16")]; tensor attn_9_transpose_x_0 = const()[name = tensor("attn_9_transpose_x_0"), val = tensor(false)]; tensor attn_9_transpose_y_0 = const()[name = tensor("attn_9_transpose_y_0"), val = tensor(true)]; tensor attn_9_cast_fp16 = matmul(transpose_x = attn_9_transpose_x_0, transpose_y = attn_9_transpose_y_0, x = var_672_cast_fp16, y = var_670_cast_fp16)[name = tensor("attn_9_cast_fp16")]; tensor var_675 = const()[name = tensor("op_675"), val = tensor([1, 1024, 1, 1])]; tensor input_21_cast_fp16 = reshape(shape = var_675, x = attn_9_cast_fp16)[name = tensor("input_21_cast_fp16")]; tensor obj_35_pad_type_0 = const()[name = tensor("obj_35_pad_type_0"), val = tensor("valid")]; tensor obj_35_strides_0 = const()[name = tensor("obj_35_strides_0"), val = tensor([1, 1])]; tensor obj_35_pad_0 = const()[name = tensor("obj_35_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_35_dilations_0 = const()[name = tensor("obj_35_dilations_0"), val = tensor([1, 1])]; tensor obj_35_groups_0 = const()[name = tensor("obj_35_groups_0"), val = tensor(1)]; tensor layers_2_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136062848))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136849344))), name = tensor("layers_2_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136849536)))]; tensor obj_35_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_35_dilations_0, groups = obj_35_groups_0, pad = obj_35_pad_0, pad_type = obj_35_pad_type_0, strides = obj_35_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16_palettized, x = input_21_cast_fp16)[name = tensor("obj_35_cast_fp16")]; tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_35_cast_fp16)[name = tensor("inputs_15_cast_fp16")]; tensor out_15_axes_0 = const()[name = tensor("out_15_axes_0"), val = tensor([1])]; tensor var_697_to_fp16 = const()[name = tensor("op_697_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_697_to_fp16, x = inputs_15_cast_fp16)[name = tensor("out_15_cast_fp16")]; tensor obj_37_gamma_0_to_fp16 = const()[name = tensor("obj_37_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136851648)))]; tensor obj_37_beta_0_to_fp16 = const()[name = tensor("obj_37_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136853760)))]; tensor obj_37_epsilon_0_to_fp16 = const()[name = tensor("obj_37_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_37_cast_fp16 = batch_norm(beta = obj_37_beta_0_to_fp16, epsilon = obj_37_epsilon_0_to_fp16, gamma = obj_37_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_15_cast_fp16)[name = tensor("obj_37_cast_fp16")]; tensor query_11_pad_type_0 = const()[name = tensor("query_11_pad_type_0"), val = tensor("valid")]; tensor query_11_strides_0 = const()[name = tensor("query_11_strides_0"), val = tensor([1, 1])]; tensor query_11_pad_0 = const()[name = tensor("query_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_11_dilations_0 = const()[name = tensor("query_11_dilations_0"), val = tensor([1, 1])]; tensor query_11_groups_0 = const()[name = tensor("query_11_groups_0"), val = tensor(1)]; tensor layers_2_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136855872))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137642368))), name = tensor("layers_2_encoder_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_2_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_2_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137642560)))]; tensor query_11_cast_fp16 = conv(bias = layers_2_encoder_attn_q_proj_bias_to_fp16, dilations = query_11_dilations_0, groups = query_11_groups_0, pad = query_11_pad_0, pad_type = query_11_pad_type_0, strides = query_11_strides_0, weight = layers_2_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_37_cast_fp16)[name = tensor("query_11_cast_fp16")]; tensor key_11_pad_type_0 = const()[name = tensor("key_11_pad_type_0"), val = tensor("valid")]; tensor key_11_strides_0 = const()[name = tensor("key_11_strides_0"), val = tensor([1, 1])]; tensor key_11_pad_0 = const()[name = tensor("key_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_11_dilations_0 = const()[name = tensor("key_11_dilations_0"), val = tensor([1, 1])]; tensor key_11_groups_0 = const()[name = tensor("key_11_groups_0"), val = tensor(1)]; tensor layers_2_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137644672))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138431168))), name = tensor("layers_2_encoder_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor key_11_cast_fp16 = conv(dilations = key_11_dilations_0, groups = key_11_groups_0, pad = key_11_pad_0, pad_type = key_11_pad_type_0, strides = key_11_strides_0, weight = layers_2_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("key_11_cast_fp16")]; tensor value_11_pad_type_0 = const()[name = tensor("value_11_pad_type_0"), val = tensor("valid")]; tensor value_11_strides_0 = const()[name = tensor("value_11_strides_0"), val = tensor([1, 1])]; tensor value_11_pad_0 = const()[name = tensor("value_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_11_dilations_0 = const()[name = tensor("value_11_dilations_0"), val = tensor([1, 1])]; tensor value_11_groups_0 = const()[name = tensor("value_11_groups_0"), val = tensor(1)]; tensor layers_2_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138431360))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(139217856))), name = tensor("layers_2_encoder_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_2_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_2_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(139218048)))]; tensor value_11_cast_fp16 = conv(bias = layers_2_encoder_attn_v_proj_bias_to_fp16, dilations = value_11_dilations_0, groups = value_11_groups_0, pad = value_11_pad_0, pad_type = value_11_pad_type_0, strides = value_11_strides_0, weight = layers_2_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("value_11_cast_fp16")]; tensor var_733 = const()[name = tensor("op_733"), val = tensor([1, 16, 64, 1])]; tensor mh_q_11_cast_fp16 = reshape(shape = var_733, x = query_11_cast_fp16)[name = tensor("mh_q_11_cast_fp16")]; tensor var_735_to_fp16 = const()[name = tensor("op_735_to_fp16"), val = tensor(0x1p-3)]; tensor var_736_cast_fp16 = mul(x = mh_q_11_cast_fp16, y = var_735_to_fp16)[name = tensor("op_736_cast_fp16")]; tensor var_739 = const()[name = tensor("op_739"), val = tensor([1, 16, 64, 1500])]; tensor var_740_cast_fp16 = reshape(shape = var_739, x = key_11_cast_fp16)[name = tensor("op_740_cast_fp16")]; tensor mh_w_17_transpose_x_0 = const()[name = tensor("mh_w_17_transpose_x_0"), val = tensor(true)]; tensor mh_w_17_transpose_y_0 = const()[name = tensor("mh_w_17_transpose_y_0"), val = tensor(false)]; tensor mh_w_17_cast_fp16 = matmul(transpose_x = mh_w_17_transpose_x_0, transpose_y = mh_w_17_transpose_y_0, x = var_736_cast_fp16, y = var_740_cast_fp16)[name = tensor("mh_w_17_cast_fp16")]; tensor obj_41_cast_fp16 = softmax(axis = var_582, x = mh_w_17_cast_fp16)[name = tensor("obj_41_cast_fp16")]; tensor var_744 = const()[name = tensor("op_744"), val = tensor([1, 16, 64, 1500])]; tensor var_745_cast_fp16 = reshape(shape = var_744, x = value_11_cast_fp16)[name = tensor("op_745_cast_fp16")]; tensor attn_11_transpose_x_0 = const()[name = tensor("attn_11_transpose_x_0"), val = tensor(false)]; tensor attn_11_transpose_y_0 = const()[name = tensor("attn_11_transpose_y_0"), val = tensor(true)]; tensor attn_11_cast_fp16 = matmul(transpose_x = attn_11_transpose_x_0, transpose_y = attn_11_transpose_y_0, x = var_745_cast_fp16, y = obj_41_cast_fp16)[name = tensor("attn_11_cast_fp16")]; tensor var_748 = const()[name = tensor("op_748"), val = tensor([1, 1024, 1, 1])]; tensor input_23_cast_fp16 = reshape(shape = var_748, x = attn_11_cast_fp16)[name = tensor("input_23_cast_fp16")]; tensor obj_39_pad_type_0 = const()[name = tensor("obj_39_pad_type_0"), val = tensor("valid")]; tensor obj_39_strides_0 = const()[name = tensor("obj_39_strides_0"), val = tensor([1, 1])]; tensor obj_39_pad_0 = const()[name = tensor("obj_39_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_39_dilations_0 = const()[name = tensor("obj_39_dilations_0"), val = tensor([1, 1])]; tensor obj_39_groups_0 = const()[name = tensor("obj_39_groups_0"), val = tensor(1)]; tensor layers_2_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(139220160))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(140006656))), name = tensor("layers_2_encoder_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_2_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_2_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(140006848)))]; tensor obj_39_cast_fp16 = conv(bias = layers_2_encoder_attn_o_proj_bias_to_fp16, dilations = obj_39_dilations_0, groups = obj_39_groups_0, pad = obj_39_pad_0, pad_type = obj_39_pad_type_0, strides = obj_39_strides_0, weight = layers_2_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_23_cast_fp16)[name = tensor("obj_39_cast_fp16")]; tensor inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = obj_39_cast_fp16)[name = tensor("inputs_17_cast_fp16")]; tensor out_17_axes_0 = const()[name = tensor("out_17_axes_0"), val = tensor([1])]; tensor var_766_to_fp16 = const()[name = tensor("op_766_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_766_to_fp16, x = inputs_17_cast_fp16)[name = tensor("out_17_cast_fp16")]; tensor input_25_gamma_0_to_fp16 = const()[name = tensor("input_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(140008960)))]; tensor input_25_beta_0_to_fp16 = const()[name = tensor("input_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(140011072)))]; tensor input_25_epsilon_0_to_fp16 = const()[name = tensor("input_25_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_25_cast_fp16 = batch_norm(beta = input_25_beta_0_to_fp16, epsilon = input_25_epsilon_0_to_fp16, gamma = input_25_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_17_cast_fp16)[name = tensor("input_25_cast_fp16")]; tensor input_27_pad_type_0 = const()[name = tensor("input_27_pad_type_0"), val = tensor("valid")]; tensor input_27_strides_0 = const()[name = tensor("input_27_strides_0"), val = tensor([1, 1])]; tensor input_27_pad_0 = const()[name = tensor("input_27_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_27_dilations_0 = const()[name = tensor("input_27_dilations_0"), val = tensor([1, 1])]; tensor input_27_groups_0 = const()[name = tensor("input_27_groups_0"), val = tensor(1)]; tensor layers_2_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(140013184))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(143158976))), name = tensor("layers_2_fc1_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; tensor layers_2_fc1_bias_to_fp16 = const()[name = tensor("layers_2_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(143159168)))]; tensor input_27_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_27_dilations_0, groups = input_27_groups_0, pad = input_27_pad_0, pad_type = input_27_pad_type_0, strides = input_27_strides_0, weight = layers_2_fc1_weight_to_fp16_palettized, x = input_25_cast_fp16)[name = tensor("input_27_cast_fp16")]; tensor input_29_mode_0 = const()[name = tensor("input_29_mode_0"), val = tensor("EXACT")]; tensor input_29_cast_fp16 = gelu(mode = input_29_mode_0, x = input_27_cast_fp16)[name = tensor("input_29_cast_fp16")]; tensor hidden_states_7_pad_type_0 = const()[name = tensor("hidden_states_7_pad_type_0"), val = tensor("valid")]; tensor hidden_states_7_strides_0 = const()[name = tensor("hidden_states_7_strides_0"), val = tensor([1, 1])]; tensor hidden_states_7_pad_0 = const()[name = tensor("hidden_states_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_7_dilations_0 = const()[name = tensor("hidden_states_7_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_7_groups_0 = const()[name = tensor("hidden_states_7_groups_0"), val = tensor(1)]; tensor layers_2_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(143167424))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(146313216))), name = tensor("layers_2_fc2_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; tensor layers_2_fc2_bias_to_fp16 = const()[name = tensor("layers_2_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(146313408)))]; tensor hidden_states_7_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = layers_2_fc2_weight_to_fp16_palettized, x = input_29_cast_fp16)[name = tensor("hidden_states_7_cast_fp16")]; tensor inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = hidden_states_7_cast_fp16)[name = tensor("inputs_19_cast_fp16")]; tensor var_801 = const()[name = tensor("op_801"), val = tensor(3)]; tensor out_19_axes_0 = const()[name = tensor("out_19_axes_0"), val = tensor([1])]; tensor var_826_to_fp16 = const()[name = tensor("op_826_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_826_to_fp16, x = inputs_19_cast_fp16)[name = tensor("out_19_cast_fp16")]; tensor obj_43_gamma_0_to_fp16 = const()[name = tensor("obj_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(146315520)))]; tensor obj_43_beta_0_to_fp16 = const()[name = tensor("obj_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(146317632)))]; tensor obj_43_epsilon_0_to_fp16 = const()[name = tensor("obj_43_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_43_cast_fp16 = batch_norm(beta = obj_43_beta_0_to_fp16, epsilon = obj_43_epsilon_0_to_fp16, gamma = obj_43_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_19_cast_fp16)[name = tensor("obj_43_cast_fp16")]; tensor query_13_pad_type_0 = const()[name = tensor("query_13_pad_type_0"), val = tensor("valid")]; tensor query_13_strides_0 = const()[name = tensor("query_13_strides_0"), val = tensor([1, 1])]; tensor query_13_pad_0 = const()[name = tensor("query_13_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_13_dilations_0 = const()[name = tensor("query_13_dilations_0"), val = tensor([1, 1])]; tensor query_13_groups_0 = const()[name = tensor("query_13_groups_0"), val = tensor(1)]; tensor layers_3_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(146319744))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(147106240))), name = tensor("layers_3_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(147106432)))]; tensor query_13_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = query_13_dilations_0, groups = query_13_groups_0, pad = query_13_pad_0, pad_type = query_13_pad_type_0, strides = query_13_strides_0, weight = layers_3_self_attn_q_proj_weight_to_fp16_palettized, x = obj_43_cast_fp16)[name = tensor("query_13_cast_fp16")]; tensor current_key_7_pad_type_0 = const()[name = tensor("current_key_7_pad_type_0"), val = tensor("valid")]; tensor current_key_7_strides_0 = const()[name = tensor("current_key_7_strides_0"), val = tensor([1, 1])]; tensor current_key_7_pad_0 = const()[name = tensor("current_key_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_7_dilations_0 = const()[name = tensor("current_key_7_dilations_0"), val = tensor([1, 1])]; tensor current_key_7_groups_0 = const()[name = tensor("current_key_7_groups_0"), val = tensor(1)]; tensor layers_3_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(147108544))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(147895040))), name = tensor("layers_3_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor current_key_7_cast_fp16 = conv(dilations = current_key_7_dilations_0, groups = current_key_7_groups_0, pad = current_key_7_pad_0, pad_type = current_key_7_pad_type_0, strides = current_key_7_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16_palettized, x = obj_43_cast_fp16)[name = tensor("current_key_7_cast_fp16")]; tensor current_value_7_pad_type_0 = const()[name = tensor("current_value_7_pad_type_0"), val = tensor("valid")]; tensor current_value_7_strides_0 = const()[name = tensor("current_value_7_strides_0"), val = tensor([1, 1])]; tensor current_value_7_pad_0 = const()[name = tensor("current_value_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_7_dilations_0 = const()[name = tensor("current_value_7_dilations_0"), val = tensor([1, 1])]; tensor current_value_7_groups_0 = const()[name = tensor("current_value_7_groups_0"), val = tensor(1)]; tensor layers_3_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(147895232))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(148681728))), name = tensor("layers_3_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(148681920)))]; tensor current_value_7_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = current_value_7_dilations_0, groups = current_value_7_groups_0, pad = current_value_7_pad_0, pad_type = current_value_7_pad_type_0, strides = current_value_7_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16_palettized, x = obj_43_cast_fp16)[name = tensor("current_value_7_cast_fp16")]; tensor var_865_cast_fp16 = mul(x = var_87_cast_fp16_3, y = var_207_cast_fp16)[name = tensor("op_865_cast_fp16")]; tensor var_866_cast_fp16 = mul(x = current_key_7_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_866_cast_fp16")]; tensor key_13_cast_fp16 = add(x = var_865_cast_fp16, y = var_866_cast_fp16)[name = tensor("key_13_cast_fp16")]; tensor var_869_cast_fp16 = mul(x = var_114_cast_fp16_3, y = var_207_cast_fp16)[name = tensor("op_869_cast_fp16")]; tensor var_870_cast_fp16 = mul(x = current_value_7_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_870_cast_fp16")]; tensor value_13_cast_fp16 = add(x = var_869_cast_fp16, y = var_870_cast_fp16)[name = tensor("value_13_cast_fp16")]; tensor var_874 = const()[name = tensor("op_874"), val = tensor([1, 16, 64, 1])]; tensor mh_q_13_cast_fp16 = reshape(shape = var_874, x = query_13_cast_fp16)[name = tensor("mh_q_13_cast_fp16")]; tensor var_876_to_fp16 = const()[name = tensor("op_876_to_fp16"), val = tensor(0x1p-3)]; tensor var_877_cast_fp16 = mul(x = mh_q_13_cast_fp16, y = var_876_to_fp16)[name = tensor("op_877_cast_fp16")]; tensor var_880 = const()[name = tensor("op_880"), val = tensor([1, 16, 64, 448])]; tensor var_881_cast_fp16 = reshape(shape = var_880, x = key_13_cast_fp16)[name = tensor("op_881_cast_fp16")]; tensor mh_w_19_transpose_x_0 = const()[name = tensor("mh_w_19_transpose_x_0"), val = tensor(true)]; tensor mh_w_19_transpose_y_0 = const()[name = tensor("mh_w_19_transpose_y_0"), val = tensor(false)]; tensor mh_w_19_cast_fp16 = matmul(transpose_x = mh_w_19_transpose_x_0, transpose_y = mh_w_19_transpose_y_0, x = var_877_cast_fp16, y = var_881_cast_fp16)[name = tensor("mh_w_19_cast_fp16")]; tensor mh_w_21_cast_fp16 = add(x = mh_w_19_cast_fp16, y = var_229_cast_fp16)[name = tensor("mh_w_21_cast_fp16")]; tensor var_889_cast_fp16 = softmax(axis = var_801, x = mh_w_21_cast_fp16)[name = tensor("op_889_cast_fp16")]; tensor var_890 = const()[name = tensor("op_890"), val = tensor([1, 16, 64, 448])]; tensor var_891_cast_fp16 = reshape(shape = var_890, x = value_13_cast_fp16)[name = tensor("op_891_cast_fp16")]; tensor attn_13_transpose_x_0 = const()[name = tensor("attn_13_transpose_x_0"), val = tensor(false)]; tensor attn_13_transpose_y_0 = const()[name = tensor("attn_13_transpose_y_0"), val = tensor(true)]; tensor attn_13_cast_fp16 = matmul(transpose_x = attn_13_transpose_x_0, transpose_y = attn_13_transpose_y_0, x = var_891_cast_fp16, y = var_889_cast_fp16)[name = tensor("attn_13_cast_fp16")]; tensor var_894 = const()[name = tensor("op_894"), val = tensor([1, 1024, 1, 1])]; tensor input_31_cast_fp16 = reshape(shape = var_894, x = attn_13_cast_fp16)[name = tensor("input_31_cast_fp16")]; tensor obj_49_pad_type_0 = const()[name = tensor("obj_49_pad_type_0"), val = tensor("valid")]; tensor obj_49_strides_0 = const()[name = tensor("obj_49_strides_0"), val = tensor([1, 1])]; tensor obj_49_pad_0 = const()[name = tensor("obj_49_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_49_dilations_0 = const()[name = tensor("obj_49_dilations_0"), val = tensor([1, 1])]; tensor obj_49_groups_0 = const()[name = tensor("obj_49_groups_0"), val = tensor(1)]; tensor layers_3_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(148684032))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(149470528))), name = tensor("layers_3_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(149470720)))]; tensor obj_49_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_49_dilations_0, groups = obj_49_groups_0, pad = obj_49_pad_0, pad_type = obj_49_pad_type_0, strides = obj_49_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16_palettized, x = input_31_cast_fp16)[name = tensor("obj_49_cast_fp16")]; tensor inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = obj_49_cast_fp16)[name = tensor("inputs_21_cast_fp16")]; tensor out_21_axes_0 = const()[name = tensor("out_21_axes_0"), val = tensor([1])]; tensor var_916_to_fp16 = const()[name = tensor("op_916_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_916_to_fp16, x = inputs_21_cast_fp16)[name = tensor("out_21_cast_fp16")]; tensor obj_51_gamma_0_to_fp16 = const()[name = tensor("obj_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(149472832)))]; tensor obj_51_beta_0_to_fp16 = const()[name = tensor("obj_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(149474944)))]; tensor obj_51_epsilon_0_to_fp16 = const()[name = tensor("obj_51_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_51_cast_fp16 = batch_norm(beta = obj_51_beta_0_to_fp16, epsilon = obj_51_epsilon_0_to_fp16, gamma = obj_51_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_21_cast_fp16)[name = tensor("obj_51_cast_fp16")]; tensor query_15_pad_type_0 = const()[name = tensor("query_15_pad_type_0"), val = tensor("valid")]; tensor query_15_strides_0 = const()[name = tensor("query_15_strides_0"), val = tensor([1, 1])]; tensor query_15_pad_0 = const()[name = tensor("query_15_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_15_dilations_0 = const()[name = tensor("query_15_dilations_0"), val = tensor([1, 1])]; tensor query_15_groups_0 = const()[name = tensor("query_15_groups_0"), val = tensor(1)]; tensor layers_3_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(149477056))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(150263552))), name = tensor("layers_3_encoder_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_3_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_3_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(150263744)))]; tensor query_15_cast_fp16 = conv(bias = layers_3_encoder_attn_q_proj_bias_to_fp16, dilations = query_15_dilations_0, groups = query_15_groups_0, pad = query_15_pad_0, pad_type = query_15_pad_type_0, strides = query_15_strides_0, weight = layers_3_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_51_cast_fp16)[name = tensor("query_15_cast_fp16")]; tensor key_15_pad_type_0 = const()[name = tensor("key_15_pad_type_0"), val = tensor("valid")]; tensor key_15_strides_0 = const()[name = tensor("key_15_strides_0"), val = tensor([1, 1])]; tensor key_15_pad_0 = const()[name = tensor("key_15_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_15_dilations_0 = const()[name = tensor("key_15_dilations_0"), val = tensor([1, 1])]; tensor key_15_groups_0 = const()[name = tensor("key_15_groups_0"), val = tensor(1)]; tensor layers_3_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(150265856))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151052352))), name = tensor("layers_3_encoder_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor key_15_cast_fp16 = conv(dilations = key_15_dilations_0, groups = key_15_groups_0, pad = key_15_pad_0, pad_type = key_15_pad_type_0, strides = key_15_strides_0, weight = layers_3_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("key_15_cast_fp16")]; tensor value_15_pad_type_0 = const()[name = tensor("value_15_pad_type_0"), val = tensor("valid")]; tensor value_15_strides_0 = const()[name = tensor("value_15_strides_0"), val = tensor([1, 1])]; tensor value_15_pad_0 = const()[name = tensor("value_15_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_15_dilations_0 = const()[name = tensor("value_15_dilations_0"), val = tensor([1, 1])]; tensor value_15_groups_0 = const()[name = tensor("value_15_groups_0"), val = tensor(1)]; tensor layers_3_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151052544))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151839040))), name = tensor("layers_3_encoder_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_3_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_3_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151839232)))]; tensor value_15_cast_fp16 = conv(bias = layers_3_encoder_attn_v_proj_bias_to_fp16, dilations = value_15_dilations_0, groups = value_15_groups_0, pad = value_15_pad_0, pad_type = value_15_pad_type_0, strides = value_15_strides_0, weight = layers_3_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("value_15_cast_fp16")]; tensor var_952 = const()[name = tensor("op_952"), val = tensor([1, 16, 64, 1])]; tensor mh_q_15_cast_fp16 = reshape(shape = var_952, x = query_15_cast_fp16)[name = tensor("mh_q_15_cast_fp16")]; tensor var_954_to_fp16 = const()[name = tensor("op_954_to_fp16"), val = tensor(0x1p-3)]; tensor var_955_cast_fp16 = mul(x = mh_q_15_cast_fp16, y = var_954_to_fp16)[name = tensor("op_955_cast_fp16")]; tensor var_958 = const()[name = tensor("op_958"), val = tensor([1, 16, 64, 1500])]; tensor var_959_cast_fp16 = reshape(shape = var_958, x = key_15_cast_fp16)[name = tensor("op_959_cast_fp16")]; tensor mh_w_23_transpose_x_0 = const()[name = tensor("mh_w_23_transpose_x_0"), val = tensor(true)]; tensor mh_w_23_transpose_y_0 = const()[name = tensor("mh_w_23_transpose_y_0"), val = tensor(false)]; tensor mh_w_23_cast_fp16 = matmul(transpose_x = mh_w_23_transpose_x_0, transpose_y = mh_w_23_transpose_y_0, x = var_955_cast_fp16, y = var_959_cast_fp16)[name = tensor("mh_w_23_cast_fp16")]; tensor obj_55_cast_fp16 = softmax(axis = var_801, x = mh_w_23_cast_fp16)[name = tensor("obj_55_cast_fp16")]; tensor var_963 = const()[name = tensor("op_963"), val = tensor([1, 16, 64, 1500])]; tensor var_964_cast_fp16 = reshape(shape = var_963, x = value_15_cast_fp16)[name = tensor("op_964_cast_fp16")]; tensor attn_15_transpose_x_0 = const()[name = tensor("attn_15_transpose_x_0"), val = tensor(false)]; tensor attn_15_transpose_y_0 = const()[name = tensor("attn_15_transpose_y_0"), val = tensor(true)]; tensor attn_15_cast_fp16 = matmul(transpose_x = attn_15_transpose_x_0, transpose_y = attn_15_transpose_y_0, x = var_964_cast_fp16, y = obj_55_cast_fp16)[name = tensor("attn_15_cast_fp16")]; tensor var_967 = const()[name = tensor("op_967"), val = tensor([1, 1024, 1, 1])]; tensor input_33_cast_fp16 = reshape(shape = var_967, x = attn_15_cast_fp16)[name = tensor("input_33_cast_fp16")]; tensor obj_53_pad_type_0 = const()[name = tensor("obj_53_pad_type_0"), val = tensor("valid")]; tensor obj_53_strides_0 = const()[name = tensor("obj_53_strides_0"), val = tensor([1, 1])]; tensor obj_53_pad_0 = const()[name = tensor("obj_53_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_53_dilations_0 = const()[name = tensor("obj_53_dilations_0"), val = tensor([1, 1])]; tensor obj_53_groups_0 = const()[name = tensor("obj_53_groups_0"), val = tensor(1)]; tensor layers_3_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151841344))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152627840))), name = tensor("layers_3_encoder_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_3_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_3_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152628032)))]; tensor obj_53_cast_fp16 = conv(bias = layers_3_encoder_attn_o_proj_bias_to_fp16, dilations = obj_53_dilations_0, groups = obj_53_groups_0, pad = obj_53_pad_0, pad_type = obj_53_pad_type_0, strides = obj_53_strides_0, weight = layers_3_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_33_cast_fp16)[name = tensor("obj_53_cast_fp16")]; tensor inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_53_cast_fp16)[name = tensor("inputs_23_cast_fp16")]; tensor out_23_axes_0 = const()[name = tensor("out_23_axes_0"), val = tensor([1])]; tensor var_985_to_fp16 = const()[name = tensor("op_985_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_985_to_fp16, x = inputs_23_cast_fp16)[name = tensor("out_23_cast_fp16")]; tensor input_35_gamma_0_to_fp16 = const()[name = tensor("input_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152630144)))]; tensor input_35_beta_0_to_fp16 = const()[name = tensor("input_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152632256)))]; tensor input_35_epsilon_0_to_fp16 = const()[name = tensor("input_35_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_35_cast_fp16 = batch_norm(beta = input_35_beta_0_to_fp16, epsilon = input_35_epsilon_0_to_fp16, gamma = input_35_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_23_cast_fp16)[name = tensor("input_35_cast_fp16")]; tensor input_37_pad_type_0 = const()[name = tensor("input_37_pad_type_0"), val = tensor("valid")]; tensor input_37_strides_0 = const()[name = tensor("input_37_strides_0"), val = tensor([1, 1])]; tensor input_37_pad_0 = const()[name = tensor("input_37_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_37_dilations_0 = const()[name = tensor("input_37_dilations_0"), val = tensor([1, 1])]; tensor input_37_groups_0 = const()[name = tensor("input_37_groups_0"), val = tensor(1)]; tensor layers_3_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152634368))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(155780160))), name = tensor("layers_3_fc1_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; tensor layers_3_fc1_bias_to_fp16 = const()[name = tensor("layers_3_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(155780352)))]; tensor input_37_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = input_37_dilations_0, groups = input_37_groups_0, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = input_37_strides_0, weight = layers_3_fc1_weight_to_fp16_palettized, x = input_35_cast_fp16)[name = tensor("input_37_cast_fp16")]; tensor input_39_mode_0 = const()[name = tensor("input_39_mode_0"), val = tensor("EXACT")]; tensor input_39_cast_fp16 = gelu(mode = input_39_mode_0, x = input_37_cast_fp16)[name = tensor("input_39_cast_fp16")]; tensor hidden_states_9_pad_type_0 = const()[name = tensor("hidden_states_9_pad_type_0"), val = tensor("valid")]; tensor hidden_states_9_strides_0 = const()[name = tensor("hidden_states_9_strides_0"), val = tensor([1, 1])]; tensor hidden_states_9_pad_0 = const()[name = tensor("hidden_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_9_dilations_0 = const()[name = tensor("hidden_states_9_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_9_groups_0 = const()[name = tensor("hidden_states_9_groups_0"), val = tensor(1)]; tensor layers_3_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(155788608))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158934400))), name = tensor("layers_3_fc2_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; tensor layers_3_fc2_bias_to_fp16 = const()[name = tensor("layers_3_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158934592)))]; tensor hidden_states_9_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = layers_3_fc2_weight_to_fp16_palettized, x = input_39_cast_fp16)[name = tensor("hidden_states_9_cast_fp16")]; tensor inputs_25_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_9_cast_fp16)[name = tensor("inputs_25_cast_fp16")]; tensor var_1020 = const()[name = tensor("op_1020"), val = tensor(3)]; tensor out_25_axes_0 = const()[name = tensor("out_25_axes_0"), val = tensor([1])]; tensor var_1045_to_fp16 = const()[name = tensor("op_1045_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_25_cast_fp16 = layer_norm(axes = out_25_axes_0, epsilon = var_1045_to_fp16, x = inputs_25_cast_fp16)[name = tensor("out_25_cast_fp16")]; tensor obj_57_gamma_0_to_fp16 = const()[name = tensor("obj_57_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158936704)))]; tensor obj_57_beta_0_to_fp16 = const()[name = tensor("obj_57_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158938816)))]; tensor obj_57_epsilon_0_to_fp16 = const()[name = tensor("obj_57_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_57_cast_fp16 = batch_norm(beta = obj_57_beta_0_to_fp16, epsilon = obj_57_epsilon_0_to_fp16, gamma = obj_57_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_25_cast_fp16)[name = tensor("obj_57_cast_fp16")]; tensor query_17_pad_type_0 = const()[name = tensor("query_17_pad_type_0"), val = tensor("valid")]; tensor query_17_strides_0 = const()[name = tensor("query_17_strides_0"), val = tensor([1, 1])]; tensor query_17_pad_0 = const()[name = tensor("query_17_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_17_dilations_0 = const()[name = tensor("query_17_dilations_0"), val = tensor([1, 1])]; tensor query_17_groups_0 = const()[name = tensor("query_17_groups_0"), val = tensor(1)]; tensor layers_4_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158940928))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(159727424))), name = tensor("layers_4_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(159727616)))]; tensor query_17_cast_fp16 = conv(bias = layers_4_self_attn_q_proj_bias_to_fp16, dilations = query_17_dilations_0, groups = query_17_groups_0, pad = query_17_pad_0, pad_type = query_17_pad_type_0, strides = query_17_strides_0, weight = layers_4_self_attn_q_proj_weight_to_fp16_palettized, x = obj_57_cast_fp16)[name = tensor("query_17_cast_fp16")]; tensor current_key_9_pad_type_0 = const()[name = tensor("current_key_9_pad_type_0"), val = tensor("valid")]; tensor current_key_9_strides_0 = const()[name = tensor("current_key_9_strides_0"), val = tensor([1, 1])]; tensor current_key_9_pad_0 = const()[name = tensor("current_key_9_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_9_dilations_0 = const()[name = tensor("current_key_9_dilations_0"), val = tensor([1, 1])]; tensor current_key_9_groups_0 = const()[name = tensor("current_key_9_groups_0"), val = tensor(1)]; tensor layers_4_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(159729728))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(160516224))), name = tensor("layers_4_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor current_key_9_cast_fp16 = conv(dilations = current_key_9_dilations_0, groups = current_key_9_groups_0, pad = current_key_9_pad_0, pad_type = current_key_9_pad_type_0, strides = current_key_9_strides_0, weight = layers_4_self_attn_k_proj_weight_to_fp16_palettized, x = obj_57_cast_fp16)[name = tensor("current_key_9_cast_fp16")]; tensor current_value_9_pad_type_0 = const()[name = tensor("current_value_9_pad_type_0"), val = tensor("valid")]; tensor current_value_9_strides_0 = const()[name = tensor("current_value_9_strides_0"), val = tensor([1, 1])]; tensor current_value_9_pad_0 = const()[name = tensor("current_value_9_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_9_dilations_0 = const()[name = tensor("current_value_9_dilations_0"), val = tensor([1, 1])]; tensor current_value_9_groups_0 = const()[name = tensor("current_value_9_groups_0"), val = tensor(1)]; tensor layers_4_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(160516416))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161302912))), name = tensor("layers_4_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161303104)))]; tensor current_value_9_cast_fp16 = conv(bias = layers_4_self_attn_v_proj_bias_to_fp16, dilations = current_value_9_dilations_0, groups = current_value_9_groups_0, pad = current_value_9_pad_0, pad_type = current_value_9_pad_type_0, strides = current_value_9_strides_0, weight = layers_4_self_attn_v_proj_weight_to_fp16_palettized, x = obj_57_cast_fp16)[name = tensor("current_value_9_cast_fp16")]; tensor var_1084_cast_fp16 = mul(x = var_87_cast_fp16_4, y = var_207_cast_fp16)[name = tensor("op_1084_cast_fp16")]; tensor var_1085_cast_fp16 = mul(x = current_key_9_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_1085_cast_fp16")]; tensor key_17_cast_fp16 = add(x = var_1084_cast_fp16, y = var_1085_cast_fp16)[name = tensor("key_17_cast_fp16")]; tensor var_1088_cast_fp16 = mul(x = var_114_cast_fp16_4, y = var_207_cast_fp16)[name = tensor("op_1088_cast_fp16")]; tensor var_1089_cast_fp16 = mul(x = current_value_9_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_1089_cast_fp16")]; tensor value_17_cast_fp16 = add(x = var_1088_cast_fp16, y = var_1089_cast_fp16)[name = tensor("value_17_cast_fp16")]; tensor var_1093 = const()[name = tensor("op_1093"), val = tensor([1, 16, 64, 1])]; tensor mh_q_17_cast_fp16 = reshape(shape = var_1093, x = query_17_cast_fp16)[name = tensor("mh_q_17_cast_fp16")]; tensor var_1095_to_fp16 = const()[name = tensor("op_1095_to_fp16"), val = tensor(0x1p-3)]; tensor var_1096_cast_fp16 = mul(x = mh_q_17_cast_fp16, y = var_1095_to_fp16)[name = tensor("op_1096_cast_fp16")]; tensor var_1099 = const()[name = tensor("op_1099"), val = tensor([1, 16, 64, 448])]; tensor var_1100_cast_fp16 = reshape(shape = var_1099, x = key_17_cast_fp16)[name = tensor("op_1100_cast_fp16")]; tensor mh_w_25_transpose_x_0 = const()[name = tensor("mh_w_25_transpose_x_0"), val = tensor(true)]; tensor mh_w_25_transpose_y_0 = const()[name = tensor("mh_w_25_transpose_y_0"), val = tensor(false)]; tensor mh_w_25_cast_fp16 = matmul(transpose_x = mh_w_25_transpose_x_0, transpose_y = mh_w_25_transpose_y_0, x = var_1096_cast_fp16, y = var_1100_cast_fp16)[name = tensor("mh_w_25_cast_fp16")]; tensor mh_w_27_cast_fp16 = add(x = mh_w_25_cast_fp16, y = var_229_cast_fp16)[name = tensor("mh_w_27_cast_fp16")]; tensor var_1108_cast_fp16 = softmax(axis = var_1020, x = mh_w_27_cast_fp16)[name = tensor("op_1108_cast_fp16")]; tensor var_1109 = const()[name = tensor("op_1109"), val = tensor([1, 16, 64, 448])]; tensor var_1110_cast_fp16 = reshape(shape = var_1109, x = value_17_cast_fp16)[name = tensor("op_1110_cast_fp16")]; tensor attn_17_transpose_x_0 = const()[name = tensor("attn_17_transpose_x_0"), val = tensor(false)]; tensor attn_17_transpose_y_0 = const()[name = tensor("attn_17_transpose_y_0"), val = tensor(true)]; tensor attn_17_cast_fp16 = matmul(transpose_x = attn_17_transpose_x_0, transpose_y = attn_17_transpose_y_0, x = var_1110_cast_fp16, y = var_1108_cast_fp16)[name = tensor("attn_17_cast_fp16")]; tensor var_1113 = const()[name = tensor("op_1113"), val = tensor([1, 1024, 1, 1])]; tensor input_41_cast_fp16 = reshape(shape = var_1113, x = attn_17_cast_fp16)[name = tensor("input_41_cast_fp16")]; tensor obj_63_pad_type_0 = const()[name = tensor("obj_63_pad_type_0"), val = tensor("valid")]; tensor obj_63_strides_0 = const()[name = tensor("obj_63_strides_0"), val = tensor([1, 1])]; tensor obj_63_pad_0 = const()[name = tensor("obj_63_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_63_dilations_0 = const()[name = tensor("obj_63_dilations_0"), val = tensor([1, 1])]; tensor obj_63_groups_0 = const()[name = tensor("obj_63_groups_0"), val = tensor(1)]; tensor layers_4_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161305216))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162091712))), name = tensor("layers_4_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_4_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_4_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162091904)))]; tensor obj_63_cast_fp16 = conv(bias = layers_4_self_attn_o_proj_bias_to_fp16, dilations = obj_63_dilations_0, groups = obj_63_groups_0, pad = obj_63_pad_0, pad_type = obj_63_pad_type_0, strides = obj_63_strides_0, weight = layers_4_self_attn_o_proj_weight_to_fp16_palettized, x = input_41_cast_fp16)[name = tensor("obj_63_cast_fp16")]; tensor inputs_27_cast_fp16 = add(x = inputs_25_cast_fp16, y = obj_63_cast_fp16)[name = tensor("inputs_27_cast_fp16")]; tensor out_27_axes_0 = const()[name = tensor("out_27_axes_0"), val = tensor([1])]; tensor var_1135_to_fp16 = const()[name = tensor("op_1135_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_27_cast_fp16 = layer_norm(axes = out_27_axes_0, epsilon = var_1135_to_fp16, x = inputs_27_cast_fp16)[name = tensor("out_27_cast_fp16")]; tensor obj_65_gamma_0_to_fp16 = const()[name = tensor("obj_65_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162094016)))]; tensor obj_65_beta_0_to_fp16 = const()[name = tensor("obj_65_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162096128)))]; tensor obj_65_epsilon_0_to_fp16 = const()[name = tensor("obj_65_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_65_cast_fp16 = batch_norm(beta = obj_65_beta_0_to_fp16, epsilon = obj_65_epsilon_0_to_fp16, gamma = obj_65_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_27_cast_fp16)[name = tensor("obj_65_cast_fp16")]; tensor query_19_pad_type_0 = const()[name = tensor("query_19_pad_type_0"), val = tensor("valid")]; tensor query_19_strides_0 = const()[name = tensor("query_19_strides_0"), val = tensor([1, 1])]; tensor query_19_pad_0 = const()[name = tensor("query_19_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_19_dilations_0 = const()[name = tensor("query_19_dilations_0"), val = tensor([1, 1])]; tensor query_19_groups_0 = const()[name = tensor("query_19_groups_0"), val = tensor(1)]; tensor layers_4_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162098240))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162884736))), name = tensor("layers_4_encoder_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_4_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_4_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162884928)))]; tensor query_19_cast_fp16 = conv(bias = layers_4_encoder_attn_q_proj_bias_to_fp16, dilations = query_19_dilations_0, groups = query_19_groups_0, pad = query_19_pad_0, pad_type = query_19_pad_type_0, strides = query_19_strides_0, weight = layers_4_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_65_cast_fp16)[name = tensor("query_19_cast_fp16")]; tensor key_19_pad_type_0 = const()[name = tensor("key_19_pad_type_0"), val = tensor("valid")]; tensor key_19_strides_0 = const()[name = tensor("key_19_strides_0"), val = tensor([1, 1])]; tensor key_19_pad_0 = const()[name = tensor("key_19_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_19_dilations_0 = const()[name = tensor("key_19_dilations_0"), val = tensor([1, 1])]; tensor key_19_groups_0 = const()[name = tensor("key_19_groups_0"), val = tensor(1)]; tensor layers_4_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162887040))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163673536))), name = tensor("layers_4_encoder_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor key_19_cast_fp16 = conv(dilations = key_19_dilations_0, groups = key_19_groups_0, pad = key_19_pad_0, pad_type = key_19_pad_type_0, strides = key_19_strides_0, weight = layers_4_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("key_19_cast_fp16")]; tensor value_19_pad_type_0 = const()[name = tensor("value_19_pad_type_0"), val = tensor("valid")]; tensor value_19_strides_0 = const()[name = tensor("value_19_strides_0"), val = tensor([1, 1])]; tensor value_19_pad_0 = const()[name = tensor("value_19_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_19_dilations_0 = const()[name = tensor("value_19_dilations_0"), val = tensor([1, 1])]; tensor value_19_groups_0 = const()[name = tensor("value_19_groups_0"), val = tensor(1)]; tensor layers_4_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163673728))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(164460224))), name = tensor("layers_4_encoder_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_4_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_4_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(164460416)))]; tensor value_19_cast_fp16 = conv(bias = layers_4_encoder_attn_v_proj_bias_to_fp16, dilations = value_19_dilations_0, groups = value_19_groups_0, pad = value_19_pad_0, pad_type = value_19_pad_type_0, strides = value_19_strides_0, weight = layers_4_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("value_19_cast_fp16")]; tensor var_1171 = const()[name = tensor("op_1171"), val = tensor([1, 16, 64, 1])]; tensor mh_q_19_cast_fp16 = reshape(shape = var_1171, x = query_19_cast_fp16)[name = tensor("mh_q_19_cast_fp16")]; tensor var_1173_to_fp16 = const()[name = tensor("op_1173_to_fp16"), val = tensor(0x1p-3)]; tensor var_1174_cast_fp16 = mul(x = mh_q_19_cast_fp16, y = var_1173_to_fp16)[name = tensor("op_1174_cast_fp16")]; tensor var_1177 = const()[name = tensor("op_1177"), val = tensor([1, 16, 64, 1500])]; tensor var_1178_cast_fp16 = reshape(shape = var_1177, x = key_19_cast_fp16)[name = tensor("op_1178_cast_fp16")]; tensor mh_w_29_transpose_x_0 = const()[name = tensor("mh_w_29_transpose_x_0"), val = tensor(true)]; tensor mh_w_29_transpose_y_0 = const()[name = tensor("mh_w_29_transpose_y_0"), val = tensor(false)]; tensor mh_w_29_cast_fp16 = matmul(transpose_x = mh_w_29_transpose_x_0, transpose_y = mh_w_29_transpose_y_0, x = var_1174_cast_fp16, y = var_1178_cast_fp16)[name = tensor("mh_w_29_cast_fp16")]; tensor obj_69_cast_fp16 = softmax(axis = var_1020, x = mh_w_29_cast_fp16)[name = tensor("obj_69_cast_fp16")]; tensor var_1182 = const()[name = tensor("op_1182"), val = tensor([1, 16, 64, 1500])]; tensor var_1183_cast_fp16 = reshape(shape = var_1182, x = value_19_cast_fp16)[name = tensor("op_1183_cast_fp16")]; tensor attn_19_transpose_x_0 = const()[name = tensor("attn_19_transpose_x_0"), val = tensor(false)]; tensor attn_19_transpose_y_0 = const()[name = tensor("attn_19_transpose_y_0"), val = tensor(true)]; tensor attn_19_cast_fp16 = matmul(transpose_x = attn_19_transpose_x_0, transpose_y = attn_19_transpose_y_0, x = var_1183_cast_fp16, y = obj_69_cast_fp16)[name = tensor("attn_19_cast_fp16")]; tensor var_1186 = const()[name = tensor("op_1186"), val = tensor([1, 1024, 1, 1])]; tensor input_43_cast_fp16 = reshape(shape = var_1186, x = attn_19_cast_fp16)[name = tensor("input_43_cast_fp16")]; tensor obj_67_pad_type_0 = const()[name = tensor("obj_67_pad_type_0"), val = tensor("valid")]; tensor obj_67_strides_0 = const()[name = tensor("obj_67_strides_0"), val = tensor([1, 1])]; tensor obj_67_pad_0 = const()[name = tensor("obj_67_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_67_dilations_0 = const()[name = tensor("obj_67_dilations_0"), val = tensor([1, 1])]; tensor obj_67_groups_0 = const()[name = tensor("obj_67_groups_0"), val = tensor(1)]; tensor layers_4_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(164462528))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165249024))), name = tensor("layers_4_encoder_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_4_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_4_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165249216)))]; tensor obj_67_cast_fp16 = conv(bias = layers_4_encoder_attn_o_proj_bias_to_fp16, dilations = obj_67_dilations_0, groups = obj_67_groups_0, pad = obj_67_pad_0, pad_type = obj_67_pad_type_0, strides = obj_67_strides_0, weight = layers_4_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_43_cast_fp16)[name = tensor("obj_67_cast_fp16")]; tensor inputs_29_cast_fp16 = add(x = inputs_27_cast_fp16, y = obj_67_cast_fp16)[name = tensor("inputs_29_cast_fp16")]; tensor out_29_axes_0 = const()[name = tensor("out_29_axes_0"), val = tensor([1])]; tensor var_1204_to_fp16 = const()[name = tensor("op_1204_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_29_cast_fp16 = layer_norm(axes = out_29_axes_0, epsilon = var_1204_to_fp16, x = inputs_29_cast_fp16)[name = tensor("out_29_cast_fp16")]; tensor input_45_gamma_0_to_fp16 = const()[name = tensor("input_45_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165251328)))]; tensor input_45_beta_0_to_fp16 = const()[name = tensor("input_45_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165253440)))]; tensor input_45_epsilon_0_to_fp16 = const()[name = tensor("input_45_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_45_cast_fp16 = batch_norm(beta = input_45_beta_0_to_fp16, epsilon = input_45_epsilon_0_to_fp16, gamma = input_45_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_29_cast_fp16)[name = tensor("input_45_cast_fp16")]; tensor input_47_pad_type_0 = const()[name = tensor("input_47_pad_type_0"), val = tensor("valid")]; tensor input_47_strides_0 = const()[name = tensor("input_47_strides_0"), val = tensor([1, 1])]; tensor input_47_pad_0 = const()[name = tensor("input_47_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_47_dilations_0 = const()[name = tensor("input_47_dilations_0"), val = tensor([1, 1])]; tensor input_47_groups_0 = const()[name = tensor("input_47_groups_0"), val = tensor(1)]; tensor layers_4_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165255552))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(168401344))), name = tensor("layers_4_fc1_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; tensor layers_4_fc1_bias_to_fp16 = const()[name = tensor("layers_4_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(168401536)))]; tensor input_47_cast_fp16 = conv(bias = layers_4_fc1_bias_to_fp16, dilations = input_47_dilations_0, groups = input_47_groups_0, pad = input_47_pad_0, pad_type = input_47_pad_type_0, strides = input_47_strides_0, weight = layers_4_fc1_weight_to_fp16_palettized, x = input_45_cast_fp16)[name = tensor("input_47_cast_fp16")]; tensor input_49_mode_0 = const()[name = tensor("input_49_mode_0"), val = tensor("EXACT")]; tensor input_49_cast_fp16 = gelu(mode = input_49_mode_0, x = input_47_cast_fp16)[name = tensor("input_49_cast_fp16")]; tensor hidden_states_11_pad_type_0 = const()[name = tensor("hidden_states_11_pad_type_0"), val = tensor("valid")]; tensor hidden_states_11_strides_0 = const()[name = tensor("hidden_states_11_strides_0"), val = tensor([1, 1])]; tensor hidden_states_11_pad_0 = const()[name = tensor("hidden_states_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_11_dilations_0 = const()[name = tensor("hidden_states_11_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_11_groups_0 = const()[name = tensor("hidden_states_11_groups_0"), val = tensor(1)]; tensor layers_4_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(168409792))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(171555584))), name = tensor("layers_4_fc2_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; tensor layers_4_fc2_bias_to_fp16 = const()[name = tensor("layers_4_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(171555776)))]; tensor hidden_states_11_cast_fp16 = conv(bias = layers_4_fc2_bias_to_fp16, dilations = hidden_states_11_dilations_0, groups = hidden_states_11_groups_0, pad = hidden_states_11_pad_0, pad_type = hidden_states_11_pad_type_0, strides = hidden_states_11_strides_0, weight = layers_4_fc2_weight_to_fp16_palettized, x = input_49_cast_fp16)[name = tensor("hidden_states_11_cast_fp16")]; tensor inputs_31_cast_fp16 = add(x = inputs_29_cast_fp16, y = hidden_states_11_cast_fp16)[name = tensor("inputs_31_cast_fp16")]; tensor var_1239 = const()[name = tensor("op_1239"), val = tensor(3)]; tensor out_31_axes_0 = const()[name = tensor("out_31_axes_0"), val = tensor([1])]; tensor var_1264_to_fp16 = const()[name = tensor("op_1264_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_31_cast_fp16 = layer_norm(axes = out_31_axes_0, epsilon = var_1264_to_fp16, x = inputs_31_cast_fp16)[name = tensor("out_31_cast_fp16")]; tensor obj_71_gamma_0_to_fp16 = const()[name = tensor("obj_71_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(171557888)))]; tensor obj_71_beta_0_to_fp16 = const()[name = tensor("obj_71_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(171560000)))]; tensor obj_71_epsilon_0_to_fp16 = const()[name = tensor("obj_71_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_71_cast_fp16 = batch_norm(beta = obj_71_beta_0_to_fp16, epsilon = obj_71_epsilon_0_to_fp16, gamma = obj_71_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_31_cast_fp16)[name = tensor("obj_71_cast_fp16")]; tensor query_21_pad_type_0 = const()[name = tensor("query_21_pad_type_0"), val = tensor("valid")]; tensor query_21_strides_0 = const()[name = tensor("query_21_strides_0"), val = tensor([1, 1])]; tensor query_21_pad_0 = const()[name = tensor("query_21_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_21_dilations_0 = const()[name = tensor("query_21_dilations_0"), val = tensor([1, 1])]; tensor query_21_groups_0 = const()[name = tensor("query_21_groups_0"), val = tensor(1)]; tensor layers_5_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(171562112))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(172348608))), name = tensor("layers_5_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(172348800)))]; tensor query_21_cast_fp16 = conv(bias = layers_5_self_attn_q_proj_bias_to_fp16, dilations = query_21_dilations_0, groups = query_21_groups_0, pad = query_21_pad_0, pad_type = query_21_pad_type_0, strides = query_21_strides_0, weight = layers_5_self_attn_q_proj_weight_to_fp16_palettized, x = obj_71_cast_fp16)[name = tensor("query_21_cast_fp16")]; tensor current_key_11_pad_type_0 = const()[name = tensor("current_key_11_pad_type_0"), val = tensor("valid")]; tensor current_key_11_strides_0 = const()[name = tensor("current_key_11_strides_0"), val = tensor([1, 1])]; tensor current_key_11_pad_0 = const()[name = tensor("current_key_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_11_dilations_0 = const()[name = tensor("current_key_11_dilations_0"), val = tensor([1, 1])]; tensor current_key_11_groups_0 = const()[name = tensor("current_key_11_groups_0"), val = tensor(1)]; tensor layers_5_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(172350912))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(173137408))), name = tensor("layers_5_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor current_key_11_cast_fp16 = conv(dilations = current_key_11_dilations_0, groups = current_key_11_groups_0, pad = current_key_11_pad_0, pad_type = current_key_11_pad_type_0, strides = current_key_11_strides_0, weight = layers_5_self_attn_k_proj_weight_to_fp16_palettized, x = obj_71_cast_fp16)[name = tensor("current_key_11_cast_fp16")]; tensor current_value_11_pad_type_0 = const()[name = tensor("current_value_11_pad_type_0"), val = tensor("valid")]; tensor current_value_11_strides_0 = const()[name = tensor("current_value_11_strides_0"), val = tensor([1, 1])]; tensor current_value_11_pad_0 = const()[name = tensor("current_value_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_11_dilations_0 = const()[name = tensor("current_value_11_dilations_0"), val = tensor([1, 1])]; tensor current_value_11_groups_0 = const()[name = tensor("current_value_11_groups_0"), val = tensor(1)]; tensor layers_5_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(173137600))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(173924096))), name = tensor("layers_5_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(173924288)))]; tensor current_value_11_cast_fp16 = conv(bias = layers_5_self_attn_v_proj_bias_to_fp16, dilations = current_value_11_dilations_0, groups = current_value_11_groups_0, pad = current_value_11_pad_0, pad_type = current_value_11_pad_type_0, strides = current_value_11_strides_0, weight = layers_5_self_attn_v_proj_weight_to_fp16_palettized, x = obj_71_cast_fp16)[name = tensor("current_value_11_cast_fp16")]; tensor var_1303_cast_fp16 = mul(x = var_87_cast_fp16_5, y = var_207_cast_fp16)[name = tensor("op_1303_cast_fp16")]; tensor var_1304_cast_fp16 = mul(x = current_key_11_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_1304_cast_fp16")]; tensor key_21_cast_fp16 = add(x = var_1303_cast_fp16, y = var_1304_cast_fp16)[name = tensor("key_21_cast_fp16")]; tensor var_1307_cast_fp16 = mul(x = var_114_cast_fp16_5, y = var_207_cast_fp16)[name = tensor("op_1307_cast_fp16")]; tensor var_1308_cast_fp16 = mul(x = current_value_11_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_1308_cast_fp16")]; tensor value_21_cast_fp16 = add(x = var_1307_cast_fp16, y = var_1308_cast_fp16)[name = tensor("value_21_cast_fp16")]; tensor var_1312 = const()[name = tensor("op_1312"), val = tensor([1, 16, 64, 1])]; tensor mh_q_21_cast_fp16 = reshape(shape = var_1312, x = query_21_cast_fp16)[name = tensor("mh_q_21_cast_fp16")]; tensor var_1314_to_fp16 = const()[name = tensor("op_1314_to_fp16"), val = tensor(0x1p-3)]; tensor var_1315_cast_fp16 = mul(x = mh_q_21_cast_fp16, y = var_1314_to_fp16)[name = tensor("op_1315_cast_fp16")]; tensor var_1318 = const()[name = tensor("op_1318"), val = tensor([1, 16, 64, 448])]; tensor var_1319_cast_fp16 = reshape(shape = var_1318, x = key_21_cast_fp16)[name = tensor("op_1319_cast_fp16")]; tensor mh_w_31_transpose_x_0 = const()[name = tensor("mh_w_31_transpose_x_0"), val = tensor(true)]; tensor mh_w_31_transpose_y_0 = const()[name = tensor("mh_w_31_transpose_y_0"), val = tensor(false)]; tensor mh_w_31_cast_fp16 = matmul(transpose_x = mh_w_31_transpose_x_0, transpose_y = mh_w_31_transpose_y_0, x = var_1315_cast_fp16, y = var_1319_cast_fp16)[name = tensor("mh_w_31_cast_fp16")]; tensor mh_w_33_cast_fp16 = add(x = mh_w_31_cast_fp16, y = var_229_cast_fp16)[name = tensor("mh_w_33_cast_fp16")]; tensor var_1327_cast_fp16 = softmax(axis = var_1239, x = mh_w_33_cast_fp16)[name = tensor("op_1327_cast_fp16")]; tensor var_1328 = const()[name = tensor("op_1328"), val = tensor([1, 16, 64, 448])]; tensor var_1329_cast_fp16 = reshape(shape = var_1328, x = value_21_cast_fp16)[name = tensor("op_1329_cast_fp16")]; tensor attn_21_transpose_x_0 = const()[name = tensor("attn_21_transpose_x_0"), val = tensor(false)]; tensor attn_21_transpose_y_0 = const()[name = tensor("attn_21_transpose_y_0"), val = tensor(true)]; tensor attn_21_cast_fp16 = matmul(transpose_x = attn_21_transpose_x_0, transpose_y = attn_21_transpose_y_0, x = var_1329_cast_fp16, y = var_1327_cast_fp16)[name = tensor("attn_21_cast_fp16")]; tensor var_1332 = const()[name = tensor("op_1332"), val = tensor([1, 1024, 1, 1])]; tensor input_51_cast_fp16 = reshape(shape = var_1332, x = attn_21_cast_fp16)[name = tensor("input_51_cast_fp16")]; tensor obj_77_pad_type_0 = const()[name = tensor("obj_77_pad_type_0"), val = tensor("valid")]; tensor obj_77_strides_0 = const()[name = tensor("obj_77_strides_0"), val = tensor([1, 1])]; tensor obj_77_pad_0 = const()[name = tensor("obj_77_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_77_dilations_0 = const()[name = tensor("obj_77_dilations_0"), val = tensor([1, 1])]; tensor obj_77_groups_0 = const()[name = tensor("obj_77_groups_0"), val = tensor(1)]; tensor layers_5_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(173926400))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(174712896))), name = tensor("layers_5_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_5_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_5_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(174713088)))]; tensor obj_77_cast_fp16 = conv(bias = layers_5_self_attn_o_proj_bias_to_fp16, dilations = obj_77_dilations_0, groups = obj_77_groups_0, pad = obj_77_pad_0, pad_type = obj_77_pad_type_0, strides = obj_77_strides_0, weight = layers_5_self_attn_o_proj_weight_to_fp16_palettized, x = input_51_cast_fp16)[name = tensor("obj_77_cast_fp16")]; tensor inputs_33_cast_fp16 = add(x = inputs_31_cast_fp16, y = obj_77_cast_fp16)[name = tensor("inputs_33_cast_fp16")]; tensor out_33_axes_0 = const()[name = tensor("out_33_axes_0"), val = tensor([1])]; tensor var_1354_to_fp16 = const()[name = tensor("op_1354_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_33_cast_fp16 = layer_norm(axes = out_33_axes_0, epsilon = var_1354_to_fp16, x = inputs_33_cast_fp16)[name = tensor("out_33_cast_fp16")]; tensor obj_79_gamma_0_to_fp16 = const()[name = tensor("obj_79_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(174715200)))]; tensor obj_79_beta_0_to_fp16 = const()[name = tensor("obj_79_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(174717312)))]; tensor obj_79_epsilon_0_to_fp16 = const()[name = tensor("obj_79_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_79_cast_fp16 = batch_norm(beta = obj_79_beta_0_to_fp16, epsilon = obj_79_epsilon_0_to_fp16, gamma = obj_79_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_33_cast_fp16)[name = tensor("obj_79_cast_fp16")]; tensor query_23_pad_type_0 = const()[name = tensor("query_23_pad_type_0"), val = tensor("valid")]; tensor query_23_strides_0 = const()[name = tensor("query_23_strides_0"), val = tensor([1, 1])]; tensor query_23_pad_0 = const()[name = tensor("query_23_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_23_dilations_0 = const()[name = tensor("query_23_dilations_0"), val = tensor([1, 1])]; tensor query_23_groups_0 = const()[name = tensor("query_23_groups_0"), val = tensor(1)]; tensor layers_5_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(174719424))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(175505920))), name = tensor("layers_5_encoder_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_5_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_5_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(175506112)))]; tensor query_23_cast_fp16 = conv(bias = layers_5_encoder_attn_q_proj_bias_to_fp16, dilations = query_23_dilations_0, groups = query_23_groups_0, pad = query_23_pad_0, pad_type = query_23_pad_type_0, strides = query_23_strides_0, weight = layers_5_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_79_cast_fp16)[name = tensor("query_23_cast_fp16")]; tensor key_23_pad_type_0 = const()[name = tensor("key_23_pad_type_0"), val = tensor("valid")]; tensor key_23_strides_0 = const()[name = tensor("key_23_strides_0"), val = tensor([1, 1])]; tensor key_23_pad_0 = const()[name = tensor("key_23_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_23_dilations_0 = const()[name = tensor("key_23_dilations_0"), val = tensor([1, 1])]; tensor key_23_groups_0 = const()[name = tensor("key_23_groups_0"), val = tensor(1)]; tensor layers_5_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(175508224))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176294720))), name = tensor("layers_5_encoder_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor key_23_cast_fp16 = conv(dilations = key_23_dilations_0, groups = key_23_groups_0, pad = key_23_pad_0, pad_type = key_23_pad_type_0, strides = key_23_strides_0, weight = layers_5_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("key_23_cast_fp16")]; tensor value_23_pad_type_0 = const()[name = tensor("value_23_pad_type_0"), val = tensor("valid")]; tensor value_23_strides_0 = const()[name = tensor("value_23_strides_0"), val = tensor([1, 1])]; tensor value_23_pad_0 = const()[name = tensor("value_23_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_23_dilations_0 = const()[name = tensor("value_23_dilations_0"), val = tensor([1, 1])]; tensor value_23_groups_0 = const()[name = tensor("value_23_groups_0"), val = tensor(1)]; tensor layers_5_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176294912))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(177081408))), name = tensor("layers_5_encoder_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_5_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_5_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(177081600)))]; tensor value_23_cast_fp16 = conv(bias = layers_5_encoder_attn_v_proj_bias_to_fp16, dilations = value_23_dilations_0, groups = value_23_groups_0, pad = value_23_pad_0, pad_type = value_23_pad_type_0, strides = value_23_strides_0, weight = layers_5_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("value_23_cast_fp16")]; tensor var_1390 = const()[name = tensor("op_1390"), val = tensor([1, 16, 64, 1])]; tensor mh_q_23_cast_fp16 = reshape(shape = var_1390, x = query_23_cast_fp16)[name = tensor("mh_q_23_cast_fp16")]; tensor var_1392_to_fp16 = const()[name = tensor("op_1392_to_fp16"), val = tensor(0x1p-3)]; tensor var_1393_cast_fp16 = mul(x = mh_q_23_cast_fp16, y = var_1392_to_fp16)[name = tensor("op_1393_cast_fp16")]; tensor var_1396 = const()[name = tensor("op_1396"), val = tensor([1, 16, 64, 1500])]; tensor var_1397_cast_fp16 = reshape(shape = var_1396, x = key_23_cast_fp16)[name = tensor("op_1397_cast_fp16")]; tensor mh_w_35_transpose_x_0 = const()[name = tensor("mh_w_35_transpose_x_0"), val = tensor(true)]; tensor mh_w_35_transpose_y_0 = const()[name = tensor("mh_w_35_transpose_y_0"), val = tensor(false)]; tensor mh_w_35_cast_fp16 = matmul(transpose_x = mh_w_35_transpose_x_0, transpose_y = mh_w_35_transpose_y_0, x = var_1393_cast_fp16, y = var_1397_cast_fp16)[name = tensor("mh_w_35_cast_fp16")]; tensor obj_83_cast_fp16 = softmax(axis = var_1239, x = mh_w_35_cast_fp16)[name = tensor("obj_83_cast_fp16")]; tensor var_1401 = const()[name = tensor("op_1401"), val = tensor([1, 16, 64, 1500])]; tensor var_1402_cast_fp16 = reshape(shape = var_1401, x = value_23_cast_fp16)[name = tensor("op_1402_cast_fp16")]; tensor attn_23_transpose_x_0 = const()[name = tensor("attn_23_transpose_x_0"), val = tensor(false)]; tensor attn_23_transpose_y_0 = const()[name = tensor("attn_23_transpose_y_0"), val = tensor(true)]; tensor attn_23_cast_fp16 = matmul(transpose_x = attn_23_transpose_x_0, transpose_y = attn_23_transpose_y_0, x = var_1402_cast_fp16, y = obj_83_cast_fp16)[name = tensor("attn_23_cast_fp16")]; tensor var_1405 = const()[name = tensor("op_1405"), val = tensor([1, 1024, 1, 1])]; tensor input_53_cast_fp16 = reshape(shape = var_1405, x = attn_23_cast_fp16)[name = tensor("input_53_cast_fp16")]; tensor obj_81_pad_type_0 = const()[name = tensor("obj_81_pad_type_0"), val = tensor("valid")]; tensor obj_81_strides_0 = const()[name = tensor("obj_81_strides_0"), val = tensor([1, 1])]; tensor obj_81_pad_0 = const()[name = tensor("obj_81_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_81_dilations_0 = const()[name = tensor("obj_81_dilations_0"), val = tensor([1, 1])]; tensor obj_81_groups_0 = const()[name = tensor("obj_81_groups_0"), val = tensor(1)]; tensor layers_5_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(177083712))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(177870208))), name = tensor("layers_5_encoder_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_5_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_5_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(177870400)))]; tensor obj_81_cast_fp16 = conv(bias = layers_5_encoder_attn_o_proj_bias_to_fp16, dilations = obj_81_dilations_0, groups = obj_81_groups_0, pad = obj_81_pad_0, pad_type = obj_81_pad_type_0, strides = obj_81_strides_0, weight = layers_5_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_53_cast_fp16)[name = tensor("obj_81_cast_fp16")]; tensor inputs_35_cast_fp16 = add(x = inputs_33_cast_fp16, y = obj_81_cast_fp16)[name = tensor("inputs_35_cast_fp16")]; tensor out_35_axes_0 = const()[name = tensor("out_35_axes_0"), val = tensor([1])]; tensor var_1423_to_fp16 = const()[name = tensor("op_1423_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_35_cast_fp16 = layer_norm(axes = out_35_axes_0, epsilon = var_1423_to_fp16, x = inputs_35_cast_fp16)[name = tensor("out_35_cast_fp16")]; tensor input_55_gamma_0_to_fp16 = const()[name = tensor("input_55_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(177872512)))]; tensor input_55_beta_0_to_fp16 = const()[name = tensor("input_55_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(177874624)))]; tensor input_55_epsilon_0_to_fp16 = const()[name = tensor("input_55_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_55_cast_fp16 = batch_norm(beta = input_55_beta_0_to_fp16, epsilon = input_55_epsilon_0_to_fp16, gamma = input_55_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_35_cast_fp16)[name = tensor("input_55_cast_fp16")]; tensor input_57_pad_type_0 = const()[name = tensor("input_57_pad_type_0"), val = tensor("valid")]; tensor input_57_strides_0 = const()[name = tensor("input_57_strides_0"), val = tensor([1, 1])]; tensor input_57_pad_0 = const()[name = tensor("input_57_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_57_dilations_0 = const()[name = tensor("input_57_dilations_0"), val = tensor([1, 1])]; tensor input_57_groups_0 = const()[name = tensor("input_57_groups_0"), val = tensor(1)]; tensor layers_5_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(177876736))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(181022528))), name = tensor("layers_5_fc1_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; tensor layers_5_fc1_bias_to_fp16 = const()[name = tensor("layers_5_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(181022720)))]; tensor input_57_cast_fp16 = conv(bias = layers_5_fc1_bias_to_fp16, dilations = input_57_dilations_0, groups = input_57_groups_0, pad = input_57_pad_0, pad_type = input_57_pad_type_0, strides = input_57_strides_0, weight = layers_5_fc1_weight_to_fp16_palettized, x = input_55_cast_fp16)[name = tensor("input_57_cast_fp16")]; tensor input_59_mode_0 = const()[name = tensor("input_59_mode_0"), val = tensor("EXACT")]; tensor input_59_cast_fp16 = gelu(mode = input_59_mode_0, x = input_57_cast_fp16)[name = tensor("input_59_cast_fp16")]; tensor hidden_states_13_pad_type_0 = const()[name = tensor("hidden_states_13_pad_type_0"), val = tensor("valid")]; tensor hidden_states_13_strides_0 = const()[name = tensor("hidden_states_13_strides_0"), val = tensor([1, 1])]; tensor hidden_states_13_pad_0 = const()[name = tensor("hidden_states_13_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_13_dilations_0 = const()[name = tensor("hidden_states_13_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_13_groups_0 = const()[name = tensor("hidden_states_13_groups_0"), val = tensor(1)]; tensor layers_5_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(181030976))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184176768))), name = tensor("layers_5_fc2_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; tensor layers_5_fc2_bias_to_fp16 = const()[name = tensor("layers_5_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184176960)))]; tensor hidden_states_13_cast_fp16 = conv(bias = layers_5_fc2_bias_to_fp16, dilations = hidden_states_13_dilations_0, groups = hidden_states_13_groups_0, pad = hidden_states_13_pad_0, pad_type = hidden_states_13_pad_type_0, strides = hidden_states_13_strides_0, weight = layers_5_fc2_weight_to_fp16_palettized, x = input_59_cast_fp16)[name = tensor("hidden_states_13_cast_fp16")]; tensor inputs_37_cast_fp16 = add(x = inputs_35_cast_fp16, y = hidden_states_13_cast_fp16)[name = tensor("inputs_37_cast_fp16")]; tensor var_1458 = const()[name = tensor("op_1458"), val = tensor(3)]; tensor out_37_axes_0 = const()[name = tensor("out_37_axes_0"), val = tensor([1])]; tensor var_1483_to_fp16 = const()[name = tensor("op_1483_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_37_cast_fp16 = layer_norm(axes = out_37_axes_0, epsilon = var_1483_to_fp16, x = inputs_37_cast_fp16)[name = tensor("out_37_cast_fp16")]; tensor obj_85_gamma_0_to_fp16 = const()[name = tensor("obj_85_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184179072)))]; tensor obj_85_beta_0_to_fp16 = const()[name = tensor("obj_85_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184181184)))]; tensor obj_85_epsilon_0_to_fp16 = const()[name = tensor("obj_85_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_85_cast_fp16 = batch_norm(beta = obj_85_beta_0_to_fp16, epsilon = obj_85_epsilon_0_to_fp16, gamma = obj_85_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_37_cast_fp16)[name = tensor("obj_85_cast_fp16")]; tensor query_25_pad_type_0 = const()[name = tensor("query_25_pad_type_0"), val = tensor("valid")]; tensor query_25_strides_0 = const()[name = tensor("query_25_strides_0"), val = tensor([1, 1])]; tensor query_25_pad_0 = const()[name = tensor("query_25_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_25_dilations_0 = const()[name = tensor("query_25_dilations_0"), val = tensor([1, 1])]; tensor query_25_groups_0 = const()[name = tensor("query_25_groups_0"), val = tensor(1)]; tensor layers_6_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184183296))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184969792))), name = tensor("layers_6_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184969984)))]; tensor query_25_cast_fp16 = conv(bias = layers_6_self_attn_q_proj_bias_to_fp16, dilations = query_25_dilations_0, groups = query_25_groups_0, pad = query_25_pad_0, pad_type = query_25_pad_type_0, strides = query_25_strides_0, weight = layers_6_self_attn_q_proj_weight_to_fp16_palettized, x = obj_85_cast_fp16)[name = tensor("query_25_cast_fp16")]; tensor current_key_13_pad_type_0 = const()[name = tensor("current_key_13_pad_type_0"), val = tensor("valid")]; tensor current_key_13_strides_0 = const()[name = tensor("current_key_13_strides_0"), val = tensor([1, 1])]; tensor current_key_13_pad_0 = const()[name = tensor("current_key_13_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_13_dilations_0 = const()[name = tensor("current_key_13_dilations_0"), val = tensor([1, 1])]; tensor current_key_13_groups_0 = const()[name = tensor("current_key_13_groups_0"), val = tensor(1)]; tensor layers_6_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184972096))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(185758592))), name = tensor("layers_6_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor current_key_13_cast_fp16 = conv(dilations = current_key_13_dilations_0, groups = current_key_13_groups_0, pad = current_key_13_pad_0, pad_type = current_key_13_pad_type_0, strides = current_key_13_strides_0, weight = layers_6_self_attn_k_proj_weight_to_fp16_palettized, x = obj_85_cast_fp16)[name = tensor("current_key_13_cast_fp16")]; tensor current_value_13_pad_type_0 = const()[name = tensor("current_value_13_pad_type_0"), val = tensor("valid")]; tensor current_value_13_strides_0 = const()[name = tensor("current_value_13_strides_0"), val = tensor([1, 1])]; tensor current_value_13_pad_0 = const()[name = tensor("current_value_13_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_13_dilations_0 = const()[name = tensor("current_value_13_dilations_0"), val = tensor([1, 1])]; tensor current_value_13_groups_0 = const()[name = tensor("current_value_13_groups_0"), val = tensor(1)]; tensor layers_6_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(185758784))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186545280))), name = tensor("layers_6_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186545472)))]; tensor current_value_13_cast_fp16 = conv(bias = layers_6_self_attn_v_proj_bias_to_fp16, dilations = current_value_13_dilations_0, groups = current_value_13_groups_0, pad = current_value_13_pad_0, pad_type = current_value_13_pad_type_0, strides = current_value_13_strides_0, weight = layers_6_self_attn_v_proj_weight_to_fp16_palettized, x = obj_85_cast_fp16)[name = tensor("current_value_13_cast_fp16")]; tensor var_1522_cast_fp16 = mul(x = var_87_cast_fp16_6, y = var_207_cast_fp16)[name = tensor("op_1522_cast_fp16")]; tensor var_1523_cast_fp16 = mul(x = current_key_13_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_1523_cast_fp16")]; tensor key_25_cast_fp16 = add(x = var_1522_cast_fp16, y = var_1523_cast_fp16)[name = tensor("key_25_cast_fp16")]; tensor var_1526_cast_fp16 = mul(x = var_114_cast_fp16_6, y = var_207_cast_fp16)[name = tensor("op_1526_cast_fp16")]; tensor var_1527_cast_fp16 = mul(x = current_value_13_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_1527_cast_fp16")]; tensor value_25_cast_fp16 = add(x = var_1526_cast_fp16, y = var_1527_cast_fp16)[name = tensor("value_25_cast_fp16")]; tensor var_1531 = const()[name = tensor("op_1531"), val = tensor([1, 16, 64, 1])]; tensor mh_q_25_cast_fp16 = reshape(shape = var_1531, x = query_25_cast_fp16)[name = tensor("mh_q_25_cast_fp16")]; tensor var_1533_to_fp16 = const()[name = tensor("op_1533_to_fp16"), val = tensor(0x1p-3)]; tensor var_1534_cast_fp16 = mul(x = mh_q_25_cast_fp16, y = var_1533_to_fp16)[name = tensor("op_1534_cast_fp16")]; tensor var_1537 = const()[name = tensor("op_1537"), val = tensor([1, 16, 64, 448])]; tensor var_1538_cast_fp16 = reshape(shape = var_1537, x = key_25_cast_fp16)[name = tensor("op_1538_cast_fp16")]; tensor mh_w_37_transpose_x_0 = const()[name = tensor("mh_w_37_transpose_x_0"), val = tensor(true)]; tensor mh_w_37_transpose_y_0 = const()[name = tensor("mh_w_37_transpose_y_0"), val = tensor(false)]; tensor mh_w_37_cast_fp16 = matmul(transpose_x = mh_w_37_transpose_x_0, transpose_y = mh_w_37_transpose_y_0, x = var_1534_cast_fp16, y = var_1538_cast_fp16)[name = tensor("mh_w_37_cast_fp16")]; tensor mh_w_39_cast_fp16 = add(x = mh_w_37_cast_fp16, y = var_229_cast_fp16)[name = tensor("mh_w_39_cast_fp16")]; tensor var_1546_cast_fp16 = softmax(axis = var_1458, x = mh_w_39_cast_fp16)[name = tensor("op_1546_cast_fp16")]; tensor var_1547 = const()[name = tensor("op_1547"), val = tensor([1, 16, 64, 448])]; tensor var_1548_cast_fp16 = reshape(shape = var_1547, x = value_25_cast_fp16)[name = tensor("op_1548_cast_fp16")]; tensor attn_25_transpose_x_0 = const()[name = tensor("attn_25_transpose_x_0"), val = tensor(false)]; tensor attn_25_transpose_y_0 = const()[name = tensor("attn_25_transpose_y_0"), val = tensor(true)]; tensor attn_25_cast_fp16 = matmul(transpose_x = attn_25_transpose_x_0, transpose_y = attn_25_transpose_y_0, x = var_1548_cast_fp16, y = var_1546_cast_fp16)[name = tensor("attn_25_cast_fp16")]; tensor var_1551 = const()[name = tensor("op_1551"), val = tensor([1, 1024, 1, 1])]; tensor input_61_cast_fp16 = reshape(shape = var_1551, x = attn_25_cast_fp16)[name = tensor("input_61_cast_fp16")]; tensor obj_91_pad_type_0 = const()[name = tensor("obj_91_pad_type_0"), val = tensor("valid")]; tensor obj_91_strides_0 = const()[name = tensor("obj_91_strides_0"), val = tensor([1, 1])]; tensor obj_91_pad_0 = const()[name = tensor("obj_91_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_91_dilations_0 = const()[name = tensor("obj_91_dilations_0"), val = tensor([1, 1])]; tensor obj_91_groups_0 = const()[name = tensor("obj_91_groups_0"), val = tensor(1)]; tensor layers_6_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186547584))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(187334080))), name = tensor("layers_6_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_6_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_6_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(187334272)))]; tensor obj_91_cast_fp16 = conv(bias = layers_6_self_attn_o_proj_bias_to_fp16, dilations = obj_91_dilations_0, groups = obj_91_groups_0, pad = obj_91_pad_0, pad_type = obj_91_pad_type_0, strides = obj_91_strides_0, weight = layers_6_self_attn_o_proj_weight_to_fp16_palettized, x = input_61_cast_fp16)[name = tensor("obj_91_cast_fp16")]; tensor inputs_39_cast_fp16 = add(x = inputs_37_cast_fp16, y = obj_91_cast_fp16)[name = tensor("inputs_39_cast_fp16")]; tensor out_39_axes_0 = const()[name = tensor("out_39_axes_0"), val = tensor([1])]; tensor var_1573_to_fp16 = const()[name = tensor("op_1573_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_39_cast_fp16 = layer_norm(axes = out_39_axes_0, epsilon = var_1573_to_fp16, x = inputs_39_cast_fp16)[name = tensor("out_39_cast_fp16")]; tensor obj_93_gamma_0_to_fp16 = const()[name = tensor("obj_93_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(187336384)))]; tensor obj_93_beta_0_to_fp16 = const()[name = tensor("obj_93_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(187338496)))]; tensor obj_93_epsilon_0_to_fp16 = const()[name = tensor("obj_93_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_93_cast_fp16 = batch_norm(beta = obj_93_beta_0_to_fp16, epsilon = obj_93_epsilon_0_to_fp16, gamma = obj_93_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_39_cast_fp16)[name = tensor("obj_93_cast_fp16")]; tensor query_27_pad_type_0 = const()[name = tensor("query_27_pad_type_0"), val = tensor("valid")]; tensor query_27_strides_0 = const()[name = tensor("query_27_strides_0"), val = tensor([1, 1])]; tensor query_27_pad_0 = const()[name = tensor("query_27_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_27_dilations_0 = const()[name = tensor("query_27_dilations_0"), val = tensor([1, 1])]; tensor query_27_groups_0 = const()[name = tensor("query_27_groups_0"), val = tensor(1)]; tensor layers_6_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(187340608))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(188127104))), name = tensor("layers_6_encoder_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_6_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_6_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(188127296)))]; tensor query_27_cast_fp16 = conv(bias = layers_6_encoder_attn_q_proj_bias_to_fp16, dilations = query_27_dilations_0, groups = query_27_groups_0, pad = query_27_pad_0, pad_type = query_27_pad_type_0, strides = query_27_strides_0, weight = layers_6_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_93_cast_fp16)[name = tensor("query_27_cast_fp16")]; tensor key_27_pad_type_0 = const()[name = tensor("key_27_pad_type_0"), val = tensor("valid")]; tensor key_27_strides_0 = const()[name = tensor("key_27_strides_0"), val = tensor([1, 1])]; tensor key_27_pad_0 = const()[name = tensor("key_27_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_27_dilations_0 = const()[name = tensor("key_27_dilations_0"), val = tensor([1, 1])]; tensor key_27_groups_0 = const()[name = tensor("key_27_groups_0"), val = tensor(1)]; tensor layers_6_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(188129408))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(188915904))), name = tensor("layers_6_encoder_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor key_27_cast_fp16 = conv(dilations = key_27_dilations_0, groups = key_27_groups_0, pad = key_27_pad_0, pad_type = key_27_pad_type_0, strides = key_27_strides_0, weight = layers_6_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("key_27_cast_fp16")]; tensor value_27_pad_type_0 = const()[name = tensor("value_27_pad_type_0"), val = tensor("valid")]; tensor value_27_strides_0 = const()[name = tensor("value_27_strides_0"), val = tensor([1, 1])]; tensor value_27_pad_0 = const()[name = tensor("value_27_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_27_dilations_0 = const()[name = tensor("value_27_dilations_0"), val = tensor([1, 1])]; tensor value_27_groups_0 = const()[name = tensor("value_27_groups_0"), val = tensor(1)]; tensor layers_6_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(188916096))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(189702592))), name = tensor("layers_6_encoder_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_6_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_6_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(189702784)))]; tensor value_27_cast_fp16 = conv(bias = layers_6_encoder_attn_v_proj_bias_to_fp16, dilations = value_27_dilations_0, groups = value_27_groups_0, pad = value_27_pad_0, pad_type = value_27_pad_type_0, strides = value_27_strides_0, weight = layers_6_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("value_27_cast_fp16")]; tensor var_1609 = const()[name = tensor("op_1609"), val = tensor([1, 16, 64, 1])]; tensor mh_q_27_cast_fp16 = reshape(shape = var_1609, x = query_27_cast_fp16)[name = tensor("mh_q_27_cast_fp16")]; tensor var_1611_to_fp16 = const()[name = tensor("op_1611_to_fp16"), val = tensor(0x1p-3)]; tensor var_1612_cast_fp16 = mul(x = mh_q_27_cast_fp16, y = var_1611_to_fp16)[name = tensor("op_1612_cast_fp16")]; tensor var_1615 = const()[name = tensor("op_1615"), val = tensor([1, 16, 64, 1500])]; tensor var_1616_cast_fp16 = reshape(shape = var_1615, x = key_27_cast_fp16)[name = tensor("op_1616_cast_fp16")]; tensor mh_w_41_transpose_x_0 = const()[name = tensor("mh_w_41_transpose_x_0"), val = tensor(true)]; tensor mh_w_41_transpose_y_0 = const()[name = tensor("mh_w_41_transpose_y_0"), val = tensor(false)]; tensor mh_w_41_cast_fp16 = matmul(transpose_x = mh_w_41_transpose_x_0, transpose_y = mh_w_41_transpose_y_0, x = var_1612_cast_fp16, y = var_1616_cast_fp16)[name = tensor("mh_w_41_cast_fp16")]; tensor obj_97_cast_fp16 = softmax(axis = var_1458, x = mh_w_41_cast_fp16)[name = tensor("obj_97_cast_fp16")]; tensor var_1620 = const()[name = tensor("op_1620"), val = tensor([1, 16, 64, 1500])]; tensor var_1621_cast_fp16 = reshape(shape = var_1620, x = value_27_cast_fp16)[name = tensor("op_1621_cast_fp16")]; tensor attn_27_transpose_x_0 = const()[name = tensor("attn_27_transpose_x_0"), val = tensor(false)]; tensor attn_27_transpose_y_0 = const()[name = tensor("attn_27_transpose_y_0"), val = tensor(true)]; tensor attn_27_cast_fp16 = matmul(transpose_x = attn_27_transpose_x_0, transpose_y = attn_27_transpose_y_0, x = var_1621_cast_fp16, y = obj_97_cast_fp16)[name = tensor("attn_27_cast_fp16")]; tensor var_1624 = const()[name = tensor("op_1624"), val = tensor([1, 1024, 1, 1])]; tensor input_63_cast_fp16 = reshape(shape = var_1624, x = attn_27_cast_fp16)[name = tensor("input_63_cast_fp16")]; tensor obj_95_pad_type_0 = const()[name = tensor("obj_95_pad_type_0"), val = tensor("valid")]; tensor obj_95_strides_0 = const()[name = tensor("obj_95_strides_0"), val = tensor([1, 1])]; tensor obj_95_pad_0 = const()[name = tensor("obj_95_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_95_dilations_0 = const()[name = tensor("obj_95_dilations_0"), val = tensor([1, 1])]; tensor obj_95_groups_0 = const()[name = tensor("obj_95_groups_0"), val = tensor(1)]; tensor layers_6_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(189704896))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190491392))), name = tensor("layers_6_encoder_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_6_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_6_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190491584)))]; tensor obj_95_cast_fp16 = conv(bias = layers_6_encoder_attn_o_proj_bias_to_fp16, dilations = obj_95_dilations_0, groups = obj_95_groups_0, pad = obj_95_pad_0, pad_type = obj_95_pad_type_0, strides = obj_95_strides_0, weight = layers_6_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_63_cast_fp16)[name = tensor("obj_95_cast_fp16")]; tensor inputs_41_cast_fp16 = add(x = inputs_39_cast_fp16, y = obj_95_cast_fp16)[name = tensor("inputs_41_cast_fp16")]; tensor out_41_axes_0 = const()[name = tensor("out_41_axes_0"), val = tensor([1])]; tensor var_1642_to_fp16 = const()[name = tensor("op_1642_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_41_cast_fp16 = layer_norm(axes = out_41_axes_0, epsilon = var_1642_to_fp16, x = inputs_41_cast_fp16)[name = tensor("out_41_cast_fp16")]; tensor input_65_gamma_0_to_fp16 = const()[name = tensor("input_65_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190493696)))]; tensor input_65_beta_0_to_fp16 = const()[name = tensor("input_65_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190495808)))]; tensor input_65_epsilon_0_to_fp16 = const()[name = tensor("input_65_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_65_cast_fp16 = batch_norm(beta = input_65_beta_0_to_fp16, epsilon = input_65_epsilon_0_to_fp16, gamma = input_65_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_41_cast_fp16)[name = tensor("input_65_cast_fp16")]; tensor input_67_pad_type_0 = const()[name = tensor("input_67_pad_type_0"), val = tensor("valid")]; tensor input_67_strides_0 = const()[name = tensor("input_67_strides_0"), val = tensor([1, 1])]; tensor input_67_pad_0 = const()[name = tensor("input_67_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_67_dilations_0 = const()[name = tensor("input_67_dilations_0"), val = tensor([1, 1])]; tensor input_67_groups_0 = const()[name = tensor("input_67_groups_0"), val = tensor(1)]; tensor layers_6_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190497920))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(193643712))), name = tensor("layers_6_fc1_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; tensor layers_6_fc1_bias_to_fp16 = const()[name = tensor("layers_6_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(193643904)))]; tensor input_67_cast_fp16 = conv(bias = layers_6_fc1_bias_to_fp16, dilations = input_67_dilations_0, groups = input_67_groups_0, pad = input_67_pad_0, pad_type = input_67_pad_type_0, strides = input_67_strides_0, weight = layers_6_fc1_weight_to_fp16_palettized, x = input_65_cast_fp16)[name = tensor("input_67_cast_fp16")]; tensor input_69_mode_0 = const()[name = tensor("input_69_mode_0"), val = tensor("EXACT")]; tensor input_69_cast_fp16 = gelu(mode = input_69_mode_0, x = input_67_cast_fp16)[name = tensor("input_69_cast_fp16")]; tensor hidden_states_15_pad_type_0 = const()[name = tensor("hidden_states_15_pad_type_0"), val = tensor("valid")]; tensor hidden_states_15_strides_0 = const()[name = tensor("hidden_states_15_strides_0"), val = tensor([1, 1])]; tensor hidden_states_15_pad_0 = const()[name = tensor("hidden_states_15_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_15_dilations_0 = const()[name = tensor("hidden_states_15_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_15_groups_0 = const()[name = tensor("hidden_states_15_groups_0"), val = tensor(1)]; tensor layers_6_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(193652160))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(196797952))), name = tensor("layers_6_fc2_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; tensor layers_6_fc2_bias_to_fp16 = const()[name = tensor("layers_6_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(196798144)))]; tensor hidden_states_15_cast_fp16 = conv(bias = layers_6_fc2_bias_to_fp16, dilations = hidden_states_15_dilations_0, groups = hidden_states_15_groups_0, pad = hidden_states_15_pad_0, pad_type = hidden_states_15_pad_type_0, strides = hidden_states_15_strides_0, weight = layers_6_fc2_weight_to_fp16_palettized, x = input_69_cast_fp16)[name = tensor("hidden_states_15_cast_fp16")]; tensor inputs_43_cast_fp16 = add(x = inputs_41_cast_fp16, y = hidden_states_15_cast_fp16)[name = tensor("inputs_43_cast_fp16")]; tensor var_1677 = const()[name = tensor("op_1677"), val = tensor(3)]; tensor out_43_axes_0 = const()[name = tensor("out_43_axes_0"), val = tensor([1])]; tensor var_1702_to_fp16 = const()[name = tensor("op_1702_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_43_cast_fp16 = layer_norm(axes = out_43_axes_0, epsilon = var_1702_to_fp16, x = inputs_43_cast_fp16)[name = tensor("out_43_cast_fp16")]; tensor obj_99_gamma_0_to_fp16 = const()[name = tensor("obj_99_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(196800256)))]; tensor obj_99_beta_0_to_fp16 = const()[name = tensor("obj_99_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(196802368)))]; tensor obj_99_epsilon_0_to_fp16 = const()[name = tensor("obj_99_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_99_cast_fp16 = batch_norm(beta = obj_99_beta_0_to_fp16, epsilon = obj_99_epsilon_0_to_fp16, gamma = obj_99_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_43_cast_fp16)[name = tensor("obj_99_cast_fp16")]; tensor query_29_pad_type_0 = const()[name = tensor("query_29_pad_type_0"), val = tensor("valid")]; tensor query_29_strides_0 = const()[name = tensor("query_29_strides_0"), val = tensor([1, 1])]; tensor query_29_pad_0 = const()[name = tensor("query_29_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_29_dilations_0 = const()[name = tensor("query_29_dilations_0"), val = tensor([1, 1])]; tensor query_29_groups_0 = const()[name = tensor("query_29_groups_0"), val = tensor(1)]; tensor layers_7_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(196804480))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(197590976))), name = tensor("layers_7_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(197591168)))]; tensor query_29_cast_fp16 = conv(bias = layers_7_self_attn_q_proj_bias_to_fp16, dilations = query_29_dilations_0, groups = query_29_groups_0, pad = query_29_pad_0, pad_type = query_29_pad_type_0, strides = query_29_strides_0, weight = layers_7_self_attn_q_proj_weight_to_fp16_palettized, x = obj_99_cast_fp16)[name = tensor("query_29_cast_fp16")]; tensor current_key_15_pad_type_0 = const()[name = tensor("current_key_15_pad_type_0"), val = tensor("valid")]; tensor current_key_15_strides_0 = const()[name = tensor("current_key_15_strides_0"), val = tensor([1, 1])]; tensor current_key_15_pad_0 = const()[name = tensor("current_key_15_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_15_dilations_0 = const()[name = tensor("current_key_15_dilations_0"), val = tensor([1, 1])]; tensor current_key_15_groups_0 = const()[name = tensor("current_key_15_groups_0"), val = tensor(1)]; tensor layers_7_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(197593280))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(198379776))), name = tensor("layers_7_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor current_key_15_cast_fp16 = conv(dilations = current_key_15_dilations_0, groups = current_key_15_groups_0, pad = current_key_15_pad_0, pad_type = current_key_15_pad_type_0, strides = current_key_15_strides_0, weight = layers_7_self_attn_k_proj_weight_to_fp16_palettized, x = obj_99_cast_fp16)[name = tensor("current_key_15_cast_fp16")]; tensor current_value_15_pad_type_0 = const()[name = tensor("current_value_15_pad_type_0"), val = tensor("valid")]; tensor current_value_15_strides_0 = const()[name = tensor("current_value_15_strides_0"), val = tensor([1, 1])]; tensor current_value_15_pad_0 = const()[name = tensor("current_value_15_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_15_dilations_0 = const()[name = tensor("current_value_15_dilations_0"), val = tensor([1, 1])]; tensor current_value_15_groups_0 = const()[name = tensor("current_value_15_groups_0"), val = tensor(1)]; tensor layers_7_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(198379968))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(199166464))), name = tensor("layers_7_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(199166656)))]; tensor current_value_15_cast_fp16 = conv(bias = layers_7_self_attn_v_proj_bias_to_fp16, dilations = current_value_15_dilations_0, groups = current_value_15_groups_0, pad = current_value_15_pad_0, pad_type = current_value_15_pad_type_0, strides = current_value_15_strides_0, weight = layers_7_self_attn_v_proj_weight_to_fp16_palettized, x = obj_99_cast_fp16)[name = tensor("current_value_15_cast_fp16")]; tensor var_1741_cast_fp16 = mul(x = var_87_cast_fp16_7, y = var_207_cast_fp16)[name = tensor("op_1741_cast_fp16")]; tensor var_1742_cast_fp16 = mul(x = current_key_15_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_1742_cast_fp16")]; tensor key_29_cast_fp16 = add(x = var_1741_cast_fp16, y = var_1742_cast_fp16)[name = tensor("key_29_cast_fp16")]; tensor var_1745_cast_fp16 = mul(x = var_114_cast_fp16_7, y = var_207_cast_fp16)[name = tensor("op_1745_cast_fp16")]; tensor var_1746_cast_fp16 = mul(x = current_value_15_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_1746_cast_fp16")]; tensor value_29_cast_fp16 = add(x = var_1745_cast_fp16, y = var_1746_cast_fp16)[name = tensor("value_29_cast_fp16")]; tensor var_1750 = const()[name = tensor("op_1750"), val = tensor([1, 16, 64, 1])]; tensor mh_q_29_cast_fp16 = reshape(shape = var_1750, x = query_29_cast_fp16)[name = tensor("mh_q_29_cast_fp16")]; tensor var_1752_to_fp16 = const()[name = tensor("op_1752_to_fp16"), val = tensor(0x1p-3)]; tensor var_1753_cast_fp16 = mul(x = mh_q_29_cast_fp16, y = var_1752_to_fp16)[name = tensor("op_1753_cast_fp16")]; tensor var_1756 = const()[name = tensor("op_1756"), val = tensor([1, 16, 64, 448])]; tensor var_1757_cast_fp16 = reshape(shape = var_1756, x = key_29_cast_fp16)[name = tensor("op_1757_cast_fp16")]; tensor mh_w_43_transpose_x_0 = const()[name = tensor("mh_w_43_transpose_x_0"), val = tensor(true)]; tensor mh_w_43_transpose_y_0 = const()[name = tensor("mh_w_43_transpose_y_0"), val = tensor(false)]; tensor mh_w_43_cast_fp16 = matmul(transpose_x = mh_w_43_transpose_x_0, transpose_y = mh_w_43_transpose_y_0, x = var_1753_cast_fp16, y = var_1757_cast_fp16)[name = tensor("mh_w_43_cast_fp16")]; tensor mh_w_45_cast_fp16 = add(x = mh_w_43_cast_fp16, y = var_229_cast_fp16)[name = tensor("mh_w_45_cast_fp16")]; tensor var_1765_cast_fp16 = softmax(axis = var_1677, x = mh_w_45_cast_fp16)[name = tensor("op_1765_cast_fp16")]; tensor var_1766 = const()[name = tensor("op_1766"), val = tensor([1, 16, 64, 448])]; tensor var_1767_cast_fp16 = reshape(shape = var_1766, x = value_29_cast_fp16)[name = tensor("op_1767_cast_fp16")]; tensor attn_29_transpose_x_0 = const()[name = tensor("attn_29_transpose_x_0"), val = tensor(false)]; tensor attn_29_transpose_y_0 = const()[name = tensor("attn_29_transpose_y_0"), val = tensor(true)]; tensor attn_29_cast_fp16 = matmul(transpose_x = attn_29_transpose_x_0, transpose_y = attn_29_transpose_y_0, x = var_1767_cast_fp16, y = var_1765_cast_fp16)[name = tensor("attn_29_cast_fp16")]; tensor var_1770 = const()[name = tensor("op_1770"), val = tensor([1, 1024, 1, 1])]; tensor input_71_cast_fp16 = reshape(shape = var_1770, x = attn_29_cast_fp16)[name = tensor("input_71_cast_fp16")]; tensor obj_105_pad_type_0 = const()[name = tensor("obj_105_pad_type_0"), val = tensor("valid")]; tensor obj_105_strides_0 = const()[name = tensor("obj_105_strides_0"), val = tensor([1, 1])]; tensor obj_105_pad_0 = const()[name = tensor("obj_105_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_105_dilations_0 = const()[name = tensor("obj_105_dilations_0"), val = tensor([1, 1])]; tensor obj_105_groups_0 = const()[name = tensor("obj_105_groups_0"), val = tensor(1)]; tensor layers_7_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(199168768))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(199955264))), name = tensor("layers_7_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_7_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_7_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(199955456)))]; tensor obj_105_cast_fp16 = conv(bias = layers_7_self_attn_o_proj_bias_to_fp16, dilations = obj_105_dilations_0, groups = obj_105_groups_0, pad = obj_105_pad_0, pad_type = obj_105_pad_type_0, strides = obj_105_strides_0, weight = layers_7_self_attn_o_proj_weight_to_fp16_palettized, x = input_71_cast_fp16)[name = tensor("obj_105_cast_fp16")]; tensor inputs_45_cast_fp16 = add(x = inputs_43_cast_fp16, y = obj_105_cast_fp16)[name = tensor("inputs_45_cast_fp16")]; tensor out_45_axes_0 = const()[name = tensor("out_45_axes_0"), val = tensor([1])]; tensor var_1792_to_fp16 = const()[name = tensor("op_1792_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_45_cast_fp16 = layer_norm(axes = out_45_axes_0, epsilon = var_1792_to_fp16, x = inputs_45_cast_fp16)[name = tensor("out_45_cast_fp16")]; tensor obj_107_gamma_0_to_fp16 = const()[name = tensor("obj_107_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(199957568)))]; tensor obj_107_beta_0_to_fp16 = const()[name = tensor("obj_107_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(199959680)))]; tensor obj_107_epsilon_0_to_fp16 = const()[name = tensor("obj_107_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_107_cast_fp16 = batch_norm(beta = obj_107_beta_0_to_fp16, epsilon = obj_107_epsilon_0_to_fp16, gamma = obj_107_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_45_cast_fp16)[name = tensor("obj_107_cast_fp16")]; tensor query_31_pad_type_0 = const()[name = tensor("query_31_pad_type_0"), val = tensor("valid")]; tensor query_31_strides_0 = const()[name = tensor("query_31_strides_0"), val = tensor([1, 1])]; tensor query_31_pad_0 = const()[name = tensor("query_31_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_31_dilations_0 = const()[name = tensor("query_31_dilations_0"), val = tensor([1, 1])]; tensor query_31_groups_0 = const()[name = tensor("query_31_groups_0"), val = tensor(1)]; tensor layers_7_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(199961792))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(200748288))), name = tensor("layers_7_encoder_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_7_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_7_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(200748480)))]; tensor query_31_cast_fp16 = conv(bias = layers_7_encoder_attn_q_proj_bias_to_fp16, dilations = query_31_dilations_0, groups = query_31_groups_0, pad = query_31_pad_0, pad_type = query_31_pad_type_0, strides = query_31_strides_0, weight = layers_7_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_107_cast_fp16)[name = tensor("query_31_cast_fp16")]; tensor key_31_pad_type_0 = const()[name = tensor("key_31_pad_type_0"), val = tensor("valid")]; tensor key_31_strides_0 = const()[name = tensor("key_31_strides_0"), val = tensor([1, 1])]; tensor key_31_pad_0 = const()[name = tensor("key_31_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_31_dilations_0 = const()[name = tensor("key_31_dilations_0"), val = tensor([1, 1])]; tensor key_31_groups_0 = const()[name = tensor("key_31_groups_0"), val = tensor(1)]; tensor layers_7_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(200750592))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(201537088))), name = tensor("layers_7_encoder_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor key_31_cast_fp16 = conv(dilations = key_31_dilations_0, groups = key_31_groups_0, pad = key_31_pad_0, pad_type = key_31_pad_type_0, strides = key_31_strides_0, weight = layers_7_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("key_31_cast_fp16")]; tensor value_31_pad_type_0 = const()[name = tensor("value_31_pad_type_0"), val = tensor("valid")]; tensor value_31_strides_0 = const()[name = tensor("value_31_strides_0"), val = tensor([1, 1])]; tensor value_31_pad_0 = const()[name = tensor("value_31_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_31_dilations_0 = const()[name = tensor("value_31_dilations_0"), val = tensor([1, 1])]; tensor value_31_groups_0 = const()[name = tensor("value_31_groups_0"), val = tensor(1)]; tensor layers_7_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(201537280))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(202323776))), name = tensor("layers_7_encoder_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_7_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_7_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(202323968)))]; tensor value_31_cast_fp16 = conv(bias = layers_7_encoder_attn_v_proj_bias_to_fp16, dilations = value_31_dilations_0, groups = value_31_groups_0, pad = value_31_pad_0, pad_type = value_31_pad_type_0, strides = value_31_strides_0, weight = layers_7_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("value_31_cast_fp16")]; tensor var_1828 = const()[name = tensor("op_1828"), val = tensor([1, 16, 64, 1])]; tensor mh_q_31_cast_fp16 = reshape(shape = var_1828, x = query_31_cast_fp16)[name = tensor("mh_q_31_cast_fp16")]; tensor var_1830_to_fp16 = const()[name = tensor("op_1830_to_fp16"), val = tensor(0x1p-3)]; tensor var_1831_cast_fp16 = mul(x = mh_q_31_cast_fp16, y = var_1830_to_fp16)[name = tensor("op_1831_cast_fp16")]; tensor var_1834 = const()[name = tensor("op_1834"), val = tensor([1, 16, 64, 1500])]; tensor var_1835_cast_fp16 = reshape(shape = var_1834, x = key_31_cast_fp16)[name = tensor("op_1835_cast_fp16")]; tensor mh_w_47_transpose_x_0 = const()[name = tensor("mh_w_47_transpose_x_0"), val = tensor(true)]; tensor mh_w_47_transpose_y_0 = const()[name = tensor("mh_w_47_transpose_y_0"), val = tensor(false)]; tensor mh_w_47_cast_fp16 = matmul(transpose_x = mh_w_47_transpose_x_0, transpose_y = mh_w_47_transpose_y_0, x = var_1831_cast_fp16, y = var_1835_cast_fp16)[name = tensor("mh_w_47_cast_fp16")]; tensor obj_111_cast_fp16 = softmax(axis = var_1677, x = mh_w_47_cast_fp16)[name = tensor("obj_111_cast_fp16")]; tensor var_1839 = const()[name = tensor("op_1839"), val = tensor([1, 16, 64, 1500])]; tensor var_1840_cast_fp16 = reshape(shape = var_1839, x = value_31_cast_fp16)[name = tensor("op_1840_cast_fp16")]; tensor attn_31_transpose_x_0 = const()[name = tensor("attn_31_transpose_x_0"), val = tensor(false)]; tensor attn_31_transpose_y_0 = const()[name = tensor("attn_31_transpose_y_0"), val = tensor(true)]; tensor attn_31_cast_fp16 = matmul(transpose_x = attn_31_transpose_x_0, transpose_y = attn_31_transpose_y_0, x = var_1840_cast_fp16, y = obj_111_cast_fp16)[name = tensor("attn_31_cast_fp16")]; tensor var_1843 = const()[name = tensor("op_1843"), val = tensor([1, 1024, 1, 1])]; tensor input_73_cast_fp16 = reshape(shape = var_1843, x = attn_31_cast_fp16)[name = tensor("input_73_cast_fp16")]; tensor obj_109_pad_type_0 = const()[name = tensor("obj_109_pad_type_0"), val = tensor("valid")]; tensor obj_109_strides_0 = const()[name = tensor("obj_109_strides_0"), val = tensor([1, 1])]; tensor obj_109_pad_0 = const()[name = tensor("obj_109_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_109_dilations_0 = const()[name = tensor("obj_109_dilations_0"), val = tensor([1, 1])]; tensor obj_109_groups_0 = const()[name = tensor("obj_109_groups_0"), val = tensor(1)]; tensor layers_7_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(202326080))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(203112576))), name = tensor("layers_7_encoder_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_7_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_7_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(203112768)))]; tensor obj_109_cast_fp16 = conv(bias = layers_7_encoder_attn_o_proj_bias_to_fp16, dilations = obj_109_dilations_0, groups = obj_109_groups_0, pad = obj_109_pad_0, pad_type = obj_109_pad_type_0, strides = obj_109_strides_0, weight = layers_7_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_73_cast_fp16)[name = tensor("obj_109_cast_fp16")]; tensor inputs_47_cast_fp16 = add(x = inputs_45_cast_fp16, y = obj_109_cast_fp16)[name = tensor("inputs_47_cast_fp16")]; tensor out_47_axes_0 = const()[name = tensor("out_47_axes_0"), val = tensor([1])]; tensor var_1861_to_fp16 = const()[name = tensor("op_1861_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_47_cast_fp16 = layer_norm(axes = out_47_axes_0, epsilon = var_1861_to_fp16, x = inputs_47_cast_fp16)[name = tensor("out_47_cast_fp16")]; tensor input_75_gamma_0_to_fp16 = const()[name = tensor("input_75_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(203114880)))]; tensor input_75_beta_0_to_fp16 = const()[name = tensor("input_75_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(203116992)))]; tensor input_75_epsilon_0_to_fp16 = const()[name = tensor("input_75_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_75_cast_fp16 = batch_norm(beta = input_75_beta_0_to_fp16, epsilon = input_75_epsilon_0_to_fp16, gamma = input_75_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_47_cast_fp16)[name = tensor("input_75_cast_fp16")]; tensor input_77_pad_type_0 = const()[name = tensor("input_77_pad_type_0"), val = tensor("valid")]; tensor input_77_strides_0 = const()[name = tensor("input_77_strides_0"), val = tensor([1, 1])]; tensor input_77_pad_0 = const()[name = tensor("input_77_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_77_dilations_0 = const()[name = tensor("input_77_dilations_0"), val = tensor([1, 1])]; tensor input_77_groups_0 = const()[name = tensor("input_77_groups_0"), val = tensor(1)]; tensor layers_7_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(203119104))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(206264896))), name = tensor("layers_7_fc1_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; tensor layers_7_fc1_bias_to_fp16 = const()[name = tensor("layers_7_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(206265088)))]; tensor input_77_cast_fp16 = conv(bias = layers_7_fc1_bias_to_fp16, dilations = input_77_dilations_0, groups = input_77_groups_0, pad = input_77_pad_0, pad_type = input_77_pad_type_0, strides = input_77_strides_0, weight = layers_7_fc1_weight_to_fp16_palettized, x = input_75_cast_fp16)[name = tensor("input_77_cast_fp16")]; tensor input_79_mode_0 = const()[name = tensor("input_79_mode_0"), val = tensor("EXACT")]; tensor input_79_cast_fp16 = gelu(mode = input_79_mode_0, x = input_77_cast_fp16)[name = tensor("input_79_cast_fp16")]; tensor hidden_states_17_pad_type_0 = const()[name = tensor("hidden_states_17_pad_type_0"), val = tensor("valid")]; tensor hidden_states_17_strides_0 = const()[name = tensor("hidden_states_17_strides_0"), val = tensor([1, 1])]; tensor hidden_states_17_pad_0 = const()[name = tensor("hidden_states_17_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_17_dilations_0 = const()[name = tensor("hidden_states_17_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_17_groups_0 = const()[name = tensor("hidden_states_17_groups_0"), val = tensor(1)]; tensor layers_7_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(206273344))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(209419136))), name = tensor("layers_7_fc2_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; tensor layers_7_fc2_bias_to_fp16 = const()[name = tensor("layers_7_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(209419328)))]; tensor hidden_states_17_cast_fp16 = conv(bias = layers_7_fc2_bias_to_fp16, dilations = hidden_states_17_dilations_0, groups = hidden_states_17_groups_0, pad = hidden_states_17_pad_0, pad_type = hidden_states_17_pad_type_0, strides = hidden_states_17_strides_0, weight = layers_7_fc2_weight_to_fp16_palettized, x = input_79_cast_fp16)[name = tensor("hidden_states_17_cast_fp16")]; tensor inputs_49_cast_fp16 = add(x = inputs_47_cast_fp16, y = hidden_states_17_cast_fp16)[name = tensor("inputs_49_cast_fp16")]; tensor var_1896 = const()[name = tensor("op_1896"), val = tensor(3)]; tensor out_49_axes_0 = const()[name = tensor("out_49_axes_0"), val = tensor([1])]; tensor var_1921_to_fp16 = const()[name = tensor("op_1921_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_49_cast_fp16 = layer_norm(axes = out_49_axes_0, epsilon = var_1921_to_fp16, x = inputs_49_cast_fp16)[name = tensor("out_49_cast_fp16")]; tensor obj_113_gamma_0_to_fp16 = const()[name = tensor("obj_113_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(209421440)))]; tensor obj_113_beta_0_to_fp16 = const()[name = tensor("obj_113_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(209423552)))]; tensor obj_113_epsilon_0_to_fp16 = const()[name = tensor("obj_113_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_113_cast_fp16 = batch_norm(beta = obj_113_beta_0_to_fp16, epsilon = obj_113_epsilon_0_to_fp16, gamma = obj_113_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_49_cast_fp16)[name = tensor("obj_113_cast_fp16")]; tensor query_33_pad_type_0 = const()[name = tensor("query_33_pad_type_0"), val = tensor("valid")]; tensor query_33_strides_0 = const()[name = tensor("query_33_strides_0"), val = tensor([1, 1])]; tensor query_33_pad_0 = const()[name = tensor("query_33_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_33_dilations_0 = const()[name = tensor("query_33_dilations_0"), val = tensor([1, 1])]; tensor query_33_groups_0 = const()[name = tensor("query_33_groups_0"), val = tensor(1)]; tensor layers_8_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(209425664))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(210212160))), name = tensor("layers_8_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(210212352)))]; tensor query_33_cast_fp16 = conv(bias = layers_8_self_attn_q_proj_bias_to_fp16, dilations = query_33_dilations_0, groups = query_33_groups_0, pad = query_33_pad_0, pad_type = query_33_pad_type_0, strides = query_33_strides_0, weight = layers_8_self_attn_q_proj_weight_to_fp16_palettized, x = obj_113_cast_fp16)[name = tensor("query_33_cast_fp16")]; tensor current_key_17_pad_type_0 = const()[name = tensor("current_key_17_pad_type_0"), val = tensor("valid")]; tensor current_key_17_strides_0 = const()[name = tensor("current_key_17_strides_0"), val = tensor([1, 1])]; tensor current_key_17_pad_0 = const()[name = tensor("current_key_17_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_17_dilations_0 = const()[name = tensor("current_key_17_dilations_0"), val = tensor([1, 1])]; tensor current_key_17_groups_0 = const()[name = tensor("current_key_17_groups_0"), val = tensor(1)]; tensor layers_8_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(210214464))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211000960))), name = tensor("layers_8_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor current_key_17_cast_fp16 = conv(dilations = current_key_17_dilations_0, groups = current_key_17_groups_0, pad = current_key_17_pad_0, pad_type = current_key_17_pad_type_0, strides = current_key_17_strides_0, weight = layers_8_self_attn_k_proj_weight_to_fp16_palettized, x = obj_113_cast_fp16)[name = tensor("current_key_17_cast_fp16")]; tensor current_value_17_pad_type_0 = const()[name = tensor("current_value_17_pad_type_0"), val = tensor("valid")]; tensor current_value_17_strides_0 = const()[name = tensor("current_value_17_strides_0"), val = tensor([1, 1])]; tensor current_value_17_pad_0 = const()[name = tensor("current_value_17_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_17_dilations_0 = const()[name = tensor("current_value_17_dilations_0"), val = tensor([1, 1])]; tensor current_value_17_groups_0 = const()[name = tensor("current_value_17_groups_0"), val = tensor(1)]; tensor layers_8_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211001152))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211787648))), name = tensor("layers_8_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211787840)))]; tensor current_value_17_cast_fp16 = conv(bias = layers_8_self_attn_v_proj_bias_to_fp16, dilations = current_value_17_dilations_0, groups = current_value_17_groups_0, pad = current_value_17_pad_0, pad_type = current_value_17_pad_type_0, strides = current_value_17_strides_0, weight = layers_8_self_attn_v_proj_weight_to_fp16_palettized, x = obj_113_cast_fp16)[name = tensor("current_value_17_cast_fp16")]; tensor var_1960_cast_fp16 = mul(x = var_87_cast_fp16_8, y = var_207_cast_fp16)[name = tensor("op_1960_cast_fp16")]; tensor var_1961_cast_fp16 = mul(x = current_key_17_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_1961_cast_fp16")]; tensor key_33_cast_fp16 = add(x = var_1960_cast_fp16, y = var_1961_cast_fp16)[name = tensor("key_33_cast_fp16")]; tensor var_1964_cast_fp16 = mul(x = var_114_cast_fp16_8, y = var_207_cast_fp16)[name = tensor("op_1964_cast_fp16")]; tensor var_1965_cast_fp16 = mul(x = current_value_17_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_1965_cast_fp16")]; tensor value_33_cast_fp16 = add(x = var_1964_cast_fp16, y = var_1965_cast_fp16)[name = tensor("value_33_cast_fp16")]; tensor var_1969 = const()[name = tensor("op_1969"), val = tensor([1, 16, 64, 1])]; tensor mh_q_33_cast_fp16 = reshape(shape = var_1969, x = query_33_cast_fp16)[name = tensor("mh_q_33_cast_fp16")]; tensor var_1971_to_fp16 = const()[name = tensor("op_1971_to_fp16"), val = tensor(0x1p-3)]; tensor var_1972_cast_fp16 = mul(x = mh_q_33_cast_fp16, y = var_1971_to_fp16)[name = tensor("op_1972_cast_fp16")]; tensor var_1975 = const()[name = tensor("op_1975"), val = tensor([1, 16, 64, 448])]; tensor var_1976_cast_fp16 = reshape(shape = var_1975, x = key_33_cast_fp16)[name = tensor("op_1976_cast_fp16")]; tensor mh_w_49_transpose_x_0 = const()[name = tensor("mh_w_49_transpose_x_0"), val = tensor(true)]; tensor mh_w_49_transpose_y_0 = const()[name = tensor("mh_w_49_transpose_y_0"), val = tensor(false)]; tensor mh_w_49_cast_fp16 = matmul(transpose_x = mh_w_49_transpose_x_0, transpose_y = mh_w_49_transpose_y_0, x = var_1972_cast_fp16, y = var_1976_cast_fp16)[name = tensor("mh_w_49_cast_fp16")]; tensor mh_w_51_cast_fp16 = add(x = mh_w_49_cast_fp16, y = var_229_cast_fp16)[name = tensor("mh_w_51_cast_fp16")]; tensor var_1984_cast_fp16 = softmax(axis = var_1896, x = mh_w_51_cast_fp16)[name = tensor("op_1984_cast_fp16")]; tensor var_1985 = const()[name = tensor("op_1985"), val = tensor([1, 16, 64, 448])]; tensor var_1986_cast_fp16 = reshape(shape = var_1985, x = value_33_cast_fp16)[name = tensor("op_1986_cast_fp16")]; tensor attn_33_transpose_x_0 = const()[name = tensor("attn_33_transpose_x_0"), val = tensor(false)]; tensor attn_33_transpose_y_0 = const()[name = tensor("attn_33_transpose_y_0"), val = tensor(true)]; tensor attn_33_cast_fp16 = matmul(transpose_x = attn_33_transpose_x_0, transpose_y = attn_33_transpose_y_0, x = var_1986_cast_fp16, y = var_1984_cast_fp16)[name = tensor("attn_33_cast_fp16")]; tensor var_1989 = const()[name = tensor("op_1989"), val = tensor([1, 1024, 1, 1])]; tensor input_81_cast_fp16 = reshape(shape = var_1989, x = attn_33_cast_fp16)[name = tensor("input_81_cast_fp16")]; tensor obj_119_pad_type_0 = const()[name = tensor("obj_119_pad_type_0"), val = tensor("valid")]; tensor obj_119_strides_0 = const()[name = tensor("obj_119_strides_0"), val = tensor([1, 1])]; tensor obj_119_pad_0 = const()[name = tensor("obj_119_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_119_dilations_0 = const()[name = tensor("obj_119_dilations_0"), val = tensor([1, 1])]; tensor obj_119_groups_0 = const()[name = tensor("obj_119_groups_0"), val = tensor(1)]; tensor layers_8_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211789952))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(212576448))), name = tensor("layers_8_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_8_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_8_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(212576640)))]; tensor obj_119_cast_fp16 = conv(bias = layers_8_self_attn_o_proj_bias_to_fp16, dilations = obj_119_dilations_0, groups = obj_119_groups_0, pad = obj_119_pad_0, pad_type = obj_119_pad_type_0, strides = obj_119_strides_0, weight = layers_8_self_attn_o_proj_weight_to_fp16_palettized, x = input_81_cast_fp16)[name = tensor("obj_119_cast_fp16")]; tensor inputs_51_cast_fp16 = add(x = inputs_49_cast_fp16, y = obj_119_cast_fp16)[name = tensor("inputs_51_cast_fp16")]; tensor out_51_axes_0 = const()[name = tensor("out_51_axes_0"), val = tensor([1])]; tensor var_2011_to_fp16 = const()[name = tensor("op_2011_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_51_cast_fp16 = layer_norm(axes = out_51_axes_0, epsilon = var_2011_to_fp16, x = inputs_51_cast_fp16)[name = tensor("out_51_cast_fp16")]; tensor obj_121_gamma_0_to_fp16 = const()[name = tensor("obj_121_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(212578752)))]; tensor obj_121_beta_0_to_fp16 = const()[name = tensor("obj_121_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(212580864)))]; tensor obj_121_epsilon_0_to_fp16 = const()[name = tensor("obj_121_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_121_cast_fp16 = batch_norm(beta = obj_121_beta_0_to_fp16, epsilon = obj_121_epsilon_0_to_fp16, gamma = obj_121_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_51_cast_fp16)[name = tensor("obj_121_cast_fp16")]; tensor query_35_pad_type_0 = const()[name = tensor("query_35_pad_type_0"), val = tensor("valid")]; tensor query_35_strides_0 = const()[name = tensor("query_35_strides_0"), val = tensor([1, 1])]; tensor query_35_pad_0 = const()[name = tensor("query_35_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_35_dilations_0 = const()[name = tensor("query_35_dilations_0"), val = tensor([1, 1])]; tensor query_35_groups_0 = const()[name = tensor("query_35_groups_0"), val = tensor(1)]; tensor layers_8_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(212582976))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(213369472))), name = tensor("layers_8_encoder_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_8_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_8_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(213369664)))]; tensor query_35_cast_fp16 = conv(bias = layers_8_encoder_attn_q_proj_bias_to_fp16, dilations = query_35_dilations_0, groups = query_35_groups_0, pad = query_35_pad_0, pad_type = query_35_pad_type_0, strides = query_35_strides_0, weight = layers_8_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_121_cast_fp16)[name = tensor("query_35_cast_fp16")]; tensor key_35_pad_type_0 = const()[name = tensor("key_35_pad_type_0"), val = tensor("valid")]; tensor key_35_strides_0 = const()[name = tensor("key_35_strides_0"), val = tensor([1, 1])]; tensor key_35_pad_0 = const()[name = tensor("key_35_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_35_dilations_0 = const()[name = tensor("key_35_dilations_0"), val = tensor([1, 1])]; tensor key_35_groups_0 = const()[name = tensor("key_35_groups_0"), val = tensor(1)]; tensor layers_8_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(213371776))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(214158272))), name = tensor("layers_8_encoder_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor key_35_cast_fp16 = conv(dilations = key_35_dilations_0, groups = key_35_groups_0, pad = key_35_pad_0, pad_type = key_35_pad_type_0, strides = key_35_strides_0, weight = layers_8_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("key_35_cast_fp16")]; tensor value_35_pad_type_0 = const()[name = tensor("value_35_pad_type_0"), val = tensor("valid")]; tensor value_35_strides_0 = const()[name = tensor("value_35_strides_0"), val = tensor([1, 1])]; tensor value_35_pad_0 = const()[name = tensor("value_35_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_35_dilations_0 = const()[name = tensor("value_35_dilations_0"), val = tensor([1, 1])]; tensor value_35_groups_0 = const()[name = tensor("value_35_groups_0"), val = tensor(1)]; tensor layers_8_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(214158464))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(214944960))), name = tensor("layers_8_encoder_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_8_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_8_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(214945152)))]; tensor value_35_cast_fp16 = conv(bias = layers_8_encoder_attn_v_proj_bias_to_fp16, dilations = value_35_dilations_0, groups = value_35_groups_0, pad = value_35_pad_0, pad_type = value_35_pad_type_0, strides = value_35_strides_0, weight = layers_8_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("value_35_cast_fp16")]; tensor var_2047 = const()[name = tensor("op_2047"), val = tensor([1, 16, 64, 1])]; tensor mh_q_35_cast_fp16 = reshape(shape = var_2047, x = query_35_cast_fp16)[name = tensor("mh_q_35_cast_fp16")]; tensor var_2049_to_fp16 = const()[name = tensor("op_2049_to_fp16"), val = tensor(0x1p-3)]; tensor var_2050_cast_fp16 = mul(x = mh_q_35_cast_fp16, y = var_2049_to_fp16)[name = tensor("op_2050_cast_fp16")]; tensor var_2053 = const()[name = tensor("op_2053"), val = tensor([1, 16, 64, 1500])]; tensor var_2054_cast_fp16 = reshape(shape = var_2053, x = key_35_cast_fp16)[name = tensor("op_2054_cast_fp16")]; tensor mh_w_53_transpose_x_0 = const()[name = tensor("mh_w_53_transpose_x_0"), val = tensor(true)]; tensor mh_w_53_transpose_y_0 = const()[name = tensor("mh_w_53_transpose_y_0"), val = tensor(false)]; tensor mh_w_53_cast_fp16 = matmul(transpose_x = mh_w_53_transpose_x_0, transpose_y = mh_w_53_transpose_y_0, x = var_2050_cast_fp16, y = var_2054_cast_fp16)[name = tensor("mh_w_53_cast_fp16")]; tensor obj_125_cast_fp16 = softmax(axis = var_1896, x = mh_w_53_cast_fp16)[name = tensor("obj_125_cast_fp16")]; tensor var_2058 = const()[name = tensor("op_2058"), val = tensor([1, 16, 64, 1500])]; tensor var_2059_cast_fp16 = reshape(shape = var_2058, x = value_35_cast_fp16)[name = tensor("op_2059_cast_fp16")]; tensor attn_35_transpose_x_0 = const()[name = tensor("attn_35_transpose_x_0"), val = tensor(false)]; tensor attn_35_transpose_y_0 = const()[name = tensor("attn_35_transpose_y_0"), val = tensor(true)]; tensor attn_35_cast_fp16 = matmul(transpose_x = attn_35_transpose_x_0, transpose_y = attn_35_transpose_y_0, x = var_2059_cast_fp16, y = obj_125_cast_fp16)[name = tensor("attn_35_cast_fp16")]; tensor var_2062 = const()[name = tensor("op_2062"), val = tensor([1, 1024, 1, 1])]; tensor input_83_cast_fp16 = reshape(shape = var_2062, x = attn_35_cast_fp16)[name = tensor("input_83_cast_fp16")]; tensor obj_123_pad_type_0 = const()[name = tensor("obj_123_pad_type_0"), val = tensor("valid")]; tensor obj_123_strides_0 = const()[name = tensor("obj_123_strides_0"), val = tensor([1, 1])]; tensor obj_123_pad_0 = const()[name = tensor("obj_123_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_123_dilations_0 = const()[name = tensor("obj_123_dilations_0"), val = tensor([1, 1])]; tensor obj_123_groups_0 = const()[name = tensor("obj_123_groups_0"), val = tensor(1)]; tensor layers_8_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(214947264))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215733760))), name = tensor("layers_8_encoder_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_8_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_8_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215733952)))]; tensor obj_123_cast_fp16 = conv(bias = layers_8_encoder_attn_o_proj_bias_to_fp16, dilations = obj_123_dilations_0, groups = obj_123_groups_0, pad = obj_123_pad_0, pad_type = obj_123_pad_type_0, strides = obj_123_strides_0, weight = layers_8_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_83_cast_fp16)[name = tensor("obj_123_cast_fp16")]; tensor inputs_53_cast_fp16 = add(x = inputs_51_cast_fp16, y = obj_123_cast_fp16)[name = tensor("inputs_53_cast_fp16")]; tensor out_53_axes_0 = const()[name = tensor("out_53_axes_0"), val = tensor([1])]; tensor var_2080_to_fp16 = const()[name = tensor("op_2080_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_53_cast_fp16 = layer_norm(axes = out_53_axes_0, epsilon = var_2080_to_fp16, x = inputs_53_cast_fp16)[name = tensor("out_53_cast_fp16")]; tensor input_85_gamma_0_to_fp16 = const()[name = tensor("input_85_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215736064)))]; tensor input_85_beta_0_to_fp16 = const()[name = tensor("input_85_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215738176)))]; tensor input_85_epsilon_0_to_fp16 = const()[name = tensor("input_85_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_85_cast_fp16 = batch_norm(beta = input_85_beta_0_to_fp16, epsilon = input_85_epsilon_0_to_fp16, gamma = input_85_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_53_cast_fp16)[name = tensor("input_85_cast_fp16")]; tensor input_87_pad_type_0 = const()[name = tensor("input_87_pad_type_0"), val = tensor("valid")]; tensor input_87_strides_0 = const()[name = tensor("input_87_strides_0"), val = tensor([1, 1])]; tensor input_87_pad_0 = const()[name = tensor("input_87_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_87_dilations_0 = const()[name = tensor("input_87_dilations_0"), val = tensor([1, 1])]; tensor input_87_groups_0 = const()[name = tensor("input_87_groups_0"), val = tensor(1)]; tensor layers_8_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215740288))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(218886080))), name = tensor("layers_8_fc1_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; tensor layers_8_fc1_bias_to_fp16 = const()[name = tensor("layers_8_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(218886272)))]; tensor input_87_cast_fp16 = conv(bias = layers_8_fc1_bias_to_fp16, dilations = input_87_dilations_0, groups = input_87_groups_0, pad = input_87_pad_0, pad_type = input_87_pad_type_0, strides = input_87_strides_0, weight = layers_8_fc1_weight_to_fp16_palettized, x = input_85_cast_fp16)[name = tensor("input_87_cast_fp16")]; tensor input_89_mode_0 = const()[name = tensor("input_89_mode_0"), val = tensor("EXACT")]; tensor input_89_cast_fp16 = gelu(mode = input_89_mode_0, x = input_87_cast_fp16)[name = tensor("input_89_cast_fp16")]; tensor hidden_states_19_pad_type_0 = const()[name = tensor("hidden_states_19_pad_type_0"), val = tensor("valid")]; tensor hidden_states_19_strides_0 = const()[name = tensor("hidden_states_19_strides_0"), val = tensor([1, 1])]; tensor hidden_states_19_pad_0 = const()[name = tensor("hidden_states_19_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_19_dilations_0 = const()[name = tensor("hidden_states_19_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_19_groups_0 = const()[name = tensor("hidden_states_19_groups_0"), val = tensor(1)]; tensor layers_8_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(218894528))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222040320))), name = tensor("layers_8_fc2_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; tensor layers_8_fc2_bias_to_fp16 = const()[name = tensor("layers_8_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222040512)))]; tensor hidden_states_19_cast_fp16 = conv(bias = layers_8_fc2_bias_to_fp16, dilations = hidden_states_19_dilations_0, groups = hidden_states_19_groups_0, pad = hidden_states_19_pad_0, pad_type = hidden_states_19_pad_type_0, strides = hidden_states_19_strides_0, weight = layers_8_fc2_weight_to_fp16_palettized, x = input_89_cast_fp16)[name = tensor("hidden_states_19_cast_fp16")]; tensor inputs_55_cast_fp16 = add(x = inputs_53_cast_fp16, y = hidden_states_19_cast_fp16)[name = tensor("inputs_55_cast_fp16")]; tensor var_2115 = const()[name = tensor("op_2115"), val = tensor(3)]; tensor out_55_axes_0 = const()[name = tensor("out_55_axes_0"), val = tensor([1])]; tensor var_2140_to_fp16 = const()[name = tensor("op_2140_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_55_cast_fp16 = layer_norm(axes = out_55_axes_0, epsilon = var_2140_to_fp16, x = inputs_55_cast_fp16)[name = tensor("out_55_cast_fp16")]; tensor obj_127_gamma_0_to_fp16 = const()[name = tensor("obj_127_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222042624)))]; tensor obj_127_beta_0_to_fp16 = const()[name = tensor("obj_127_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222044736)))]; tensor obj_127_epsilon_0_to_fp16 = const()[name = tensor("obj_127_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_127_cast_fp16 = batch_norm(beta = obj_127_beta_0_to_fp16, epsilon = obj_127_epsilon_0_to_fp16, gamma = obj_127_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_55_cast_fp16)[name = tensor("obj_127_cast_fp16")]; tensor query_37_pad_type_0 = const()[name = tensor("query_37_pad_type_0"), val = tensor("valid")]; tensor query_37_strides_0 = const()[name = tensor("query_37_strides_0"), val = tensor([1, 1])]; tensor query_37_pad_0 = const()[name = tensor("query_37_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_37_dilations_0 = const()[name = tensor("query_37_dilations_0"), val = tensor([1, 1])]; tensor query_37_groups_0 = const()[name = tensor("query_37_groups_0"), val = tensor(1)]; tensor layers_9_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222046848))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222833344))), name = tensor("layers_9_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222833536)))]; tensor query_37_cast_fp16 = conv(bias = layers_9_self_attn_q_proj_bias_to_fp16, dilations = query_37_dilations_0, groups = query_37_groups_0, pad = query_37_pad_0, pad_type = query_37_pad_type_0, strides = query_37_strides_0, weight = layers_9_self_attn_q_proj_weight_to_fp16_palettized, x = obj_127_cast_fp16)[name = tensor("query_37_cast_fp16")]; tensor current_key_19_pad_type_0 = const()[name = tensor("current_key_19_pad_type_0"), val = tensor("valid")]; tensor current_key_19_strides_0 = const()[name = tensor("current_key_19_strides_0"), val = tensor([1, 1])]; tensor current_key_19_pad_0 = const()[name = tensor("current_key_19_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_19_dilations_0 = const()[name = tensor("current_key_19_dilations_0"), val = tensor([1, 1])]; tensor current_key_19_groups_0 = const()[name = tensor("current_key_19_groups_0"), val = tensor(1)]; tensor layers_9_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222835648))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(223622144))), name = tensor("layers_9_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor current_key_19_cast_fp16 = conv(dilations = current_key_19_dilations_0, groups = current_key_19_groups_0, pad = current_key_19_pad_0, pad_type = current_key_19_pad_type_0, strides = current_key_19_strides_0, weight = layers_9_self_attn_k_proj_weight_to_fp16_palettized, x = obj_127_cast_fp16)[name = tensor("current_key_19_cast_fp16")]; tensor current_value_19_pad_type_0 = const()[name = tensor("current_value_19_pad_type_0"), val = tensor("valid")]; tensor current_value_19_strides_0 = const()[name = tensor("current_value_19_strides_0"), val = tensor([1, 1])]; tensor current_value_19_pad_0 = const()[name = tensor("current_value_19_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_19_dilations_0 = const()[name = tensor("current_value_19_dilations_0"), val = tensor([1, 1])]; tensor current_value_19_groups_0 = const()[name = tensor("current_value_19_groups_0"), val = tensor(1)]; tensor layers_9_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(223622336))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224408832))), name = tensor("layers_9_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224409024)))]; tensor current_value_19_cast_fp16 = conv(bias = layers_9_self_attn_v_proj_bias_to_fp16, dilations = current_value_19_dilations_0, groups = current_value_19_groups_0, pad = current_value_19_pad_0, pad_type = current_value_19_pad_type_0, strides = current_value_19_strides_0, weight = layers_9_self_attn_v_proj_weight_to_fp16_palettized, x = obj_127_cast_fp16)[name = tensor("current_value_19_cast_fp16")]; tensor var_2179_cast_fp16 = mul(x = var_87_cast_fp16_9, y = var_207_cast_fp16)[name = tensor("op_2179_cast_fp16")]; tensor var_2180_cast_fp16 = mul(x = current_key_19_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_2180_cast_fp16")]; tensor key_37_cast_fp16 = add(x = var_2179_cast_fp16, y = var_2180_cast_fp16)[name = tensor("key_37_cast_fp16")]; tensor var_2183_cast_fp16 = mul(x = var_114_cast_fp16_9, y = var_207_cast_fp16)[name = tensor("op_2183_cast_fp16")]; tensor var_2184_cast_fp16 = mul(x = current_value_19_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_2184_cast_fp16")]; tensor value_37_cast_fp16 = add(x = var_2183_cast_fp16, y = var_2184_cast_fp16)[name = tensor("value_37_cast_fp16")]; tensor var_2188 = const()[name = tensor("op_2188"), val = tensor([1, 16, 64, 1])]; tensor mh_q_37_cast_fp16 = reshape(shape = var_2188, x = query_37_cast_fp16)[name = tensor("mh_q_37_cast_fp16")]; tensor var_2190_to_fp16 = const()[name = tensor("op_2190_to_fp16"), val = tensor(0x1p-3)]; tensor var_2191_cast_fp16 = mul(x = mh_q_37_cast_fp16, y = var_2190_to_fp16)[name = tensor("op_2191_cast_fp16")]; tensor var_2194 = const()[name = tensor("op_2194"), val = tensor([1, 16, 64, 448])]; tensor var_2195_cast_fp16 = reshape(shape = var_2194, x = key_37_cast_fp16)[name = tensor("op_2195_cast_fp16")]; tensor mh_w_55_transpose_x_0 = const()[name = tensor("mh_w_55_transpose_x_0"), val = tensor(true)]; tensor mh_w_55_transpose_y_0 = const()[name = tensor("mh_w_55_transpose_y_0"), val = tensor(false)]; tensor mh_w_55_cast_fp16 = matmul(transpose_x = mh_w_55_transpose_x_0, transpose_y = mh_w_55_transpose_y_0, x = var_2191_cast_fp16, y = var_2195_cast_fp16)[name = tensor("mh_w_55_cast_fp16")]; tensor mh_w_57_cast_fp16 = add(x = mh_w_55_cast_fp16, y = var_229_cast_fp16)[name = tensor("mh_w_57_cast_fp16")]; tensor var_2203_cast_fp16 = softmax(axis = var_2115, x = mh_w_57_cast_fp16)[name = tensor("op_2203_cast_fp16")]; tensor var_2204 = const()[name = tensor("op_2204"), val = tensor([1, 16, 64, 448])]; tensor var_2205_cast_fp16 = reshape(shape = var_2204, x = value_37_cast_fp16)[name = tensor("op_2205_cast_fp16")]; tensor attn_37_transpose_x_0 = const()[name = tensor("attn_37_transpose_x_0"), val = tensor(false)]; tensor attn_37_transpose_y_0 = const()[name = tensor("attn_37_transpose_y_0"), val = tensor(true)]; tensor attn_37_cast_fp16 = matmul(transpose_x = attn_37_transpose_x_0, transpose_y = attn_37_transpose_y_0, x = var_2205_cast_fp16, y = var_2203_cast_fp16)[name = tensor("attn_37_cast_fp16")]; tensor var_2208 = const()[name = tensor("op_2208"), val = tensor([1, 1024, 1, 1])]; tensor input_91_cast_fp16 = reshape(shape = var_2208, x = attn_37_cast_fp16)[name = tensor("input_91_cast_fp16")]; tensor obj_133_pad_type_0 = const()[name = tensor("obj_133_pad_type_0"), val = tensor("valid")]; tensor obj_133_strides_0 = const()[name = tensor("obj_133_strides_0"), val = tensor([1, 1])]; tensor obj_133_pad_0 = const()[name = tensor("obj_133_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_133_dilations_0 = const()[name = tensor("obj_133_dilations_0"), val = tensor([1, 1])]; tensor obj_133_groups_0 = const()[name = tensor("obj_133_groups_0"), val = tensor(1)]; tensor layers_9_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224411136))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(225197632))), name = tensor("layers_9_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_9_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_9_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(225197824)))]; tensor obj_133_cast_fp16 = conv(bias = layers_9_self_attn_o_proj_bias_to_fp16, dilations = obj_133_dilations_0, groups = obj_133_groups_0, pad = obj_133_pad_0, pad_type = obj_133_pad_type_0, strides = obj_133_strides_0, weight = layers_9_self_attn_o_proj_weight_to_fp16_palettized, x = input_91_cast_fp16)[name = tensor("obj_133_cast_fp16")]; tensor inputs_57_cast_fp16 = add(x = inputs_55_cast_fp16, y = obj_133_cast_fp16)[name = tensor("inputs_57_cast_fp16")]; tensor out_57_axes_0 = const()[name = tensor("out_57_axes_0"), val = tensor([1])]; tensor var_2230_to_fp16 = const()[name = tensor("op_2230_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_57_cast_fp16 = layer_norm(axes = out_57_axes_0, epsilon = var_2230_to_fp16, x = inputs_57_cast_fp16)[name = tensor("out_57_cast_fp16")]; tensor obj_135_gamma_0_to_fp16 = const()[name = tensor("obj_135_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(225199936)))]; tensor obj_135_beta_0_to_fp16 = const()[name = tensor("obj_135_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(225202048)))]; tensor obj_135_epsilon_0_to_fp16 = const()[name = tensor("obj_135_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_135_cast_fp16 = batch_norm(beta = obj_135_beta_0_to_fp16, epsilon = obj_135_epsilon_0_to_fp16, gamma = obj_135_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_57_cast_fp16)[name = tensor("obj_135_cast_fp16")]; tensor query_39_pad_type_0 = const()[name = tensor("query_39_pad_type_0"), val = tensor("valid")]; tensor query_39_strides_0 = const()[name = tensor("query_39_strides_0"), val = tensor([1, 1])]; tensor query_39_pad_0 = const()[name = tensor("query_39_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_39_dilations_0 = const()[name = tensor("query_39_dilations_0"), val = tensor([1, 1])]; tensor query_39_groups_0 = const()[name = tensor("query_39_groups_0"), val = tensor(1)]; tensor layers_9_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(225204160))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(225990656))), name = tensor("layers_9_encoder_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_9_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_9_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(225990848)))]; tensor query_39_cast_fp16 = conv(bias = layers_9_encoder_attn_q_proj_bias_to_fp16, dilations = query_39_dilations_0, groups = query_39_groups_0, pad = query_39_pad_0, pad_type = query_39_pad_type_0, strides = query_39_strides_0, weight = layers_9_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_135_cast_fp16)[name = tensor("query_39_cast_fp16")]; tensor key_39_pad_type_0 = const()[name = tensor("key_39_pad_type_0"), val = tensor("valid")]; tensor key_39_strides_0 = const()[name = tensor("key_39_strides_0"), val = tensor([1, 1])]; tensor key_39_pad_0 = const()[name = tensor("key_39_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_39_dilations_0 = const()[name = tensor("key_39_dilations_0"), val = tensor([1, 1])]; tensor key_39_groups_0 = const()[name = tensor("key_39_groups_0"), val = tensor(1)]; tensor layers_9_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(225992960))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(226779456))), name = tensor("layers_9_encoder_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor key_39_cast_fp16 = conv(dilations = key_39_dilations_0, groups = key_39_groups_0, pad = key_39_pad_0, pad_type = key_39_pad_type_0, strides = key_39_strides_0, weight = layers_9_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("key_39_cast_fp16")]; tensor value_39_pad_type_0 = const()[name = tensor("value_39_pad_type_0"), val = tensor("valid")]; tensor value_39_strides_0 = const()[name = tensor("value_39_strides_0"), val = tensor([1, 1])]; tensor value_39_pad_0 = const()[name = tensor("value_39_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_39_dilations_0 = const()[name = tensor("value_39_dilations_0"), val = tensor([1, 1])]; tensor value_39_groups_0 = const()[name = tensor("value_39_groups_0"), val = tensor(1)]; tensor layers_9_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(226779648))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(227566144))), name = tensor("layers_9_encoder_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_9_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_9_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(227566336)))]; tensor value_39_cast_fp16 = conv(bias = layers_9_encoder_attn_v_proj_bias_to_fp16, dilations = value_39_dilations_0, groups = value_39_groups_0, pad = value_39_pad_0, pad_type = value_39_pad_type_0, strides = value_39_strides_0, weight = layers_9_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("value_39_cast_fp16")]; tensor var_2266 = const()[name = tensor("op_2266"), val = tensor([1, 16, 64, 1])]; tensor mh_q_39_cast_fp16 = reshape(shape = var_2266, x = query_39_cast_fp16)[name = tensor("mh_q_39_cast_fp16")]; tensor var_2268_to_fp16 = const()[name = tensor("op_2268_to_fp16"), val = tensor(0x1p-3)]; tensor var_2269_cast_fp16 = mul(x = mh_q_39_cast_fp16, y = var_2268_to_fp16)[name = tensor("op_2269_cast_fp16")]; tensor var_2272 = const()[name = tensor("op_2272"), val = tensor([1, 16, 64, 1500])]; tensor var_2273_cast_fp16 = reshape(shape = var_2272, x = key_39_cast_fp16)[name = tensor("op_2273_cast_fp16")]; tensor mh_w_59_transpose_x_0 = const()[name = tensor("mh_w_59_transpose_x_0"), val = tensor(true)]; tensor mh_w_59_transpose_y_0 = const()[name = tensor("mh_w_59_transpose_y_0"), val = tensor(false)]; tensor mh_w_59_cast_fp16 = matmul(transpose_x = mh_w_59_transpose_x_0, transpose_y = mh_w_59_transpose_y_0, x = var_2269_cast_fp16, y = var_2273_cast_fp16)[name = tensor("mh_w_59_cast_fp16")]; tensor obj_139_cast_fp16 = softmax(axis = var_2115, x = mh_w_59_cast_fp16)[name = tensor("obj_139_cast_fp16")]; tensor var_2277 = const()[name = tensor("op_2277"), val = tensor([1, 16, 64, 1500])]; tensor var_2278_cast_fp16 = reshape(shape = var_2277, x = value_39_cast_fp16)[name = tensor("op_2278_cast_fp16")]; tensor attn_39_transpose_x_0 = const()[name = tensor("attn_39_transpose_x_0"), val = tensor(false)]; tensor attn_39_transpose_y_0 = const()[name = tensor("attn_39_transpose_y_0"), val = tensor(true)]; tensor attn_39_cast_fp16 = matmul(transpose_x = attn_39_transpose_x_0, transpose_y = attn_39_transpose_y_0, x = var_2278_cast_fp16, y = obj_139_cast_fp16)[name = tensor("attn_39_cast_fp16")]; tensor var_2281 = const()[name = tensor("op_2281"), val = tensor([1, 1024, 1, 1])]; tensor input_93_cast_fp16 = reshape(shape = var_2281, x = attn_39_cast_fp16)[name = tensor("input_93_cast_fp16")]; tensor obj_137_pad_type_0 = const()[name = tensor("obj_137_pad_type_0"), val = tensor("valid")]; tensor obj_137_strides_0 = const()[name = tensor("obj_137_strides_0"), val = tensor([1, 1])]; tensor obj_137_pad_0 = const()[name = tensor("obj_137_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_137_dilations_0 = const()[name = tensor("obj_137_dilations_0"), val = tensor([1, 1])]; tensor obj_137_groups_0 = const()[name = tensor("obj_137_groups_0"), val = tensor(1)]; tensor layers_9_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(227568448))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(228354944))), name = tensor("layers_9_encoder_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_9_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_9_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(228355136)))]; tensor obj_137_cast_fp16 = conv(bias = layers_9_encoder_attn_o_proj_bias_to_fp16, dilations = obj_137_dilations_0, groups = obj_137_groups_0, pad = obj_137_pad_0, pad_type = obj_137_pad_type_0, strides = obj_137_strides_0, weight = layers_9_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_93_cast_fp16)[name = tensor("obj_137_cast_fp16")]; tensor inputs_59_cast_fp16 = add(x = inputs_57_cast_fp16, y = obj_137_cast_fp16)[name = tensor("inputs_59_cast_fp16")]; tensor out_59_axes_0 = const()[name = tensor("out_59_axes_0"), val = tensor([1])]; tensor var_2299_to_fp16 = const()[name = tensor("op_2299_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_59_cast_fp16 = layer_norm(axes = out_59_axes_0, epsilon = var_2299_to_fp16, x = inputs_59_cast_fp16)[name = tensor("out_59_cast_fp16")]; tensor input_95_gamma_0_to_fp16 = const()[name = tensor("input_95_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(228357248)))]; tensor input_95_beta_0_to_fp16 = const()[name = tensor("input_95_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(228359360)))]; tensor input_95_epsilon_0_to_fp16 = const()[name = tensor("input_95_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_95_cast_fp16 = batch_norm(beta = input_95_beta_0_to_fp16, epsilon = input_95_epsilon_0_to_fp16, gamma = input_95_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_59_cast_fp16)[name = tensor("input_95_cast_fp16")]; tensor input_97_pad_type_0 = const()[name = tensor("input_97_pad_type_0"), val = tensor("valid")]; tensor input_97_strides_0 = const()[name = tensor("input_97_strides_0"), val = tensor([1, 1])]; tensor input_97_pad_0 = const()[name = tensor("input_97_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_97_dilations_0 = const()[name = tensor("input_97_dilations_0"), val = tensor([1, 1])]; tensor input_97_groups_0 = const()[name = tensor("input_97_groups_0"), val = tensor(1)]; tensor layers_9_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(228361472))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(231507264))), name = tensor("layers_9_fc1_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; tensor layers_9_fc1_bias_to_fp16 = const()[name = tensor("layers_9_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(231507456)))]; tensor input_97_cast_fp16 = conv(bias = layers_9_fc1_bias_to_fp16, dilations = input_97_dilations_0, groups = input_97_groups_0, pad = input_97_pad_0, pad_type = input_97_pad_type_0, strides = input_97_strides_0, weight = layers_9_fc1_weight_to_fp16_palettized, x = input_95_cast_fp16)[name = tensor("input_97_cast_fp16")]; tensor input_99_mode_0 = const()[name = tensor("input_99_mode_0"), val = tensor("EXACT")]; tensor input_99_cast_fp16 = gelu(mode = input_99_mode_0, x = input_97_cast_fp16)[name = tensor("input_99_cast_fp16")]; tensor hidden_states_21_pad_type_0 = const()[name = tensor("hidden_states_21_pad_type_0"), val = tensor("valid")]; tensor hidden_states_21_strides_0 = const()[name = tensor("hidden_states_21_strides_0"), val = tensor([1, 1])]; tensor hidden_states_21_pad_0 = const()[name = tensor("hidden_states_21_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_21_dilations_0 = const()[name = tensor("hidden_states_21_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_21_groups_0 = const()[name = tensor("hidden_states_21_groups_0"), val = tensor(1)]; tensor layers_9_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(231515712))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(234661504))), name = tensor("layers_9_fc2_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; tensor layers_9_fc2_bias_to_fp16 = const()[name = tensor("layers_9_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(234661696)))]; tensor hidden_states_21_cast_fp16 = conv(bias = layers_9_fc2_bias_to_fp16, dilations = hidden_states_21_dilations_0, groups = hidden_states_21_groups_0, pad = hidden_states_21_pad_0, pad_type = hidden_states_21_pad_type_0, strides = hidden_states_21_strides_0, weight = layers_9_fc2_weight_to_fp16_palettized, x = input_99_cast_fp16)[name = tensor("hidden_states_21_cast_fp16")]; tensor inputs_61_cast_fp16 = add(x = inputs_59_cast_fp16, y = hidden_states_21_cast_fp16)[name = tensor("inputs_61_cast_fp16")]; tensor var_2334 = const()[name = tensor("op_2334"), val = tensor(3)]; tensor out_61_axes_0 = const()[name = tensor("out_61_axes_0"), val = tensor([1])]; tensor var_2359_to_fp16 = const()[name = tensor("op_2359_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_61_cast_fp16 = layer_norm(axes = out_61_axes_0, epsilon = var_2359_to_fp16, x = inputs_61_cast_fp16)[name = tensor("out_61_cast_fp16")]; tensor obj_141_gamma_0_to_fp16 = const()[name = tensor("obj_141_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(234663808)))]; tensor obj_141_beta_0_to_fp16 = const()[name = tensor("obj_141_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(234665920)))]; tensor obj_141_epsilon_0_to_fp16 = const()[name = tensor("obj_141_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_141_cast_fp16 = batch_norm(beta = obj_141_beta_0_to_fp16, epsilon = obj_141_epsilon_0_to_fp16, gamma = obj_141_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_61_cast_fp16)[name = tensor("obj_141_cast_fp16")]; tensor query_41_pad_type_0 = const()[name = tensor("query_41_pad_type_0"), val = tensor("valid")]; tensor query_41_strides_0 = const()[name = tensor("query_41_strides_0"), val = tensor([1, 1])]; tensor query_41_pad_0 = const()[name = tensor("query_41_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_41_dilations_0 = const()[name = tensor("query_41_dilations_0"), val = tensor([1, 1])]; tensor query_41_groups_0 = const()[name = tensor("query_41_groups_0"), val = tensor(1)]; tensor layers_10_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(234668032))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(235454528))), name = tensor("layers_10_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(235454720)))]; tensor query_41_cast_fp16 = conv(bias = layers_10_self_attn_q_proj_bias_to_fp16, dilations = query_41_dilations_0, groups = query_41_groups_0, pad = query_41_pad_0, pad_type = query_41_pad_type_0, strides = query_41_strides_0, weight = layers_10_self_attn_q_proj_weight_to_fp16_palettized, x = obj_141_cast_fp16)[name = tensor("query_41_cast_fp16")]; tensor current_key_21_pad_type_0 = const()[name = tensor("current_key_21_pad_type_0"), val = tensor("valid")]; tensor current_key_21_strides_0 = const()[name = tensor("current_key_21_strides_0"), val = tensor([1, 1])]; tensor current_key_21_pad_0 = const()[name = tensor("current_key_21_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_21_dilations_0 = const()[name = tensor("current_key_21_dilations_0"), val = tensor([1, 1])]; tensor current_key_21_groups_0 = const()[name = tensor("current_key_21_groups_0"), val = tensor(1)]; tensor layers_10_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(235456832))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236243328))), name = tensor("layers_10_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor current_key_21_cast_fp16 = conv(dilations = current_key_21_dilations_0, groups = current_key_21_groups_0, pad = current_key_21_pad_0, pad_type = current_key_21_pad_type_0, strides = current_key_21_strides_0, weight = layers_10_self_attn_k_proj_weight_to_fp16_palettized, x = obj_141_cast_fp16)[name = tensor("current_key_21_cast_fp16")]; tensor current_value_21_pad_type_0 = const()[name = tensor("current_value_21_pad_type_0"), val = tensor("valid")]; tensor current_value_21_strides_0 = const()[name = tensor("current_value_21_strides_0"), val = tensor([1, 1])]; tensor current_value_21_pad_0 = const()[name = tensor("current_value_21_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_21_dilations_0 = const()[name = tensor("current_value_21_dilations_0"), val = tensor([1, 1])]; tensor current_value_21_groups_0 = const()[name = tensor("current_value_21_groups_0"), val = tensor(1)]; tensor layers_10_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236243520))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237030016))), name = tensor("layers_10_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237030208)))]; tensor current_value_21_cast_fp16 = conv(bias = layers_10_self_attn_v_proj_bias_to_fp16, dilations = current_value_21_dilations_0, groups = current_value_21_groups_0, pad = current_value_21_pad_0, pad_type = current_value_21_pad_type_0, strides = current_value_21_strides_0, weight = layers_10_self_attn_v_proj_weight_to_fp16_palettized, x = obj_141_cast_fp16)[name = tensor("current_value_21_cast_fp16")]; tensor var_2398_cast_fp16 = mul(x = var_87_cast_fp16_10, y = var_207_cast_fp16)[name = tensor("op_2398_cast_fp16")]; tensor var_2399_cast_fp16 = mul(x = current_key_21_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_2399_cast_fp16")]; tensor key_41_cast_fp16 = add(x = var_2398_cast_fp16, y = var_2399_cast_fp16)[name = tensor("key_41_cast_fp16")]; tensor var_2402_cast_fp16 = mul(x = var_114_cast_fp16_10, y = var_207_cast_fp16)[name = tensor("op_2402_cast_fp16")]; tensor var_2403_cast_fp16 = mul(x = current_value_21_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_2403_cast_fp16")]; tensor value_41_cast_fp16 = add(x = var_2402_cast_fp16, y = var_2403_cast_fp16)[name = tensor("value_41_cast_fp16")]; tensor var_2407 = const()[name = tensor("op_2407"), val = tensor([1, 16, 64, 1])]; tensor mh_q_41_cast_fp16 = reshape(shape = var_2407, x = query_41_cast_fp16)[name = tensor("mh_q_41_cast_fp16")]; tensor var_2409_to_fp16 = const()[name = tensor("op_2409_to_fp16"), val = tensor(0x1p-3)]; tensor var_2410_cast_fp16 = mul(x = mh_q_41_cast_fp16, y = var_2409_to_fp16)[name = tensor("op_2410_cast_fp16")]; tensor var_2413 = const()[name = tensor("op_2413"), val = tensor([1, 16, 64, 448])]; tensor var_2414_cast_fp16 = reshape(shape = var_2413, x = key_41_cast_fp16)[name = tensor("op_2414_cast_fp16")]; tensor mh_w_61_transpose_x_0 = const()[name = tensor("mh_w_61_transpose_x_0"), val = tensor(true)]; tensor mh_w_61_transpose_y_0 = const()[name = tensor("mh_w_61_transpose_y_0"), val = tensor(false)]; tensor mh_w_61_cast_fp16 = matmul(transpose_x = mh_w_61_transpose_x_0, transpose_y = mh_w_61_transpose_y_0, x = var_2410_cast_fp16, y = var_2414_cast_fp16)[name = tensor("mh_w_61_cast_fp16")]; tensor mh_w_63_cast_fp16 = add(x = mh_w_61_cast_fp16, y = var_229_cast_fp16)[name = tensor("mh_w_63_cast_fp16")]; tensor var_2422_cast_fp16 = softmax(axis = var_2334, x = mh_w_63_cast_fp16)[name = tensor("op_2422_cast_fp16")]; tensor var_2423 = const()[name = tensor("op_2423"), val = tensor([1, 16, 64, 448])]; tensor var_2424_cast_fp16 = reshape(shape = var_2423, x = value_41_cast_fp16)[name = tensor("op_2424_cast_fp16")]; tensor attn_41_transpose_x_0 = const()[name = tensor("attn_41_transpose_x_0"), val = tensor(false)]; tensor attn_41_transpose_y_0 = const()[name = tensor("attn_41_transpose_y_0"), val = tensor(true)]; tensor attn_41_cast_fp16 = matmul(transpose_x = attn_41_transpose_x_0, transpose_y = attn_41_transpose_y_0, x = var_2424_cast_fp16, y = var_2422_cast_fp16)[name = tensor("attn_41_cast_fp16")]; tensor var_2427 = const()[name = tensor("op_2427"), val = tensor([1, 1024, 1, 1])]; tensor input_101_cast_fp16 = reshape(shape = var_2427, x = attn_41_cast_fp16)[name = tensor("input_101_cast_fp16")]; tensor obj_147_pad_type_0 = const()[name = tensor("obj_147_pad_type_0"), val = tensor("valid")]; tensor obj_147_strides_0 = const()[name = tensor("obj_147_strides_0"), val = tensor([1, 1])]; tensor obj_147_pad_0 = const()[name = tensor("obj_147_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_147_dilations_0 = const()[name = tensor("obj_147_dilations_0"), val = tensor([1, 1])]; tensor obj_147_groups_0 = const()[name = tensor("obj_147_groups_0"), val = tensor(1)]; tensor layers_10_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237032320))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237818816))), name = tensor("layers_10_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_10_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_10_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237819008)))]; tensor obj_147_cast_fp16 = conv(bias = layers_10_self_attn_o_proj_bias_to_fp16, dilations = obj_147_dilations_0, groups = obj_147_groups_0, pad = obj_147_pad_0, pad_type = obj_147_pad_type_0, strides = obj_147_strides_0, weight = layers_10_self_attn_o_proj_weight_to_fp16_palettized, x = input_101_cast_fp16)[name = tensor("obj_147_cast_fp16")]; tensor inputs_63_cast_fp16 = add(x = inputs_61_cast_fp16, y = obj_147_cast_fp16)[name = tensor("inputs_63_cast_fp16")]; tensor out_63_axes_0 = const()[name = tensor("out_63_axes_0"), val = tensor([1])]; tensor var_2449_to_fp16 = const()[name = tensor("op_2449_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_63_cast_fp16 = layer_norm(axes = out_63_axes_0, epsilon = var_2449_to_fp16, x = inputs_63_cast_fp16)[name = tensor("out_63_cast_fp16")]; tensor obj_149_gamma_0_to_fp16 = const()[name = tensor("obj_149_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237821120)))]; tensor obj_149_beta_0_to_fp16 = const()[name = tensor("obj_149_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237823232)))]; tensor obj_149_epsilon_0_to_fp16 = const()[name = tensor("obj_149_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_149_cast_fp16 = batch_norm(beta = obj_149_beta_0_to_fp16, epsilon = obj_149_epsilon_0_to_fp16, gamma = obj_149_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_63_cast_fp16)[name = tensor("obj_149_cast_fp16")]; tensor query_43_pad_type_0 = const()[name = tensor("query_43_pad_type_0"), val = tensor("valid")]; tensor query_43_strides_0 = const()[name = tensor("query_43_strides_0"), val = tensor([1, 1])]; tensor query_43_pad_0 = const()[name = tensor("query_43_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_43_dilations_0 = const()[name = tensor("query_43_dilations_0"), val = tensor([1, 1])]; tensor query_43_groups_0 = const()[name = tensor("query_43_groups_0"), val = tensor(1)]; tensor layers_10_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237825344))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(238611840))), name = tensor("layers_10_encoder_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_10_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_10_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(238612032)))]; tensor query_43_cast_fp16 = conv(bias = layers_10_encoder_attn_q_proj_bias_to_fp16, dilations = query_43_dilations_0, groups = query_43_groups_0, pad = query_43_pad_0, pad_type = query_43_pad_type_0, strides = query_43_strides_0, weight = layers_10_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_149_cast_fp16)[name = tensor("query_43_cast_fp16")]; tensor key_43_pad_type_0 = const()[name = tensor("key_43_pad_type_0"), val = tensor("valid")]; tensor key_43_strides_0 = const()[name = tensor("key_43_strides_0"), val = tensor([1, 1])]; tensor key_43_pad_0 = const()[name = tensor("key_43_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_43_dilations_0 = const()[name = tensor("key_43_dilations_0"), val = tensor([1, 1])]; tensor key_43_groups_0 = const()[name = tensor("key_43_groups_0"), val = tensor(1)]; tensor layers_10_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(238614144))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(239400640))), name = tensor("layers_10_encoder_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor key_43_cast_fp16 = conv(dilations = key_43_dilations_0, groups = key_43_groups_0, pad = key_43_pad_0, pad_type = key_43_pad_type_0, strides = key_43_strides_0, weight = layers_10_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("key_43_cast_fp16")]; tensor value_43_pad_type_0 = const()[name = tensor("value_43_pad_type_0"), val = tensor("valid")]; tensor value_43_strides_0 = const()[name = tensor("value_43_strides_0"), val = tensor([1, 1])]; tensor value_43_pad_0 = const()[name = tensor("value_43_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_43_dilations_0 = const()[name = tensor("value_43_dilations_0"), val = tensor([1, 1])]; tensor value_43_groups_0 = const()[name = tensor("value_43_groups_0"), val = tensor(1)]; tensor layers_10_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(239400832))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240187328))), name = tensor("layers_10_encoder_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_10_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_10_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240187520)))]; tensor value_43_cast_fp16 = conv(bias = layers_10_encoder_attn_v_proj_bias_to_fp16, dilations = value_43_dilations_0, groups = value_43_groups_0, pad = value_43_pad_0, pad_type = value_43_pad_type_0, strides = value_43_strides_0, weight = layers_10_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("value_43_cast_fp16")]; tensor var_2485 = const()[name = tensor("op_2485"), val = tensor([1, 16, 64, 1])]; tensor mh_q_43_cast_fp16 = reshape(shape = var_2485, x = query_43_cast_fp16)[name = tensor("mh_q_43_cast_fp16")]; tensor var_2487_to_fp16 = const()[name = tensor("op_2487_to_fp16"), val = tensor(0x1p-3)]; tensor var_2488_cast_fp16 = mul(x = mh_q_43_cast_fp16, y = var_2487_to_fp16)[name = tensor("op_2488_cast_fp16")]; tensor var_2491 = const()[name = tensor("op_2491"), val = tensor([1, 16, 64, 1500])]; tensor var_2492_cast_fp16 = reshape(shape = var_2491, x = key_43_cast_fp16)[name = tensor("op_2492_cast_fp16")]; tensor mh_w_65_transpose_x_0 = const()[name = tensor("mh_w_65_transpose_x_0"), val = tensor(true)]; tensor mh_w_65_transpose_y_0 = const()[name = tensor("mh_w_65_transpose_y_0"), val = tensor(false)]; tensor mh_w_65_cast_fp16 = matmul(transpose_x = mh_w_65_transpose_x_0, transpose_y = mh_w_65_transpose_y_0, x = var_2488_cast_fp16, y = var_2492_cast_fp16)[name = tensor("mh_w_65_cast_fp16")]; tensor obj_153_cast_fp16 = softmax(axis = var_2334, x = mh_w_65_cast_fp16)[name = tensor("obj_153_cast_fp16")]; tensor var_2496 = const()[name = tensor("op_2496"), val = tensor([1, 16, 64, 1500])]; tensor var_2497_cast_fp16 = reshape(shape = var_2496, x = value_43_cast_fp16)[name = tensor("op_2497_cast_fp16")]; tensor attn_43_transpose_x_0 = const()[name = tensor("attn_43_transpose_x_0"), val = tensor(false)]; tensor attn_43_transpose_y_0 = const()[name = tensor("attn_43_transpose_y_0"), val = tensor(true)]; tensor attn_43_cast_fp16 = matmul(transpose_x = attn_43_transpose_x_0, transpose_y = attn_43_transpose_y_0, x = var_2497_cast_fp16, y = obj_153_cast_fp16)[name = tensor("attn_43_cast_fp16")]; tensor var_2500 = const()[name = tensor("op_2500"), val = tensor([1, 1024, 1, 1])]; tensor input_103_cast_fp16 = reshape(shape = var_2500, x = attn_43_cast_fp16)[name = tensor("input_103_cast_fp16")]; tensor obj_151_pad_type_0 = const()[name = tensor("obj_151_pad_type_0"), val = tensor("valid")]; tensor obj_151_strides_0 = const()[name = tensor("obj_151_strides_0"), val = tensor([1, 1])]; tensor obj_151_pad_0 = const()[name = tensor("obj_151_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_151_dilations_0 = const()[name = tensor("obj_151_dilations_0"), val = tensor([1, 1])]; tensor obj_151_groups_0 = const()[name = tensor("obj_151_groups_0"), val = tensor(1)]; tensor layers_10_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240189632))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240976128))), name = tensor("layers_10_encoder_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_10_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_10_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240976320)))]; tensor obj_151_cast_fp16 = conv(bias = layers_10_encoder_attn_o_proj_bias_to_fp16, dilations = obj_151_dilations_0, groups = obj_151_groups_0, pad = obj_151_pad_0, pad_type = obj_151_pad_type_0, strides = obj_151_strides_0, weight = layers_10_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_103_cast_fp16)[name = tensor("obj_151_cast_fp16")]; tensor inputs_65_cast_fp16 = add(x = inputs_63_cast_fp16, y = obj_151_cast_fp16)[name = tensor("inputs_65_cast_fp16")]; tensor out_65_axes_0 = const()[name = tensor("out_65_axes_0"), val = tensor([1])]; tensor var_2518_to_fp16 = const()[name = tensor("op_2518_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_65_cast_fp16 = layer_norm(axes = out_65_axes_0, epsilon = var_2518_to_fp16, x = inputs_65_cast_fp16)[name = tensor("out_65_cast_fp16")]; tensor input_105_gamma_0_to_fp16 = const()[name = tensor("input_105_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240978432)))]; tensor input_105_beta_0_to_fp16 = const()[name = tensor("input_105_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240980544)))]; tensor input_105_epsilon_0_to_fp16 = const()[name = tensor("input_105_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_105_cast_fp16 = batch_norm(beta = input_105_beta_0_to_fp16, epsilon = input_105_epsilon_0_to_fp16, gamma = input_105_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_65_cast_fp16)[name = tensor("input_105_cast_fp16")]; tensor input_107_pad_type_0 = const()[name = tensor("input_107_pad_type_0"), val = tensor("valid")]; tensor input_107_strides_0 = const()[name = tensor("input_107_strides_0"), val = tensor([1, 1])]; tensor input_107_pad_0 = const()[name = tensor("input_107_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_107_dilations_0 = const()[name = tensor("input_107_dilations_0"), val = tensor([1, 1])]; tensor input_107_groups_0 = const()[name = tensor("input_107_groups_0"), val = tensor(1)]; tensor layers_10_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240982656))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(244128448))), name = tensor("layers_10_fc1_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; tensor layers_10_fc1_bias_to_fp16 = const()[name = tensor("layers_10_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(244128640)))]; tensor input_107_cast_fp16 = conv(bias = layers_10_fc1_bias_to_fp16, dilations = input_107_dilations_0, groups = input_107_groups_0, pad = input_107_pad_0, pad_type = input_107_pad_type_0, strides = input_107_strides_0, weight = layers_10_fc1_weight_to_fp16_palettized, x = input_105_cast_fp16)[name = tensor("input_107_cast_fp16")]; tensor input_109_mode_0 = const()[name = tensor("input_109_mode_0"), val = tensor("EXACT")]; tensor input_109_cast_fp16 = gelu(mode = input_109_mode_0, x = input_107_cast_fp16)[name = tensor("input_109_cast_fp16")]; tensor hidden_states_23_pad_type_0 = const()[name = tensor("hidden_states_23_pad_type_0"), val = tensor("valid")]; tensor hidden_states_23_strides_0 = const()[name = tensor("hidden_states_23_strides_0"), val = tensor([1, 1])]; tensor hidden_states_23_pad_0 = const()[name = tensor("hidden_states_23_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_23_dilations_0 = const()[name = tensor("hidden_states_23_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_23_groups_0 = const()[name = tensor("hidden_states_23_groups_0"), val = tensor(1)]; tensor layers_10_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(244136896))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(247282688))), name = tensor("layers_10_fc2_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; tensor layers_10_fc2_bias_to_fp16 = const()[name = tensor("layers_10_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(247282880)))]; tensor hidden_states_23_cast_fp16 = conv(bias = layers_10_fc2_bias_to_fp16, dilations = hidden_states_23_dilations_0, groups = hidden_states_23_groups_0, pad = hidden_states_23_pad_0, pad_type = hidden_states_23_pad_type_0, strides = hidden_states_23_strides_0, weight = layers_10_fc2_weight_to_fp16_palettized, x = input_109_cast_fp16)[name = tensor("hidden_states_23_cast_fp16")]; tensor inputs_67_cast_fp16 = add(x = inputs_65_cast_fp16, y = hidden_states_23_cast_fp16)[name = tensor("inputs_67_cast_fp16")]; tensor var_2553 = const()[name = tensor("op_2553"), val = tensor(3)]; tensor out_67_axes_0 = const()[name = tensor("out_67_axes_0"), val = tensor([1])]; tensor var_2578_to_fp16 = const()[name = tensor("op_2578_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_67_cast_fp16 = layer_norm(axes = out_67_axes_0, epsilon = var_2578_to_fp16, x = inputs_67_cast_fp16)[name = tensor("out_67_cast_fp16")]; tensor obj_155_gamma_0_to_fp16 = const()[name = tensor("obj_155_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(247284992)))]; tensor obj_155_beta_0_to_fp16 = const()[name = tensor("obj_155_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(247287104)))]; tensor obj_155_epsilon_0_to_fp16 = const()[name = tensor("obj_155_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_155_cast_fp16 = batch_norm(beta = obj_155_beta_0_to_fp16, epsilon = obj_155_epsilon_0_to_fp16, gamma = obj_155_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_67_cast_fp16)[name = tensor("obj_155_cast_fp16")]; tensor query_45_pad_type_0 = const()[name = tensor("query_45_pad_type_0"), val = tensor("valid")]; tensor query_45_strides_0 = const()[name = tensor("query_45_strides_0"), val = tensor([1, 1])]; tensor query_45_pad_0 = const()[name = tensor("query_45_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_45_dilations_0 = const()[name = tensor("query_45_dilations_0"), val = tensor([1, 1])]; tensor query_45_groups_0 = const()[name = tensor("query_45_groups_0"), val = tensor(1)]; tensor layers_11_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(247289216))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(248075712))), name = tensor("layers_11_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(248075904)))]; tensor query_45_cast_fp16 = conv(bias = layers_11_self_attn_q_proj_bias_to_fp16, dilations = query_45_dilations_0, groups = query_45_groups_0, pad = query_45_pad_0, pad_type = query_45_pad_type_0, strides = query_45_strides_0, weight = layers_11_self_attn_q_proj_weight_to_fp16_palettized, x = obj_155_cast_fp16)[name = tensor("query_45_cast_fp16")]; tensor current_key_23_pad_type_0 = const()[name = tensor("current_key_23_pad_type_0"), val = tensor("valid")]; tensor current_key_23_strides_0 = const()[name = tensor("current_key_23_strides_0"), val = tensor([1, 1])]; tensor current_key_23_pad_0 = const()[name = tensor("current_key_23_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_23_dilations_0 = const()[name = tensor("current_key_23_dilations_0"), val = tensor([1, 1])]; tensor current_key_23_groups_0 = const()[name = tensor("current_key_23_groups_0"), val = tensor(1)]; tensor layers_11_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(248078016))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(248864512))), name = tensor("layers_11_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor current_key_23_cast_fp16 = conv(dilations = current_key_23_dilations_0, groups = current_key_23_groups_0, pad = current_key_23_pad_0, pad_type = current_key_23_pad_type_0, strides = current_key_23_strides_0, weight = layers_11_self_attn_k_proj_weight_to_fp16_palettized, x = obj_155_cast_fp16)[name = tensor("current_key_23_cast_fp16")]; tensor current_value_23_pad_type_0 = const()[name = tensor("current_value_23_pad_type_0"), val = tensor("valid")]; tensor current_value_23_strides_0 = const()[name = tensor("current_value_23_strides_0"), val = tensor([1, 1])]; tensor current_value_23_pad_0 = const()[name = tensor("current_value_23_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_23_dilations_0 = const()[name = tensor("current_value_23_dilations_0"), val = tensor([1, 1])]; tensor current_value_23_groups_0 = const()[name = tensor("current_value_23_groups_0"), val = tensor(1)]; tensor layers_11_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(248864704))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(249651200))), name = tensor("layers_11_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(249651392)))]; tensor current_value_23_cast_fp16 = conv(bias = layers_11_self_attn_v_proj_bias_to_fp16, dilations = current_value_23_dilations_0, groups = current_value_23_groups_0, pad = current_value_23_pad_0, pad_type = current_value_23_pad_type_0, strides = current_value_23_strides_0, weight = layers_11_self_attn_v_proj_weight_to_fp16_palettized, x = obj_155_cast_fp16)[name = tensor("current_value_23_cast_fp16")]; tensor var_2617_cast_fp16 = mul(x = var_87_cast_fp16_11, y = var_207_cast_fp16)[name = tensor("op_2617_cast_fp16")]; tensor var_2618_cast_fp16 = mul(x = current_key_23_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_2618_cast_fp16")]; tensor key_45_cast_fp16 = add(x = var_2617_cast_fp16, y = var_2618_cast_fp16)[name = tensor("key_45_cast_fp16")]; tensor var_2621_cast_fp16 = mul(x = var_114_cast_fp16_11, y = var_207_cast_fp16)[name = tensor("op_2621_cast_fp16")]; tensor var_2622_cast_fp16 = mul(x = current_value_23_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_2622_cast_fp16")]; tensor value_45_cast_fp16 = add(x = var_2621_cast_fp16, y = var_2622_cast_fp16)[name = tensor("value_45_cast_fp16")]; tensor var_2626 = const()[name = tensor("op_2626"), val = tensor([1, 16, 64, 1])]; tensor mh_q_45_cast_fp16 = reshape(shape = var_2626, x = query_45_cast_fp16)[name = tensor("mh_q_45_cast_fp16")]; tensor var_2628_to_fp16 = const()[name = tensor("op_2628_to_fp16"), val = tensor(0x1p-3)]; tensor var_2629_cast_fp16 = mul(x = mh_q_45_cast_fp16, y = var_2628_to_fp16)[name = tensor("op_2629_cast_fp16")]; tensor var_2632 = const()[name = tensor("op_2632"), val = tensor([1, 16, 64, 448])]; tensor var_2633_cast_fp16 = reshape(shape = var_2632, x = key_45_cast_fp16)[name = tensor("op_2633_cast_fp16")]; tensor mh_w_67_transpose_x_0 = const()[name = tensor("mh_w_67_transpose_x_0"), val = tensor(true)]; tensor mh_w_67_transpose_y_0 = const()[name = tensor("mh_w_67_transpose_y_0"), val = tensor(false)]; tensor mh_w_67_cast_fp16 = matmul(transpose_x = mh_w_67_transpose_x_0, transpose_y = mh_w_67_transpose_y_0, x = var_2629_cast_fp16, y = var_2633_cast_fp16)[name = tensor("mh_w_67_cast_fp16")]; tensor mh_w_69_cast_fp16 = add(x = mh_w_67_cast_fp16, y = var_229_cast_fp16)[name = tensor("mh_w_69_cast_fp16")]; tensor var_2641_cast_fp16 = softmax(axis = var_2553, x = mh_w_69_cast_fp16)[name = tensor("op_2641_cast_fp16")]; tensor var_2642 = const()[name = tensor("op_2642"), val = tensor([1, 16, 64, 448])]; tensor var_2643_cast_fp16 = reshape(shape = var_2642, x = value_45_cast_fp16)[name = tensor("op_2643_cast_fp16")]; tensor attn_45_transpose_x_0 = const()[name = tensor("attn_45_transpose_x_0"), val = tensor(false)]; tensor attn_45_transpose_y_0 = const()[name = tensor("attn_45_transpose_y_0"), val = tensor(true)]; tensor attn_45_cast_fp16 = matmul(transpose_x = attn_45_transpose_x_0, transpose_y = attn_45_transpose_y_0, x = var_2643_cast_fp16, y = var_2641_cast_fp16)[name = tensor("attn_45_cast_fp16")]; tensor var_2646 = const()[name = tensor("op_2646"), val = tensor([1, 1024, 1, 1])]; tensor input_111_cast_fp16 = reshape(shape = var_2646, x = attn_45_cast_fp16)[name = tensor("input_111_cast_fp16")]; tensor obj_161_pad_type_0 = const()[name = tensor("obj_161_pad_type_0"), val = tensor("valid")]; tensor obj_161_strides_0 = const()[name = tensor("obj_161_strides_0"), val = tensor([1, 1])]; tensor obj_161_pad_0 = const()[name = tensor("obj_161_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_161_dilations_0 = const()[name = tensor("obj_161_dilations_0"), val = tensor([1, 1])]; tensor obj_161_groups_0 = const()[name = tensor("obj_161_groups_0"), val = tensor(1)]; tensor layers_11_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(249653504))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250440000))), name = tensor("layers_11_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_11_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_11_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250440192)))]; tensor obj_161_cast_fp16 = conv(bias = layers_11_self_attn_o_proj_bias_to_fp16, dilations = obj_161_dilations_0, groups = obj_161_groups_0, pad = obj_161_pad_0, pad_type = obj_161_pad_type_0, strides = obj_161_strides_0, weight = layers_11_self_attn_o_proj_weight_to_fp16_palettized, x = input_111_cast_fp16)[name = tensor("obj_161_cast_fp16")]; tensor inputs_69_cast_fp16 = add(x = inputs_67_cast_fp16, y = obj_161_cast_fp16)[name = tensor("inputs_69_cast_fp16")]; tensor out_69_axes_0 = const()[name = tensor("out_69_axes_0"), val = tensor([1])]; tensor var_2668_to_fp16 = const()[name = tensor("op_2668_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_69_cast_fp16 = layer_norm(axes = out_69_axes_0, epsilon = var_2668_to_fp16, x = inputs_69_cast_fp16)[name = tensor("out_69_cast_fp16")]; tensor obj_163_gamma_0_to_fp16 = const()[name = tensor("obj_163_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250442304)))]; tensor obj_163_beta_0_to_fp16 = const()[name = tensor("obj_163_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250444416)))]; tensor obj_163_epsilon_0_to_fp16 = const()[name = tensor("obj_163_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_163_cast_fp16 = batch_norm(beta = obj_163_beta_0_to_fp16, epsilon = obj_163_epsilon_0_to_fp16, gamma = obj_163_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_69_cast_fp16)[name = tensor("obj_163_cast_fp16")]; tensor query_47_pad_type_0 = const()[name = tensor("query_47_pad_type_0"), val = tensor("valid")]; tensor query_47_strides_0 = const()[name = tensor("query_47_strides_0"), val = tensor([1, 1])]; tensor query_47_pad_0 = const()[name = tensor("query_47_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_47_dilations_0 = const()[name = tensor("query_47_dilations_0"), val = tensor([1, 1])]; tensor query_47_groups_0 = const()[name = tensor("query_47_groups_0"), val = tensor(1)]; tensor layers_11_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250446528))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(251233024))), name = tensor("layers_11_encoder_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_11_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_11_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(251233216)))]; tensor query_47_cast_fp16 = conv(bias = layers_11_encoder_attn_q_proj_bias_to_fp16, dilations = query_47_dilations_0, groups = query_47_groups_0, pad = query_47_pad_0, pad_type = query_47_pad_type_0, strides = query_47_strides_0, weight = layers_11_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_163_cast_fp16)[name = tensor("query_47_cast_fp16")]; tensor key_47_pad_type_0 = const()[name = tensor("key_47_pad_type_0"), val = tensor("valid")]; tensor key_47_strides_0 = const()[name = tensor("key_47_strides_0"), val = tensor([1, 1])]; tensor key_47_pad_0 = const()[name = tensor("key_47_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_47_dilations_0 = const()[name = tensor("key_47_dilations_0"), val = tensor([1, 1])]; tensor key_47_groups_0 = const()[name = tensor("key_47_groups_0"), val = tensor(1)]; tensor layers_11_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(251235328))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(252021824))), name = tensor("layers_11_encoder_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor key_47_cast_fp16 = conv(dilations = key_47_dilations_0, groups = key_47_groups_0, pad = key_47_pad_0, pad_type = key_47_pad_type_0, strides = key_47_strides_0, weight = layers_11_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("key_47_cast_fp16")]; tensor value_47_pad_type_0 = const()[name = tensor("value_47_pad_type_0"), val = tensor("valid")]; tensor value_47_strides_0 = const()[name = tensor("value_47_strides_0"), val = tensor([1, 1])]; tensor value_47_pad_0 = const()[name = tensor("value_47_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_47_dilations_0 = const()[name = tensor("value_47_dilations_0"), val = tensor([1, 1])]; tensor value_47_groups_0 = const()[name = tensor("value_47_groups_0"), val = tensor(1)]; tensor layers_11_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(252022016))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(252808512))), name = tensor("layers_11_encoder_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_11_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_11_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(252808704)))]; tensor value_47_cast_fp16 = conv(bias = layers_11_encoder_attn_v_proj_bias_to_fp16, dilations = value_47_dilations_0, groups = value_47_groups_0, pad = value_47_pad_0, pad_type = value_47_pad_type_0, strides = value_47_strides_0, weight = layers_11_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("value_47_cast_fp16")]; tensor var_2704 = const()[name = tensor("op_2704"), val = tensor([1, 16, 64, 1])]; tensor mh_q_47_cast_fp16 = reshape(shape = var_2704, x = query_47_cast_fp16)[name = tensor("mh_q_47_cast_fp16")]; tensor var_2706_to_fp16 = const()[name = tensor("op_2706_to_fp16"), val = tensor(0x1p-3)]; tensor var_2707_cast_fp16 = mul(x = mh_q_47_cast_fp16, y = var_2706_to_fp16)[name = tensor("op_2707_cast_fp16")]; tensor var_2710 = const()[name = tensor("op_2710"), val = tensor([1, 16, 64, 1500])]; tensor var_2711_cast_fp16 = reshape(shape = var_2710, x = key_47_cast_fp16)[name = tensor("op_2711_cast_fp16")]; tensor mh_w_71_transpose_x_0 = const()[name = tensor("mh_w_71_transpose_x_0"), val = tensor(true)]; tensor mh_w_71_transpose_y_0 = const()[name = tensor("mh_w_71_transpose_y_0"), val = tensor(false)]; tensor mh_w_71_cast_fp16 = matmul(transpose_x = mh_w_71_transpose_x_0, transpose_y = mh_w_71_transpose_y_0, x = var_2707_cast_fp16, y = var_2711_cast_fp16)[name = tensor("mh_w_71_cast_fp16")]; tensor obj_167_cast_fp16 = softmax(axis = var_2553, x = mh_w_71_cast_fp16)[name = tensor("obj_167_cast_fp16")]; tensor var_2715 = const()[name = tensor("op_2715"), val = tensor([1, 16, 64, 1500])]; tensor var_2716_cast_fp16 = reshape(shape = var_2715, x = value_47_cast_fp16)[name = tensor("op_2716_cast_fp16")]; tensor attn_47_transpose_x_0 = const()[name = tensor("attn_47_transpose_x_0"), val = tensor(false)]; tensor attn_47_transpose_y_0 = const()[name = tensor("attn_47_transpose_y_0"), val = tensor(true)]; tensor attn_47_cast_fp16 = matmul(transpose_x = attn_47_transpose_x_0, transpose_y = attn_47_transpose_y_0, x = var_2716_cast_fp16, y = obj_167_cast_fp16)[name = tensor("attn_47_cast_fp16")]; tensor var_2719 = const()[name = tensor("op_2719"), val = tensor([1, 1024, 1, 1])]; tensor input_113_cast_fp16 = reshape(shape = var_2719, x = attn_47_cast_fp16)[name = tensor("input_113_cast_fp16")]; tensor obj_165_pad_type_0 = const()[name = tensor("obj_165_pad_type_0"), val = tensor("valid")]; tensor obj_165_strides_0 = const()[name = tensor("obj_165_strides_0"), val = tensor([1, 1])]; tensor obj_165_pad_0 = const()[name = tensor("obj_165_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_165_dilations_0 = const()[name = tensor("obj_165_dilations_0"), val = tensor([1, 1])]; tensor obj_165_groups_0 = const()[name = tensor("obj_165_groups_0"), val = tensor(1)]; tensor layers_11_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(252810816))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(253597312))), name = tensor("layers_11_encoder_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_11_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_11_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(253597504)))]; tensor obj_165_cast_fp16 = conv(bias = layers_11_encoder_attn_o_proj_bias_to_fp16, dilations = obj_165_dilations_0, groups = obj_165_groups_0, pad = obj_165_pad_0, pad_type = obj_165_pad_type_0, strides = obj_165_strides_0, weight = layers_11_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_113_cast_fp16)[name = tensor("obj_165_cast_fp16")]; tensor inputs_71_cast_fp16 = add(x = inputs_69_cast_fp16, y = obj_165_cast_fp16)[name = tensor("inputs_71_cast_fp16")]; tensor out_71_axes_0 = const()[name = tensor("out_71_axes_0"), val = tensor([1])]; tensor var_2737_to_fp16 = const()[name = tensor("op_2737_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_71_cast_fp16 = layer_norm(axes = out_71_axes_0, epsilon = var_2737_to_fp16, x = inputs_71_cast_fp16)[name = tensor("out_71_cast_fp16")]; tensor input_115_gamma_0_to_fp16 = const()[name = tensor("input_115_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(253599616)))]; tensor input_115_beta_0_to_fp16 = const()[name = tensor("input_115_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(253601728)))]; tensor input_115_epsilon_0_to_fp16 = const()[name = tensor("input_115_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_115_cast_fp16 = batch_norm(beta = input_115_beta_0_to_fp16, epsilon = input_115_epsilon_0_to_fp16, gamma = input_115_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_71_cast_fp16)[name = tensor("input_115_cast_fp16")]; tensor input_117_pad_type_0 = const()[name = tensor("input_117_pad_type_0"), val = tensor("valid")]; tensor input_117_strides_0 = const()[name = tensor("input_117_strides_0"), val = tensor([1, 1])]; tensor input_117_pad_0 = const()[name = tensor("input_117_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_117_dilations_0 = const()[name = tensor("input_117_dilations_0"), val = tensor([1, 1])]; tensor input_117_groups_0 = const()[name = tensor("input_117_groups_0"), val = tensor(1)]; tensor layers_11_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(253603840))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(256749632))), name = tensor("layers_11_fc1_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; tensor layers_11_fc1_bias_to_fp16 = const()[name = tensor("layers_11_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(256749824)))]; tensor input_117_cast_fp16 = conv(bias = layers_11_fc1_bias_to_fp16, dilations = input_117_dilations_0, groups = input_117_groups_0, pad = input_117_pad_0, pad_type = input_117_pad_type_0, strides = input_117_strides_0, weight = layers_11_fc1_weight_to_fp16_palettized, x = input_115_cast_fp16)[name = tensor("input_117_cast_fp16")]; tensor input_119_mode_0 = const()[name = tensor("input_119_mode_0"), val = tensor("EXACT")]; tensor input_119_cast_fp16 = gelu(mode = input_119_mode_0, x = input_117_cast_fp16)[name = tensor("input_119_cast_fp16")]; tensor hidden_states_25_pad_type_0 = const()[name = tensor("hidden_states_25_pad_type_0"), val = tensor("valid")]; tensor hidden_states_25_strides_0 = const()[name = tensor("hidden_states_25_strides_0"), val = tensor([1, 1])]; tensor hidden_states_25_pad_0 = const()[name = tensor("hidden_states_25_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_25_dilations_0 = const()[name = tensor("hidden_states_25_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_25_groups_0 = const()[name = tensor("hidden_states_25_groups_0"), val = tensor(1)]; tensor layers_11_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(256758080))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(259903872))), name = tensor("layers_11_fc2_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; tensor layers_11_fc2_bias_to_fp16 = const()[name = tensor("layers_11_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(259904064)))]; tensor hidden_states_25_cast_fp16 = conv(bias = layers_11_fc2_bias_to_fp16, dilations = hidden_states_25_dilations_0, groups = hidden_states_25_groups_0, pad = hidden_states_25_pad_0, pad_type = hidden_states_25_pad_type_0, strides = hidden_states_25_strides_0, weight = layers_11_fc2_weight_to_fp16_palettized, x = input_119_cast_fp16)[name = tensor("hidden_states_25_cast_fp16")]; tensor inputs_73_cast_fp16 = add(x = inputs_71_cast_fp16, y = hidden_states_25_cast_fp16)[name = tensor("inputs_73_cast_fp16")]; tensor var_2772 = const()[name = tensor("op_2772"), val = tensor(3)]; tensor out_73_axes_0 = const()[name = tensor("out_73_axes_0"), val = tensor([1])]; tensor var_2797_to_fp16 = const()[name = tensor("op_2797_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_73_cast_fp16 = layer_norm(axes = out_73_axes_0, epsilon = var_2797_to_fp16, x = inputs_73_cast_fp16)[name = tensor("out_73_cast_fp16")]; tensor obj_169_gamma_0_to_fp16 = const()[name = tensor("obj_169_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(259906176)))]; tensor obj_169_beta_0_to_fp16 = const()[name = tensor("obj_169_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(259908288)))]; tensor obj_169_epsilon_0_to_fp16 = const()[name = tensor("obj_169_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_169_cast_fp16 = batch_norm(beta = obj_169_beta_0_to_fp16, epsilon = obj_169_epsilon_0_to_fp16, gamma = obj_169_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_73_cast_fp16)[name = tensor("obj_169_cast_fp16")]; tensor query_49_pad_type_0 = const()[name = tensor("query_49_pad_type_0"), val = tensor("valid")]; tensor query_49_strides_0 = const()[name = tensor("query_49_strides_0"), val = tensor([1, 1])]; tensor query_49_pad_0 = const()[name = tensor("query_49_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_49_dilations_0 = const()[name = tensor("query_49_dilations_0"), val = tensor([1, 1])]; tensor query_49_groups_0 = const()[name = tensor("query_49_groups_0"), val = tensor(1)]; tensor layers_12_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(259910400))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(260696896))), name = tensor("layers_12_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_12_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_12_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(260697088)))]; tensor query_49_cast_fp16 = conv(bias = layers_12_self_attn_q_proj_bias_to_fp16, dilations = query_49_dilations_0, groups = query_49_groups_0, pad = query_49_pad_0, pad_type = query_49_pad_type_0, strides = query_49_strides_0, weight = layers_12_self_attn_q_proj_weight_to_fp16_palettized, x = obj_169_cast_fp16)[name = tensor("query_49_cast_fp16")]; tensor current_key_25_pad_type_0 = const()[name = tensor("current_key_25_pad_type_0"), val = tensor("valid")]; tensor current_key_25_strides_0 = const()[name = tensor("current_key_25_strides_0"), val = tensor([1, 1])]; tensor current_key_25_pad_0 = const()[name = tensor("current_key_25_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_25_dilations_0 = const()[name = tensor("current_key_25_dilations_0"), val = tensor([1, 1])]; tensor current_key_25_groups_0 = const()[name = tensor("current_key_25_groups_0"), val = tensor(1)]; tensor layers_12_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(260699200))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261485696))), name = tensor("layers_12_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor current_key_25_cast_fp16 = conv(dilations = current_key_25_dilations_0, groups = current_key_25_groups_0, pad = current_key_25_pad_0, pad_type = current_key_25_pad_type_0, strides = current_key_25_strides_0, weight = layers_12_self_attn_k_proj_weight_to_fp16_palettized, x = obj_169_cast_fp16)[name = tensor("current_key_25_cast_fp16")]; tensor current_value_25_pad_type_0 = const()[name = tensor("current_value_25_pad_type_0"), val = tensor("valid")]; tensor current_value_25_strides_0 = const()[name = tensor("current_value_25_strides_0"), val = tensor([1, 1])]; tensor current_value_25_pad_0 = const()[name = tensor("current_value_25_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_25_dilations_0 = const()[name = tensor("current_value_25_dilations_0"), val = tensor([1, 1])]; tensor current_value_25_groups_0 = const()[name = tensor("current_value_25_groups_0"), val = tensor(1)]; tensor layers_12_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261485888))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(262272384))), name = tensor("layers_12_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_12_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_12_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(262272576)))]; tensor current_value_25_cast_fp16 = conv(bias = layers_12_self_attn_v_proj_bias_to_fp16, dilations = current_value_25_dilations_0, groups = current_value_25_groups_0, pad = current_value_25_pad_0, pad_type = current_value_25_pad_type_0, strides = current_value_25_strides_0, weight = layers_12_self_attn_v_proj_weight_to_fp16_palettized, x = obj_169_cast_fp16)[name = tensor("current_value_25_cast_fp16")]; tensor var_2836_cast_fp16 = mul(x = var_87_cast_fp16_12, y = var_207_cast_fp16)[name = tensor("op_2836_cast_fp16")]; tensor var_2837_cast_fp16 = mul(x = current_key_25_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_2837_cast_fp16")]; tensor key_49_cast_fp16 = add(x = var_2836_cast_fp16, y = var_2837_cast_fp16)[name = tensor("key_49_cast_fp16")]; tensor var_2840_cast_fp16 = mul(x = var_114_cast_fp16_12, y = var_207_cast_fp16)[name = tensor("op_2840_cast_fp16")]; tensor var_2841_cast_fp16 = mul(x = current_value_25_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_2841_cast_fp16")]; tensor value_49_cast_fp16 = add(x = var_2840_cast_fp16, y = var_2841_cast_fp16)[name = tensor("value_49_cast_fp16")]; tensor var_2845 = const()[name = tensor("op_2845"), val = tensor([1, 16, 64, 1])]; tensor mh_q_49_cast_fp16 = reshape(shape = var_2845, x = query_49_cast_fp16)[name = tensor("mh_q_49_cast_fp16")]; tensor var_2847_to_fp16 = const()[name = tensor("op_2847_to_fp16"), val = tensor(0x1p-3)]; tensor var_2848_cast_fp16 = mul(x = mh_q_49_cast_fp16, y = var_2847_to_fp16)[name = tensor("op_2848_cast_fp16")]; tensor var_2851 = const()[name = tensor("op_2851"), val = tensor([1, 16, 64, 448])]; tensor var_2852_cast_fp16 = reshape(shape = var_2851, x = key_49_cast_fp16)[name = tensor("op_2852_cast_fp16")]; tensor mh_w_73_transpose_x_0 = const()[name = tensor("mh_w_73_transpose_x_0"), val = tensor(true)]; tensor mh_w_73_transpose_y_0 = const()[name = tensor("mh_w_73_transpose_y_0"), val = tensor(false)]; tensor mh_w_73_cast_fp16 = matmul(transpose_x = mh_w_73_transpose_x_0, transpose_y = mh_w_73_transpose_y_0, x = var_2848_cast_fp16, y = var_2852_cast_fp16)[name = tensor("mh_w_73_cast_fp16")]; tensor mh_w_75_cast_fp16 = add(x = mh_w_73_cast_fp16, y = var_229_cast_fp16)[name = tensor("mh_w_75_cast_fp16")]; tensor var_2860_cast_fp16 = softmax(axis = var_2772, x = mh_w_75_cast_fp16)[name = tensor("op_2860_cast_fp16")]; tensor var_2861 = const()[name = tensor("op_2861"), val = tensor([1, 16, 64, 448])]; tensor var_2862_cast_fp16 = reshape(shape = var_2861, x = value_49_cast_fp16)[name = tensor("op_2862_cast_fp16")]; tensor attn_49_transpose_x_0 = const()[name = tensor("attn_49_transpose_x_0"), val = tensor(false)]; tensor attn_49_transpose_y_0 = const()[name = tensor("attn_49_transpose_y_0"), val = tensor(true)]; tensor attn_49_cast_fp16 = matmul(transpose_x = attn_49_transpose_x_0, transpose_y = attn_49_transpose_y_0, x = var_2862_cast_fp16, y = var_2860_cast_fp16)[name = tensor("attn_49_cast_fp16")]; tensor var_2865 = const()[name = tensor("op_2865"), val = tensor([1, 1024, 1, 1])]; tensor input_121_cast_fp16 = reshape(shape = var_2865, x = attn_49_cast_fp16)[name = tensor("input_121_cast_fp16")]; tensor obj_175_pad_type_0 = const()[name = tensor("obj_175_pad_type_0"), val = tensor("valid")]; tensor obj_175_strides_0 = const()[name = tensor("obj_175_strides_0"), val = tensor([1, 1])]; tensor obj_175_pad_0 = const()[name = tensor("obj_175_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_175_dilations_0 = const()[name = tensor("obj_175_dilations_0"), val = tensor([1, 1])]; tensor obj_175_groups_0 = const()[name = tensor("obj_175_groups_0"), val = tensor(1)]; tensor layers_12_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(262274688))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263061184))), name = tensor("layers_12_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_12_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_12_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263061376)))]; tensor obj_175_cast_fp16 = conv(bias = layers_12_self_attn_o_proj_bias_to_fp16, dilations = obj_175_dilations_0, groups = obj_175_groups_0, pad = obj_175_pad_0, pad_type = obj_175_pad_type_0, strides = obj_175_strides_0, weight = layers_12_self_attn_o_proj_weight_to_fp16_palettized, x = input_121_cast_fp16)[name = tensor("obj_175_cast_fp16")]; tensor inputs_75_cast_fp16 = add(x = inputs_73_cast_fp16, y = obj_175_cast_fp16)[name = tensor("inputs_75_cast_fp16")]; tensor out_75_axes_0 = const()[name = tensor("out_75_axes_0"), val = tensor([1])]; tensor var_2887_to_fp16 = const()[name = tensor("op_2887_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_75_cast_fp16 = layer_norm(axes = out_75_axes_0, epsilon = var_2887_to_fp16, x = inputs_75_cast_fp16)[name = tensor("out_75_cast_fp16")]; tensor obj_177_gamma_0_to_fp16 = const()[name = tensor("obj_177_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263063488)))]; tensor obj_177_beta_0_to_fp16 = const()[name = tensor("obj_177_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263065600)))]; tensor obj_177_epsilon_0_to_fp16 = const()[name = tensor("obj_177_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_177_cast_fp16 = batch_norm(beta = obj_177_beta_0_to_fp16, epsilon = obj_177_epsilon_0_to_fp16, gamma = obj_177_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_75_cast_fp16)[name = tensor("obj_177_cast_fp16")]; tensor query_51_pad_type_0 = const()[name = tensor("query_51_pad_type_0"), val = tensor("valid")]; tensor query_51_strides_0 = const()[name = tensor("query_51_strides_0"), val = tensor([1, 1])]; tensor query_51_pad_0 = const()[name = tensor("query_51_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_51_dilations_0 = const()[name = tensor("query_51_dilations_0"), val = tensor([1, 1])]; tensor query_51_groups_0 = const()[name = tensor("query_51_groups_0"), val = tensor(1)]; tensor layers_12_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263067712))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263854208))), name = tensor("layers_12_encoder_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_12_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_12_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263854400)))]; tensor query_51_cast_fp16 = conv(bias = layers_12_encoder_attn_q_proj_bias_to_fp16, dilations = query_51_dilations_0, groups = query_51_groups_0, pad = query_51_pad_0, pad_type = query_51_pad_type_0, strides = query_51_strides_0, weight = layers_12_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_177_cast_fp16)[name = tensor("query_51_cast_fp16")]; tensor key_51_pad_type_0 = const()[name = tensor("key_51_pad_type_0"), val = tensor("valid")]; tensor key_51_strides_0 = const()[name = tensor("key_51_strides_0"), val = tensor([1, 1])]; tensor key_51_pad_0 = const()[name = tensor("key_51_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_51_dilations_0 = const()[name = tensor("key_51_dilations_0"), val = tensor([1, 1])]; tensor key_51_groups_0 = const()[name = tensor("key_51_groups_0"), val = tensor(1)]; tensor layers_12_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263856512))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(264643008))), name = tensor("layers_12_encoder_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor key_51_cast_fp16 = conv(dilations = key_51_dilations_0, groups = key_51_groups_0, pad = key_51_pad_0, pad_type = key_51_pad_type_0, strides = key_51_strides_0, weight = layers_12_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("key_51_cast_fp16")]; tensor value_51_pad_type_0 = const()[name = tensor("value_51_pad_type_0"), val = tensor("valid")]; tensor value_51_strides_0 = const()[name = tensor("value_51_strides_0"), val = tensor([1, 1])]; tensor value_51_pad_0 = const()[name = tensor("value_51_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_51_dilations_0 = const()[name = tensor("value_51_dilations_0"), val = tensor([1, 1])]; tensor value_51_groups_0 = const()[name = tensor("value_51_groups_0"), val = tensor(1)]; tensor layers_12_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(264643200))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265429696))), name = tensor("layers_12_encoder_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_12_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_12_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265429888)))]; tensor value_51_cast_fp16 = conv(bias = layers_12_encoder_attn_v_proj_bias_to_fp16, dilations = value_51_dilations_0, groups = value_51_groups_0, pad = value_51_pad_0, pad_type = value_51_pad_type_0, strides = value_51_strides_0, weight = layers_12_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("value_51_cast_fp16")]; tensor var_2923 = const()[name = tensor("op_2923"), val = tensor([1, 16, 64, 1])]; tensor mh_q_51_cast_fp16 = reshape(shape = var_2923, x = query_51_cast_fp16)[name = tensor("mh_q_51_cast_fp16")]; tensor var_2925_to_fp16 = const()[name = tensor("op_2925_to_fp16"), val = tensor(0x1p-3)]; tensor var_2926_cast_fp16 = mul(x = mh_q_51_cast_fp16, y = var_2925_to_fp16)[name = tensor("op_2926_cast_fp16")]; tensor var_2929 = const()[name = tensor("op_2929"), val = tensor([1, 16, 64, 1500])]; tensor var_2930_cast_fp16 = reshape(shape = var_2929, x = key_51_cast_fp16)[name = tensor("op_2930_cast_fp16")]; tensor mh_w_77_transpose_x_0 = const()[name = tensor("mh_w_77_transpose_x_0"), val = tensor(true)]; tensor mh_w_77_transpose_y_0 = const()[name = tensor("mh_w_77_transpose_y_0"), val = tensor(false)]; tensor mh_w_77_cast_fp16 = matmul(transpose_x = mh_w_77_transpose_x_0, transpose_y = mh_w_77_transpose_y_0, x = var_2926_cast_fp16, y = var_2930_cast_fp16)[name = tensor("mh_w_77_cast_fp16")]; tensor obj_181_cast_fp16 = softmax(axis = var_2772, x = mh_w_77_cast_fp16)[name = tensor("obj_181_cast_fp16")]; tensor var_2934 = const()[name = tensor("op_2934"), val = tensor([1, 16, 64, 1500])]; tensor var_2935_cast_fp16 = reshape(shape = var_2934, x = value_51_cast_fp16)[name = tensor("op_2935_cast_fp16")]; tensor attn_51_transpose_x_0 = const()[name = tensor("attn_51_transpose_x_0"), val = tensor(false)]; tensor attn_51_transpose_y_0 = const()[name = tensor("attn_51_transpose_y_0"), val = tensor(true)]; tensor attn_51_cast_fp16 = matmul(transpose_x = attn_51_transpose_x_0, transpose_y = attn_51_transpose_y_0, x = var_2935_cast_fp16, y = obj_181_cast_fp16)[name = tensor("attn_51_cast_fp16")]; tensor var_2938 = const()[name = tensor("op_2938"), val = tensor([1, 1024, 1, 1])]; tensor input_123_cast_fp16 = reshape(shape = var_2938, x = attn_51_cast_fp16)[name = tensor("input_123_cast_fp16")]; tensor obj_179_pad_type_0 = const()[name = tensor("obj_179_pad_type_0"), val = tensor("valid")]; tensor obj_179_strides_0 = const()[name = tensor("obj_179_strides_0"), val = tensor([1, 1])]; tensor obj_179_pad_0 = const()[name = tensor("obj_179_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_179_dilations_0 = const()[name = tensor("obj_179_dilations_0"), val = tensor([1, 1])]; tensor obj_179_groups_0 = const()[name = tensor("obj_179_groups_0"), val = tensor(1)]; tensor layers_12_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265432000))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(266218496))), name = tensor("layers_12_encoder_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_12_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_12_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(266218688)))]; tensor obj_179_cast_fp16 = conv(bias = layers_12_encoder_attn_o_proj_bias_to_fp16, dilations = obj_179_dilations_0, groups = obj_179_groups_0, pad = obj_179_pad_0, pad_type = obj_179_pad_type_0, strides = obj_179_strides_0, weight = layers_12_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_123_cast_fp16)[name = tensor("obj_179_cast_fp16")]; tensor inputs_77_cast_fp16 = add(x = inputs_75_cast_fp16, y = obj_179_cast_fp16)[name = tensor("inputs_77_cast_fp16")]; tensor out_77_axes_0 = const()[name = tensor("out_77_axes_0"), val = tensor([1])]; tensor var_2956_to_fp16 = const()[name = tensor("op_2956_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_77_cast_fp16 = layer_norm(axes = out_77_axes_0, epsilon = var_2956_to_fp16, x = inputs_77_cast_fp16)[name = tensor("out_77_cast_fp16")]; tensor input_125_gamma_0_to_fp16 = const()[name = tensor("input_125_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(266220800)))]; tensor input_125_beta_0_to_fp16 = const()[name = tensor("input_125_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(266222912)))]; tensor input_125_epsilon_0_to_fp16 = const()[name = tensor("input_125_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_125_cast_fp16 = batch_norm(beta = input_125_beta_0_to_fp16, epsilon = input_125_epsilon_0_to_fp16, gamma = input_125_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_77_cast_fp16)[name = tensor("input_125_cast_fp16")]; tensor input_127_pad_type_0 = const()[name = tensor("input_127_pad_type_0"), val = tensor("valid")]; tensor input_127_strides_0 = const()[name = tensor("input_127_strides_0"), val = tensor([1, 1])]; tensor input_127_pad_0 = const()[name = tensor("input_127_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_127_dilations_0 = const()[name = tensor("input_127_dilations_0"), val = tensor([1, 1])]; tensor input_127_groups_0 = const()[name = tensor("input_127_groups_0"), val = tensor(1)]; tensor layers_12_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(266225024))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(269370816))), name = tensor("layers_12_fc1_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; tensor layers_12_fc1_bias_to_fp16 = const()[name = tensor("layers_12_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(269371008)))]; tensor input_127_cast_fp16 = conv(bias = layers_12_fc1_bias_to_fp16, dilations = input_127_dilations_0, groups = input_127_groups_0, pad = input_127_pad_0, pad_type = input_127_pad_type_0, strides = input_127_strides_0, weight = layers_12_fc1_weight_to_fp16_palettized, x = input_125_cast_fp16)[name = tensor("input_127_cast_fp16")]; tensor input_129_mode_0 = const()[name = tensor("input_129_mode_0"), val = tensor("EXACT")]; tensor input_129_cast_fp16 = gelu(mode = input_129_mode_0, x = input_127_cast_fp16)[name = tensor("input_129_cast_fp16")]; tensor hidden_states_27_pad_type_0 = const()[name = tensor("hidden_states_27_pad_type_0"), val = tensor("valid")]; tensor hidden_states_27_strides_0 = const()[name = tensor("hidden_states_27_strides_0"), val = tensor([1, 1])]; tensor hidden_states_27_pad_0 = const()[name = tensor("hidden_states_27_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_27_dilations_0 = const()[name = tensor("hidden_states_27_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_27_groups_0 = const()[name = tensor("hidden_states_27_groups_0"), val = tensor(1)]; tensor layers_12_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(269379264))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(272525056))), name = tensor("layers_12_fc2_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; tensor layers_12_fc2_bias_to_fp16 = const()[name = tensor("layers_12_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(272525248)))]; tensor hidden_states_27_cast_fp16 = conv(bias = layers_12_fc2_bias_to_fp16, dilations = hidden_states_27_dilations_0, groups = hidden_states_27_groups_0, pad = hidden_states_27_pad_0, pad_type = hidden_states_27_pad_type_0, strides = hidden_states_27_strides_0, weight = layers_12_fc2_weight_to_fp16_palettized, x = input_129_cast_fp16)[name = tensor("hidden_states_27_cast_fp16")]; tensor inputs_79_cast_fp16 = add(x = inputs_77_cast_fp16, y = hidden_states_27_cast_fp16)[name = tensor("inputs_79_cast_fp16")]; tensor var_2991 = const()[name = tensor("op_2991"), val = tensor(3)]; tensor out_79_axes_0 = const()[name = tensor("out_79_axes_0"), val = tensor([1])]; tensor var_3016_to_fp16 = const()[name = tensor("op_3016_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_79_cast_fp16 = layer_norm(axes = out_79_axes_0, epsilon = var_3016_to_fp16, x = inputs_79_cast_fp16)[name = tensor("out_79_cast_fp16")]; tensor obj_183_gamma_0_to_fp16 = const()[name = tensor("obj_183_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(272527360)))]; tensor obj_183_beta_0_to_fp16 = const()[name = tensor("obj_183_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(272529472)))]; tensor obj_183_epsilon_0_to_fp16 = const()[name = tensor("obj_183_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_183_cast_fp16 = batch_norm(beta = obj_183_beta_0_to_fp16, epsilon = obj_183_epsilon_0_to_fp16, gamma = obj_183_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_79_cast_fp16)[name = tensor("obj_183_cast_fp16")]; tensor query_53_pad_type_0 = const()[name = tensor("query_53_pad_type_0"), val = tensor("valid")]; tensor query_53_strides_0 = const()[name = tensor("query_53_strides_0"), val = tensor([1, 1])]; tensor query_53_pad_0 = const()[name = tensor("query_53_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_53_dilations_0 = const()[name = tensor("query_53_dilations_0"), val = tensor([1, 1])]; tensor query_53_groups_0 = const()[name = tensor("query_53_groups_0"), val = tensor(1)]; tensor layers_13_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(272531584))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(273318080))), name = tensor("layers_13_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_13_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_13_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(273318272)))]; tensor query_53_cast_fp16 = conv(bias = layers_13_self_attn_q_proj_bias_to_fp16, dilations = query_53_dilations_0, groups = query_53_groups_0, pad = query_53_pad_0, pad_type = query_53_pad_type_0, strides = query_53_strides_0, weight = layers_13_self_attn_q_proj_weight_to_fp16_palettized, x = obj_183_cast_fp16)[name = tensor("query_53_cast_fp16")]; tensor current_key_27_pad_type_0 = const()[name = tensor("current_key_27_pad_type_0"), val = tensor("valid")]; tensor current_key_27_strides_0 = const()[name = tensor("current_key_27_strides_0"), val = tensor([1, 1])]; tensor current_key_27_pad_0 = const()[name = tensor("current_key_27_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_27_dilations_0 = const()[name = tensor("current_key_27_dilations_0"), val = tensor([1, 1])]; tensor current_key_27_groups_0 = const()[name = tensor("current_key_27_groups_0"), val = tensor(1)]; tensor layers_13_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(273320384))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(274106880))), name = tensor("layers_13_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor current_key_27_cast_fp16 = conv(dilations = current_key_27_dilations_0, groups = current_key_27_groups_0, pad = current_key_27_pad_0, pad_type = current_key_27_pad_type_0, strides = current_key_27_strides_0, weight = layers_13_self_attn_k_proj_weight_to_fp16_palettized, x = obj_183_cast_fp16)[name = tensor("current_key_27_cast_fp16")]; tensor current_value_27_pad_type_0 = const()[name = tensor("current_value_27_pad_type_0"), val = tensor("valid")]; tensor current_value_27_strides_0 = const()[name = tensor("current_value_27_strides_0"), val = tensor([1, 1])]; tensor current_value_27_pad_0 = const()[name = tensor("current_value_27_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_27_dilations_0 = const()[name = tensor("current_value_27_dilations_0"), val = tensor([1, 1])]; tensor current_value_27_groups_0 = const()[name = tensor("current_value_27_groups_0"), val = tensor(1)]; tensor layers_13_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(274107072))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(274893568))), name = tensor("layers_13_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_13_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_13_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(274893760)))]; tensor current_value_27_cast_fp16 = conv(bias = layers_13_self_attn_v_proj_bias_to_fp16, dilations = current_value_27_dilations_0, groups = current_value_27_groups_0, pad = current_value_27_pad_0, pad_type = current_value_27_pad_type_0, strides = current_value_27_strides_0, weight = layers_13_self_attn_v_proj_weight_to_fp16_palettized, x = obj_183_cast_fp16)[name = tensor("current_value_27_cast_fp16")]; tensor var_3055_cast_fp16 = mul(x = var_87_cast_fp16_13, y = var_207_cast_fp16)[name = tensor("op_3055_cast_fp16")]; tensor var_3056_cast_fp16 = mul(x = current_key_27_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_3056_cast_fp16")]; tensor key_53_cast_fp16 = add(x = var_3055_cast_fp16, y = var_3056_cast_fp16)[name = tensor("key_53_cast_fp16")]; tensor var_3059_cast_fp16 = mul(x = var_114_cast_fp16_13, y = var_207_cast_fp16)[name = tensor("op_3059_cast_fp16")]; tensor var_3060_cast_fp16 = mul(x = current_value_27_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_3060_cast_fp16")]; tensor value_53_cast_fp16 = add(x = var_3059_cast_fp16, y = var_3060_cast_fp16)[name = tensor("value_53_cast_fp16")]; tensor var_3064 = const()[name = tensor("op_3064"), val = tensor([1, 16, 64, 1])]; tensor mh_q_53_cast_fp16 = reshape(shape = var_3064, x = query_53_cast_fp16)[name = tensor("mh_q_53_cast_fp16")]; tensor var_3066_to_fp16 = const()[name = tensor("op_3066_to_fp16"), val = tensor(0x1p-3)]; tensor var_3067_cast_fp16 = mul(x = mh_q_53_cast_fp16, y = var_3066_to_fp16)[name = tensor("op_3067_cast_fp16")]; tensor var_3070 = const()[name = tensor("op_3070"), val = tensor([1, 16, 64, 448])]; tensor var_3071_cast_fp16 = reshape(shape = var_3070, x = key_53_cast_fp16)[name = tensor("op_3071_cast_fp16")]; tensor mh_w_79_transpose_x_0 = const()[name = tensor("mh_w_79_transpose_x_0"), val = tensor(true)]; tensor mh_w_79_transpose_y_0 = const()[name = tensor("mh_w_79_transpose_y_0"), val = tensor(false)]; tensor mh_w_79_cast_fp16 = matmul(transpose_x = mh_w_79_transpose_x_0, transpose_y = mh_w_79_transpose_y_0, x = var_3067_cast_fp16, y = var_3071_cast_fp16)[name = tensor("mh_w_79_cast_fp16")]; tensor mh_w_81_cast_fp16 = add(x = mh_w_79_cast_fp16, y = var_229_cast_fp16)[name = tensor("mh_w_81_cast_fp16")]; tensor var_3079_cast_fp16 = softmax(axis = var_2991, x = mh_w_81_cast_fp16)[name = tensor("op_3079_cast_fp16")]; tensor var_3080 = const()[name = tensor("op_3080"), val = tensor([1, 16, 64, 448])]; tensor var_3081_cast_fp16 = reshape(shape = var_3080, x = value_53_cast_fp16)[name = tensor("op_3081_cast_fp16")]; tensor attn_53_transpose_x_0 = const()[name = tensor("attn_53_transpose_x_0"), val = tensor(false)]; tensor attn_53_transpose_y_0 = const()[name = tensor("attn_53_transpose_y_0"), val = tensor(true)]; tensor attn_53_cast_fp16 = matmul(transpose_x = attn_53_transpose_x_0, transpose_y = attn_53_transpose_y_0, x = var_3081_cast_fp16, y = var_3079_cast_fp16)[name = tensor("attn_53_cast_fp16")]; tensor var_3084 = const()[name = tensor("op_3084"), val = tensor([1, 1024, 1, 1])]; tensor input_131_cast_fp16 = reshape(shape = var_3084, x = attn_53_cast_fp16)[name = tensor("input_131_cast_fp16")]; tensor obj_189_pad_type_0 = const()[name = tensor("obj_189_pad_type_0"), val = tensor("valid")]; tensor obj_189_strides_0 = const()[name = tensor("obj_189_strides_0"), val = tensor([1, 1])]; tensor obj_189_pad_0 = const()[name = tensor("obj_189_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_189_dilations_0 = const()[name = tensor("obj_189_dilations_0"), val = tensor([1, 1])]; tensor obj_189_groups_0 = const()[name = tensor("obj_189_groups_0"), val = tensor(1)]; tensor layers_13_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(274895872))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(275682368))), name = tensor("layers_13_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_13_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_13_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(275682560)))]; tensor obj_189_cast_fp16 = conv(bias = layers_13_self_attn_o_proj_bias_to_fp16, dilations = obj_189_dilations_0, groups = obj_189_groups_0, pad = obj_189_pad_0, pad_type = obj_189_pad_type_0, strides = obj_189_strides_0, weight = layers_13_self_attn_o_proj_weight_to_fp16_palettized, x = input_131_cast_fp16)[name = tensor("obj_189_cast_fp16")]; tensor inputs_81_cast_fp16 = add(x = inputs_79_cast_fp16, y = obj_189_cast_fp16)[name = tensor("inputs_81_cast_fp16")]; tensor out_81_axes_0 = const()[name = tensor("out_81_axes_0"), val = tensor([1])]; tensor var_3106_to_fp16 = const()[name = tensor("op_3106_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_81_cast_fp16 = layer_norm(axes = out_81_axes_0, epsilon = var_3106_to_fp16, x = inputs_81_cast_fp16)[name = tensor("out_81_cast_fp16")]; tensor obj_191_gamma_0_to_fp16 = const()[name = tensor("obj_191_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(275684672)))]; tensor obj_191_beta_0_to_fp16 = const()[name = tensor("obj_191_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(275686784)))]; tensor obj_191_epsilon_0_to_fp16 = const()[name = tensor("obj_191_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_191_cast_fp16 = batch_norm(beta = obj_191_beta_0_to_fp16, epsilon = obj_191_epsilon_0_to_fp16, gamma = obj_191_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_81_cast_fp16)[name = tensor("obj_191_cast_fp16")]; tensor query_55_pad_type_0 = const()[name = tensor("query_55_pad_type_0"), val = tensor("valid")]; tensor query_55_strides_0 = const()[name = tensor("query_55_strides_0"), val = tensor([1, 1])]; tensor query_55_pad_0 = const()[name = tensor("query_55_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_55_dilations_0 = const()[name = tensor("query_55_dilations_0"), val = tensor([1, 1])]; tensor query_55_groups_0 = const()[name = tensor("query_55_groups_0"), val = tensor(1)]; tensor layers_13_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(275688896))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(276475392))), name = tensor("layers_13_encoder_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_13_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_13_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(276475584)))]; tensor query_55_cast_fp16 = conv(bias = layers_13_encoder_attn_q_proj_bias_to_fp16, dilations = query_55_dilations_0, groups = query_55_groups_0, pad = query_55_pad_0, pad_type = query_55_pad_type_0, strides = query_55_strides_0, weight = layers_13_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_191_cast_fp16)[name = tensor("query_55_cast_fp16")]; tensor key_55_pad_type_0 = const()[name = tensor("key_55_pad_type_0"), val = tensor("valid")]; tensor key_55_strides_0 = const()[name = tensor("key_55_strides_0"), val = tensor([1, 1])]; tensor key_55_pad_0 = const()[name = tensor("key_55_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_55_dilations_0 = const()[name = tensor("key_55_dilations_0"), val = tensor([1, 1])]; tensor key_55_groups_0 = const()[name = tensor("key_55_groups_0"), val = tensor(1)]; tensor layers_13_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(276477696))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(277264192))), name = tensor("layers_13_encoder_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor key_55_cast_fp16 = conv(dilations = key_55_dilations_0, groups = key_55_groups_0, pad = key_55_pad_0, pad_type = key_55_pad_type_0, strides = key_55_strides_0, weight = layers_13_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("key_55_cast_fp16")]; tensor value_55_pad_type_0 = const()[name = tensor("value_55_pad_type_0"), val = tensor("valid")]; tensor value_55_strides_0 = const()[name = tensor("value_55_strides_0"), val = tensor([1, 1])]; tensor value_55_pad_0 = const()[name = tensor("value_55_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_55_dilations_0 = const()[name = tensor("value_55_dilations_0"), val = tensor([1, 1])]; tensor value_55_groups_0 = const()[name = tensor("value_55_groups_0"), val = tensor(1)]; tensor layers_13_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(277264384))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(278050880))), name = tensor("layers_13_encoder_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_13_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_13_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(278051072)))]; tensor value_55_cast_fp16 = conv(bias = layers_13_encoder_attn_v_proj_bias_to_fp16, dilations = value_55_dilations_0, groups = value_55_groups_0, pad = value_55_pad_0, pad_type = value_55_pad_type_0, strides = value_55_strides_0, weight = layers_13_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("value_55_cast_fp16")]; tensor var_3142 = const()[name = tensor("op_3142"), val = tensor([1, 16, 64, 1])]; tensor mh_q_55_cast_fp16 = reshape(shape = var_3142, x = query_55_cast_fp16)[name = tensor("mh_q_55_cast_fp16")]; tensor var_3144_to_fp16 = const()[name = tensor("op_3144_to_fp16"), val = tensor(0x1p-3)]; tensor var_3145_cast_fp16 = mul(x = mh_q_55_cast_fp16, y = var_3144_to_fp16)[name = tensor("op_3145_cast_fp16")]; tensor var_3148 = const()[name = tensor("op_3148"), val = tensor([1, 16, 64, 1500])]; tensor var_3149_cast_fp16 = reshape(shape = var_3148, x = key_55_cast_fp16)[name = tensor("op_3149_cast_fp16")]; tensor mh_w_83_transpose_x_0 = const()[name = tensor("mh_w_83_transpose_x_0"), val = tensor(true)]; tensor mh_w_83_transpose_y_0 = const()[name = tensor("mh_w_83_transpose_y_0"), val = tensor(false)]; tensor mh_w_83_cast_fp16 = matmul(transpose_x = mh_w_83_transpose_x_0, transpose_y = mh_w_83_transpose_y_0, x = var_3145_cast_fp16, y = var_3149_cast_fp16)[name = tensor("mh_w_83_cast_fp16")]; tensor obj_195_cast_fp16 = softmax(axis = var_2991, x = mh_w_83_cast_fp16)[name = tensor("obj_195_cast_fp16")]; tensor var_3153 = const()[name = tensor("op_3153"), val = tensor([1, 16, 64, 1500])]; tensor var_3154_cast_fp16 = reshape(shape = var_3153, x = value_55_cast_fp16)[name = tensor("op_3154_cast_fp16")]; tensor attn_55_transpose_x_0 = const()[name = tensor("attn_55_transpose_x_0"), val = tensor(false)]; tensor attn_55_transpose_y_0 = const()[name = tensor("attn_55_transpose_y_0"), val = tensor(true)]; tensor attn_55_cast_fp16 = matmul(transpose_x = attn_55_transpose_x_0, transpose_y = attn_55_transpose_y_0, x = var_3154_cast_fp16, y = obj_195_cast_fp16)[name = tensor("attn_55_cast_fp16")]; tensor var_3157 = const()[name = tensor("op_3157"), val = tensor([1, 1024, 1, 1])]; tensor input_133_cast_fp16 = reshape(shape = var_3157, x = attn_55_cast_fp16)[name = tensor("input_133_cast_fp16")]; tensor obj_193_pad_type_0 = const()[name = tensor("obj_193_pad_type_0"), val = tensor("valid")]; tensor obj_193_strides_0 = const()[name = tensor("obj_193_strides_0"), val = tensor([1, 1])]; tensor obj_193_pad_0 = const()[name = tensor("obj_193_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_193_dilations_0 = const()[name = tensor("obj_193_dilations_0"), val = tensor([1, 1])]; tensor obj_193_groups_0 = const()[name = tensor("obj_193_groups_0"), val = tensor(1)]; tensor layers_13_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(278053184))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(278839680))), name = tensor("layers_13_encoder_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_13_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_13_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(278839872)))]; tensor obj_193_cast_fp16 = conv(bias = layers_13_encoder_attn_o_proj_bias_to_fp16, dilations = obj_193_dilations_0, groups = obj_193_groups_0, pad = obj_193_pad_0, pad_type = obj_193_pad_type_0, strides = obj_193_strides_0, weight = layers_13_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_133_cast_fp16)[name = tensor("obj_193_cast_fp16")]; tensor inputs_83_cast_fp16 = add(x = inputs_81_cast_fp16, y = obj_193_cast_fp16)[name = tensor("inputs_83_cast_fp16")]; tensor out_83_axes_0 = const()[name = tensor("out_83_axes_0"), val = tensor([1])]; tensor var_3178_to_fp16 = const()[name = tensor("op_3178_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_83_cast_fp16 = layer_norm(axes = out_83_axes_0, epsilon = var_3178_to_fp16, x = inputs_83_cast_fp16)[name = tensor("out_83_cast_fp16")]; tensor input_135_gamma_0_to_fp16 = const()[name = tensor("input_135_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(278841984)))]; tensor input_135_beta_0_to_fp16 = const()[name = tensor("input_135_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(278844096)))]; tensor input_135_epsilon_0_to_fp16 = const()[name = tensor("input_135_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_135_cast_fp16 = batch_norm(beta = input_135_beta_0_to_fp16, epsilon = input_135_epsilon_0_to_fp16, gamma = input_135_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_83_cast_fp16)[name = tensor("input_135_cast_fp16")]; tensor input_137_pad_type_0 = const()[name = tensor("input_137_pad_type_0"), val = tensor("valid")]; tensor input_137_strides_0 = const()[name = tensor("input_137_strides_0"), val = tensor([1, 1])]; tensor input_137_pad_0 = const()[name = tensor("input_137_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_137_dilations_0 = const()[name = tensor("input_137_dilations_0"), val = tensor([1, 1])]; tensor input_137_groups_0 = const()[name = tensor("input_137_groups_0"), val = tensor(1)]; tensor layers_13_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(278846208))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(281992000))), name = tensor("layers_13_fc1_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; tensor layers_13_fc1_bias_to_fp16 = const()[name = tensor("layers_13_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(281992192)))]; tensor input_137_cast_fp16 = conv(bias = layers_13_fc1_bias_to_fp16, dilations = input_137_dilations_0, groups = input_137_groups_0, pad = input_137_pad_0, pad_type = input_137_pad_type_0, strides = input_137_strides_0, weight = layers_13_fc1_weight_to_fp16_palettized, x = input_135_cast_fp16)[name = tensor("input_137_cast_fp16")]; tensor input_139_mode_0 = const()[name = tensor("input_139_mode_0"), val = tensor("EXACT")]; tensor input_139_cast_fp16 = gelu(mode = input_139_mode_0, x = input_137_cast_fp16)[name = tensor("input_139_cast_fp16")]; tensor hidden_states_29_pad_type_0 = const()[name = tensor("hidden_states_29_pad_type_0"), val = tensor("valid")]; tensor hidden_states_29_strides_0 = const()[name = tensor("hidden_states_29_strides_0"), val = tensor([1, 1])]; tensor hidden_states_29_pad_0 = const()[name = tensor("hidden_states_29_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_29_dilations_0 = const()[name = tensor("hidden_states_29_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_29_groups_0 = const()[name = tensor("hidden_states_29_groups_0"), val = tensor(1)]; tensor layers_13_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(282000448))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(285146240))), name = tensor("layers_13_fc2_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; tensor layers_13_fc2_bias_to_fp16 = const()[name = tensor("layers_13_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(285146432)))]; tensor hidden_states_29_cast_fp16 = conv(bias = layers_13_fc2_bias_to_fp16, dilations = hidden_states_29_dilations_0, groups = hidden_states_29_groups_0, pad = hidden_states_29_pad_0, pad_type = hidden_states_29_pad_type_0, strides = hidden_states_29_strides_0, weight = layers_13_fc2_weight_to_fp16_palettized, x = input_139_cast_fp16)[name = tensor("hidden_states_29_cast_fp16")]; tensor inputs_85_cast_fp16 = add(x = inputs_83_cast_fp16, y = hidden_states_29_cast_fp16)[name = tensor("inputs_85_cast_fp16")]; tensor var_3214 = const()[name = tensor("op_3214"), val = tensor(3)]; tensor out_85_axes_0 = const()[name = tensor("out_85_axes_0"), val = tensor([1])]; tensor var_3239_to_fp16 = const()[name = tensor("op_3239_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_85_cast_fp16 = layer_norm(axes = out_85_axes_0, epsilon = var_3239_to_fp16, x = inputs_85_cast_fp16)[name = tensor("out_85_cast_fp16")]; tensor obj_197_gamma_0_to_fp16 = const()[name = tensor("obj_197_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(285148544)))]; tensor obj_197_beta_0_to_fp16 = const()[name = tensor("obj_197_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(285150656)))]; tensor obj_197_epsilon_0_to_fp16 = const()[name = tensor("obj_197_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_197_cast_fp16 = batch_norm(beta = obj_197_beta_0_to_fp16, epsilon = obj_197_epsilon_0_to_fp16, gamma = obj_197_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_85_cast_fp16)[name = tensor("obj_197_cast_fp16")]; tensor query_57_pad_type_0 = const()[name = tensor("query_57_pad_type_0"), val = tensor("valid")]; tensor query_57_strides_0 = const()[name = tensor("query_57_strides_0"), val = tensor([1, 1])]; tensor query_57_pad_0 = const()[name = tensor("query_57_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_57_dilations_0 = const()[name = tensor("query_57_dilations_0"), val = tensor([1, 1])]; tensor query_57_groups_0 = const()[name = tensor("query_57_groups_0"), val = tensor(1)]; tensor layers_14_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(285152768))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(285939264))), name = tensor("layers_14_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_14_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_14_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(285939456)))]; tensor query_57_cast_fp16 = conv(bias = layers_14_self_attn_q_proj_bias_to_fp16, dilations = query_57_dilations_0, groups = query_57_groups_0, pad = query_57_pad_0, pad_type = query_57_pad_type_0, strides = query_57_strides_0, weight = layers_14_self_attn_q_proj_weight_to_fp16_palettized, x = obj_197_cast_fp16)[name = tensor("query_57_cast_fp16")]; tensor current_key_29_pad_type_0 = const()[name = tensor("current_key_29_pad_type_0"), val = tensor("valid")]; tensor current_key_29_strides_0 = const()[name = tensor("current_key_29_strides_0"), val = tensor([1, 1])]; tensor current_key_29_pad_0 = const()[name = tensor("current_key_29_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_29_dilations_0 = const()[name = tensor("current_key_29_dilations_0"), val = tensor([1, 1])]; tensor current_key_29_groups_0 = const()[name = tensor("current_key_29_groups_0"), val = tensor(1)]; tensor layers_14_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(285941568))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(286728064))), name = tensor("layers_14_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor current_key_29_cast_fp16 = conv(dilations = current_key_29_dilations_0, groups = current_key_29_groups_0, pad = current_key_29_pad_0, pad_type = current_key_29_pad_type_0, strides = current_key_29_strides_0, weight = layers_14_self_attn_k_proj_weight_to_fp16_palettized, x = obj_197_cast_fp16)[name = tensor("current_key_29_cast_fp16")]; tensor current_value_29_pad_type_0 = const()[name = tensor("current_value_29_pad_type_0"), val = tensor("valid")]; tensor current_value_29_strides_0 = const()[name = tensor("current_value_29_strides_0"), val = tensor([1, 1])]; tensor current_value_29_pad_0 = const()[name = tensor("current_value_29_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_29_dilations_0 = const()[name = tensor("current_value_29_dilations_0"), val = tensor([1, 1])]; tensor current_value_29_groups_0 = const()[name = tensor("current_value_29_groups_0"), val = tensor(1)]; tensor layers_14_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(286728256))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(287514752))), name = tensor("layers_14_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_14_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_14_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(287514944)))]; tensor current_value_29_cast_fp16 = conv(bias = layers_14_self_attn_v_proj_bias_to_fp16, dilations = current_value_29_dilations_0, groups = current_value_29_groups_0, pad = current_value_29_pad_0, pad_type = current_value_29_pad_type_0, strides = current_value_29_strides_0, weight = layers_14_self_attn_v_proj_weight_to_fp16_palettized, x = obj_197_cast_fp16)[name = tensor("current_value_29_cast_fp16")]; tensor var_3278_cast_fp16 = mul(x = var_87_cast_fp16_14, y = var_207_cast_fp16)[name = tensor("op_3278_cast_fp16")]; tensor var_3279_cast_fp16 = mul(x = current_key_29_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_3279_cast_fp16")]; tensor key_57_cast_fp16 = add(x = var_3278_cast_fp16, y = var_3279_cast_fp16)[name = tensor("key_57_cast_fp16")]; tensor var_3282_cast_fp16 = mul(x = var_114_cast_fp16_14, y = var_207_cast_fp16)[name = tensor("op_3282_cast_fp16")]; tensor var_3283_cast_fp16 = mul(x = current_value_29_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_3283_cast_fp16")]; tensor value_57_cast_fp16 = add(x = var_3282_cast_fp16, y = var_3283_cast_fp16)[name = tensor("value_57_cast_fp16")]; tensor var_3287 = const()[name = tensor("op_3287"), val = tensor([1, 16, 64, 1])]; tensor mh_q_57_cast_fp16 = reshape(shape = var_3287, x = query_57_cast_fp16)[name = tensor("mh_q_57_cast_fp16")]; tensor var_3289_to_fp16 = const()[name = tensor("op_3289_to_fp16"), val = tensor(0x1p-3)]; tensor var_3290_cast_fp16 = mul(x = mh_q_57_cast_fp16, y = var_3289_to_fp16)[name = tensor("op_3290_cast_fp16")]; tensor var_3293 = const()[name = tensor("op_3293"), val = tensor([1, 16, 64, 448])]; tensor var_3294_cast_fp16 = reshape(shape = var_3293, x = key_57_cast_fp16)[name = tensor("op_3294_cast_fp16")]; tensor mh_w_85_transpose_x_0 = const()[name = tensor("mh_w_85_transpose_x_0"), val = tensor(true)]; tensor mh_w_85_transpose_y_0 = const()[name = tensor("mh_w_85_transpose_y_0"), val = tensor(false)]; tensor mh_w_85_cast_fp16 = matmul(transpose_x = mh_w_85_transpose_x_0, transpose_y = mh_w_85_transpose_y_0, x = var_3290_cast_fp16, y = var_3294_cast_fp16)[name = tensor("mh_w_85_cast_fp16")]; tensor mh_w_87_cast_fp16 = add(x = mh_w_85_cast_fp16, y = var_229_cast_fp16)[name = tensor("mh_w_87_cast_fp16")]; tensor var_3302_cast_fp16 = softmax(axis = var_3214, x = mh_w_87_cast_fp16)[name = tensor("op_3302_cast_fp16")]; tensor var_3303 = const()[name = tensor("op_3303"), val = tensor([1, 16, 64, 448])]; tensor var_3304_cast_fp16 = reshape(shape = var_3303, x = value_57_cast_fp16)[name = tensor("op_3304_cast_fp16")]; tensor attn_57_transpose_x_0 = const()[name = tensor("attn_57_transpose_x_0"), val = tensor(false)]; tensor attn_57_transpose_y_0 = const()[name = tensor("attn_57_transpose_y_0"), val = tensor(true)]; tensor attn_57_cast_fp16 = matmul(transpose_x = attn_57_transpose_x_0, transpose_y = attn_57_transpose_y_0, x = var_3304_cast_fp16, y = var_3302_cast_fp16)[name = tensor("attn_57_cast_fp16")]; tensor var_3307 = const()[name = tensor("op_3307"), val = tensor([1, 1024, 1, 1])]; tensor input_141_cast_fp16 = reshape(shape = var_3307, x = attn_57_cast_fp16)[name = tensor("input_141_cast_fp16")]; tensor obj_203_pad_type_0 = const()[name = tensor("obj_203_pad_type_0"), val = tensor("valid")]; tensor obj_203_strides_0 = const()[name = tensor("obj_203_strides_0"), val = tensor([1, 1])]; tensor obj_203_pad_0 = const()[name = tensor("obj_203_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_203_dilations_0 = const()[name = tensor("obj_203_dilations_0"), val = tensor([1, 1])]; tensor obj_203_groups_0 = const()[name = tensor("obj_203_groups_0"), val = tensor(1)]; tensor layers_14_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(287517056))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(288303552))), name = tensor("layers_14_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_14_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_14_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(288303744)))]; tensor obj_203_cast_fp16 = conv(bias = layers_14_self_attn_o_proj_bias_to_fp16, dilations = obj_203_dilations_0, groups = obj_203_groups_0, pad = obj_203_pad_0, pad_type = obj_203_pad_type_0, strides = obj_203_strides_0, weight = layers_14_self_attn_o_proj_weight_to_fp16_palettized, x = input_141_cast_fp16)[name = tensor("obj_203_cast_fp16")]; tensor inputs_87_cast_fp16 = add(x = inputs_85_cast_fp16, y = obj_203_cast_fp16)[name = tensor("inputs_87_cast_fp16")]; tensor out_87_axes_0 = const()[name = tensor("out_87_axes_0"), val = tensor([1])]; tensor var_3329_to_fp16 = const()[name = tensor("op_3329_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_87_cast_fp16 = layer_norm(axes = out_87_axes_0, epsilon = var_3329_to_fp16, x = inputs_87_cast_fp16)[name = tensor("out_87_cast_fp16")]; tensor obj_205_gamma_0_to_fp16 = const()[name = tensor("obj_205_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(288305856)))]; tensor obj_205_beta_0_to_fp16 = const()[name = tensor("obj_205_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(288307968)))]; tensor obj_205_epsilon_0_to_fp16 = const()[name = tensor("obj_205_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_205_cast_fp16 = batch_norm(beta = obj_205_beta_0_to_fp16, epsilon = obj_205_epsilon_0_to_fp16, gamma = obj_205_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_87_cast_fp16)[name = tensor("obj_205_cast_fp16")]; tensor query_59_pad_type_0 = const()[name = tensor("query_59_pad_type_0"), val = tensor("valid")]; tensor query_59_strides_0 = const()[name = tensor("query_59_strides_0"), val = tensor([1, 1])]; tensor query_59_pad_0 = const()[name = tensor("query_59_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_59_dilations_0 = const()[name = tensor("query_59_dilations_0"), val = tensor([1, 1])]; tensor query_59_groups_0 = const()[name = tensor("query_59_groups_0"), val = tensor(1)]; tensor layers_14_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(288310080))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289096576))), name = tensor("layers_14_encoder_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_14_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_14_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289096768)))]; tensor query_59_cast_fp16 = conv(bias = layers_14_encoder_attn_q_proj_bias_to_fp16, dilations = query_59_dilations_0, groups = query_59_groups_0, pad = query_59_pad_0, pad_type = query_59_pad_type_0, strides = query_59_strides_0, weight = layers_14_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_205_cast_fp16)[name = tensor("query_59_cast_fp16")]; tensor key_59_pad_type_0 = const()[name = tensor("key_59_pad_type_0"), val = tensor("valid")]; tensor key_59_strides_0 = const()[name = tensor("key_59_strides_0"), val = tensor([1, 1])]; tensor key_59_pad_0 = const()[name = tensor("key_59_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_59_dilations_0 = const()[name = tensor("key_59_dilations_0"), val = tensor([1, 1])]; tensor key_59_groups_0 = const()[name = tensor("key_59_groups_0"), val = tensor(1)]; tensor layers_14_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289098880))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289885376))), name = tensor("layers_14_encoder_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor key_59_cast_fp16 = conv(dilations = key_59_dilations_0, groups = key_59_groups_0, pad = key_59_pad_0, pad_type = key_59_pad_type_0, strides = key_59_strides_0, weight = layers_14_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("key_59_cast_fp16")]; tensor value_59_pad_type_0 = const()[name = tensor("value_59_pad_type_0"), val = tensor("valid")]; tensor value_59_strides_0 = const()[name = tensor("value_59_strides_0"), val = tensor([1, 1])]; tensor value_59_pad_0 = const()[name = tensor("value_59_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_59_dilations_0 = const()[name = tensor("value_59_dilations_0"), val = tensor([1, 1])]; tensor value_59_groups_0 = const()[name = tensor("value_59_groups_0"), val = tensor(1)]; tensor layers_14_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289885568))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(290672064))), name = tensor("layers_14_encoder_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_14_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_14_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(290672256)))]; tensor value_59_cast_fp16 = conv(bias = layers_14_encoder_attn_v_proj_bias_to_fp16, dilations = value_59_dilations_0, groups = value_59_groups_0, pad = value_59_pad_0, pad_type = value_59_pad_type_0, strides = value_59_strides_0, weight = layers_14_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("value_59_cast_fp16")]; tensor var_3365 = const()[name = tensor("op_3365"), val = tensor([1, 16, 64, 1])]; tensor mh_q_59_cast_fp16 = reshape(shape = var_3365, x = query_59_cast_fp16)[name = tensor("mh_q_59_cast_fp16")]; tensor var_3367_to_fp16 = const()[name = tensor("op_3367_to_fp16"), val = tensor(0x1p-3)]; tensor var_3368_cast_fp16 = mul(x = mh_q_59_cast_fp16, y = var_3367_to_fp16)[name = tensor("op_3368_cast_fp16")]; tensor var_3371 = const()[name = tensor("op_3371"), val = tensor([1, 16, 64, 1500])]; tensor var_3372_cast_fp16 = reshape(shape = var_3371, x = key_59_cast_fp16)[name = tensor("op_3372_cast_fp16")]; tensor mh_w_89_transpose_x_0 = const()[name = tensor("mh_w_89_transpose_x_0"), val = tensor(true)]; tensor mh_w_89_transpose_y_0 = const()[name = tensor("mh_w_89_transpose_y_0"), val = tensor(false)]; tensor mh_w_89_cast_fp16 = matmul(transpose_x = mh_w_89_transpose_x_0, transpose_y = mh_w_89_transpose_y_0, x = var_3368_cast_fp16, y = var_3372_cast_fp16)[name = tensor("mh_w_89_cast_fp16")]; tensor obj_209_cast_fp16 = softmax(axis = var_3214, x = mh_w_89_cast_fp16)[name = tensor("obj_209_cast_fp16")]; tensor var_3376 = const()[name = tensor("op_3376"), val = tensor([1, 16, 64, 1500])]; tensor var_3377_cast_fp16 = reshape(shape = var_3376, x = value_59_cast_fp16)[name = tensor("op_3377_cast_fp16")]; tensor attn_59_transpose_x_0 = const()[name = tensor("attn_59_transpose_x_0"), val = tensor(false)]; tensor attn_59_transpose_y_0 = const()[name = tensor("attn_59_transpose_y_0"), val = tensor(true)]; tensor attn_59_cast_fp16 = matmul(transpose_x = attn_59_transpose_x_0, transpose_y = attn_59_transpose_y_0, x = var_3377_cast_fp16, y = obj_209_cast_fp16)[name = tensor("attn_59_cast_fp16")]; tensor var_3380 = const()[name = tensor("op_3380"), val = tensor([1, 1024, 1, 1])]; tensor input_143_cast_fp16 = reshape(shape = var_3380, x = attn_59_cast_fp16)[name = tensor("input_143_cast_fp16")]; tensor obj_207_pad_type_0 = const()[name = tensor("obj_207_pad_type_0"), val = tensor("valid")]; tensor obj_207_strides_0 = const()[name = tensor("obj_207_strides_0"), val = tensor([1, 1])]; tensor obj_207_pad_0 = const()[name = tensor("obj_207_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_207_dilations_0 = const()[name = tensor("obj_207_dilations_0"), val = tensor([1, 1])]; tensor obj_207_groups_0 = const()[name = tensor("obj_207_groups_0"), val = tensor(1)]; tensor layers_14_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(290674368))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(291460864))), name = tensor("layers_14_encoder_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_14_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_14_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(291461056)))]; tensor obj_207_cast_fp16 = conv(bias = layers_14_encoder_attn_o_proj_bias_to_fp16, dilations = obj_207_dilations_0, groups = obj_207_groups_0, pad = obj_207_pad_0, pad_type = obj_207_pad_type_0, strides = obj_207_strides_0, weight = layers_14_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_143_cast_fp16)[name = tensor("obj_207_cast_fp16")]; tensor inputs_89_cast_fp16 = add(x = inputs_87_cast_fp16, y = obj_207_cast_fp16)[name = tensor("inputs_89_cast_fp16")]; tensor out_89_axes_0 = const()[name = tensor("out_89_axes_0"), val = tensor([1])]; tensor var_3398_to_fp16 = const()[name = tensor("op_3398_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_89_cast_fp16 = layer_norm(axes = out_89_axes_0, epsilon = var_3398_to_fp16, x = inputs_89_cast_fp16)[name = tensor("out_89_cast_fp16")]; tensor input_145_gamma_0_to_fp16 = const()[name = tensor("input_145_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(291463168)))]; tensor input_145_beta_0_to_fp16 = const()[name = tensor("input_145_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(291465280)))]; tensor input_145_epsilon_0_to_fp16 = const()[name = tensor("input_145_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_145_cast_fp16 = batch_norm(beta = input_145_beta_0_to_fp16, epsilon = input_145_epsilon_0_to_fp16, gamma = input_145_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_89_cast_fp16)[name = tensor("input_145_cast_fp16")]; tensor input_147_pad_type_0 = const()[name = tensor("input_147_pad_type_0"), val = tensor("valid")]; tensor input_147_strides_0 = const()[name = tensor("input_147_strides_0"), val = tensor([1, 1])]; tensor input_147_pad_0 = const()[name = tensor("input_147_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_147_dilations_0 = const()[name = tensor("input_147_dilations_0"), val = tensor([1, 1])]; tensor input_147_groups_0 = const()[name = tensor("input_147_groups_0"), val = tensor(1)]; tensor layers_14_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(291467392))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(294613184))), name = tensor("layers_14_fc1_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; tensor layers_14_fc1_bias_to_fp16 = const()[name = tensor("layers_14_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(294613376)))]; tensor input_147_cast_fp16 = conv(bias = layers_14_fc1_bias_to_fp16, dilations = input_147_dilations_0, groups = input_147_groups_0, pad = input_147_pad_0, pad_type = input_147_pad_type_0, strides = input_147_strides_0, weight = layers_14_fc1_weight_to_fp16_palettized, x = input_145_cast_fp16)[name = tensor("input_147_cast_fp16")]; tensor input_149_mode_0 = const()[name = tensor("input_149_mode_0"), val = tensor("EXACT")]; tensor input_149_cast_fp16 = gelu(mode = input_149_mode_0, x = input_147_cast_fp16)[name = tensor("input_149_cast_fp16")]; tensor hidden_states_31_pad_type_0 = const()[name = tensor("hidden_states_31_pad_type_0"), val = tensor("valid")]; tensor hidden_states_31_strides_0 = const()[name = tensor("hidden_states_31_strides_0"), val = tensor([1, 1])]; tensor hidden_states_31_pad_0 = const()[name = tensor("hidden_states_31_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_31_dilations_0 = const()[name = tensor("hidden_states_31_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_31_groups_0 = const()[name = tensor("hidden_states_31_groups_0"), val = tensor(1)]; tensor layers_14_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(294621632))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(297767424))), name = tensor("layers_14_fc2_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; tensor layers_14_fc2_bias_to_fp16 = const()[name = tensor("layers_14_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(297767616)))]; tensor hidden_states_31_cast_fp16 = conv(bias = layers_14_fc2_bias_to_fp16, dilations = hidden_states_31_dilations_0, groups = hidden_states_31_groups_0, pad = hidden_states_31_pad_0, pad_type = hidden_states_31_pad_type_0, strides = hidden_states_31_strides_0, weight = layers_14_fc2_weight_to_fp16_palettized, x = input_149_cast_fp16)[name = tensor("hidden_states_31_cast_fp16")]; tensor inputs_91_cast_fp16 = add(x = inputs_89_cast_fp16, y = hidden_states_31_cast_fp16)[name = tensor("inputs_91_cast_fp16")]; tensor var_3433 = const()[name = tensor("op_3433"), val = tensor(3)]; tensor out_91_axes_0 = const()[name = tensor("out_91_axes_0"), val = tensor([1])]; tensor var_3458_to_fp16 = const()[name = tensor("op_3458_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_91_cast_fp16 = layer_norm(axes = out_91_axes_0, epsilon = var_3458_to_fp16, x = inputs_91_cast_fp16)[name = tensor("out_91_cast_fp16")]; tensor obj_211_gamma_0_to_fp16 = const()[name = tensor("obj_211_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(297769728)))]; tensor obj_211_beta_0_to_fp16 = const()[name = tensor("obj_211_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(297771840)))]; tensor obj_211_epsilon_0_to_fp16 = const()[name = tensor("obj_211_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_211_cast_fp16 = batch_norm(beta = obj_211_beta_0_to_fp16, epsilon = obj_211_epsilon_0_to_fp16, gamma = obj_211_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_91_cast_fp16)[name = tensor("obj_211_cast_fp16")]; tensor query_61_pad_type_0 = const()[name = tensor("query_61_pad_type_0"), val = tensor("valid")]; tensor query_61_strides_0 = const()[name = tensor("query_61_strides_0"), val = tensor([1, 1])]; tensor query_61_pad_0 = const()[name = tensor("query_61_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_61_dilations_0 = const()[name = tensor("query_61_dilations_0"), val = tensor([1, 1])]; tensor query_61_groups_0 = const()[name = tensor("query_61_groups_0"), val = tensor(1)]; tensor layers_15_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(297773952))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(298560448))), name = tensor("layers_15_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_15_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_15_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(298560640)))]; tensor query_61_cast_fp16 = conv(bias = layers_15_self_attn_q_proj_bias_to_fp16, dilations = query_61_dilations_0, groups = query_61_groups_0, pad = query_61_pad_0, pad_type = query_61_pad_type_0, strides = query_61_strides_0, weight = layers_15_self_attn_q_proj_weight_to_fp16_palettized, x = obj_211_cast_fp16)[name = tensor("query_61_cast_fp16")]; tensor current_key_31_pad_type_0 = const()[name = tensor("current_key_31_pad_type_0"), val = tensor("valid")]; tensor current_key_31_strides_0 = const()[name = tensor("current_key_31_strides_0"), val = tensor([1, 1])]; tensor current_key_31_pad_0 = const()[name = tensor("current_key_31_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_31_dilations_0 = const()[name = tensor("current_key_31_dilations_0"), val = tensor([1, 1])]; tensor current_key_31_groups_0 = const()[name = tensor("current_key_31_groups_0"), val = tensor(1)]; tensor layers_15_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(298562752))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(299349248))), name = tensor("layers_15_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor current_key_31_cast_fp16 = conv(dilations = current_key_31_dilations_0, groups = current_key_31_groups_0, pad = current_key_31_pad_0, pad_type = current_key_31_pad_type_0, strides = current_key_31_strides_0, weight = layers_15_self_attn_k_proj_weight_to_fp16_palettized, x = obj_211_cast_fp16)[name = tensor("current_key_31_cast_fp16")]; tensor current_value_31_pad_type_0 = const()[name = tensor("current_value_31_pad_type_0"), val = tensor("valid")]; tensor current_value_31_strides_0 = const()[name = tensor("current_value_31_strides_0"), val = tensor([1, 1])]; tensor current_value_31_pad_0 = const()[name = tensor("current_value_31_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_31_dilations_0 = const()[name = tensor("current_value_31_dilations_0"), val = tensor([1, 1])]; tensor current_value_31_groups_0 = const()[name = tensor("current_value_31_groups_0"), val = tensor(1)]; tensor layers_15_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(299349440))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300135936))), name = tensor("layers_15_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_15_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_15_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300136128)))]; tensor current_value_31_cast_fp16 = conv(bias = layers_15_self_attn_v_proj_bias_to_fp16, dilations = current_value_31_dilations_0, groups = current_value_31_groups_0, pad = current_value_31_pad_0, pad_type = current_value_31_pad_type_0, strides = current_value_31_strides_0, weight = layers_15_self_attn_v_proj_weight_to_fp16_palettized, x = obj_211_cast_fp16)[name = tensor("current_value_31_cast_fp16")]; tensor var_3497_cast_fp16 = mul(x = var_87_cast_fp16_15, y = var_207_cast_fp16)[name = tensor("op_3497_cast_fp16")]; tensor var_3498_cast_fp16 = mul(x = current_key_31_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_3498_cast_fp16")]; tensor key_61_cast_fp16 = add(x = var_3497_cast_fp16, y = var_3498_cast_fp16)[name = tensor("key_61_cast_fp16")]; tensor var_3501_cast_fp16 = mul(x = var_114_cast_fp16_15, y = var_207_cast_fp16)[name = tensor("op_3501_cast_fp16")]; tensor var_3502_cast_fp16 = mul(x = current_value_31_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_3502_cast_fp16")]; tensor value_61_cast_fp16 = add(x = var_3501_cast_fp16, y = var_3502_cast_fp16)[name = tensor("value_61_cast_fp16")]; tensor var_3506 = const()[name = tensor("op_3506"), val = tensor([1, 16, 64, 1])]; tensor mh_q_61_cast_fp16 = reshape(shape = var_3506, x = query_61_cast_fp16)[name = tensor("mh_q_61_cast_fp16")]; tensor var_3508_to_fp16 = const()[name = tensor("op_3508_to_fp16"), val = tensor(0x1p-3)]; tensor var_3509_cast_fp16 = mul(x = mh_q_61_cast_fp16, y = var_3508_to_fp16)[name = tensor("op_3509_cast_fp16")]; tensor var_3512 = const()[name = tensor("op_3512"), val = tensor([1, 16, 64, 448])]; tensor var_3513_cast_fp16 = reshape(shape = var_3512, x = key_61_cast_fp16)[name = tensor("op_3513_cast_fp16")]; tensor mh_w_91_transpose_x_0 = const()[name = tensor("mh_w_91_transpose_x_0"), val = tensor(true)]; tensor mh_w_91_transpose_y_0 = const()[name = tensor("mh_w_91_transpose_y_0"), val = tensor(false)]; tensor mh_w_91_cast_fp16 = matmul(transpose_x = mh_w_91_transpose_x_0, transpose_y = mh_w_91_transpose_y_0, x = var_3509_cast_fp16, y = var_3513_cast_fp16)[name = tensor("mh_w_91_cast_fp16")]; tensor mh_w_93_cast_fp16 = add(x = mh_w_91_cast_fp16, y = var_229_cast_fp16)[name = tensor("mh_w_93_cast_fp16")]; tensor var_3521_cast_fp16 = softmax(axis = var_3433, x = mh_w_93_cast_fp16)[name = tensor("op_3521_cast_fp16")]; tensor var_3522 = const()[name = tensor("op_3522"), val = tensor([1, 16, 64, 448])]; tensor var_3523_cast_fp16 = reshape(shape = var_3522, x = value_61_cast_fp16)[name = tensor("op_3523_cast_fp16")]; tensor attn_61_transpose_x_0 = const()[name = tensor("attn_61_transpose_x_0"), val = tensor(false)]; tensor attn_61_transpose_y_0 = const()[name = tensor("attn_61_transpose_y_0"), val = tensor(true)]; tensor attn_61_cast_fp16 = matmul(transpose_x = attn_61_transpose_x_0, transpose_y = attn_61_transpose_y_0, x = var_3523_cast_fp16, y = var_3521_cast_fp16)[name = tensor("attn_61_cast_fp16")]; tensor var_3526 = const()[name = tensor("op_3526"), val = tensor([1, 1024, 1, 1])]; tensor input_151_cast_fp16 = reshape(shape = var_3526, x = attn_61_cast_fp16)[name = tensor("input_151_cast_fp16")]; tensor obj_217_pad_type_0 = const()[name = tensor("obj_217_pad_type_0"), val = tensor("valid")]; tensor obj_217_strides_0 = const()[name = tensor("obj_217_strides_0"), val = tensor([1, 1])]; tensor obj_217_pad_0 = const()[name = tensor("obj_217_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_217_dilations_0 = const()[name = tensor("obj_217_dilations_0"), val = tensor([1, 1])]; tensor obj_217_groups_0 = const()[name = tensor("obj_217_groups_0"), val = tensor(1)]; tensor layers_15_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300138240))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300924736))), name = tensor("layers_15_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_15_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_15_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300924928)))]; tensor obj_217_cast_fp16 = conv(bias = layers_15_self_attn_o_proj_bias_to_fp16, dilations = obj_217_dilations_0, groups = obj_217_groups_0, pad = obj_217_pad_0, pad_type = obj_217_pad_type_0, strides = obj_217_strides_0, weight = layers_15_self_attn_o_proj_weight_to_fp16_palettized, x = input_151_cast_fp16)[name = tensor("obj_217_cast_fp16")]; tensor inputs_93_cast_fp16 = add(x = inputs_91_cast_fp16, y = obj_217_cast_fp16)[name = tensor("inputs_93_cast_fp16")]; tensor out_93_axes_0 = const()[name = tensor("out_93_axes_0"), val = tensor([1])]; tensor var_3548_to_fp16 = const()[name = tensor("op_3548_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_93_cast_fp16 = layer_norm(axes = out_93_axes_0, epsilon = var_3548_to_fp16, x = inputs_93_cast_fp16)[name = tensor("out_93_cast_fp16")]; tensor obj_219_gamma_0_to_fp16 = const()[name = tensor("obj_219_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300927040)))]; tensor obj_219_beta_0_to_fp16 = const()[name = tensor("obj_219_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300929152)))]; tensor obj_219_epsilon_0_to_fp16 = const()[name = tensor("obj_219_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_219_cast_fp16 = batch_norm(beta = obj_219_beta_0_to_fp16, epsilon = obj_219_epsilon_0_to_fp16, gamma = obj_219_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_93_cast_fp16)[name = tensor("obj_219_cast_fp16")]; tensor query_63_pad_type_0 = const()[name = tensor("query_63_pad_type_0"), val = tensor("valid")]; tensor query_63_strides_0 = const()[name = tensor("query_63_strides_0"), val = tensor([1, 1])]; tensor query_63_pad_0 = const()[name = tensor("query_63_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_63_dilations_0 = const()[name = tensor("query_63_dilations_0"), val = tensor([1, 1])]; tensor query_63_groups_0 = const()[name = tensor("query_63_groups_0"), val = tensor(1)]; tensor layers_15_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300931264))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(301717760))), name = tensor("layers_15_encoder_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_15_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_15_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(301717952)))]; tensor query_63_cast_fp16 = conv(bias = layers_15_encoder_attn_q_proj_bias_to_fp16, dilations = query_63_dilations_0, groups = query_63_groups_0, pad = query_63_pad_0, pad_type = query_63_pad_type_0, strides = query_63_strides_0, weight = layers_15_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_219_cast_fp16)[name = tensor("query_63_cast_fp16")]; tensor key_63_pad_type_0 = const()[name = tensor("key_63_pad_type_0"), val = tensor("valid")]; tensor key_63_strides_0 = const()[name = tensor("key_63_strides_0"), val = tensor([1, 1])]; tensor key_63_pad_0 = const()[name = tensor("key_63_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_63_dilations_0 = const()[name = tensor("key_63_dilations_0"), val = tensor([1, 1])]; tensor key_63_groups_0 = const()[name = tensor("key_63_groups_0"), val = tensor(1)]; tensor layers_15_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(301720064))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(302506560))), name = tensor("layers_15_encoder_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor key_63_cast_fp16 = conv(dilations = key_63_dilations_0, groups = key_63_groups_0, pad = key_63_pad_0, pad_type = key_63_pad_type_0, strides = key_63_strides_0, weight = layers_15_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("key_63_cast_fp16")]; tensor value_63_pad_type_0 = const()[name = tensor("value_63_pad_type_0"), val = tensor("valid")]; tensor value_63_strides_0 = const()[name = tensor("value_63_strides_0"), val = tensor([1, 1])]; tensor value_63_pad_0 = const()[name = tensor("value_63_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_63_dilations_0 = const()[name = tensor("value_63_dilations_0"), val = tensor([1, 1])]; tensor value_63_groups_0 = const()[name = tensor("value_63_groups_0"), val = tensor(1)]; tensor layers_15_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(302506752))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303293248))), name = tensor("layers_15_encoder_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_15_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_15_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303293440)))]; tensor value_63_cast_fp16 = conv(bias = layers_15_encoder_attn_v_proj_bias_to_fp16, dilations = value_63_dilations_0, groups = value_63_groups_0, pad = value_63_pad_0, pad_type = value_63_pad_type_0, strides = value_63_strides_0, weight = layers_15_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("value_63_cast_fp16")]; tensor var_3584 = const()[name = tensor("op_3584"), val = tensor([1, 16, 64, 1])]; tensor mh_q_63_cast_fp16 = reshape(shape = var_3584, x = query_63_cast_fp16)[name = tensor("mh_q_63_cast_fp16")]; tensor var_3586_to_fp16 = const()[name = tensor("op_3586_to_fp16"), val = tensor(0x1p-3)]; tensor var_3587_cast_fp16 = mul(x = mh_q_63_cast_fp16, y = var_3586_to_fp16)[name = tensor("op_3587_cast_fp16")]; tensor var_3590 = const()[name = tensor("op_3590"), val = tensor([1, 16, 64, 1500])]; tensor var_3591_cast_fp16 = reshape(shape = var_3590, x = key_63_cast_fp16)[name = tensor("op_3591_cast_fp16")]; tensor mh_w_95_transpose_x_0 = const()[name = tensor("mh_w_95_transpose_x_0"), val = tensor(true)]; tensor mh_w_95_transpose_y_0 = const()[name = tensor("mh_w_95_transpose_y_0"), val = tensor(false)]; tensor mh_w_95_cast_fp16 = matmul(transpose_x = mh_w_95_transpose_x_0, transpose_y = mh_w_95_transpose_y_0, x = var_3587_cast_fp16, y = var_3591_cast_fp16)[name = tensor("mh_w_95_cast_fp16")]; tensor obj_223_cast_fp16 = softmax(axis = var_3433, x = mh_w_95_cast_fp16)[name = tensor("obj_223_cast_fp16")]; tensor var_3595 = const()[name = tensor("op_3595"), val = tensor([1, 16, 64, 1500])]; tensor var_3596_cast_fp16 = reshape(shape = var_3595, x = value_63_cast_fp16)[name = tensor("op_3596_cast_fp16")]; tensor attn_63_transpose_x_0 = const()[name = tensor("attn_63_transpose_x_0"), val = tensor(false)]; tensor attn_63_transpose_y_0 = const()[name = tensor("attn_63_transpose_y_0"), val = tensor(true)]; tensor attn_63_cast_fp16 = matmul(transpose_x = attn_63_transpose_x_0, transpose_y = attn_63_transpose_y_0, x = var_3596_cast_fp16, y = obj_223_cast_fp16)[name = tensor("attn_63_cast_fp16")]; tensor var_3599 = const()[name = tensor("op_3599"), val = tensor([1, 1024, 1, 1])]; tensor input_153_cast_fp16 = reshape(shape = var_3599, x = attn_63_cast_fp16)[name = tensor("input_153_cast_fp16")]; tensor obj_221_pad_type_0 = const()[name = tensor("obj_221_pad_type_0"), val = tensor("valid")]; tensor obj_221_strides_0 = const()[name = tensor("obj_221_strides_0"), val = tensor([1, 1])]; tensor obj_221_pad_0 = const()[name = tensor("obj_221_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_221_dilations_0 = const()[name = tensor("obj_221_dilations_0"), val = tensor([1, 1])]; tensor obj_221_groups_0 = const()[name = tensor("obj_221_groups_0"), val = tensor(1)]; tensor layers_15_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303295552))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(304082048))), name = tensor("layers_15_encoder_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_15_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_15_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(304082240)))]; tensor obj_221_cast_fp16 = conv(bias = layers_15_encoder_attn_o_proj_bias_to_fp16, dilations = obj_221_dilations_0, groups = obj_221_groups_0, pad = obj_221_pad_0, pad_type = obj_221_pad_type_0, strides = obj_221_strides_0, weight = layers_15_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_153_cast_fp16)[name = tensor("obj_221_cast_fp16")]; tensor inputs_95_cast_fp16 = add(x = inputs_93_cast_fp16, y = obj_221_cast_fp16)[name = tensor("inputs_95_cast_fp16")]; tensor out_95_axes_0 = const()[name = tensor("out_95_axes_0"), val = tensor([1])]; tensor var_3620_to_fp16 = const()[name = tensor("op_3620_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_95_cast_fp16 = layer_norm(axes = out_95_axes_0, epsilon = var_3620_to_fp16, x = inputs_95_cast_fp16)[name = tensor("out_95_cast_fp16")]; tensor input_155_gamma_0_to_fp16 = const()[name = tensor("input_155_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(304084352)))]; tensor input_155_beta_0_to_fp16 = const()[name = tensor("input_155_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(304086464)))]; tensor input_155_epsilon_0_to_fp16 = const()[name = tensor("input_155_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_155_cast_fp16 = batch_norm(beta = input_155_beta_0_to_fp16, epsilon = input_155_epsilon_0_to_fp16, gamma = input_155_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_95_cast_fp16)[name = tensor("input_155_cast_fp16")]; tensor input_157_pad_type_0 = const()[name = tensor("input_157_pad_type_0"), val = tensor("valid")]; tensor input_157_strides_0 = const()[name = tensor("input_157_strides_0"), val = tensor([1, 1])]; tensor input_157_pad_0 = const()[name = tensor("input_157_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_157_dilations_0 = const()[name = tensor("input_157_dilations_0"), val = tensor([1, 1])]; tensor input_157_groups_0 = const()[name = tensor("input_157_groups_0"), val = tensor(1)]; tensor layers_15_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(304088576))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(307234368))), name = tensor("layers_15_fc1_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; tensor layers_15_fc1_bias_to_fp16 = const()[name = tensor("layers_15_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(307234560)))]; tensor input_157_cast_fp16 = conv(bias = layers_15_fc1_bias_to_fp16, dilations = input_157_dilations_0, groups = input_157_groups_0, pad = input_157_pad_0, pad_type = input_157_pad_type_0, strides = input_157_strides_0, weight = layers_15_fc1_weight_to_fp16_palettized, x = input_155_cast_fp16)[name = tensor("input_157_cast_fp16")]; tensor input_159_mode_0 = const()[name = tensor("input_159_mode_0"), val = tensor("EXACT")]; tensor input_159_cast_fp16 = gelu(mode = input_159_mode_0, x = input_157_cast_fp16)[name = tensor("input_159_cast_fp16")]; tensor hidden_states_33_pad_type_0 = const()[name = tensor("hidden_states_33_pad_type_0"), val = tensor("valid")]; tensor hidden_states_33_strides_0 = const()[name = tensor("hidden_states_33_strides_0"), val = tensor([1, 1])]; tensor hidden_states_33_pad_0 = const()[name = tensor("hidden_states_33_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_33_dilations_0 = const()[name = tensor("hidden_states_33_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_33_groups_0 = const()[name = tensor("hidden_states_33_groups_0"), val = tensor(1)]; tensor layers_15_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(307242816))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(310388608))), name = tensor("layers_15_fc2_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; tensor layers_15_fc2_bias_to_fp16 = const()[name = tensor("layers_15_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(310388800)))]; tensor hidden_states_33_cast_fp16 = conv(bias = layers_15_fc2_bias_to_fp16, dilations = hidden_states_33_dilations_0, groups = hidden_states_33_groups_0, pad = hidden_states_33_pad_0, pad_type = hidden_states_33_pad_type_0, strides = hidden_states_33_strides_0, weight = layers_15_fc2_weight_to_fp16_palettized, x = input_159_cast_fp16)[name = tensor("hidden_states_33_cast_fp16")]; tensor inputs_97_cast_fp16 = add(x = inputs_95_cast_fp16, y = hidden_states_33_cast_fp16)[name = tensor("inputs_97_cast_fp16")]; tensor var_3656 = const()[name = tensor("op_3656"), val = tensor(3)]; tensor out_97_axes_0 = const()[name = tensor("out_97_axes_0"), val = tensor([1])]; tensor var_3681_to_fp16 = const()[name = tensor("op_3681_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_97_cast_fp16 = layer_norm(axes = out_97_axes_0, epsilon = var_3681_to_fp16, x = inputs_97_cast_fp16)[name = tensor("out_97_cast_fp16")]; tensor obj_225_gamma_0_to_fp16 = const()[name = tensor("obj_225_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(310390912)))]; tensor obj_225_beta_0_to_fp16 = const()[name = tensor("obj_225_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(310393024)))]; tensor obj_225_epsilon_0_to_fp16 = const()[name = tensor("obj_225_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_225_cast_fp16 = batch_norm(beta = obj_225_beta_0_to_fp16, epsilon = obj_225_epsilon_0_to_fp16, gamma = obj_225_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_97_cast_fp16)[name = tensor("obj_225_cast_fp16")]; tensor query_65_pad_type_0 = const()[name = tensor("query_65_pad_type_0"), val = tensor("valid")]; tensor query_65_strides_0 = const()[name = tensor("query_65_strides_0"), val = tensor([1, 1])]; tensor query_65_pad_0 = const()[name = tensor("query_65_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_65_dilations_0 = const()[name = tensor("query_65_dilations_0"), val = tensor([1, 1])]; tensor query_65_groups_0 = const()[name = tensor("query_65_groups_0"), val = tensor(1)]; tensor layers_16_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(310395136))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(311181632))), name = tensor("layers_16_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_16_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_16_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(311181824)))]; tensor query_65_cast_fp16 = conv(bias = layers_16_self_attn_q_proj_bias_to_fp16, dilations = query_65_dilations_0, groups = query_65_groups_0, pad = query_65_pad_0, pad_type = query_65_pad_type_0, strides = query_65_strides_0, weight = layers_16_self_attn_q_proj_weight_to_fp16_palettized, x = obj_225_cast_fp16)[name = tensor("query_65_cast_fp16")]; tensor current_key_33_pad_type_0 = const()[name = tensor("current_key_33_pad_type_0"), val = tensor("valid")]; tensor current_key_33_strides_0 = const()[name = tensor("current_key_33_strides_0"), val = tensor([1, 1])]; tensor current_key_33_pad_0 = const()[name = tensor("current_key_33_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_33_dilations_0 = const()[name = tensor("current_key_33_dilations_0"), val = tensor([1, 1])]; tensor current_key_33_groups_0 = const()[name = tensor("current_key_33_groups_0"), val = tensor(1)]; tensor layers_16_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(311183936))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(311970432))), name = tensor("layers_16_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor current_key_33_cast_fp16 = conv(dilations = current_key_33_dilations_0, groups = current_key_33_groups_0, pad = current_key_33_pad_0, pad_type = current_key_33_pad_type_0, strides = current_key_33_strides_0, weight = layers_16_self_attn_k_proj_weight_to_fp16_palettized, x = obj_225_cast_fp16)[name = tensor("current_key_33_cast_fp16")]; tensor current_value_33_pad_type_0 = const()[name = tensor("current_value_33_pad_type_0"), val = tensor("valid")]; tensor current_value_33_strides_0 = const()[name = tensor("current_value_33_strides_0"), val = tensor([1, 1])]; tensor current_value_33_pad_0 = const()[name = tensor("current_value_33_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_33_dilations_0 = const()[name = tensor("current_value_33_dilations_0"), val = tensor([1, 1])]; tensor current_value_33_groups_0 = const()[name = tensor("current_value_33_groups_0"), val = tensor(1)]; tensor layers_16_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(311970624))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(312757120))), name = tensor("layers_16_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_16_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_16_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(312757312)))]; tensor current_value_33_cast_fp16 = conv(bias = layers_16_self_attn_v_proj_bias_to_fp16, dilations = current_value_33_dilations_0, groups = current_value_33_groups_0, pad = current_value_33_pad_0, pad_type = current_value_33_pad_type_0, strides = current_value_33_strides_0, weight = layers_16_self_attn_v_proj_weight_to_fp16_palettized, x = obj_225_cast_fp16)[name = tensor("current_value_33_cast_fp16")]; tensor var_3720_cast_fp16 = mul(x = var_87_cast_fp16_16, y = var_207_cast_fp16)[name = tensor("op_3720_cast_fp16")]; tensor var_3721_cast_fp16 = mul(x = current_key_33_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_3721_cast_fp16")]; tensor key_65_cast_fp16 = add(x = var_3720_cast_fp16, y = var_3721_cast_fp16)[name = tensor("key_65_cast_fp16")]; tensor var_3724_cast_fp16 = mul(x = var_114_cast_fp16_16, y = var_207_cast_fp16)[name = tensor("op_3724_cast_fp16")]; tensor var_3725_cast_fp16 = mul(x = current_value_33_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_3725_cast_fp16")]; tensor value_65_cast_fp16 = add(x = var_3724_cast_fp16, y = var_3725_cast_fp16)[name = tensor("value_65_cast_fp16")]; tensor var_3729 = const()[name = tensor("op_3729"), val = tensor([1, 16, 64, 1])]; tensor mh_q_65_cast_fp16 = reshape(shape = var_3729, x = query_65_cast_fp16)[name = tensor("mh_q_65_cast_fp16")]; tensor var_3731_to_fp16 = const()[name = tensor("op_3731_to_fp16"), val = tensor(0x1p-3)]; tensor var_3732_cast_fp16 = mul(x = mh_q_65_cast_fp16, y = var_3731_to_fp16)[name = tensor("op_3732_cast_fp16")]; tensor var_3735 = const()[name = tensor("op_3735"), val = tensor([1, 16, 64, 448])]; tensor var_3736_cast_fp16 = reshape(shape = var_3735, x = key_65_cast_fp16)[name = tensor("op_3736_cast_fp16")]; tensor mh_w_97_transpose_x_0 = const()[name = tensor("mh_w_97_transpose_x_0"), val = tensor(true)]; tensor mh_w_97_transpose_y_0 = const()[name = tensor("mh_w_97_transpose_y_0"), val = tensor(false)]; tensor mh_w_97_cast_fp16 = matmul(transpose_x = mh_w_97_transpose_x_0, transpose_y = mh_w_97_transpose_y_0, x = var_3732_cast_fp16, y = var_3736_cast_fp16)[name = tensor("mh_w_97_cast_fp16")]; tensor mh_w_99_cast_fp16 = add(x = mh_w_97_cast_fp16, y = var_229_cast_fp16)[name = tensor("mh_w_99_cast_fp16")]; tensor var_3744_cast_fp16 = softmax(axis = var_3656, x = mh_w_99_cast_fp16)[name = tensor("op_3744_cast_fp16")]; tensor var_3745 = const()[name = tensor("op_3745"), val = tensor([1, 16, 64, 448])]; tensor var_3746_cast_fp16 = reshape(shape = var_3745, x = value_65_cast_fp16)[name = tensor("op_3746_cast_fp16")]; tensor attn_65_transpose_x_0 = const()[name = tensor("attn_65_transpose_x_0"), val = tensor(false)]; tensor attn_65_transpose_y_0 = const()[name = tensor("attn_65_transpose_y_0"), val = tensor(true)]; tensor attn_65_cast_fp16 = matmul(transpose_x = attn_65_transpose_x_0, transpose_y = attn_65_transpose_y_0, x = var_3746_cast_fp16, y = var_3744_cast_fp16)[name = tensor("attn_65_cast_fp16")]; tensor var_3749 = const()[name = tensor("op_3749"), val = tensor([1, 1024, 1, 1])]; tensor input_161_cast_fp16 = reshape(shape = var_3749, x = attn_65_cast_fp16)[name = tensor("input_161_cast_fp16")]; tensor obj_231_pad_type_0 = const()[name = tensor("obj_231_pad_type_0"), val = tensor("valid")]; tensor obj_231_strides_0 = const()[name = tensor("obj_231_strides_0"), val = tensor([1, 1])]; tensor obj_231_pad_0 = const()[name = tensor("obj_231_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_231_dilations_0 = const()[name = tensor("obj_231_dilations_0"), val = tensor([1, 1])]; tensor obj_231_groups_0 = const()[name = tensor("obj_231_groups_0"), val = tensor(1)]; tensor layers_16_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(312759424))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(313545920))), name = tensor("layers_16_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_16_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_16_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(313546112)))]; tensor obj_231_cast_fp16 = conv(bias = layers_16_self_attn_o_proj_bias_to_fp16, dilations = obj_231_dilations_0, groups = obj_231_groups_0, pad = obj_231_pad_0, pad_type = obj_231_pad_type_0, strides = obj_231_strides_0, weight = layers_16_self_attn_o_proj_weight_to_fp16_palettized, x = input_161_cast_fp16)[name = tensor("obj_231_cast_fp16")]; tensor inputs_99_cast_fp16 = add(x = inputs_97_cast_fp16, y = obj_231_cast_fp16)[name = tensor("inputs_99_cast_fp16")]; tensor out_99_axes_0 = const()[name = tensor("out_99_axes_0"), val = tensor([1])]; tensor var_3771_to_fp16 = const()[name = tensor("op_3771_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_99_cast_fp16 = layer_norm(axes = out_99_axes_0, epsilon = var_3771_to_fp16, x = inputs_99_cast_fp16)[name = tensor("out_99_cast_fp16")]; tensor obj_233_gamma_0_to_fp16 = const()[name = tensor("obj_233_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(313548224)))]; tensor obj_233_beta_0_to_fp16 = const()[name = tensor("obj_233_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(313550336)))]; tensor obj_233_epsilon_0_to_fp16 = const()[name = tensor("obj_233_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_233_cast_fp16 = batch_norm(beta = obj_233_beta_0_to_fp16, epsilon = obj_233_epsilon_0_to_fp16, gamma = obj_233_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_99_cast_fp16)[name = tensor("obj_233_cast_fp16")]; tensor query_67_pad_type_0 = const()[name = tensor("query_67_pad_type_0"), val = tensor("valid")]; tensor query_67_strides_0 = const()[name = tensor("query_67_strides_0"), val = tensor([1, 1])]; tensor query_67_pad_0 = const()[name = tensor("query_67_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_67_dilations_0 = const()[name = tensor("query_67_dilations_0"), val = tensor([1, 1])]; tensor query_67_groups_0 = const()[name = tensor("query_67_groups_0"), val = tensor(1)]; tensor layers_16_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(313552448))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314338944))), name = tensor("layers_16_encoder_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_16_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_16_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314339136)))]; tensor query_67_cast_fp16 = conv(bias = layers_16_encoder_attn_q_proj_bias_to_fp16, dilations = query_67_dilations_0, groups = query_67_groups_0, pad = query_67_pad_0, pad_type = query_67_pad_type_0, strides = query_67_strides_0, weight = layers_16_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_233_cast_fp16)[name = tensor("query_67_cast_fp16")]; tensor key_67_pad_type_0 = const()[name = tensor("key_67_pad_type_0"), val = tensor("valid")]; tensor key_67_strides_0 = const()[name = tensor("key_67_strides_0"), val = tensor([1, 1])]; tensor key_67_pad_0 = const()[name = tensor("key_67_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_67_dilations_0 = const()[name = tensor("key_67_dilations_0"), val = tensor([1, 1])]; tensor key_67_groups_0 = const()[name = tensor("key_67_groups_0"), val = tensor(1)]; tensor layers_16_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314341248))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(315127744))), name = tensor("layers_16_encoder_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor key_67_cast_fp16 = conv(dilations = key_67_dilations_0, groups = key_67_groups_0, pad = key_67_pad_0, pad_type = key_67_pad_type_0, strides = key_67_strides_0, weight = layers_16_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("key_67_cast_fp16")]; tensor value_67_pad_type_0 = const()[name = tensor("value_67_pad_type_0"), val = tensor("valid")]; tensor value_67_strides_0 = const()[name = tensor("value_67_strides_0"), val = tensor([1, 1])]; tensor value_67_pad_0 = const()[name = tensor("value_67_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_67_dilations_0 = const()[name = tensor("value_67_dilations_0"), val = tensor([1, 1])]; tensor value_67_groups_0 = const()[name = tensor("value_67_groups_0"), val = tensor(1)]; tensor layers_16_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(315127936))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(315914432))), name = tensor("layers_16_encoder_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_16_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_16_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(315914624)))]; tensor value_67_cast_fp16 = conv(bias = layers_16_encoder_attn_v_proj_bias_to_fp16, dilations = value_67_dilations_0, groups = value_67_groups_0, pad = value_67_pad_0, pad_type = value_67_pad_type_0, strides = value_67_strides_0, weight = layers_16_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("value_67_cast_fp16")]; tensor var_3807 = const()[name = tensor("op_3807"), val = tensor([1, 16, 64, 1])]; tensor mh_q_67_cast_fp16 = reshape(shape = var_3807, x = query_67_cast_fp16)[name = tensor("mh_q_67_cast_fp16")]; tensor var_3809_to_fp16 = const()[name = tensor("op_3809_to_fp16"), val = tensor(0x1p-3)]; tensor var_3810_cast_fp16 = mul(x = mh_q_67_cast_fp16, y = var_3809_to_fp16)[name = tensor("op_3810_cast_fp16")]; tensor var_3813 = const()[name = tensor("op_3813"), val = tensor([1, 16, 64, 1500])]; tensor var_3814_cast_fp16 = reshape(shape = var_3813, x = key_67_cast_fp16)[name = tensor("op_3814_cast_fp16")]; tensor mh_w_101_transpose_x_0 = const()[name = tensor("mh_w_101_transpose_x_0"), val = tensor(true)]; tensor mh_w_101_transpose_y_0 = const()[name = tensor("mh_w_101_transpose_y_0"), val = tensor(false)]; tensor mh_w_101_cast_fp16 = matmul(transpose_x = mh_w_101_transpose_x_0, transpose_y = mh_w_101_transpose_y_0, x = var_3810_cast_fp16, y = var_3814_cast_fp16)[name = tensor("mh_w_101_cast_fp16")]; tensor obj_237_cast_fp16 = softmax(axis = var_3656, x = mh_w_101_cast_fp16)[name = tensor("obj_237_cast_fp16")]; tensor var_3818 = const()[name = tensor("op_3818"), val = tensor([1, 16, 64, 1500])]; tensor var_3819_cast_fp16 = reshape(shape = var_3818, x = value_67_cast_fp16)[name = tensor("op_3819_cast_fp16")]; tensor attn_67_transpose_x_0 = const()[name = tensor("attn_67_transpose_x_0"), val = tensor(false)]; tensor attn_67_transpose_y_0 = const()[name = tensor("attn_67_transpose_y_0"), val = tensor(true)]; tensor attn_67_cast_fp16 = matmul(transpose_x = attn_67_transpose_x_0, transpose_y = attn_67_transpose_y_0, x = var_3819_cast_fp16, y = obj_237_cast_fp16)[name = tensor("attn_67_cast_fp16")]; tensor var_3822 = const()[name = tensor("op_3822"), val = tensor([1, 1024, 1, 1])]; tensor input_163_cast_fp16 = reshape(shape = var_3822, x = attn_67_cast_fp16)[name = tensor("input_163_cast_fp16")]; tensor obj_235_pad_type_0 = const()[name = tensor("obj_235_pad_type_0"), val = tensor("valid")]; tensor obj_235_strides_0 = const()[name = tensor("obj_235_strides_0"), val = tensor([1, 1])]; tensor obj_235_pad_0 = const()[name = tensor("obj_235_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_235_dilations_0 = const()[name = tensor("obj_235_dilations_0"), val = tensor([1, 1])]; tensor obj_235_groups_0 = const()[name = tensor("obj_235_groups_0"), val = tensor(1)]; tensor layers_16_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(315916736))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316703232))), name = tensor("layers_16_encoder_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_16_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_16_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316703424)))]; tensor obj_235_cast_fp16 = conv(bias = layers_16_encoder_attn_o_proj_bias_to_fp16, dilations = obj_235_dilations_0, groups = obj_235_groups_0, pad = obj_235_pad_0, pad_type = obj_235_pad_type_0, strides = obj_235_strides_0, weight = layers_16_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_163_cast_fp16)[name = tensor("obj_235_cast_fp16")]; tensor inputs_101_cast_fp16 = add(x = inputs_99_cast_fp16, y = obj_235_cast_fp16)[name = tensor("inputs_101_cast_fp16")]; tensor out_101_axes_0 = const()[name = tensor("out_101_axes_0"), val = tensor([1])]; tensor var_3843_to_fp16 = const()[name = tensor("op_3843_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_101_cast_fp16 = layer_norm(axes = out_101_axes_0, epsilon = var_3843_to_fp16, x = inputs_101_cast_fp16)[name = tensor("out_101_cast_fp16")]; tensor input_165_gamma_0_to_fp16 = const()[name = tensor("input_165_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316705536)))]; tensor input_165_beta_0_to_fp16 = const()[name = tensor("input_165_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316707648)))]; tensor input_165_epsilon_0_to_fp16 = const()[name = tensor("input_165_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_165_cast_fp16 = batch_norm(beta = input_165_beta_0_to_fp16, epsilon = input_165_epsilon_0_to_fp16, gamma = input_165_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_101_cast_fp16)[name = tensor("input_165_cast_fp16")]; tensor input_167_pad_type_0 = const()[name = tensor("input_167_pad_type_0"), val = tensor("valid")]; tensor input_167_strides_0 = const()[name = tensor("input_167_strides_0"), val = tensor([1, 1])]; tensor input_167_pad_0 = const()[name = tensor("input_167_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_167_dilations_0 = const()[name = tensor("input_167_dilations_0"), val = tensor([1, 1])]; tensor input_167_groups_0 = const()[name = tensor("input_167_groups_0"), val = tensor(1)]; tensor layers_16_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316709760))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(319855552))), name = tensor("layers_16_fc1_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; tensor layers_16_fc1_bias_to_fp16 = const()[name = tensor("layers_16_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(319855744)))]; tensor input_167_cast_fp16 = conv(bias = layers_16_fc1_bias_to_fp16, dilations = input_167_dilations_0, groups = input_167_groups_0, pad = input_167_pad_0, pad_type = input_167_pad_type_0, strides = input_167_strides_0, weight = layers_16_fc1_weight_to_fp16_palettized, x = input_165_cast_fp16)[name = tensor("input_167_cast_fp16")]; tensor input_169_mode_0 = const()[name = tensor("input_169_mode_0"), val = tensor("EXACT")]; tensor input_169_cast_fp16 = gelu(mode = input_169_mode_0, x = input_167_cast_fp16)[name = tensor("input_169_cast_fp16")]; tensor hidden_states_35_pad_type_0 = const()[name = tensor("hidden_states_35_pad_type_0"), val = tensor("valid")]; tensor hidden_states_35_strides_0 = const()[name = tensor("hidden_states_35_strides_0"), val = tensor([1, 1])]; tensor hidden_states_35_pad_0 = const()[name = tensor("hidden_states_35_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_35_dilations_0 = const()[name = tensor("hidden_states_35_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_35_groups_0 = const()[name = tensor("hidden_states_35_groups_0"), val = tensor(1)]; tensor layers_16_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(319864000))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(323009792))), name = tensor("layers_16_fc2_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; tensor layers_16_fc2_bias_to_fp16 = const()[name = tensor("layers_16_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(323009984)))]; tensor hidden_states_35_cast_fp16 = conv(bias = layers_16_fc2_bias_to_fp16, dilations = hidden_states_35_dilations_0, groups = hidden_states_35_groups_0, pad = hidden_states_35_pad_0, pad_type = hidden_states_35_pad_type_0, strides = hidden_states_35_strides_0, weight = layers_16_fc2_weight_to_fp16_palettized, x = input_169_cast_fp16)[name = tensor("hidden_states_35_cast_fp16")]; tensor inputs_103_cast_fp16 = add(x = inputs_101_cast_fp16, y = hidden_states_35_cast_fp16)[name = tensor("inputs_103_cast_fp16")]; tensor var_3879 = const()[name = tensor("op_3879"), val = tensor(3)]; tensor out_103_axes_0 = const()[name = tensor("out_103_axes_0"), val = tensor([1])]; tensor var_3904_to_fp16 = const()[name = tensor("op_3904_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_103_cast_fp16 = layer_norm(axes = out_103_axes_0, epsilon = var_3904_to_fp16, x = inputs_103_cast_fp16)[name = tensor("out_103_cast_fp16")]; tensor obj_239_gamma_0_to_fp16 = const()[name = tensor("obj_239_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(323012096)))]; tensor obj_239_beta_0_to_fp16 = const()[name = tensor("obj_239_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(323014208)))]; tensor obj_239_epsilon_0_to_fp16 = const()[name = tensor("obj_239_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_239_cast_fp16 = batch_norm(beta = obj_239_beta_0_to_fp16, epsilon = obj_239_epsilon_0_to_fp16, gamma = obj_239_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_103_cast_fp16)[name = tensor("obj_239_cast_fp16")]; tensor query_69_pad_type_0 = const()[name = tensor("query_69_pad_type_0"), val = tensor("valid")]; tensor query_69_strides_0 = const()[name = tensor("query_69_strides_0"), val = tensor([1, 1])]; tensor query_69_pad_0 = const()[name = tensor("query_69_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_69_dilations_0 = const()[name = tensor("query_69_dilations_0"), val = tensor([1, 1])]; tensor query_69_groups_0 = const()[name = tensor("query_69_groups_0"), val = tensor(1)]; tensor layers_17_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(323016320))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(323802816))), name = tensor("layers_17_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_17_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_17_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(323803008)))]; tensor query_69_cast_fp16 = conv(bias = layers_17_self_attn_q_proj_bias_to_fp16, dilations = query_69_dilations_0, groups = query_69_groups_0, pad = query_69_pad_0, pad_type = query_69_pad_type_0, strides = query_69_strides_0, weight = layers_17_self_attn_q_proj_weight_to_fp16_palettized, x = obj_239_cast_fp16)[name = tensor("query_69_cast_fp16")]; tensor current_key_35_pad_type_0 = const()[name = tensor("current_key_35_pad_type_0"), val = tensor("valid")]; tensor current_key_35_strides_0 = const()[name = tensor("current_key_35_strides_0"), val = tensor([1, 1])]; tensor current_key_35_pad_0 = const()[name = tensor("current_key_35_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_35_dilations_0 = const()[name = tensor("current_key_35_dilations_0"), val = tensor([1, 1])]; tensor current_key_35_groups_0 = const()[name = tensor("current_key_35_groups_0"), val = tensor(1)]; tensor layers_17_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(323805120))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(324591616))), name = tensor("layers_17_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor current_key_35_cast_fp16 = conv(dilations = current_key_35_dilations_0, groups = current_key_35_groups_0, pad = current_key_35_pad_0, pad_type = current_key_35_pad_type_0, strides = current_key_35_strides_0, weight = layers_17_self_attn_k_proj_weight_to_fp16_palettized, x = obj_239_cast_fp16)[name = tensor("current_key_35_cast_fp16")]; tensor current_value_35_pad_type_0 = const()[name = tensor("current_value_35_pad_type_0"), val = tensor("valid")]; tensor current_value_35_strides_0 = const()[name = tensor("current_value_35_strides_0"), val = tensor([1, 1])]; tensor current_value_35_pad_0 = const()[name = tensor("current_value_35_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_35_dilations_0 = const()[name = tensor("current_value_35_dilations_0"), val = tensor([1, 1])]; tensor current_value_35_groups_0 = const()[name = tensor("current_value_35_groups_0"), val = tensor(1)]; tensor layers_17_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(324591808))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(325378304))), name = tensor("layers_17_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_17_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_17_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(325378496)))]; tensor current_value_35_cast_fp16 = conv(bias = layers_17_self_attn_v_proj_bias_to_fp16, dilations = current_value_35_dilations_0, groups = current_value_35_groups_0, pad = current_value_35_pad_0, pad_type = current_value_35_pad_type_0, strides = current_value_35_strides_0, weight = layers_17_self_attn_v_proj_weight_to_fp16_palettized, x = obj_239_cast_fp16)[name = tensor("current_value_35_cast_fp16")]; tensor var_3943_cast_fp16 = mul(x = var_87_cast_fp16_17, y = var_207_cast_fp16)[name = tensor("op_3943_cast_fp16")]; tensor var_3944_cast_fp16 = mul(x = current_key_35_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_3944_cast_fp16")]; tensor key_69_cast_fp16 = add(x = var_3943_cast_fp16, y = var_3944_cast_fp16)[name = tensor("key_69_cast_fp16")]; tensor var_3947_cast_fp16 = mul(x = var_114_cast_fp16_17, y = var_207_cast_fp16)[name = tensor("op_3947_cast_fp16")]; tensor var_3948_cast_fp16 = mul(x = current_value_35_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_3948_cast_fp16")]; tensor value_69_cast_fp16 = add(x = var_3947_cast_fp16, y = var_3948_cast_fp16)[name = tensor("value_69_cast_fp16")]; tensor var_3952 = const()[name = tensor("op_3952"), val = tensor([1, 16, 64, 1])]; tensor mh_q_69_cast_fp16 = reshape(shape = var_3952, x = query_69_cast_fp16)[name = tensor("mh_q_69_cast_fp16")]; tensor var_3954_to_fp16 = const()[name = tensor("op_3954_to_fp16"), val = tensor(0x1p-3)]; tensor var_3955_cast_fp16 = mul(x = mh_q_69_cast_fp16, y = var_3954_to_fp16)[name = tensor("op_3955_cast_fp16")]; tensor var_3958 = const()[name = tensor("op_3958"), val = tensor([1, 16, 64, 448])]; tensor var_3959_cast_fp16 = reshape(shape = var_3958, x = key_69_cast_fp16)[name = tensor("op_3959_cast_fp16")]; tensor mh_w_103_transpose_x_0 = const()[name = tensor("mh_w_103_transpose_x_0"), val = tensor(true)]; tensor mh_w_103_transpose_y_0 = const()[name = tensor("mh_w_103_transpose_y_0"), val = tensor(false)]; tensor mh_w_103_cast_fp16 = matmul(transpose_x = mh_w_103_transpose_x_0, transpose_y = mh_w_103_transpose_y_0, x = var_3955_cast_fp16, y = var_3959_cast_fp16)[name = tensor("mh_w_103_cast_fp16")]; tensor mh_w_105_cast_fp16 = add(x = mh_w_103_cast_fp16, y = var_229_cast_fp16)[name = tensor("mh_w_105_cast_fp16")]; tensor var_3967_cast_fp16 = softmax(axis = var_3879, x = mh_w_105_cast_fp16)[name = tensor("op_3967_cast_fp16")]; tensor var_3968 = const()[name = tensor("op_3968"), val = tensor([1, 16, 64, 448])]; tensor var_3969_cast_fp16 = reshape(shape = var_3968, x = value_69_cast_fp16)[name = tensor("op_3969_cast_fp16")]; tensor attn_69_transpose_x_0 = const()[name = tensor("attn_69_transpose_x_0"), val = tensor(false)]; tensor attn_69_transpose_y_0 = const()[name = tensor("attn_69_transpose_y_0"), val = tensor(true)]; tensor attn_69_cast_fp16 = matmul(transpose_x = attn_69_transpose_x_0, transpose_y = attn_69_transpose_y_0, x = var_3969_cast_fp16, y = var_3967_cast_fp16)[name = tensor("attn_69_cast_fp16")]; tensor var_3972 = const()[name = tensor("op_3972"), val = tensor([1, 1024, 1, 1])]; tensor input_171_cast_fp16 = reshape(shape = var_3972, x = attn_69_cast_fp16)[name = tensor("input_171_cast_fp16")]; tensor obj_245_pad_type_0 = const()[name = tensor("obj_245_pad_type_0"), val = tensor("valid")]; tensor obj_245_strides_0 = const()[name = tensor("obj_245_strides_0"), val = tensor([1, 1])]; tensor obj_245_pad_0 = const()[name = tensor("obj_245_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_245_dilations_0 = const()[name = tensor("obj_245_dilations_0"), val = tensor([1, 1])]; tensor obj_245_groups_0 = const()[name = tensor("obj_245_groups_0"), val = tensor(1)]; tensor layers_17_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(325380608))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(326167104))), name = tensor("layers_17_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_17_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_17_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(326167296)))]; tensor obj_245_cast_fp16 = conv(bias = layers_17_self_attn_o_proj_bias_to_fp16, dilations = obj_245_dilations_0, groups = obj_245_groups_0, pad = obj_245_pad_0, pad_type = obj_245_pad_type_0, strides = obj_245_strides_0, weight = layers_17_self_attn_o_proj_weight_to_fp16_palettized, x = input_171_cast_fp16)[name = tensor("obj_245_cast_fp16")]; tensor inputs_105_cast_fp16 = add(x = inputs_103_cast_fp16, y = obj_245_cast_fp16)[name = tensor("inputs_105_cast_fp16")]; tensor out_105_axes_0 = const()[name = tensor("out_105_axes_0"), val = tensor([1])]; tensor var_3994_to_fp16 = const()[name = tensor("op_3994_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_105_cast_fp16 = layer_norm(axes = out_105_axes_0, epsilon = var_3994_to_fp16, x = inputs_105_cast_fp16)[name = tensor("out_105_cast_fp16")]; tensor obj_247_gamma_0_to_fp16 = const()[name = tensor("obj_247_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(326169408)))]; tensor obj_247_beta_0_to_fp16 = const()[name = tensor("obj_247_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(326171520)))]; tensor obj_247_epsilon_0_to_fp16 = const()[name = tensor("obj_247_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_247_cast_fp16 = batch_norm(beta = obj_247_beta_0_to_fp16, epsilon = obj_247_epsilon_0_to_fp16, gamma = obj_247_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_105_cast_fp16)[name = tensor("obj_247_cast_fp16")]; tensor query_71_pad_type_0 = const()[name = tensor("query_71_pad_type_0"), val = tensor("valid")]; tensor query_71_strides_0 = const()[name = tensor("query_71_strides_0"), val = tensor([1, 1])]; tensor query_71_pad_0 = const()[name = tensor("query_71_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_71_dilations_0 = const()[name = tensor("query_71_dilations_0"), val = tensor([1, 1])]; tensor query_71_groups_0 = const()[name = tensor("query_71_groups_0"), val = tensor(1)]; tensor layers_17_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(326173632))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(326960128))), name = tensor("layers_17_encoder_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_17_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_17_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(326960320)))]; tensor query_71_cast_fp16 = conv(bias = layers_17_encoder_attn_q_proj_bias_to_fp16, dilations = query_71_dilations_0, groups = query_71_groups_0, pad = query_71_pad_0, pad_type = query_71_pad_type_0, strides = query_71_strides_0, weight = layers_17_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_247_cast_fp16)[name = tensor("query_71_cast_fp16")]; tensor key_71_pad_type_0 = const()[name = tensor("key_71_pad_type_0"), val = tensor("valid")]; tensor key_71_strides_0 = const()[name = tensor("key_71_strides_0"), val = tensor([1, 1])]; tensor key_71_pad_0 = const()[name = tensor("key_71_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_71_dilations_0 = const()[name = tensor("key_71_dilations_0"), val = tensor([1, 1])]; tensor key_71_groups_0 = const()[name = tensor("key_71_groups_0"), val = tensor(1)]; tensor layers_17_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(326962432))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(327748928))), name = tensor("layers_17_encoder_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor key_71_cast_fp16 = conv(dilations = key_71_dilations_0, groups = key_71_groups_0, pad = key_71_pad_0, pad_type = key_71_pad_type_0, strides = key_71_strides_0, weight = layers_17_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("key_71_cast_fp16")]; tensor value_71_pad_type_0 = const()[name = tensor("value_71_pad_type_0"), val = tensor("valid")]; tensor value_71_strides_0 = const()[name = tensor("value_71_strides_0"), val = tensor([1, 1])]; tensor value_71_pad_0 = const()[name = tensor("value_71_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_71_dilations_0 = const()[name = tensor("value_71_dilations_0"), val = tensor([1, 1])]; tensor value_71_groups_0 = const()[name = tensor("value_71_groups_0"), val = tensor(1)]; tensor layers_17_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(327749120))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(328535616))), name = tensor("layers_17_encoder_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_17_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_17_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(328535808)))]; tensor value_71_cast_fp16 = conv(bias = layers_17_encoder_attn_v_proj_bias_to_fp16, dilations = value_71_dilations_0, groups = value_71_groups_0, pad = value_71_pad_0, pad_type = value_71_pad_type_0, strides = value_71_strides_0, weight = layers_17_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("value_71_cast_fp16")]; tensor var_4030 = const()[name = tensor("op_4030"), val = tensor([1, 16, 64, 1])]; tensor mh_q_71_cast_fp16 = reshape(shape = var_4030, x = query_71_cast_fp16)[name = tensor("mh_q_71_cast_fp16")]; tensor var_4032_to_fp16 = const()[name = tensor("op_4032_to_fp16"), val = tensor(0x1p-3)]; tensor var_4033_cast_fp16 = mul(x = mh_q_71_cast_fp16, y = var_4032_to_fp16)[name = tensor("op_4033_cast_fp16")]; tensor var_4036 = const()[name = tensor("op_4036"), val = tensor([1, 16, 64, 1500])]; tensor var_4037_cast_fp16 = reshape(shape = var_4036, x = key_71_cast_fp16)[name = tensor("op_4037_cast_fp16")]; tensor mh_w_107_transpose_x_0 = const()[name = tensor("mh_w_107_transpose_x_0"), val = tensor(true)]; tensor mh_w_107_transpose_y_0 = const()[name = tensor("mh_w_107_transpose_y_0"), val = tensor(false)]; tensor mh_w_107_cast_fp16 = matmul(transpose_x = mh_w_107_transpose_x_0, transpose_y = mh_w_107_transpose_y_0, x = var_4033_cast_fp16, y = var_4037_cast_fp16)[name = tensor("mh_w_107_cast_fp16")]; tensor obj_251_cast_fp16 = softmax(axis = var_3879, x = mh_w_107_cast_fp16)[name = tensor("obj_251_cast_fp16")]; tensor var_4041 = const()[name = tensor("op_4041"), val = tensor([1, 16, 64, 1500])]; tensor var_4042_cast_fp16 = reshape(shape = var_4041, x = value_71_cast_fp16)[name = tensor("op_4042_cast_fp16")]; tensor attn_71_transpose_x_0 = const()[name = tensor("attn_71_transpose_x_0"), val = tensor(false)]; tensor attn_71_transpose_y_0 = const()[name = tensor("attn_71_transpose_y_0"), val = tensor(true)]; tensor attn_71_cast_fp16 = matmul(transpose_x = attn_71_transpose_x_0, transpose_y = attn_71_transpose_y_0, x = var_4042_cast_fp16, y = obj_251_cast_fp16)[name = tensor("attn_71_cast_fp16")]; tensor var_4045 = const()[name = tensor("op_4045"), val = tensor([1, 1024, 1, 1])]; tensor input_173_cast_fp16 = reshape(shape = var_4045, x = attn_71_cast_fp16)[name = tensor("input_173_cast_fp16")]; tensor obj_249_pad_type_0 = const()[name = tensor("obj_249_pad_type_0"), val = tensor("valid")]; tensor obj_249_strides_0 = const()[name = tensor("obj_249_strides_0"), val = tensor([1, 1])]; tensor obj_249_pad_0 = const()[name = tensor("obj_249_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_249_dilations_0 = const()[name = tensor("obj_249_dilations_0"), val = tensor([1, 1])]; tensor obj_249_groups_0 = const()[name = tensor("obj_249_groups_0"), val = tensor(1)]; tensor layers_17_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(328537920))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329324416))), name = tensor("layers_17_encoder_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_17_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_17_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329324608)))]; tensor obj_249_cast_fp16 = conv(bias = layers_17_encoder_attn_o_proj_bias_to_fp16, dilations = obj_249_dilations_0, groups = obj_249_groups_0, pad = obj_249_pad_0, pad_type = obj_249_pad_type_0, strides = obj_249_strides_0, weight = layers_17_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_173_cast_fp16)[name = tensor("obj_249_cast_fp16")]; tensor inputs_107_cast_fp16 = add(x = inputs_105_cast_fp16, y = obj_249_cast_fp16)[name = tensor("inputs_107_cast_fp16")]; tensor out_107_axes_0 = const()[name = tensor("out_107_axes_0"), val = tensor([1])]; tensor var_4063_to_fp16 = const()[name = tensor("op_4063_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_107_cast_fp16 = layer_norm(axes = out_107_axes_0, epsilon = var_4063_to_fp16, x = inputs_107_cast_fp16)[name = tensor("out_107_cast_fp16")]; tensor input_175_gamma_0_to_fp16 = const()[name = tensor("input_175_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329326720)))]; tensor input_175_beta_0_to_fp16 = const()[name = tensor("input_175_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329328832)))]; tensor input_175_epsilon_0_to_fp16 = const()[name = tensor("input_175_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_175_cast_fp16 = batch_norm(beta = input_175_beta_0_to_fp16, epsilon = input_175_epsilon_0_to_fp16, gamma = input_175_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_107_cast_fp16)[name = tensor("input_175_cast_fp16")]; tensor input_177_pad_type_0 = const()[name = tensor("input_177_pad_type_0"), val = tensor("valid")]; tensor input_177_strides_0 = const()[name = tensor("input_177_strides_0"), val = tensor([1, 1])]; tensor input_177_pad_0 = const()[name = tensor("input_177_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_177_dilations_0 = const()[name = tensor("input_177_dilations_0"), val = tensor([1, 1])]; tensor input_177_groups_0 = const()[name = tensor("input_177_groups_0"), val = tensor(1)]; tensor layers_17_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329330944))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(332476736))), name = tensor("layers_17_fc1_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; tensor layers_17_fc1_bias_to_fp16 = const()[name = tensor("layers_17_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(332476928)))]; tensor input_177_cast_fp16 = conv(bias = layers_17_fc1_bias_to_fp16, dilations = input_177_dilations_0, groups = input_177_groups_0, pad = input_177_pad_0, pad_type = input_177_pad_type_0, strides = input_177_strides_0, weight = layers_17_fc1_weight_to_fp16_palettized, x = input_175_cast_fp16)[name = tensor("input_177_cast_fp16")]; tensor input_179_mode_0 = const()[name = tensor("input_179_mode_0"), val = tensor("EXACT")]; tensor input_179_cast_fp16 = gelu(mode = input_179_mode_0, x = input_177_cast_fp16)[name = tensor("input_179_cast_fp16")]; tensor hidden_states_37_pad_type_0 = const()[name = tensor("hidden_states_37_pad_type_0"), val = tensor("valid")]; tensor hidden_states_37_strides_0 = const()[name = tensor("hidden_states_37_strides_0"), val = tensor([1, 1])]; tensor hidden_states_37_pad_0 = const()[name = tensor("hidden_states_37_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_37_dilations_0 = const()[name = tensor("hidden_states_37_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_37_groups_0 = const()[name = tensor("hidden_states_37_groups_0"), val = tensor(1)]; tensor layers_17_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(332485184))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335630976))), name = tensor("layers_17_fc2_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; tensor layers_17_fc2_bias_to_fp16 = const()[name = tensor("layers_17_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335631168)))]; tensor hidden_states_37_cast_fp16 = conv(bias = layers_17_fc2_bias_to_fp16, dilations = hidden_states_37_dilations_0, groups = hidden_states_37_groups_0, pad = hidden_states_37_pad_0, pad_type = hidden_states_37_pad_type_0, strides = hidden_states_37_strides_0, weight = layers_17_fc2_weight_to_fp16_palettized, x = input_179_cast_fp16)[name = tensor("hidden_states_37_cast_fp16")]; tensor inputs_109_cast_fp16 = add(x = inputs_107_cast_fp16, y = hidden_states_37_cast_fp16)[name = tensor("inputs_109_cast_fp16")]; tensor var_4098 = const()[name = tensor("op_4098"), val = tensor(3)]; tensor out_109_axes_0 = const()[name = tensor("out_109_axes_0"), val = tensor([1])]; tensor var_4123_to_fp16 = const()[name = tensor("op_4123_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_109_cast_fp16 = layer_norm(axes = out_109_axes_0, epsilon = var_4123_to_fp16, x = inputs_109_cast_fp16)[name = tensor("out_109_cast_fp16")]; tensor obj_253_gamma_0_to_fp16 = const()[name = tensor("obj_253_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335633280)))]; tensor obj_253_beta_0_to_fp16 = const()[name = tensor("obj_253_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335635392)))]; tensor obj_253_epsilon_0_to_fp16 = const()[name = tensor("obj_253_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_253_cast_fp16 = batch_norm(beta = obj_253_beta_0_to_fp16, epsilon = obj_253_epsilon_0_to_fp16, gamma = obj_253_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_109_cast_fp16)[name = tensor("obj_253_cast_fp16")]; tensor query_73_pad_type_0 = const()[name = tensor("query_73_pad_type_0"), val = tensor("valid")]; tensor query_73_strides_0 = const()[name = tensor("query_73_strides_0"), val = tensor([1, 1])]; tensor query_73_pad_0 = const()[name = tensor("query_73_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_73_dilations_0 = const()[name = tensor("query_73_dilations_0"), val = tensor([1, 1])]; tensor query_73_groups_0 = const()[name = tensor("query_73_groups_0"), val = tensor(1)]; tensor layers_18_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335637504))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(336424000))), name = tensor("layers_18_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_18_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_18_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(336424192)))]; tensor query_73_cast_fp16 = conv(bias = layers_18_self_attn_q_proj_bias_to_fp16, dilations = query_73_dilations_0, groups = query_73_groups_0, pad = query_73_pad_0, pad_type = query_73_pad_type_0, strides = query_73_strides_0, weight = layers_18_self_attn_q_proj_weight_to_fp16_palettized, x = obj_253_cast_fp16)[name = tensor("query_73_cast_fp16")]; tensor current_key_37_pad_type_0 = const()[name = tensor("current_key_37_pad_type_0"), val = tensor("valid")]; tensor current_key_37_strides_0 = const()[name = tensor("current_key_37_strides_0"), val = tensor([1, 1])]; tensor current_key_37_pad_0 = const()[name = tensor("current_key_37_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_37_dilations_0 = const()[name = tensor("current_key_37_dilations_0"), val = tensor([1, 1])]; tensor current_key_37_groups_0 = const()[name = tensor("current_key_37_groups_0"), val = tensor(1)]; tensor layers_18_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(336426304))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337212800))), name = tensor("layers_18_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor current_key_37_cast_fp16 = conv(dilations = current_key_37_dilations_0, groups = current_key_37_groups_0, pad = current_key_37_pad_0, pad_type = current_key_37_pad_type_0, strides = current_key_37_strides_0, weight = layers_18_self_attn_k_proj_weight_to_fp16_palettized, x = obj_253_cast_fp16)[name = tensor("current_key_37_cast_fp16")]; tensor current_value_37_pad_type_0 = const()[name = tensor("current_value_37_pad_type_0"), val = tensor("valid")]; tensor current_value_37_strides_0 = const()[name = tensor("current_value_37_strides_0"), val = tensor([1, 1])]; tensor current_value_37_pad_0 = const()[name = tensor("current_value_37_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_37_dilations_0 = const()[name = tensor("current_value_37_dilations_0"), val = tensor([1, 1])]; tensor current_value_37_groups_0 = const()[name = tensor("current_value_37_groups_0"), val = tensor(1)]; tensor layers_18_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337212992))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337999488))), name = tensor("layers_18_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_18_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_18_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337999680)))]; tensor current_value_37_cast_fp16 = conv(bias = layers_18_self_attn_v_proj_bias_to_fp16, dilations = current_value_37_dilations_0, groups = current_value_37_groups_0, pad = current_value_37_pad_0, pad_type = current_value_37_pad_type_0, strides = current_value_37_strides_0, weight = layers_18_self_attn_v_proj_weight_to_fp16_palettized, x = obj_253_cast_fp16)[name = tensor("current_value_37_cast_fp16")]; tensor var_4162_cast_fp16 = mul(x = var_87_cast_fp16_18, y = var_207_cast_fp16)[name = tensor("op_4162_cast_fp16")]; tensor var_4163_cast_fp16 = mul(x = current_key_37_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_4163_cast_fp16")]; tensor key_73_cast_fp16 = add(x = var_4162_cast_fp16, y = var_4163_cast_fp16)[name = tensor("key_73_cast_fp16")]; tensor var_4166_cast_fp16 = mul(x = var_114_cast_fp16_18, y = var_207_cast_fp16)[name = tensor("op_4166_cast_fp16")]; tensor var_4167_cast_fp16 = mul(x = current_value_37_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_4167_cast_fp16")]; tensor value_73_cast_fp16 = add(x = var_4166_cast_fp16, y = var_4167_cast_fp16)[name = tensor("value_73_cast_fp16")]; tensor var_4171 = const()[name = tensor("op_4171"), val = tensor([1, 16, 64, 1])]; tensor mh_q_73_cast_fp16 = reshape(shape = var_4171, x = query_73_cast_fp16)[name = tensor("mh_q_73_cast_fp16")]; tensor var_4173_to_fp16 = const()[name = tensor("op_4173_to_fp16"), val = tensor(0x1p-3)]; tensor var_4174_cast_fp16 = mul(x = mh_q_73_cast_fp16, y = var_4173_to_fp16)[name = tensor("op_4174_cast_fp16")]; tensor var_4177 = const()[name = tensor("op_4177"), val = tensor([1, 16, 64, 448])]; tensor var_4178_cast_fp16 = reshape(shape = var_4177, x = key_73_cast_fp16)[name = tensor("op_4178_cast_fp16")]; tensor mh_w_109_transpose_x_0 = const()[name = tensor("mh_w_109_transpose_x_0"), val = tensor(true)]; tensor mh_w_109_transpose_y_0 = const()[name = tensor("mh_w_109_transpose_y_0"), val = tensor(false)]; tensor mh_w_109_cast_fp16 = matmul(transpose_x = mh_w_109_transpose_x_0, transpose_y = mh_w_109_transpose_y_0, x = var_4174_cast_fp16, y = var_4178_cast_fp16)[name = tensor("mh_w_109_cast_fp16")]; tensor mh_w_111_cast_fp16 = add(x = mh_w_109_cast_fp16, y = var_229_cast_fp16)[name = tensor("mh_w_111_cast_fp16")]; tensor var_4186_cast_fp16 = softmax(axis = var_4098, x = mh_w_111_cast_fp16)[name = tensor("op_4186_cast_fp16")]; tensor var_4187 = const()[name = tensor("op_4187"), val = tensor([1, 16, 64, 448])]; tensor var_4188_cast_fp16 = reshape(shape = var_4187, x = value_73_cast_fp16)[name = tensor("op_4188_cast_fp16")]; tensor attn_73_transpose_x_0 = const()[name = tensor("attn_73_transpose_x_0"), val = tensor(false)]; tensor attn_73_transpose_y_0 = const()[name = tensor("attn_73_transpose_y_0"), val = tensor(true)]; tensor attn_73_cast_fp16 = matmul(transpose_x = attn_73_transpose_x_0, transpose_y = attn_73_transpose_y_0, x = var_4188_cast_fp16, y = var_4186_cast_fp16)[name = tensor("attn_73_cast_fp16")]; tensor var_4191 = const()[name = tensor("op_4191"), val = tensor([1, 1024, 1, 1])]; tensor input_181_cast_fp16 = reshape(shape = var_4191, x = attn_73_cast_fp16)[name = tensor("input_181_cast_fp16")]; tensor obj_259_pad_type_0 = const()[name = tensor("obj_259_pad_type_0"), val = tensor("valid")]; tensor obj_259_strides_0 = const()[name = tensor("obj_259_strides_0"), val = tensor([1, 1])]; tensor obj_259_pad_0 = const()[name = tensor("obj_259_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_259_dilations_0 = const()[name = tensor("obj_259_dilations_0"), val = tensor([1, 1])]; tensor obj_259_groups_0 = const()[name = tensor("obj_259_groups_0"), val = tensor(1)]; tensor layers_18_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(338001792))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(338788288))), name = tensor("layers_18_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_18_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_18_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(338788480)))]; tensor obj_259_cast_fp16 = conv(bias = layers_18_self_attn_o_proj_bias_to_fp16, dilations = obj_259_dilations_0, groups = obj_259_groups_0, pad = obj_259_pad_0, pad_type = obj_259_pad_type_0, strides = obj_259_strides_0, weight = layers_18_self_attn_o_proj_weight_to_fp16_palettized, x = input_181_cast_fp16)[name = tensor("obj_259_cast_fp16")]; tensor inputs_111_cast_fp16 = add(x = inputs_109_cast_fp16, y = obj_259_cast_fp16)[name = tensor("inputs_111_cast_fp16")]; tensor out_111_axes_0 = const()[name = tensor("out_111_axes_0"), val = tensor([1])]; tensor var_4213_to_fp16 = const()[name = tensor("op_4213_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_111_cast_fp16 = layer_norm(axes = out_111_axes_0, epsilon = var_4213_to_fp16, x = inputs_111_cast_fp16)[name = tensor("out_111_cast_fp16")]; tensor obj_261_gamma_0_to_fp16 = const()[name = tensor("obj_261_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(338790592)))]; tensor obj_261_beta_0_to_fp16 = const()[name = tensor("obj_261_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(338792704)))]; tensor obj_261_epsilon_0_to_fp16 = const()[name = tensor("obj_261_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_261_cast_fp16 = batch_norm(beta = obj_261_beta_0_to_fp16, epsilon = obj_261_epsilon_0_to_fp16, gamma = obj_261_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_111_cast_fp16)[name = tensor("obj_261_cast_fp16")]; tensor query_75_pad_type_0 = const()[name = tensor("query_75_pad_type_0"), val = tensor("valid")]; tensor query_75_strides_0 = const()[name = tensor("query_75_strides_0"), val = tensor([1, 1])]; tensor query_75_pad_0 = const()[name = tensor("query_75_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_75_dilations_0 = const()[name = tensor("query_75_dilations_0"), val = tensor([1, 1])]; tensor query_75_groups_0 = const()[name = tensor("query_75_groups_0"), val = tensor(1)]; tensor layers_18_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(338794816))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(339581312))), name = tensor("layers_18_encoder_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_18_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_18_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(339581504)))]; tensor query_75_cast_fp16 = conv(bias = layers_18_encoder_attn_q_proj_bias_to_fp16, dilations = query_75_dilations_0, groups = query_75_groups_0, pad = query_75_pad_0, pad_type = query_75_pad_type_0, strides = query_75_strides_0, weight = layers_18_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_261_cast_fp16)[name = tensor("query_75_cast_fp16")]; tensor key_75_pad_type_0 = const()[name = tensor("key_75_pad_type_0"), val = tensor("valid")]; tensor key_75_strides_0 = const()[name = tensor("key_75_strides_0"), val = tensor([1, 1])]; tensor key_75_pad_0 = const()[name = tensor("key_75_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_75_dilations_0 = const()[name = tensor("key_75_dilations_0"), val = tensor([1, 1])]; tensor key_75_groups_0 = const()[name = tensor("key_75_groups_0"), val = tensor(1)]; tensor layers_18_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(339583616))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(340370112))), name = tensor("layers_18_encoder_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor key_75_cast_fp16 = conv(dilations = key_75_dilations_0, groups = key_75_groups_0, pad = key_75_pad_0, pad_type = key_75_pad_type_0, strides = key_75_strides_0, weight = layers_18_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("key_75_cast_fp16")]; tensor value_75_pad_type_0 = const()[name = tensor("value_75_pad_type_0"), val = tensor("valid")]; tensor value_75_strides_0 = const()[name = tensor("value_75_strides_0"), val = tensor([1, 1])]; tensor value_75_pad_0 = const()[name = tensor("value_75_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_75_dilations_0 = const()[name = tensor("value_75_dilations_0"), val = tensor([1, 1])]; tensor value_75_groups_0 = const()[name = tensor("value_75_groups_0"), val = tensor(1)]; tensor layers_18_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(340370304))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(341156800))), name = tensor("layers_18_encoder_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_18_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_18_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(341156992)))]; tensor value_75_cast_fp16 = conv(bias = layers_18_encoder_attn_v_proj_bias_to_fp16, dilations = value_75_dilations_0, groups = value_75_groups_0, pad = value_75_pad_0, pad_type = value_75_pad_type_0, strides = value_75_strides_0, weight = layers_18_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("value_75_cast_fp16")]; tensor var_4249 = const()[name = tensor("op_4249"), val = tensor([1, 16, 64, 1])]; tensor mh_q_75_cast_fp16 = reshape(shape = var_4249, x = query_75_cast_fp16)[name = tensor("mh_q_75_cast_fp16")]; tensor var_4251_to_fp16 = const()[name = tensor("op_4251_to_fp16"), val = tensor(0x1p-3)]; tensor var_4252_cast_fp16 = mul(x = mh_q_75_cast_fp16, y = var_4251_to_fp16)[name = tensor("op_4252_cast_fp16")]; tensor var_4255 = const()[name = tensor("op_4255"), val = tensor([1, 16, 64, 1500])]; tensor var_4256_cast_fp16 = reshape(shape = var_4255, x = key_75_cast_fp16)[name = tensor("op_4256_cast_fp16")]; tensor mh_w_113_transpose_x_0 = const()[name = tensor("mh_w_113_transpose_x_0"), val = tensor(true)]; tensor mh_w_113_transpose_y_0 = const()[name = tensor("mh_w_113_transpose_y_0"), val = tensor(false)]; tensor mh_w_113_cast_fp16 = matmul(transpose_x = mh_w_113_transpose_x_0, transpose_y = mh_w_113_transpose_y_0, x = var_4252_cast_fp16, y = var_4256_cast_fp16)[name = tensor("mh_w_113_cast_fp16")]; tensor obj_265_cast_fp16 = softmax(axis = var_4098, x = mh_w_113_cast_fp16)[name = tensor("obj_265_cast_fp16")]; tensor var_4260 = const()[name = tensor("op_4260"), val = tensor([1, 16, 64, 1500])]; tensor var_4261_cast_fp16 = reshape(shape = var_4260, x = value_75_cast_fp16)[name = tensor("op_4261_cast_fp16")]; tensor attn_75_transpose_x_0 = const()[name = tensor("attn_75_transpose_x_0"), val = tensor(false)]; tensor attn_75_transpose_y_0 = const()[name = tensor("attn_75_transpose_y_0"), val = tensor(true)]; tensor attn_75_cast_fp16 = matmul(transpose_x = attn_75_transpose_x_0, transpose_y = attn_75_transpose_y_0, x = var_4261_cast_fp16, y = obj_265_cast_fp16)[name = tensor("attn_75_cast_fp16")]; tensor var_4264 = const()[name = tensor("op_4264"), val = tensor([1, 1024, 1, 1])]; tensor input_183_cast_fp16 = reshape(shape = var_4264, x = attn_75_cast_fp16)[name = tensor("input_183_cast_fp16")]; tensor obj_263_pad_type_0 = const()[name = tensor("obj_263_pad_type_0"), val = tensor("valid")]; tensor obj_263_strides_0 = const()[name = tensor("obj_263_strides_0"), val = tensor([1, 1])]; tensor obj_263_pad_0 = const()[name = tensor("obj_263_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_263_dilations_0 = const()[name = tensor("obj_263_dilations_0"), val = tensor([1, 1])]; tensor obj_263_groups_0 = const()[name = tensor("obj_263_groups_0"), val = tensor(1)]; tensor layers_18_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(341159104))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(341945600))), name = tensor("layers_18_encoder_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_18_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_18_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(341945792)))]; tensor obj_263_cast_fp16 = conv(bias = layers_18_encoder_attn_o_proj_bias_to_fp16, dilations = obj_263_dilations_0, groups = obj_263_groups_0, pad = obj_263_pad_0, pad_type = obj_263_pad_type_0, strides = obj_263_strides_0, weight = layers_18_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_183_cast_fp16)[name = tensor("obj_263_cast_fp16")]; tensor inputs_113_cast_fp16 = add(x = inputs_111_cast_fp16, y = obj_263_cast_fp16)[name = tensor("inputs_113_cast_fp16")]; tensor out_113_axes_0 = const()[name = tensor("out_113_axes_0"), val = tensor([1])]; tensor var_4282_to_fp16 = const()[name = tensor("op_4282_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_113_cast_fp16 = layer_norm(axes = out_113_axes_0, epsilon = var_4282_to_fp16, x = inputs_113_cast_fp16)[name = tensor("out_113_cast_fp16")]; tensor input_185_gamma_0_to_fp16 = const()[name = tensor("input_185_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(341947904)))]; tensor input_185_beta_0_to_fp16 = const()[name = tensor("input_185_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(341950016)))]; tensor input_185_epsilon_0_to_fp16 = const()[name = tensor("input_185_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_185_cast_fp16 = batch_norm(beta = input_185_beta_0_to_fp16, epsilon = input_185_epsilon_0_to_fp16, gamma = input_185_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_113_cast_fp16)[name = tensor("input_185_cast_fp16")]; tensor input_187_pad_type_0 = const()[name = tensor("input_187_pad_type_0"), val = tensor("valid")]; tensor input_187_strides_0 = const()[name = tensor("input_187_strides_0"), val = tensor([1, 1])]; tensor input_187_pad_0 = const()[name = tensor("input_187_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_187_dilations_0 = const()[name = tensor("input_187_dilations_0"), val = tensor([1, 1])]; tensor input_187_groups_0 = const()[name = tensor("input_187_groups_0"), val = tensor(1)]; tensor layers_18_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(341952128))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(345097920))), name = tensor("layers_18_fc1_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; tensor layers_18_fc1_bias_to_fp16 = const()[name = tensor("layers_18_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(345098112)))]; tensor input_187_cast_fp16 = conv(bias = layers_18_fc1_bias_to_fp16, dilations = input_187_dilations_0, groups = input_187_groups_0, pad = input_187_pad_0, pad_type = input_187_pad_type_0, strides = input_187_strides_0, weight = layers_18_fc1_weight_to_fp16_palettized, x = input_185_cast_fp16)[name = tensor("input_187_cast_fp16")]; tensor input_189_mode_0 = const()[name = tensor("input_189_mode_0"), val = tensor("EXACT")]; tensor input_189_cast_fp16 = gelu(mode = input_189_mode_0, x = input_187_cast_fp16)[name = tensor("input_189_cast_fp16")]; tensor hidden_states_39_pad_type_0 = const()[name = tensor("hidden_states_39_pad_type_0"), val = tensor("valid")]; tensor hidden_states_39_strides_0 = const()[name = tensor("hidden_states_39_strides_0"), val = tensor([1, 1])]; tensor hidden_states_39_pad_0 = const()[name = tensor("hidden_states_39_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_39_dilations_0 = const()[name = tensor("hidden_states_39_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_39_groups_0 = const()[name = tensor("hidden_states_39_groups_0"), val = tensor(1)]; tensor layers_18_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(345106368))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(348252160))), name = tensor("layers_18_fc2_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; tensor layers_18_fc2_bias_to_fp16 = const()[name = tensor("layers_18_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(348252352)))]; tensor hidden_states_39_cast_fp16 = conv(bias = layers_18_fc2_bias_to_fp16, dilations = hidden_states_39_dilations_0, groups = hidden_states_39_groups_0, pad = hidden_states_39_pad_0, pad_type = hidden_states_39_pad_type_0, strides = hidden_states_39_strides_0, weight = layers_18_fc2_weight_to_fp16_palettized, x = input_189_cast_fp16)[name = tensor("hidden_states_39_cast_fp16")]; tensor inputs_115_cast_fp16 = add(x = inputs_113_cast_fp16, y = hidden_states_39_cast_fp16)[name = tensor("inputs_115_cast_fp16")]; tensor var_4317 = const()[name = tensor("op_4317"), val = tensor(3)]; tensor out_115_axes_0 = const()[name = tensor("out_115_axes_0"), val = tensor([1])]; tensor var_4342_to_fp16 = const()[name = tensor("op_4342_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_115_cast_fp16 = layer_norm(axes = out_115_axes_0, epsilon = var_4342_to_fp16, x = inputs_115_cast_fp16)[name = tensor("out_115_cast_fp16")]; tensor obj_267_gamma_0_to_fp16 = const()[name = tensor("obj_267_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(348254464)))]; tensor obj_267_beta_0_to_fp16 = const()[name = tensor("obj_267_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(348256576)))]; tensor obj_267_epsilon_0_to_fp16 = const()[name = tensor("obj_267_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_267_cast_fp16 = batch_norm(beta = obj_267_beta_0_to_fp16, epsilon = obj_267_epsilon_0_to_fp16, gamma = obj_267_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_115_cast_fp16)[name = tensor("obj_267_cast_fp16")]; tensor query_77_pad_type_0 = const()[name = tensor("query_77_pad_type_0"), val = tensor("valid")]; tensor query_77_strides_0 = const()[name = tensor("query_77_strides_0"), val = tensor([1, 1])]; tensor query_77_pad_0 = const()[name = tensor("query_77_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_77_dilations_0 = const()[name = tensor("query_77_dilations_0"), val = tensor([1, 1])]; tensor query_77_groups_0 = const()[name = tensor("query_77_groups_0"), val = tensor(1)]; tensor layers_19_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(348258688))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(349045184))), name = tensor("layers_19_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_19_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_19_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(349045376)))]; tensor query_77_cast_fp16 = conv(bias = layers_19_self_attn_q_proj_bias_to_fp16, dilations = query_77_dilations_0, groups = query_77_groups_0, pad = query_77_pad_0, pad_type = query_77_pad_type_0, strides = query_77_strides_0, weight = layers_19_self_attn_q_proj_weight_to_fp16_palettized, x = obj_267_cast_fp16)[name = tensor("query_77_cast_fp16")]; tensor current_key_39_pad_type_0 = const()[name = tensor("current_key_39_pad_type_0"), val = tensor("valid")]; tensor current_key_39_strides_0 = const()[name = tensor("current_key_39_strides_0"), val = tensor([1, 1])]; tensor current_key_39_pad_0 = const()[name = tensor("current_key_39_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_39_dilations_0 = const()[name = tensor("current_key_39_dilations_0"), val = tensor([1, 1])]; tensor current_key_39_groups_0 = const()[name = tensor("current_key_39_groups_0"), val = tensor(1)]; tensor layers_19_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(349047488))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(349833984))), name = tensor("layers_19_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor current_key_39_cast_fp16 = conv(dilations = current_key_39_dilations_0, groups = current_key_39_groups_0, pad = current_key_39_pad_0, pad_type = current_key_39_pad_type_0, strides = current_key_39_strides_0, weight = layers_19_self_attn_k_proj_weight_to_fp16_palettized, x = obj_267_cast_fp16)[name = tensor("current_key_39_cast_fp16")]; tensor current_value_39_pad_type_0 = const()[name = tensor("current_value_39_pad_type_0"), val = tensor("valid")]; tensor current_value_39_strides_0 = const()[name = tensor("current_value_39_strides_0"), val = tensor([1, 1])]; tensor current_value_39_pad_0 = const()[name = tensor("current_value_39_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_39_dilations_0 = const()[name = tensor("current_value_39_dilations_0"), val = tensor([1, 1])]; tensor current_value_39_groups_0 = const()[name = tensor("current_value_39_groups_0"), val = tensor(1)]; tensor layers_19_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(349834176))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(350620672))), name = tensor("layers_19_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_19_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_19_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(350620864)))]; tensor current_value_39_cast_fp16 = conv(bias = layers_19_self_attn_v_proj_bias_to_fp16, dilations = current_value_39_dilations_0, groups = current_value_39_groups_0, pad = current_value_39_pad_0, pad_type = current_value_39_pad_type_0, strides = current_value_39_strides_0, weight = layers_19_self_attn_v_proj_weight_to_fp16_palettized, x = obj_267_cast_fp16)[name = tensor("current_value_39_cast_fp16")]; tensor var_4381_cast_fp16 = mul(x = var_87_cast_fp16_19, y = var_207_cast_fp16)[name = tensor("op_4381_cast_fp16")]; tensor var_4382_cast_fp16 = mul(x = current_key_39_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_4382_cast_fp16")]; tensor key_77_cast_fp16 = add(x = var_4381_cast_fp16, y = var_4382_cast_fp16)[name = tensor("key_77_cast_fp16")]; tensor var_4385_cast_fp16 = mul(x = var_114_cast_fp16_19, y = var_207_cast_fp16)[name = tensor("op_4385_cast_fp16")]; tensor var_4386_cast_fp16 = mul(x = current_value_39_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_4386_cast_fp16")]; tensor value_77_cast_fp16 = add(x = var_4385_cast_fp16, y = var_4386_cast_fp16)[name = tensor("value_77_cast_fp16")]; tensor var_4390 = const()[name = tensor("op_4390"), val = tensor([1, 16, 64, 1])]; tensor mh_q_77_cast_fp16 = reshape(shape = var_4390, x = query_77_cast_fp16)[name = tensor("mh_q_77_cast_fp16")]; tensor var_4392_to_fp16 = const()[name = tensor("op_4392_to_fp16"), val = tensor(0x1p-3)]; tensor var_4393_cast_fp16 = mul(x = mh_q_77_cast_fp16, y = var_4392_to_fp16)[name = tensor("op_4393_cast_fp16")]; tensor var_4396 = const()[name = tensor("op_4396"), val = tensor([1, 16, 64, 448])]; tensor var_4397_cast_fp16 = reshape(shape = var_4396, x = key_77_cast_fp16)[name = tensor("op_4397_cast_fp16")]; tensor mh_w_115_transpose_x_0 = const()[name = tensor("mh_w_115_transpose_x_0"), val = tensor(true)]; tensor mh_w_115_transpose_y_0 = const()[name = tensor("mh_w_115_transpose_y_0"), val = tensor(false)]; tensor mh_w_115_cast_fp16 = matmul(transpose_x = mh_w_115_transpose_x_0, transpose_y = mh_w_115_transpose_y_0, x = var_4393_cast_fp16, y = var_4397_cast_fp16)[name = tensor("mh_w_115_cast_fp16")]; tensor mh_w_117_cast_fp16 = add(x = mh_w_115_cast_fp16, y = var_229_cast_fp16)[name = tensor("mh_w_117_cast_fp16")]; tensor var_4405_cast_fp16 = softmax(axis = var_4317, x = mh_w_117_cast_fp16)[name = tensor("op_4405_cast_fp16")]; tensor var_4406 = const()[name = tensor("op_4406"), val = tensor([1, 16, 64, 448])]; tensor var_4407_cast_fp16 = reshape(shape = var_4406, x = value_77_cast_fp16)[name = tensor("op_4407_cast_fp16")]; tensor attn_77_transpose_x_0 = const()[name = tensor("attn_77_transpose_x_0"), val = tensor(false)]; tensor attn_77_transpose_y_0 = const()[name = tensor("attn_77_transpose_y_0"), val = tensor(true)]; tensor attn_77_cast_fp16 = matmul(transpose_x = attn_77_transpose_x_0, transpose_y = attn_77_transpose_y_0, x = var_4407_cast_fp16, y = var_4405_cast_fp16)[name = tensor("attn_77_cast_fp16")]; tensor var_4410 = const()[name = tensor("op_4410"), val = tensor([1, 1024, 1, 1])]; tensor input_191_cast_fp16 = reshape(shape = var_4410, x = attn_77_cast_fp16)[name = tensor("input_191_cast_fp16")]; tensor obj_273_pad_type_0 = const()[name = tensor("obj_273_pad_type_0"), val = tensor("valid")]; tensor obj_273_strides_0 = const()[name = tensor("obj_273_strides_0"), val = tensor([1, 1])]; tensor obj_273_pad_0 = const()[name = tensor("obj_273_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_273_dilations_0 = const()[name = tensor("obj_273_dilations_0"), val = tensor([1, 1])]; tensor obj_273_groups_0 = const()[name = tensor("obj_273_groups_0"), val = tensor(1)]; tensor layers_19_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(350622976))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(351409472))), name = tensor("layers_19_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_19_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_19_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(351409664)))]; tensor obj_273_cast_fp16 = conv(bias = layers_19_self_attn_o_proj_bias_to_fp16, dilations = obj_273_dilations_0, groups = obj_273_groups_0, pad = obj_273_pad_0, pad_type = obj_273_pad_type_0, strides = obj_273_strides_0, weight = layers_19_self_attn_o_proj_weight_to_fp16_palettized, x = input_191_cast_fp16)[name = tensor("obj_273_cast_fp16")]; tensor inputs_117_cast_fp16 = add(x = inputs_115_cast_fp16, y = obj_273_cast_fp16)[name = tensor("inputs_117_cast_fp16")]; tensor out_117_axes_0 = const()[name = tensor("out_117_axes_0"), val = tensor([1])]; tensor var_4432_to_fp16 = const()[name = tensor("op_4432_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_117_cast_fp16 = layer_norm(axes = out_117_axes_0, epsilon = var_4432_to_fp16, x = inputs_117_cast_fp16)[name = tensor("out_117_cast_fp16")]; tensor obj_275_gamma_0_to_fp16 = const()[name = tensor("obj_275_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(351411776)))]; tensor obj_275_beta_0_to_fp16 = const()[name = tensor("obj_275_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(351413888)))]; tensor obj_275_epsilon_0_to_fp16 = const()[name = tensor("obj_275_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_275_cast_fp16 = batch_norm(beta = obj_275_beta_0_to_fp16, epsilon = obj_275_epsilon_0_to_fp16, gamma = obj_275_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_117_cast_fp16)[name = tensor("obj_275_cast_fp16")]; tensor query_79_pad_type_0 = const()[name = tensor("query_79_pad_type_0"), val = tensor("valid")]; tensor query_79_strides_0 = const()[name = tensor("query_79_strides_0"), val = tensor([1, 1])]; tensor query_79_pad_0 = const()[name = tensor("query_79_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_79_dilations_0 = const()[name = tensor("query_79_dilations_0"), val = tensor([1, 1])]; tensor query_79_groups_0 = const()[name = tensor("query_79_groups_0"), val = tensor(1)]; tensor layers_19_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(351416000))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(352202496))), name = tensor("layers_19_encoder_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_19_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_19_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(352202688)))]; tensor query_79_cast_fp16 = conv(bias = layers_19_encoder_attn_q_proj_bias_to_fp16, dilations = query_79_dilations_0, groups = query_79_groups_0, pad = query_79_pad_0, pad_type = query_79_pad_type_0, strides = query_79_strides_0, weight = layers_19_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_275_cast_fp16)[name = tensor("query_79_cast_fp16")]; tensor key_79_pad_type_0 = const()[name = tensor("key_79_pad_type_0"), val = tensor("valid")]; tensor key_79_strides_0 = const()[name = tensor("key_79_strides_0"), val = tensor([1, 1])]; tensor key_79_pad_0 = const()[name = tensor("key_79_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_79_dilations_0 = const()[name = tensor("key_79_dilations_0"), val = tensor([1, 1])]; tensor key_79_groups_0 = const()[name = tensor("key_79_groups_0"), val = tensor(1)]; tensor layers_19_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(352204800))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(352991296))), name = tensor("layers_19_encoder_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor key_79_cast_fp16 = conv(dilations = key_79_dilations_0, groups = key_79_groups_0, pad = key_79_pad_0, pad_type = key_79_pad_type_0, strides = key_79_strides_0, weight = layers_19_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("key_79_cast_fp16")]; tensor value_79_pad_type_0 = const()[name = tensor("value_79_pad_type_0"), val = tensor("valid")]; tensor value_79_strides_0 = const()[name = tensor("value_79_strides_0"), val = tensor([1, 1])]; tensor value_79_pad_0 = const()[name = tensor("value_79_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_79_dilations_0 = const()[name = tensor("value_79_dilations_0"), val = tensor([1, 1])]; tensor value_79_groups_0 = const()[name = tensor("value_79_groups_0"), val = tensor(1)]; tensor layers_19_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(352991488))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(353777984))), name = tensor("layers_19_encoder_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_19_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_19_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(353778176)))]; tensor value_79_cast_fp16 = conv(bias = layers_19_encoder_attn_v_proj_bias_to_fp16, dilations = value_79_dilations_0, groups = value_79_groups_0, pad = value_79_pad_0, pad_type = value_79_pad_type_0, strides = value_79_strides_0, weight = layers_19_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("value_79_cast_fp16")]; tensor var_4468 = const()[name = tensor("op_4468"), val = tensor([1, 16, 64, 1])]; tensor mh_q_79_cast_fp16 = reshape(shape = var_4468, x = query_79_cast_fp16)[name = tensor("mh_q_79_cast_fp16")]; tensor var_4470_to_fp16 = const()[name = tensor("op_4470_to_fp16"), val = tensor(0x1p-3)]; tensor var_4471_cast_fp16 = mul(x = mh_q_79_cast_fp16, y = var_4470_to_fp16)[name = tensor("op_4471_cast_fp16")]; tensor var_4474 = const()[name = tensor("op_4474"), val = tensor([1, 16, 64, 1500])]; tensor var_4475_cast_fp16 = reshape(shape = var_4474, x = key_79_cast_fp16)[name = tensor("op_4475_cast_fp16")]; tensor mh_w_119_transpose_x_0 = const()[name = tensor("mh_w_119_transpose_x_0"), val = tensor(true)]; tensor mh_w_119_transpose_y_0 = const()[name = tensor("mh_w_119_transpose_y_0"), val = tensor(false)]; tensor mh_w_119_cast_fp16 = matmul(transpose_x = mh_w_119_transpose_x_0, transpose_y = mh_w_119_transpose_y_0, x = var_4471_cast_fp16, y = var_4475_cast_fp16)[name = tensor("mh_w_119_cast_fp16")]; tensor obj_279_cast_fp16 = softmax(axis = var_4317, x = mh_w_119_cast_fp16)[name = tensor("obj_279_cast_fp16")]; tensor var_4479 = const()[name = tensor("op_4479"), val = tensor([1, 16, 64, 1500])]; tensor var_4480_cast_fp16 = reshape(shape = var_4479, x = value_79_cast_fp16)[name = tensor("op_4480_cast_fp16")]; tensor attn_79_transpose_x_0 = const()[name = tensor("attn_79_transpose_x_0"), val = tensor(false)]; tensor attn_79_transpose_y_0 = const()[name = tensor("attn_79_transpose_y_0"), val = tensor(true)]; tensor attn_79_cast_fp16 = matmul(transpose_x = attn_79_transpose_x_0, transpose_y = attn_79_transpose_y_0, x = var_4480_cast_fp16, y = obj_279_cast_fp16)[name = tensor("attn_79_cast_fp16")]; tensor var_4483 = const()[name = tensor("op_4483"), val = tensor([1, 1024, 1, 1])]; tensor input_193_cast_fp16 = reshape(shape = var_4483, x = attn_79_cast_fp16)[name = tensor("input_193_cast_fp16")]; tensor obj_277_pad_type_0 = const()[name = tensor("obj_277_pad_type_0"), val = tensor("valid")]; tensor obj_277_strides_0 = const()[name = tensor("obj_277_strides_0"), val = tensor([1, 1])]; tensor obj_277_pad_0 = const()[name = tensor("obj_277_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_277_dilations_0 = const()[name = tensor("obj_277_dilations_0"), val = tensor([1, 1])]; tensor obj_277_groups_0 = const()[name = tensor("obj_277_groups_0"), val = tensor(1)]; tensor layers_19_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(353780288))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(354566784))), name = tensor("layers_19_encoder_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_19_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_19_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(354566976)))]; tensor obj_277_cast_fp16 = conv(bias = layers_19_encoder_attn_o_proj_bias_to_fp16, dilations = obj_277_dilations_0, groups = obj_277_groups_0, pad = obj_277_pad_0, pad_type = obj_277_pad_type_0, strides = obj_277_strides_0, weight = layers_19_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_193_cast_fp16)[name = tensor("obj_277_cast_fp16")]; tensor inputs_119_cast_fp16 = add(x = inputs_117_cast_fp16, y = obj_277_cast_fp16)[name = tensor("inputs_119_cast_fp16")]; tensor out_119_axes_0 = const()[name = tensor("out_119_axes_0"), val = tensor([1])]; tensor var_4501_to_fp16 = const()[name = tensor("op_4501_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_119_cast_fp16 = layer_norm(axes = out_119_axes_0, epsilon = var_4501_to_fp16, x = inputs_119_cast_fp16)[name = tensor("out_119_cast_fp16")]; tensor input_195_gamma_0_to_fp16 = const()[name = tensor("input_195_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(354569088)))]; tensor input_195_beta_0_to_fp16 = const()[name = tensor("input_195_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(354571200)))]; tensor input_195_epsilon_0_to_fp16 = const()[name = tensor("input_195_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_195_cast_fp16 = batch_norm(beta = input_195_beta_0_to_fp16, epsilon = input_195_epsilon_0_to_fp16, gamma = input_195_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_119_cast_fp16)[name = tensor("input_195_cast_fp16")]; tensor input_197_pad_type_0 = const()[name = tensor("input_197_pad_type_0"), val = tensor("valid")]; tensor input_197_strides_0 = const()[name = tensor("input_197_strides_0"), val = tensor([1, 1])]; tensor input_197_pad_0 = const()[name = tensor("input_197_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_197_dilations_0 = const()[name = tensor("input_197_dilations_0"), val = tensor([1, 1])]; tensor input_197_groups_0 = const()[name = tensor("input_197_groups_0"), val = tensor(1)]; tensor layers_19_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(354573312))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(357719104))), name = tensor("layers_19_fc1_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; tensor layers_19_fc1_bias_to_fp16 = const()[name = tensor("layers_19_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(357719296)))]; tensor input_197_cast_fp16 = conv(bias = layers_19_fc1_bias_to_fp16, dilations = input_197_dilations_0, groups = input_197_groups_0, pad = input_197_pad_0, pad_type = input_197_pad_type_0, strides = input_197_strides_0, weight = layers_19_fc1_weight_to_fp16_palettized, x = input_195_cast_fp16)[name = tensor("input_197_cast_fp16")]; tensor input_199_mode_0 = const()[name = tensor("input_199_mode_0"), val = tensor("EXACT")]; tensor input_199_cast_fp16 = gelu(mode = input_199_mode_0, x = input_197_cast_fp16)[name = tensor("input_199_cast_fp16")]; tensor hidden_states_41_pad_type_0 = const()[name = tensor("hidden_states_41_pad_type_0"), val = tensor("valid")]; tensor hidden_states_41_strides_0 = const()[name = tensor("hidden_states_41_strides_0"), val = tensor([1, 1])]; tensor hidden_states_41_pad_0 = const()[name = tensor("hidden_states_41_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_41_dilations_0 = const()[name = tensor("hidden_states_41_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_41_groups_0 = const()[name = tensor("hidden_states_41_groups_0"), val = tensor(1)]; tensor layers_19_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(357727552))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(360873344))), name = tensor("layers_19_fc2_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; tensor layers_19_fc2_bias_to_fp16 = const()[name = tensor("layers_19_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(360873536)))]; tensor hidden_states_41_cast_fp16 = conv(bias = layers_19_fc2_bias_to_fp16, dilations = hidden_states_41_dilations_0, groups = hidden_states_41_groups_0, pad = hidden_states_41_pad_0, pad_type = hidden_states_41_pad_type_0, strides = hidden_states_41_strides_0, weight = layers_19_fc2_weight_to_fp16_palettized, x = input_199_cast_fp16)[name = tensor("hidden_states_41_cast_fp16")]; tensor inputs_121_cast_fp16 = add(x = inputs_119_cast_fp16, y = hidden_states_41_cast_fp16)[name = tensor("inputs_121_cast_fp16")]; tensor var_4536 = const()[name = tensor("op_4536"), val = tensor(3)]; tensor out_121_axes_0 = const()[name = tensor("out_121_axes_0"), val = tensor([1])]; tensor var_4561_to_fp16 = const()[name = tensor("op_4561_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_121_cast_fp16 = layer_norm(axes = out_121_axes_0, epsilon = var_4561_to_fp16, x = inputs_121_cast_fp16)[name = tensor("out_121_cast_fp16")]; tensor obj_281_gamma_0_to_fp16 = const()[name = tensor("obj_281_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(360875648)))]; tensor obj_281_beta_0_to_fp16 = const()[name = tensor("obj_281_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(360877760)))]; tensor obj_281_epsilon_0_to_fp16 = const()[name = tensor("obj_281_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_281_cast_fp16 = batch_norm(beta = obj_281_beta_0_to_fp16, epsilon = obj_281_epsilon_0_to_fp16, gamma = obj_281_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_121_cast_fp16)[name = tensor("obj_281_cast_fp16")]; tensor query_81_pad_type_0 = const()[name = tensor("query_81_pad_type_0"), val = tensor("valid")]; tensor query_81_strides_0 = const()[name = tensor("query_81_strides_0"), val = tensor([1, 1])]; tensor query_81_pad_0 = const()[name = tensor("query_81_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_81_dilations_0 = const()[name = tensor("query_81_dilations_0"), val = tensor([1, 1])]; tensor query_81_groups_0 = const()[name = tensor("query_81_groups_0"), val = tensor(1)]; tensor layers_20_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(360879872))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(361666368))), name = tensor("layers_20_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_20_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_20_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(361666560)))]; tensor query_81_cast_fp16 = conv(bias = layers_20_self_attn_q_proj_bias_to_fp16, dilations = query_81_dilations_0, groups = query_81_groups_0, pad = query_81_pad_0, pad_type = query_81_pad_type_0, strides = query_81_strides_0, weight = layers_20_self_attn_q_proj_weight_to_fp16_palettized, x = obj_281_cast_fp16)[name = tensor("query_81_cast_fp16")]; tensor current_key_41_pad_type_0 = const()[name = tensor("current_key_41_pad_type_0"), val = tensor("valid")]; tensor current_key_41_strides_0 = const()[name = tensor("current_key_41_strides_0"), val = tensor([1, 1])]; tensor current_key_41_pad_0 = const()[name = tensor("current_key_41_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_41_dilations_0 = const()[name = tensor("current_key_41_dilations_0"), val = tensor([1, 1])]; tensor current_key_41_groups_0 = const()[name = tensor("current_key_41_groups_0"), val = tensor(1)]; tensor layers_20_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(361668672))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(362455168))), name = tensor("layers_20_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor current_key_41_cast_fp16 = conv(dilations = current_key_41_dilations_0, groups = current_key_41_groups_0, pad = current_key_41_pad_0, pad_type = current_key_41_pad_type_0, strides = current_key_41_strides_0, weight = layers_20_self_attn_k_proj_weight_to_fp16_palettized, x = obj_281_cast_fp16)[name = tensor("current_key_41_cast_fp16")]; tensor current_value_41_pad_type_0 = const()[name = tensor("current_value_41_pad_type_0"), val = tensor("valid")]; tensor current_value_41_strides_0 = const()[name = tensor("current_value_41_strides_0"), val = tensor([1, 1])]; tensor current_value_41_pad_0 = const()[name = tensor("current_value_41_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_41_dilations_0 = const()[name = tensor("current_value_41_dilations_0"), val = tensor([1, 1])]; tensor current_value_41_groups_0 = const()[name = tensor("current_value_41_groups_0"), val = tensor(1)]; tensor layers_20_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(362455360))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(363241856))), name = tensor("layers_20_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_20_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_20_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(363242048)))]; tensor current_value_41_cast_fp16 = conv(bias = layers_20_self_attn_v_proj_bias_to_fp16, dilations = current_value_41_dilations_0, groups = current_value_41_groups_0, pad = current_value_41_pad_0, pad_type = current_value_41_pad_type_0, strides = current_value_41_strides_0, weight = layers_20_self_attn_v_proj_weight_to_fp16_palettized, x = obj_281_cast_fp16)[name = tensor("current_value_41_cast_fp16")]; tensor var_4600_cast_fp16 = mul(x = var_87_cast_fp16_20, y = var_207_cast_fp16)[name = tensor("op_4600_cast_fp16")]; tensor var_4601_cast_fp16 = mul(x = current_key_41_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_4601_cast_fp16")]; tensor key_81_cast_fp16 = add(x = var_4600_cast_fp16, y = var_4601_cast_fp16)[name = tensor("key_81_cast_fp16")]; tensor var_4604_cast_fp16 = mul(x = var_114_cast_fp16_20, y = var_207_cast_fp16)[name = tensor("op_4604_cast_fp16")]; tensor var_4605_cast_fp16 = mul(x = current_value_41_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_4605_cast_fp16")]; tensor value_81_cast_fp16 = add(x = var_4604_cast_fp16, y = var_4605_cast_fp16)[name = tensor("value_81_cast_fp16")]; tensor var_4609 = const()[name = tensor("op_4609"), val = tensor([1, 16, 64, 1])]; tensor mh_q_81_cast_fp16 = reshape(shape = var_4609, x = query_81_cast_fp16)[name = tensor("mh_q_81_cast_fp16")]; tensor var_4611_to_fp16 = const()[name = tensor("op_4611_to_fp16"), val = tensor(0x1p-3)]; tensor var_4612_cast_fp16 = mul(x = mh_q_81_cast_fp16, y = var_4611_to_fp16)[name = tensor("op_4612_cast_fp16")]; tensor var_4615 = const()[name = tensor("op_4615"), val = tensor([1, 16, 64, 448])]; tensor var_4616_cast_fp16 = reshape(shape = var_4615, x = key_81_cast_fp16)[name = tensor("op_4616_cast_fp16")]; tensor mh_w_121_transpose_x_0 = const()[name = tensor("mh_w_121_transpose_x_0"), val = tensor(true)]; tensor mh_w_121_transpose_y_0 = const()[name = tensor("mh_w_121_transpose_y_0"), val = tensor(false)]; tensor mh_w_121_cast_fp16 = matmul(transpose_x = mh_w_121_transpose_x_0, transpose_y = mh_w_121_transpose_y_0, x = var_4612_cast_fp16, y = var_4616_cast_fp16)[name = tensor("mh_w_121_cast_fp16")]; tensor mh_w_123_cast_fp16 = add(x = mh_w_121_cast_fp16, y = var_229_cast_fp16)[name = tensor("mh_w_123_cast_fp16")]; tensor var_4624_cast_fp16 = softmax(axis = var_4536, x = mh_w_123_cast_fp16)[name = tensor("op_4624_cast_fp16")]; tensor var_4625 = const()[name = tensor("op_4625"), val = tensor([1, 16, 64, 448])]; tensor var_4626_cast_fp16 = reshape(shape = var_4625, x = value_81_cast_fp16)[name = tensor("op_4626_cast_fp16")]; tensor attn_81_transpose_x_0 = const()[name = tensor("attn_81_transpose_x_0"), val = tensor(false)]; tensor attn_81_transpose_y_0 = const()[name = tensor("attn_81_transpose_y_0"), val = tensor(true)]; tensor attn_81_cast_fp16 = matmul(transpose_x = attn_81_transpose_x_0, transpose_y = attn_81_transpose_y_0, x = var_4626_cast_fp16, y = var_4624_cast_fp16)[name = tensor("attn_81_cast_fp16")]; tensor var_4629 = const()[name = tensor("op_4629"), val = tensor([1, 1024, 1, 1])]; tensor input_201_cast_fp16 = reshape(shape = var_4629, x = attn_81_cast_fp16)[name = tensor("input_201_cast_fp16")]; tensor obj_287_pad_type_0 = const()[name = tensor("obj_287_pad_type_0"), val = tensor("valid")]; tensor obj_287_strides_0 = const()[name = tensor("obj_287_strides_0"), val = tensor([1, 1])]; tensor obj_287_pad_0 = const()[name = tensor("obj_287_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_287_dilations_0 = const()[name = tensor("obj_287_dilations_0"), val = tensor([1, 1])]; tensor obj_287_groups_0 = const()[name = tensor("obj_287_groups_0"), val = tensor(1)]; tensor layers_20_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(363244160))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364030656))), name = tensor("layers_20_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_20_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_20_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364030848)))]; tensor obj_287_cast_fp16 = conv(bias = layers_20_self_attn_o_proj_bias_to_fp16, dilations = obj_287_dilations_0, groups = obj_287_groups_0, pad = obj_287_pad_0, pad_type = obj_287_pad_type_0, strides = obj_287_strides_0, weight = layers_20_self_attn_o_proj_weight_to_fp16_palettized, x = input_201_cast_fp16)[name = tensor("obj_287_cast_fp16")]; tensor inputs_123_cast_fp16 = add(x = inputs_121_cast_fp16, y = obj_287_cast_fp16)[name = tensor("inputs_123_cast_fp16")]; tensor out_123_axes_0 = const()[name = tensor("out_123_axes_0"), val = tensor([1])]; tensor var_4651_to_fp16 = const()[name = tensor("op_4651_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_123_cast_fp16 = layer_norm(axes = out_123_axes_0, epsilon = var_4651_to_fp16, x = inputs_123_cast_fp16)[name = tensor("out_123_cast_fp16")]; tensor obj_289_gamma_0_to_fp16 = const()[name = tensor("obj_289_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364032960)))]; tensor obj_289_beta_0_to_fp16 = const()[name = tensor("obj_289_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364035072)))]; tensor obj_289_epsilon_0_to_fp16 = const()[name = tensor("obj_289_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_289_cast_fp16 = batch_norm(beta = obj_289_beta_0_to_fp16, epsilon = obj_289_epsilon_0_to_fp16, gamma = obj_289_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_123_cast_fp16)[name = tensor("obj_289_cast_fp16")]; tensor query_83_pad_type_0 = const()[name = tensor("query_83_pad_type_0"), val = tensor("valid")]; tensor query_83_strides_0 = const()[name = tensor("query_83_strides_0"), val = tensor([1, 1])]; tensor query_83_pad_0 = const()[name = tensor("query_83_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_83_dilations_0 = const()[name = tensor("query_83_dilations_0"), val = tensor([1, 1])]; tensor query_83_groups_0 = const()[name = tensor("query_83_groups_0"), val = tensor(1)]; tensor layers_20_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364037184))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364823680))), name = tensor("layers_20_encoder_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_20_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_20_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364823872)))]; tensor query_83_cast_fp16 = conv(bias = layers_20_encoder_attn_q_proj_bias_to_fp16, dilations = query_83_dilations_0, groups = query_83_groups_0, pad = query_83_pad_0, pad_type = query_83_pad_type_0, strides = query_83_strides_0, weight = layers_20_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_289_cast_fp16)[name = tensor("query_83_cast_fp16")]; tensor key_83_pad_type_0 = const()[name = tensor("key_83_pad_type_0"), val = tensor("valid")]; tensor key_83_strides_0 = const()[name = tensor("key_83_strides_0"), val = tensor([1, 1])]; tensor key_83_pad_0 = const()[name = tensor("key_83_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_83_dilations_0 = const()[name = tensor("key_83_dilations_0"), val = tensor([1, 1])]; tensor key_83_groups_0 = const()[name = tensor("key_83_groups_0"), val = tensor(1)]; tensor layers_20_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364825984))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(365612480))), name = tensor("layers_20_encoder_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor key_83_cast_fp16 = conv(dilations = key_83_dilations_0, groups = key_83_groups_0, pad = key_83_pad_0, pad_type = key_83_pad_type_0, strides = key_83_strides_0, weight = layers_20_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("key_83_cast_fp16")]; tensor value_83_pad_type_0 = const()[name = tensor("value_83_pad_type_0"), val = tensor("valid")]; tensor value_83_strides_0 = const()[name = tensor("value_83_strides_0"), val = tensor([1, 1])]; tensor value_83_pad_0 = const()[name = tensor("value_83_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_83_dilations_0 = const()[name = tensor("value_83_dilations_0"), val = tensor([1, 1])]; tensor value_83_groups_0 = const()[name = tensor("value_83_groups_0"), val = tensor(1)]; tensor layers_20_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(365612672))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(366399168))), name = tensor("layers_20_encoder_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_20_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_20_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(366399360)))]; tensor value_83_cast_fp16 = conv(bias = layers_20_encoder_attn_v_proj_bias_to_fp16, dilations = value_83_dilations_0, groups = value_83_groups_0, pad = value_83_pad_0, pad_type = value_83_pad_type_0, strides = value_83_strides_0, weight = layers_20_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("value_83_cast_fp16")]; tensor var_4687 = const()[name = tensor("op_4687"), val = tensor([1, 16, 64, 1])]; tensor mh_q_83_cast_fp16 = reshape(shape = var_4687, x = query_83_cast_fp16)[name = tensor("mh_q_83_cast_fp16")]; tensor var_4689_to_fp16 = const()[name = tensor("op_4689_to_fp16"), val = tensor(0x1p-3)]; tensor var_4690_cast_fp16 = mul(x = mh_q_83_cast_fp16, y = var_4689_to_fp16)[name = tensor("op_4690_cast_fp16")]; tensor var_4693 = const()[name = tensor("op_4693"), val = tensor([1, 16, 64, 1500])]; tensor var_4694_cast_fp16 = reshape(shape = var_4693, x = key_83_cast_fp16)[name = tensor("op_4694_cast_fp16")]; tensor mh_w_125_transpose_x_0 = const()[name = tensor("mh_w_125_transpose_x_0"), val = tensor(true)]; tensor mh_w_125_transpose_y_0 = const()[name = tensor("mh_w_125_transpose_y_0"), val = tensor(false)]; tensor mh_w_125_cast_fp16 = matmul(transpose_x = mh_w_125_transpose_x_0, transpose_y = mh_w_125_transpose_y_0, x = var_4690_cast_fp16, y = var_4694_cast_fp16)[name = tensor("mh_w_125_cast_fp16")]; tensor obj_293_cast_fp16 = softmax(axis = var_4536, x = mh_w_125_cast_fp16)[name = tensor("obj_293_cast_fp16")]; tensor var_4698 = const()[name = tensor("op_4698"), val = tensor([1, 16, 64, 1500])]; tensor var_4699_cast_fp16 = reshape(shape = var_4698, x = value_83_cast_fp16)[name = tensor("op_4699_cast_fp16")]; tensor attn_83_transpose_x_0 = const()[name = tensor("attn_83_transpose_x_0"), val = tensor(false)]; tensor attn_83_transpose_y_0 = const()[name = tensor("attn_83_transpose_y_0"), val = tensor(true)]; tensor attn_83_cast_fp16 = matmul(transpose_x = attn_83_transpose_x_0, transpose_y = attn_83_transpose_y_0, x = var_4699_cast_fp16, y = obj_293_cast_fp16)[name = tensor("attn_83_cast_fp16")]; tensor var_4702 = const()[name = tensor("op_4702"), val = tensor([1, 1024, 1, 1])]; tensor input_203_cast_fp16 = reshape(shape = var_4702, x = attn_83_cast_fp16)[name = tensor("input_203_cast_fp16")]; tensor obj_291_pad_type_0 = const()[name = tensor("obj_291_pad_type_0"), val = tensor("valid")]; tensor obj_291_strides_0 = const()[name = tensor("obj_291_strides_0"), val = tensor([1, 1])]; tensor obj_291_pad_0 = const()[name = tensor("obj_291_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_291_dilations_0 = const()[name = tensor("obj_291_dilations_0"), val = tensor([1, 1])]; tensor obj_291_groups_0 = const()[name = tensor("obj_291_groups_0"), val = tensor(1)]; tensor layers_20_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(366401472))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(367187968))), name = tensor("layers_20_encoder_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_20_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_20_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(367188160)))]; tensor obj_291_cast_fp16 = conv(bias = layers_20_encoder_attn_o_proj_bias_to_fp16, dilations = obj_291_dilations_0, groups = obj_291_groups_0, pad = obj_291_pad_0, pad_type = obj_291_pad_type_0, strides = obj_291_strides_0, weight = layers_20_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_203_cast_fp16)[name = tensor("obj_291_cast_fp16")]; tensor inputs_125_cast_fp16 = add(x = inputs_123_cast_fp16, y = obj_291_cast_fp16)[name = tensor("inputs_125_cast_fp16")]; tensor out_125_axes_0 = const()[name = tensor("out_125_axes_0"), val = tensor([1])]; tensor var_4723_to_fp16 = const()[name = tensor("op_4723_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_125_cast_fp16 = layer_norm(axes = out_125_axes_0, epsilon = var_4723_to_fp16, x = inputs_125_cast_fp16)[name = tensor("out_125_cast_fp16")]; tensor input_205_gamma_0_to_fp16 = const()[name = tensor("input_205_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(367190272)))]; tensor input_205_beta_0_to_fp16 = const()[name = tensor("input_205_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(367192384)))]; tensor input_205_epsilon_0_to_fp16 = const()[name = tensor("input_205_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_205_cast_fp16 = batch_norm(beta = input_205_beta_0_to_fp16, epsilon = input_205_epsilon_0_to_fp16, gamma = input_205_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_125_cast_fp16)[name = tensor("input_205_cast_fp16")]; tensor input_207_pad_type_0 = const()[name = tensor("input_207_pad_type_0"), val = tensor("valid")]; tensor input_207_strides_0 = const()[name = tensor("input_207_strides_0"), val = tensor([1, 1])]; tensor input_207_pad_0 = const()[name = tensor("input_207_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_207_dilations_0 = const()[name = tensor("input_207_dilations_0"), val = tensor([1, 1])]; tensor input_207_groups_0 = const()[name = tensor("input_207_groups_0"), val = tensor(1)]; tensor layers_20_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(367194496))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(370340288))), name = tensor("layers_20_fc1_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; tensor layers_20_fc1_bias_to_fp16 = const()[name = tensor("layers_20_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(370340480)))]; tensor input_207_cast_fp16 = conv(bias = layers_20_fc1_bias_to_fp16, dilations = input_207_dilations_0, groups = input_207_groups_0, pad = input_207_pad_0, pad_type = input_207_pad_type_0, strides = input_207_strides_0, weight = layers_20_fc1_weight_to_fp16_palettized, x = input_205_cast_fp16)[name = tensor("input_207_cast_fp16")]; tensor input_209_mode_0 = const()[name = tensor("input_209_mode_0"), val = tensor("EXACT")]; tensor input_209_cast_fp16 = gelu(mode = input_209_mode_0, x = input_207_cast_fp16)[name = tensor("input_209_cast_fp16")]; tensor hidden_states_43_pad_type_0 = const()[name = tensor("hidden_states_43_pad_type_0"), val = tensor("valid")]; tensor hidden_states_43_strides_0 = const()[name = tensor("hidden_states_43_strides_0"), val = tensor([1, 1])]; tensor hidden_states_43_pad_0 = const()[name = tensor("hidden_states_43_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_43_dilations_0 = const()[name = tensor("hidden_states_43_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_43_groups_0 = const()[name = tensor("hidden_states_43_groups_0"), val = tensor(1)]; tensor layers_20_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(370348736))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(373494528))), name = tensor("layers_20_fc2_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; tensor layers_20_fc2_bias_to_fp16 = const()[name = tensor("layers_20_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(373494720)))]; tensor hidden_states_43_cast_fp16 = conv(bias = layers_20_fc2_bias_to_fp16, dilations = hidden_states_43_dilations_0, groups = hidden_states_43_groups_0, pad = hidden_states_43_pad_0, pad_type = hidden_states_43_pad_type_0, strides = hidden_states_43_strides_0, weight = layers_20_fc2_weight_to_fp16_palettized, x = input_209_cast_fp16)[name = tensor("hidden_states_43_cast_fp16")]; tensor inputs_127_cast_fp16 = add(x = inputs_125_cast_fp16, y = hidden_states_43_cast_fp16)[name = tensor("inputs_127_cast_fp16")]; tensor var_4759 = const()[name = tensor("op_4759"), val = tensor(3)]; tensor out_127_axes_0 = const()[name = tensor("out_127_axes_0"), val = tensor([1])]; tensor var_4784_to_fp16 = const()[name = tensor("op_4784_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_127_cast_fp16 = layer_norm(axes = out_127_axes_0, epsilon = var_4784_to_fp16, x = inputs_127_cast_fp16)[name = tensor("out_127_cast_fp16")]; tensor obj_295_gamma_0_to_fp16 = const()[name = tensor("obj_295_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(373496832)))]; tensor obj_295_beta_0_to_fp16 = const()[name = tensor("obj_295_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(373498944)))]; tensor obj_295_epsilon_0_to_fp16 = const()[name = tensor("obj_295_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_295_cast_fp16 = batch_norm(beta = obj_295_beta_0_to_fp16, epsilon = obj_295_epsilon_0_to_fp16, gamma = obj_295_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_127_cast_fp16)[name = tensor("obj_295_cast_fp16")]; tensor query_85_pad_type_0 = const()[name = tensor("query_85_pad_type_0"), val = tensor("valid")]; tensor query_85_strides_0 = const()[name = tensor("query_85_strides_0"), val = tensor([1, 1])]; tensor query_85_pad_0 = const()[name = tensor("query_85_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_85_dilations_0 = const()[name = tensor("query_85_dilations_0"), val = tensor([1, 1])]; tensor query_85_groups_0 = const()[name = tensor("query_85_groups_0"), val = tensor(1)]; tensor layers_21_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(373501056))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(374287552))), name = tensor("layers_21_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_21_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_21_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(374287744)))]; tensor query_85_cast_fp16 = conv(bias = layers_21_self_attn_q_proj_bias_to_fp16, dilations = query_85_dilations_0, groups = query_85_groups_0, pad = query_85_pad_0, pad_type = query_85_pad_type_0, strides = query_85_strides_0, weight = layers_21_self_attn_q_proj_weight_to_fp16_palettized, x = obj_295_cast_fp16)[name = tensor("query_85_cast_fp16")]; tensor current_key_43_pad_type_0 = const()[name = tensor("current_key_43_pad_type_0"), val = tensor("valid")]; tensor current_key_43_strides_0 = const()[name = tensor("current_key_43_strides_0"), val = tensor([1, 1])]; tensor current_key_43_pad_0 = const()[name = tensor("current_key_43_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_43_dilations_0 = const()[name = tensor("current_key_43_dilations_0"), val = tensor([1, 1])]; tensor current_key_43_groups_0 = const()[name = tensor("current_key_43_groups_0"), val = tensor(1)]; tensor layers_21_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(374289856))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(375076352))), name = tensor("layers_21_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor current_key_43_cast_fp16 = conv(dilations = current_key_43_dilations_0, groups = current_key_43_groups_0, pad = current_key_43_pad_0, pad_type = current_key_43_pad_type_0, strides = current_key_43_strides_0, weight = layers_21_self_attn_k_proj_weight_to_fp16_palettized, x = obj_295_cast_fp16)[name = tensor("current_key_43_cast_fp16")]; tensor current_value_43_pad_type_0 = const()[name = tensor("current_value_43_pad_type_0"), val = tensor("valid")]; tensor current_value_43_strides_0 = const()[name = tensor("current_value_43_strides_0"), val = tensor([1, 1])]; tensor current_value_43_pad_0 = const()[name = tensor("current_value_43_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_43_dilations_0 = const()[name = tensor("current_value_43_dilations_0"), val = tensor([1, 1])]; tensor current_value_43_groups_0 = const()[name = tensor("current_value_43_groups_0"), val = tensor(1)]; tensor layers_21_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(375076544))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(375863040))), name = tensor("layers_21_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_21_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_21_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(375863232)))]; tensor current_value_43_cast_fp16 = conv(bias = layers_21_self_attn_v_proj_bias_to_fp16, dilations = current_value_43_dilations_0, groups = current_value_43_groups_0, pad = current_value_43_pad_0, pad_type = current_value_43_pad_type_0, strides = current_value_43_strides_0, weight = layers_21_self_attn_v_proj_weight_to_fp16_palettized, x = obj_295_cast_fp16)[name = tensor("current_value_43_cast_fp16")]; tensor var_4823_cast_fp16 = mul(x = var_87_cast_fp16_21, y = var_207_cast_fp16)[name = tensor("op_4823_cast_fp16")]; tensor var_4824_cast_fp16 = mul(x = current_key_43_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_4824_cast_fp16")]; tensor key_85_cast_fp16 = add(x = var_4823_cast_fp16, y = var_4824_cast_fp16)[name = tensor("key_85_cast_fp16")]; tensor var_4827_cast_fp16 = mul(x = var_114_cast_fp16_21, y = var_207_cast_fp16)[name = tensor("op_4827_cast_fp16")]; tensor var_4828_cast_fp16 = mul(x = current_value_43_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_4828_cast_fp16")]; tensor value_85_cast_fp16 = add(x = var_4827_cast_fp16, y = var_4828_cast_fp16)[name = tensor("value_85_cast_fp16")]; tensor var_4832 = const()[name = tensor("op_4832"), val = tensor([1, 16, 64, 1])]; tensor mh_q_85_cast_fp16 = reshape(shape = var_4832, x = query_85_cast_fp16)[name = tensor("mh_q_85_cast_fp16")]; tensor var_4834_to_fp16 = const()[name = tensor("op_4834_to_fp16"), val = tensor(0x1p-3)]; tensor var_4835_cast_fp16 = mul(x = mh_q_85_cast_fp16, y = var_4834_to_fp16)[name = tensor("op_4835_cast_fp16")]; tensor var_4838 = const()[name = tensor("op_4838"), val = tensor([1, 16, 64, 448])]; tensor var_4839_cast_fp16 = reshape(shape = var_4838, x = key_85_cast_fp16)[name = tensor("op_4839_cast_fp16")]; tensor mh_w_127_transpose_x_0 = const()[name = tensor("mh_w_127_transpose_x_0"), val = tensor(true)]; tensor mh_w_127_transpose_y_0 = const()[name = tensor("mh_w_127_transpose_y_0"), val = tensor(false)]; tensor mh_w_127_cast_fp16 = matmul(transpose_x = mh_w_127_transpose_x_0, transpose_y = mh_w_127_transpose_y_0, x = var_4835_cast_fp16, y = var_4839_cast_fp16)[name = tensor("mh_w_127_cast_fp16")]; tensor mh_w_129_cast_fp16 = add(x = mh_w_127_cast_fp16, y = var_229_cast_fp16)[name = tensor("mh_w_129_cast_fp16")]; tensor var_4847_cast_fp16 = softmax(axis = var_4759, x = mh_w_129_cast_fp16)[name = tensor("op_4847_cast_fp16")]; tensor var_4848 = const()[name = tensor("op_4848"), val = tensor([1, 16, 64, 448])]; tensor var_4849_cast_fp16 = reshape(shape = var_4848, x = value_85_cast_fp16)[name = tensor("op_4849_cast_fp16")]; tensor attn_85_transpose_x_0 = const()[name = tensor("attn_85_transpose_x_0"), val = tensor(false)]; tensor attn_85_transpose_y_0 = const()[name = tensor("attn_85_transpose_y_0"), val = tensor(true)]; tensor attn_85_cast_fp16 = matmul(transpose_x = attn_85_transpose_x_0, transpose_y = attn_85_transpose_y_0, x = var_4849_cast_fp16, y = var_4847_cast_fp16)[name = tensor("attn_85_cast_fp16")]; tensor var_4852 = const()[name = tensor("op_4852"), val = tensor([1, 1024, 1, 1])]; tensor input_211_cast_fp16 = reshape(shape = var_4852, x = attn_85_cast_fp16)[name = tensor("input_211_cast_fp16")]; tensor obj_301_pad_type_0 = const()[name = tensor("obj_301_pad_type_0"), val = tensor("valid")]; tensor obj_301_strides_0 = const()[name = tensor("obj_301_strides_0"), val = tensor([1, 1])]; tensor obj_301_pad_0 = const()[name = tensor("obj_301_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_301_dilations_0 = const()[name = tensor("obj_301_dilations_0"), val = tensor([1, 1])]; tensor obj_301_groups_0 = const()[name = tensor("obj_301_groups_0"), val = tensor(1)]; tensor layers_21_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(375865344))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(376651840))), name = tensor("layers_21_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_21_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_21_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(376652032)))]; tensor obj_301_cast_fp16 = conv(bias = layers_21_self_attn_o_proj_bias_to_fp16, dilations = obj_301_dilations_0, groups = obj_301_groups_0, pad = obj_301_pad_0, pad_type = obj_301_pad_type_0, strides = obj_301_strides_0, weight = layers_21_self_attn_o_proj_weight_to_fp16_palettized, x = input_211_cast_fp16)[name = tensor("obj_301_cast_fp16")]; tensor inputs_129_cast_fp16 = add(x = inputs_127_cast_fp16, y = obj_301_cast_fp16)[name = tensor("inputs_129_cast_fp16")]; tensor out_129_axes_0 = const()[name = tensor("out_129_axes_0"), val = tensor([1])]; tensor var_4874_to_fp16 = const()[name = tensor("op_4874_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_129_cast_fp16 = layer_norm(axes = out_129_axes_0, epsilon = var_4874_to_fp16, x = inputs_129_cast_fp16)[name = tensor("out_129_cast_fp16")]; tensor obj_303_gamma_0_to_fp16 = const()[name = tensor("obj_303_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(376654144)))]; tensor obj_303_beta_0_to_fp16 = const()[name = tensor("obj_303_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(376656256)))]; tensor obj_303_epsilon_0_to_fp16 = const()[name = tensor("obj_303_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_303_cast_fp16 = batch_norm(beta = obj_303_beta_0_to_fp16, epsilon = obj_303_epsilon_0_to_fp16, gamma = obj_303_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_129_cast_fp16)[name = tensor("obj_303_cast_fp16")]; tensor query_87_pad_type_0 = const()[name = tensor("query_87_pad_type_0"), val = tensor("valid")]; tensor query_87_strides_0 = const()[name = tensor("query_87_strides_0"), val = tensor([1, 1])]; tensor query_87_pad_0 = const()[name = tensor("query_87_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_87_dilations_0 = const()[name = tensor("query_87_dilations_0"), val = tensor([1, 1])]; tensor query_87_groups_0 = const()[name = tensor("query_87_groups_0"), val = tensor(1)]; tensor layers_21_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(376658368))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(377444864))), name = tensor("layers_21_encoder_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_21_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_21_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(377445056)))]; tensor query_87_cast_fp16 = conv(bias = layers_21_encoder_attn_q_proj_bias_to_fp16, dilations = query_87_dilations_0, groups = query_87_groups_0, pad = query_87_pad_0, pad_type = query_87_pad_type_0, strides = query_87_strides_0, weight = layers_21_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_303_cast_fp16)[name = tensor("query_87_cast_fp16")]; tensor key_87_pad_type_0 = const()[name = tensor("key_87_pad_type_0"), val = tensor("valid")]; tensor key_87_strides_0 = const()[name = tensor("key_87_strides_0"), val = tensor([1, 1])]; tensor key_87_pad_0 = const()[name = tensor("key_87_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_87_dilations_0 = const()[name = tensor("key_87_dilations_0"), val = tensor([1, 1])]; tensor key_87_groups_0 = const()[name = tensor("key_87_groups_0"), val = tensor(1)]; tensor layers_21_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(377447168))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(378233664))), name = tensor("layers_21_encoder_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor key_87_cast_fp16 = conv(dilations = key_87_dilations_0, groups = key_87_groups_0, pad = key_87_pad_0, pad_type = key_87_pad_type_0, strides = key_87_strides_0, weight = layers_21_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("key_87_cast_fp16")]; tensor value_87_pad_type_0 = const()[name = tensor("value_87_pad_type_0"), val = tensor("valid")]; tensor value_87_strides_0 = const()[name = tensor("value_87_strides_0"), val = tensor([1, 1])]; tensor value_87_pad_0 = const()[name = tensor("value_87_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_87_dilations_0 = const()[name = tensor("value_87_dilations_0"), val = tensor([1, 1])]; tensor value_87_groups_0 = const()[name = tensor("value_87_groups_0"), val = tensor(1)]; tensor layers_21_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(378233856))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(379020352))), name = tensor("layers_21_encoder_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_21_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_21_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(379020544)))]; tensor value_87_cast_fp16 = conv(bias = layers_21_encoder_attn_v_proj_bias_to_fp16, dilations = value_87_dilations_0, groups = value_87_groups_0, pad = value_87_pad_0, pad_type = value_87_pad_type_0, strides = value_87_strides_0, weight = layers_21_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("value_87_cast_fp16")]; tensor var_4910 = const()[name = tensor("op_4910"), val = tensor([1, 16, 64, 1])]; tensor mh_q_87_cast_fp16 = reshape(shape = var_4910, x = query_87_cast_fp16)[name = tensor("mh_q_87_cast_fp16")]; tensor var_4912_to_fp16 = const()[name = tensor("op_4912_to_fp16"), val = tensor(0x1p-3)]; tensor var_4913_cast_fp16 = mul(x = mh_q_87_cast_fp16, y = var_4912_to_fp16)[name = tensor("op_4913_cast_fp16")]; tensor var_4916 = const()[name = tensor("op_4916"), val = tensor([1, 16, 64, 1500])]; tensor var_4917_cast_fp16 = reshape(shape = var_4916, x = key_87_cast_fp16)[name = tensor("op_4917_cast_fp16")]; tensor mh_w_131_transpose_x_0 = const()[name = tensor("mh_w_131_transpose_x_0"), val = tensor(true)]; tensor mh_w_131_transpose_y_0 = const()[name = tensor("mh_w_131_transpose_y_0"), val = tensor(false)]; tensor mh_w_131_cast_fp16 = matmul(transpose_x = mh_w_131_transpose_x_0, transpose_y = mh_w_131_transpose_y_0, x = var_4913_cast_fp16, y = var_4917_cast_fp16)[name = tensor("mh_w_131_cast_fp16")]; tensor obj_307_cast_fp16 = softmax(axis = var_4759, x = mh_w_131_cast_fp16)[name = tensor("obj_307_cast_fp16")]; tensor var_4921 = const()[name = tensor("op_4921"), val = tensor([1, 16, 64, 1500])]; tensor var_4922_cast_fp16 = reshape(shape = var_4921, x = value_87_cast_fp16)[name = tensor("op_4922_cast_fp16")]; tensor attn_87_transpose_x_0 = const()[name = tensor("attn_87_transpose_x_0"), val = tensor(false)]; tensor attn_87_transpose_y_0 = const()[name = tensor("attn_87_transpose_y_0"), val = tensor(true)]; tensor attn_87_cast_fp16 = matmul(transpose_x = attn_87_transpose_x_0, transpose_y = attn_87_transpose_y_0, x = var_4922_cast_fp16, y = obj_307_cast_fp16)[name = tensor("attn_87_cast_fp16")]; tensor var_4925 = const()[name = tensor("op_4925"), val = tensor([1, 1024, 1, 1])]; tensor input_213_cast_fp16 = reshape(shape = var_4925, x = attn_87_cast_fp16)[name = tensor("input_213_cast_fp16")]; tensor obj_305_pad_type_0 = const()[name = tensor("obj_305_pad_type_0"), val = tensor("valid")]; tensor obj_305_strides_0 = const()[name = tensor("obj_305_strides_0"), val = tensor([1, 1])]; tensor obj_305_pad_0 = const()[name = tensor("obj_305_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_305_dilations_0 = const()[name = tensor("obj_305_dilations_0"), val = tensor([1, 1])]; tensor obj_305_groups_0 = const()[name = tensor("obj_305_groups_0"), val = tensor(1)]; tensor layers_21_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(379022656))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(379809152))), name = tensor("layers_21_encoder_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_21_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_21_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(379809344)))]; tensor obj_305_cast_fp16 = conv(bias = layers_21_encoder_attn_o_proj_bias_to_fp16, dilations = obj_305_dilations_0, groups = obj_305_groups_0, pad = obj_305_pad_0, pad_type = obj_305_pad_type_0, strides = obj_305_strides_0, weight = layers_21_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_213_cast_fp16)[name = tensor("obj_305_cast_fp16")]; tensor inputs_131_cast_fp16 = add(x = inputs_129_cast_fp16, y = obj_305_cast_fp16)[name = tensor("inputs_131_cast_fp16")]; tensor out_131_axes_0 = const()[name = tensor("out_131_axes_0"), val = tensor([1])]; tensor var_4943_to_fp16 = const()[name = tensor("op_4943_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_131_cast_fp16 = layer_norm(axes = out_131_axes_0, epsilon = var_4943_to_fp16, x = inputs_131_cast_fp16)[name = tensor("out_131_cast_fp16")]; tensor input_215_gamma_0_to_fp16 = const()[name = tensor("input_215_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(379811456)))]; tensor input_215_beta_0_to_fp16 = const()[name = tensor("input_215_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(379813568)))]; tensor input_215_epsilon_0_to_fp16 = const()[name = tensor("input_215_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_215_cast_fp16 = batch_norm(beta = input_215_beta_0_to_fp16, epsilon = input_215_epsilon_0_to_fp16, gamma = input_215_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_131_cast_fp16)[name = tensor("input_215_cast_fp16")]; tensor input_217_pad_type_0 = const()[name = tensor("input_217_pad_type_0"), val = tensor("valid")]; tensor input_217_strides_0 = const()[name = tensor("input_217_strides_0"), val = tensor([1, 1])]; tensor input_217_pad_0 = const()[name = tensor("input_217_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_217_dilations_0 = const()[name = tensor("input_217_dilations_0"), val = tensor([1, 1])]; tensor input_217_groups_0 = const()[name = tensor("input_217_groups_0"), val = tensor(1)]; tensor layers_21_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(379815680))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(382961472))), name = tensor("layers_21_fc1_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; tensor layers_21_fc1_bias_to_fp16 = const()[name = tensor("layers_21_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(382961664)))]; tensor input_217_cast_fp16 = conv(bias = layers_21_fc1_bias_to_fp16, dilations = input_217_dilations_0, groups = input_217_groups_0, pad = input_217_pad_0, pad_type = input_217_pad_type_0, strides = input_217_strides_0, weight = layers_21_fc1_weight_to_fp16_palettized, x = input_215_cast_fp16)[name = tensor("input_217_cast_fp16")]; tensor input_219_mode_0 = const()[name = tensor("input_219_mode_0"), val = tensor("EXACT")]; tensor input_219_cast_fp16 = gelu(mode = input_219_mode_0, x = input_217_cast_fp16)[name = tensor("input_219_cast_fp16")]; tensor hidden_states_45_pad_type_0 = const()[name = tensor("hidden_states_45_pad_type_0"), val = tensor("valid")]; tensor hidden_states_45_strides_0 = const()[name = tensor("hidden_states_45_strides_0"), val = tensor([1, 1])]; tensor hidden_states_45_pad_0 = const()[name = tensor("hidden_states_45_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_45_dilations_0 = const()[name = tensor("hidden_states_45_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_45_groups_0 = const()[name = tensor("hidden_states_45_groups_0"), val = tensor(1)]; tensor layers_21_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(382969920))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(386115712))), name = tensor("layers_21_fc2_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; tensor layers_21_fc2_bias_to_fp16 = const()[name = tensor("layers_21_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(386115904)))]; tensor hidden_states_45_cast_fp16 = conv(bias = layers_21_fc2_bias_to_fp16, dilations = hidden_states_45_dilations_0, groups = hidden_states_45_groups_0, pad = hidden_states_45_pad_0, pad_type = hidden_states_45_pad_type_0, strides = hidden_states_45_strides_0, weight = layers_21_fc2_weight_to_fp16_palettized, x = input_219_cast_fp16)[name = tensor("hidden_states_45_cast_fp16")]; tensor inputs_133_cast_fp16 = add(x = inputs_131_cast_fp16, y = hidden_states_45_cast_fp16)[name = tensor("inputs_133_cast_fp16")]; tensor var_4978 = const()[name = tensor("op_4978"), val = tensor(3)]; tensor out_133_axes_0 = const()[name = tensor("out_133_axes_0"), val = tensor([1])]; tensor var_5003_to_fp16 = const()[name = tensor("op_5003_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_133_cast_fp16 = layer_norm(axes = out_133_axes_0, epsilon = var_5003_to_fp16, x = inputs_133_cast_fp16)[name = tensor("out_133_cast_fp16")]; tensor obj_309_gamma_0_to_fp16 = const()[name = tensor("obj_309_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(386118016)))]; tensor obj_309_beta_0_to_fp16 = const()[name = tensor("obj_309_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(386120128)))]; tensor obj_309_epsilon_0_to_fp16 = const()[name = tensor("obj_309_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_309_cast_fp16 = batch_norm(beta = obj_309_beta_0_to_fp16, epsilon = obj_309_epsilon_0_to_fp16, gamma = obj_309_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_133_cast_fp16)[name = tensor("obj_309_cast_fp16")]; tensor query_89_pad_type_0 = const()[name = tensor("query_89_pad_type_0"), val = tensor("valid")]; tensor query_89_strides_0 = const()[name = tensor("query_89_strides_0"), val = tensor([1, 1])]; tensor query_89_pad_0 = const()[name = tensor("query_89_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_89_dilations_0 = const()[name = tensor("query_89_dilations_0"), val = tensor([1, 1])]; tensor query_89_groups_0 = const()[name = tensor("query_89_groups_0"), val = tensor(1)]; tensor layers_22_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(386122240))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(386908736))), name = tensor("layers_22_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_22_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_22_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(386908928)))]; tensor query_89_cast_fp16 = conv(bias = layers_22_self_attn_q_proj_bias_to_fp16, dilations = query_89_dilations_0, groups = query_89_groups_0, pad = query_89_pad_0, pad_type = query_89_pad_type_0, strides = query_89_strides_0, weight = layers_22_self_attn_q_proj_weight_to_fp16_palettized, x = obj_309_cast_fp16)[name = tensor("query_89_cast_fp16")]; tensor current_key_45_pad_type_0 = const()[name = tensor("current_key_45_pad_type_0"), val = tensor("valid")]; tensor current_key_45_strides_0 = const()[name = tensor("current_key_45_strides_0"), val = tensor([1, 1])]; tensor current_key_45_pad_0 = const()[name = tensor("current_key_45_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_45_dilations_0 = const()[name = tensor("current_key_45_dilations_0"), val = tensor([1, 1])]; tensor current_key_45_groups_0 = const()[name = tensor("current_key_45_groups_0"), val = tensor(1)]; tensor layers_22_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(386911040))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(387697536))), name = tensor("layers_22_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor current_key_45_cast_fp16 = conv(dilations = current_key_45_dilations_0, groups = current_key_45_groups_0, pad = current_key_45_pad_0, pad_type = current_key_45_pad_type_0, strides = current_key_45_strides_0, weight = layers_22_self_attn_k_proj_weight_to_fp16_palettized, x = obj_309_cast_fp16)[name = tensor("current_key_45_cast_fp16")]; tensor current_value_45_pad_type_0 = const()[name = tensor("current_value_45_pad_type_0"), val = tensor("valid")]; tensor current_value_45_strides_0 = const()[name = tensor("current_value_45_strides_0"), val = tensor([1, 1])]; tensor current_value_45_pad_0 = const()[name = tensor("current_value_45_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_45_dilations_0 = const()[name = tensor("current_value_45_dilations_0"), val = tensor([1, 1])]; tensor current_value_45_groups_0 = const()[name = tensor("current_value_45_groups_0"), val = tensor(1)]; tensor layers_22_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(387697728))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(388484224))), name = tensor("layers_22_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_22_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_22_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(388484416)))]; tensor current_value_45_cast_fp16 = conv(bias = layers_22_self_attn_v_proj_bias_to_fp16, dilations = current_value_45_dilations_0, groups = current_value_45_groups_0, pad = current_value_45_pad_0, pad_type = current_value_45_pad_type_0, strides = current_value_45_strides_0, weight = layers_22_self_attn_v_proj_weight_to_fp16_palettized, x = obj_309_cast_fp16)[name = tensor("current_value_45_cast_fp16")]; tensor var_5042_cast_fp16 = mul(x = var_87_cast_fp16_22, y = var_207_cast_fp16)[name = tensor("op_5042_cast_fp16")]; tensor var_5043_cast_fp16 = mul(x = current_key_45_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_5043_cast_fp16")]; tensor key_89_cast_fp16 = add(x = var_5042_cast_fp16, y = var_5043_cast_fp16)[name = tensor("key_89_cast_fp16")]; tensor var_5046_cast_fp16 = mul(x = var_114_cast_fp16_22, y = var_207_cast_fp16)[name = tensor("op_5046_cast_fp16")]; tensor var_5047_cast_fp16 = mul(x = current_value_45_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_5047_cast_fp16")]; tensor value_89_cast_fp16 = add(x = var_5046_cast_fp16, y = var_5047_cast_fp16)[name = tensor("value_89_cast_fp16")]; tensor var_5051 = const()[name = tensor("op_5051"), val = tensor([1, 16, 64, 1])]; tensor mh_q_89_cast_fp16 = reshape(shape = var_5051, x = query_89_cast_fp16)[name = tensor("mh_q_89_cast_fp16")]; tensor var_5053_to_fp16 = const()[name = tensor("op_5053_to_fp16"), val = tensor(0x1p-3)]; tensor var_5054_cast_fp16 = mul(x = mh_q_89_cast_fp16, y = var_5053_to_fp16)[name = tensor("op_5054_cast_fp16")]; tensor var_5057 = const()[name = tensor("op_5057"), val = tensor([1, 16, 64, 448])]; tensor var_5058_cast_fp16 = reshape(shape = var_5057, x = key_89_cast_fp16)[name = tensor("op_5058_cast_fp16")]; tensor mh_w_133_transpose_x_0 = const()[name = tensor("mh_w_133_transpose_x_0"), val = tensor(true)]; tensor mh_w_133_transpose_y_0 = const()[name = tensor("mh_w_133_transpose_y_0"), val = tensor(false)]; tensor mh_w_133_cast_fp16 = matmul(transpose_x = mh_w_133_transpose_x_0, transpose_y = mh_w_133_transpose_y_0, x = var_5054_cast_fp16, y = var_5058_cast_fp16)[name = tensor("mh_w_133_cast_fp16")]; tensor mh_w_135_cast_fp16 = add(x = mh_w_133_cast_fp16, y = var_229_cast_fp16)[name = tensor("mh_w_135_cast_fp16")]; tensor var_5066_cast_fp16 = softmax(axis = var_4978, x = mh_w_135_cast_fp16)[name = tensor("op_5066_cast_fp16")]; tensor var_5067 = const()[name = tensor("op_5067"), val = tensor([1, 16, 64, 448])]; tensor var_5068_cast_fp16 = reshape(shape = var_5067, x = value_89_cast_fp16)[name = tensor("op_5068_cast_fp16")]; tensor attn_89_transpose_x_0 = const()[name = tensor("attn_89_transpose_x_0"), val = tensor(false)]; tensor attn_89_transpose_y_0 = const()[name = tensor("attn_89_transpose_y_0"), val = tensor(true)]; tensor attn_89_cast_fp16 = matmul(transpose_x = attn_89_transpose_x_0, transpose_y = attn_89_transpose_y_0, x = var_5068_cast_fp16, y = var_5066_cast_fp16)[name = tensor("attn_89_cast_fp16")]; tensor var_5071 = const()[name = tensor("op_5071"), val = tensor([1, 1024, 1, 1])]; tensor input_221_cast_fp16 = reshape(shape = var_5071, x = attn_89_cast_fp16)[name = tensor("input_221_cast_fp16")]; tensor obj_315_pad_type_0 = const()[name = tensor("obj_315_pad_type_0"), val = tensor("valid")]; tensor obj_315_strides_0 = const()[name = tensor("obj_315_strides_0"), val = tensor([1, 1])]; tensor obj_315_pad_0 = const()[name = tensor("obj_315_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_315_dilations_0 = const()[name = tensor("obj_315_dilations_0"), val = tensor([1, 1])]; tensor obj_315_groups_0 = const()[name = tensor("obj_315_groups_0"), val = tensor(1)]; tensor layers_22_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(388486528))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(389273024))), name = tensor("layers_22_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_22_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_22_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(389273216)))]; tensor obj_315_cast_fp16 = conv(bias = layers_22_self_attn_o_proj_bias_to_fp16, dilations = obj_315_dilations_0, groups = obj_315_groups_0, pad = obj_315_pad_0, pad_type = obj_315_pad_type_0, strides = obj_315_strides_0, weight = layers_22_self_attn_o_proj_weight_to_fp16_palettized, x = input_221_cast_fp16)[name = tensor("obj_315_cast_fp16")]; tensor inputs_135_cast_fp16 = add(x = inputs_133_cast_fp16, y = obj_315_cast_fp16)[name = tensor("inputs_135_cast_fp16")]; tensor out_135_axes_0 = const()[name = tensor("out_135_axes_0"), val = tensor([1])]; tensor var_5093_to_fp16 = const()[name = tensor("op_5093_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_135_cast_fp16 = layer_norm(axes = out_135_axes_0, epsilon = var_5093_to_fp16, x = inputs_135_cast_fp16)[name = tensor("out_135_cast_fp16")]; tensor obj_317_gamma_0_to_fp16 = const()[name = tensor("obj_317_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(389275328)))]; tensor obj_317_beta_0_to_fp16 = const()[name = tensor("obj_317_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(389277440)))]; tensor obj_317_epsilon_0_to_fp16 = const()[name = tensor("obj_317_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_317_cast_fp16 = batch_norm(beta = obj_317_beta_0_to_fp16, epsilon = obj_317_epsilon_0_to_fp16, gamma = obj_317_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_135_cast_fp16)[name = tensor("obj_317_cast_fp16")]; tensor query_91_pad_type_0 = const()[name = tensor("query_91_pad_type_0"), val = tensor("valid")]; tensor query_91_strides_0 = const()[name = tensor("query_91_strides_0"), val = tensor([1, 1])]; tensor query_91_pad_0 = const()[name = tensor("query_91_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_91_dilations_0 = const()[name = tensor("query_91_dilations_0"), val = tensor([1, 1])]; tensor query_91_groups_0 = const()[name = tensor("query_91_groups_0"), val = tensor(1)]; tensor layers_22_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(389279552))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(390066048))), name = tensor("layers_22_encoder_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_22_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_22_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(390066240)))]; tensor query_91_cast_fp16 = conv(bias = layers_22_encoder_attn_q_proj_bias_to_fp16, dilations = query_91_dilations_0, groups = query_91_groups_0, pad = query_91_pad_0, pad_type = query_91_pad_type_0, strides = query_91_strides_0, weight = layers_22_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_317_cast_fp16)[name = tensor("query_91_cast_fp16")]; tensor key_91_pad_type_0 = const()[name = tensor("key_91_pad_type_0"), val = tensor("valid")]; tensor key_91_strides_0 = const()[name = tensor("key_91_strides_0"), val = tensor([1, 1])]; tensor key_91_pad_0 = const()[name = tensor("key_91_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_91_dilations_0 = const()[name = tensor("key_91_dilations_0"), val = tensor([1, 1])]; tensor key_91_groups_0 = const()[name = tensor("key_91_groups_0"), val = tensor(1)]; tensor layers_22_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(390068352))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(390854848))), name = tensor("layers_22_encoder_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor key_91_cast_fp16 = conv(dilations = key_91_dilations_0, groups = key_91_groups_0, pad = key_91_pad_0, pad_type = key_91_pad_type_0, strides = key_91_strides_0, weight = layers_22_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("key_91_cast_fp16")]; tensor value_91_pad_type_0 = const()[name = tensor("value_91_pad_type_0"), val = tensor("valid")]; tensor value_91_strides_0 = const()[name = tensor("value_91_strides_0"), val = tensor([1, 1])]; tensor value_91_pad_0 = const()[name = tensor("value_91_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_91_dilations_0 = const()[name = tensor("value_91_dilations_0"), val = tensor([1, 1])]; tensor value_91_groups_0 = const()[name = tensor("value_91_groups_0"), val = tensor(1)]; tensor layers_22_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(390855040))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(391641536))), name = tensor("layers_22_encoder_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_22_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_22_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(391641728)))]; tensor value_91_cast_fp16 = conv(bias = layers_22_encoder_attn_v_proj_bias_to_fp16, dilations = value_91_dilations_0, groups = value_91_groups_0, pad = value_91_pad_0, pad_type = value_91_pad_type_0, strides = value_91_strides_0, weight = layers_22_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("value_91_cast_fp16")]; tensor var_5129 = const()[name = tensor("op_5129"), val = tensor([1, 16, 64, 1])]; tensor mh_q_91_cast_fp16 = reshape(shape = var_5129, x = query_91_cast_fp16)[name = tensor("mh_q_91_cast_fp16")]; tensor var_5131_to_fp16 = const()[name = tensor("op_5131_to_fp16"), val = tensor(0x1p-3)]; tensor var_5132_cast_fp16 = mul(x = mh_q_91_cast_fp16, y = var_5131_to_fp16)[name = tensor("op_5132_cast_fp16")]; tensor var_5135 = const()[name = tensor("op_5135"), val = tensor([1, 16, 64, 1500])]; tensor var_5136_cast_fp16 = reshape(shape = var_5135, x = key_91_cast_fp16)[name = tensor("op_5136_cast_fp16")]; tensor mh_w_137_transpose_x_0 = const()[name = tensor("mh_w_137_transpose_x_0"), val = tensor(true)]; tensor mh_w_137_transpose_y_0 = const()[name = tensor("mh_w_137_transpose_y_0"), val = tensor(false)]; tensor mh_w_137_cast_fp16 = matmul(transpose_x = mh_w_137_transpose_x_0, transpose_y = mh_w_137_transpose_y_0, x = var_5132_cast_fp16, y = var_5136_cast_fp16)[name = tensor("mh_w_137_cast_fp16")]; tensor obj_321_cast_fp16 = softmax(axis = var_4978, x = mh_w_137_cast_fp16)[name = tensor("obj_321_cast_fp16")]; tensor var_5140 = const()[name = tensor("op_5140"), val = tensor([1, 16, 64, 1500])]; tensor var_5141_cast_fp16 = reshape(shape = var_5140, x = value_91_cast_fp16)[name = tensor("op_5141_cast_fp16")]; tensor attn_91_transpose_x_0 = const()[name = tensor("attn_91_transpose_x_0"), val = tensor(false)]; tensor attn_91_transpose_y_0 = const()[name = tensor("attn_91_transpose_y_0"), val = tensor(true)]; tensor attn_91_cast_fp16 = matmul(transpose_x = attn_91_transpose_x_0, transpose_y = attn_91_transpose_y_0, x = var_5141_cast_fp16, y = obj_321_cast_fp16)[name = tensor("attn_91_cast_fp16")]; tensor var_5144 = const()[name = tensor("op_5144"), val = tensor([1, 1024, 1, 1])]; tensor input_223_cast_fp16 = reshape(shape = var_5144, x = attn_91_cast_fp16)[name = tensor("input_223_cast_fp16")]; tensor obj_319_pad_type_0 = const()[name = tensor("obj_319_pad_type_0"), val = tensor("valid")]; tensor obj_319_strides_0 = const()[name = tensor("obj_319_strides_0"), val = tensor([1, 1])]; tensor obj_319_pad_0 = const()[name = tensor("obj_319_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_319_dilations_0 = const()[name = tensor("obj_319_dilations_0"), val = tensor([1, 1])]; tensor obj_319_groups_0 = const()[name = tensor("obj_319_groups_0"), val = tensor(1)]; tensor layers_22_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(391643840))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(392430336))), name = tensor("layers_22_encoder_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_22_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_22_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(392430528)))]; tensor obj_319_cast_fp16 = conv(bias = layers_22_encoder_attn_o_proj_bias_to_fp16, dilations = obj_319_dilations_0, groups = obj_319_groups_0, pad = obj_319_pad_0, pad_type = obj_319_pad_type_0, strides = obj_319_strides_0, weight = layers_22_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_223_cast_fp16)[name = tensor("obj_319_cast_fp16")]; tensor inputs_137_cast_fp16 = add(x = inputs_135_cast_fp16, y = obj_319_cast_fp16)[name = tensor("inputs_137_cast_fp16")]; tensor out_137_axes_0 = const()[name = tensor("out_137_axes_0"), val = tensor([1])]; tensor var_5162_to_fp16 = const()[name = tensor("op_5162_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_137_cast_fp16 = layer_norm(axes = out_137_axes_0, epsilon = var_5162_to_fp16, x = inputs_137_cast_fp16)[name = tensor("out_137_cast_fp16")]; tensor input_225_gamma_0_to_fp16 = const()[name = tensor("input_225_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(392432640)))]; tensor input_225_beta_0_to_fp16 = const()[name = tensor("input_225_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(392434752)))]; tensor input_225_epsilon_0_to_fp16 = const()[name = tensor("input_225_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_225_cast_fp16 = batch_norm(beta = input_225_beta_0_to_fp16, epsilon = input_225_epsilon_0_to_fp16, gamma = input_225_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_137_cast_fp16)[name = tensor("input_225_cast_fp16")]; tensor input_227_pad_type_0 = const()[name = tensor("input_227_pad_type_0"), val = tensor("valid")]; tensor input_227_strides_0 = const()[name = tensor("input_227_strides_0"), val = tensor([1, 1])]; tensor input_227_pad_0 = const()[name = tensor("input_227_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_227_dilations_0 = const()[name = tensor("input_227_dilations_0"), val = tensor([1, 1])]; tensor input_227_groups_0 = const()[name = tensor("input_227_groups_0"), val = tensor(1)]; tensor layers_22_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(392436864))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(395582656))), name = tensor("layers_22_fc1_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; tensor layers_22_fc1_bias_to_fp16 = const()[name = tensor("layers_22_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(395582848)))]; tensor input_227_cast_fp16 = conv(bias = layers_22_fc1_bias_to_fp16, dilations = input_227_dilations_0, groups = input_227_groups_0, pad = input_227_pad_0, pad_type = input_227_pad_type_0, strides = input_227_strides_0, weight = layers_22_fc1_weight_to_fp16_palettized, x = input_225_cast_fp16)[name = tensor("input_227_cast_fp16")]; tensor input_229_mode_0 = const()[name = tensor("input_229_mode_0"), val = tensor("EXACT")]; tensor input_229_cast_fp16 = gelu(mode = input_229_mode_0, x = input_227_cast_fp16)[name = tensor("input_229_cast_fp16")]; tensor hidden_states_47_pad_type_0 = const()[name = tensor("hidden_states_47_pad_type_0"), val = tensor("valid")]; tensor hidden_states_47_strides_0 = const()[name = tensor("hidden_states_47_strides_0"), val = tensor([1, 1])]; tensor hidden_states_47_pad_0 = const()[name = tensor("hidden_states_47_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_47_dilations_0 = const()[name = tensor("hidden_states_47_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_47_groups_0 = const()[name = tensor("hidden_states_47_groups_0"), val = tensor(1)]; tensor layers_22_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(395591104))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(398736896))), name = tensor("layers_22_fc2_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; tensor layers_22_fc2_bias_to_fp16 = const()[name = tensor("layers_22_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(398737088)))]; tensor hidden_states_47_cast_fp16 = conv(bias = layers_22_fc2_bias_to_fp16, dilations = hidden_states_47_dilations_0, groups = hidden_states_47_groups_0, pad = hidden_states_47_pad_0, pad_type = hidden_states_47_pad_type_0, strides = hidden_states_47_strides_0, weight = layers_22_fc2_weight_to_fp16_palettized, x = input_229_cast_fp16)[name = tensor("hidden_states_47_cast_fp16")]; tensor inputs_139_cast_fp16 = add(x = inputs_137_cast_fp16, y = hidden_states_47_cast_fp16)[name = tensor("inputs_139_cast_fp16")]; tensor var_5197 = const()[name = tensor("op_5197"), val = tensor(3)]; tensor out_139_axes_0 = const()[name = tensor("out_139_axes_0"), val = tensor([1])]; tensor var_5222_to_fp16 = const()[name = tensor("op_5222_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_139_cast_fp16 = layer_norm(axes = out_139_axes_0, epsilon = var_5222_to_fp16, x = inputs_139_cast_fp16)[name = tensor("out_139_cast_fp16")]; tensor obj_323_gamma_0_to_fp16 = const()[name = tensor("obj_323_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(398739200)))]; tensor obj_323_beta_0_to_fp16 = const()[name = tensor("obj_323_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(398741312)))]; tensor obj_323_epsilon_0_to_fp16 = const()[name = tensor("obj_323_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_323_cast_fp16 = batch_norm(beta = obj_323_beta_0_to_fp16, epsilon = obj_323_epsilon_0_to_fp16, gamma = obj_323_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_139_cast_fp16)[name = tensor("obj_323_cast_fp16")]; tensor query_93_pad_type_0 = const()[name = tensor("query_93_pad_type_0"), val = tensor("valid")]; tensor query_93_strides_0 = const()[name = tensor("query_93_strides_0"), val = tensor([1, 1])]; tensor query_93_pad_0 = const()[name = tensor("query_93_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_93_dilations_0 = const()[name = tensor("query_93_dilations_0"), val = tensor([1, 1])]; tensor query_93_groups_0 = const()[name = tensor("query_93_groups_0"), val = tensor(1)]; tensor layers_23_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(398743424))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(399529920))), name = tensor("layers_23_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_23_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_23_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(399530112)))]; tensor query_93_cast_fp16 = conv(bias = layers_23_self_attn_q_proj_bias_to_fp16, dilations = query_93_dilations_0, groups = query_93_groups_0, pad = query_93_pad_0, pad_type = query_93_pad_type_0, strides = query_93_strides_0, weight = layers_23_self_attn_q_proj_weight_to_fp16_palettized, x = obj_323_cast_fp16)[name = tensor("query_93_cast_fp16")]; tensor current_key_pad_type_0 = const()[name = tensor("current_key_pad_type_0"), val = tensor("valid")]; tensor current_key_strides_0 = const()[name = tensor("current_key_strides_0"), val = tensor([1, 1])]; tensor current_key_pad_0 = const()[name = tensor("current_key_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_dilations_0 = const()[name = tensor("current_key_dilations_0"), val = tensor([1, 1])]; tensor current_key_groups_0 = const()[name = tensor("current_key_groups_0"), val = tensor(1)]; tensor layers_23_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(399532224))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(400318720))), name = tensor("layers_23_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor current_key_cast_fp16 = conv(dilations = current_key_dilations_0, groups = current_key_groups_0, pad = current_key_pad_0, pad_type = current_key_pad_type_0, strides = current_key_strides_0, weight = layers_23_self_attn_k_proj_weight_to_fp16_palettized, x = obj_323_cast_fp16)[name = tensor("current_key_cast_fp16")]; tensor current_value_pad_type_0 = const()[name = tensor("current_value_pad_type_0"), val = tensor("valid")]; tensor current_value_strides_0 = const()[name = tensor("current_value_strides_0"), val = tensor([1, 1])]; tensor current_value_pad_0 = const()[name = tensor("current_value_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_dilations_0 = const()[name = tensor("current_value_dilations_0"), val = tensor([1, 1])]; tensor current_value_groups_0 = const()[name = tensor("current_value_groups_0"), val = tensor(1)]; tensor layers_23_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(400318912))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(401105408))), name = tensor("layers_23_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_23_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_23_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(401105600)))]; tensor current_value_cast_fp16 = conv(bias = layers_23_self_attn_v_proj_bias_to_fp16, dilations = current_value_dilations_0, groups = current_value_groups_0, pad = current_value_pad_0, pad_type = current_value_pad_type_0, strides = current_value_strides_0, weight = layers_23_self_attn_v_proj_weight_to_fp16_palettized, x = obj_323_cast_fp16)[name = tensor("current_value_cast_fp16")]; tensor var_5261_cast_fp16 = mul(x = var_87_cast_fp16_23, y = var_207_cast_fp16)[name = tensor("op_5261_cast_fp16")]; tensor var_5262_cast_fp16 = mul(x = current_key_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_5262_cast_fp16")]; tensor key_93_cast_fp16 = add(x = var_5261_cast_fp16, y = var_5262_cast_fp16)[name = tensor("key_93_cast_fp16")]; tensor var_5265_cast_fp16 = mul(x = var_114_cast_fp16_23, y = var_207_cast_fp16)[name = tensor("op_5265_cast_fp16")]; tensor var_5266_cast_fp16 = mul(x = current_value_cast_fp16, y = var_205_cast_fp16)[name = tensor("op_5266_cast_fp16")]; tensor value_93_cast_fp16 = add(x = var_5265_cast_fp16, y = var_5266_cast_fp16)[name = tensor("value_93_cast_fp16")]; tensor var_5270 = const()[name = tensor("op_5270"), val = tensor([1, 16, 64, 1])]; tensor mh_q_93_cast_fp16 = reshape(shape = var_5270, x = query_93_cast_fp16)[name = tensor("mh_q_93_cast_fp16")]; tensor var_5272_to_fp16 = const()[name = tensor("op_5272_to_fp16"), val = tensor(0x1p-3)]; tensor var_5273_cast_fp16 = mul(x = mh_q_93_cast_fp16, y = var_5272_to_fp16)[name = tensor("op_5273_cast_fp16")]; tensor var_5276 = const()[name = tensor("op_5276"), val = tensor([1, 16, 64, 448])]; tensor var_5277_cast_fp16 = reshape(shape = var_5276, x = key_93_cast_fp16)[name = tensor("op_5277_cast_fp16")]; tensor mh_w_139_transpose_x_0 = const()[name = tensor("mh_w_139_transpose_x_0"), val = tensor(true)]; tensor mh_w_139_transpose_y_0 = const()[name = tensor("mh_w_139_transpose_y_0"), val = tensor(false)]; tensor mh_w_139_cast_fp16 = matmul(transpose_x = mh_w_139_transpose_x_0, transpose_y = mh_w_139_transpose_y_0, x = var_5273_cast_fp16, y = var_5277_cast_fp16)[name = tensor("mh_w_139_cast_fp16")]; tensor mh_w_141_cast_fp16 = add(x = mh_w_139_cast_fp16, y = var_229_cast_fp16)[name = tensor("mh_w_141_cast_fp16")]; tensor var_5285_cast_fp16 = softmax(axis = var_5197, x = mh_w_141_cast_fp16)[name = tensor("op_5285_cast_fp16")]; tensor var_5286 = const()[name = tensor("op_5286"), val = tensor([1, 16, 64, 448])]; tensor var_5287_cast_fp16 = reshape(shape = var_5286, x = value_93_cast_fp16)[name = tensor("op_5287_cast_fp16")]; tensor attn_93_transpose_x_0 = const()[name = tensor("attn_93_transpose_x_0"), val = tensor(false)]; tensor attn_93_transpose_y_0 = const()[name = tensor("attn_93_transpose_y_0"), val = tensor(true)]; tensor attn_93_cast_fp16 = matmul(transpose_x = attn_93_transpose_x_0, transpose_y = attn_93_transpose_y_0, x = var_5287_cast_fp16, y = var_5285_cast_fp16)[name = tensor("attn_93_cast_fp16")]; tensor var_5290 = const()[name = tensor("op_5290"), val = tensor([1, 1024, 1, 1])]; tensor input_231_cast_fp16 = reshape(shape = var_5290, x = attn_93_cast_fp16)[name = tensor("input_231_cast_fp16")]; tensor obj_329_pad_type_0 = const()[name = tensor("obj_329_pad_type_0"), val = tensor("valid")]; tensor obj_329_strides_0 = const()[name = tensor("obj_329_strides_0"), val = tensor([1, 1])]; tensor obj_329_pad_0 = const()[name = tensor("obj_329_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_329_dilations_0 = const()[name = tensor("obj_329_dilations_0"), val = tensor([1, 1])]; tensor obj_329_groups_0 = const()[name = tensor("obj_329_groups_0"), val = tensor(1)]; tensor layers_23_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(401107712))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(401894208))), name = tensor("layers_23_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_23_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_23_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(401894400)))]; tensor obj_329_cast_fp16 = conv(bias = layers_23_self_attn_o_proj_bias_to_fp16, dilations = obj_329_dilations_0, groups = obj_329_groups_0, pad = obj_329_pad_0, pad_type = obj_329_pad_type_0, strides = obj_329_strides_0, weight = layers_23_self_attn_o_proj_weight_to_fp16_palettized, x = input_231_cast_fp16)[name = tensor("obj_329_cast_fp16")]; tensor inputs_141_cast_fp16 = add(x = inputs_139_cast_fp16, y = obj_329_cast_fp16)[name = tensor("inputs_141_cast_fp16")]; tensor out_141_axes_0 = const()[name = tensor("out_141_axes_0"), val = tensor([1])]; tensor var_5312_to_fp16 = const()[name = tensor("op_5312_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_141_cast_fp16 = layer_norm(axes = out_141_axes_0, epsilon = var_5312_to_fp16, x = inputs_141_cast_fp16)[name = tensor("out_141_cast_fp16")]; tensor obj_331_gamma_0_to_fp16 = const()[name = tensor("obj_331_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(401896512)))]; tensor obj_331_beta_0_to_fp16 = const()[name = tensor("obj_331_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(401898624)))]; tensor obj_331_epsilon_0_to_fp16 = const()[name = tensor("obj_331_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_331_cast_fp16 = batch_norm(beta = obj_331_beta_0_to_fp16, epsilon = obj_331_epsilon_0_to_fp16, gamma = obj_331_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_141_cast_fp16)[name = tensor("obj_331_cast_fp16")]; tensor query_pad_type_0 = const()[name = tensor("query_pad_type_0"), val = tensor("valid")]; tensor query_strides_0 = const()[name = tensor("query_strides_0"), val = tensor([1, 1])]; tensor query_pad_0 = const()[name = tensor("query_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_dilations_0 = const()[name = tensor("query_dilations_0"), val = tensor([1, 1])]; tensor query_groups_0 = const()[name = tensor("query_groups_0"), val = tensor(1)]; tensor layers_23_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(401900736))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(402687232))), name = tensor("layers_23_encoder_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_23_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_23_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(402687424)))]; tensor query_cast_fp16 = conv(bias = layers_23_encoder_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_23_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_331_cast_fp16)[name = tensor("query_cast_fp16")]; tensor key_pad_type_0 = const()[name = tensor("key_pad_type_0"), val = tensor("valid")]; tensor key_strides_0 = const()[name = tensor("key_strides_0"), val = tensor([1, 1])]; tensor key_pad_0 = const()[name = tensor("key_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_dilations_0 = const()[name = tensor("key_dilations_0"), val = tensor([1, 1])]; tensor key_groups_0 = const()[name = tensor("key_groups_0"), val = tensor(1)]; tensor layers_23_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(402689536))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(403476032))), name = tensor("layers_23_encoder_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor key_cast_fp16 = conv(dilations = key_dilations_0, groups = key_groups_0, pad = key_pad_0, pad_type = key_pad_type_0, strides = key_strides_0, weight = layers_23_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("key_cast_fp16")]; tensor value_pad_type_0 = const()[name = tensor("value_pad_type_0"), val = tensor("valid")]; tensor value_strides_0 = const()[name = tensor("value_strides_0"), val = tensor([1, 1])]; tensor value_pad_0 = const()[name = tensor("value_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_dilations_0 = const()[name = tensor("value_dilations_0"), val = tensor([1, 1])]; tensor value_groups_0 = const()[name = tensor("value_groups_0"), val = tensor(1)]; tensor layers_23_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(403476224))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(404262720))), name = tensor("layers_23_encoder_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_23_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_23_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(404262912)))]; tensor value_cast_fp16 = conv(bias = layers_23_encoder_attn_v_proj_bias_to_fp16, dilations = value_dilations_0, groups = value_groups_0, pad = value_pad_0, pad_type = value_pad_type_0, strides = value_strides_0, weight = layers_23_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("value_cast_fp16")]; tensor var_5348 = const()[name = tensor("op_5348"), val = tensor([1, 16, 64, 1])]; tensor mh_q_cast_fp16 = reshape(shape = var_5348, x = query_cast_fp16)[name = tensor("mh_q_cast_fp16")]; tensor var_5350_to_fp16 = const()[name = tensor("op_5350_to_fp16"), val = tensor(0x1p-3)]; tensor var_5351_cast_fp16 = mul(x = mh_q_cast_fp16, y = var_5350_to_fp16)[name = tensor("op_5351_cast_fp16")]; tensor var_5354 = const()[name = tensor("op_5354"), val = tensor([1, 16, 64, 1500])]; tensor var_5355_cast_fp16 = reshape(shape = var_5354, x = key_cast_fp16)[name = tensor("op_5355_cast_fp16")]; tensor mh_w_transpose_x_0 = const()[name = tensor("mh_w_transpose_x_0"), val = tensor(true)]; tensor mh_w_transpose_y_0 = const()[name = tensor("mh_w_transpose_y_0"), val = tensor(false)]; tensor mh_w_cast_fp16 = matmul(transpose_x = mh_w_transpose_x_0, transpose_y = mh_w_transpose_y_0, x = var_5351_cast_fp16, y = var_5355_cast_fp16)[name = tensor("mh_w_cast_fp16")]; tensor obj_335_cast_fp16 = softmax(axis = var_5197, x = mh_w_cast_fp16)[name = tensor("obj_335_cast_fp16")]; tensor var_5359 = const()[name = tensor("op_5359"), val = tensor([1, 16, 64, 1500])]; tensor var_5360_cast_fp16 = reshape(shape = var_5359, x = value_cast_fp16)[name = tensor("op_5360_cast_fp16")]; tensor attn_transpose_x_0 = const()[name = tensor("attn_transpose_x_0"), val = tensor(false)]; tensor attn_transpose_y_0 = const()[name = tensor("attn_transpose_y_0"), val = tensor(true)]; tensor attn_cast_fp16 = matmul(transpose_x = attn_transpose_x_0, transpose_y = attn_transpose_y_0, x = var_5360_cast_fp16, y = obj_335_cast_fp16)[name = tensor("attn_cast_fp16")]; tensor var_5363 = const()[name = tensor("op_5363"), val = tensor([1, 1024, 1, 1])]; tensor input_233_cast_fp16 = reshape(shape = var_5363, x = attn_cast_fp16)[name = tensor("input_233_cast_fp16")]; tensor obj_333_pad_type_0 = const()[name = tensor("obj_333_pad_type_0"), val = tensor("valid")]; tensor obj_333_strides_0 = const()[name = tensor("obj_333_strides_0"), val = tensor([1, 1])]; tensor obj_333_pad_0 = const()[name = tensor("obj_333_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_333_dilations_0 = const()[name = tensor("obj_333_dilations_0"), val = tensor([1, 1])]; tensor obj_333_groups_0 = const()[name = tensor("obj_333_groups_0"), val = tensor(1)]; tensor layers_23_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(404265024))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(405051520))), name = tensor("layers_23_encoder_attn_o_proj_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; tensor layers_23_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_23_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(405051712)))]; tensor obj_333_cast_fp16 = conv(bias = layers_23_encoder_attn_o_proj_bias_to_fp16, dilations = obj_333_dilations_0, groups = obj_333_groups_0, pad = obj_333_pad_0, pad_type = obj_333_pad_type_0, strides = obj_333_strides_0, weight = layers_23_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_233_cast_fp16)[name = tensor("obj_333_cast_fp16")]; tensor inputs_143_cast_fp16 = add(x = inputs_141_cast_fp16, y = obj_333_cast_fp16)[name = tensor("inputs_143_cast_fp16")]; tensor out_143_axes_0 = const()[name = tensor("out_143_axes_0"), val = tensor([1])]; tensor var_5384_to_fp16 = const()[name = tensor("op_5384_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_143_cast_fp16 = layer_norm(axes = out_143_axes_0, epsilon = var_5384_to_fp16, x = inputs_143_cast_fp16)[name = tensor("out_143_cast_fp16")]; tensor input_235_gamma_0_to_fp16 = const()[name = tensor("input_235_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(405053824)))]; tensor input_235_beta_0_to_fp16 = const()[name = tensor("input_235_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(405055936)))]; tensor input_235_epsilon_0_to_fp16 = const()[name = tensor("input_235_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_235_cast_fp16 = batch_norm(beta = input_235_beta_0_to_fp16, epsilon = input_235_epsilon_0_to_fp16, gamma = input_235_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_143_cast_fp16)[name = tensor("input_235_cast_fp16")]; tensor input_237_pad_type_0 = const()[name = tensor("input_237_pad_type_0"), val = tensor("valid")]; tensor input_237_strides_0 = const()[name = tensor("input_237_strides_0"), val = tensor([1, 1])]; tensor input_237_pad_0 = const()[name = tensor("input_237_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_237_dilations_0 = const()[name = tensor("input_237_dilations_0"), val = tensor([1, 1])]; tensor input_237_groups_0 = const()[name = tensor("input_237_groups_0"), val = tensor(1)]; tensor layers_23_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(405058048))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(408203840))), name = tensor("layers_23_fc1_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; tensor layers_23_fc1_bias_to_fp16 = const()[name = tensor("layers_23_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(408204032)))]; tensor input_237_cast_fp16 = conv(bias = layers_23_fc1_bias_to_fp16, dilations = input_237_dilations_0, groups = input_237_groups_0, pad = input_237_pad_0, pad_type = input_237_pad_type_0, strides = input_237_strides_0, weight = layers_23_fc1_weight_to_fp16_palettized, x = input_235_cast_fp16)[name = tensor("input_237_cast_fp16")]; tensor input_mode_0 = const()[name = tensor("input_mode_0"), val = tensor("EXACT")]; tensor input_cast_fp16 = gelu(mode = input_mode_0, x = input_237_cast_fp16)[name = tensor("input_cast_fp16")]; tensor hidden_states_49_pad_type_0 = const()[name = tensor("hidden_states_49_pad_type_0"), val = tensor("valid")]; tensor hidden_states_49_strides_0 = const()[name = tensor("hidden_states_49_strides_0"), val = tensor([1, 1])]; tensor hidden_states_49_pad_0 = const()[name = tensor("hidden_states_49_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_49_dilations_0 = const()[name = tensor("hidden_states_49_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_49_groups_0 = const()[name = tensor("hidden_states_49_groups_0"), val = tensor(1)]; tensor layers_23_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(408212288))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411358080))), name = tensor("layers_23_fc2_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; tensor layers_23_fc2_bias_to_fp16 = const()[name = tensor("layers_23_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411358272)))]; tensor hidden_states_49_cast_fp16 = conv(bias = layers_23_fc2_bias_to_fp16, dilations = hidden_states_49_dilations_0, groups = hidden_states_49_groups_0, pad = hidden_states_49_pad_0, pad_type = hidden_states_49_pad_type_0, strides = hidden_states_49_strides_0, weight = layers_23_fc2_weight_to_fp16_palettized, x = input_cast_fp16)[name = tensor("hidden_states_49_cast_fp16")]; tensor inputs_cast_fp16 = add(x = inputs_143_cast_fp16, y = hidden_states_49_cast_fp16)[name = tensor("inputs_cast_fp16")]; tensor out_axes_0 = const()[name = tensor("out_axes_0"), val = tensor([1])]; tensor var_5427_to_fp16 = const()[name = tensor("op_5427_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_5427_to_fp16, x = inputs_cast_fp16)[name = tensor("out_cast_fp16")]; tensor hidden_states_gamma_0_to_fp16 = const()[name = tensor("hidden_states_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411360384)))]; tensor hidden_states_beta_0_to_fp16 = const()[name = tensor("hidden_states_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411362496)))]; tensor hidden_states_epsilon_0_to_fp16 = const()[name = tensor("hidden_states_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor hidden_states_cast_fp16 = batch_norm(beta = hidden_states_beta_0_to_fp16, epsilon = hidden_states_epsilon_0_to_fp16, gamma = hidden_states_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_cast_fp16)[name = tensor("hidden_states_cast_fp16")]; tensor var_5438_axes_0 = const()[name = tensor("op_5438_axes_0"), val = tensor([2])]; tensor var_5438_cast_fp16 = squeeze(axes = var_5438_axes_0, x = hidden_states_cast_fp16)[name = tensor("op_5438_cast_fp16")]; tensor var_5441_perm_0 = const()[name = tensor("op_5441_perm_0"), val = tensor([0, 2, 1])]; tensor linear_0_bias_0_to_fp16 = const()[name = tensor("linear_0_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411364608)))]; tensor var_5441_cast_fp16 = transpose(perm = var_5441_perm_0, x = var_5438_cast_fp16)[name = tensor("transpose_0")]; tensor logits = linear(bias = linear_0_bias_0_to_fp16, weight = embed_tokens_weight_to_fp16, x = var_5441_cast_fp16)[name = tensor("linear_0_cast_fp16")]; tensor var_5445 = const()[name = tensor("op_5445"), val = tensor(1)]; tensor obj_339_interleave_0 = const()[name = tensor("obj_339_interleave_0"), val = tensor(false)]; tensor key_cache_updates = concat(axis = var_5445, interleave = obj_339_interleave_0, values = (current_key_1_cast_fp16, current_key_3_cast_fp16, current_key_5_cast_fp16, current_key_7_cast_fp16, current_key_9_cast_fp16, current_key_11_cast_fp16, current_key_13_cast_fp16, current_key_15_cast_fp16, current_key_17_cast_fp16, current_key_19_cast_fp16, current_key_21_cast_fp16, current_key_23_cast_fp16, current_key_25_cast_fp16, current_key_27_cast_fp16, current_key_29_cast_fp16, current_key_31_cast_fp16, current_key_33_cast_fp16, current_key_35_cast_fp16, current_key_37_cast_fp16, current_key_39_cast_fp16, current_key_41_cast_fp16, current_key_43_cast_fp16, current_key_45_cast_fp16, current_key_cast_fp16))[name = tensor("obj_339_cast_fp16")]; tensor var_5448 = const()[name = tensor("op_5448"), val = tensor(1)]; tensor obj_341_interleave_0 = const()[name = tensor("obj_341_interleave_0"), val = tensor(false)]; tensor value_cache_updates = concat(axis = var_5448, interleave = obj_341_interleave_0, values = (current_value_1_cast_fp16, current_value_3_cast_fp16, current_value_5_cast_fp16, current_value_7_cast_fp16, current_value_9_cast_fp16, current_value_11_cast_fp16, current_value_13_cast_fp16, current_value_15_cast_fp16, current_value_17_cast_fp16, current_value_19_cast_fp16, current_value_21_cast_fp16, current_value_23_cast_fp16, current_value_25_cast_fp16, current_value_27_cast_fp16, current_value_29_cast_fp16, current_value_31_cast_fp16, current_value_33_cast_fp16, current_value_35_cast_fp16, current_value_37_cast_fp16, current_value_39_cast_fp16, current_value_41_cast_fp16, current_value_43_cast_fp16, current_value_45_cast_fp16, current_value_cast_fp16))[name = tensor("obj_341_cast_fp16")]; tensor var_5459_begin_0 = const()[name = tensor("op_5459_begin_0"), val = tensor([0, 15, 0, 0])]; tensor var_5459_end_0 = const()[name = tensor("op_5459_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_5459_end_mask_0 = const()[name = tensor("op_5459_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_5459_cast_fp16 = slice_by_index(begin = var_5459_begin_0, end = var_5459_end_0, end_mask = var_5459_end_mask_0, x = obj_195_cast_fp16)[name = tensor("op_5459_cast_fp16")]; tensor var_5462_begin_0 = const()[name = tensor("op_5462_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5462_end_0 = const()[name = tensor("op_5462_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_5462_end_mask_0 = const()[name = tensor("op_5462_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_5462_squeeze_mask_0 = const()[name = tensor("op_5462_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_5462_cast_fp16 = slice_by_index(begin = var_5462_begin_0, end = var_5462_end_0, end_mask = var_5462_end_mask_0, squeeze_mask = var_5462_squeeze_mask_0, x = var_5459_cast_fp16)[name = tensor("op_5462_cast_fp16")]; tensor var_5477_begin_0 = const()[name = tensor("op_5477_begin_0"), val = tensor([0, 4, 0, 0])]; tensor var_5477_end_0 = const()[name = tensor("op_5477_end_0"), val = tensor([1, 5, 1, 1500])]; tensor var_5477_end_mask_0 = const()[name = tensor("op_5477_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5477_cast_fp16 = slice_by_index(begin = var_5477_begin_0, end = var_5477_end_0, end_mask = var_5477_end_mask_0, x = obj_223_cast_fp16)[name = tensor("op_5477_cast_fp16")]; tensor var_5480_begin_0 = const()[name = tensor("op_5480_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5480_end_0 = const()[name = tensor("op_5480_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_5480_end_mask_0 = const()[name = tensor("op_5480_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_5480_squeeze_mask_0 = const()[name = tensor("op_5480_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_5480_cast_fp16 = slice_by_index(begin = var_5480_begin_0, end = var_5480_end_0, end_mask = var_5480_end_mask_0, squeeze_mask = var_5480_squeeze_mask_0, x = var_5477_cast_fp16)[name = tensor("op_5480_cast_fp16")]; tensor var_5495_begin_0 = const()[name = tensor("op_5495_begin_0"), val = tensor([0, 15, 0, 0])]; tensor var_5495_end_0 = const()[name = tensor("op_5495_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_5495_end_mask_0 = const()[name = tensor("op_5495_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_5495_cast_fp16 = slice_by_index(begin = var_5495_begin_0, end = var_5495_end_0, end_mask = var_5495_end_mask_0, x = obj_223_cast_fp16)[name = tensor("op_5495_cast_fp16")]; tensor var_5498_begin_0 = const()[name = tensor("op_5498_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5498_end_0 = const()[name = tensor("op_5498_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_5498_end_mask_0 = const()[name = tensor("op_5498_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_5498_squeeze_mask_0 = const()[name = tensor("op_5498_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_5498_cast_fp16 = slice_by_index(begin = var_5498_begin_0, end = var_5498_end_0, end_mask = var_5498_end_mask_0, squeeze_mask = var_5498_squeeze_mask_0, x = var_5495_cast_fp16)[name = tensor("op_5498_cast_fp16")]; tensor var_5513_begin_0 = const()[name = tensor("op_5513_begin_0"), val = tensor([0, 1, 0, 0])]; tensor var_5513_end_0 = const()[name = tensor("op_5513_end_0"), val = tensor([1, 2, 1, 1500])]; tensor var_5513_end_mask_0 = const()[name = tensor("op_5513_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5513_cast_fp16 = slice_by_index(begin = var_5513_begin_0, end = var_5513_end_0, end_mask = var_5513_end_mask_0, x = obj_237_cast_fp16)[name = tensor("op_5513_cast_fp16")]; tensor var_5516_begin_0 = const()[name = tensor("op_5516_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5516_end_0 = const()[name = tensor("op_5516_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_5516_end_mask_0 = const()[name = tensor("op_5516_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_5516_squeeze_mask_0 = const()[name = tensor("op_5516_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_5516_cast_fp16 = slice_by_index(begin = var_5516_begin_0, end = var_5516_end_0, end_mask = var_5516_end_mask_0, squeeze_mask = var_5516_squeeze_mask_0, x = var_5513_cast_fp16)[name = tensor("op_5516_cast_fp16")]; tensor var_5531_begin_0 = const()[name = tensor("op_5531_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5531_end_0 = const()[name = tensor("op_5531_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_5531_end_mask_0 = const()[name = tensor("op_5531_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5531_cast_fp16 = slice_by_index(begin = var_5531_begin_0, end = var_5531_end_0, end_mask = var_5531_end_mask_0, x = obj_293_cast_fp16)[name = tensor("op_5531_cast_fp16")]; tensor var_5534_begin_0 = const()[name = tensor("op_5534_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5534_end_0 = const()[name = tensor("op_5534_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_5534_end_mask_0 = const()[name = tensor("op_5534_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_5534_squeeze_mask_0 = const()[name = tensor("op_5534_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_5534_cast_fp16 = slice_by_index(begin = var_5534_begin_0, end = var_5534_end_0, end_mask = var_5534_end_mask_0, squeeze_mask = var_5534_squeeze_mask_0, x = var_5531_cast_fp16)[name = tensor("op_5534_cast_fp16")]; tensor var_5549_begin_0 = const()[name = tensor("op_5549_begin_0"), val = tensor([0, 4, 0, 0])]; tensor var_5549_end_0 = const()[name = tensor("op_5549_end_0"), val = tensor([1, 5, 1, 1500])]; tensor var_5549_end_mask_0 = const()[name = tensor("op_5549_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5549_cast_fp16 = slice_by_index(begin = var_5549_begin_0, end = var_5549_end_0, end_mask = var_5549_end_mask_0, x = obj_335_cast_fp16)[name = tensor("op_5549_cast_fp16")]; tensor var_5552_begin_0 = const()[name = tensor("op_5552_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5552_end_0 = const()[name = tensor("op_5552_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_5552_end_mask_0 = const()[name = tensor("op_5552_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_5552_squeeze_mask_0 = const()[name = tensor("op_5552_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_5552_cast_fp16 = slice_by_index(begin = var_5552_begin_0, end = var_5552_end_0, end_mask = var_5552_end_mask_0, squeeze_mask = var_5552_squeeze_mask_0, x = var_5549_cast_fp16)[name = tensor("op_5552_cast_fp16")]; tensor var_5559 = const()[name = tensor("op_5559"), val = tensor(1)]; tensor var_5560_interleave_0 = const()[name = tensor("op_5560_interleave_0"), val = tensor(false)]; tensor var_5560_cast_fp16 = concat(axis = var_5559, interleave = var_5560_interleave_0, values = (var_5462_cast_fp16, var_5480_cast_fp16, var_5498_cast_fp16, var_5516_cast_fp16, var_5534_cast_fp16, var_5552_cast_fp16))[name = tensor("op_5560_cast_fp16")]; tensor obj_axes_0 = const()[name = tensor("obj_axes_0"), val = tensor([1])]; tensor obj_keep_dims_0 = const()[name = tensor("obj_keep_dims_0"), val = tensor(false)]; tensor alignment_heads_weights = reduce_mean(axes = obj_axes_0, keep_dims = obj_keep_dims_0, x = var_5560_cast_fp16)[name = tensor("obj_cast_fp16")]; } -> (logits, key_cache_updates, value_cache_updates, alignment_heads_weights); }