diff --git "a/qwen3_tts/speech_decoder/12hz-0.6b-customvoice/W8A16/SpeechDecoder.mlmodelc/model.mil" "b/qwen3_tts/speech_decoder/12hz-0.6b-customvoice/W8A16/SpeechDecoder.mlmodelc/model.mil" new file mode 100644--- /dev/null +++ "b/qwen3_tts/speech_decoder/12hz-0.6b-customvoice/W8A16/SpeechDecoder.mlmodelc/model.mil" @@ -0,0 +1,2109 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3510.2.1"}, {"coremlc-version", "3500.32.1"}})] +{ + func main(tensor audio_codes, tensor cache_length, tensor hidden_context, tensor key_cache, tensor key_padding_mask, tensor kv_cache_update_mask, tensor value_cache) { + int32 var_28 = const()[name = string("op_28"), val = int32(-1)]; + int32 var_32 = const()[name = string("op_32"), val = int32(1)]; + tensor codes_1_begin_0 = const()[name = string("codes_1_begin_0"), val = tensor([0, 0, 0])]; + tensor codes_1_end_0 = const()[name = string("codes_1_end_0"), val = tensor([1, 1, 1])]; + tensor codes_1_end_mask_0 = const()[name = string("codes_1_end_mask_0"), val = tensor([true, false, true])]; + tensor codes_1 = slice_by_index(begin = codes_1_begin_0, end = codes_1_end_0, end_mask = codes_1_end_mask_0, x = audio_codes)[name = string("codes_1")]; + tensor var_295 = const()[name = string("op_295"), val = tensor([1, 0, 2])]; + tensor squeeze_0_axes_0 = const()[name = string("squeeze_0_axes_0"), val = tensor([0])]; + string var_296_to_int16_dtype_0 = const()[name = string("op_296_to_int16_dtype_0"), val = string("int16")]; + tensor var_296 = transpose(perm = var_295, x = codes_1)[name = string("transpose_29")]; + tensor var_296_to_int16 = cast(dtype = var_296_to_int16_dtype_0, x = var_296)[name = string("cast_18")]; + tensor squeeze_0_cast_uint16 = squeeze(axes = squeeze_0_axes_0, x = var_296_to_int16)[name = string("squeeze_0_cast_uint16")]; + int32 quantized_1_batch_dims_0 = const()[name = string("quantized_1_batch_dims_0"), val = int32(0)]; + bool quantized_1_validate_indices_0 = const()[name = string("quantized_1_validate_indices_0"), val = bool(false)]; + tensor weight_1_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(524416))))[name = string("weight_1_to_fp16_palettized")]; + string cast_216_dtype_0 = const()[name = string("cast_216_dtype_0"), val = string("int32")]; + int32 greater_equal_0_y_0 = const()[name = string("greater_equal_0_y_0"), val = int32(0)]; + tensor cast_216 = cast(dtype = cast_216_dtype_0, x = squeeze_0_cast_uint16)[name = string("cast_17")]; + tensor greater_equal_0 = greater_equal(x = cast_216, y = greater_equal_0_y_0)[name = string("greater_equal_0")]; + int32 slice_by_index_0 = const()[name = string("slice_by_index_0"), val = int32(2048)]; + tensor add_0 = add(x = cast_216, y = slice_by_index_0)[name = string("add_0")]; + tensor select_0 = select(a = cast_216, b = add_0, cond = greater_equal_0)[name = string("select_0")]; + int32 quantized_1_cast_fp16_cast_uint16_axis_0 = const()[name = string("quantized_1_cast_fp16_cast_uint16_axis_0"), val = int32(0)]; + string select_0_to_uint16_dtype_0 = const()[name = string("select_0_to_uint16_dtype_0"), val = string("uint16")]; + tensor select_0_to_uint16 = cast(dtype = select_0_to_uint16_dtype_0, x = select_0)[name = string("cast_16")]; + tensor quantized_1_cast_fp16_cast_uint16_cast_uint16 = gather(axis = quantized_1_cast_fp16_cast_uint16_axis_0, batch_dims = quantized_1_batch_dims_0, indices = select_0_to_uint16, validate_indices = quantized_1_validate_indices_0, x = weight_1_to_fp16_palettized)[name = string("quantized_1_cast_fp16_cast_uint16_cast_uint16")]; + tensor var_304 = const()[name = string("op_304"), val = tensor([0, 2, 1])]; + tensor input_3_axes_0 = const()[name = string("input_3_axes_0"), val = tensor([2])]; + tensor var_305_cast_fp16 = transpose(perm = var_304, x = quantized_1_cast_fp16_cast_uint16_cast_uint16)[name = string("transpose_28")]; + tensor input_3_cast_fp16 = expand_dims(axes = input_3_axes_0, x = var_305_cast_fp16)[name = string("input_3_cast_fp16")]; + string quantized_pad_type_0 = const()[name = string("quantized_pad_type_0"), val = string("valid")]; + tensor quantized_strides_0 = const()[name = string("quantized_strides_0"), val = tensor([1, 1])]; + tensor quantized_pad_0 = const()[name = string("quantized_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor quantized_dilations_0 = const()[name = string("quantized_dilations_0"), val = tensor([1, 1])]; + int32 quantized_groups_0 = const()[name = string("quantized_groups_0"), val = int32(1)]; + tensor decoder_quantizer_rvq_first_output_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(524992))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(656128))))[name = string("decoder_quantizer_rvq_first_output_proj_weight_to_fp16_palettized")]; + tensor quantized_cast_fp16 = conv(dilations = quantized_dilations_0, groups = quantized_groups_0, pad = quantized_pad_0, pad_type = quantized_pad_type_0, strides = quantized_strides_0, weight = decoder_quantizer_rvq_first_output_proj_weight_to_fp16_palettized, x = input_3_cast_fp16)[name = string("quantized_cast_fp16")]; + tensor codes_begin_0 = const()[name = string("codes_begin_0"), val = tensor([0, 1, 0])]; + tensor codes_end_0 = const()[name = string("codes_end_0"), val = tensor([1, 16, 1])]; + tensor codes_end_mask_0 = const()[name = string("codes_end_mask_0"), val = tensor([true, true, true])]; + tensor codes = slice_by_index(begin = codes_begin_0, end = codes_end_0, end_mask = codes_end_mask_0, x = audio_codes)[name = string("codes")]; + tensor var_315 = const()[name = string("op_315"), val = tensor([1, 0, 2])]; + tensor var_317_split_sizes_0 = const()[name = string("op_317_split_sizes_0"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(656704)))]; + int32 var_317_axis_0 = const()[name = string("op_317_axis_0"), val = int32(0)]; + tensor var_316 = transpose(perm = var_315, x = codes)[name = string("transpose_27")]; + tensor var_317_0, tensor var_317_1, tensor var_317_2, tensor var_317_3, tensor var_317_4, tensor var_317_5, tensor var_317_6, tensor var_317_7, tensor var_317_8, tensor var_317_9, tensor var_317_10, tensor var_317_11, tensor var_317_12, tensor var_317_13, tensor var_317_14 = split(axis = var_317_axis_0, split_sizes = var_317_split_sizes_0, x = var_316)[name = string("op_317")]; + tensor squeeze_1_axes_0 = const()[name = string("squeeze_1_axes_0"), val = tensor([0])]; + string var_317_0_to_uint16_dtype_0 = const()[name = string("op_317_0_to_uint16_dtype_0"), val = string("uint16")]; + tensor var_317_0_to_uint16 = cast(dtype = var_317_0_to_uint16_dtype_0, x = var_317_0)[name = string("cast_15")]; + tensor squeeze_1_cast_uint16 = squeeze(axes = squeeze_1_axes_0, x = var_317_0_to_uint16)[name = string("squeeze_1_cast_uint16")]; + tensor squeeze_2_axes_0 = const()[name = string("squeeze_2_axes_0"), val = tensor([0])]; + string var_317_1_to_uint16_dtype_0 = const()[name = string("op_317_1_to_uint16_dtype_0"), val = string("uint16")]; + tensor var_317_1_to_uint16 = cast(dtype = var_317_1_to_uint16_dtype_0, x = var_317_1)[name = string("cast_14")]; + tensor squeeze_2_cast_uint16 = squeeze(axes = squeeze_2_axes_0, x = var_317_1_to_uint16)[name = string("squeeze_2_cast_uint16")]; + tensor squeeze_3_axes_0 = const()[name = string("squeeze_3_axes_0"), val = tensor([0])]; + string var_317_2_to_uint16_dtype_0 = const()[name = string("op_317_2_to_uint16_dtype_0"), val = string("uint16")]; + tensor var_317_2_to_uint16 = cast(dtype = var_317_2_to_uint16_dtype_0, x = var_317_2)[name = string("cast_13")]; + tensor squeeze_3_cast_uint16 = squeeze(axes = squeeze_3_axes_0, x = var_317_2_to_uint16)[name = string("squeeze_3_cast_uint16")]; + tensor squeeze_4_axes_0 = const()[name = string("squeeze_4_axes_0"), val = tensor([0])]; + string var_317_3_to_uint16_dtype_0 = const()[name = string("op_317_3_to_uint16_dtype_0"), val = string("uint16")]; + tensor var_317_3_to_uint16 = cast(dtype = var_317_3_to_uint16_dtype_0, x = var_317_3)[name = string("cast_12")]; + tensor squeeze_4_cast_uint16 = squeeze(axes = squeeze_4_axes_0, x = var_317_3_to_uint16)[name = string("squeeze_4_cast_uint16")]; + tensor squeeze_5_axes_0 = const()[name = string("squeeze_5_axes_0"), val = tensor([0])]; + string var_317_4_to_uint16_dtype_0 = const()[name = string("op_317_4_to_uint16_dtype_0"), val = string("uint16")]; + tensor var_317_4_to_uint16 = cast(dtype = var_317_4_to_uint16_dtype_0, x = var_317_4)[name = string("cast_11")]; + tensor squeeze_5_cast_uint16 = squeeze(axes = squeeze_5_axes_0, x = var_317_4_to_uint16)[name = string("squeeze_5_cast_uint16")]; + tensor squeeze_6_axes_0 = const()[name = string("squeeze_6_axes_0"), val = tensor([0])]; + string var_317_5_to_uint16_dtype_0 = const()[name = string("op_317_5_to_uint16_dtype_0"), val = string("uint16")]; + tensor var_317_5_to_uint16 = cast(dtype = var_317_5_to_uint16_dtype_0, x = var_317_5)[name = string("cast_10")]; + tensor squeeze_6_cast_uint16 = squeeze(axes = squeeze_6_axes_0, x = var_317_5_to_uint16)[name = string("squeeze_6_cast_uint16")]; + tensor squeeze_7_axes_0 = const()[name = string("squeeze_7_axes_0"), val = tensor([0])]; + string var_317_6_to_uint16_dtype_0 = const()[name = string("op_317_6_to_uint16_dtype_0"), val = string("uint16")]; + tensor var_317_6_to_uint16 = cast(dtype = var_317_6_to_uint16_dtype_0, x = var_317_6)[name = string("cast_9")]; + tensor squeeze_7_cast_uint16 = squeeze(axes = squeeze_7_axes_0, x = var_317_6_to_uint16)[name = string("squeeze_7_cast_uint16")]; + tensor squeeze_8_axes_0 = const()[name = string("squeeze_8_axes_0"), val = tensor([0])]; + string var_317_7_to_uint16_dtype_0 = const()[name = string("op_317_7_to_uint16_dtype_0"), val = string("uint16")]; + tensor var_317_7_to_uint16 = cast(dtype = var_317_7_to_uint16_dtype_0, x = var_317_7)[name = string("cast_8")]; + tensor squeeze_8_cast_uint16 = squeeze(axes = squeeze_8_axes_0, x = var_317_7_to_uint16)[name = string("squeeze_8_cast_uint16")]; + tensor squeeze_9_axes_0 = const()[name = string("squeeze_9_axes_0"), val = tensor([0])]; + string var_317_8_to_uint16_dtype_0 = const()[name = string("op_317_8_to_uint16_dtype_0"), val = string("uint16")]; + tensor var_317_8_to_uint16 = cast(dtype = var_317_8_to_uint16_dtype_0, x = var_317_8)[name = string("cast_7")]; + tensor squeeze_9_cast_uint16 = squeeze(axes = squeeze_9_axes_0, x = var_317_8_to_uint16)[name = string("squeeze_9_cast_uint16")]; + tensor squeeze_10_axes_0 = const()[name = string("squeeze_10_axes_0"), val = tensor([0])]; + string var_317_9_to_uint16_dtype_0 = const()[name = string("op_317_9_to_uint16_dtype_0"), val = string("uint16")]; + tensor var_317_9_to_uint16 = cast(dtype = var_317_9_to_uint16_dtype_0, x = var_317_9)[name = string("cast_6")]; + tensor squeeze_10_cast_uint16 = squeeze(axes = squeeze_10_axes_0, x = var_317_9_to_uint16)[name = string("squeeze_10_cast_uint16")]; + tensor squeeze_11_axes_0 = const()[name = string("squeeze_11_axes_0"), val = tensor([0])]; + string var_317_10_to_uint16_dtype_0 = const()[name = string("op_317_10_to_uint16_dtype_0"), val = string("uint16")]; + tensor var_317_10_to_uint16 = cast(dtype = var_317_10_to_uint16_dtype_0, x = var_317_10)[name = string("cast_5")]; + tensor squeeze_11_cast_uint16 = squeeze(axes = squeeze_11_axes_0, x = var_317_10_to_uint16)[name = string("squeeze_11_cast_uint16")]; + tensor squeeze_12_axes_0 = const()[name = string("squeeze_12_axes_0"), val = tensor([0])]; + string var_317_11_to_uint16_dtype_0 = const()[name = string("op_317_11_to_uint16_dtype_0"), val = string("uint16")]; + tensor var_317_11_to_uint16 = cast(dtype = var_317_11_to_uint16_dtype_0, x = var_317_11)[name = string("cast_4")]; + tensor squeeze_12_cast_uint16 = squeeze(axes = squeeze_12_axes_0, x = var_317_11_to_uint16)[name = string("squeeze_12_cast_uint16")]; + tensor squeeze_13_axes_0 = const()[name = string("squeeze_13_axes_0"), val = tensor([0])]; + string var_317_12_to_uint16_dtype_0 = const()[name = string("op_317_12_to_uint16_dtype_0"), val = string("uint16")]; + tensor var_317_12_to_uint16 = cast(dtype = var_317_12_to_uint16_dtype_0, x = var_317_12)[name = string("cast_3")]; + tensor squeeze_13_cast_uint16 = squeeze(axes = squeeze_13_axes_0, x = var_317_12_to_uint16)[name = string("squeeze_13_cast_uint16")]; + tensor squeeze_14_axes_0 = const()[name = string("squeeze_14_axes_0"), val = tensor([0])]; + string var_317_13_to_uint16_dtype_0 = const()[name = string("op_317_13_to_uint16_dtype_0"), val = string("uint16")]; + tensor var_317_13_to_uint16 = cast(dtype = var_317_13_to_uint16_dtype_0, x = var_317_13)[name = string("cast_2")]; + tensor squeeze_14_cast_uint16 = squeeze(axes = squeeze_14_axes_0, x = var_317_13_to_uint16)[name = string("squeeze_14_cast_uint16")]; + tensor squeeze_15_axes_0 = const()[name = string("squeeze_15_axes_0"), val = tensor([0])]; + string var_317_14_to_uint16_dtype_0 = const()[name = string("op_317_14_to_uint16_dtype_0"), val = string("uint16")]; + tensor var_317_14_to_uint16 = cast(dtype = var_317_14_to_uint16_dtype_0, x = var_317_14)[name = string("cast_1")]; + tensor squeeze_15_cast_uint16 = squeeze(axes = squeeze_15_axes_0, x = var_317_14_to_uint16)[name = string("squeeze_15_cast_uint16")]; + int32 quantized_3_axis_0 = const()[name = string("quantized_3_axis_0"), val = int32(0)]; + int32 quantized_3_batch_dims_0 = const()[name = string("quantized_3_batch_dims_0"), val = int32(0)]; + bool quantized_3_validate_indices_0 = const()[name = string("quantized_3_validate_indices_0"), val = bool(false)]; + tensor weight_5_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(656832))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1181184))))[name = string("weight_5_to_fp16_palettized")]; + tensor quantized_3_cast_fp16_cast_uint16 = gather(axis = quantized_3_axis_0, batch_dims = quantized_3_batch_dims_0, indices = squeeze_1_cast_uint16, validate_indices = quantized_3_validate_indices_0, x = weight_5_to_fp16_palettized)[name = string("quantized_3_cast_fp16_cast_uint16")]; + tensor var_338 = const()[name = string("op_338"), val = tensor([0, 2, 1])]; + tensor quantized_7_axes_0 = const()[name = string("quantized_7_axes_0"), val = tensor([2])]; + tensor var_339_cast_fp16 = transpose(perm = var_338, x = quantized_3_cast_fp16_cast_uint16)[name = string("transpose_26")]; + tensor quantized_7_cast_fp16 = expand_dims(axes = quantized_7_axes_0, x = var_339_cast_fp16)[name = string("quantized_7_cast_fp16")]; + int32 quantized_5_axis_0 = const()[name = string("quantized_5_axis_0"), val = int32(0)]; + int32 quantized_5_batch_dims_0 = const()[name = string("quantized_5_batch_dims_0"), val = int32(0)]; + bool quantized_5_validate_indices_0 = const()[name = string("quantized_5_validate_indices_0"), val = bool(false)]; + tensor weight_7_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1181760))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1706112))))[name = string("weight_7_to_fp16_palettized")]; + tensor quantized_5_cast_fp16_cast_uint16 = gather(axis = quantized_5_axis_0, batch_dims = quantized_5_batch_dims_0, indices = squeeze_2_cast_uint16, validate_indices = quantized_5_validate_indices_0, x = weight_7_to_fp16_palettized)[name = string("quantized_5_cast_fp16_cast_uint16")]; + tensor var_346 = const()[name = string("op_346"), val = tensor([0, 2, 1])]; + tensor layer_out_1_axes_0 = const()[name = string("layer_out_1_axes_0"), val = tensor([2])]; + tensor var_347_cast_fp16 = transpose(perm = var_346, x = quantized_5_cast_fp16_cast_uint16)[name = string("transpose_25")]; + tensor layer_out_1_cast_fp16 = expand_dims(axes = layer_out_1_axes_0, x = var_347_cast_fp16)[name = string("layer_out_1_cast_fp16")]; + tensor quantized_11_cast_fp16 = add(x = quantized_7_cast_fp16, y = layer_out_1_cast_fp16)[name = string("quantized_11_cast_fp16")]; + int32 quantized_9_axis_0 = const()[name = string("quantized_9_axis_0"), val = int32(0)]; + int32 quantized_9_batch_dims_0 = const()[name = string("quantized_9_batch_dims_0"), val = int32(0)]; + bool quantized_9_validate_indices_0 = const()[name = string("quantized_9_validate_indices_0"), val = bool(false)]; + tensor weight_9_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1706688))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2231040))))[name = string("weight_9_to_fp16_palettized")]; + tensor quantized_9_cast_fp16_cast_uint16 = gather(axis = quantized_9_axis_0, batch_dims = quantized_9_batch_dims_0, indices = squeeze_3_cast_uint16, validate_indices = quantized_9_validate_indices_0, x = weight_9_to_fp16_palettized)[name = string("quantized_9_cast_fp16_cast_uint16")]; + tensor var_355 = const()[name = string("op_355"), val = tensor([0, 2, 1])]; + tensor layer_out_3_axes_0 = const()[name = string("layer_out_3_axes_0"), val = tensor([2])]; + tensor var_356_cast_fp16 = transpose(perm = var_355, x = quantized_9_cast_fp16_cast_uint16)[name = string("transpose_24")]; + tensor layer_out_3_cast_fp16 = expand_dims(axes = layer_out_3_axes_0, x = var_356_cast_fp16)[name = string("layer_out_3_cast_fp16")]; + tensor quantized_15_cast_fp16 = add(x = quantized_11_cast_fp16, y = layer_out_3_cast_fp16)[name = string("quantized_15_cast_fp16")]; + int32 quantized_13_axis_0 = const()[name = string("quantized_13_axis_0"), val = int32(0)]; + int32 quantized_13_batch_dims_0 = const()[name = string("quantized_13_batch_dims_0"), val = int32(0)]; + bool quantized_13_validate_indices_0 = const()[name = string("quantized_13_validate_indices_0"), val = bool(false)]; + tensor weight_11_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2231616))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2755968))))[name = string("weight_11_to_fp16_palettized")]; + tensor quantized_13_cast_fp16_cast_uint16 = gather(axis = quantized_13_axis_0, batch_dims = quantized_13_batch_dims_0, indices = squeeze_4_cast_uint16, validate_indices = quantized_13_validate_indices_0, x = weight_11_to_fp16_palettized)[name = string("quantized_13_cast_fp16_cast_uint16")]; + tensor var_364 = const()[name = string("op_364"), val = tensor([0, 2, 1])]; + tensor layer_out_5_axes_0 = const()[name = string("layer_out_5_axes_0"), val = tensor([2])]; + tensor var_365_cast_fp16 = transpose(perm = var_364, x = quantized_13_cast_fp16_cast_uint16)[name = string("transpose_23")]; + tensor layer_out_5_cast_fp16 = expand_dims(axes = layer_out_5_axes_0, x = var_365_cast_fp16)[name = string("layer_out_5_cast_fp16")]; + tensor quantized_19_cast_fp16 = add(x = quantized_15_cast_fp16, y = layer_out_5_cast_fp16)[name = string("quantized_19_cast_fp16")]; + int32 quantized_17_axis_0 = const()[name = string("quantized_17_axis_0"), val = int32(0)]; + int32 quantized_17_batch_dims_0 = const()[name = string("quantized_17_batch_dims_0"), val = int32(0)]; + bool quantized_17_validate_indices_0 = const()[name = string("quantized_17_validate_indices_0"), val = bool(false)]; + tensor weight_13_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2756544))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3280896))))[name = string("weight_13_to_fp16_palettized")]; + tensor quantized_17_cast_fp16_cast_uint16 = gather(axis = quantized_17_axis_0, batch_dims = quantized_17_batch_dims_0, indices = squeeze_5_cast_uint16, validate_indices = quantized_17_validate_indices_0, x = weight_13_to_fp16_palettized)[name = string("quantized_17_cast_fp16_cast_uint16")]; + tensor var_373 = const()[name = string("op_373"), val = tensor([0, 2, 1])]; + tensor layer_out_7_axes_0 = const()[name = string("layer_out_7_axes_0"), val = tensor([2])]; + tensor var_374_cast_fp16 = transpose(perm = var_373, x = quantized_17_cast_fp16_cast_uint16)[name = string("transpose_22")]; + tensor layer_out_7_cast_fp16 = expand_dims(axes = layer_out_7_axes_0, x = var_374_cast_fp16)[name = string("layer_out_7_cast_fp16")]; + tensor quantized_23_cast_fp16 = add(x = quantized_19_cast_fp16, y = layer_out_7_cast_fp16)[name = string("quantized_23_cast_fp16")]; + int32 quantized_21_axis_0 = const()[name = string("quantized_21_axis_0"), val = int32(0)]; + int32 quantized_21_batch_dims_0 = const()[name = string("quantized_21_batch_dims_0"), val = int32(0)]; + bool quantized_21_validate_indices_0 = const()[name = string("quantized_21_validate_indices_0"), val = bool(false)]; + tensor weight_15_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3281472))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3805824))))[name = string("weight_15_to_fp16_palettized")]; + tensor quantized_21_cast_fp16_cast_uint16 = gather(axis = quantized_21_axis_0, batch_dims = quantized_21_batch_dims_0, indices = squeeze_6_cast_uint16, validate_indices = quantized_21_validate_indices_0, x = weight_15_to_fp16_palettized)[name = string("quantized_21_cast_fp16_cast_uint16")]; + tensor var_382 = const()[name = string("op_382"), val = tensor([0, 2, 1])]; + tensor layer_out_9_axes_0 = const()[name = string("layer_out_9_axes_0"), val = tensor([2])]; + tensor var_383_cast_fp16 = transpose(perm = var_382, x = quantized_21_cast_fp16_cast_uint16)[name = string("transpose_21")]; + tensor layer_out_9_cast_fp16 = expand_dims(axes = layer_out_9_axes_0, x = var_383_cast_fp16)[name = string("layer_out_9_cast_fp16")]; + tensor quantized_27_cast_fp16 = add(x = quantized_23_cast_fp16, y = layer_out_9_cast_fp16)[name = string("quantized_27_cast_fp16")]; + int32 quantized_25_axis_0 = const()[name = string("quantized_25_axis_0"), val = int32(0)]; + int32 quantized_25_batch_dims_0 = const()[name = string("quantized_25_batch_dims_0"), val = int32(0)]; + bool quantized_25_validate_indices_0 = const()[name = string("quantized_25_validate_indices_0"), val = bool(false)]; + tensor weight_17_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3806400))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4330752))))[name = string("weight_17_to_fp16_palettized")]; + tensor quantized_25_cast_fp16_cast_uint16 = gather(axis = quantized_25_axis_0, batch_dims = quantized_25_batch_dims_0, indices = squeeze_7_cast_uint16, validate_indices = quantized_25_validate_indices_0, x = weight_17_to_fp16_palettized)[name = string("quantized_25_cast_fp16_cast_uint16")]; + tensor var_391 = const()[name = string("op_391"), val = tensor([0, 2, 1])]; + tensor layer_out_11_axes_0 = const()[name = string("layer_out_11_axes_0"), val = tensor([2])]; + tensor var_392_cast_fp16 = transpose(perm = var_391, x = quantized_25_cast_fp16_cast_uint16)[name = string("transpose_20")]; + tensor layer_out_11_cast_fp16 = expand_dims(axes = layer_out_11_axes_0, x = var_392_cast_fp16)[name = string("layer_out_11_cast_fp16")]; + tensor quantized_31_cast_fp16 = add(x = quantized_27_cast_fp16, y = layer_out_11_cast_fp16)[name = string("quantized_31_cast_fp16")]; + int32 quantized_29_axis_0 = const()[name = string("quantized_29_axis_0"), val = int32(0)]; + int32 quantized_29_batch_dims_0 = const()[name = string("quantized_29_batch_dims_0"), val = int32(0)]; + bool quantized_29_validate_indices_0 = const()[name = string("quantized_29_validate_indices_0"), val = bool(false)]; + tensor weight_19_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4331328))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4855680))))[name = string("weight_19_to_fp16_palettized")]; + tensor quantized_29_cast_fp16_cast_uint16 = gather(axis = quantized_29_axis_0, batch_dims = quantized_29_batch_dims_0, indices = squeeze_8_cast_uint16, validate_indices = quantized_29_validate_indices_0, x = weight_19_to_fp16_palettized)[name = string("quantized_29_cast_fp16_cast_uint16")]; + tensor var_400 = const()[name = string("op_400"), val = tensor([0, 2, 1])]; + tensor layer_out_13_axes_0 = const()[name = string("layer_out_13_axes_0"), val = tensor([2])]; + tensor var_401_cast_fp16 = transpose(perm = var_400, x = quantized_29_cast_fp16_cast_uint16)[name = string("transpose_19")]; + tensor layer_out_13_cast_fp16 = expand_dims(axes = layer_out_13_axes_0, x = var_401_cast_fp16)[name = string("layer_out_13_cast_fp16")]; + tensor quantized_35_cast_fp16 = add(x = quantized_31_cast_fp16, y = layer_out_13_cast_fp16)[name = string("quantized_35_cast_fp16")]; + int32 quantized_33_axis_0 = const()[name = string("quantized_33_axis_0"), val = int32(0)]; + int32 quantized_33_batch_dims_0 = const()[name = string("quantized_33_batch_dims_0"), val = int32(0)]; + bool quantized_33_validate_indices_0 = const()[name = string("quantized_33_validate_indices_0"), val = bool(false)]; + tensor weight_21_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4856256))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5380608))))[name = string("weight_21_to_fp16_palettized")]; + tensor quantized_33_cast_fp16_cast_uint16 = gather(axis = quantized_33_axis_0, batch_dims = quantized_33_batch_dims_0, indices = squeeze_9_cast_uint16, validate_indices = quantized_33_validate_indices_0, x = weight_21_to_fp16_palettized)[name = string("quantized_33_cast_fp16_cast_uint16")]; + tensor var_409 = const()[name = string("op_409"), val = tensor([0, 2, 1])]; + tensor layer_out_15_axes_0 = const()[name = string("layer_out_15_axes_0"), val = tensor([2])]; + tensor var_410_cast_fp16 = transpose(perm = var_409, x = quantized_33_cast_fp16_cast_uint16)[name = string("transpose_18")]; + tensor layer_out_15_cast_fp16 = expand_dims(axes = layer_out_15_axes_0, x = var_410_cast_fp16)[name = string("layer_out_15_cast_fp16")]; + tensor quantized_39_cast_fp16 = add(x = quantized_35_cast_fp16, y = layer_out_15_cast_fp16)[name = string("quantized_39_cast_fp16")]; + int32 quantized_37_axis_0 = const()[name = string("quantized_37_axis_0"), val = int32(0)]; + int32 quantized_37_batch_dims_0 = const()[name = string("quantized_37_batch_dims_0"), val = int32(0)]; + bool quantized_37_validate_indices_0 = const()[name = string("quantized_37_validate_indices_0"), val = bool(false)]; + tensor weight_23_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5381184))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5905536))))[name = string("weight_23_to_fp16_palettized")]; + tensor quantized_37_cast_fp16_cast_uint16 = gather(axis = quantized_37_axis_0, batch_dims = quantized_37_batch_dims_0, indices = squeeze_10_cast_uint16, validate_indices = quantized_37_validate_indices_0, x = weight_23_to_fp16_palettized)[name = string("quantized_37_cast_fp16_cast_uint16")]; + tensor var_418 = const()[name = string("op_418"), val = tensor([0, 2, 1])]; + tensor layer_out_17_axes_0 = const()[name = string("layer_out_17_axes_0"), val = tensor([2])]; + tensor var_419_cast_fp16 = transpose(perm = var_418, x = quantized_37_cast_fp16_cast_uint16)[name = string("transpose_17")]; + tensor layer_out_17_cast_fp16 = expand_dims(axes = layer_out_17_axes_0, x = var_419_cast_fp16)[name = string("layer_out_17_cast_fp16")]; + tensor quantized_43_cast_fp16 = add(x = quantized_39_cast_fp16, y = layer_out_17_cast_fp16)[name = string("quantized_43_cast_fp16")]; + int32 quantized_41_axis_0 = const()[name = string("quantized_41_axis_0"), val = int32(0)]; + int32 quantized_41_batch_dims_0 = const()[name = string("quantized_41_batch_dims_0"), val = int32(0)]; + bool quantized_41_validate_indices_0 = const()[name = string("quantized_41_validate_indices_0"), val = bool(false)]; + tensor weight_25_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5906112))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6430464))))[name = string("weight_25_to_fp16_palettized")]; + tensor quantized_41_cast_fp16_cast_uint16 = gather(axis = quantized_41_axis_0, batch_dims = quantized_41_batch_dims_0, indices = squeeze_11_cast_uint16, validate_indices = quantized_41_validate_indices_0, x = weight_25_to_fp16_palettized)[name = string("quantized_41_cast_fp16_cast_uint16")]; + tensor var_427 = const()[name = string("op_427"), val = tensor([0, 2, 1])]; + tensor layer_out_19_axes_0 = const()[name = string("layer_out_19_axes_0"), val = tensor([2])]; + tensor var_428_cast_fp16 = transpose(perm = var_427, x = quantized_41_cast_fp16_cast_uint16)[name = string("transpose_16")]; + tensor layer_out_19_cast_fp16 = expand_dims(axes = layer_out_19_axes_0, x = var_428_cast_fp16)[name = string("layer_out_19_cast_fp16")]; + tensor quantized_47_cast_fp16 = add(x = quantized_43_cast_fp16, y = layer_out_19_cast_fp16)[name = string("quantized_47_cast_fp16")]; + int32 quantized_45_axis_0 = const()[name = string("quantized_45_axis_0"), val = int32(0)]; + int32 quantized_45_batch_dims_0 = const()[name = string("quantized_45_batch_dims_0"), val = int32(0)]; + bool quantized_45_validate_indices_0 = const()[name = string("quantized_45_validate_indices_0"), val = bool(false)]; + tensor weight_27_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6431040))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6955392))))[name = string("weight_27_to_fp16_palettized")]; + tensor quantized_45_cast_fp16_cast_uint16 = gather(axis = quantized_45_axis_0, batch_dims = quantized_45_batch_dims_0, indices = squeeze_12_cast_uint16, validate_indices = quantized_45_validate_indices_0, x = weight_27_to_fp16_palettized)[name = string("quantized_45_cast_fp16_cast_uint16")]; + tensor var_436 = const()[name = string("op_436"), val = tensor([0, 2, 1])]; + tensor layer_out_21_axes_0 = const()[name = string("layer_out_21_axes_0"), val = tensor([2])]; + tensor var_437_cast_fp16 = transpose(perm = var_436, x = quantized_45_cast_fp16_cast_uint16)[name = string("transpose_15")]; + tensor layer_out_21_cast_fp16 = expand_dims(axes = layer_out_21_axes_0, x = var_437_cast_fp16)[name = string("layer_out_21_cast_fp16")]; + tensor quantized_51_cast_fp16 = add(x = quantized_47_cast_fp16, y = layer_out_21_cast_fp16)[name = string("quantized_51_cast_fp16")]; + int32 quantized_49_axis_0 = const()[name = string("quantized_49_axis_0"), val = int32(0)]; + int32 quantized_49_batch_dims_0 = const()[name = string("quantized_49_batch_dims_0"), val = int32(0)]; + bool quantized_49_validate_indices_0 = const()[name = string("quantized_49_validate_indices_0"), val = bool(false)]; + tensor weight_29_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6955968))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7480320))))[name = string("weight_29_to_fp16_palettized")]; + tensor quantized_49_cast_fp16_cast_uint16 = gather(axis = quantized_49_axis_0, batch_dims = quantized_49_batch_dims_0, indices = squeeze_13_cast_uint16, validate_indices = quantized_49_validate_indices_0, x = weight_29_to_fp16_palettized)[name = string("quantized_49_cast_fp16_cast_uint16")]; + tensor var_445 = const()[name = string("op_445"), val = tensor([0, 2, 1])]; + tensor layer_out_23_axes_0 = const()[name = string("layer_out_23_axes_0"), val = tensor([2])]; + tensor var_446_cast_fp16 = transpose(perm = var_445, x = quantized_49_cast_fp16_cast_uint16)[name = string("transpose_14")]; + tensor layer_out_23_cast_fp16 = expand_dims(axes = layer_out_23_axes_0, x = var_446_cast_fp16)[name = string("layer_out_23_cast_fp16")]; + tensor quantized_55_cast_fp16 = add(x = quantized_51_cast_fp16, y = layer_out_23_cast_fp16)[name = string("quantized_55_cast_fp16")]; + int32 quantized_53_axis_0 = const()[name = string("quantized_53_axis_0"), val = int32(0)]; + int32 quantized_53_batch_dims_0 = const()[name = string("quantized_53_batch_dims_0"), val = int32(0)]; + bool quantized_53_validate_indices_0 = const()[name = string("quantized_53_validate_indices_0"), val = bool(false)]; + tensor weight_31_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7480896))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8005248))))[name = string("weight_31_to_fp16_palettized")]; + tensor quantized_53_cast_fp16_cast_uint16 = gather(axis = quantized_53_axis_0, batch_dims = quantized_53_batch_dims_0, indices = squeeze_14_cast_uint16, validate_indices = quantized_53_validate_indices_0, x = weight_31_to_fp16_palettized)[name = string("quantized_53_cast_fp16_cast_uint16")]; + tensor var_454 = const()[name = string("op_454"), val = tensor([0, 2, 1])]; + tensor layer_out_25_axes_0 = const()[name = string("layer_out_25_axes_0"), val = tensor([2])]; + tensor var_455_cast_fp16 = transpose(perm = var_454, x = quantized_53_cast_fp16_cast_uint16)[name = string("transpose_13")]; + tensor layer_out_25_cast_fp16 = expand_dims(axes = layer_out_25_axes_0, x = var_455_cast_fp16)[name = string("layer_out_25_cast_fp16")]; + tensor quantized_59_cast_fp16 = add(x = quantized_55_cast_fp16, y = layer_out_25_cast_fp16)[name = string("quantized_59_cast_fp16")]; + int32 quantized_57_axis_0 = const()[name = string("quantized_57_axis_0"), val = int32(0)]; + int32 quantized_57_batch_dims_0 = const()[name = string("quantized_57_batch_dims_0"), val = int32(0)]; + bool quantized_57_validate_indices_0 = const()[name = string("quantized_57_validate_indices_0"), val = bool(false)]; + tensor weight_33_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8005824))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8530176))))[name = string("weight_33_to_fp16_palettized")]; + tensor quantized_57_cast_fp16_cast_uint16 = gather(axis = quantized_57_axis_0, batch_dims = quantized_57_batch_dims_0, indices = squeeze_15_cast_uint16, validate_indices = quantized_57_validate_indices_0, x = weight_33_to_fp16_palettized)[name = string("quantized_57_cast_fp16_cast_uint16")]; + tensor var_463 = const()[name = string("op_463"), val = tensor([0, 2, 1])]; + tensor layer_out_axes_0 = const()[name = string("layer_out_axes_0"), val = tensor([2])]; + tensor var_464_cast_fp16 = transpose(perm = var_463, x = quantized_57_cast_fp16_cast_uint16)[name = string("transpose_12")]; + tensor layer_out_cast_fp16 = expand_dims(axes = layer_out_axes_0, x = var_464_cast_fp16)[name = string("layer_out_cast_fp16")]; + tensor input_35_cast_fp16 = add(x = quantized_59_cast_fp16, y = layer_out_cast_fp16)[name = string("input_35_cast_fp16")]; + string var_472_pad_type_0 = const()[name = string("op_472_pad_type_0"), val = string("valid")]; + tensor var_472_strides_0 = const()[name = string("op_472_strides_0"), val = tensor([1, 1])]; + tensor var_472_pad_0 = const()[name = string("op_472_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_472_dilations_0 = const()[name = string("op_472_dilations_0"), val = tensor([1, 1])]; + int32 var_472_groups_0 = const()[name = string("op_472_groups_0"), val = int32(1)]; + tensor decoder_quantizer_rvq_rest_output_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8530752))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8661888))))[name = string("decoder_quantizer_rvq_rest_output_proj_weight_to_fp16_palettized")]; + tensor var_472_cast_fp16 = conv(dilations = var_472_dilations_0, groups = var_472_groups_0, pad = var_472_pad_0, pad_type = var_472_pad_type_0, strides = var_472_strides_0, weight = decoder_quantizer_rvq_rest_output_proj_weight_to_fp16_palettized, x = input_35_cast_fp16)[name = string("op_472_cast_fp16")]; + tensor x_1_cast_fp16 = add(x = quantized_cast_fp16, y = var_472_cast_fp16)[name = string("x_1_cast_fp16")]; + tensor input_37_pad_0 = const()[name = string("input_37_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 2, 0])]; + string input_37_mode_0 = const()[name = string("input_37_mode_0"), val = string("constant")]; + fp16 const_16_to_fp16 = const()[name = string("const_16_to_fp16"), val = fp16(0x0p+0)]; + tensor input_37_cast_fp16 = pad(constant_val = const_16_to_fp16, mode = input_37_mode_0, pad = input_37_pad_0, x = x_1_cast_fp16)[name = string("input_37_cast_fp16")]; + string input_39_pad_type_0 = const()[name = string("input_39_pad_type_0"), val = string("valid")]; + tensor input_39_strides_0 = const()[name = string("input_39_strides_0"), val = tensor([1, 1])]; + tensor input_39_pad_0 = const()[name = string("input_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_39_dilations_0 = const()[name = string("input_39_dilations_0"), val = tensor([1, 1])]; + int32 input_39_groups_0 = const()[name = string("input_39_groups_0"), val = int32(1)]; + tensor decoder_pre_conv_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8662464))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10235392))))[name = string("decoder_pre_conv_conv_weight_to_fp16_palettized")]; + tensor decoder_pre_conv_conv_bias_to_fp16 = const()[name = string("decoder_pre_conv_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10235968)))]; + tensor input_39_cast_fp16 = conv(bias = decoder_pre_conv_conv_bias_to_fp16, dilations = input_39_dilations_0, groups = input_39_groups_0, pad = input_39_pad_0, pad_type = input_39_pad_type_0, strides = input_39_strides_0, weight = decoder_pre_conv_conv_weight_to_fp16_palettized, x = input_37_cast_fp16)[name = string("input_39_cast_fp16")]; + string inputs_1_pad_type_0 = const()[name = string("inputs_1_pad_type_0"), val = string("valid")]; + tensor inputs_1_strides_0 = const()[name = string("inputs_1_strides_0"), val = tensor([1, 1])]; + tensor inputs_1_pad_0 = const()[name = string("inputs_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor inputs_1_dilations_0 = const()[name = string("inputs_1_dilations_0"), val = tensor([1, 1])]; + int32 inputs_1_groups_0 = const()[name = string("inputs_1_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_input_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10238080))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10762432))))[name = string("decoder_pre_transformer_input_proj_weight_to_fp16_palettized")]; + tensor decoder_pre_transformer_input_proj_bias_to_fp16 = const()[name = string("decoder_pre_transformer_input_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10763008)))]; + tensor inputs_1_cast_fp16 = conv(bias = decoder_pre_transformer_input_proj_bias_to_fp16, dilations = inputs_1_dilations_0, groups = inputs_1_groups_0, pad = inputs_1_pad_0, pad_type = inputs_1_pad_type_0, strides = inputs_1_strides_0, weight = decoder_pre_transformer_input_proj_weight_to_fp16_palettized, x = input_39_cast_fp16)[name = string("inputs_1_cast_fp16")]; + int32 pos_cos_1_axis_0 = const()[name = string("pos_cos_1_axis_0"), val = int32(0)]; + int32 pos_cos_1_batch_dims_0 = const()[name = string("pos_cos_1_batch_dims_0"), val = int32(0)]; + bool pos_cos_1_validate_indices_0 = const()[name = string("pos_cos_1_validate_indices_0"), val = bool(false)]; + tensor decoder_pre_transformer_position_embeddings_cos_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10764096))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10780544))))[name = string("decoder_pre_transformer_position_embeddings_cos_weight_to_fp16_palettized")]; + string cache_length_to_uint16_dtype_0 = const()[name = string("cache_length_to_uint16_dtype_0"), val = string("uint16")]; + tensor cache_length_to_uint16 = cast(dtype = cache_length_to_uint16_dtype_0, x = cache_length)[name = string("cast_0")]; + tensor pos_cos_1_cast_fp16_cast_uint16 = gather(axis = pos_cos_1_axis_0, batch_dims = pos_cos_1_batch_dims_0, indices = cache_length_to_uint16, validate_indices = pos_cos_1_validate_indices_0, x = decoder_pre_transformer_position_embeddings_cos_weight_to_fp16_palettized)[name = string("pos_cos_1_cast_fp16_cast_uint16")]; + tensor pos_cos_axes_0 = const()[name = string("pos_cos_axes_0"), val = tensor([2])]; + tensor pos_cos_cast_fp16 = expand_dims(axes = pos_cos_axes_0, x = pos_cos_1_cast_fp16_cast_uint16)[name = string("pos_cos_cast_fp16")]; + int32 pos_sin_1_axis_0 = const()[name = string("pos_sin_1_axis_0"), val = int32(0)]; + int32 pos_sin_1_batch_dims_0 = const()[name = string("pos_sin_1_batch_dims_0"), val = int32(0)]; + bool pos_sin_1_validate_indices_0 = const()[name = string("pos_sin_1_validate_indices_0"), val = bool(false)]; + tensor decoder_pre_transformer_position_embeddings_sin_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10781120))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10797568))))[name = string("decoder_pre_transformer_position_embeddings_sin_weight_to_fp16_palettized")]; + tensor pos_sin_1_cast_fp16_cast_uint16 = gather(axis = pos_sin_1_axis_0, batch_dims = pos_sin_1_batch_dims_0, indices = cache_length_to_uint16, validate_indices = pos_sin_1_validate_indices_0, x = decoder_pre_transformer_position_embeddings_sin_weight_to_fp16_palettized)[name = string("pos_sin_1_cast_fp16_cast_uint16")]; + tensor pos_sin_axes_0 = const()[name = string("pos_sin_axes_0"), val = tensor([2])]; + tensor pos_sin_cast_fp16 = expand_dims(axes = pos_sin_axes_0, x = pos_sin_1_cast_fp16_cast_uint16)[name = string("pos_sin_cast_fp16")]; + tensor tile_0 = const()[name = string("tile_0"), val = tensor([1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024])]; + int32 var_522_axis_0 = const()[name = string("op_522_axis_0"), val = int32(1)]; + tensor var_522_cast_fp16_0, tensor var_522_cast_fp16_1, tensor var_522_cast_fp16_2, tensor var_522_cast_fp16_3, tensor var_522_cast_fp16_4, tensor var_522_cast_fp16_5, tensor var_522_cast_fp16_6, tensor var_522_cast_fp16_7 = split(axis = var_522_axis_0, split_sizes = tile_0, x = key_cache)[name = string("op_522_cast_fp16")]; + tensor tile_1 = const()[name = string("tile_1"), val = tensor([1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024])]; + int32 var_531_axis_0 = const()[name = string("op_531_axis_0"), val = int32(1)]; + tensor var_531_cast_fp16_0, tensor var_531_cast_fp16_1, tensor var_531_cast_fp16_2, tensor var_531_cast_fp16_3, tensor var_531_cast_fp16_4, tensor var_531_cast_fp16_5, tensor var_531_cast_fp16_6, tensor var_531_cast_fp16_7 = split(axis = var_531_axis_0, split_sizes = tile_1, x = value_cache)[name = string("op_531_cast_fp16")]; + tensor inputs_sq_1_cast_fp16 = mul(x = inputs_1_cast_fp16, y = inputs_1_cast_fp16)[name = string("inputs_sq_1_cast_fp16")]; + tensor variance_1_axes_0 = const()[name = string("variance_1_axes_0"), val = tensor([1])]; + bool variance_1_keep_dims_0 = const()[name = string("variance_1_keep_dims_0"), val = bool(true)]; + tensor variance_1_cast_fp16 = reduce_mean(axes = variance_1_axes_0, keep_dims = variance_1_keep_dims_0, x = inputs_sq_1_cast_fp16)[name = string("variance_1_cast_fp16")]; + fp16 var_550_to_fp16 = const()[name = string("op_550_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_551_cast_fp16 = add(x = variance_1_cast_fp16, y = var_550_to_fp16)[name = string("op_551_cast_fp16")]; + fp32 var_552_epsilon_0 = const()[name = string("op_552_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_552_cast_fp16 = rsqrt(epsilon = var_552_epsilon_0, x = var_551_cast_fp16)[name = string("op_552_cast_fp16")]; + tensor hidden_states_1_cast_fp16 = mul(x = inputs_1_cast_fp16, y = var_552_cast_fp16)[name = string("hidden_states_1_cast_fp16")]; + tensor w_1_to_fp16 = const()[name = string("w_1_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10798144)))]; + tensor x_3_cast_fp16 = mul(x = w_1_to_fp16, y = hidden_states_1_cast_fp16)[name = string("x_3_cast_fp16")]; + string q_1_pad_type_0 = const()[name = string("q_1_pad_type_0"), val = string("valid")]; + tensor q_1_strides_0 = const()[name = string("q_1_strides_0"), val = tensor([1, 1])]; + tensor q_1_pad_0 = const()[name = string("q_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_1_dilations_0 = const()[name = string("q_1_dilations_0"), val = tensor([1, 1])]; + int32 q_1_groups_0 = const()[name = string("q_1_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_0_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10799232))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11323584))))[name = string("decoder_pre_transformer_layers_0_self_attn_q_proj_weight_to_fp16_palettized")]; + tensor q_1_cast_fp16 = conv(dilations = q_1_dilations_0, groups = q_1_groups_0, pad = q_1_pad_0, pad_type = q_1_pad_type_0, strides = q_1_strides_0, weight = decoder_pre_transformer_layers_0_self_attn_q_proj_weight_to_fp16_palettized, x = x_3_cast_fp16)[name = string("q_1_cast_fp16")]; + string k_1_pad_type_0 = const()[name = string("k_1_pad_type_0"), val = string("valid")]; + tensor k_1_strides_0 = const()[name = string("k_1_strides_0"), val = tensor([1, 1])]; + tensor k_1_pad_0 = const()[name = string("k_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_1_dilations_0 = const()[name = string("k_1_dilations_0"), val = tensor([1, 1])]; + int32 k_1_groups_0 = const()[name = string("k_1_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_0_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11324160))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11848512))))[name = string("decoder_pre_transformer_layers_0_self_attn_k_proj_weight_to_fp16_palettized")]; + tensor k_1_cast_fp16 = conv(dilations = k_1_dilations_0, groups = k_1_groups_0, pad = k_1_pad_0, pad_type = k_1_pad_type_0, strides = k_1_strides_0, weight = decoder_pre_transformer_layers_0_self_attn_k_proj_weight_to_fp16_palettized, x = x_3_cast_fp16)[name = string("k_1_cast_fp16")]; + string v_1_pad_type_0 = const()[name = string("v_1_pad_type_0"), val = string("valid")]; + tensor v_1_strides_0 = const()[name = string("v_1_strides_0"), val = tensor([1, 1])]; + tensor v_1_pad_0 = const()[name = string("v_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_1_dilations_0 = const()[name = string("v_1_dilations_0"), val = tensor([1, 1])]; + int32 v_1_groups_0 = const()[name = string("v_1_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_0_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11849088))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12373440))))[name = string("decoder_pre_transformer_layers_0_self_attn_v_proj_weight_to_fp16_palettized")]; + tensor v_1_cast_fp16 = conv(dilations = v_1_dilations_0, groups = v_1_groups_0, pad = v_1_pad_0, pad_type = v_1_pad_type_0, strides = v_1_strides_0, weight = decoder_pre_transformer_layers_0_self_attn_v_proj_weight_to_fp16_palettized, x = x_3_cast_fp16)[name = string("v_1_cast_fp16")]; + tensor var_584 = const()[name = string("op_584"), val = tensor([16, 64, 1, 1])]; + tensor q_3_cast_fp16 = reshape(shape = var_584, x = q_1_cast_fp16)[name = string("q_3_cast_fp16")]; + tensor var_589 = const()[name = string("op_589"), val = tensor([16, 64, 1, 1])]; + tensor k_3_cast_fp16 = reshape(shape = var_589, x = k_1_cast_fp16)[name = string("k_3_cast_fp16")]; + tensor cos_expanded_1_axes_0 = const()[name = string("cos_expanded_1_axes_0"), val = tensor([3])]; + tensor cos_expanded_1_cast_fp16 = expand_dims(axes = cos_expanded_1_axes_0, x = pos_cos_cast_fp16)[name = string("cos_expanded_1_cast_fp16")]; + tensor sin_expanded_1_axes_0 = const()[name = string("sin_expanded_1_axes_0"), val = tensor([3])]; + tensor sin_expanded_1_cast_fp16 = expand_dims(axes = sin_expanded_1_axes_0, x = pos_sin_cast_fp16)[name = string("sin_expanded_1_cast_fp16")]; + tensor var_593 = const()[name = string("op_593"), val = tensor([16, 1, 1, 1])]; + tensor cos_1_cast_fp16 = tile(reps = var_593, x = cos_expanded_1_cast_fp16)[name = string("cos_1_cast_fp16")]; + tensor var_595 = const()[name = string("op_595"), val = tensor([16, 1, 1, 1])]; + tensor sin_1_cast_fp16 = tile(reps = var_595, x = sin_expanded_1_cast_fp16)[name = string("sin_1_cast_fp16")]; + tensor var_601_cast_fp16 = mul(x = q_3_cast_fp16, y = cos_1_cast_fp16)[name = string("op_601_cast_fp16")]; + tensor var_606_begin_0 = const()[name = string("op_606_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_606_end_0 = const()[name = string("op_606_end_0"), val = tensor([16, 32, 1, 1])]; + tensor var_606_end_mask_0 = const()[name = string("op_606_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_606_cast_fp16 = slice_by_index(begin = var_606_begin_0, end = var_606_end_0, end_mask = var_606_end_mask_0, x = q_3_cast_fp16)[name = string("op_606_cast_fp16")]; + tensor var_613_begin_0 = const()[name = string("op_613_begin_0"), val = tensor([0, 32, 0, 0])]; + tensor var_613_end_0 = const()[name = string("op_613_end_0"), val = tensor([16, 64, 1, 1])]; + tensor var_613_end_mask_0 = const()[name = string("op_613_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_613_cast_fp16 = slice_by_index(begin = var_613_begin_0, end = var_613_end_0, end_mask = var_613_end_mask_0, x = q_3_cast_fp16)[name = string("op_613_cast_fp16")]; + fp16 const_24_promoted_to_fp16 = const()[name = string("const_24_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_616_cast_fp16 = mul(x = var_613_cast_fp16, y = const_24_promoted_to_fp16)[name = string("op_616_cast_fp16")]; + bool var_618_interleave_0 = const()[name = string("op_618_interleave_0"), val = bool(false)]; + tensor var_618_cast_fp16 = concat(axis = var_32, interleave = var_618_interleave_0, values = (var_616_cast_fp16, var_606_cast_fp16))[name = string("op_618_cast_fp16")]; + tensor var_619_cast_fp16 = mul(x = var_618_cast_fp16, y = sin_1_cast_fp16)[name = string("op_619_cast_fp16")]; + tensor q_rotated_1_cast_fp16 = add(x = var_601_cast_fp16, y = var_619_cast_fp16)[name = string("q_rotated_1_cast_fp16")]; + tensor var_621_cast_fp16 = mul(x = k_3_cast_fp16, y = cos_1_cast_fp16)[name = string("op_621_cast_fp16")]; + tensor var_626_begin_0 = const()[name = string("op_626_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_626_end_0 = const()[name = string("op_626_end_0"), val = tensor([16, 32, 1, 1])]; + tensor var_626_end_mask_0 = const()[name = string("op_626_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_626_cast_fp16 = slice_by_index(begin = var_626_begin_0, end = var_626_end_0, end_mask = var_626_end_mask_0, x = k_3_cast_fp16)[name = string("op_626_cast_fp16")]; + tensor var_633_begin_0 = const()[name = string("op_633_begin_0"), val = tensor([0, 32, 0, 0])]; + tensor var_633_end_0 = const()[name = string("op_633_end_0"), val = tensor([16, 64, 1, 1])]; + tensor var_633_end_mask_0 = const()[name = string("op_633_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_633_cast_fp16 = slice_by_index(begin = var_633_begin_0, end = var_633_end_0, end_mask = var_633_end_mask_0, x = k_3_cast_fp16)[name = string("op_633_cast_fp16")]; + fp16 const_27_promoted_to_fp16 = const()[name = string("const_27_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_636_cast_fp16 = mul(x = var_633_cast_fp16, y = const_27_promoted_to_fp16)[name = string("op_636_cast_fp16")]; + bool var_638_interleave_0 = const()[name = string("op_638_interleave_0"), val = bool(false)]; + tensor var_638_cast_fp16 = concat(axis = var_32, interleave = var_638_interleave_0, values = (var_636_cast_fp16, var_626_cast_fp16))[name = string("op_638_cast_fp16")]; + tensor var_639_cast_fp16 = mul(x = var_638_cast_fp16, y = sin_1_cast_fp16)[name = string("op_639_cast_fp16")]; + tensor k_rotated_1_cast_fp16 = add(x = var_621_cast_fp16, y = var_639_cast_fp16)[name = string("k_rotated_1_cast_fp16")]; + tensor var_643 = const()[name = string("op_643"), val = tensor([1, 1024, 1, 1])]; + tensor current_key_1_cast_fp16 = reshape(shape = var_643, x = k_rotated_1_cast_fp16)[name = string("current_key_1_cast_fp16")]; + tensor var_645_axes_0 = const()[name = string("op_645_axes_0"), val = tensor([1])]; + tensor var_645_cast_fp16 = expand_dims(axes = var_645_axes_0, x = kv_cache_update_mask)[name = string("op_645_cast_fp16")]; + tensor update_mask_1_axes_0 = const()[name = string("update_mask_1_axes_0"), val = tensor([2])]; + tensor update_mask_1_cast_fp16 = expand_dims(axes = update_mask_1_axes_0, x = var_645_cast_fp16)[name = string("update_mask_1_cast_fp16")]; + fp16 var_32_promoted_to_fp16 = const()[name = string("op_32_promoted_to_fp16"), val = fp16(0x1p+0)]; + tensor var_647_cast_fp16 = sub(x = var_32_promoted_to_fp16, y = update_mask_1_cast_fp16)[name = string("op_647_cast_fp16")]; + tensor var_648_cast_fp16 = mul(x = var_522_cast_fp16_0, y = var_647_cast_fp16)[name = string("op_648_cast_fp16")]; + tensor var_649_cast_fp16 = mul(x = current_key_1_cast_fp16, y = update_mask_1_cast_fp16)[name = string("op_649_cast_fp16")]; + tensor key_cache_updated_1_cast_fp16 = add(x = var_648_cast_fp16, y = var_649_cast_fp16)[name = string("key_cache_updated_1_cast_fp16")]; + tensor var_652_cast_fp16 = mul(x = var_531_cast_fp16_0, y = var_647_cast_fp16)[name = string("op_652_cast_fp16")]; + tensor var_653_cast_fp16 = mul(x = v_1_cast_fp16, y = update_mask_1_cast_fp16)[name = string("op_653_cast_fp16")]; + tensor value_cache_updated_1_cast_fp16 = add(x = var_652_cast_fp16, y = var_653_cast_fp16)[name = string("value_cache_updated_1_cast_fp16")]; + tensor var_655 = const()[name = string("op_655"), val = tensor([1, 16, 64, 1])]; + tensor q_7_cast_fp16 = reshape(shape = var_655, x = q_rotated_1_cast_fp16)[name = string("q_7_cast_fp16")]; + tensor var_658 = const()[name = string("op_658"), val = tensor([1, 16, 64, 256])]; + tensor k_for_attn_1_cast_fp16 = reshape(shape = var_658, x = key_cache_updated_1_cast_fp16)[name = string("k_for_attn_1_cast_fp16")]; + tensor var_660 = const()[name = string("op_660"), val = tensor([1, 16, 64, 256])]; + tensor v_for_attn_1_cast_fp16 = reshape(shape = var_660, x = value_cache_updated_1_cast_fp16)[name = string("v_for_attn_1_cast_fp16")]; + bool var_664_transpose_x_1 = const()[name = string("op_664_transpose_x_1"), val = bool(true)]; + bool var_664_transpose_y_1 = const()[name = string("op_664_transpose_y_1"), val = bool(false)]; + tensor var_664_cast_fp16 = matmul(transpose_x = var_664_transpose_x_1, transpose_y = var_664_transpose_y_1, x = q_7_cast_fp16, y = k_for_attn_1_cast_fp16)[name = string("op_664_cast_fp16")]; + fp16 var_665_to_fp16 = const()[name = string("op_665_to_fp16"), val = fp16(0x1p-3)]; + tensor attn_weights_1_cast_fp16 = mul(x = var_664_cast_fp16, y = var_665_to_fp16)[name = string("attn_weights_1_cast_fp16")]; + tensor var_667_axes_0 = const()[name = string("op_667_axes_0"), val = tensor([1])]; + tensor var_667_cast_fp16 = expand_dims(axes = var_667_axes_0, x = key_padding_mask)[name = string("op_667_cast_fp16")]; + tensor attn_mask_1_axes_0 = const()[name = string("attn_mask_1_axes_0"), val = tensor([2])]; + tensor attn_mask_1_cast_fp16 = expand_dims(axes = attn_mask_1_axes_0, x = var_667_cast_fp16)[name = string("attn_mask_1_cast_fp16")]; + tensor input_41_cast_fp16 = add(x = attn_weights_1_cast_fp16, y = attn_mask_1_cast_fp16)[name = string("input_41_cast_fp16")]; + tensor attn_weights_3_cast_fp16 = softmax(axis = var_28, x = input_41_cast_fp16)[name = string("attn_weights_3_cast_fp16")]; + bool attn_output_1_transpose_x_1 = const()[name = string("attn_output_1_transpose_x_1"), val = bool(false)]; + bool attn_output_1_transpose_y_1 = const()[name = string("attn_output_1_transpose_y_1"), val = bool(true)]; + tensor attn_output_1_cast_fp16 = matmul(transpose_x = attn_output_1_transpose_x_1, transpose_y = attn_output_1_transpose_y_1, x = attn_weights_3_cast_fp16, y = v_for_attn_1_cast_fp16)[name = string("attn_output_1_cast_fp16")]; + tensor var_674 = const()[name = string("op_674"), val = tensor([0, 1, 3, 2])]; + tensor var_676 = const()[name = string("op_676"), val = tensor([1, -1, 1, 1])]; + tensor attn_output_3_cast_fp16 = transpose(perm = var_674, x = attn_output_1_cast_fp16)[name = string("transpose_11")]; + tensor input_43_cast_fp16 = reshape(shape = var_676, x = attn_output_3_cast_fp16)[name = string("input_43_cast_fp16")]; + string x_5_pad_type_0 = const()[name = string("x_5_pad_type_0"), val = string("valid")]; + tensor x_5_strides_0 = const()[name = string("x_5_strides_0"), val = tensor([1, 1])]; + tensor x_5_pad_0 = const()[name = string("x_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_5_dilations_0 = const()[name = string("x_5_dilations_0"), val = tensor([1, 1])]; + int32 x_5_groups_0 = const()[name = string("x_5_groups_0"), val = int32(1)]; + tensor op_691_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12374016))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12898368))))[name = string("op_691_weight_0_to_fp16_palettized")]; + tensor var_691_bias_0_to_fp16 = const()[name = string("op_691_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12898944)))]; + tensor var_691_cast_fp16 = conv(bias = var_691_bias_0_to_fp16, dilations = x_5_dilations_0, groups = x_5_groups_0, pad = x_5_pad_0, pad_type = x_5_pad_type_0, strides = x_5_strides_0, weight = op_691_weight_0_to_fp16_palettized, x = input_43_cast_fp16)[name = string("op_691_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = var_691_cast_fp16)[name = string("inputs_3_cast_fp16")]; + tensor inputs_sq_3_cast_fp16 = mul(x = inputs_3_cast_fp16, y = inputs_3_cast_fp16)[name = string("inputs_sq_3_cast_fp16")]; + tensor variance_3_axes_0 = const()[name = string("variance_3_axes_0"), val = tensor([1])]; + bool variance_3_keep_dims_0 = const()[name = string("variance_3_keep_dims_0"), val = bool(true)]; + tensor variance_3_cast_fp16 = reduce_mean(axes = variance_3_axes_0, keep_dims = variance_3_keep_dims_0, x = inputs_sq_3_cast_fp16)[name = string("variance_3_cast_fp16")]; + fp16 var_697_to_fp16 = const()[name = string("op_697_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_698_cast_fp16 = add(x = variance_3_cast_fp16, y = var_697_to_fp16)[name = string("op_698_cast_fp16")]; + fp32 var_699_epsilon_0 = const()[name = string("op_699_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_699_cast_fp16 = rsqrt(epsilon = var_699_epsilon_0, x = var_698_cast_fp16)[name = string("op_699_cast_fp16")]; + tensor hidden_states_3_cast_fp16 = mul(x = inputs_3_cast_fp16, y = var_699_cast_fp16)[name = string("hidden_states_3_cast_fp16")]; + tensor w_3_to_fp16 = const()[name = string("w_3_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12900032)))]; + tensor input_45_cast_fp16 = mul(x = w_3_to_fp16, y = hidden_states_3_cast_fp16)[name = string("input_45_cast_fp16")]; + string input_47_pad_type_0 = const()[name = string("input_47_pad_type_0"), val = string("valid")]; + tensor input_47_strides_0 = const()[name = string("input_47_strides_0"), val = tensor([1, 1])]; + tensor input_47_pad_0 = const()[name = string("input_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_47_dilations_0 = const()[name = string("input_47_dilations_0"), val = tensor([1, 1])]; + int32 input_47_groups_0 = const()[name = string("input_47_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_0_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12901120))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13425472))))[name = string("decoder_pre_transformer_layers_0_mlp_gate_proj_weight_to_fp16_palettized")]; + tensor input_47_cast_fp16 = conv(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 = decoder_pre_transformer_layers_0_mlp_gate_proj_weight_to_fp16_palettized, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")]; + tensor var_713_cast_fp16 = silu(x = input_47_cast_fp16)[name = string("op_713_cast_fp16")]; + string var_719_pad_type_0 = const()[name = string("op_719_pad_type_0"), val = string("valid")]; + tensor var_719_strides_0 = const()[name = string("op_719_strides_0"), val = tensor([1, 1])]; + tensor var_719_pad_0 = const()[name = string("op_719_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_719_dilations_0 = const()[name = string("op_719_dilations_0"), val = tensor([1, 1])]; + int32 var_719_groups_0 = const()[name = string("op_719_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_0_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13426048))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13950400))))[name = string("decoder_pre_transformer_layers_0_mlp_up_proj_weight_to_fp16_palettized")]; + tensor var_719_cast_fp16 = conv(dilations = var_719_dilations_0, groups = var_719_groups_0, pad = var_719_pad_0, pad_type = var_719_pad_type_0, strides = var_719_strides_0, weight = decoder_pre_transformer_layers_0_mlp_up_proj_weight_to_fp16_palettized, x = input_45_cast_fp16)[name = string("op_719_cast_fp16")]; + tensor input_49_cast_fp16 = mul(x = var_713_cast_fp16, y = var_719_cast_fp16)[name = string("input_49_cast_fp16")]; + string x_7_pad_type_0 = const()[name = string("x_7_pad_type_0"), val = string("valid")]; + tensor x_7_strides_0 = const()[name = string("x_7_strides_0"), val = tensor([1, 1])]; + tensor x_7_pad_0 = const()[name = string("x_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_7_dilations_0 = const()[name = string("x_7_dilations_0"), val = tensor([1, 1])]; + int32 x_7_groups_0 = const()[name = string("x_7_groups_0"), val = int32(1)]; + tensor op_730_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13950976))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14475328))))[name = string("op_730_weight_0_to_fp16_palettized")]; + tensor var_730_bias_0_to_fp16 = const()[name = string("op_730_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14475904)))]; + tensor var_730_cast_fp16 = conv(bias = var_730_bias_0_to_fp16, dilations = x_7_dilations_0, groups = x_7_groups_0, pad = x_7_pad_0, pad_type = x_7_pad_type_0, strides = x_7_strides_0, weight = op_730_weight_0_to_fp16_palettized, x = input_49_cast_fp16)[name = string("op_730_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = var_730_cast_fp16)[name = string("inputs_5_cast_fp16")]; + tensor inputs_sq_5_cast_fp16 = mul(x = inputs_5_cast_fp16, y = inputs_5_cast_fp16)[name = string("inputs_sq_5_cast_fp16")]; + tensor variance_5_axes_0 = const()[name = string("variance_5_axes_0"), val = tensor([1])]; + bool variance_5_keep_dims_0 = const()[name = string("variance_5_keep_dims_0"), val = bool(true)]; + tensor variance_5_cast_fp16 = reduce_mean(axes = variance_5_axes_0, keep_dims = variance_5_keep_dims_0, x = inputs_sq_5_cast_fp16)[name = string("variance_5_cast_fp16")]; + fp16 var_746_to_fp16 = const()[name = string("op_746_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_747_cast_fp16 = add(x = variance_5_cast_fp16, y = var_746_to_fp16)[name = string("op_747_cast_fp16")]; + fp32 var_748_epsilon_0 = const()[name = string("op_748_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_748_cast_fp16 = rsqrt(epsilon = var_748_epsilon_0, x = var_747_cast_fp16)[name = string("op_748_cast_fp16")]; + tensor hidden_states_5_cast_fp16 = mul(x = inputs_5_cast_fp16, y = var_748_cast_fp16)[name = string("hidden_states_5_cast_fp16")]; + tensor w_5_to_fp16 = const()[name = string("w_5_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14476992)))]; + tensor x_9_cast_fp16 = mul(x = w_5_to_fp16, y = hidden_states_5_cast_fp16)[name = string("x_9_cast_fp16")]; + string q_9_pad_type_0 = const()[name = string("q_9_pad_type_0"), val = string("valid")]; + tensor q_9_strides_0 = const()[name = string("q_9_strides_0"), val = tensor([1, 1])]; + tensor q_9_pad_0 = const()[name = string("q_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_9_dilations_0 = const()[name = string("q_9_dilations_0"), val = tensor([1, 1])]; + int32 q_9_groups_0 = const()[name = string("q_9_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_1_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14478080))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15002432))))[name = string("decoder_pre_transformer_layers_1_self_attn_q_proj_weight_to_fp16_palettized")]; + tensor q_9_cast_fp16 = conv(dilations = q_9_dilations_0, groups = q_9_groups_0, pad = q_9_pad_0, pad_type = q_9_pad_type_0, strides = q_9_strides_0, weight = decoder_pre_transformer_layers_1_self_attn_q_proj_weight_to_fp16_palettized, x = x_9_cast_fp16)[name = string("q_9_cast_fp16")]; + string k_5_pad_type_0 = const()[name = string("k_5_pad_type_0"), val = string("valid")]; + tensor k_5_strides_0 = const()[name = string("k_5_strides_0"), val = tensor([1, 1])]; + tensor k_5_pad_0 = const()[name = string("k_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_5_dilations_0 = const()[name = string("k_5_dilations_0"), val = tensor([1, 1])]; + int32 k_5_groups_0 = const()[name = string("k_5_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_1_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15003008))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15527360))))[name = string("decoder_pre_transformer_layers_1_self_attn_k_proj_weight_to_fp16_palettized")]; + tensor k_5_cast_fp16 = conv(dilations = k_5_dilations_0, groups = k_5_groups_0, pad = k_5_pad_0, pad_type = k_5_pad_type_0, strides = k_5_strides_0, weight = decoder_pre_transformer_layers_1_self_attn_k_proj_weight_to_fp16_palettized, x = x_9_cast_fp16)[name = string("k_5_cast_fp16")]; + string v_3_pad_type_0 = const()[name = string("v_3_pad_type_0"), val = string("valid")]; + tensor v_3_strides_0 = const()[name = string("v_3_strides_0"), val = tensor([1, 1])]; + tensor v_3_pad_0 = const()[name = string("v_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_3_dilations_0 = const()[name = string("v_3_dilations_0"), val = tensor([1, 1])]; + int32 v_3_groups_0 = const()[name = string("v_3_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_1_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15527936))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16052288))))[name = string("decoder_pre_transformer_layers_1_self_attn_v_proj_weight_to_fp16_palettized")]; + tensor v_3_cast_fp16 = conv(dilations = v_3_dilations_0, groups = v_3_groups_0, pad = v_3_pad_0, pad_type = v_3_pad_type_0, strides = v_3_strides_0, weight = decoder_pre_transformer_layers_1_self_attn_v_proj_weight_to_fp16_palettized, x = x_9_cast_fp16)[name = string("v_3_cast_fp16")]; + tensor var_780 = const()[name = string("op_780"), val = tensor([16, 64, 1, 1])]; + tensor q_11_cast_fp16 = reshape(shape = var_780, x = q_9_cast_fp16)[name = string("q_11_cast_fp16")]; + tensor var_785 = const()[name = string("op_785"), val = tensor([16, 64, 1, 1])]; + tensor k_7_cast_fp16 = reshape(shape = var_785, x = k_5_cast_fp16)[name = string("k_7_cast_fp16")]; + tensor var_797_cast_fp16 = mul(x = q_11_cast_fp16, y = cos_1_cast_fp16)[name = string("op_797_cast_fp16")]; + tensor var_802_begin_0 = const()[name = string("op_802_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_802_end_0 = const()[name = string("op_802_end_0"), val = tensor([16, 32, 1, 1])]; + tensor var_802_end_mask_0 = const()[name = string("op_802_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_802_cast_fp16 = slice_by_index(begin = var_802_begin_0, end = var_802_end_0, end_mask = var_802_end_mask_0, x = q_11_cast_fp16)[name = string("op_802_cast_fp16")]; + tensor var_809_begin_0 = const()[name = string("op_809_begin_0"), val = tensor([0, 32, 0, 0])]; + tensor var_809_end_0 = const()[name = string("op_809_end_0"), val = tensor([16, 64, 1, 1])]; + tensor var_809_end_mask_0 = const()[name = string("op_809_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_809_cast_fp16 = slice_by_index(begin = var_809_begin_0, end = var_809_end_0, end_mask = var_809_end_mask_0, x = q_11_cast_fp16)[name = string("op_809_cast_fp16")]; + fp16 const_32_promoted_to_fp16 = const()[name = string("const_32_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_812_cast_fp16 = mul(x = var_809_cast_fp16, y = const_32_promoted_to_fp16)[name = string("op_812_cast_fp16")]; + bool var_814_interleave_0 = const()[name = string("op_814_interleave_0"), val = bool(false)]; + tensor var_814_cast_fp16 = concat(axis = var_32, interleave = var_814_interleave_0, values = (var_812_cast_fp16, var_802_cast_fp16))[name = string("op_814_cast_fp16")]; + tensor var_815_cast_fp16 = mul(x = var_814_cast_fp16, y = sin_1_cast_fp16)[name = string("op_815_cast_fp16")]; + tensor q_rotated_3_cast_fp16 = add(x = var_797_cast_fp16, y = var_815_cast_fp16)[name = string("q_rotated_3_cast_fp16")]; + tensor var_817_cast_fp16 = mul(x = k_7_cast_fp16, y = cos_1_cast_fp16)[name = string("op_817_cast_fp16")]; + tensor var_822_begin_0 = const()[name = string("op_822_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_822_end_0 = const()[name = string("op_822_end_0"), val = tensor([16, 32, 1, 1])]; + tensor var_822_end_mask_0 = const()[name = string("op_822_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_822_cast_fp16 = slice_by_index(begin = var_822_begin_0, end = var_822_end_0, end_mask = var_822_end_mask_0, x = k_7_cast_fp16)[name = string("op_822_cast_fp16")]; + tensor var_829_begin_0 = const()[name = string("op_829_begin_0"), val = tensor([0, 32, 0, 0])]; + tensor var_829_end_0 = const()[name = string("op_829_end_0"), val = tensor([16, 64, 1, 1])]; + tensor var_829_end_mask_0 = const()[name = string("op_829_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_829_cast_fp16 = slice_by_index(begin = var_829_begin_0, end = var_829_end_0, end_mask = var_829_end_mask_0, x = k_7_cast_fp16)[name = string("op_829_cast_fp16")]; + fp16 const_35_promoted_to_fp16 = const()[name = string("const_35_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_832_cast_fp16 = mul(x = var_829_cast_fp16, y = const_35_promoted_to_fp16)[name = string("op_832_cast_fp16")]; + bool var_834_interleave_0 = const()[name = string("op_834_interleave_0"), val = bool(false)]; + tensor var_834_cast_fp16 = concat(axis = var_32, interleave = var_834_interleave_0, values = (var_832_cast_fp16, var_822_cast_fp16))[name = string("op_834_cast_fp16")]; + tensor var_835_cast_fp16 = mul(x = var_834_cast_fp16, y = sin_1_cast_fp16)[name = string("op_835_cast_fp16")]; + tensor k_rotated_3_cast_fp16 = add(x = var_817_cast_fp16, y = var_835_cast_fp16)[name = string("k_rotated_3_cast_fp16")]; + tensor var_839 = const()[name = string("op_839"), val = tensor([1, 1024, 1, 1])]; + tensor current_key_3_cast_fp16 = reshape(shape = var_839, x = k_rotated_3_cast_fp16)[name = string("current_key_3_cast_fp16")]; + tensor var_844_cast_fp16 = mul(x = var_522_cast_fp16_1, y = var_647_cast_fp16)[name = string("op_844_cast_fp16")]; + tensor var_845_cast_fp16 = mul(x = current_key_3_cast_fp16, y = update_mask_1_cast_fp16)[name = string("op_845_cast_fp16")]; + tensor key_cache_updated_3_cast_fp16 = add(x = var_844_cast_fp16, y = var_845_cast_fp16)[name = string("key_cache_updated_3_cast_fp16")]; + tensor var_848_cast_fp16 = mul(x = var_531_cast_fp16_1, y = var_647_cast_fp16)[name = string("op_848_cast_fp16")]; + tensor var_849_cast_fp16 = mul(x = v_3_cast_fp16, y = update_mask_1_cast_fp16)[name = string("op_849_cast_fp16")]; + tensor value_cache_updated_3_cast_fp16 = add(x = var_848_cast_fp16, y = var_849_cast_fp16)[name = string("value_cache_updated_3_cast_fp16")]; + tensor var_851 = const()[name = string("op_851"), val = tensor([1, 16, 64, 1])]; + tensor q_15_cast_fp16 = reshape(shape = var_851, x = q_rotated_3_cast_fp16)[name = string("q_15_cast_fp16")]; + tensor var_854 = const()[name = string("op_854"), val = tensor([1, 16, 64, 256])]; + tensor k_for_attn_3_cast_fp16 = reshape(shape = var_854, x = key_cache_updated_3_cast_fp16)[name = string("k_for_attn_3_cast_fp16")]; + tensor var_856 = const()[name = string("op_856"), val = tensor([1, 16, 64, 256])]; + tensor v_for_attn_3_cast_fp16 = reshape(shape = var_856, x = value_cache_updated_3_cast_fp16)[name = string("v_for_attn_3_cast_fp16")]; + bool var_860_transpose_x_1 = const()[name = string("op_860_transpose_x_1"), val = bool(true)]; + bool var_860_transpose_y_1 = const()[name = string("op_860_transpose_y_1"), val = bool(false)]; + tensor var_860_cast_fp16 = matmul(transpose_x = var_860_transpose_x_1, transpose_y = var_860_transpose_y_1, x = q_15_cast_fp16, y = k_for_attn_3_cast_fp16)[name = string("op_860_cast_fp16")]; + fp16 var_861_to_fp16 = const()[name = string("op_861_to_fp16"), val = fp16(0x1p-3)]; + tensor attn_weights_5_cast_fp16 = mul(x = var_860_cast_fp16, y = var_861_to_fp16)[name = string("attn_weights_5_cast_fp16")]; + tensor input_51_cast_fp16 = add(x = attn_weights_5_cast_fp16, y = attn_mask_1_cast_fp16)[name = string("input_51_cast_fp16")]; + tensor attn_weights_7_cast_fp16 = softmax(axis = var_28, x = input_51_cast_fp16)[name = string("attn_weights_7_cast_fp16")]; + bool attn_output_5_transpose_x_1 = const()[name = string("attn_output_5_transpose_x_1"), val = bool(false)]; + bool attn_output_5_transpose_y_1 = const()[name = string("attn_output_5_transpose_y_1"), val = bool(true)]; + tensor attn_output_5_cast_fp16 = matmul(transpose_x = attn_output_5_transpose_x_1, transpose_y = attn_output_5_transpose_y_1, x = attn_weights_7_cast_fp16, y = v_for_attn_3_cast_fp16)[name = string("attn_output_5_cast_fp16")]; + tensor var_870 = const()[name = string("op_870"), val = tensor([0, 1, 3, 2])]; + tensor var_872 = const()[name = string("op_872"), val = tensor([1, -1, 1, 1])]; + tensor attn_output_7_cast_fp16 = transpose(perm = var_870, x = attn_output_5_cast_fp16)[name = string("transpose_10")]; + tensor input_53_cast_fp16 = reshape(shape = var_872, x = attn_output_7_cast_fp16)[name = string("input_53_cast_fp16")]; + string x_11_pad_type_0 = const()[name = string("x_11_pad_type_0"), val = string("valid")]; + tensor x_11_strides_0 = const()[name = string("x_11_strides_0"), val = tensor([1, 1])]; + tensor x_11_pad_0 = const()[name = string("x_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_11_dilations_0 = const()[name = string("x_11_dilations_0"), val = tensor([1, 1])]; + int32 x_11_groups_0 = const()[name = string("x_11_groups_0"), val = int32(1)]; + tensor op_887_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16052864))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16577216))))[name = string("op_887_weight_0_to_fp16_palettized")]; + tensor var_887_bias_0_to_fp16 = const()[name = string("op_887_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16577792)))]; + tensor var_887_cast_fp16 = conv(bias = var_887_bias_0_to_fp16, dilations = x_11_dilations_0, groups = x_11_groups_0, pad = x_11_pad_0, pad_type = x_11_pad_type_0, strides = x_11_strides_0, weight = op_887_weight_0_to_fp16_palettized, x = input_53_cast_fp16)[name = string("op_887_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = var_887_cast_fp16)[name = string("inputs_7_cast_fp16")]; + tensor inputs_sq_7_cast_fp16 = mul(x = inputs_7_cast_fp16, y = inputs_7_cast_fp16)[name = string("inputs_sq_7_cast_fp16")]; + tensor variance_7_axes_0 = const()[name = string("variance_7_axes_0"), val = tensor([1])]; + bool variance_7_keep_dims_0 = const()[name = string("variance_7_keep_dims_0"), val = bool(true)]; + tensor variance_7_cast_fp16 = reduce_mean(axes = variance_7_axes_0, keep_dims = variance_7_keep_dims_0, x = inputs_sq_7_cast_fp16)[name = string("variance_7_cast_fp16")]; + fp16 var_893_to_fp16 = const()[name = string("op_893_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_894_cast_fp16 = add(x = variance_7_cast_fp16, y = var_893_to_fp16)[name = string("op_894_cast_fp16")]; + fp32 var_895_epsilon_0 = const()[name = string("op_895_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_895_cast_fp16 = rsqrt(epsilon = var_895_epsilon_0, x = var_894_cast_fp16)[name = string("op_895_cast_fp16")]; + tensor hidden_states_7_cast_fp16 = mul(x = inputs_7_cast_fp16, y = var_895_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; + tensor w_7_to_fp16 = const()[name = string("w_7_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16578880)))]; + tensor input_55_cast_fp16 = mul(x = w_7_to_fp16, y = hidden_states_7_cast_fp16)[name = string("input_55_cast_fp16")]; + string input_57_pad_type_0 = const()[name = string("input_57_pad_type_0"), val = string("valid")]; + tensor input_57_strides_0 = const()[name = string("input_57_strides_0"), val = tensor([1, 1])]; + tensor input_57_pad_0 = const()[name = string("input_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_57_dilations_0 = const()[name = string("input_57_dilations_0"), val = tensor([1, 1])]; + int32 input_57_groups_0 = const()[name = string("input_57_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_1_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16579968))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17104320))))[name = string("decoder_pre_transformer_layers_1_mlp_gate_proj_weight_to_fp16_palettized")]; + tensor input_57_cast_fp16 = conv(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 = decoder_pre_transformer_layers_1_mlp_gate_proj_weight_to_fp16_palettized, x = input_55_cast_fp16)[name = string("input_57_cast_fp16")]; + tensor var_909_cast_fp16 = silu(x = input_57_cast_fp16)[name = string("op_909_cast_fp16")]; + string var_915_pad_type_0 = const()[name = string("op_915_pad_type_0"), val = string("valid")]; + tensor var_915_strides_0 = const()[name = string("op_915_strides_0"), val = tensor([1, 1])]; + tensor var_915_pad_0 = const()[name = string("op_915_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_915_dilations_0 = const()[name = string("op_915_dilations_0"), val = tensor([1, 1])]; + int32 var_915_groups_0 = const()[name = string("op_915_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_1_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17104896))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17629248))))[name = string("decoder_pre_transformer_layers_1_mlp_up_proj_weight_to_fp16_palettized")]; + tensor var_915_cast_fp16 = conv(dilations = var_915_dilations_0, groups = var_915_groups_0, pad = var_915_pad_0, pad_type = var_915_pad_type_0, strides = var_915_strides_0, weight = decoder_pre_transformer_layers_1_mlp_up_proj_weight_to_fp16_palettized, x = input_55_cast_fp16)[name = string("op_915_cast_fp16")]; + tensor input_59_cast_fp16 = mul(x = var_909_cast_fp16, y = var_915_cast_fp16)[name = string("input_59_cast_fp16")]; + string x_13_pad_type_0 = const()[name = string("x_13_pad_type_0"), val = string("valid")]; + tensor x_13_strides_0 = const()[name = string("x_13_strides_0"), val = tensor([1, 1])]; + tensor x_13_pad_0 = const()[name = string("x_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_13_dilations_0 = const()[name = string("x_13_dilations_0"), val = tensor([1, 1])]; + int32 x_13_groups_0 = const()[name = string("x_13_groups_0"), val = int32(1)]; + tensor op_926_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17629824))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18154176))))[name = string("op_926_weight_0_to_fp16_palettized")]; + tensor var_926_bias_0_to_fp16 = const()[name = string("op_926_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18154752)))]; + tensor var_926_cast_fp16 = conv(bias = var_926_bias_0_to_fp16, dilations = x_13_dilations_0, groups = x_13_groups_0, pad = x_13_pad_0, pad_type = x_13_pad_type_0, strides = x_13_strides_0, weight = op_926_weight_0_to_fp16_palettized, x = input_59_cast_fp16)[name = string("op_926_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = var_926_cast_fp16)[name = string("inputs_9_cast_fp16")]; + tensor inputs_sq_9_cast_fp16 = mul(x = inputs_9_cast_fp16, y = inputs_9_cast_fp16)[name = string("inputs_sq_9_cast_fp16")]; + tensor variance_9_axes_0 = const()[name = string("variance_9_axes_0"), val = tensor([1])]; + bool variance_9_keep_dims_0 = const()[name = string("variance_9_keep_dims_0"), val = bool(true)]; + tensor variance_9_cast_fp16 = reduce_mean(axes = variance_9_axes_0, keep_dims = variance_9_keep_dims_0, x = inputs_sq_9_cast_fp16)[name = string("variance_9_cast_fp16")]; + fp16 var_942_to_fp16 = const()[name = string("op_942_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_943_cast_fp16 = add(x = variance_9_cast_fp16, y = var_942_to_fp16)[name = string("op_943_cast_fp16")]; + fp32 var_944_epsilon_0 = const()[name = string("op_944_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_944_cast_fp16 = rsqrt(epsilon = var_944_epsilon_0, x = var_943_cast_fp16)[name = string("op_944_cast_fp16")]; + tensor hidden_states_9_cast_fp16 = mul(x = inputs_9_cast_fp16, y = var_944_cast_fp16)[name = string("hidden_states_9_cast_fp16")]; + tensor w_9_to_fp16 = const()[name = string("w_9_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18155840)))]; + tensor x_15_cast_fp16 = mul(x = w_9_to_fp16, y = hidden_states_9_cast_fp16)[name = string("x_15_cast_fp16")]; + string q_17_pad_type_0 = const()[name = string("q_17_pad_type_0"), val = string("valid")]; + tensor q_17_strides_0 = const()[name = string("q_17_strides_0"), val = tensor([1, 1])]; + tensor q_17_pad_0 = const()[name = string("q_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_17_dilations_0 = const()[name = string("q_17_dilations_0"), val = tensor([1, 1])]; + int32 q_17_groups_0 = const()[name = string("q_17_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_2_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18156928))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18681280))))[name = string("decoder_pre_transformer_layers_2_self_attn_q_proj_weight_to_fp16_palettized")]; + tensor q_17_cast_fp16 = conv(dilations = q_17_dilations_0, groups = q_17_groups_0, pad = q_17_pad_0, pad_type = q_17_pad_type_0, strides = q_17_strides_0, weight = decoder_pre_transformer_layers_2_self_attn_q_proj_weight_to_fp16_palettized, x = x_15_cast_fp16)[name = string("q_17_cast_fp16")]; + string k_9_pad_type_0 = const()[name = string("k_9_pad_type_0"), val = string("valid")]; + tensor k_9_strides_0 = const()[name = string("k_9_strides_0"), val = tensor([1, 1])]; + tensor k_9_pad_0 = const()[name = string("k_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_9_dilations_0 = const()[name = string("k_9_dilations_0"), val = tensor([1, 1])]; + int32 k_9_groups_0 = const()[name = string("k_9_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_2_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18681856))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19206208))))[name = string("decoder_pre_transformer_layers_2_self_attn_k_proj_weight_to_fp16_palettized")]; + tensor k_9_cast_fp16 = conv(dilations = k_9_dilations_0, groups = k_9_groups_0, pad = k_9_pad_0, pad_type = k_9_pad_type_0, strides = k_9_strides_0, weight = decoder_pre_transformer_layers_2_self_attn_k_proj_weight_to_fp16_palettized, x = x_15_cast_fp16)[name = string("k_9_cast_fp16")]; + string v_5_pad_type_0 = const()[name = string("v_5_pad_type_0"), val = string("valid")]; + tensor v_5_strides_0 = const()[name = string("v_5_strides_0"), val = tensor([1, 1])]; + tensor v_5_pad_0 = const()[name = string("v_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_5_dilations_0 = const()[name = string("v_5_dilations_0"), val = tensor([1, 1])]; + int32 v_5_groups_0 = const()[name = string("v_5_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_2_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19206784))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19731136))))[name = string("decoder_pre_transformer_layers_2_self_attn_v_proj_weight_to_fp16_palettized")]; + tensor v_5_cast_fp16 = conv(dilations = v_5_dilations_0, groups = v_5_groups_0, pad = v_5_pad_0, pad_type = v_5_pad_type_0, strides = v_5_strides_0, weight = decoder_pre_transformer_layers_2_self_attn_v_proj_weight_to_fp16_palettized, x = x_15_cast_fp16)[name = string("v_5_cast_fp16")]; + tensor var_976 = const()[name = string("op_976"), val = tensor([16, 64, 1, 1])]; + tensor q_19_cast_fp16 = reshape(shape = var_976, x = q_17_cast_fp16)[name = string("q_19_cast_fp16")]; + tensor var_981 = const()[name = string("op_981"), val = tensor([16, 64, 1, 1])]; + tensor k_11_cast_fp16 = reshape(shape = var_981, x = k_9_cast_fp16)[name = string("k_11_cast_fp16")]; + tensor var_993_cast_fp16 = mul(x = q_19_cast_fp16, y = cos_1_cast_fp16)[name = string("op_993_cast_fp16")]; + tensor var_998_begin_0 = const()[name = string("op_998_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_998_end_0 = const()[name = string("op_998_end_0"), val = tensor([16, 32, 1, 1])]; + tensor var_998_end_mask_0 = const()[name = string("op_998_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_998_cast_fp16 = slice_by_index(begin = var_998_begin_0, end = var_998_end_0, end_mask = var_998_end_mask_0, x = q_19_cast_fp16)[name = string("op_998_cast_fp16")]; + tensor var_1005_begin_0 = const()[name = string("op_1005_begin_0"), val = tensor([0, 32, 0, 0])]; + tensor var_1005_end_0 = const()[name = string("op_1005_end_0"), val = tensor([16, 64, 1, 1])]; + tensor var_1005_end_mask_0 = const()[name = string("op_1005_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1005_cast_fp16 = slice_by_index(begin = var_1005_begin_0, end = var_1005_end_0, end_mask = var_1005_end_mask_0, x = q_19_cast_fp16)[name = string("op_1005_cast_fp16")]; + fp16 const_40_promoted_to_fp16 = const()[name = string("const_40_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_1008_cast_fp16 = mul(x = var_1005_cast_fp16, y = const_40_promoted_to_fp16)[name = string("op_1008_cast_fp16")]; + bool var_1010_interleave_0 = const()[name = string("op_1010_interleave_0"), val = bool(false)]; + tensor var_1010_cast_fp16 = concat(axis = var_32, interleave = var_1010_interleave_0, values = (var_1008_cast_fp16, var_998_cast_fp16))[name = string("op_1010_cast_fp16")]; + tensor var_1011_cast_fp16 = mul(x = var_1010_cast_fp16, y = sin_1_cast_fp16)[name = string("op_1011_cast_fp16")]; + tensor q_rotated_5_cast_fp16 = add(x = var_993_cast_fp16, y = var_1011_cast_fp16)[name = string("q_rotated_5_cast_fp16")]; + tensor var_1013_cast_fp16 = mul(x = k_11_cast_fp16, y = cos_1_cast_fp16)[name = string("op_1013_cast_fp16")]; + tensor var_1018_begin_0 = const()[name = string("op_1018_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1018_end_0 = const()[name = string("op_1018_end_0"), val = tensor([16, 32, 1, 1])]; + tensor var_1018_end_mask_0 = const()[name = string("op_1018_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1018_cast_fp16 = slice_by_index(begin = var_1018_begin_0, end = var_1018_end_0, end_mask = var_1018_end_mask_0, x = k_11_cast_fp16)[name = string("op_1018_cast_fp16")]; + tensor var_1025_begin_0 = const()[name = string("op_1025_begin_0"), val = tensor([0, 32, 0, 0])]; + tensor var_1025_end_0 = const()[name = string("op_1025_end_0"), val = tensor([16, 64, 1, 1])]; + tensor var_1025_end_mask_0 = const()[name = string("op_1025_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1025_cast_fp16 = slice_by_index(begin = var_1025_begin_0, end = var_1025_end_0, end_mask = var_1025_end_mask_0, x = k_11_cast_fp16)[name = string("op_1025_cast_fp16")]; + fp16 const_43_promoted_to_fp16 = const()[name = string("const_43_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_1028_cast_fp16 = mul(x = var_1025_cast_fp16, y = const_43_promoted_to_fp16)[name = string("op_1028_cast_fp16")]; + bool var_1030_interleave_0 = const()[name = string("op_1030_interleave_0"), val = bool(false)]; + tensor var_1030_cast_fp16 = concat(axis = var_32, interleave = var_1030_interleave_0, values = (var_1028_cast_fp16, var_1018_cast_fp16))[name = string("op_1030_cast_fp16")]; + tensor var_1031_cast_fp16 = mul(x = var_1030_cast_fp16, y = sin_1_cast_fp16)[name = string("op_1031_cast_fp16")]; + tensor k_rotated_5_cast_fp16 = add(x = var_1013_cast_fp16, y = var_1031_cast_fp16)[name = string("k_rotated_5_cast_fp16")]; + tensor var_1035 = const()[name = string("op_1035"), val = tensor([1, 1024, 1, 1])]; + tensor current_key_5_cast_fp16 = reshape(shape = var_1035, x = k_rotated_5_cast_fp16)[name = string("current_key_5_cast_fp16")]; + tensor var_1040_cast_fp16 = mul(x = var_522_cast_fp16_2, y = var_647_cast_fp16)[name = string("op_1040_cast_fp16")]; + tensor var_1041_cast_fp16 = mul(x = current_key_5_cast_fp16, y = update_mask_1_cast_fp16)[name = string("op_1041_cast_fp16")]; + tensor key_cache_updated_5_cast_fp16 = add(x = var_1040_cast_fp16, y = var_1041_cast_fp16)[name = string("key_cache_updated_5_cast_fp16")]; + tensor var_1044_cast_fp16 = mul(x = var_531_cast_fp16_2, y = var_647_cast_fp16)[name = string("op_1044_cast_fp16")]; + tensor var_1045_cast_fp16 = mul(x = v_5_cast_fp16, y = update_mask_1_cast_fp16)[name = string("op_1045_cast_fp16")]; + tensor value_cache_updated_5_cast_fp16 = add(x = var_1044_cast_fp16, y = var_1045_cast_fp16)[name = string("value_cache_updated_5_cast_fp16")]; + tensor var_1047 = const()[name = string("op_1047"), val = tensor([1, 16, 64, 1])]; + tensor q_23_cast_fp16 = reshape(shape = var_1047, x = q_rotated_5_cast_fp16)[name = string("q_23_cast_fp16")]; + tensor var_1050 = const()[name = string("op_1050"), val = tensor([1, 16, 64, 256])]; + tensor k_for_attn_5_cast_fp16 = reshape(shape = var_1050, x = key_cache_updated_5_cast_fp16)[name = string("k_for_attn_5_cast_fp16")]; + tensor var_1052 = const()[name = string("op_1052"), val = tensor([1, 16, 64, 256])]; + tensor v_for_attn_5_cast_fp16 = reshape(shape = var_1052, x = value_cache_updated_5_cast_fp16)[name = string("v_for_attn_5_cast_fp16")]; + bool var_1056_transpose_x_1 = const()[name = string("op_1056_transpose_x_1"), val = bool(true)]; + bool var_1056_transpose_y_1 = const()[name = string("op_1056_transpose_y_1"), val = bool(false)]; + tensor var_1056_cast_fp16 = matmul(transpose_x = var_1056_transpose_x_1, transpose_y = var_1056_transpose_y_1, x = q_23_cast_fp16, y = k_for_attn_5_cast_fp16)[name = string("op_1056_cast_fp16")]; + fp16 var_1057_to_fp16 = const()[name = string("op_1057_to_fp16"), val = fp16(0x1p-3)]; + tensor attn_weights_9_cast_fp16 = mul(x = var_1056_cast_fp16, y = var_1057_to_fp16)[name = string("attn_weights_9_cast_fp16")]; + tensor input_61_cast_fp16 = add(x = attn_weights_9_cast_fp16, y = attn_mask_1_cast_fp16)[name = string("input_61_cast_fp16")]; + tensor attn_weights_11_cast_fp16 = softmax(axis = var_28, x = input_61_cast_fp16)[name = string("attn_weights_11_cast_fp16")]; + bool attn_output_9_transpose_x_1 = const()[name = string("attn_output_9_transpose_x_1"), val = bool(false)]; + bool attn_output_9_transpose_y_1 = const()[name = string("attn_output_9_transpose_y_1"), val = bool(true)]; + tensor attn_output_9_cast_fp16 = matmul(transpose_x = attn_output_9_transpose_x_1, transpose_y = attn_output_9_transpose_y_1, x = attn_weights_11_cast_fp16, y = v_for_attn_5_cast_fp16)[name = string("attn_output_9_cast_fp16")]; + tensor var_1066 = const()[name = string("op_1066"), val = tensor([0, 1, 3, 2])]; + tensor var_1068 = const()[name = string("op_1068"), val = tensor([1, -1, 1, 1])]; + tensor attn_output_11_cast_fp16 = transpose(perm = var_1066, x = attn_output_9_cast_fp16)[name = string("transpose_9")]; + tensor input_63_cast_fp16 = reshape(shape = var_1068, x = attn_output_11_cast_fp16)[name = string("input_63_cast_fp16")]; + string x_17_pad_type_0 = const()[name = string("x_17_pad_type_0"), val = string("valid")]; + tensor x_17_strides_0 = const()[name = string("x_17_strides_0"), val = tensor([1, 1])]; + tensor x_17_pad_0 = const()[name = string("x_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_17_dilations_0 = const()[name = string("x_17_dilations_0"), val = tensor([1, 1])]; + int32 x_17_groups_0 = const()[name = string("x_17_groups_0"), val = int32(1)]; + tensor op_1083_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19731712))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20256064))))[name = string("op_1083_weight_0_to_fp16_palettized")]; + tensor var_1083_bias_0_to_fp16 = const()[name = string("op_1083_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20256640)))]; + tensor var_1083_cast_fp16 = conv(bias = var_1083_bias_0_to_fp16, dilations = x_17_dilations_0, groups = x_17_groups_0, pad = x_17_pad_0, pad_type = x_17_pad_type_0, strides = x_17_strides_0, weight = op_1083_weight_0_to_fp16_palettized, x = input_63_cast_fp16)[name = string("op_1083_cast_fp16")]; + tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = var_1083_cast_fp16)[name = string("inputs_11_cast_fp16")]; + tensor inputs_sq_11_cast_fp16 = mul(x = inputs_11_cast_fp16, y = inputs_11_cast_fp16)[name = string("inputs_sq_11_cast_fp16")]; + tensor variance_11_axes_0 = const()[name = string("variance_11_axes_0"), val = tensor([1])]; + bool variance_11_keep_dims_0 = const()[name = string("variance_11_keep_dims_0"), val = bool(true)]; + tensor variance_11_cast_fp16 = reduce_mean(axes = variance_11_axes_0, keep_dims = variance_11_keep_dims_0, x = inputs_sq_11_cast_fp16)[name = string("variance_11_cast_fp16")]; + fp16 var_1089_to_fp16 = const()[name = string("op_1089_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1090_cast_fp16 = add(x = variance_11_cast_fp16, y = var_1089_to_fp16)[name = string("op_1090_cast_fp16")]; + fp32 var_1091_epsilon_0 = const()[name = string("op_1091_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_1091_cast_fp16 = rsqrt(epsilon = var_1091_epsilon_0, x = var_1090_cast_fp16)[name = string("op_1091_cast_fp16")]; + tensor hidden_states_11_cast_fp16 = mul(x = inputs_11_cast_fp16, y = var_1091_cast_fp16)[name = string("hidden_states_11_cast_fp16")]; + tensor w_11_to_fp16 = const()[name = string("w_11_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20257728)))]; + tensor input_65_cast_fp16 = mul(x = w_11_to_fp16, y = hidden_states_11_cast_fp16)[name = string("input_65_cast_fp16")]; + string input_67_pad_type_0 = const()[name = string("input_67_pad_type_0"), val = string("valid")]; + tensor input_67_strides_0 = const()[name = string("input_67_strides_0"), val = tensor([1, 1])]; + tensor input_67_pad_0 = const()[name = string("input_67_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_67_dilations_0 = const()[name = string("input_67_dilations_0"), val = tensor([1, 1])]; + int32 input_67_groups_0 = const()[name = string("input_67_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_2_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20258816))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20783168))))[name = string("decoder_pre_transformer_layers_2_mlp_gate_proj_weight_to_fp16_palettized")]; + tensor input_67_cast_fp16 = conv(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 = decoder_pre_transformer_layers_2_mlp_gate_proj_weight_to_fp16_palettized, x = input_65_cast_fp16)[name = string("input_67_cast_fp16")]; + tensor var_1105_cast_fp16 = silu(x = input_67_cast_fp16)[name = string("op_1105_cast_fp16")]; + string var_1111_pad_type_0 = const()[name = string("op_1111_pad_type_0"), val = string("valid")]; + tensor var_1111_strides_0 = const()[name = string("op_1111_strides_0"), val = tensor([1, 1])]; + tensor var_1111_pad_0 = const()[name = string("op_1111_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1111_dilations_0 = const()[name = string("op_1111_dilations_0"), val = tensor([1, 1])]; + int32 var_1111_groups_0 = const()[name = string("op_1111_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_2_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20783744))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21308096))))[name = string("decoder_pre_transformer_layers_2_mlp_up_proj_weight_to_fp16_palettized")]; + tensor var_1111_cast_fp16 = conv(dilations = var_1111_dilations_0, groups = var_1111_groups_0, pad = var_1111_pad_0, pad_type = var_1111_pad_type_0, strides = var_1111_strides_0, weight = decoder_pre_transformer_layers_2_mlp_up_proj_weight_to_fp16_palettized, x = input_65_cast_fp16)[name = string("op_1111_cast_fp16")]; + tensor input_69_cast_fp16 = mul(x = var_1105_cast_fp16, y = var_1111_cast_fp16)[name = string("input_69_cast_fp16")]; + string x_19_pad_type_0 = const()[name = string("x_19_pad_type_0"), val = string("valid")]; + tensor x_19_strides_0 = const()[name = string("x_19_strides_0"), val = tensor([1, 1])]; + tensor x_19_pad_0 = const()[name = string("x_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_19_dilations_0 = const()[name = string("x_19_dilations_0"), val = tensor([1, 1])]; + int32 x_19_groups_0 = const()[name = string("x_19_groups_0"), val = int32(1)]; + tensor op_1122_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21308672))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21833024))))[name = string("op_1122_weight_0_to_fp16_palettized")]; + tensor var_1122_bias_0_to_fp16 = const()[name = string("op_1122_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21833600)))]; + tensor var_1122_cast_fp16 = conv(bias = var_1122_bias_0_to_fp16, dilations = x_19_dilations_0, groups = x_19_groups_0, pad = x_19_pad_0, pad_type = x_19_pad_type_0, strides = x_19_strides_0, weight = op_1122_weight_0_to_fp16_palettized, x = input_69_cast_fp16)[name = string("op_1122_cast_fp16")]; + tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = var_1122_cast_fp16)[name = string("inputs_13_cast_fp16")]; + tensor inputs_sq_13_cast_fp16 = mul(x = inputs_13_cast_fp16, y = inputs_13_cast_fp16)[name = string("inputs_sq_13_cast_fp16")]; + tensor variance_13_axes_0 = const()[name = string("variance_13_axes_0"), val = tensor([1])]; + bool variance_13_keep_dims_0 = const()[name = string("variance_13_keep_dims_0"), val = bool(true)]; + tensor variance_13_cast_fp16 = reduce_mean(axes = variance_13_axes_0, keep_dims = variance_13_keep_dims_0, x = inputs_sq_13_cast_fp16)[name = string("variance_13_cast_fp16")]; + fp16 var_1138_to_fp16 = const()[name = string("op_1138_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1139_cast_fp16 = add(x = variance_13_cast_fp16, y = var_1138_to_fp16)[name = string("op_1139_cast_fp16")]; + fp32 var_1140_epsilon_0 = const()[name = string("op_1140_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_1140_cast_fp16 = rsqrt(epsilon = var_1140_epsilon_0, x = var_1139_cast_fp16)[name = string("op_1140_cast_fp16")]; + tensor hidden_states_13_cast_fp16 = mul(x = inputs_13_cast_fp16, y = var_1140_cast_fp16)[name = string("hidden_states_13_cast_fp16")]; + tensor w_13_to_fp16 = const()[name = string("w_13_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21834688)))]; + tensor x_21_cast_fp16 = mul(x = w_13_to_fp16, y = hidden_states_13_cast_fp16)[name = string("x_21_cast_fp16")]; + string q_25_pad_type_0 = const()[name = string("q_25_pad_type_0"), val = string("valid")]; + tensor q_25_strides_0 = const()[name = string("q_25_strides_0"), val = tensor([1, 1])]; + tensor q_25_pad_0 = const()[name = string("q_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_25_dilations_0 = const()[name = string("q_25_dilations_0"), val = tensor([1, 1])]; + int32 q_25_groups_0 = const()[name = string("q_25_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_3_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21835776))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22360128))))[name = string("decoder_pre_transformer_layers_3_self_attn_q_proj_weight_to_fp16_palettized")]; + tensor q_25_cast_fp16 = conv(dilations = q_25_dilations_0, groups = q_25_groups_0, pad = q_25_pad_0, pad_type = q_25_pad_type_0, strides = q_25_strides_0, weight = decoder_pre_transformer_layers_3_self_attn_q_proj_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("q_25_cast_fp16")]; + string k_13_pad_type_0 = const()[name = string("k_13_pad_type_0"), val = string("valid")]; + tensor k_13_strides_0 = const()[name = string("k_13_strides_0"), val = tensor([1, 1])]; + tensor k_13_pad_0 = const()[name = string("k_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_13_dilations_0 = const()[name = string("k_13_dilations_0"), val = tensor([1, 1])]; + int32 k_13_groups_0 = const()[name = string("k_13_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_3_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22360704))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22885056))))[name = string("decoder_pre_transformer_layers_3_self_attn_k_proj_weight_to_fp16_palettized")]; + tensor k_13_cast_fp16 = conv(dilations = k_13_dilations_0, groups = k_13_groups_0, pad = k_13_pad_0, pad_type = k_13_pad_type_0, strides = k_13_strides_0, weight = decoder_pre_transformer_layers_3_self_attn_k_proj_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("k_13_cast_fp16")]; + string v_7_pad_type_0 = const()[name = string("v_7_pad_type_0"), val = string("valid")]; + tensor v_7_strides_0 = const()[name = string("v_7_strides_0"), val = tensor([1, 1])]; + tensor v_7_pad_0 = const()[name = string("v_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_7_dilations_0 = const()[name = string("v_7_dilations_0"), val = tensor([1, 1])]; + int32 v_7_groups_0 = const()[name = string("v_7_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_3_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22885632))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23409984))))[name = string("decoder_pre_transformer_layers_3_self_attn_v_proj_weight_to_fp16_palettized")]; + tensor v_7_cast_fp16 = conv(dilations = v_7_dilations_0, groups = v_7_groups_0, pad = v_7_pad_0, pad_type = v_7_pad_type_0, strides = v_7_strides_0, weight = decoder_pre_transformer_layers_3_self_attn_v_proj_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("v_7_cast_fp16")]; + tensor var_1172 = const()[name = string("op_1172"), val = tensor([16, 64, 1, 1])]; + tensor q_27_cast_fp16 = reshape(shape = var_1172, x = q_25_cast_fp16)[name = string("q_27_cast_fp16")]; + tensor var_1177 = const()[name = string("op_1177"), val = tensor([16, 64, 1, 1])]; + tensor k_15_cast_fp16 = reshape(shape = var_1177, x = k_13_cast_fp16)[name = string("k_15_cast_fp16")]; + tensor var_1189_cast_fp16 = mul(x = q_27_cast_fp16, y = cos_1_cast_fp16)[name = string("op_1189_cast_fp16")]; + tensor var_1194_begin_0 = const()[name = string("op_1194_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1194_end_0 = const()[name = string("op_1194_end_0"), val = tensor([16, 32, 1, 1])]; + tensor var_1194_end_mask_0 = const()[name = string("op_1194_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1194_cast_fp16 = slice_by_index(begin = var_1194_begin_0, end = var_1194_end_0, end_mask = var_1194_end_mask_0, x = q_27_cast_fp16)[name = string("op_1194_cast_fp16")]; + tensor var_1201_begin_0 = const()[name = string("op_1201_begin_0"), val = tensor([0, 32, 0, 0])]; + tensor var_1201_end_0 = const()[name = string("op_1201_end_0"), val = tensor([16, 64, 1, 1])]; + tensor var_1201_end_mask_0 = const()[name = string("op_1201_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1201_cast_fp16 = slice_by_index(begin = var_1201_begin_0, end = var_1201_end_0, end_mask = var_1201_end_mask_0, x = q_27_cast_fp16)[name = string("op_1201_cast_fp16")]; + fp16 const_48_promoted_to_fp16 = const()[name = string("const_48_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_1204_cast_fp16 = mul(x = var_1201_cast_fp16, y = const_48_promoted_to_fp16)[name = string("op_1204_cast_fp16")]; + bool var_1206_interleave_0 = const()[name = string("op_1206_interleave_0"), val = bool(false)]; + tensor var_1206_cast_fp16 = concat(axis = var_32, interleave = var_1206_interleave_0, values = (var_1204_cast_fp16, var_1194_cast_fp16))[name = string("op_1206_cast_fp16")]; + tensor var_1207_cast_fp16 = mul(x = var_1206_cast_fp16, y = sin_1_cast_fp16)[name = string("op_1207_cast_fp16")]; + tensor q_rotated_7_cast_fp16 = add(x = var_1189_cast_fp16, y = var_1207_cast_fp16)[name = string("q_rotated_7_cast_fp16")]; + tensor var_1209_cast_fp16 = mul(x = k_15_cast_fp16, y = cos_1_cast_fp16)[name = string("op_1209_cast_fp16")]; + tensor var_1214_begin_0 = const()[name = string("op_1214_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1214_end_0 = const()[name = string("op_1214_end_0"), val = tensor([16, 32, 1, 1])]; + tensor var_1214_end_mask_0 = const()[name = string("op_1214_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1214_cast_fp16 = slice_by_index(begin = var_1214_begin_0, end = var_1214_end_0, end_mask = var_1214_end_mask_0, x = k_15_cast_fp16)[name = string("op_1214_cast_fp16")]; + tensor var_1221_begin_0 = const()[name = string("op_1221_begin_0"), val = tensor([0, 32, 0, 0])]; + tensor var_1221_end_0 = const()[name = string("op_1221_end_0"), val = tensor([16, 64, 1, 1])]; + tensor var_1221_end_mask_0 = const()[name = string("op_1221_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1221_cast_fp16 = slice_by_index(begin = var_1221_begin_0, end = var_1221_end_0, end_mask = var_1221_end_mask_0, x = k_15_cast_fp16)[name = string("op_1221_cast_fp16")]; + fp16 const_51_promoted_to_fp16 = const()[name = string("const_51_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_1224_cast_fp16 = mul(x = var_1221_cast_fp16, y = const_51_promoted_to_fp16)[name = string("op_1224_cast_fp16")]; + bool var_1226_interleave_0 = const()[name = string("op_1226_interleave_0"), val = bool(false)]; + tensor var_1226_cast_fp16 = concat(axis = var_32, interleave = var_1226_interleave_0, values = (var_1224_cast_fp16, var_1214_cast_fp16))[name = string("op_1226_cast_fp16")]; + tensor var_1227_cast_fp16 = mul(x = var_1226_cast_fp16, y = sin_1_cast_fp16)[name = string("op_1227_cast_fp16")]; + tensor k_rotated_7_cast_fp16 = add(x = var_1209_cast_fp16, y = var_1227_cast_fp16)[name = string("k_rotated_7_cast_fp16")]; + tensor var_1231 = const()[name = string("op_1231"), val = tensor([1, 1024, 1, 1])]; + tensor current_key_7_cast_fp16 = reshape(shape = var_1231, x = k_rotated_7_cast_fp16)[name = string("current_key_7_cast_fp16")]; + tensor var_1236_cast_fp16 = mul(x = var_522_cast_fp16_3, y = var_647_cast_fp16)[name = string("op_1236_cast_fp16")]; + tensor var_1237_cast_fp16 = mul(x = current_key_7_cast_fp16, y = update_mask_1_cast_fp16)[name = string("op_1237_cast_fp16")]; + tensor key_cache_updated_7_cast_fp16 = add(x = var_1236_cast_fp16, y = var_1237_cast_fp16)[name = string("key_cache_updated_7_cast_fp16")]; + tensor var_1240_cast_fp16 = mul(x = var_531_cast_fp16_3, y = var_647_cast_fp16)[name = string("op_1240_cast_fp16")]; + tensor var_1241_cast_fp16 = mul(x = v_7_cast_fp16, y = update_mask_1_cast_fp16)[name = string("op_1241_cast_fp16")]; + tensor value_cache_updated_7_cast_fp16 = add(x = var_1240_cast_fp16, y = var_1241_cast_fp16)[name = string("value_cache_updated_7_cast_fp16")]; + tensor var_1243 = const()[name = string("op_1243"), val = tensor([1, 16, 64, 1])]; + tensor q_31_cast_fp16 = reshape(shape = var_1243, x = q_rotated_7_cast_fp16)[name = string("q_31_cast_fp16")]; + tensor var_1246 = const()[name = string("op_1246"), val = tensor([1, 16, 64, 256])]; + tensor k_for_attn_7_cast_fp16 = reshape(shape = var_1246, x = key_cache_updated_7_cast_fp16)[name = string("k_for_attn_7_cast_fp16")]; + tensor var_1248 = const()[name = string("op_1248"), val = tensor([1, 16, 64, 256])]; + tensor v_for_attn_7_cast_fp16 = reshape(shape = var_1248, x = value_cache_updated_7_cast_fp16)[name = string("v_for_attn_7_cast_fp16")]; + bool var_1252_transpose_x_1 = const()[name = string("op_1252_transpose_x_1"), val = bool(true)]; + bool var_1252_transpose_y_1 = const()[name = string("op_1252_transpose_y_1"), val = bool(false)]; + tensor var_1252_cast_fp16 = matmul(transpose_x = var_1252_transpose_x_1, transpose_y = var_1252_transpose_y_1, x = q_31_cast_fp16, y = k_for_attn_7_cast_fp16)[name = string("op_1252_cast_fp16")]; + fp16 var_1253_to_fp16 = const()[name = string("op_1253_to_fp16"), val = fp16(0x1p-3)]; + tensor attn_weights_13_cast_fp16 = mul(x = var_1252_cast_fp16, y = var_1253_to_fp16)[name = string("attn_weights_13_cast_fp16")]; + tensor input_71_cast_fp16 = add(x = attn_weights_13_cast_fp16, y = attn_mask_1_cast_fp16)[name = string("input_71_cast_fp16")]; + tensor attn_weights_15_cast_fp16 = softmax(axis = var_28, x = input_71_cast_fp16)[name = string("attn_weights_15_cast_fp16")]; + bool attn_output_13_transpose_x_1 = const()[name = string("attn_output_13_transpose_x_1"), val = bool(false)]; + bool attn_output_13_transpose_y_1 = const()[name = string("attn_output_13_transpose_y_1"), val = bool(true)]; + tensor attn_output_13_cast_fp16 = matmul(transpose_x = attn_output_13_transpose_x_1, transpose_y = attn_output_13_transpose_y_1, x = attn_weights_15_cast_fp16, y = v_for_attn_7_cast_fp16)[name = string("attn_output_13_cast_fp16")]; + tensor var_1262 = const()[name = string("op_1262"), val = tensor([0, 1, 3, 2])]; + tensor var_1264 = const()[name = string("op_1264"), val = tensor([1, -1, 1, 1])]; + tensor attn_output_15_cast_fp16 = transpose(perm = var_1262, x = attn_output_13_cast_fp16)[name = string("transpose_8")]; + tensor input_73_cast_fp16 = reshape(shape = var_1264, x = attn_output_15_cast_fp16)[name = string("input_73_cast_fp16")]; + string x_23_pad_type_0 = const()[name = string("x_23_pad_type_0"), val = string("valid")]; + tensor x_23_strides_0 = const()[name = string("x_23_strides_0"), val = tensor([1, 1])]; + tensor x_23_pad_0 = const()[name = string("x_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_23_dilations_0 = const()[name = string("x_23_dilations_0"), val = tensor([1, 1])]; + int32 x_23_groups_0 = const()[name = string("x_23_groups_0"), val = int32(1)]; + tensor op_1279_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23410560))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23934912))))[name = string("op_1279_weight_0_to_fp16_palettized")]; + tensor var_1279_bias_0_to_fp16 = const()[name = string("op_1279_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23935488)))]; + tensor var_1279_cast_fp16 = conv(bias = var_1279_bias_0_to_fp16, dilations = x_23_dilations_0, groups = x_23_groups_0, pad = x_23_pad_0, pad_type = x_23_pad_type_0, strides = x_23_strides_0, weight = op_1279_weight_0_to_fp16_palettized, x = input_73_cast_fp16)[name = string("op_1279_cast_fp16")]; + tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = var_1279_cast_fp16)[name = string("inputs_15_cast_fp16")]; + tensor inputs_sq_15_cast_fp16 = mul(x = inputs_15_cast_fp16, y = inputs_15_cast_fp16)[name = string("inputs_sq_15_cast_fp16")]; + tensor variance_15_axes_0 = const()[name = string("variance_15_axes_0"), val = tensor([1])]; + bool variance_15_keep_dims_0 = const()[name = string("variance_15_keep_dims_0"), val = bool(true)]; + tensor variance_15_cast_fp16 = reduce_mean(axes = variance_15_axes_0, keep_dims = variance_15_keep_dims_0, x = inputs_sq_15_cast_fp16)[name = string("variance_15_cast_fp16")]; + fp16 var_1285_to_fp16 = const()[name = string("op_1285_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1286_cast_fp16 = add(x = variance_15_cast_fp16, y = var_1285_to_fp16)[name = string("op_1286_cast_fp16")]; + fp32 var_1287_epsilon_0 = const()[name = string("op_1287_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_1287_cast_fp16 = rsqrt(epsilon = var_1287_epsilon_0, x = var_1286_cast_fp16)[name = string("op_1287_cast_fp16")]; + tensor hidden_states_15_cast_fp16 = mul(x = inputs_15_cast_fp16, y = var_1287_cast_fp16)[name = string("hidden_states_15_cast_fp16")]; + tensor w_15_to_fp16 = const()[name = string("w_15_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23936576)))]; + tensor input_75_cast_fp16 = mul(x = w_15_to_fp16, y = hidden_states_15_cast_fp16)[name = string("input_75_cast_fp16")]; + string input_77_pad_type_0 = const()[name = string("input_77_pad_type_0"), val = string("valid")]; + tensor input_77_strides_0 = const()[name = string("input_77_strides_0"), val = tensor([1, 1])]; + tensor input_77_pad_0 = const()[name = string("input_77_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_77_dilations_0 = const()[name = string("input_77_dilations_0"), val = tensor([1, 1])]; + int32 input_77_groups_0 = const()[name = string("input_77_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_3_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23937664))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24462016))))[name = string("decoder_pre_transformer_layers_3_mlp_gate_proj_weight_to_fp16_palettized")]; + tensor input_77_cast_fp16 = conv(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 = decoder_pre_transformer_layers_3_mlp_gate_proj_weight_to_fp16_palettized, x = input_75_cast_fp16)[name = string("input_77_cast_fp16")]; + tensor var_1301_cast_fp16 = silu(x = input_77_cast_fp16)[name = string("op_1301_cast_fp16")]; + string var_1307_pad_type_0 = const()[name = string("op_1307_pad_type_0"), val = string("valid")]; + tensor var_1307_strides_0 = const()[name = string("op_1307_strides_0"), val = tensor([1, 1])]; + tensor var_1307_pad_0 = const()[name = string("op_1307_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1307_dilations_0 = const()[name = string("op_1307_dilations_0"), val = tensor([1, 1])]; + int32 var_1307_groups_0 = const()[name = string("op_1307_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_3_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24462592))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24986944))))[name = string("decoder_pre_transformer_layers_3_mlp_up_proj_weight_to_fp16_palettized")]; + tensor var_1307_cast_fp16 = conv(dilations = var_1307_dilations_0, groups = var_1307_groups_0, pad = var_1307_pad_0, pad_type = var_1307_pad_type_0, strides = var_1307_strides_0, weight = decoder_pre_transformer_layers_3_mlp_up_proj_weight_to_fp16_palettized, x = input_75_cast_fp16)[name = string("op_1307_cast_fp16")]; + tensor input_79_cast_fp16 = mul(x = var_1301_cast_fp16, y = var_1307_cast_fp16)[name = string("input_79_cast_fp16")]; + string x_25_pad_type_0 = const()[name = string("x_25_pad_type_0"), val = string("valid")]; + tensor x_25_strides_0 = const()[name = string("x_25_strides_0"), val = tensor([1, 1])]; + tensor x_25_pad_0 = const()[name = string("x_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_25_dilations_0 = const()[name = string("x_25_dilations_0"), val = tensor([1, 1])]; + int32 x_25_groups_0 = const()[name = string("x_25_groups_0"), val = int32(1)]; + tensor op_1318_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24987520))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25511872))))[name = string("op_1318_weight_0_to_fp16_palettized")]; + tensor var_1318_bias_0_to_fp16 = const()[name = string("op_1318_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25512448)))]; + tensor var_1318_cast_fp16 = conv(bias = var_1318_bias_0_to_fp16, dilations = x_25_dilations_0, groups = x_25_groups_0, pad = x_25_pad_0, pad_type = x_25_pad_type_0, strides = x_25_strides_0, weight = op_1318_weight_0_to_fp16_palettized, x = input_79_cast_fp16)[name = string("op_1318_cast_fp16")]; + tensor inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = var_1318_cast_fp16)[name = string("inputs_17_cast_fp16")]; + tensor inputs_sq_17_cast_fp16 = mul(x = inputs_17_cast_fp16, y = inputs_17_cast_fp16)[name = string("inputs_sq_17_cast_fp16")]; + tensor variance_17_axes_0 = const()[name = string("variance_17_axes_0"), val = tensor([1])]; + bool variance_17_keep_dims_0 = const()[name = string("variance_17_keep_dims_0"), val = bool(true)]; + tensor variance_17_cast_fp16 = reduce_mean(axes = variance_17_axes_0, keep_dims = variance_17_keep_dims_0, x = inputs_sq_17_cast_fp16)[name = string("variance_17_cast_fp16")]; + fp16 var_1334_to_fp16 = const()[name = string("op_1334_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1335_cast_fp16 = add(x = variance_17_cast_fp16, y = var_1334_to_fp16)[name = string("op_1335_cast_fp16")]; + fp32 var_1336_epsilon_0 = const()[name = string("op_1336_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_1336_cast_fp16 = rsqrt(epsilon = var_1336_epsilon_0, x = var_1335_cast_fp16)[name = string("op_1336_cast_fp16")]; + tensor hidden_states_17_cast_fp16 = mul(x = inputs_17_cast_fp16, y = var_1336_cast_fp16)[name = string("hidden_states_17_cast_fp16")]; + tensor w_17_to_fp16 = const()[name = string("w_17_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25513536)))]; + tensor x_27_cast_fp16 = mul(x = w_17_to_fp16, y = hidden_states_17_cast_fp16)[name = string("x_27_cast_fp16")]; + string q_33_pad_type_0 = const()[name = string("q_33_pad_type_0"), val = string("valid")]; + tensor q_33_strides_0 = const()[name = string("q_33_strides_0"), val = tensor([1, 1])]; + tensor q_33_pad_0 = const()[name = string("q_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_33_dilations_0 = const()[name = string("q_33_dilations_0"), val = tensor([1, 1])]; + int32 q_33_groups_0 = const()[name = string("q_33_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_4_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25514624))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26038976))))[name = string("decoder_pre_transformer_layers_4_self_attn_q_proj_weight_to_fp16_palettized")]; + tensor q_33_cast_fp16 = conv(dilations = q_33_dilations_0, groups = q_33_groups_0, pad = q_33_pad_0, pad_type = q_33_pad_type_0, strides = q_33_strides_0, weight = decoder_pre_transformer_layers_4_self_attn_q_proj_weight_to_fp16_palettized, x = x_27_cast_fp16)[name = string("q_33_cast_fp16")]; + string k_17_pad_type_0 = const()[name = string("k_17_pad_type_0"), val = string("valid")]; + tensor k_17_strides_0 = const()[name = string("k_17_strides_0"), val = tensor([1, 1])]; + tensor k_17_pad_0 = const()[name = string("k_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_17_dilations_0 = const()[name = string("k_17_dilations_0"), val = tensor([1, 1])]; + int32 k_17_groups_0 = const()[name = string("k_17_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_4_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26039552))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26563904))))[name = string("decoder_pre_transformer_layers_4_self_attn_k_proj_weight_to_fp16_palettized")]; + tensor k_17_cast_fp16 = conv(dilations = k_17_dilations_0, groups = k_17_groups_0, pad = k_17_pad_0, pad_type = k_17_pad_type_0, strides = k_17_strides_0, weight = decoder_pre_transformer_layers_4_self_attn_k_proj_weight_to_fp16_palettized, x = x_27_cast_fp16)[name = string("k_17_cast_fp16")]; + string v_9_pad_type_0 = const()[name = string("v_9_pad_type_0"), val = string("valid")]; + tensor v_9_strides_0 = const()[name = string("v_9_strides_0"), val = tensor([1, 1])]; + tensor v_9_pad_0 = const()[name = string("v_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_9_dilations_0 = const()[name = string("v_9_dilations_0"), val = tensor([1, 1])]; + int32 v_9_groups_0 = const()[name = string("v_9_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_4_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26564480))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27088832))))[name = string("decoder_pre_transformer_layers_4_self_attn_v_proj_weight_to_fp16_palettized")]; + tensor v_9_cast_fp16 = conv(dilations = v_9_dilations_0, groups = v_9_groups_0, pad = v_9_pad_0, pad_type = v_9_pad_type_0, strides = v_9_strides_0, weight = decoder_pre_transformer_layers_4_self_attn_v_proj_weight_to_fp16_palettized, x = x_27_cast_fp16)[name = string("v_9_cast_fp16")]; + tensor var_1368 = const()[name = string("op_1368"), val = tensor([16, 64, 1, 1])]; + tensor q_35_cast_fp16 = reshape(shape = var_1368, x = q_33_cast_fp16)[name = string("q_35_cast_fp16")]; + tensor var_1373 = const()[name = string("op_1373"), val = tensor([16, 64, 1, 1])]; + tensor k_19_cast_fp16 = reshape(shape = var_1373, x = k_17_cast_fp16)[name = string("k_19_cast_fp16")]; + tensor var_1385_cast_fp16 = mul(x = q_35_cast_fp16, y = cos_1_cast_fp16)[name = string("op_1385_cast_fp16")]; + tensor var_1390_begin_0 = const()[name = string("op_1390_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1390_end_0 = const()[name = string("op_1390_end_0"), val = tensor([16, 32, 1, 1])]; + tensor var_1390_end_mask_0 = const()[name = string("op_1390_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1390_cast_fp16 = slice_by_index(begin = var_1390_begin_0, end = var_1390_end_0, end_mask = var_1390_end_mask_0, x = q_35_cast_fp16)[name = string("op_1390_cast_fp16")]; + tensor var_1397_begin_0 = const()[name = string("op_1397_begin_0"), val = tensor([0, 32, 0, 0])]; + tensor var_1397_end_0 = const()[name = string("op_1397_end_0"), val = tensor([16, 64, 1, 1])]; + tensor var_1397_end_mask_0 = const()[name = string("op_1397_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1397_cast_fp16 = slice_by_index(begin = var_1397_begin_0, end = var_1397_end_0, end_mask = var_1397_end_mask_0, x = q_35_cast_fp16)[name = string("op_1397_cast_fp16")]; + fp16 const_56_promoted_to_fp16 = const()[name = string("const_56_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_1400_cast_fp16 = mul(x = var_1397_cast_fp16, y = const_56_promoted_to_fp16)[name = string("op_1400_cast_fp16")]; + bool var_1402_interleave_0 = const()[name = string("op_1402_interleave_0"), val = bool(false)]; + tensor var_1402_cast_fp16 = concat(axis = var_32, interleave = var_1402_interleave_0, values = (var_1400_cast_fp16, var_1390_cast_fp16))[name = string("op_1402_cast_fp16")]; + tensor var_1403_cast_fp16 = mul(x = var_1402_cast_fp16, y = sin_1_cast_fp16)[name = string("op_1403_cast_fp16")]; + tensor q_rotated_9_cast_fp16 = add(x = var_1385_cast_fp16, y = var_1403_cast_fp16)[name = string("q_rotated_9_cast_fp16")]; + tensor var_1405_cast_fp16 = mul(x = k_19_cast_fp16, y = cos_1_cast_fp16)[name = string("op_1405_cast_fp16")]; + tensor var_1410_begin_0 = const()[name = string("op_1410_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1410_end_0 = const()[name = string("op_1410_end_0"), val = tensor([16, 32, 1, 1])]; + tensor var_1410_end_mask_0 = const()[name = string("op_1410_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1410_cast_fp16 = slice_by_index(begin = var_1410_begin_0, end = var_1410_end_0, end_mask = var_1410_end_mask_0, x = k_19_cast_fp16)[name = string("op_1410_cast_fp16")]; + tensor var_1417_begin_0 = const()[name = string("op_1417_begin_0"), val = tensor([0, 32, 0, 0])]; + tensor var_1417_end_0 = const()[name = string("op_1417_end_0"), val = tensor([16, 64, 1, 1])]; + tensor var_1417_end_mask_0 = const()[name = string("op_1417_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1417_cast_fp16 = slice_by_index(begin = var_1417_begin_0, end = var_1417_end_0, end_mask = var_1417_end_mask_0, x = k_19_cast_fp16)[name = string("op_1417_cast_fp16")]; + fp16 const_59_promoted_to_fp16 = const()[name = string("const_59_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_1420_cast_fp16 = mul(x = var_1417_cast_fp16, y = const_59_promoted_to_fp16)[name = string("op_1420_cast_fp16")]; + bool var_1422_interleave_0 = const()[name = string("op_1422_interleave_0"), val = bool(false)]; + tensor var_1422_cast_fp16 = concat(axis = var_32, interleave = var_1422_interleave_0, values = (var_1420_cast_fp16, var_1410_cast_fp16))[name = string("op_1422_cast_fp16")]; + tensor var_1423_cast_fp16 = mul(x = var_1422_cast_fp16, y = sin_1_cast_fp16)[name = string("op_1423_cast_fp16")]; + tensor k_rotated_9_cast_fp16 = add(x = var_1405_cast_fp16, y = var_1423_cast_fp16)[name = string("k_rotated_9_cast_fp16")]; + tensor var_1427 = const()[name = string("op_1427"), val = tensor([1, 1024, 1, 1])]; + tensor current_key_9_cast_fp16 = reshape(shape = var_1427, x = k_rotated_9_cast_fp16)[name = string("current_key_9_cast_fp16")]; + tensor var_1432_cast_fp16 = mul(x = var_522_cast_fp16_4, y = var_647_cast_fp16)[name = string("op_1432_cast_fp16")]; + tensor var_1433_cast_fp16 = mul(x = current_key_9_cast_fp16, y = update_mask_1_cast_fp16)[name = string("op_1433_cast_fp16")]; + tensor key_cache_updated_9_cast_fp16 = add(x = var_1432_cast_fp16, y = var_1433_cast_fp16)[name = string("key_cache_updated_9_cast_fp16")]; + tensor var_1436_cast_fp16 = mul(x = var_531_cast_fp16_4, y = var_647_cast_fp16)[name = string("op_1436_cast_fp16")]; + tensor var_1437_cast_fp16 = mul(x = v_9_cast_fp16, y = update_mask_1_cast_fp16)[name = string("op_1437_cast_fp16")]; + tensor value_cache_updated_9_cast_fp16 = add(x = var_1436_cast_fp16, y = var_1437_cast_fp16)[name = string("value_cache_updated_9_cast_fp16")]; + tensor var_1439 = const()[name = string("op_1439"), val = tensor([1, 16, 64, 1])]; + tensor q_39_cast_fp16 = reshape(shape = var_1439, x = q_rotated_9_cast_fp16)[name = string("q_39_cast_fp16")]; + tensor var_1442 = const()[name = string("op_1442"), val = tensor([1, 16, 64, 256])]; + tensor k_for_attn_9_cast_fp16 = reshape(shape = var_1442, x = key_cache_updated_9_cast_fp16)[name = string("k_for_attn_9_cast_fp16")]; + tensor var_1444 = const()[name = string("op_1444"), val = tensor([1, 16, 64, 256])]; + tensor v_for_attn_9_cast_fp16 = reshape(shape = var_1444, x = value_cache_updated_9_cast_fp16)[name = string("v_for_attn_9_cast_fp16")]; + bool var_1448_transpose_x_1 = const()[name = string("op_1448_transpose_x_1"), val = bool(true)]; + bool var_1448_transpose_y_1 = const()[name = string("op_1448_transpose_y_1"), val = bool(false)]; + tensor var_1448_cast_fp16 = matmul(transpose_x = var_1448_transpose_x_1, transpose_y = var_1448_transpose_y_1, x = q_39_cast_fp16, y = k_for_attn_9_cast_fp16)[name = string("op_1448_cast_fp16")]; + fp16 var_1449_to_fp16 = const()[name = string("op_1449_to_fp16"), val = fp16(0x1p-3)]; + tensor attn_weights_17_cast_fp16 = mul(x = var_1448_cast_fp16, y = var_1449_to_fp16)[name = string("attn_weights_17_cast_fp16")]; + tensor input_81_cast_fp16 = add(x = attn_weights_17_cast_fp16, y = attn_mask_1_cast_fp16)[name = string("input_81_cast_fp16")]; + tensor attn_weights_19_cast_fp16 = softmax(axis = var_28, x = input_81_cast_fp16)[name = string("attn_weights_19_cast_fp16")]; + bool attn_output_17_transpose_x_1 = const()[name = string("attn_output_17_transpose_x_1"), val = bool(false)]; + bool attn_output_17_transpose_y_1 = const()[name = string("attn_output_17_transpose_y_1"), val = bool(true)]; + tensor attn_output_17_cast_fp16 = matmul(transpose_x = attn_output_17_transpose_x_1, transpose_y = attn_output_17_transpose_y_1, x = attn_weights_19_cast_fp16, y = v_for_attn_9_cast_fp16)[name = string("attn_output_17_cast_fp16")]; + tensor var_1458 = const()[name = string("op_1458"), val = tensor([0, 1, 3, 2])]; + tensor var_1460 = const()[name = string("op_1460"), val = tensor([1, -1, 1, 1])]; + tensor attn_output_19_cast_fp16 = transpose(perm = var_1458, x = attn_output_17_cast_fp16)[name = string("transpose_7")]; + tensor input_83_cast_fp16 = reshape(shape = var_1460, x = attn_output_19_cast_fp16)[name = string("input_83_cast_fp16")]; + string x_29_pad_type_0 = const()[name = string("x_29_pad_type_0"), val = string("valid")]; + tensor x_29_strides_0 = const()[name = string("x_29_strides_0"), val = tensor([1, 1])]; + tensor x_29_pad_0 = const()[name = string("x_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_29_dilations_0 = const()[name = string("x_29_dilations_0"), val = tensor([1, 1])]; + int32 x_29_groups_0 = const()[name = string("x_29_groups_0"), val = int32(1)]; + tensor op_1475_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27089408))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27613760))))[name = string("op_1475_weight_0_to_fp16_palettized")]; + tensor var_1475_bias_0_to_fp16 = const()[name = string("op_1475_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27614336)))]; + tensor var_1475_cast_fp16 = conv(bias = var_1475_bias_0_to_fp16, dilations = x_29_dilations_0, groups = x_29_groups_0, pad = x_29_pad_0, pad_type = x_29_pad_type_0, strides = x_29_strides_0, weight = op_1475_weight_0_to_fp16_palettized, x = input_83_cast_fp16)[name = string("op_1475_cast_fp16")]; + tensor inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = var_1475_cast_fp16)[name = string("inputs_19_cast_fp16")]; + tensor inputs_sq_19_cast_fp16 = mul(x = inputs_19_cast_fp16, y = inputs_19_cast_fp16)[name = string("inputs_sq_19_cast_fp16")]; + tensor variance_19_axes_0 = const()[name = string("variance_19_axes_0"), val = tensor([1])]; + bool variance_19_keep_dims_0 = const()[name = string("variance_19_keep_dims_0"), val = bool(true)]; + tensor variance_19_cast_fp16 = reduce_mean(axes = variance_19_axes_0, keep_dims = variance_19_keep_dims_0, x = inputs_sq_19_cast_fp16)[name = string("variance_19_cast_fp16")]; + fp16 var_1481_to_fp16 = const()[name = string("op_1481_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1482_cast_fp16 = add(x = variance_19_cast_fp16, y = var_1481_to_fp16)[name = string("op_1482_cast_fp16")]; + fp32 var_1483_epsilon_0 = const()[name = string("op_1483_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_1483_cast_fp16 = rsqrt(epsilon = var_1483_epsilon_0, x = var_1482_cast_fp16)[name = string("op_1483_cast_fp16")]; + tensor hidden_states_19_cast_fp16 = mul(x = inputs_19_cast_fp16, y = var_1483_cast_fp16)[name = string("hidden_states_19_cast_fp16")]; + tensor w_19_to_fp16 = const()[name = string("w_19_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27615424)))]; + tensor input_85_cast_fp16 = mul(x = w_19_to_fp16, y = hidden_states_19_cast_fp16)[name = string("input_85_cast_fp16")]; + string input_87_pad_type_0 = const()[name = string("input_87_pad_type_0"), val = string("valid")]; + tensor input_87_strides_0 = const()[name = string("input_87_strides_0"), val = tensor([1, 1])]; + tensor input_87_pad_0 = const()[name = string("input_87_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_87_dilations_0 = const()[name = string("input_87_dilations_0"), val = tensor([1, 1])]; + int32 input_87_groups_0 = const()[name = string("input_87_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_4_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27616512))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28140864))))[name = string("decoder_pre_transformer_layers_4_mlp_gate_proj_weight_to_fp16_palettized")]; + tensor input_87_cast_fp16 = conv(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 = decoder_pre_transformer_layers_4_mlp_gate_proj_weight_to_fp16_palettized, x = input_85_cast_fp16)[name = string("input_87_cast_fp16")]; + tensor var_1497_cast_fp16 = silu(x = input_87_cast_fp16)[name = string("op_1497_cast_fp16")]; + string var_1503_pad_type_0 = const()[name = string("op_1503_pad_type_0"), val = string("valid")]; + tensor var_1503_strides_0 = const()[name = string("op_1503_strides_0"), val = tensor([1, 1])]; + tensor var_1503_pad_0 = const()[name = string("op_1503_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1503_dilations_0 = const()[name = string("op_1503_dilations_0"), val = tensor([1, 1])]; + int32 var_1503_groups_0 = const()[name = string("op_1503_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_4_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28141440))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28665792))))[name = string("decoder_pre_transformer_layers_4_mlp_up_proj_weight_to_fp16_palettized")]; + tensor var_1503_cast_fp16 = conv(dilations = var_1503_dilations_0, groups = var_1503_groups_0, pad = var_1503_pad_0, pad_type = var_1503_pad_type_0, strides = var_1503_strides_0, weight = decoder_pre_transformer_layers_4_mlp_up_proj_weight_to_fp16_palettized, x = input_85_cast_fp16)[name = string("op_1503_cast_fp16")]; + tensor input_89_cast_fp16 = mul(x = var_1497_cast_fp16, y = var_1503_cast_fp16)[name = string("input_89_cast_fp16")]; + string x_31_pad_type_0 = const()[name = string("x_31_pad_type_0"), val = string("valid")]; + tensor x_31_strides_0 = const()[name = string("x_31_strides_0"), val = tensor([1, 1])]; + tensor x_31_pad_0 = const()[name = string("x_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_31_dilations_0 = const()[name = string("x_31_dilations_0"), val = tensor([1, 1])]; + int32 x_31_groups_0 = const()[name = string("x_31_groups_0"), val = int32(1)]; + tensor op_1514_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28666368))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29190720))))[name = string("op_1514_weight_0_to_fp16_palettized")]; + tensor var_1514_bias_0_to_fp16 = const()[name = string("op_1514_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29191296)))]; + tensor var_1514_cast_fp16 = conv(bias = var_1514_bias_0_to_fp16, dilations = x_31_dilations_0, groups = x_31_groups_0, pad = x_31_pad_0, pad_type = x_31_pad_type_0, strides = x_31_strides_0, weight = op_1514_weight_0_to_fp16_palettized, x = input_89_cast_fp16)[name = string("op_1514_cast_fp16")]; + tensor inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = var_1514_cast_fp16)[name = string("inputs_21_cast_fp16")]; + tensor inputs_sq_21_cast_fp16 = mul(x = inputs_21_cast_fp16, y = inputs_21_cast_fp16)[name = string("inputs_sq_21_cast_fp16")]; + tensor variance_21_axes_0 = const()[name = string("variance_21_axes_0"), val = tensor([1])]; + bool variance_21_keep_dims_0 = const()[name = string("variance_21_keep_dims_0"), val = bool(true)]; + tensor variance_21_cast_fp16 = reduce_mean(axes = variance_21_axes_0, keep_dims = variance_21_keep_dims_0, x = inputs_sq_21_cast_fp16)[name = string("variance_21_cast_fp16")]; + fp16 var_1530_to_fp16 = const()[name = string("op_1530_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1531_cast_fp16 = add(x = variance_21_cast_fp16, y = var_1530_to_fp16)[name = string("op_1531_cast_fp16")]; + fp32 var_1532_epsilon_0 = const()[name = string("op_1532_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_1532_cast_fp16 = rsqrt(epsilon = var_1532_epsilon_0, x = var_1531_cast_fp16)[name = string("op_1532_cast_fp16")]; + tensor hidden_states_21_cast_fp16 = mul(x = inputs_21_cast_fp16, y = var_1532_cast_fp16)[name = string("hidden_states_21_cast_fp16")]; + tensor w_21_to_fp16 = const()[name = string("w_21_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29192384)))]; + tensor x_33_cast_fp16 = mul(x = w_21_to_fp16, y = hidden_states_21_cast_fp16)[name = string("x_33_cast_fp16")]; + string q_41_pad_type_0 = const()[name = string("q_41_pad_type_0"), val = string("valid")]; + tensor q_41_strides_0 = const()[name = string("q_41_strides_0"), val = tensor([1, 1])]; + tensor q_41_pad_0 = const()[name = string("q_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_41_dilations_0 = const()[name = string("q_41_dilations_0"), val = tensor([1, 1])]; + int32 q_41_groups_0 = const()[name = string("q_41_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_5_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29193472))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29717824))))[name = string("decoder_pre_transformer_layers_5_self_attn_q_proj_weight_to_fp16_palettized")]; + tensor q_41_cast_fp16 = conv(dilations = q_41_dilations_0, groups = q_41_groups_0, pad = q_41_pad_0, pad_type = q_41_pad_type_0, strides = q_41_strides_0, weight = decoder_pre_transformer_layers_5_self_attn_q_proj_weight_to_fp16_palettized, x = x_33_cast_fp16)[name = string("q_41_cast_fp16")]; + string k_21_pad_type_0 = const()[name = string("k_21_pad_type_0"), val = string("valid")]; + tensor k_21_strides_0 = const()[name = string("k_21_strides_0"), val = tensor([1, 1])]; + tensor k_21_pad_0 = const()[name = string("k_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_21_dilations_0 = const()[name = string("k_21_dilations_0"), val = tensor([1, 1])]; + int32 k_21_groups_0 = const()[name = string("k_21_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_5_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29718400))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30242752))))[name = string("decoder_pre_transformer_layers_5_self_attn_k_proj_weight_to_fp16_palettized")]; + tensor k_21_cast_fp16 = conv(dilations = k_21_dilations_0, groups = k_21_groups_0, pad = k_21_pad_0, pad_type = k_21_pad_type_0, strides = k_21_strides_0, weight = decoder_pre_transformer_layers_5_self_attn_k_proj_weight_to_fp16_palettized, x = x_33_cast_fp16)[name = string("k_21_cast_fp16")]; + string v_11_pad_type_0 = const()[name = string("v_11_pad_type_0"), val = string("valid")]; + tensor v_11_strides_0 = const()[name = string("v_11_strides_0"), val = tensor([1, 1])]; + tensor v_11_pad_0 = const()[name = string("v_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_11_dilations_0 = const()[name = string("v_11_dilations_0"), val = tensor([1, 1])]; + int32 v_11_groups_0 = const()[name = string("v_11_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_5_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30243328))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30767680))))[name = string("decoder_pre_transformer_layers_5_self_attn_v_proj_weight_to_fp16_palettized")]; + tensor v_11_cast_fp16 = conv(dilations = v_11_dilations_0, groups = v_11_groups_0, pad = v_11_pad_0, pad_type = v_11_pad_type_0, strides = v_11_strides_0, weight = decoder_pre_transformer_layers_5_self_attn_v_proj_weight_to_fp16_palettized, x = x_33_cast_fp16)[name = string("v_11_cast_fp16")]; + tensor var_1564 = const()[name = string("op_1564"), val = tensor([16, 64, 1, 1])]; + tensor q_43_cast_fp16 = reshape(shape = var_1564, x = q_41_cast_fp16)[name = string("q_43_cast_fp16")]; + tensor var_1569 = const()[name = string("op_1569"), val = tensor([16, 64, 1, 1])]; + tensor k_23_cast_fp16 = reshape(shape = var_1569, x = k_21_cast_fp16)[name = string("k_23_cast_fp16")]; + tensor var_1581_cast_fp16 = mul(x = q_43_cast_fp16, y = cos_1_cast_fp16)[name = string("op_1581_cast_fp16")]; + tensor var_1586_begin_0 = const()[name = string("op_1586_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1586_end_0 = const()[name = string("op_1586_end_0"), val = tensor([16, 32, 1, 1])]; + tensor var_1586_end_mask_0 = const()[name = string("op_1586_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1586_cast_fp16 = slice_by_index(begin = var_1586_begin_0, end = var_1586_end_0, end_mask = var_1586_end_mask_0, x = q_43_cast_fp16)[name = string("op_1586_cast_fp16")]; + tensor var_1593_begin_0 = const()[name = string("op_1593_begin_0"), val = tensor([0, 32, 0, 0])]; + tensor var_1593_end_0 = const()[name = string("op_1593_end_0"), val = tensor([16, 64, 1, 1])]; + tensor var_1593_end_mask_0 = const()[name = string("op_1593_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1593_cast_fp16 = slice_by_index(begin = var_1593_begin_0, end = var_1593_end_0, end_mask = var_1593_end_mask_0, x = q_43_cast_fp16)[name = string("op_1593_cast_fp16")]; + fp16 const_64_promoted_to_fp16 = const()[name = string("const_64_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_1596_cast_fp16 = mul(x = var_1593_cast_fp16, y = const_64_promoted_to_fp16)[name = string("op_1596_cast_fp16")]; + bool var_1598_interleave_0 = const()[name = string("op_1598_interleave_0"), val = bool(false)]; + tensor var_1598_cast_fp16 = concat(axis = var_32, interleave = var_1598_interleave_0, values = (var_1596_cast_fp16, var_1586_cast_fp16))[name = string("op_1598_cast_fp16")]; + tensor var_1599_cast_fp16 = mul(x = var_1598_cast_fp16, y = sin_1_cast_fp16)[name = string("op_1599_cast_fp16")]; + tensor q_rotated_11_cast_fp16 = add(x = var_1581_cast_fp16, y = var_1599_cast_fp16)[name = string("q_rotated_11_cast_fp16")]; + tensor var_1601_cast_fp16 = mul(x = k_23_cast_fp16, y = cos_1_cast_fp16)[name = string("op_1601_cast_fp16")]; + tensor var_1606_begin_0 = const()[name = string("op_1606_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1606_end_0 = const()[name = string("op_1606_end_0"), val = tensor([16, 32, 1, 1])]; + tensor var_1606_end_mask_0 = const()[name = string("op_1606_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1606_cast_fp16 = slice_by_index(begin = var_1606_begin_0, end = var_1606_end_0, end_mask = var_1606_end_mask_0, x = k_23_cast_fp16)[name = string("op_1606_cast_fp16")]; + tensor var_1613_begin_0 = const()[name = string("op_1613_begin_0"), val = tensor([0, 32, 0, 0])]; + tensor var_1613_end_0 = const()[name = string("op_1613_end_0"), val = tensor([16, 64, 1, 1])]; + tensor var_1613_end_mask_0 = const()[name = string("op_1613_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1613_cast_fp16 = slice_by_index(begin = var_1613_begin_0, end = var_1613_end_0, end_mask = var_1613_end_mask_0, x = k_23_cast_fp16)[name = string("op_1613_cast_fp16")]; + fp16 const_67_promoted_to_fp16 = const()[name = string("const_67_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_1616_cast_fp16 = mul(x = var_1613_cast_fp16, y = const_67_promoted_to_fp16)[name = string("op_1616_cast_fp16")]; + bool var_1618_interleave_0 = const()[name = string("op_1618_interleave_0"), val = bool(false)]; + tensor var_1618_cast_fp16 = concat(axis = var_32, interleave = var_1618_interleave_0, values = (var_1616_cast_fp16, var_1606_cast_fp16))[name = string("op_1618_cast_fp16")]; + tensor var_1619_cast_fp16 = mul(x = var_1618_cast_fp16, y = sin_1_cast_fp16)[name = string("op_1619_cast_fp16")]; + tensor k_rotated_11_cast_fp16 = add(x = var_1601_cast_fp16, y = var_1619_cast_fp16)[name = string("k_rotated_11_cast_fp16")]; + tensor var_1623 = const()[name = string("op_1623"), val = tensor([1, 1024, 1, 1])]; + tensor current_key_11_cast_fp16 = reshape(shape = var_1623, x = k_rotated_11_cast_fp16)[name = string("current_key_11_cast_fp16")]; + tensor var_1628_cast_fp16 = mul(x = var_522_cast_fp16_5, y = var_647_cast_fp16)[name = string("op_1628_cast_fp16")]; + tensor var_1629_cast_fp16 = mul(x = current_key_11_cast_fp16, y = update_mask_1_cast_fp16)[name = string("op_1629_cast_fp16")]; + tensor key_cache_updated_11_cast_fp16 = add(x = var_1628_cast_fp16, y = var_1629_cast_fp16)[name = string("key_cache_updated_11_cast_fp16")]; + tensor var_1632_cast_fp16 = mul(x = var_531_cast_fp16_5, y = var_647_cast_fp16)[name = string("op_1632_cast_fp16")]; + tensor var_1633_cast_fp16 = mul(x = v_11_cast_fp16, y = update_mask_1_cast_fp16)[name = string("op_1633_cast_fp16")]; + tensor value_cache_updated_11_cast_fp16 = add(x = var_1632_cast_fp16, y = var_1633_cast_fp16)[name = string("value_cache_updated_11_cast_fp16")]; + tensor var_1635 = const()[name = string("op_1635"), val = tensor([1, 16, 64, 1])]; + tensor q_47_cast_fp16 = reshape(shape = var_1635, x = q_rotated_11_cast_fp16)[name = string("q_47_cast_fp16")]; + tensor var_1638 = const()[name = string("op_1638"), val = tensor([1, 16, 64, 256])]; + tensor k_for_attn_11_cast_fp16 = reshape(shape = var_1638, x = key_cache_updated_11_cast_fp16)[name = string("k_for_attn_11_cast_fp16")]; + tensor var_1640 = const()[name = string("op_1640"), val = tensor([1, 16, 64, 256])]; + tensor v_for_attn_11_cast_fp16 = reshape(shape = var_1640, x = value_cache_updated_11_cast_fp16)[name = string("v_for_attn_11_cast_fp16")]; + bool var_1644_transpose_x_1 = const()[name = string("op_1644_transpose_x_1"), val = bool(true)]; + bool var_1644_transpose_y_1 = const()[name = string("op_1644_transpose_y_1"), val = bool(false)]; + tensor var_1644_cast_fp16 = matmul(transpose_x = var_1644_transpose_x_1, transpose_y = var_1644_transpose_y_1, x = q_47_cast_fp16, y = k_for_attn_11_cast_fp16)[name = string("op_1644_cast_fp16")]; + fp16 var_1645_to_fp16 = const()[name = string("op_1645_to_fp16"), val = fp16(0x1p-3)]; + tensor attn_weights_21_cast_fp16 = mul(x = var_1644_cast_fp16, y = var_1645_to_fp16)[name = string("attn_weights_21_cast_fp16")]; + tensor input_91_cast_fp16 = add(x = attn_weights_21_cast_fp16, y = attn_mask_1_cast_fp16)[name = string("input_91_cast_fp16")]; + tensor attn_weights_23_cast_fp16 = softmax(axis = var_28, x = input_91_cast_fp16)[name = string("attn_weights_23_cast_fp16")]; + bool attn_output_21_transpose_x_1 = const()[name = string("attn_output_21_transpose_x_1"), val = bool(false)]; + bool attn_output_21_transpose_y_1 = const()[name = string("attn_output_21_transpose_y_1"), val = bool(true)]; + tensor attn_output_21_cast_fp16 = matmul(transpose_x = attn_output_21_transpose_x_1, transpose_y = attn_output_21_transpose_y_1, x = attn_weights_23_cast_fp16, y = v_for_attn_11_cast_fp16)[name = string("attn_output_21_cast_fp16")]; + tensor var_1654 = const()[name = string("op_1654"), val = tensor([0, 1, 3, 2])]; + tensor var_1656 = const()[name = string("op_1656"), val = tensor([1, -1, 1, 1])]; + tensor attn_output_23_cast_fp16 = transpose(perm = var_1654, x = attn_output_21_cast_fp16)[name = string("transpose_6")]; + tensor input_93_cast_fp16 = reshape(shape = var_1656, x = attn_output_23_cast_fp16)[name = string("input_93_cast_fp16")]; + string x_35_pad_type_0 = const()[name = string("x_35_pad_type_0"), val = string("valid")]; + tensor x_35_strides_0 = const()[name = string("x_35_strides_0"), val = tensor([1, 1])]; + tensor x_35_pad_0 = const()[name = string("x_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_35_dilations_0 = const()[name = string("x_35_dilations_0"), val = tensor([1, 1])]; + int32 x_35_groups_0 = const()[name = string("x_35_groups_0"), val = int32(1)]; + tensor op_1671_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30768256))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31292608))))[name = string("op_1671_weight_0_to_fp16_palettized")]; + tensor var_1671_bias_0_to_fp16 = const()[name = string("op_1671_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31293184)))]; + tensor var_1671_cast_fp16 = conv(bias = var_1671_bias_0_to_fp16, dilations = x_35_dilations_0, groups = x_35_groups_0, pad = x_35_pad_0, pad_type = x_35_pad_type_0, strides = x_35_strides_0, weight = op_1671_weight_0_to_fp16_palettized, x = input_93_cast_fp16)[name = string("op_1671_cast_fp16")]; + tensor inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = var_1671_cast_fp16)[name = string("inputs_23_cast_fp16")]; + tensor inputs_sq_23_cast_fp16 = mul(x = inputs_23_cast_fp16, y = inputs_23_cast_fp16)[name = string("inputs_sq_23_cast_fp16")]; + tensor variance_23_axes_0 = const()[name = string("variance_23_axes_0"), val = tensor([1])]; + bool variance_23_keep_dims_0 = const()[name = string("variance_23_keep_dims_0"), val = bool(true)]; + tensor variance_23_cast_fp16 = reduce_mean(axes = variance_23_axes_0, keep_dims = variance_23_keep_dims_0, x = inputs_sq_23_cast_fp16)[name = string("variance_23_cast_fp16")]; + fp16 var_1677_to_fp16 = const()[name = string("op_1677_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1678_cast_fp16 = add(x = variance_23_cast_fp16, y = var_1677_to_fp16)[name = string("op_1678_cast_fp16")]; + fp32 var_1679_epsilon_0 = const()[name = string("op_1679_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_1679_cast_fp16 = rsqrt(epsilon = var_1679_epsilon_0, x = var_1678_cast_fp16)[name = string("op_1679_cast_fp16")]; + tensor hidden_states_23_cast_fp16 = mul(x = inputs_23_cast_fp16, y = var_1679_cast_fp16)[name = string("hidden_states_23_cast_fp16")]; + tensor w_23_to_fp16 = const()[name = string("w_23_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31294272)))]; + tensor input_95_cast_fp16 = mul(x = w_23_to_fp16, y = hidden_states_23_cast_fp16)[name = string("input_95_cast_fp16")]; + string input_97_pad_type_0 = const()[name = string("input_97_pad_type_0"), val = string("valid")]; + tensor input_97_strides_0 = const()[name = string("input_97_strides_0"), val = tensor([1, 1])]; + tensor input_97_pad_0 = const()[name = string("input_97_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_97_dilations_0 = const()[name = string("input_97_dilations_0"), val = tensor([1, 1])]; + int32 input_97_groups_0 = const()[name = string("input_97_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_5_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31295360))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31819712))))[name = string("decoder_pre_transformer_layers_5_mlp_gate_proj_weight_to_fp16_palettized")]; + tensor input_97_cast_fp16 = conv(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 = decoder_pre_transformer_layers_5_mlp_gate_proj_weight_to_fp16_palettized, x = input_95_cast_fp16)[name = string("input_97_cast_fp16")]; + tensor var_1693_cast_fp16 = silu(x = input_97_cast_fp16)[name = string("op_1693_cast_fp16")]; + string var_1699_pad_type_0 = const()[name = string("op_1699_pad_type_0"), val = string("valid")]; + tensor var_1699_strides_0 = const()[name = string("op_1699_strides_0"), val = tensor([1, 1])]; + tensor var_1699_pad_0 = const()[name = string("op_1699_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1699_dilations_0 = const()[name = string("op_1699_dilations_0"), val = tensor([1, 1])]; + int32 var_1699_groups_0 = const()[name = string("op_1699_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_5_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31820288))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32344640))))[name = string("decoder_pre_transformer_layers_5_mlp_up_proj_weight_to_fp16_palettized")]; + tensor var_1699_cast_fp16 = conv(dilations = var_1699_dilations_0, groups = var_1699_groups_0, pad = var_1699_pad_0, pad_type = var_1699_pad_type_0, strides = var_1699_strides_0, weight = decoder_pre_transformer_layers_5_mlp_up_proj_weight_to_fp16_palettized, x = input_95_cast_fp16)[name = string("op_1699_cast_fp16")]; + tensor input_99_cast_fp16 = mul(x = var_1693_cast_fp16, y = var_1699_cast_fp16)[name = string("input_99_cast_fp16")]; + string x_37_pad_type_0 = const()[name = string("x_37_pad_type_0"), val = string("valid")]; + tensor x_37_strides_0 = const()[name = string("x_37_strides_0"), val = tensor([1, 1])]; + tensor x_37_pad_0 = const()[name = string("x_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_37_dilations_0 = const()[name = string("x_37_dilations_0"), val = tensor([1, 1])]; + int32 x_37_groups_0 = const()[name = string("x_37_groups_0"), val = int32(1)]; + tensor op_1710_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32345216))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32869568))))[name = string("op_1710_weight_0_to_fp16_palettized")]; + tensor var_1710_bias_0_to_fp16 = const()[name = string("op_1710_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32870144)))]; + tensor var_1710_cast_fp16 = conv(bias = var_1710_bias_0_to_fp16, dilations = x_37_dilations_0, groups = x_37_groups_0, pad = x_37_pad_0, pad_type = x_37_pad_type_0, strides = x_37_strides_0, weight = op_1710_weight_0_to_fp16_palettized, x = input_99_cast_fp16)[name = string("op_1710_cast_fp16")]; + tensor inputs_25_cast_fp16 = add(x = inputs_23_cast_fp16, y = var_1710_cast_fp16)[name = string("inputs_25_cast_fp16")]; + tensor inputs_sq_25_cast_fp16 = mul(x = inputs_25_cast_fp16, y = inputs_25_cast_fp16)[name = string("inputs_sq_25_cast_fp16")]; + tensor variance_25_axes_0 = const()[name = string("variance_25_axes_0"), val = tensor([1])]; + bool variance_25_keep_dims_0 = const()[name = string("variance_25_keep_dims_0"), val = bool(true)]; + tensor variance_25_cast_fp16 = reduce_mean(axes = variance_25_axes_0, keep_dims = variance_25_keep_dims_0, x = inputs_sq_25_cast_fp16)[name = string("variance_25_cast_fp16")]; + fp16 var_1726_to_fp16 = const()[name = string("op_1726_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1727_cast_fp16 = add(x = variance_25_cast_fp16, y = var_1726_to_fp16)[name = string("op_1727_cast_fp16")]; + fp32 var_1728_epsilon_0 = const()[name = string("op_1728_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_1728_cast_fp16 = rsqrt(epsilon = var_1728_epsilon_0, x = var_1727_cast_fp16)[name = string("op_1728_cast_fp16")]; + tensor hidden_states_25_cast_fp16 = mul(x = inputs_25_cast_fp16, y = var_1728_cast_fp16)[name = string("hidden_states_25_cast_fp16")]; + tensor w_25_to_fp16 = const()[name = string("w_25_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32871232)))]; + tensor x_39_cast_fp16 = mul(x = w_25_to_fp16, y = hidden_states_25_cast_fp16)[name = string("x_39_cast_fp16")]; + string q_49_pad_type_0 = const()[name = string("q_49_pad_type_0"), val = string("valid")]; + tensor q_49_strides_0 = const()[name = string("q_49_strides_0"), val = tensor([1, 1])]; + tensor q_49_pad_0 = const()[name = string("q_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_49_dilations_0 = const()[name = string("q_49_dilations_0"), val = tensor([1, 1])]; + int32 q_49_groups_0 = const()[name = string("q_49_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_6_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32872320))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33396672))))[name = string("decoder_pre_transformer_layers_6_self_attn_q_proj_weight_to_fp16_palettized")]; + tensor q_49_cast_fp16 = conv(dilations = q_49_dilations_0, groups = q_49_groups_0, pad = q_49_pad_0, pad_type = q_49_pad_type_0, strides = q_49_strides_0, weight = decoder_pre_transformer_layers_6_self_attn_q_proj_weight_to_fp16_palettized, x = x_39_cast_fp16)[name = string("q_49_cast_fp16")]; + string k_25_pad_type_0 = const()[name = string("k_25_pad_type_0"), val = string("valid")]; + tensor k_25_strides_0 = const()[name = string("k_25_strides_0"), val = tensor([1, 1])]; + tensor k_25_pad_0 = const()[name = string("k_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_25_dilations_0 = const()[name = string("k_25_dilations_0"), val = tensor([1, 1])]; + int32 k_25_groups_0 = const()[name = string("k_25_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_6_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33397248))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33921600))))[name = string("decoder_pre_transformer_layers_6_self_attn_k_proj_weight_to_fp16_palettized")]; + tensor k_25_cast_fp16 = conv(dilations = k_25_dilations_0, groups = k_25_groups_0, pad = k_25_pad_0, pad_type = k_25_pad_type_0, strides = k_25_strides_0, weight = decoder_pre_transformer_layers_6_self_attn_k_proj_weight_to_fp16_palettized, x = x_39_cast_fp16)[name = string("k_25_cast_fp16")]; + string v_13_pad_type_0 = const()[name = string("v_13_pad_type_0"), val = string("valid")]; + tensor v_13_strides_0 = const()[name = string("v_13_strides_0"), val = tensor([1, 1])]; + tensor v_13_pad_0 = const()[name = string("v_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_13_dilations_0 = const()[name = string("v_13_dilations_0"), val = tensor([1, 1])]; + int32 v_13_groups_0 = const()[name = string("v_13_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_6_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33922176))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34446528))))[name = string("decoder_pre_transformer_layers_6_self_attn_v_proj_weight_to_fp16_palettized")]; + tensor v_13_cast_fp16 = conv(dilations = v_13_dilations_0, groups = v_13_groups_0, pad = v_13_pad_0, pad_type = v_13_pad_type_0, strides = v_13_strides_0, weight = decoder_pre_transformer_layers_6_self_attn_v_proj_weight_to_fp16_palettized, x = x_39_cast_fp16)[name = string("v_13_cast_fp16")]; + tensor var_1760 = const()[name = string("op_1760"), val = tensor([16, 64, 1, 1])]; + tensor q_51_cast_fp16 = reshape(shape = var_1760, x = q_49_cast_fp16)[name = string("q_51_cast_fp16")]; + tensor var_1765 = const()[name = string("op_1765"), val = tensor([16, 64, 1, 1])]; + tensor k_27_cast_fp16 = reshape(shape = var_1765, x = k_25_cast_fp16)[name = string("k_27_cast_fp16")]; + tensor var_1777_cast_fp16 = mul(x = q_51_cast_fp16, y = cos_1_cast_fp16)[name = string("op_1777_cast_fp16")]; + tensor var_1782_begin_0 = const()[name = string("op_1782_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1782_end_0 = const()[name = string("op_1782_end_0"), val = tensor([16, 32, 1, 1])]; + tensor var_1782_end_mask_0 = const()[name = string("op_1782_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1782_cast_fp16 = slice_by_index(begin = var_1782_begin_0, end = var_1782_end_0, end_mask = var_1782_end_mask_0, x = q_51_cast_fp16)[name = string("op_1782_cast_fp16")]; + tensor var_1789_begin_0 = const()[name = string("op_1789_begin_0"), val = tensor([0, 32, 0, 0])]; + tensor var_1789_end_0 = const()[name = string("op_1789_end_0"), val = tensor([16, 64, 1, 1])]; + tensor var_1789_end_mask_0 = const()[name = string("op_1789_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1789_cast_fp16 = slice_by_index(begin = var_1789_begin_0, end = var_1789_end_0, end_mask = var_1789_end_mask_0, x = q_51_cast_fp16)[name = string("op_1789_cast_fp16")]; + fp16 const_72_promoted_to_fp16 = const()[name = string("const_72_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_1792_cast_fp16 = mul(x = var_1789_cast_fp16, y = const_72_promoted_to_fp16)[name = string("op_1792_cast_fp16")]; + bool var_1794_interleave_0 = const()[name = string("op_1794_interleave_0"), val = bool(false)]; + tensor var_1794_cast_fp16 = concat(axis = var_32, interleave = var_1794_interleave_0, values = (var_1792_cast_fp16, var_1782_cast_fp16))[name = string("op_1794_cast_fp16")]; + tensor var_1795_cast_fp16 = mul(x = var_1794_cast_fp16, y = sin_1_cast_fp16)[name = string("op_1795_cast_fp16")]; + tensor q_rotated_13_cast_fp16 = add(x = var_1777_cast_fp16, y = var_1795_cast_fp16)[name = string("q_rotated_13_cast_fp16")]; + tensor var_1797_cast_fp16 = mul(x = k_27_cast_fp16, y = cos_1_cast_fp16)[name = string("op_1797_cast_fp16")]; + tensor var_1802_begin_0 = const()[name = string("op_1802_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1802_end_0 = const()[name = string("op_1802_end_0"), val = tensor([16, 32, 1, 1])]; + tensor var_1802_end_mask_0 = const()[name = string("op_1802_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1802_cast_fp16 = slice_by_index(begin = var_1802_begin_0, end = var_1802_end_0, end_mask = var_1802_end_mask_0, x = k_27_cast_fp16)[name = string("op_1802_cast_fp16")]; + tensor var_1809_begin_0 = const()[name = string("op_1809_begin_0"), val = tensor([0, 32, 0, 0])]; + tensor var_1809_end_0 = const()[name = string("op_1809_end_0"), val = tensor([16, 64, 1, 1])]; + tensor var_1809_end_mask_0 = const()[name = string("op_1809_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1809_cast_fp16 = slice_by_index(begin = var_1809_begin_0, end = var_1809_end_0, end_mask = var_1809_end_mask_0, x = k_27_cast_fp16)[name = string("op_1809_cast_fp16")]; + fp16 const_75_promoted_to_fp16 = const()[name = string("const_75_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_1812_cast_fp16 = mul(x = var_1809_cast_fp16, y = const_75_promoted_to_fp16)[name = string("op_1812_cast_fp16")]; + bool var_1814_interleave_0 = const()[name = string("op_1814_interleave_0"), val = bool(false)]; + tensor var_1814_cast_fp16 = concat(axis = var_32, interleave = var_1814_interleave_0, values = (var_1812_cast_fp16, var_1802_cast_fp16))[name = string("op_1814_cast_fp16")]; + tensor var_1815_cast_fp16 = mul(x = var_1814_cast_fp16, y = sin_1_cast_fp16)[name = string("op_1815_cast_fp16")]; + tensor k_rotated_13_cast_fp16 = add(x = var_1797_cast_fp16, y = var_1815_cast_fp16)[name = string("k_rotated_13_cast_fp16")]; + tensor var_1819 = const()[name = string("op_1819"), val = tensor([1, 1024, 1, 1])]; + tensor current_key_13_cast_fp16 = reshape(shape = var_1819, x = k_rotated_13_cast_fp16)[name = string("current_key_13_cast_fp16")]; + tensor var_1824_cast_fp16 = mul(x = var_522_cast_fp16_6, y = var_647_cast_fp16)[name = string("op_1824_cast_fp16")]; + tensor var_1825_cast_fp16 = mul(x = current_key_13_cast_fp16, y = update_mask_1_cast_fp16)[name = string("op_1825_cast_fp16")]; + tensor key_cache_updated_13_cast_fp16 = add(x = var_1824_cast_fp16, y = var_1825_cast_fp16)[name = string("key_cache_updated_13_cast_fp16")]; + tensor var_1828_cast_fp16 = mul(x = var_531_cast_fp16_6, y = var_647_cast_fp16)[name = string("op_1828_cast_fp16")]; + tensor var_1829_cast_fp16 = mul(x = v_13_cast_fp16, y = update_mask_1_cast_fp16)[name = string("op_1829_cast_fp16")]; + tensor value_cache_updated_13_cast_fp16 = add(x = var_1828_cast_fp16, y = var_1829_cast_fp16)[name = string("value_cache_updated_13_cast_fp16")]; + tensor var_1831 = const()[name = string("op_1831"), val = tensor([1, 16, 64, 1])]; + tensor q_55_cast_fp16 = reshape(shape = var_1831, x = q_rotated_13_cast_fp16)[name = string("q_55_cast_fp16")]; + tensor var_1834 = const()[name = string("op_1834"), val = tensor([1, 16, 64, 256])]; + tensor k_for_attn_13_cast_fp16 = reshape(shape = var_1834, x = key_cache_updated_13_cast_fp16)[name = string("k_for_attn_13_cast_fp16")]; + tensor var_1836 = const()[name = string("op_1836"), val = tensor([1, 16, 64, 256])]; + tensor v_for_attn_13_cast_fp16 = reshape(shape = var_1836, x = value_cache_updated_13_cast_fp16)[name = string("v_for_attn_13_cast_fp16")]; + bool var_1840_transpose_x_1 = const()[name = string("op_1840_transpose_x_1"), val = bool(true)]; + bool var_1840_transpose_y_1 = const()[name = string("op_1840_transpose_y_1"), val = bool(false)]; + tensor var_1840_cast_fp16 = matmul(transpose_x = var_1840_transpose_x_1, transpose_y = var_1840_transpose_y_1, x = q_55_cast_fp16, y = k_for_attn_13_cast_fp16)[name = string("op_1840_cast_fp16")]; + fp16 var_1841_to_fp16 = const()[name = string("op_1841_to_fp16"), val = fp16(0x1p-3)]; + tensor attn_weights_25_cast_fp16 = mul(x = var_1840_cast_fp16, y = var_1841_to_fp16)[name = string("attn_weights_25_cast_fp16")]; + tensor input_101_cast_fp16 = add(x = attn_weights_25_cast_fp16, y = attn_mask_1_cast_fp16)[name = string("input_101_cast_fp16")]; + tensor attn_weights_27_cast_fp16 = softmax(axis = var_28, x = input_101_cast_fp16)[name = string("attn_weights_27_cast_fp16")]; + bool attn_output_25_transpose_x_1 = const()[name = string("attn_output_25_transpose_x_1"), val = bool(false)]; + bool attn_output_25_transpose_y_1 = const()[name = string("attn_output_25_transpose_y_1"), val = bool(true)]; + tensor attn_output_25_cast_fp16 = matmul(transpose_x = attn_output_25_transpose_x_1, transpose_y = attn_output_25_transpose_y_1, x = attn_weights_27_cast_fp16, y = v_for_attn_13_cast_fp16)[name = string("attn_output_25_cast_fp16")]; + tensor var_1850 = const()[name = string("op_1850"), val = tensor([0, 1, 3, 2])]; + tensor var_1852 = const()[name = string("op_1852"), val = tensor([1, -1, 1, 1])]; + tensor attn_output_27_cast_fp16 = transpose(perm = var_1850, x = attn_output_25_cast_fp16)[name = string("transpose_5")]; + tensor input_103_cast_fp16 = reshape(shape = var_1852, x = attn_output_27_cast_fp16)[name = string("input_103_cast_fp16")]; + string x_41_pad_type_0 = const()[name = string("x_41_pad_type_0"), val = string("valid")]; + tensor x_41_strides_0 = const()[name = string("x_41_strides_0"), val = tensor([1, 1])]; + tensor x_41_pad_0 = const()[name = string("x_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_41_dilations_0 = const()[name = string("x_41_dilations_0"), val = tensor([1, 1])]; + int32 x_41_groups_0 = const()[name = string("x_41_groups_0"), val = int32(1)]; + tensor op_1867_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34447104))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34971456))))[name = string("op_1867_weight_0_to_fp16_palettized")]; + tensor var_1867_bias_0_to_fp16 = const()[name = string("op_1867_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34972032)))]; + tensor var_1867_cast_fp16 = conv(bias = var_1867_bias_0_to_fp16, dilations = x_41_dilations_0, groups = x_41_groups_0, pad = x_41_pad_0, pad_type = x_41_pad_type_0, strides = x_41_strides_0, weight = op_1867_weight_0_to_fp16_palettized, x = input_103_cast_fp16)[name = string("op_1867_cast_fp16")]; + tensor inputs_27_cast_fp16 = add(x = inputs_25_cast_fp16, y = var_1867_cast_fp16)[name = string("inputs_27_cast_fp16")]; + tensor inputs_sq_27_cast_fp16 = mul(x = inputs_27_cast_fp16, y = inputs_27_cast_fp16)[name = string("inputs_sq_27_cast_fp16")]; + tensor variance_27_axes_0 = const()[name = string("variance_27_axes_0"), val = tensor([1])]; + bool variance_27_keep_dims_0 = const()[name = string("variance_27_keep_dims_0"), val = bool(true)]; + tensor variance_27_cast_fp16 = reduce_mean(axes = variance_27_axes_0, keep_dims = variance_27_keep_dims_0, x = inputs_sq_27_cast_fp16)[name = string("variance_27_cast_fp16")]; + fp16 var_1873_to_fp16 = const()[name = string("op_1873_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1874_cast_fp16 = add(x = variance_27_cast_fp16, y = var_1873_to_fp16)[name = string("op_1874_cast_fp16")]; + fp32 var_1875_epsilon_0 = const()[name = string("op_1875_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_1875_cast_fp16 = rsqrt(epsilon = var_1875_epsilon_0, x = var_1874_cast_fp16)[name = string("op_1875_cast_fp16")]; + tensor hidden_states_27_cast_fp16 = mul(x = inputs_27_cast_fp16, y = var_1875_cast_fp16)[name = string("hidden_states_27_cast_fp16")]; + tensor w_27_to_fp16 = const()[name = string("w_27_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34973120)))]; + tensor input_105_cast_fp16 = mul(x = w_27_to_fp16, y = hidden_states_27_cast_fp16)[name = string("input_105_cast_fp16")]; + string input_107_pad_type_0 = const()[name = string("input_107_pad_type_0"), val = string("valid")]; + tensor input_107_strides_0 = const()[name = string("input_107_strides_0"), val = tensor([1, 1])]; + tensor input_107_pad_0 = const()[name = string("input_107_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_107_dilations_0 = const()[name = string("input_107_dilations_0"), val = tensor([1, 1])]; + int32 input_107_groups_0 = const()[name = string("input_107_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_6_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34974208))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35498560))))[name = string("decoder_pre_transformer_layers_6_mlp_gate_proj_weight_to_fp16_palettized")]; + tensor input_107_cast_fp16 = conv(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 = decoder_pre_transformer_layers_6_mlp_gate_proj_weight_to_fp16_palettized, x = input_105_cast_fp16)[name = string("input_107_cast_fp16")]; + tensor var_1889_cast_fp16 = silu(x = input_107_cast_fp16)[name = string("op_1889_cast_fp16")]; + string var_1895_pad_type_0 = const()[name = string("op_1895_pad_type_0"), val = string("valid")]; + tensor var_1895_strides_0 = const()[name = string("op_1895_strides_0"), val = tensor([1, 1])]; + tensor var_1895_pad_0 = const()[name = string("op_1895_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1895_dilations_0 = const()[name = string("op_1895_dilations_0"), val = tensor([1, 1])]; + int32 var_1895_groups_0 = const()[name = string("op_1895_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_6_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35499136))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36023488))))[name = string("decoder_pre_transformer_layers_6_mlp_up_proj_weight_to_fp16_palettized")]; + tensor var_1895_cast_fp16 = conv(dilations = var_1895_dilations_0, groups = var_1895_groups_0, pad = var_1895_pad_0, pad_type = var_1895_pad_type_0, strides = var_1895_strides_0, weight = decoder_pre_transformer_layers_6_mlp_up_proj_weight_to_fp16_palettized, x = input_105_cast_fp16)[name = string("op_1895_cast_fp16")]; + tensor input_109_cast_fp16 = mul(x = var_1889_cast_fp16, y = var_1895_cast_fp16)[name = string("input_109_cast_fp16")]; + string x_43_pad_type_0 = const()[name = string("x_43_pad_type_0"), val = string("valid")]; + tensor x_43_strides_0 = const()[name = string("x_43_strides_0"), val = tensor([1, 1])]; + tensor x_43_pad_0 = const()[name = string("x_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_43_dilations_0 = const()[name = string("x_43_dilations_0"), val = tensor([1, 1])]; + int32 x_43_groups_0 = const()[name = string("x_43_groups_0"), val = int32(1)]; + tensor op_1906_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36024064))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36548416))))[name = string("op_1906_weight_0_to_fp16_palettized")]; + tensor var_1906_bias_0_to_fp16 = const()[name = string("op_1906_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36548992)))]; + tensor var_1906_cast_fp16 = conv(bias = var_1906_bias_0_to_fp16, dilations = x_43_dilations_0, groups = x_43_groups_0, pad = x_43_pad_0, pad_type = x_43_pad_type_0, strides = x_43_strides_0, weight = op_1906_weight_0_to_fp16_palettized, x = input_109_cast_fp16)[name = string("op_1906_cast_fp16")]; + tensor inputs_29_cast_fp16 = add(x = inputs_27_cast_fp16, y = var_1906_cast_fp16)[name = string("inputs_29_cast_fp16")]; + tensor inputs_sq_29_cast_fp16 = mul(x = inputs_29_cast_fp16, y = inputs_29_cast_fp16)[name = string("inputs_sq_29_cast_fp16")]; + tensor variance_29_axes_0 = const()[name = string("variance_29_axes_0"), val = tensor([1])]; + bool variance_29_keep_dims_0 = const()[name = string("variance_29_keep_dims_0"), val = bool(true)]; + tensor variance_29_cast_fp16 = reduce_mean(axes = variance_29_axes_0, keep_dims = variance_29_keep_dims_0, x = inputs_sq_29_cast_fp16)[name = string("variance_29_cast_fp16")]; + fp16 var_1922_to_fp16 = const()[name = string("op_1922_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1923_cast_fp16 = add(x = variance_29_cast_fp16, y = var_1922_to_fp16)[name = string("op_1923_cast_fp16")]; + fp32 var_1924_epsilon_0 = const()[name = string("op_1924_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_1924_cast_fp16 = rsqrt(epsilon = var_1924_epsilon_0, x = var_1923_cast_fp16)[name = string("op_1924_cast_fp16")]; + tensor hidden_states_29_cast_fp16 = mul(x = inputs_29_cast_fp16, y = var_1924_cast_fp16)[name = string("hidden_states_29_cast_fp16")]; + tensor w_29_to_fp16 = const()[name = string("w_29_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36550080)))]; + tensor x_45_cast_fp16 = mul(x = w_29_to_fp16, y = hidden_states_29_cast_fp16)[name = string("x_45_cast_fp16")]; + string q_57_pad_type_0 = const()[name = string("q_57_pad_type_0"), val = string("valid")]; + tensor q_57_strides_0 = const()[name = string("q_57_strides_0"), val = tensor([1, 1])]; + tensor q_57_pad_0 = const()[name = string("q_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_57_dilations_0 = const()[name = string("q_57_dilations_0"), val = tensor([1, 1])]; + int32 q_57_groups_0 = const()[name = string("q_57_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_7_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36551168))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37075520))))[name = string("decoder_pre_transformer_layers_7_self_attn_q_proj_weight_to_fp16_palettized")]; + tensor q_57_cast_fp16 = conv(dilations = q_57_dilations_0, groups = q_57_groups_0, pad = q_57_pad_0, pad_type = q_57_pad_type_0, strides = q_57_strides_0, weight = decoder_pre_transformer_layers_7_self_attn_q_proj_weight_to_fp16_palettized, x = x_45_cast_fp16)[name = string("q_57_cast_fp16")]; + string k_29_pad_type_0 = const()[name = string("k_29_pad_type_0"), val = string("valid")]; + tensor k_29_strides_0 = const()[name = string("k_29_strides_0"), val = tensor([1, 1])]; + tensor k_29_pad_0 = const()[name = string("k_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_29_dilations_0 = const()[name = string("k_29_dilations_0"), val = tensor([1, 1])]; + int32 k_29_groups_0 = const()[name = string("k_29_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_7_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37076096))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37600448))))[name = string("decoder_pre_transformer_layers_7_self_attn_k_proj_weight_to_fp16_palettized")]; + tensor k_29_cast_fp16 = conv(dilations = k_29_dilations_0, groups = k_29_groups_0, pad = k_29_pad_0, pad_type = k_29_pad_type_0, strides = k_29_strides_0, weight = decoder_pre_transformer_layers_7_self_attn_k_proj_weight_to_fp16_palettized, x = x_45_cast_fp16)[name = string("k_29_cast_fp16")]; + string v_pad_type_0 = const()[name = string("v_pad_type_0"), val = string("valid")]; + tensor v_strides_0 = const()[name = string("v_strides_0"), val = tensor([1, 1])]; + tensor v_pad_0 = const()[name = string("v_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_dilations_0 = const()[name = string("v_dilations_0"), val = tensor([1, 1])]; + int32 v_groups_0 = const()[name = string("v_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_7_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37601024))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38125376))))[name = string("decoder_pre_transformer_layers_7_self_attn_v_proj_weight_to_fp16_palettized")]; + tensor v_cast_fp16 = conv(dilations = v_dilations_0, groups = v_groups_0, pad = v_pad_0, pad_type = v_pad_type_0, strides = v_strides_0, weight = decoder_pre_transformer_layers_7_self_attn_v_proj_weight_to_fp16_palettized, x = x_45_cast_fp16)[name = string("v_cast_fp16")]; + tensor var_1956 = const()[name = string("op_1956"), val = tensor([16, 64, 1, 1])]; + tensor q_59_cast_fp16 = reshape(shape = var_1956, x = q_57_cast_fp16)[name = string("q_59_cast_fp16")]; + tensor var_1961 = const()[name = string("op_1961"), val = tensor([16, 64, 1, 1])]; + tensor k_cast_fp16 = reshape(shape = var_1961, x = k_29_cast_fp16)[name = string("k_cast_fp16")]; + tensor var_1973_cast_fp16 = mul(x = q_59_cast_fp16, y = cos_1_cast_fp16)[name = string("op_1973_cast_fp16")]; + tensor var_1978_begin_0 = const()[name = string("op_1978_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1978_end_0 = const()[name = string("op_1978_end_0"), val = tensor([16, 32, 1, 1])]; + tensor var_1978_end_mask_0 = const()[name = string("op_1978_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1978_cast_fp16 = slice_by_index(begin = var_1978_begin_0, end = var_1978_end_0, end_mask = var_1978_end_mask_0, x = q_59_cast_fp16)[name = string("op_1978_cast_fp16")]; + tensor var_1985_begin_0 = const()[name = string("op_1985_begin_0"), val = tensor([0, 32, 0, 0])]; + tensor var_1985_end_0 = const()[name = string("op_1985_end_0"), val = tensor([16, 64, 1, 1])]; + tensor var_1985_end_mask_0 = const()[name = string("op_1985_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1985_cast_fp16 = slice_by_index(begin = var_1985_begin_0, end = var_1985_end_0, end_mask = var_1985_end_mask_0, x = q_59_cast_fp16)[name = string("op_1985_cast_fp16")]; + fp16 const_80_promoted_to_fp16 = const()[name = string("const_80_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_1988_cast_fp16 = mul(x = var_1985_cast_fp16, y = const_80_promoted_to_fp16)[name = string("op_1988_cast_fp16")]; + bool var_1990_interleave_0 = const()[name = string("op_1990_interleave_0"), val = bool(false)]; + tensor var_1990_cast_fp16 = concat(axis = var_32, interleave = var_1990_interleave_0, values = (var_1988_cast_fp16, var_1978_cast_fp16))[name = string("op_1990_cast_fp16")]; + tensor var_1991_cast_fp16 = mul(x = var_1990_cast_fp16, y = sin_1_cast_fp16)[name = string("op_1991_cast_fp16")]; + tensor q_rotated_cast_fp16 = add(x = var_1973_cast_fp16, y = var_1991_cast_fp16)[name = string("q_rotated_cast_fp16")]; + tensor var_1993_cast_fp16 = mul(x = k_cast_fp16, y = cos_1_cast_fp16)[name = string("op_1993_cast_fp16")]; + tensor var_1998_begin_0 = const()[name = string("op_1998_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1998_end_0 = const()[name = string("op_1998_end_0"), val = tensor([16, 32, 1, 1])]; + tensor var_1998_end_mask_0 = const()[name = string("op_1998_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1998_cast_fp16 = slice_by_index(begin = var_1998_begin_0, end = var_1998_end_0, end_mask = var_1998_end_mask_0, x = k_cast_fp16)[name = string("op_1998_cast_fp16")]; + tensor var_2005_begin_0 = const()[name = string("op_2005_begin_0"), val = tensor([0, 32, 0, 0])]; + tensor var_2005_end_0 = const()[name = string("op_2005_end_0"), val = tensor([16, 64, 1, 1])]; + tensor var_2005_end_mask_0 = const()[name = string("op_2005_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2005_cast_fp16 = slice_by_index(begin = var_2005_begin_0, end = var_2005_end_0, end_mask = var_2005_end_mask_0, x = k_cast_fp16)[name = string("op_2005_cast_fp16")]; + fp16 const_83_promoted_to_fp16 = const()[name = string("const_83_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_2008_cast_fp16 = mul(x = var_2005_cast_fp16, y = const_83_promoted_to_fp16)[name = string("op_2008_cast_fp16")]; + bool var_2010_interleave_0 = const()[name = string("op_2010_interleave_0"), val = bool(false)]; + tensor var_2010_cast_fp16 = concat(axis = var_32, interleave = var_2010_interleave_0, values = (var_2008_cast_fp16, var_1998_cast_fp16))[name = string("op_2010_cast_fp16")]; + tensor var_2011_cast_fp16 = mul(x = var_2010_cast_fp16, y = sin_1_cast_fp16)[name = string("op_2011_cast_fp16")]; + tensor k_rotated_cast_fp16 = add(x = var_1993_cast_fp16, y = var_2011_cast_fp16)[name = string("k_rotated_cast_fp16")]; + tensor var_2015 = const()[name = string("op_2015"), val = tensor([1, 1024, 1, 1])]; + tensor current_key_cast_fp16 = reshape(shape = var_2015, x = k_rotated_cast_fp16)[name = string("current_key_cast_fp16")]; + tensor var_2020_cast_fp16 = mul(x = var_522_cast_fp16_7, y = var_647_cast_fp16)[name = string("op_2020_cast_fp16")]; + tensor var_2021_cast_fp16 = mul(x = current_key_cast_fp16, y = update_mask_1_cast_fp16)[name = string("op_2021_cast_fp16")]; + tensor key_cache_updated_cast_fp16 = add(x = var_2020_cast_fp16, y = var_2021_cast_fp16)[name = string("key_cache_updated_cast_fp16")]; + tensor var_2024_cast_fp16 = mul(x = var_531_cast_fp16_7, y = var_647_cast_fp16)[name = string("op_2024_cast_fp16")]; + tensor var_2025_cast_fp16 = mul(x = v_cast_fp16, y = update_mask_1_cast_fp16)[name = string("op_2025_cast_fp16")]; + tensor value_cache_updated_cast_fp16 = add(x = var_2024_cast_fp16, y = var_2025_cast_fp16)[name = string("value_cache_updated_cast_fp16")]; + tensor var_2027 = const()[name = string("op_2027"), val = tensor([1, 16, 64, 1])]; + tensor q_cast_fp16 = reshape(shape = var_2027, x = q_rotated_cast_fp16)[name = string("q_cast_fp16")]; + tensor var_2030 = const()[name = string("op_2030"), val = tensor([1, 16, 64, 256])]; + tensor k_for_attn_cast_fp16 = reshape(shape = var_2030, x = key_cache_updated_cast_fp16)[name = string("k_for_attn_cast_fp16")]; + tensor var_2032 = const()[name = string("op_2032"), val = tensor([1, 16, 64, 256])]; + tensor v_for_attn_cast_fp16 = reshape(shape = var_2032, x = value_cache_updated_cast_fp16)[name = string("v_for_attn_cast_fp16")]; + bool var_2036_transpose_x_1 = const()[name = string("op_2036_transpose_x_1"), val = bool(true)]; + bool var_2036_transpose_y_1 = const()[name = string("op_2036_transpose_y_1"), val = bool(false)]; + tensor var_2036_cast_fp16 = matmul(transpose_x = var_2036_transpose_x_1, transpose_y = var_2036_transpose_y_1, x = q_cast_fp16, y = k_for_attn_cast_fp16)[name = string("op_2036_cast_fp16")]; + fp16 var_2037_to_fp16 = const()[name = string("op_2037_to_fp16"), val = fp16(0x1p-3)]; + tensor attn_weights_29_cast_fp16 = mul(x = var_2036_cast_fp16, y = var_2037_to_fp16)[name = string("attn_weights_29_cast_fp16")]; + tensor input_111_cast_fp16 = add(x = attn_weights_29_cast_fp16, y = attn_mask_1_cast_fp16)[name = string("input_111_cast_fp16")]; + tensor attn_weights_cast_fp16 = softmax(axis = var_28, x = input_111_cast_fp16)[name = string("attn_weights_cast_fp16")]; + bool attn_output_29_transpose_x_1 = const()[name = string("attn_output_29_transpose_x_1"), val = bool(false)]; + bool attn_output_29_transpose_y_1 = const()[name = string("attn_output_29_transpose_y_1"), val = bool(true)]; + tensor attn_output_29_cast_fp16 = matmul(transpose_x = attn_output_29_transpose_x_1, transpose_y = attn_output_29_transpose_y_1, x = attn_weights_cast_fp16, y = v_for_attn_cast_fp16)[name = string("attn_output_29_cast_fp16")]; + tensor var_2046 = const()[name = string("op_2046"), val = tensor([0, 1, 3, 2])]; + tensor var_2048 = const()[name = string("op_2048"), val = tensor([1, -1, 1, 1])]; + tensor attn_output_cast_fp16 = transpose(perm = var_2046, x = attn_output_29_cast_fp16)[name = string("transpose_4")]; + tensor input_113_cast_fp16 = reshape(shape = var_2048, x = attn_output_cast_fp16)[name = string("input_113_cast_fp16")]; + string x_47_pad_type_0 = const()[name = string("x_47_pad_type_0"), val = string("valid")]; + tensor x_47_strides_0 = const()[name = string("x_47_strides_0"), val = tensor([1, 1])]; + tensor x_47_pad_0 = const()[name = string("x_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_47_dilations_0 = const()[name = string("x_47_dilations_0"), val = tensor([1, 1])]; + int32 x_47_groups_0 = const()[name = string("x_47_groups_0"), val = int32(1)]; + tensor op_2063_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38125952))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38650304))))[name = string("op_2063_weight_0_to_fp16_palettized")]; + tensor var_2063_bias_0_to_fp16 = const()[name = string("op_2063_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38650880)))]; + tensor var_2063_cast_fp16 = conv(bias = var_2063_bias_0_to_fp16, dilations = x_47_dilations_0, groups = x_47_groups_0, pad = x_47_pad_0, pad_type = x_47_pad_type_0, strides = x_47_strides_0, weight = op_2063_weight_0_to_fp16_palettized, x = input_113_cast_fp16)[name = string("op_2063_cast_fp16")]; + tensor inputs_31_cast_fp16 = add(x = inputs_29_cast_fp16, y = var_2063_cast_fp16)[name = string("inputs_31_cast_fp16")]; + tensor inputs_sq_31_cast_fp16 = mul(x = inputs_31_cast_fp16, y = inputs_31_cast_fp16)[name = string("inputs_sq_31_cast_fp16")]; + tensor variance_31_axes_0 = const()[name = string("variance_31_axes_0"), val = tensor([1])]; + bool variance_31_keep_dims_0 = const()[name = string("variance_31_keep_dims_0"), val = bool(true)]; + tensor variance_31_cast_fp16 = reduce_mean(axes = variance_31_axes_0, keep_dims = variance_31_keep_dims_0, x = inputs_sq_31_cast_fp16)[name = string("variance_31_cast_fp16")]; + fp16 var_2069_to_fp16 = const()[name = string("op_2069_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_2070_cast_fp16 = add(x = variance_31_cast_fp16, y = var_2069_to_fp16)[name = string("op_2070_cast_fp16")]; + fp32 var_2071_epsilon_0 = const()[name = string("op_2071_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_2071_cast_fp16 = rsqrt(epsilon = var_2071_epsilon_0, x = var_2070_cast_fp16)[name = string("op_2071_cast_fp16")]; + tensor hidden_states_31_cast_fp16 = mul(x = inputs_31_cast_fp16, y = var_2071_cast_fp16)[name = string("hidden_states_31_cast_fp16")]; + tensor w_31_to_fp16 = const()[name = string("w_31_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38651968)))]; + tensor input_115_cast_fp16 = mul(x = w_31_to_fp16, y = hidden_states_31_cast_fp16)[name = string("input_115_cast_fp16")]; + string input_117_pad_type_0 = const()[name = string("input_117_pad_type_0"), val = string("valid")]; + tensor input_117_strides_0 = const()[name = string("input_117_strides_0"), val = tensor([1, 1])]; + tensor input_117_pad_0 = const()[name = string("input_117_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_117_dilations_0 = const()[name = string("input_117_dilations_0"), val = tensor([1, 1])]; + int32 input_117_groups_0 = const()[name = string("input_117_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_7_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38653056))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39177408))))[name = string("decoder_pre_transformer_layers_7_mlp_gate_proj_weight_to_fp16_palettized")]; + tensor input_117_cast_fp16 = conv(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 = decoder_pre_transformer_layers_7_mlp_gate_proj_weight_to_fp16_palettized, x = input_115_cast_fp16)[name = string("input_117_cast_fp16")]; + tensor var_2085_cast_fp16 = silu(x = input_117_cast_fp16)[name = string("op_2085_cast_fp16")]; + string var_2091_pad_type_0 = const()[name = string("op_2091_pad_type_0"), val = string("valid")]; + tensor var_2091_strides_0 = const()[name = string("op_2091_strides_0"), val = tensor([1, 1])]; + tensor var_2091_pad_0 = const()[name = string("op_2091_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2091_dilations_0 = const()[name = string("op_2091_dilations_0"), val = tensor([1, 1])]; + int32 var_2091_groups_0 = const()[name = string("op_2091_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_layers_7_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39177984))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39702336))))[name = string("decoder_pre_transformer_layers_7_mlp_up_proj_weight_to_fp16_palettized")]; + tensor var_2091_cast_fp16 = conv(dilations = var_2091_dilations_0, groups = var_2091_groups_0, pad = var_2091_pad_0, pad_type = var_2091_pad_type_0, strides = var_2091_strides_0, weight = decoder_pre_transformer_layers_7_mlp_up_proj_weight_to_fp16_palettized, x = input_115_cast_fp16)[name = string("op_2091_cast_fp16")]; + tensor input_119_cast_fp16 = mul(x = var_2085_cast_fp16, y = var_2091_cast_fp16)[name = string("input_119_cast_fp16")]; + string x_49_pad_type_0 = const()[name = string("x_49_pad_type_0"), val = string("valid")]; + tensor x_49_strides_0 = const()[name = string("x_49_strides_0"), val = tensor([1, 1])]; + tensor x_49_pad_0 = const()[name = string("x_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_49_dilations_0 = const()[name = string("x_49_dilations_0"), val = tensor([1, 1])]; + int32 x_49_groups_0 = const()[name = string("x_49_groups_0"), val = int32(1)]; + tensor op_2102_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39702912))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40227264))))[name = string("op_2102_weight_0_to_fp16_palettized")]; + tensor var_2102_bias_0_to_fp16 = const()[name = string("op_2102_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40227840)))]; + tensor var_2102_cast_fp16 = conv(bias = var_2102_bias_0_to_fp16, dilations = x_49_dilations_0, groups = x_49_groups_0, pad = x_49_pad_0, pad_type = x_49_pad_type_0, strides = x_49_strides_0, weight = op_2102_weight_0_to_fp16_palettized, x = input_119_cast_fp16)[name = string("op_2102_cast_fp16")]; + tensor inputs_cast_fp16 = add(x = inputs_31_cast_fp16, y = var_2102_cast_fp16)[name = string("inputs_cast_fp16")]; + tensor inputs_sq_cast_fp16 = mul(x = inputs_cast_fp16, y = inputs_cast_fp16)[name = string("inputs_sq_cast_fp16")]; + tensor variance_axes_0 = const()[name = string("variance_axes_0"), val = tensor([1])]; + bool variance_keep_dims_0 = const()[name = string("variance_keep_dims_0"), val = bool(true)]; + tensor variance_cast_fp16 = reduce_mean(axes = variance_axes_0, keep_dims = variance_keep_dims_0, x = inputs_sq_cast_fp16)[name = string("variance_cast_fp16")]; + fp16 var_2112_to_fp16 = const()[name = string("op_2112_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_2113_cast_fp16 = add(x = variance_cast_fp16, y = var_2112_to_fp16)[name = string("op_2113_cast_fp16")]; + fp32 var_2114_epsilon_0 = const()[name = string("op_2114_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_2114_cast_fp16 = rsqrt(epsilon = var_2114_epsilon_0, x = var_2113_cast_fp16)[name = string("op_2114_cast_fp16")]; + tensor hidden_states_cast_fp16 = mul(x = inputs_cast_fp16, y = var_2114_cast_fp16)[name = string("hidden_states_cast_fp16")]; + tensor w_to_fp16 = const()[name = string("w_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40228928)))]; + tensor input_121_cast_fp16 = mul(x = w_to_fp16, y = hidden_states_cast_fp16)[name = string("input_121_cast_fp16")]; + string new_hidden_pad_type_0 = const()[name = string("new_hidden_pad_type_0"), val = string("valid")]; + tensor new_hidden_strides_0 = const()[name = string("new_hidden_strides_0"), val = tensor([1, 1])]; + tensor new_hidden_pad_0 = const()[name = string("new_hidden_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor new_hidden_dilations_0 = const()[name = string("new_hidden_dilations_0"), val = tensor([1, 1])]; + int32 new_hidden_groups_0 = const()[name = string("new_hidden_groups_0"), val = int32(1)]; + tensor decoder_pre_transformer_output_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40230016))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40754368))))[name = string("decoder_pre_transformer_output_proj_weight_to_fp16_palettized")]; + tensor decoder_pre_transformer_output_proj_bias_to_fp16 = const()[name = string("decoder_pre_transformer_output_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40754944)))]; + tensor hidden_context_update = conv(bias = decoder_pre_transformer_output_proj_bias_to_fp16, dilations = new_hidden_dilations_0, groups = new_hidden_groups_0, pad = new_hidden_pad_0, pad_type = new_hidden_pad_type_0, strides = new_hidden_strides_0, weight = decoder_pre_transformer_output_proj_weight_to_fp16_palettized, x = input_121_cast_fp16)[name = string("new_hidden_cast_fp16")]; + bool var_2127_interleave_0 = const()[name = string("op_2127_interleave_0"), val = bool(false)]; + tensor key_cache_updates = concat(axis = var_32, interleave = var_2127_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_cast_fp16))[name = string("op_2127_cast_fp16")]; + bool var_2129_interleave_0 = const()[name = string("op_2129_interleave_0"), val = bool(false)]; + tensor value_cache_updates = concat(axis = var_32, interleave = var_2129_interleave_0, values = (v_1_cast_fp16, v_3_cast_fp16, v_5_cast_fp16, v_7_cast_fp16, v_9_cast_fp16, v_11_cast_fp16, v_13_cast_fp16, v_cast_fp16))[name = string("op_2129_cast_fp16")]; + bool x_51_interleave_0 = const()[name = string("x_51_interleave_0"), val = bool(false)]; + tensor x_51_cast_fp16 = concat(axis = var_28, interleave = x_51_interleave_0, values = (hidden_context, hidden_context_update))[name = string("x_51_cast_fp16")]; + string x_53_pad_type_0 = const()[name = string("x_53_pad_type_0"), val = string("valid")]; + tensor x_53_strides_0 = const()[name = string("x_53_strides_0"), val = tensor([1, 2])]; + tensor x_53_pad_0 = const()[name = string("x_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_53_dilations_0 = const()[name = string("x_53_dilations_0"), val = tensor([1, 1])]; + int32 x_53_groups_0 = const()[name = string("x_53_groups_0"), val = int32(1)]; + tensor x_53_has_output_shape_output_shape_0 = const()[name = string("x_53_has_output_shape_output_shape_0"), val = tensor([1, 1024, 1, 10])]; + tensor decoder_upsample_0_0_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40757056))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42854272))))[name = string("decoder_upsample_0_0_conv_weight_to_fp16_palettized")]; + tensor decoder_upsample_0_0_conv_bias_to_fp16 = const()[name = string("decoder_upsample_0_0_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42854848)))]; + tensor x_53_has_output_shape_cast_fp16 = conv_transpose(bias = decoder_upsample_0_0_conv_bias_to_fp16, dilations = x_53_dilations_0, groups = x_53_groups_0, output_shape = x_53_has_output_shape_output_shape_0, pad = x_53_pad_0, pad_type = x_53_pad_type_0, strides = x_53_strides_0, weight = decoder_upsample_0_0_conv_weight_to_fp16_palettized, x = x_51_cast_fp16)[name = string("x_53_has_output_shape_cast_fp16")]; + tensor input_123_pad_0 = const()[name = string("input_123_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 6, 0])]; + string input_123_mode_0 = const()[name = string("input_123_mode_0"), val = string("constant")]; + fp16 const_86_to_fp16 = const()[name = string("const_86_to_fp16"), val = fp16(0x0p+0)]; + tensor input_123_cast_fp16 = pad(constant_val = const_86_to_fp16, mode = input_123_mode_0, pad = input_123_pad_0, x = x_53_has_output_shape_cast_fp16)[name = string("input_123_cast_fp16")]; + string x_57_pad_type_0 = const()[name = string("x_57_pad_type_0"), val = string("valid")]; + int32 x_57_groups_0 = const()[name = string("x_57_groups_0"), val = int32(1024)]; + tensor x_57_strides_0 = const()[name = string("x_57_strides_0"), val = tensor([1, 1])]; + tensor x_57_pad_0 = const()[name = string("x_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_57_dilations_0 = const()[name = string("x_57_dilations_0"), val = tensor([1, 1])]; + tensor decoder_upsample_0_1_dwconv_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42856960))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42864192))))[name = string("decoder_upsample_0_1_dwconv_conv_weight_to_fp16_palettized")]; + tensor decoder_upsample_0_1_dwconv_conv_bias_to_fp16 = const()[name = string("decoder_upsample_0_1_dwconv_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42864768)))]; + tensor x_57_cast_fp16 = conv(bias = decoder_upsample_0_1_dwconv_conv_bias_to_fp16, dilations = x_57_dilations_0, groups = x_57_groups_0, pad = x_57_pad_0, pad_type = x_57_pad_type_0, strides = x_57_strides_0, weight = decoder_upsample_0_1_dwconv_conv_weight_to_fp16_palettized, x = input_123_cast_fp16)[name = string("x_57_cast_fp16")]; + tensor var_2169_axes_0 = const()[name = string("op_2169_axes_0"), val = tensor([2])]; + tensor var_2169_cast_fp16 = squeeze(axes = var_2169_axes_0, x = x_57_cast_fp16)[name = string("op_2169_cast_fp16")]; + tensor var_2170 = const()[name = string("op_2170"), val = tensor([0, 2, 1])]; + tensor x_59_axes_0 = const()[name = string("x_59_axes_0"), val = tensor([-1])]; + tensor decoder_upsample_0_1_norm_weight_to_fp16 = const()[name = string("decoder_upsample_0_1_norm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42866880)))]; + tensor decoder_upsample_0_1_norm_bias_to_fp16 = const()[name = string("decoder_upsample_0_1_norm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42868992)))]; + fp16 var_17_to_fp16 = const()[name = string("op_17_to_fp16"), val = fp16(0x1.1p-20)]; + tensor input_125_cast_fp16 = transpose(perm = var_2170, x = var_2169_cast_fp16)[name = string("transpose_3")]; + tensor x_59_cast_fp16 = layer_norm(axes = x_59_axes_0, beta = decoder_upsample_0_1_norm_bias_to_fp16, epsilon = var_17_to_fp16, gamma = decoder_upsample_0_1_norm_weight_to_fp16, x = input_125_cast_fp16)[name = string("x_59_cast_fp16")]; + tensor var_2176 = const()[name = string("op_2176"), val = tensor([0, 2, 1])]; + tensor input_127_axes_0 = const()[name = string("input_127_axes_0"), val = tensor([2])]; + tensor var_2177_cast_fp16 = transpose(perm = var_2176, x = x_59_cast_fp16)[name = string("transpose_2")]; + tensor input_127_cast_fp16 = expand_dims(axes = input_127_axes_0, x = var_2177_cast_fp16)[name = string("input_127_cast_fp16")]; + string input_129_pad_type_0 = const()[name = string("input_129_pad_type_0"), val = string("valid")]; + tensor input_129_strides_0 = const()[name = string("input_129_strides_0"), val = tensor([1, 1])]; + tensor input_129_pad_0 = const()[name = string("input_129_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_129_dilations_0 = const()[name = string("input_129_dilations_0"), val = tensor([1, 1])]; + int32 input_129_groups_0 = const()[name = string("input_129_groups_0"), val = int32(1)]; + tensor decoder_upsample_0_1_pwconv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42871104))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47065472))))[name = string("decoder_upsample_0_1_pwconv1_weight_to_fp16_palettized")]; + tensor decoder_upsample_0_1_pwconv1_bias_to_fp16 = const()[name = string("decoder_upsample_0_1_pwconv1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47066048)))]; + tensor input_129_cast_fp16 = conv(bias = decoder_upsample_0_1_pwconv1_bias_to_fp16, dilations = input_129_dilations_0, groups = input_129_groups_0, pad = input_129_pad_0, pad_type = input_129_pad_type_0, strides = input_129_strides_0, weight = decoder_upsample_0_1_pwconv1_weight_to_fp16_palettized, x = input_127_cast_fp16)[name = string("input_129_cast_fp16")]; + string input_131_mode_0 = const()[name = string("input_131_mode_0"), val = string("EXACT")]; + tensor input_131_cast_fp16 = gelu(mode = input_131_mode_0, x = input_129_cast_fp16)[name = string("input_131_cast_fp16")]; + string x_61_pad_type_0 = const()[name = string("x_61_pad_type_0"), val = string("valid")]; + tensor x_61_strides_0 = const()[name = string("x_61_strides_0"), val = tensor([1, 1])]; + tensor x_61_pad_0 = const()[name = string("x_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_61_dilations_0 = const()[name = string("x_61_dilations_0"), val = tensor([1, 1])]; + int32 x_61_groups_0 = const()[name = string("x_61_groups_0"), val = int32(1)]; + tensor x_63_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47074304))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51268672))))[name = string("x_63_weight_0_to_fp16_palettized")]; + tensor x_63_bias_0_to_fp16 = const()[name = string("x_63_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51269248)))]; + tensor x_63_cast_fp16 = conv(bias = x_63_bias_0_to_fp16, dilations = x_61_dilations_0, groups = x_61_groups_0, pad = x_61_pad_0, pad_type = x_61_pad_type_0, strides = x_61_strides_0, weight = x_63_weight_0_to_fp16_palettized, x = input_131_cast_fp16)[name = string("x_63_cast_fp16")]; + tensor x_65_cast_fp16 = add(x = x_53_has_output_shape_cast_fp16, y = x_63_cast_fp16)[name = string("x_65_cast_fp16")]; + string x_67_pad_type_0 = const()[name = string("x_67_pad_type_0"), val = string("valid")]; + tensor x_67_strides_0 = const()[name = string("x_67_strides_0"), val = tensor([1, 2])]; + tensor x_67_pad_0 = const()[name = string("x_67_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_67_dilations_0 = const()[name = string("x_67_dilations_0"), val = tensor([1, 1])]; + int32 x_67_groups_0 = const()[name = string("x_67_groups_0"), val = int32(1)]; + tensor x_67_has_output_shape_output_shape_0 = const()[name = string("x_67_has_output_shape_output_shape_0"), val = tensor([1, 1024, 1, 20])]; + tensor decoder_upsample_1_0_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51271360))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53368576))))[name = string("decoder_upsample_1_0_conv_weight_to_fp16_palettized")]; + tensor decoder_upsample_1_0_conv_bias_to_fp16 = const()[name = string("decoder_upsample_1_0_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53369152)))]; + tensor x_67_has_output_shape_cast_fp16 = conv_transpose(bias = decoder_upsample_1_0_conv_bias_to_fp16, dilations = x_67_dilations_0, groups = x_67_groups_0, output_shape = x_67_has_output_shape_output_shape_0, pad = x_67_pad_0, pad_type = x_67_pad_type_0, strides = x_67_strides_0, weight = decoder_upsample_1_0_conv_weight_to_fp16_palettized, x = x_65_cast_fp16)[name = string("x_67_has_output_shape_cast_fp16")]; + tensor input_133_pad_0 = const()[name = string("input_133_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 6, 0])]; + string input_133_mode_0 = const()[name = string("input_133_mode_0"), val = string("constant")]; + fp16 const_88_to_fp16 = const()[name = string("const_88_to_fp16"), val = fp16(0x0p+0)]; + tensor input_133_cast_fp16 = pad(constant_val = const_88_to_fp16, mode = input_133_mode_0, pad = input_133_pad_0, x = x_67_has_output_shape_cast_fp16)[name = string("input_133_cast_fp16")]; + string x_71_pad_type_0 = const()[name = string("x_71_pad_type_0"), val = string("valid")]; + int32 x_71_groups_0 = const()[name = string("x_71_groups_0"), val = int32(1024)]; + tensor x_71_strides_0 = const()[name = string("x_71_strides_0"), val = tensor([1, 1])]; + tensor x_71_pad_0 = const()[name = string("x_71_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_71_dilations_0 = const()[name = string("x_71_dilations_0"), val = tensor([1, 1])]; + tensor decoder_upsample_1_1_dwconv_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53371264))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53378496))))[name = string("decoder_upsample_1_1_dwconv_conv_weight_to_fp16_palettized")]; + tensor decoder_upsample_1_1_dwconv_conv_bias_to_fp16 = const()[name = string("decoder_upsample_1_1_dwconv_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53379072)))]; + tensor x_71_cast_fp16 = conv(bias = decoder_upsample_1_1_dwconv_conv_bias_to_fp16, dilations = x_71_dilations_0, groups = x_71_groups_0, pad = x_71_pad_0, pad_type = x_71_pad_type_0, strides = x_71_strides_0, weight = decoder_upsample_1_1_dwconv_conv_weight_to_fp16_palettized, x = input_133_cast_fp16)[name = string("x_71_cast_fp16")]; + tensor var_2231_axes_0 = const()[name = string("op_2231_axes_0"), val = tensor([2])]; + tensor var_2231_cast_fp16 = squeeze(axes = var_2231_axes_0, x = x_71_cast_fp16)[name = string("op_2231_cast_fp16")]; + tensor var_2232 = const()[name = string("op_2232"), val = tensor([0, 2, 1])]; + tensor x_73_axes_0 = const()[name = string("x_73_axes_0"), val = tensor([-1])]; + tensor decoder_upsample_1_1_norm_weight_to_fp16 = const()[name = string("decoder_upsample_1_1_norm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53381184)))]; + tensor decoder_upsample_1_1_norm_bias_to_fp16 = const()[name = string("decoder_upsample_1_1_norm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53383296)))]; + tensor input_135_cast_fp16 = transpose(perm = var_2232, x = var_2231_cast_fp16)[name = string("transpose_1")]; + tensor x_73_cast_fp16 = layer_norm(axes = x_73_axes_0, beta = decoder_upsample_1_1_norm_bias_to_fp16, epsilon = var_17_to_fp16, gamma = decoder_upsample_1_1_norm_weight_to_fp16, x = input_135_cast_fp16)[name = string("x_73_cast_fp16")]; + tensor var_2238 = const()[name = string("op_2238"), val = tensor([0, 2, 1])]; + tensor input_137_axes_0 = const()[name = string("input_137_axes_0"), val = tensor([2])]; + tensor var_2239_cast_fp16 = transpose(perm = var_2238, x = x_73_cast_fp16)[name = string("transpose_0")]; + tensor input_137_cast_fp16 = expand_dims(axes = input_137_axes_0, x = var_2239_cast_fp16)[name = string("input_137_cast_fp16")]; + string input_139_pad_type_0 = const()[name = string("input_139_pad_type_0"), val = string("valid")]; + tensor input_139_strides_0 = const()[name = string("input_139_strides_0"), val = tensor([1, 1])]; + tensor input_139_pad_0 = const()[name = string("input_139_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_139_dilations_0 = const()[name = string("input_139_dilations_0"), val = tensor([1, 1])]; + int32 input_139_groups_0 = const()[name = string("input_139_groups_0"), val = int32(1)]; + tensor decoder_upsample_1_1_pwconv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53385408))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57579776))))[name = string("decoder_upsample_1_1_pwconv1_weight_to_fp16_palettized")]; + tensor decoder_upsample_1_1_pwconv1_bias_to_fp16 = const()[name = string("decoder_upsample_1_1_pwconv1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57580352)))]; + tensor input_139_cast_fp16 = conv(bias = decoder_upsample_1_1_pwconv1_bias_to_fp16, dilations = input_139_dilations_0, groups = input_139_groups_0, pad = input_139_pad_0, pad_type = input_139_pad_type_0, strides = input_139_strides_0, weight = decoder_upsample_1_1_pwconv1_weight_to_fp16_palettized, x = input_137_cast_fp16)[name = string("input_139_cast_fp16")]; + string input_141_mode_0 = const()[name = string("input_141_mode_0"), val = string("EXACT")]; + tensor input_141_cast_fp16 = gelu(mode = input_141_mode_0, x = input_139_cast_fp16)[name = string("input_141_cast_fp16")]; + string x_75_pad_type_0 = const()[name = string("x_75_pad_type_0"), val = string("valid")]; + tensor x_75_strides_0 = const()[name = string("x_75_strides_0"), val = tensor([1, 1])]; + tensor x_75_pad_0 = const()[name = string("x_75_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_75_dilations_0 = const()[name = string("x_75_dilations_0"), val = tensor([1, 1])]; + int32 x_75_groups_0 = const()[name = string("x_75_groups_0"), val = int32(1)]; + tensor x_77_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57588608))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61782976))))[name = string("x_77_weight_0_to_fp16_palettized")]; + tensor x_77_bias_0_to_fp16 = const()[name = string("x_77_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61783552)))]; + tensor x_77_cast_fp16 = conv(bias = x_77_bias_0_to_fp16, dilations = x_75_dilations_0, groups = x_75_groups_0, pad = x_75_pad_0, pad_type = x_75_pad_type_0, strides = x_75_strides_0, weight = x_77_weight_0_to_fp16_palettized, x = input_141_cast_fp16)[name = string("x_77_cast_fp16")]; + tensor x_79_cast_fp16 = add(x = x_67_has_output_shape_cast_fp16, y = x_77_cast_fp16)[name = string("x_79_cast_fp16")]; + tensor input_143_pad_0 = const()[name = string("input_143_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 6, 0])]; + string input_143_mode_0 = const()[name = string("input_143_mode_0"), val = string("constant")]; + fp16 const_89_to_fp16 = const()[name = string("const_89_to_fp16"), val = fp16(0x0p+0)]; + tensor input_143_cast_fp16 = pad(constant_val = const_89_to_fp16, mode = input_143_mode_0, pad = input_143_pad_0, x = x_79_cast_fp16)[name = string("input_143_cast_fp16")]; + string x_81_pad_type_0 = const()[name = string("x_81_pad_type_0"), val = string("valid")]; + tensor x_81_strides_0 = const()[name = string("x_81_strides_0"), val = tensor([1, 1])]; + tensor x_81_pad_0 = const()[name = string("x_81_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_81_dilations_0 = const()[name = string("x_81_dilations_0"), val = tensor([1, 1])]; + int32 x_81_groups_0 = const()[name = string("x_81_groups_0"), val = int32(1)]; + tensor decoder_decoder_0_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61785664))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72795776))))[name = string("decoder_decoder_0_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_0_conv_bias_to_fp16 = const()[name = string("decoder_decoder_0_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72796352)))]; + tensor x_81_cast_fp16 = conv(bias = decoder_decoder_0_conv_bias_to_fp16, dilations = x_81_dilations_0, groups = x_81_groups_0, pad = x_81_pad_0, pad_type = x_81_pad_type_0, strides = x_81_strides_0, weight = decoder_decoder_0_conv_weight_to_fp16_palettized, x = input_143_cast_fp16)[name = string("x_81_cast_fp16")]; + tensor alpha_1_to_fp16 = const()[name = string("alpha_1_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72799488)))]; + tensor var_2288_cast_fp16 = mul(x = x_81_cast_fp16, y = alpha_1_to_fp16)[name = string("op_2288_cast_fp16")]; + tensor sin_val_1_cast_fp16 = sin(x = var_2288_cast_fp16)[name = string("sin_val_1_cast_fp16")]; + tensor var_2295_cast_fp16 = mul(x = sin_val_1_cast_fp16, y = sin_val_1_cast_fp16)[name = string("op_2295_cast_fp16")]; + tensor var_2292_to_fp16 = const()[name = string("op_2292_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72802624)))]; + tensor var_2296_cast_fp16 = mul(x = var_2292_to_fp16, y = var_2295_cast_fp16)[name = string("op_2296_cast_fp16")]; + tensor x_83_cast_fp16 = add(x = x_81_cast_fp16, y = var_2296_cast_fp16)[name = string("x_83_cast_fp16")]; + string x_85_pad_type_0 = const()[name = string("x_85_pad_type_0"), val = string("valid")]; + tensor x_85_strides_0 = const()[name = string("x_85_strides_0"), val = tensor([1, 8])]; + tensor x_85_pad_0 = const()[name = string("x_85_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_85_dilations_0 = const()[name = string("x_85_dilations_0"), val = tensor([1, 1])]; + int32 x_85_groups_0 = const()[name = string("x_85_groups_0"), val = int32(1)]; + tensor x_85_has_output_shape_output_shape_0 = const()[name = string("x_85_has_output_shape_output_shape_0"), val = tensor([1, 768, 1, 168])]; + tensor decoder_decoder_1_block_1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72805760))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91680192))))[name = string("decoder_decoder_1_block_1_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_1_block_1_conv_bias_to_fp16 = const()[name = string("decoder_decoder_1_block_1_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91680768)))]; + tensor x_85_has_output_shape_cast_fp16 = conv_transpose(bias = decoder_decoder_1_block_1_conv_bias_to_fp16, dilations = x_85_dilations_0, groups = x_85_groups_0, output_shape = x_85_has_output_shape_output_shape_0, pad = x_85_pad_0, pad_type = x_85_pad_type_0, strides = x_85_strides_0, weight = decoder_decoder_1_block_1_conv_weight_to_fp16_palettized, x = x_83_cast_fp16)[name = string("x_85_has_output_shape_cast_fp16")]; + tensor x_87_begin_0 = const()[name = string("x_87_begin_0"), val = tensor([0, 0, 0, 8])]; + tensor x_87_end_0 = const()[name = string("x_87_end_0"), val = tensor([1, 768, 1, 160])]; + tensor x_87_end_mask_0 = const()[name = string("x_87_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_87_cast_fp16 = slice_by_index(begin = x_87_begin_0, end = x_87_end_0, end_mask = x_87_end_mask_0, x = x_85_has_output_shape_cast_fp16)[name = string("x_87_cast_fp16")]; + tensor alpha_3_to_fp16 = const()[name = string("alpha_3_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91682368)))]; + tensor var_2328_cast_fp16 = mul(x = x_87_cast_fp16, y = alpha_3_to_fp16)[name = string("op_2328_cast_fp16")]; + tensor sin_val_3_cast_fp16 = sin(x = var_2328_cast_fp16)[name = string("sin_val_3_cast_fp16")]; + tensor var_2335_cast_fp16 = mul(x = sin_val_3_cast_fp16, y = sin_val_3_cast_fp16)[name = string("op_2335_cast_fp16")]; + tensor var_2332_to_fp16 = const()[name = string("op_2332_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91683968)))]; + tensor var_2336_cast_fp16 = mul(x = var_2332_to_fp16, y = var_2335_cast_fp16)[name = string("op_2336_cast_fp16")]; + tensor x_89_cast_fp16 = add(x = x_87_cast_fp16, y = var_2336_cast_fp16)[name = string("x_89_cast_fp16")]; + tensor input_145_pad_0 = const()[name = string("input_145_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 6, 0])]; + string input_145_mode_0 = const()[name = string("input_145_mode_0"), val = string("constant")]; + fp16 const_91_to_fp16 = const()[name = string("const_91_to_fp16"), val = fp16(0x0p+0)]; + tensor input_145_cast_fp16 = pad(constant_val = const_91_to_fp16, mode = input_145_mode_0, pad = input_145_pad_0, x = x_89_cast_fp16)[name = string("input_145_cast_fp16")]; + string x_91_pad_type_0 = const()[name = string("x_91_pad_type_0"), val = string("valid")]; + tensor x_91_strides_0 = const()[name = string("x_91_strides_0"), val = tensor([1, 1])]; + tensor x_91_pad_0 = const()[name = string("x_91_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_91_dilations_0 = const()[name = string("x_91_dilations_0"), val = tensor([1, 1])]; + int32 x_91_groups_0 = const()[name = string("x_91_groups_0"), val = int32(1)]; + tensor decoder_decoder_1_block_2_conv1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91685568))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95814400))))[name = string("decoder_decoder_1_block_2_conv1_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_1_block_2_conv1_conv_bias_to_fp16 = const()[name = string("decoder_decoder_1_block_2_conv1_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95814976)))]; + tensor x_91_cast_fp16 = conv(bias = decoder_decoder_1_block_2_conv1_conv_bias_to_fp16, dilations = x_91_dilations_0, groups = x_91_groups_0, pad = x_91_pad_0, pad_type = x_91_pad_type_0, strides = x_91_strides_0, weight = decoder_decoder_1_block_2_conv1_conv_weight_to_fp16_palettized, x = input_145_cast_fp16)[name = string("x_91_cast_fp16")]; + tensor alpha_5_to_fp16 = const()[name = string("alpha_5_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95816576)))]; + tensor var_2356_cast_fp16 = mul(x = x_91_cast_fp16, y = alpha_5_to_fp16)[name = string("op_2356_cast_fp16")]; + tensor sin_val_5_cast_fp16 = sin(x = var_2356_cast_fp16)[name = string("sin_val_5_cast_fp16")]; + tensor var_2363_cast_fp16 = mul(x = sin_val_5_cast_fp16, y = sin_val_5_cast_fp16)[name = string("op_2363_cast_fp16")]; + tensor var_2360_to_fp16 = const()[name = string("op_2360_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95818176)))]; + tensor var_2364_cast_fp16 = mul(x = var_2360_to_fp16, y = var_2363_cast_fp16)[name = string("op_2364_cast_fp16")]; + tensor x_93_cast_fp16 = add(x = x_91_cast_fp16, y = var_2364_cast_fp16)[name = string("x_93_cast_fp16")]; + string x_95_pad_type_0 = const()[name = string("x_95_pad_type_0"), val = string("valid")]; + tensor x_95_strides_0 = const()[name = string("x_95_strides_0"), val = tensor([1, 1])]; + tensor x_95_pad_0 = const()[name = string("x_95_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_95_dilations_0 = const()[name = string("x_95_dilations_0"), val = tensor([1, 1])]; + int32 x_95_groups_0 = const()[name = string("x_95_groups_0"), val = int32(1)]; + tensor decoder_decoder_1_block_2_conv2_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95819776))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96409664))))[name = string("decoder_decoder_1_block_2_conv2_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_1_block_2_conv2_conv_bias_to_fp16 = const()[name = string("decoder_decoder_1_block_2_conv2_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96410240)))]; + tensor x_95_cast_fp16 = conv(bias = decoder_decoder_1_block_2_conv2_conv_bias_to_fp16, dilations = x_95_dilations_0, groups = x_95_groups_0, pad = x_95_pad_0, pad_type = x_95_pad_type_0, strides = x_95_strides_0, weight = decoder_decoder_1_block_2_conv2_conv_weight_to_fp16_palettized, x = x_93_cast_fp16)[name = string("x_95_cast_fp16")]; + tensor x_97_cast_fp16 = add(x = x_95_cast_fp16, y = x_87_cast_fp16)[name = string("x_97_cast_fp16")]; + tensor alpha_7_to_fp16 = const()[name = string("alpha_7_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96411840)))]; + tensor var_2389_cast_fp16 = mul(x = x_97_cast_fp16, y = alpha_7_to_fp16)[name = string("op_2389_cast_fp16")]; + tensor sin_val_7_cast_fp16 = sin(x = var_2389_cast_fp16)[name = string("sin_val_7_cast_fp16")]; + tensor var_2396_cast_fp16 = mul(x = sin_val_7_cast_fp16, y = sin_val_7_cast_fp16)[name = string("op_2396_cast_fp16")]; + tensor var_2393_to_fp16 = const()[name = string("op_2393_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96413440)))]; + tensor var_2397_cast_fp16 = mul(x = var_2393_to_fp16, y = var_2396_cast_fp16)[name = string("op_2397_cast_fp16")]; + tensor x_99_cast_fp16 = add(x = x_97_cast_fp16, y = var_2397_cast_fp16)[name = string("x_99_cast_fp16")]; + tensor input_149_pad_0 = const()[name = string("input_149_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 18, 0])]; + string input_149_mode_0 = const()[name = string("input_149_mode_0"), val = string("constant")]; + fp16 const_93_to_fp16 = const()[name = string("const_93_to_fp16"), val = fp16(0x0p+0)]; + tensor input_149_cast_fp16 = pad(constant_val = const_93_to_fp16, mode = input_149_mode_0, pad = input_149_pad_0, x = x_99_cast_fp16)[name = string("input_149_cast_fp16")]; + string x_101_pad_type_0 = const()[name = string("x_101_pad_type_0"), val = string("valid")]; + tensor x_101_dilations_0 = const()[name = string("x_101_dilations_0"), val = tensor([1, 3])]; + tensor x_101_strides_0 = const()[name = string("x_101_strides_0"), val = tensor([1, 1])]; + tensor x_101_pad_0 = const()[name = string("x_101_pad_0"), val = tensor([0, 0, 0, 0])]; + int32 x_101_groups_0 = const()[name = string("x_101_groups_0"), val = int32(1)]; + tensor decoder_decoder_1_block_3_conv1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96415040))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100543872))))[name = string("decoder_decoder_1_block_3_conv1_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_1_block_3_conv1_conv_bias_to_fp16 = const()[name = string("decoder_decoder_1_block_3_conv1_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100544448)))]; + tensor x_101_cast_fp16 = conv(bias = decoder_decoder_1_block_3_conv1_conv_bias_to_fp16, dilations = x_101_dilations_0, groups = x_101_groups_0, pad = x_101_pad_0, pad_type = x_101_pad_type_0, strides = x_101_strides_0, weight = decoder_decoder_1_block_3_conv1_conv_weight_to_fp16_palettized, x = input_149_cast_fp16)[name = string("x_101_cast_fp16")]; + tensor alpha_9_to_fp16 = const()[name = string("alpha_9_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100546048)))]; + tensor var_2417_cast_fp16 = mul(x = x_101_cast_fp16, y = alpha_9_to_fp16)[name = string("op_2417_cast_fp16")]; + tensor sin_val_9_cast_fp16 = sin(x = var_2417_cast_fp16)[name = string("sin_val_9_cast_fp16")]; + tensor var_2424_cast_fp16 = mul(x = sin_val_9_cast_fp16, y = sin_val_9_cast_fp16)[name = string("op_2424_cast_fp16")]; + tensor var_2421_to_fp16 = const()[name = string("op_2421_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100547648)))]; + tensor var_2425_cast_fp16 = mul(x = var_2421_to_fp16, y = var_2424_cast_fp16)[name = string("op_2425_cast_fp16")]; + tensor x_103_cast_fp16 = add(x = x_101_cast_fp16, y = var_2425_cast_fp16)[name = string("x_103_cast_fp16")]; + string x_105_pad_type_0 = const()[name = string("x_105_pad_type_0"), val = string("valid")]; + tensor x_105_strides_0 = const()[name = string("x_105_strides_0"), val = tensor([1, 1])]; + tensor x_105_pad_0 = const()[name = string("x_105_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_105_dilations_0 = const()[name = string("x_105_dilations_0"), val = tensor([1, 1])]; + int32 x_105_groups_0 = const()[name = string("x_105_groups_0"), val = int32(1)]; + tensor decoder_decoder_1_block_3_conv2_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100549248))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101139136))))[name = string("decoder_decoder_1_block_3_conv2_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_1_block_3_conv2_conv_bias_to_fp16 = const()[name = string("decoder_decoder_1_block_3_conv2_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101139712)))]; + tensor x_105_cast_fp16 = conv(bias = decoder_decoder_1_block_3_conv2_conv_bias_to_fp16, dilations = x_105_dilations_0, groups = x_105_groups_0, pad = x_105_pad_0, pad_type = x_105_pad_type_0, strides = x_105_strides_0, weight = decoder_decoder_1_block_3_conv2_conv_weight_to_fp16_palettized, x = x_103_cast_fp16)[name = string("x_105_cast_fp16")]; + tensor x_107_cast_fp16 = add(x = x_105_cast_fp16, y = x_97_cast_fp16)[name = string("x_107_cast_fp16")]; + tensor alpha_11_to_fp16 = const()[name = string("alpha_11_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101141312)))]; + tensor var_2450_cast_fp16 = mul(x = x_107_cast_fp16, y = alpha_11_to_fp16)[name = string("op_2450_cast_fp16")]; + tensor sin_val_11_cast_fp16 = sin(x = var_2450_cast_fp16)[name = string("sin_val_11_cast_fp16")]; + tensor var_2457_cast_fp16 = mul(x = sin_val_11_cast_fp16, y = sin_val_11_cast_fp16)[name = string("op_2457_cast_fp16")]; + tensor var_2454_to_fp16 = const()[name = string("op_2454_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101142912)))]; + tensor var_2458_cast_fp16 = mul(x = var_2454_to_fp16, y = var_2457_cast_fp16)[name = string("op_2458_cast_fp16")]; + tensor x_109_cast_fp16 = add(x = x_107_cast_fp16, y = var_2458_cast_fp16)[name = string("x_109_cast_fp16")]; + tensor input_153_pad_0 = const()[name = string("input_153_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 54, 0])]; + string input_153_mode_0 = const()[name = string("input_153_mode_0"), val = string("constant")]; + fp16 const_95_to_fp16 = const()[name = string("const_95_to_fp16"), val = fp16(0x0p+0)]; + tensor input_153_cast_fp16 = pad(constant_val = const_95_to_fp16, mode = input_153_mode_0, pad = input_153_pad_0, x = x_109_cast_fp16)[name = string("input_153_cast_fp16")]; + string x_111_pad_type_0 = const()[name = string("x_111_pad_type_0"), val = string("valid")]; + tensor x_111_dilations_0 = const()[name = string("x_111_dilations_0"), val = tensor([1, 9])]; + tensor x_111_strides_0 = const()[name = string("x_111_strides_0"), val = tensor([1, 1])]; + tensor x_111_pad_0 = const()[name = string("x_111_pad_0"), val = tensor([0, 0, 0, 0])]; + int32 x_111_groups_0 = const()[name = string("x_111_groups_0"), val = int32(1)]; + tensor decoder_decoder_1_block_4_conv1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101144512))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105273344))))[name = string("decoder_decoder_1_block_4_conv1_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_1_block_4_conv1_conv_bias_to_fp16 = const()[name = string("decoder_decoder_1_block_4_conv1_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105273920)))]; + tensor x_111_cast_fp16 = conv(bias = decoder_decoder_1_block_4_conv1_conv_bias_to_fp16, dilations = x_111_dilations_0, groups = x_111_groups_0, pad = x_111_pad_0, pad_type = x_111_pad_type_0, strides = x_111_strides_0, weight = decoder_decoder_1_block_4_conv1_conv_weight_to_fp16_palettized, x = input_153_cast_fp16)[name = string("x_111_cast_fp16")]; + tensor alpha_13_to_fp16 = const()[name = string("alpha_13_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105275520)))]; + tensor var_2478_cast_fp16 = mul(x = x_111_cast_fp16, y = alpha_13_to_fp16)[name = string("op_2478_cast_fp16")]; + tensor sin_val_13_cast_fp16 = sin(x = var_2478_cast_fp16)[name = string("sin_val_13_cast_fp16")]; + tensor var_2485_cast_fp16 = mul(x = sin_val_13_cast_fp16, y = sin_val_13_cast_fp16)[name = string("op_2485_cast_fp16")]; + tensor var_2482_to_fp16 = const()[name = string("op_2482_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105277120)))]; + tensor var_2486_cast_fp16 = mul(x = var_2482_to_fp16, y = var_2485_cast_fp16)[name = string("op_2486_cast_fp16")]; + tensor x_113_cast_fp16 = add(x = x_111_cast_fp16, y = var_2486_cast_fp16)[name = string("x_113_cast_fp16")]; + string x_115_pad_type_0 = const()[name = string("x_115_pad_type_0"), val = string("valid")]; + tensor x_115_strides_0 = const()[name = string("x_115_strides_0"), val = tensor([1, 1])]; + tensor x_115_pad_0 = const()[name = string("x_115_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_115_dilations_0 = const()[name = string("x_115_dilations_0"), val = tensor([1, 1])]; + int32 x_115_groups_0 = const()[name = string("x_115_groups_0"), val = int32(1)]; + tensor decoder_decoder_1_block_4_conv2_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105278720))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105868608))))[name = string("decoder_decoder_1_block_4_conv2_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_1_block_4_conv2_conv_bias_to_fp16 = const()[name = string("decoder_decoder_1_block_4_conv2_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105869184)))]; + tensor x_115_cast_fp16 = conv(bias = decoder_decoder_1_block_4_conv2_conv_bias_to_fp16, dilations = x_115_dilations_0, groups = x_115_groups_0, pad = x_115_pad_0, pad_type = x_115_pad_type_0, strides = x_115_strides_0, weight = decoder_decoder_1_block_4_conv2_conv_weight_to_fp16_palettized, x = x_113_cast_fp16)[name = string("x_115_cast_fp16")]; + tensor x_117_cast_fp16 = add(x = x_115_cast_fp16, y = x_107_cast_fp16)[name = string("x_117_cast_fp16")]; + tensor alpha_15_to_fp16 = const()[name = string("alpha_15_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105870784)))]; + tensor var_2517_cast_fp16 = mul(x = x_117_cast_fp16, y = alpha_15_to_fp16)[name = string("op_2517_cast_fp16")]; + tensor sin_val_15_cast_fp16 = sin(x = var_2517_cast_fp16)[name = string("sin_val_15_cast_fp16")]; + tensor var_2524_cast_fp16 = mul(x = sin_val_15_cast_fp16, y = sin_val_15_cast_fp16)[name = string("op_2524_cast_fp16")]; + tensor var_2521_to_fp16 = const()[name = string("op_2521_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105872384)))]; + tensor var_2525_cast_fp16 = mul(x = var_2521_to_fp16, y = var_2524_cast_fp16)[name = string("op_2525_cast_fp16")]; + tensor x_119_cast_fp16 = add(x = x_117_cast_fp16, y = var_2525_cast_fp16)[name = string("x_119_cast_fp16")]; + string x_121_pad_type_0 = const()[name = string("x_121_pad_type_0"), val = string("valid")]; + tensor x_121_strides_0 = const()[name = string("x_121_strides_0"), val = tensor([1, 5])]; + tensor x_121_pad_0 = const()[name = string("x_121_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_121_dilations_0 = const()[name = string("x_121_dilations_0"), val = tensor([1, 1])]; + int32 x_121_groups_0 = const()[name = string("x_121_groups_0"), val = int32(1)]; + tensor x_121_has_output_shape_output_shape_0 = const()[name = string("x_121_has_output_shape_output_shape_0"), val = tensor([1, 384, 1, 765])]; + tensor decoder_decoder_2_block_1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105873984))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108823168))))[name = string("decoder_decoder_2_block_1_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_2_block_1_conv_bias_to_fp16 = const()[name = string("decoder_decoder_2_block_1_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108823744)))]; + tensor x_121_has_output_shape_cast_fp16 = conv_transpose(bias = decoder_decoder_2_block_1_conv_bias_to_fp16, dilations = x_121_dilations_0, groups = x_121_groups_0, output_shape = x_121_has_output_shape_output_shape_0, pad = x_121_pad_0, pad_type = x_121_pad_type_0, strides = x_121_strides_0, weight = decoder_decoder_2_block_1_conv_weight_to_fp16_palettized, x = x_119_cast_fp16)[name = string("x_121_has_output_shape_cast_fp16")]; + tensor x_123_begin_0 = const()[name = string("x_123_begin_0"), val = tensor([0, 0, 0, 5])]; + tensor x_123_end_0 = const()[name = string("x_123_end_0"), val = tensor([1, 384, 1, 760])]; + tensor x_123_end_mask_0 = const()[name = string("x_123_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_123_cast_fp16 = slice_by_index(begin = x_123_begin_0, end = x_123_end_0, end_mask = x_123_end_mask_0, x = x_121_has_output_shape_cast_fp16)[name = string("x_123_cast_fp16")]; + tensor alpha_17_to_fp16 = const()[name = string("alpha_17_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108824576)))]; + tensor var_2557_cast_fp16 = mul(x = x_123_cast_fp16, y = alpha_17_to_fp16)[name = string("op_2557_cast_fp16")]; + tensor sin_val_17_cast_fp16 = sin(x = var_2557_cast_fp16)[name = string("sin_val_17_cast_fp16")]; + tensor var_2564_cast_fp16 = mul(x = sin_val_17_cast_fp16, y = sin_val_17_cast_fp16)[name = string("op_2564_cast_fp16")]; + tensor var_2561_to_fp16 = const()[name = string("op_2561_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108825408)))]; + tensor var_2565_cast_fp16 = mul(x = var_2561_to_fp16, y = var_2564_cast_fp16)[name = string("op_2565_cast_fp16")]; + tensor x_125_cast_fp16 = add(x = x_123_cast_fp16, y = var_2565_cast_fp16)[name = string("x_125_cast_fp16")]; + tensor input_157_pad_0 = const()[name = string("input_157_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 6, 0])]; + string input_157_mode_0 = const()[name = string("input_157_mode_0"), val = string("constant")]; + fp16 const_98_to_fp16 = const()[name = string("const_98_to_fp16"), val = fp16(0x0p+0)]; + tensor input_157_cast_fp16 = pad(constant_val = const_98_to_fp16, mode = input_157_mode_0, pad = input_157_pad_0, x = x_125_cast_fp16)[name = string("input_157_cast_fp16")]; + string x_127_pad_type_0 = const()[name = string("x_127_pad_type_0"), val = string("valid")]; + tensor x_127_strides_0 = const()[name = string("x_127_strides_0"), val = tensor([1, 1])]; + tensor x_127_pad_0 = const()[name = string("x_127_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_127_dilations_0 = const()[name = string("x_127_dilations_0"), val = tensor([1, 1])]; + int32 x_127_groups_0 = const()[name = string("x_127_groups_0"), val = int32(1)]; + tensor decoder_decoder_2_block_2_conv1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108826240))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109858496))))[name = string("decoder_decoder_2_block_2_conv1_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_2_block_2_conv1_conv_bias_to_fp16 = const()[name = string("decoder_decoder_2_block_2_conv1_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109859072)))]; + tensor x_127_cast_fp16 = conv(bias = decoder_decoder_2_block_2_conv1_conv_bias_to_fp16, dilations = x_127_dilations_0, groups = x_127_groups_0, pad = x_127_pad_0, pad_type = x_127_pad_type_0, strides = x_127_strides_0, weight = decoder_decoder_2_block_2_conv1_conv_weight_to_fp16_palettized, x = input_157_cast_fp16)[name = string("x_127_cast_fp16")]; + tensor alpha_19_to_fp16 = const()[name = string("alpha_19_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109859904)))]; + tensor var_2585_cast_fp16 = mul(x = x_127_cast_fp16, y = alpha_19_to_fp16)[name = string("op_2585_cast_fp16")]; + tensor sin_val_19_cast_fp16 = sin(x = var_2585_cast_fp16)[name = string("sin_val_19_cast_fp16")]; + tensor var_2592_cast_fp16 = mul(x = sin_val_19_cast_fp16, y = sin_val_19_cast_fp16)[name = string("op_2592_cast_fp16")]; + tensor var_2589_to_fp16 = const()[name = string("op_2589_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109860736)))]; + tensor var_2593_cast_fp16 = mul(x = var_2589_to_fp16, y = var_2592_cast_fp16)[name = string("op_2593_cast_fp16")]; + tensor x_129_cast_fp16 = add(x = x_127_cast_fp16, y = var_2593_cast_fp16)[name = string("x_129_cast_fp16")]; + string x_131_pad_type_0 = const()[name = string("x_131_pad_type_0"), val = string("valid")]; + tensor x_131_strides_0 = const()[name = string("x_131_strides_0"), val = tensor([1, 1])]; + tensor x_131_pad_0 = const()[name = string("x_131_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_131_dilations_0 = const()[name = string("x_131_dilations_0"), val = tensor([1, 1])]; + int32 x_131_groups_0 = const()[name = string("x_131_groups_0"), val = int32(1)]; + tensor decoder_decoder_2_block_2_conv2_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109861568))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110009088))))[name = string("decoder_decoder_2_block_2_conv2_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_2_block_2_conv2_conv_bias_to_fp16 = const()[name = string("decoder_decoder_2_block_2_conv2_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110009664)))]; + tensor x_131_cast_fp16 = conv(bias = decoder_decoder_2_block_2_conv2_conv_bias_to_fp16, dilations = x_131_dilations_0, groups = x_131_groups_0, pad = x_131_pad_0, pad_type = x_131_pad_type_0, strides = x_131_strides_0, weight = decoder_decoder_2_block_2_conv2_conv_weight_to_fp16_palettized, x = x_129_cast_fp16)[name = string("x_131_cast_fp16")]; + tensor x_133_cast_fp16 = add(x = x_131_cast_fp16, y = x_123_cast_fp16)[name = string("x_133_cast_fp16")]; + tensor alpha_21_to_fp16 = const()[name = string("alpha_21_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110010496)))]; + tensor var_2618_cast_fp16 = mul(x = x_133_cast_fp16, y = alpha_21_to_fp16)[name = string("op_2618_cast_fp16")]; + tensor sin_val_21_cast_fp16 = sin(x = var_2618_cast_fp16)[name = string("sin_val_21_cast_fp16")]; + tensor var_2625_cast_fp16 = mul(x = sin_val_21_cast_fp16, y = sin_val_21_cast_fp16)[name = string("op_2625_cast_fp16")]; + tensor var_2622_to_fp16 = const()[name = string("op_2622_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110011328)))]; + tensor var_2626_cast_fp16 = mul(x = var_2622_to_fp16, y = var_2625_cast_fp16)[name = string("op_2626_cast_fp16")]; + tensor x_135_cast_fp16 = add(x = x_133_cast_fp16, y = var_2626_cast_fp16)[name = string("x_135_cast_fp16")]; + tensor input_161_pad_0 = const()[name = string("input_161_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 18, 0])]; + string input_161_mode_0 = const()[name = string("input_161_mode_0"), val = string("constant")]; + fp16 const_100_to_fp16 = const()[name = string("const_100_to_fp16"), val = fp16(0x0p+0)]; + tensor input_161_cast_fp16 = pad(constant_val = const_100_to_fp16, mode = input_161_mode_0, pad = input_161_pad_0, x = x_135_cast_fp16)[name = string("input_161_cast_fp16")]; + string x_137_pad_type_0 = const()[name = string("x_137_pad_type_0"), val = string("valid")]; + tensor x_137_dilations_0 = const()[name = string("x_137_dilations_0"), val = tensor([1, 3])]; + tensor x_137_strides_0 = const()[name = string("x_137_strides_0"), val = tensor([1, 1])]; + tensor x_137_pad_0 = const()[name = string("x_137_pad_0"), val = tensor([0, 0, 0, 0])]; + int32 x_137_groups_0 = const()[name = string("x_137_groups_0"), val = int32(1)]; + tensor decoder_decoder_2_block_3_conv1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110012160))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111044416))))[name = string("decoder_decoder_2_block_3_conv1_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_2_block_3_conv1_conv_bias_to_fp16 = const()[name = string("decoder_decoder_2_block_3_conv1_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111044992)))]; + tensor x_137_cast_fp16 = conv(bias = decoder_decoder_2_block_3_conv1_conv_bias_to_fp16, dilations = x_137_dilations_0, groups = x_137_groups_0, pad = x_137_pad_0, pad_type = x_137_pad_type_0, strides = x_137_strides_0, weight = decoder_decoder_2_block_3_conv1_conv_weight_to_fp16_palettized, x = input_161_cast_fp16)[name = string("x_137_cast_fp16")]; + tensor alpha_23_to_fp16 = const()[name = string("alpha_23_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111045824)))]; + tensor var_2646_cast_fp16 = mul(x = x_137_cast_fp16, y = alpha_23_to_fp16)[name = string("op_2646_cast_fp16")]; + tensor sin_val_23_cast_fp16 = sin(x = var_2646_cast_fp16)[name = string("sin_val_23_cast_fp16")]; + tensor var_2653_cast_fp16 = mul(x = sin_val_23_cast_fp16, y = sin_val_23_cast_fp16)[name = string("op_2653_cast_fp16")]; + tensor var_2650_to_fp16 = const()[name = string("op_2650_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111046656)))]; + tensor var_2654_cast_fp16 = mul(x = var_2650_to_fp16, y = var_2653_cast_fp16)[name = string("op_2654_cast_fp16")]; + tensor x_139_cast_fp16 = add(x = x_137_cast_fp16, y = var_2654_cast_fp16)[name = string("x_139_cast_fp16")]; + string x_141_pad_type_0 = const()[name = string("x_141_pad_type_0"), val = string("valid")]; + tensor x_141_strides_0 = const()[name = string("x_141_strides_0"), val = tensor([1, 1])]; + tensor x_141_pad_0 = const()[name = string("x_141_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_141_dilations_0 = const()[name = string("x_141_dilations_0"), val = tensor([1, 1])]; + int32 x_141_groups_0 = const()[name = string("x_141_groups_0"), val = int32(1)]; + tensor decoder_decoder_2_block_3_conv2_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111047488))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111195008))))[name = string("decoder_decoder_2_block_3_conv2_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_2_block_3_conv2_conv_bias_to_fp16 = const()[name = string("decoder_decoder_2_block_3_conv2_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111195584)))]; + tensor x_141_cast_fp16 = conv(bias = decoder_decoder_2_block_3_conv2_conv_bias_to_fp16, dilations = x_141_dilations_0, groups = x_141_groups_0, pad = x_141_pad_0, pad_type = x_141_pad_type_0, strides = x_141_strides_0, weight = decoder_decoder_2_block_3_conv2_conv_weight_to_fp16_palettized, x = x_139_cast_fp16)[name = string("x_141_cast_fp16")]; + tensor x_143_cast_fp16 = add(x = x_141_cast_fp16, y = x_133_cast_fp16)[name = string("x_143_cast_fp16")]; + tensor alpha_25_to_fp16 = const()[name = string("alpha_25_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111196416)))]; + tensor var_2679_cast_fp16 = mul(x = x_143_cast_fp16, y = alpha_25_to_fp16)[name = string("op_2679_cast_fp16")]; + tensor sin_val_25_cast_fp16 = sin(x = var_2679_cast_fp16)[name = string("sin_val_25_cast_fp16")]; + tensor var_2686_cast_fp16 = mul(x = sin_val_25_cast_fp16, y = sin_val_25_cast_fp16)[name = string("op_2686_cast_fp16")]; + tensor var_2683_to_fp16 = const()[name = string("op_2683_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111197248)))]; + tensor var_2687_cast_fp16 = mul(x = var_2683_to_fp16, y = var_2686_cast_fp16)[name = string("op_2687_cast_fp16")]; + tensor x_145_cast_fp16 = add(x = x_143_cast_fp16, y = var_2687_cast_fp16)[name = string("x_145_cast_fp16")]; + tensor input_165_pad_0 = const()[name = string("input_165_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 54, 0])]; + string input_165_mode_0 = const()[name = string("input_165_mode_0"), val = string("constant")]; + fp16 const_102_to_fp16 = const()[name = string("const_102_to_fp16"), val = fp16(0x0p+0)]; + tensor input_165_cast_fp16 = pad(constant_val = const_102_to_fp16, mode = input_165_mode_0, pad = input_165_pad_0, x = x_145_cast_fp16)[name = string("input_165_cast_fp16")]; + string x_147_pad_type_0 = const()[name = string("x_147_pad_type_0"), val = string("valid")]; + tensor x_147_dilations_0 = const()[name = string("x_147_dilations_0"), val = tensor([1, 9])]; + tensor x_147_strides_0 = const()[name = string("x_147_strides_0"), val = tensor([1, 1])]; + tensor x_147_pad_0 = const()[name = string("x_147_pad_0"), val = tensor([0, 0, 0, 0])]; + int32 x_147_groups_0 = const()[name = string("x_147_groups_0"), val = int32(1)]; + tensor decoder_decoder_2_block_4_conv1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111198080))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112230336))))[name = string("decoder_decoder_2_block_4_conv1_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_2_block_4_conv1_conv_bias_to_fp16 = const()[name = string("decoder_decoder_2_block_4_conv1_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112230912)))]; + tensor x_147_cast_fp16 = conv(bias = decoder_decoder_2_block_4_conv1_conv_bias_to_fp16, dilations = x_147_dilations_0, groups = x_147_groups_0, pad = x_147_pad_0, pad_type = x_147_pad_type_0, strides = x_147_strides_0, weight = decoder_decoder_2_block_4_conv1_conv_weight_to_fp16_palettized, x = input_165_cast_fp16)[name = string("x_147_cast_fp16")]; + tensor alpha_27_to_fp16 = const()[name = string("alpha_27_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112231744)))]; + tensor var_2707_cast_fp16 = mul(x = x_147_cast_fp16, y = alpha_27_to_fp16)[name = string("op_2707_cast_fp16")]; + tensor sin_val_27_cast_fp16 = sin(x = var_2707_cast_fp16)[name = string("sin_val_27_cast_fp16")]; + tensor var_2714_cast_fp16 = mul(x = sin_val_27_cast_fp16, y = sin_val_27_cast_fp16)[name = string("op_2714_cast_fp16")]; + tensor var_2711_to_fp16 = const()[name = string("op_2711_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112232576)))]; + tensor var_2715_cast_fp16 = mul(x = var_2711_to_fp16, y = var_2714_cast_fp16)[name = string("op_2715_cast_fp16")]; + tensor x_149_cast_fp16 = add(x = x_147_cast_fp16, y = var_2715_cast_fp16)[name = string("x_149_cast_fp16")]; + string x_151_pad_type_0 = const()[name = string("x_151_pad_type_0"), val = string("valid")]; + tensor x_151_strides_0 = const()[name = string("x_151_strides_0"), val = tensor([1, 1])]; + tensor x_151_pad_0 = const()[name = string("x_151_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_151_dilations_0 = const()[name = string("x_151_dilations_0"), val = tensor([1, 1])]; + int32 x_151_groups_0 = const()[name = string("x_151_groups_0"), val = int32(1)]; + tensor decoder_decoder_2_block_4_conv2_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112233408))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112380928))))[name = string("decoder_decoder_2_block_4_conv2_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_2_block_4_conv2_conv_bias_to_fp16 = const()[name = string("decoder_decoder_2_block_4_conv2_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112381504)))]; + tensor x_151_cast_fp16 = conv(bias = decoder_decoder_2_block_4_conv2_conv_bias_to_fp16, dilations = x_151_dilations_0, groups = x_151_groups_0, pad = x_151_pad_0, pad_type = x_151_pad_type_0, strides = x_151_strides_0, weight = decoder_decoder_2_block_4_conv2_conv_weight_to_fp16_palettized, x = x_149_cast_fp16)[name = string("x_151_cast_fp16")]; + tensor x_153_cast_fp16 = add(x = x_151_cast_fp16, y = x_143_cast_fp16)[name = string("x_153_cast_fp16")]; + tensor alpha_29_to_fp16 = const()[name = string("alpha_29_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112382336)))]; + tensor var_2746_cast_fp16 = mul(x = x_153_cast_fp16, y = alpha_29_to_fp16)[name = string("op_2746_cast_fp16")]; + tensor sin_val_29_cast_fp16 = sin(x = var_2746_cast_fp16)[name = string("sin_val_29_cast_fp16")]; + tensor var_2753_cast_fp16 = mul(x = sin_val_29_cast_fp16, y = sin_val_29_cast_fp16)[name = string("op_2753_cast_fp16")]; + tensor var_2750_to_fp16 = const()[name = string("op_2750_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112383168)))]; + tensor var_2754_cast_fp16 = mul(x = var_2750_to_fp16, y = var_2753_cast_fp16)[name = string("op_2754_cast_fp16")]; + tensor x_155_cast_fp16 = add(x = x_153_cast_fp16, y = var_2754_cast_fp16)[name = string("x_155_cast_fp16")]; + string x_157_pad_type_0 = const()[name = string("x_157_pad_type_0"), val = string("valid")]; + tensor x_157_strides_0 = const()[name = string("x_157_strides_0"), val = tensor([1, 4])]; + tensor x_157_pad_0 = const()[name = string("x_157_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_157_dilations_0 = const()[name = string("x_157_dilations_0"), val = tensor([1, 1])]; + int32 x_157_groups_0 = const()[name = string("x_157_groups_0"), val = int32(1)]; + tensor x_157_has_output_shape_output_shape_0 = const()[name = string("x_157_has_output_shape_output_shape_0"), val = tensor([1, 192, 1, 3024])]; + tensor decoder_decoder_3_block_1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112384000))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112973888))))[name = string("decoder_decoder_3_block_1_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_3_block_1_conv_bias_to_fp16 = const()[name = string("decoder_decoder_3_block_1_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112974464)))]; + tensor x_157_has_output_shape_cast_fp16 = conv_transpose(bias = decoder_decoder_3_block_1_conv_bias_to_fp16, dilations = x_157_dilations_0, groups = x_157_groups_0, output_shape = x_157_has_output_shape_output_shape_0, pad = x_157_pad_0, pad_type = x_157_pad_type_0, strides = x_157_strides_0, weight = decoder_decoder_3_block_1_conv_weight_to_fp16_palettized, x = x_155_cast_fp16)[name = string("x_157_has_output_shape_cast_fp16")]; + tensor x_159_begin_0 = const()[name = string("x_159_begin_0"), val = tensor([0, 0, 0, 4])]; + tensor x_159_end_0 = const()[name = string("x_159_end_0"), val = tensor([1, 192, 1, 3020])]; + tensor x_159_end_mask_0 = const()[name = string("x_159_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_159_cast_fp16 = slice_by_index(begin = x_159_begin_0, end = x_159_end_0, end_mask = x_159_end_mask_0, x = x_157_has_output_shape_cast_fp16)[name = string("x_159_cast_fp16")]; + tensor alpha_31_to_fp16 = const()[name = string("alpha_31_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112974912)))]; + tensor var_2786_cast_fp16 = mul(x = x_159_cast_fp16, y = alpha_31_to_fp16)[name = string("op_2786_cast_fp16")]; + tensor sin_val_31_cast_fp16 = sin(x = var_2786_cast_fp16)[name = string("sin_val_31_cast_fp16")]; + tensor var_2793_cast_fp16 = mul(x = sin_val_31_cast_fp16, y = sin_val_31_cast_fp16)[name = string("op_2793_cast_fp16")]; + tensor var_2790_to_fp16 = const()[name = string("op_2790_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112975360)))]; + tensor var_2794_cast_fp16 = mul(x = var_2790_to_fp16, y = var_2793_cast_fp16)[name = string("op_2794_cast_fp16")]; + tensor x_161_cast_fp16 = add(x = x_159_cast_fp16, y = var_2794_cast_fp16)[name = string("x_161_cast_fp16")]; + tensor input_169_pad_0 = const()[name = string("input_169_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 6, 0])]; + string input_169_mode_0 = const()[name = string("input_169_mode_0"), val = string("constant")]; + fp16 const_105_to_fp16 = const()[name = string("const_105_to_fp16"), val = fp16(0x0p+0)]; + tensor input_169_cast_fp16 = pad(constant_val = const_105_to_fp16, mode = input_169_mode_0, pad = input_169_pad_0, x = x_161_cast_fp16)[name = string("input_169_cast_fp16")]; + string x_163_pad_type_0 = const()[name = string("x_163_pad_type_0"), val = string("valid")]; + tensor x_163_strides_0 = const()[name = string("x_163_strides_0"), val = tensor([1, 1])]; + tensor x_163_pad_0 = const()[name = string("x_163_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_163_dilations_0 = const()[name = string("x_163_dilations_0"), val = tensor([1, 1])]; + int32 x_163_groups_0 = const()[name = string("x_163_groups_0"), val = int32(1)]; + tensor decoder_decoder_3_block_2_conv1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112975808))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113233920))))[name = string("decoder_decoder_3_block_2_conv1_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_3_block_2_conv1_conv_bias_to_fp16 = const()[name = string("decoder_decoder_3_block_2_conv1_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113234496)))]; + tensor x_163_cast_fp16 = conv(bias = decoder_decoder_3_block_2_conv1_conv_bias_to_fp16, dilations = x_163_dilations_0, groups = x_163_groups_0, pad = x_163_pad_0, pad_type = x_163_pad_type_0, strides = x_163_strides_0, weight = decoder_decoder_3_block_2_conv1_conv_weight_to_fp16_palettized, x = input_169_cast_fp16)[name = string("x_163_cast_fp16")]; + tensor alpha_33_to_fp16 = const()[name = string("alpha_33_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113234944)))]; + tensor var_2814_cast_fp16 = mul(x = x_163_cast_fp16, y = alpha_33_to_fp16)[name = string("op_2814_cast_fp16")]; + tensor sin_val_33_cast_fp16 = sin(x = var_2814_cast_fp16)[name = string("sin_val_33_cast_fp16")]; + tensor var_2821_cast_fp16 = mul(x = sin_val_33_cast_fp16, y = sin_val_33_cast_fp16)[name = string("op_2821_cast_fp16")]; + tensor var_2818_to_fp16 = const()[name = string("op_2818_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113235392)))]; + tensor var_2822_cast_fp16 = mul(x = var_2818_to_fp16, y = var_2821_cast_fp16)[name = string("op_2822_cast_fp16")]; + tensor x_165_cast_fp16 = add(x = x_163_cast_fp16, y = var_2822_cast_fp16)[name = string("x_165_cast_fp16")]; + string x_167_pad_type_0 = const()[name = string("x_167_pad_type_0"), val = string("valid")]; + tensor x_167_strides_0 = const()[name = string("x_167_strides_0"), val = tensor([1, 1])]; + tensor x_167_pad_0 = const()[name = string("x_167_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_167_dilations_0 = const()[name = string("x_167_dilations_0"), val = tensor([1, 1])]; + int32 x_167_groups_0 = const()[name = string("x_167_groups_0"), val = int32(1)]; + tensor decoder_decoder_3_block_2_conv2_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113235840))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113272768))))[name = string("decoder_decoder_3_block_2_conv2_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_3_block_2_conv2_conv_bias_to_fp16 = const()[name = string("decoder_decoder_3_block_2_conv2_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113273344)))]; + tensor x_167_cast_fp16 = conv(bias = decoder_decoder_3_block_2_conv2_conv_bias_to_fp16, dilations = x_167_dilations_0, groups = x_167_groups_0, pad = x_167_pad_0, pad_type = x_167_pad_type_0, strides = x_167_strides_0, weight = decoder_decoder_3_block_2_conv2_conv_weight_to_fp16_palettized, x = x_165_cast_fp16)[name = string("x_167_cast_fp16")]; + tensor x_169_cast_fp16 = add(x = x_167_cast_fp16, y = x_159_cast_fp16)[name = string("x_169_cast_fp16")]; + tensor alpha_35_to_fp16 = const()[name = string("alpha_35_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113273792)))]; + tensor var_2847_cast_fp16 = mul(x = x_169_cast_fp16, y = alpha_35_to_fp16)[name = string("op_2847_cast_fp16")]; + tensor sin_val_35_cast_fp16 = sin(x = var_2847_cast_fp16)[name = string("sin_val_35_cast_fp16")]; + tensor var_2854_cast_fp16 = mul(x = sin_val_35_cast_fp16, y = sin_val_35_cast_fp16)[name = string("op_2854_cast_fp16")]; + tensor var_2851_to_fp16 = const()[name = string("op_2851_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113274240)))]; + tensor var_2855_cast_fp16 = mul(x = var_2851_to_fp16, y = var_2854_cast_fp16)[name = string("op_2855_cast_fp16")]; + tensor x_171_cast_fp16 = add(x = x_169_cast_fp16, y = var_2855_cast_fp16)[name = string("x_171_cast_fp16")]; + tensor input_173_pad_0 = const()[name = string("input_173_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 18, 0])]; + string input_173_mode_0 = const()[name = string("input_173_mode_0"), val = string("constant")]; + fp16 const_107_to_fp16 = const()[name = string("const_107_to_fp16"), val = fp16(0x0p+0)]; + tensor input_173_cast_fp16 = pad(constant_val = const_107_to_fp16, mode = input_173_mode_0, pad = input_173_pad_0, x = x_171_cast_fp16)[name = string("input_173_cast_fp16")]; + string x_173_pad_type_0 = const()[name = string("x_173_pad_type_0"), val = string("valid")]; + tensor x_173_dilations_0 = const()[name = string("x_173_dilations_0"), val = tensor([1, 3])]; + tensor x_173_strides_0 = const()[name = string("x_173_strides_0"), val = tensor([1, 1])]; + tensor x_173_pad_0 = const()[name = string("x_173_pad_0"), val = tensor([0, 0, 0, 0])]; + int32 x_173_groups_0 = const()[name = string("x_173_groups_0"), val = int32(1)]; + tensor decoder_decoder_3_block_3_conv1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113274688))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113532800))))[name = string("decoder_decoder_3_block_3_conv1_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_3_block_3_conv1_conv_bias_to_fp16 = const()[name = string("decoder_decoder_3_block_3_conv1_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113533376)))]; + tensor x_173_cast_fp16 = conv(bias = decoder_decoder_3_block_3_conv1_conv_bias_to_fp16, dilations = x_173_dilations_0, groups = x_173_groups_0, pad = x_173_pad_0, pad_type = x_173_pad_type_0, strides = x_173_strides_0, weight = decoder_decoder_3_block_3_conv1_conv_weight_to_fp16_palettized, x = input_173_cast_fp16)[name = string("x_173_cast_fp16")]; + tensor alpha_37_to_fp16 = const()[name = string("alpha_37_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113533824)))]; + tensor var_2875_cast_fp16 = mul(x = x_173_cast_fp16, y = alpha_37_to_fp16)[name = string("op_2875_cast_fp16")]; + tensor sin_val_37_cast_fp16 = sin(x = var_2875_cast_fp16)[name = string("sin_val_37_cast_fp16")]; + tensor var_2882_cast_fp16 = mul(x = sin_val_37_cast_fp16, y = sin_val_37_cast_fp16)[name = string("op_2882_cast_fp16")]; + tensor var_2879_to_fp16 = const()[name = string("op_2879_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113534272)))]; + tensor var_2883_cast_fp16 = mul(x = var_2879_to_fp16, y = var_2882_cast_fp16)[name = string("op_2883_cast_fp16")]; + tensor x_175_cast_fp16 = add(x = x_173_cast_fp16, y = var_2883_cast_fp16)[name = string("x_175_cast_fp16")]; + string x_177_pad_type_0 = const()[name = string("x_177_pad_type_0"), val = string("valid")]; + tensor x_177_strides_0 = const()[name = string("x_177_strides_0"), val = tensor([1, 1])]; + tensor x_177_pad_0 = const()[name = string("x_177_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_177_dilations_0 = const()[name = string("x_177_dilations_0"), val = tensor([1, 1])]; + int32 x_177_groups_0 = const()[name = string("x_177_groups_0"), val = int32(1)]; + tensor decoder_decoder_3_block_3_conv2_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113534720))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113571648))))[name = string("decoder_decoder_3_block_3_conv2_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_3_block_3_conv2_conv_bias_to_fp16 = const()[name = string("decoder_decoder_3_block_3_conv2_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113572224)))]; + tensor x_177_cast_fp16 = conv(bias = decoder_decoder_3_block_3_conv2_conv_bias_to_fp16, dilations = x_177_dilations_0, groups = x_177_groups_0, pad = x_177_pad_0, pad_type = x_177_pad_type_0, strides = x_177_strides_0, weight = decoder_decoder_3_block_3_conv2_conv_weight_to_fp16_palettized, x = x_175_cast_fp16)[name = string("x_177_cast_fp16")]; + tensor x_179_cast_fp16 = add(x = x_177_cast_fp16, y = x_169_cast_fp16)[name = string("x_179_cast_fp16")]; + tensor alpha_39_to_fp16 = const()[name = string("alpha_39_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113572672)))]; + tensor var_2908_cast_fp16 = mul(x = x_179_cast_fp16, y = alpha_39_to_fp16)[name = string("op_2908_cast_fp16")]; + tensor sin_val_39_cast_fp16 = sin(x = var_2908_cast_fp16)[name = string("sin_val_39_cast_fp16")]; + tensor var_2915_cast_fp16 = mul(x = sin_val_39_cast_fp16, y = sin_val_39_cast_fp16)[name = string("op_2915_cast_fp16")]; + tensor var_2912_to_fp16 = const()[name = string("op_2912_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113573120)))]; + tensor var_2916_cast_fp16 = mul(x = var_2912_to_fp16, y = var_2915_cast_fp16)[name = string("op_2916_cast_fp16")]; + tensor x_181_cast_fp16 = add(x = x_179_cast_fp16, y = var_2916_cast_fp16)[name = string("x_181_cast_fp16")]; + tensor input_177_pad_0 = const()[name = string("input_177_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 54, 0])]; + string input_177_mode_0 = const()[name = string("input_177_mode_0"), val = string("constant")]; + fp16 const_109_to_fp16 = const()[name = string("const_109_to_fp16"), val = fp16(0x0p+0)]; + tensor input_177_cast_fp16 = pad(constant_val = const_109_to_fp16, mode = input_177_mode_0, pad = input_177_pad_0, x = x_181_cast_fp16)[name = string("input_177_cast_fp16")]; + string x_183_pad_type_0 = const()[name = string("x_183_pad_type_0"), val = string("valid")]; + tensor x_183_dilations_0 = const()[name = string("x_183_dilations_0"), val = tensor([1, 9])]; + tensor x_183_strides_0 = const()[name = string("x_183_strides_0"), val = tensor([1, 1])]; + tensor x_183_pad_0 = const()[name = string("x_183_pad_0"), val = tensor([0, 0, 0, 0])]; + int32 x_183_groups_0 = const()[name = string("x_183_groups_0"), val = int32(1)]; + tensor decoder_decoder_3_block_4_conv1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113573568))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113831680))))[name = string("decoder_decoder_3_block_4_conv1_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_3_block_4_conv1_conv_bias_to_fp16 = const()[name = string("decoder_decoder_3_block_4_conv1_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113832256)))]; + tensor x_183_cast_fp16 = conv(bias = decoder_decoder_3_block_4_conv1_conv_bias_to_fp16, dilations = x_183_dilations_0, groups = x_183_groups_0, pad = x_183_pad_0, pad_type = x_183_pad_type_0, strides = x_183_strides_0, weight = decoder_decoder_3_block_4_conv1_conv_weight_to_fp16_palettized, x = input_177_cast_fp16)[name = string("x_183_cast_fp16")]; + tensor alpha_41_to_fp16 = const()[name = string("alpha_41_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113832704)))]; + tensor var_2936_cast_fp16 = mul(x = x_183_cast_fp16, y = alpha_41_to_fp16)[name = string("op_2936_cast_fp16")]; + tensor sin_val_41_cast_fp16 = sin(x = var_2936_cast_fp16)[name = string("sin_val_41_cast_fp16")]; + tensor var_2943_cast_fp16 = mul(x = sin_val_41_cast_fp16, y = sin_val_41_cast_fp16)[name = string("op_2943_cast_fp16")]; + tensor var_2940_to_fp16 = const()[name = string("op_2940_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113833152)))]; + tensor var_2944_cast_fp16 = mul(x = var_2940_to_fp16, y = var_2943_cast_fp16)[name = string("op_2944_cast_fp16")]; + tensor x_185_cast_fp16 = add(x = x_183_cast_fp16, y = var_2944_cast_fp16)[name = string("x_185_cast_fp16")]; + string x_187_pad_type_0 = const()[name = string("x_187_pad_type_0"), val = string("valid")]; + tensor x_187_strides_0 = const()[name = string("x_187_strides_0"), val = tensor([1, 1])]; + tensor x_187_pad_0 = const()[name = string("x_187_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_187_dilations_0 = const()[name = string("x_187_dilations_0"), val = tensor([1, 1])]; + int32 x_187_groups_0 = const()[name = string("x_187_groups_0"), val = int32(1)]; + tensor decoder_decoder_3_block_4_conv2_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113833600))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113870528))))[name = string("decoder_decoder_3_block_4_conv2_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_3_block_4_conv2_conv_bias_to_fp16 = const()[name = string("decoder_decoder_3_block_4_conv2_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113871104)))]; + tensor x_187_cast_fp16 = conv(bias = decoder_decoder_3_block_4_conv2_conv_bias_to_fp16, dilations = x_187_dilations_0, groups = x_187_groups_0, pad = x_187_pad_0, pad_type = x_187_pad_type_0, strides = x_187_strides_0, weight = decoder_decoder_3_block_4_conv2_conv_weight_to_fp16_palettized, x = x_185_cast_fp16)[name = string("x_187_cast_fp16")]; + tensor x_189_cast_fp16 = add(x = x_187_cast_fp16, y = x_179_cast_fp16)[name = string("x_189_cast_fp16")]; + tensor alpha_43_to_fp16 = const()[name = string("alpha_43_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113871552)))]; + tensor var_2975_cast_fp16 = mul(x = x_189_cast_fp16, y = alpha_43_to_fp16)[name = string("op_2975_cast_fp16")]; + tensor sin_val_43_cast_fp16 = sin(x = var_2975_cast_fp16)[name = string("sin_val_43_cast_fp16")]; + tensor var_2982_cast_fp16 = mul(x = sin_val_43_cast_fp16, y = sin_val_43_cast_fp16)[name = string("op_2982_cast_fp16")]; + tensor var_2979_to_fp16 = const()[name = string("op_2979_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113872000)))]; + tensor var_2983_cast_fp16 = mul(x = var_2979_to_fp16, y = var_2982_cast_fp16)[name = string("op_2983_cast_fp16")]; + tensor x_191_cast_fp16 = add(x = x_189_cast_fp16, y = var_2983_cast_fp16)[name = string("x_191_cast_fp16")]; + string x_193_pad_type_0 = const()[name = string("x_193_pad_type_0"), val = string("valid")]; + tensor x_193_strides_0 = const()[name = string("x_193_strides_0"), val = tensor([1, 3])]; + tensor x_193_pad_0 = const()[name = string("x_193_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_193_dilations_0 = const()[name = string("x_193_dilations_0"), val = tensor([1, 1])]; + int32 x_193_groups_0 = const()[name = string("x_193_groups_0"), val = int32(1)]; + tensor x_193_has_output_shape_output_shape_0 = const()[name = string("x_193_has_output_shape_output_shape_0"), val = tensor([1, 96, 1, 9051])]; + tensor decoder_decoder_4_block_1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113872448))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113983104))))[name = string("decoder_decoder_4_block_1_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_4_block_1_conv_bias_to_fp16 = const()[name = string("decoder_decoder_4_block_1_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113983680)))]; + tensor x_193_has_output_shape_cast_fp16 = conv_transpose(bias = decoder_decoder_4_block_1_conv_bias_to_fp16, dilations = x_193_dilations_0, groups = x_193_groups_0, output_shape = x_193_has_output_shape_output_shape_0, pad = x_193_pad_0, pad_type = x_193_pad_type_0, strides = x_193_strides_0, weight = decoder_decoder_4_block_1_conv_weight_to_fp16_palettized, x = x_191_cast_fp16)[name = string("x_193_has_output_shape_cast_fp16")]; + tensor x_195_begin_0 = const()[name = string("x_195_begin_0"), val = tensor([0, 0, 0, 3])]; + tensor x_195_end_0 = const()[name = string("x_195_end_0"), val = tensor([1, 96, 1, 9048])]; + tensor x_195_end_mask_0 = const()[name = string("x_195_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_195_cast_fp16 = slice_by_index(begin = x_195_begin_0, end = x_195_end_0, end_mask = x_195_end_mask_0, x = x_193_has_output_shape_cast_fp16)[name = string("x_195_cast_fp16")]; + tensor alpha_45_to_fp16 = const()[name = string("alpha_45_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113983936)))]; + tensor var_3015_cast_fp16 = mul(x = x_195_cast_fp16, y = alpha_45_to_fp16)[name = string("op_3015_cast_fp16")]; + tensor sin_val_45_cast_fp16 = sin(x = var_3015_cast_fp16)[name = string("sin_val_45_cast_fp16")]; + tensor var_3022_cast_fp16 = mul(x = sin_val_45_cast_fp16, y = sin_val_45_cast_fp16)[name = string("op_3022_cast_fp16")]; + tensor var_3019_to_fp16 = const()[name = string("op_3019_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113984192)))]; + tensor var_3023_cast_fp16 = mul(x = var_3019_to_fp16, y = var_3022_cast_fp16)[name = string("op_3023_cast_fp16")]; + tensor x_197_cast_fp16 = add(x = x_195_cast_fp16, y = var_3023_cast_fp16)[name = string("x_197_cast_fp16")]; + tensor input_181_pad_0 = const()[name = string("input_181_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 6, 0])]; + string input_181_mode_0 = const()[name = string("input_181_mode_0"), val = string("constant")]; + fp16 const_112_to_fp16 = const()[name = string("const_112_to_fp16"), val = fp16(0x0p+0)]; + tensor input_181_cast_fp16 = pad(constant_val = const_112_to_fp16, mode = input_181_mode_0, pad = input_181_pad_0, x = x_197_cast_fp16)[name = string("input_181_cast_fp16")]; + string x_199_pad_type_0 = const()[name = string("x_199_pad_type_0"), val = string("valid")]; + tensor x_199_strides_0 = const()[name = string("x_199_strides_0"), val = tensor([1, 1])]; + tensor x_199_pad_0 = const()[name = string("x_199_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_199_dilations_0 = const()[name = string("x_199_dilations_0"), val = tensor([1, 1])]; + int32 x_199_groups_0 = const()[name = string("x_199_groups_0"), val = int32(1)]; + tensor decoder_decoder_4_block_2_conv1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113984448))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114049024))))[name = string("decoder_decoder_4_block_2_conv1_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_4_block_2_conv1_conv_bias_to_fp16 = const()[name = string("decoder_decoder_4_block_2_conv1_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114049600)))]; + tensor x_199_cast_fp16 = conv(bias = decoder_decoder_4_block_2_conv1_conv_bias_to_fp16, dilations = x_199_dilations_0, groups = x_199_groups_0, pad = x_199_pad_0, pad_type = x_199_pad_type_0, strides = x_199_strides_0, weight = decoder_decoder_4_block_2_conv1_conv_weight_to_fp16_palettized, x = input_181_cast_fp16)[name = string("x_199_cast_fp16")]; + tensor alpha_47_to_fp16 = const()[name = string("alpha_47_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114049856)))]; + tensor var_3043_cast_fp16 = mul(x = x_199_cast_fp16, y = alpha_47_to_fp16)[name = string("op_3043_cast_fp16")]; + tensor sin_val_47_cast_fp16 = sin(x = var_3043_cast_fp16)[name = string("sin_val_47_cast_fp16")]; + tensor var_3050_cast_fp16 = mul(x = sin_val_47_cast_fp16, y = sin_val_47_cast_fp16)[name = string("op_3050_cast_fp16")]; + tensor var_3047_to_fp16 = const()[name = string("op_3047_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114050112)))]; + tensor var_3051_cast_fp16 = mul(x = var_3047_to_fp16, y = var_3050_cast_fp16)[name = string("op_3051_cast_fp16")]; + tensor x_201_cast_fp16 = add(x = x_199_cast_fp16, y = var_3051_cast_fp16)[name = string("x_201_cast_fp16")]; + string x_203_pad_type_0 = const()[name = string("x_203_pad_type_0"), val = string("valid")]; + tensor x_203_strides_0 = const()[name = string("x_203_strides_0"), val = tensor([1, 1])]; + tensor x_203_pad_0 = const()[name = string("x_203_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_203_dilations_0 = const()[name = string("x_203_dilations_0"), val = tensor([1, 1])]; + int32 x_203_groups_0 = const()[name = string("x_203_groups_0"), val = int32(1)]; + tensor decoder_decoder_4_block_2_conv2_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114050368))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114059648))))[name = string("decoder_decoder_4_block_2_conv2_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_4_block_2_conv2_conv_bias_to_fp16 = const()[name = string("decoder_decoder_4_block_2_conv2_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114060224)))]; + tensor x_203_cast_fp16 = conv(bias = decoder_decoder_4_block_2_conv2_conv_bias_to_fp16, dilations = x_203_dilations_0, groups = x_203_groups_0, pad = x_203_pad_0, pad_type = x_203_pad_type_0, strides = x_203_strides_0, weight = decoder_decoder_4_block_2_conv2_conv_weight_to_fp16_palettized, x = x_201_cast_fp16)[name = string("x_203_cast_fp16")]; + tensor x_205_cast_fp16 = add(x = x_203_cast_fp16, y = x_195_cast_fp16)[name = string("x_205_cast_fp16")]; + tensor alpha_49_to_fp16 = const()[name = string("alpha_49_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114060480)))]; + tensor var_3076_cast_fp16 = mul(x = x_205_cast_fp16, y = alpha_49_to_fp16)[name = string("op_3076_cast_fp16")]; + tensor sin_val_49_cast_fp16 = sin(x = var_3076_cast_fp16)[name = string("sin_val_49_cast_fp16")]; + tensor var_3083_cast_fp16 = mul(x = sin_val_49_cast_fp16, y = sin_val_49_cast_fp16)[name = string("op_3083_cast_fp16")]; + tensor var_3080_to_fp16 = const()[name = string("op_3080_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114060736)))]; + tensor var_3084_cast_fp16 = mul(x = var_3080_to_fp16, y = var_3083_cast_fp16)[name = string("op_3084_cast_fp16")]; + tensor x_207_cast_fp16 = add(x = x_205_cast_fp16, y = var_3084_cast_fp16)[name = string("x_207_cast_fp16")]; + tensor input_185_pad_0 = const()[name = string("input_185_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 18, 0])]; + string input_185_mode_0 = const()[name = string("input_185_mode_0"), val = string("constant")]; + fp16 const_114_to_fp16 = const()[name = string("const_114_to_fp16"), val = fp16(0x0p+0)]; + tensor input_185_cast_fp16 = pad(constant_val = const_114_to_fp16, mode = input_185_mode_0, pad = input_185_pad_0, x = x_207_cast_fp16)[name = string("input_185_cast_fp16")]; + string x_209_pad_type_0 = const()[name = string("x_209_pad_type_0"), val = string("valid")]; + tensor x_209_dilations_0 = const()[name = string("x_209_dilations_0"), val = tensor([1, 3])]; + tensor x_209_strides_0 = const()[name = string("x_209_strides_0"), val = tensor([1, 1])]; + tensor x_209_pad_0 = const()[name = string("x_209_pad_0"), val = tensor([0, 0, 0, 0])]; + int32 x_209_groups_0 = const()[name = string("x_209_groups_0"), val = int32(1)]; + tensor decoder_decoder_4_block_3_conv1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114060992))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114125568))))[name = string("decoder_decoder_4_block_3_conv1_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_4_block_3_conv1_conv_bias_to_fp16 = const()[name = string("decoder_decoder_4_block_3_conv1_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114126144)))]; + tensor x_209_cast_fp16 = conv(bias = decoder_decoder_4_block_3_conv1_conv_bias_to_fp16, dilations = x_209_dilations_0, groups = x_209_groups_0, pad = x_209_pad_0, pad_type = x_209_pad_type_0, strides = x_209_strides_0, weight = decoder_decoder_4_block_3_conv1_conv_weight_to_fp16_palettized, x = input_185_cast_fp16)[name = string("x_209_cast_fp16")]; + tensor alpha_51_to_fp16 = const()[name = string("alpha_51_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114126400)))]; + tensor var_3104_cast_fp16 = mul(x = x_209_cast_fp16, y = alpha_51_to_fp16)[name = string("op_3104_cast_fp16")]; + tensor sin_val_51_cast_fp16 = sin(x = var_3104_cast_fp16)[name = string("sin_val_51_cast_fp16")]; + tensor var_3111_cast_fp16 = mul(x = sin_val_51_cast_fp16, y = sin_val_51_cast_fp16)[name = string("op_3111_cast_fp16")]; + tensor var_3108_to_fp16 = const()[name = string("op_3108_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114126656)))]; + tensor var_3112_cast_fp16 = mul(x = var_3108_to_fp16, y = var_3111_cast_fp16)[name = string("op_3112_cast_fp16")]; + tensor x_211_cast_fp16 = add(x = x_209_cast_fp16, y = var_3112_cast_fp16)[name = string("x_211_cast_fp16")]; + string x_213_pad_type_0 = const()[name = string("x_213_pad_type_0"), val = string("valid")]; + tensor x_213_strides_0 = const()[name = string("x_213_strides_0"), val = tensor([1, 1])]; + tensor x_213_pad_0 = const()[name = string("x_213_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_213_dilations_0 = const()[name = string("x_213_dilations_0"), val = tensor([1, 1])]; + int32 x_213_groups_0 = const()[name = string("x_213_groups_0"), val = int32(1)]; + tensor decoder_decoder_4_block_3_conv2_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114126912))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114136192))))[name = string("decoder_decoder_4_block_3_conv2_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_4_block_3_conv2_conv_bias_to_fp16 = const()[name = string("decoder_decoder_4_block_3_conv2_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114136768)))]; + tensor x_213_cast_fp16 = conv(bias = decoder_decoder_4_block_3_conv2_conv_bias_to_fp16, dilations = x_213_dilations_0, groups = x_213_groups_0, pad = x_213_pad_0, pad_type = x_213_pad_type_0, strides = x_213_strides_0, weight = decoder_decoder_4_block_3_conv2_conv_weight_to_fp16_palettized, x = x_211_cast_fp16)[name = string("x_213_cast_fp16")]; + tensor x_215_cast_fp16 = add(x = x_213_cast_fp16, y = x_205_cast_fp16)[name = string("x_215_cast_fp16")]; + tensor alpha_53_to_fp16 = const()[name = string("alpha_53_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114137024)))]; + tensor var_3137_cast_fp16 = mul(x = x_215_cast_fp16, y = alpha_53_to_fp16)[name = string("op_3137_cast_fp16")]; + tensor sin_val_53_cast_fp16 = sin(x = var_3137_cast_fp16)[name = string("sin_val_53_cast_fp16")]; + tensor var_3144_cast_fp16 = mul(x = sin_val_53_cast_fp16, y = sin_val_53_cast_fp16)[name = string("op_3144_cast_fp16")]; + tensor var_3141_to_fp16 = const()[name = string("op_3141_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114137280)))]; + tensor var_3145_cast_fp16 = mul(x = var_3141_to_fp16, y = var_3144_cast_fp16)[name = string("op_3145_cast_fp16")]; + tensor x_217_cast_fp16 = add(x = x_215_cast_fp16, y = var_3145_cast_fp16)[name = string("x_217_cast_fp16")]; + tensor input_189_pad_0 = const()[name = string("input_189_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 54, 0])]; + string input_189_mode_0 = const()[name = string("input_189_mode_0"), val = string("constant")]; + fp16 const_116_to_fp16 = const()[name = string("const_116_to_fp16"), val = fp16(0x0p+0)]; + tensor input_189_cast_fp16 = pad(constant_val = const_116_to_fp16, mode = input_189_mode_0, pad = input_189_pad_0, x = x_217_cast_fp16)[name = string("input_189_cast_fp16")]; + string x_219_pad_type_0 = const()[name = string("x_219_pad_type_0"), val = string("valid")]; + tensor x_219_dilations_0 = const()[name = string("x_219_dilations_0"), val = tensor([1, 9])]; + tensor x_219_strides_0 = const()[name = string("x_219_strides_0"), val = tensor([1, 1])]; + tensor x_219_pad_0 = const()[name = string("x_219_pad_0"), val = tensor([0, 0, 0, 0])]; + int32 x_219_groups_0 = const()[name = string("x_219_groups_0"), val = int32(1)]; + tensor decoder_decoder_4_block_4_conv1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114137536))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114202112))))[name = string("decoder_decoder_4_block_4_conv1_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_4_block_4_conv1_conv_bias_to_fp16 = const()[name = string("decoder_decoder_4_block_4_conv1_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114202688)))]; + tensor x_219_cast_fp16 = conv(bias = decoder_decoder_4_block_4_conv1_conv_bias_to_fp16, dilations = x_219_dilations_0, groups = x_219_groups_0, pad = x_219_pad_0, pad_type = x_219_pad_type_0, strides = x_219_strides_0, weight = decoder_decoder_4_block_4_conv1_conv_weight_to_fp16_palettized, x = input_189_cast_fp16)[name = string("x_219_cast_fp16")]; + tensor alpha_55_to_fp16 = const()[name = string("alpha_55_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114202944)))]; + tensor var_3165_cast_fp16 = mul(x = x_219_cast_fp16, y = alpha_55_to_fp16)[name = string("op_3165_cast_fp16")]; + tensor sin_val_55_cast_fp16 = sin(x = var_3165_cast_fp16)[name = string("sin_val_55_cast_fp16")]; + tensor var_3172_cast_fp16 = mul(x = sin_val_55_cast_fp16, y = sin_val_55_cast_fp16)[name = string("op_3172_cast_fp16")]; + tensor var_3169_to_fp16 = const()[name = string("op_3169_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114203200)))]; + tensor var_3173_cast_fp16 = mul(x = var_3169_to_fp16, y = var_3172_cast_fp16)[name = string("op_3173_cast_fp16")]; + tensor x_221_cast_fp16 = add(x = x_219_cast_fp16, y = var_3173_cast_fp16)[name = string("x_221_cast_fp16")]; + string x_223_pad_type_0 = const()[name = string("x_223_pad_type_0"), val = string("valid")]; + tensor x_223_strides_0 = const()[name = string("x_223_strides_0"), val = tensor([1, 1])]; + tensor x_223_pad_0 = const()[name = string("x_223_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_223_dilations_0 = const()[name = string("x_223_dilations_0"), val = tensor([1, 1])]; + int32 x_223_groups_0 = const()[name = string("x_223_groups_0"), val = int32(1)]; + tensor decoder_decoder_4_block_4_conv2_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114203456))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114212736))))[name = string("decoder_decoder_4_block_4_conv2_conv_weight_to_fp16_palettized")]; + tensor decoder_decoder_4_block_4_conv2_conv_bias_to_fp16 = const()[name = string("decoder_decoder_4_block_4_conv2_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114213312)))]; + tensor x_223_cast_fp16 = conv(bias = decoder_decoder_4_block_4_conv2_conv_bias_to_fp16, dilations = x_223_dilations_0, groups = x_223_groups_0, pad = x_223_pad_0, pad_type = x_223_pad_type_0, strides = x_223_strides_0, weight = decoder_decoder_4_block_4_conv2_conv_weight_to_fp16_palettized, x = x_221_cast_fp16)[name = string("x_223_cast_fp16")]; + tensor x_225_cast_fp16 = add(x = x_223_cast_fp16, y = x_215_cast_fp16)[name = string("x_225_cast_fp16")]; + tensor alpha_57_to_fp16 = const()[name = string("alpha_57_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114213568)))]; + tensor var_3194_cast_fp16 = mul(x = x_225_cast_fp16, y = alpha_57_to_fp16)[name = string("op_3194_cast_fp16")]; + tensor sin_val_cast_fp16 = sin(x = var_3194_cast_fp16)[name = string("sin_val_cast_fp16")]; + tensor var_3201_cast_fp16 = mul(x = sin_val_cast_fp16, y = sin_val_cast_fp16)[name = string("op_3201_cast_fp16")]; + tensor var_3198_to_fp16 = const()[name = string("op_3198_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114213824)))]; + tensor var_3202_cast_fp16 = mul(x = var_3198_to_fp16, y = var_3201_cast_fp16)[name = string("op_3202_cast_fp16")]; + tensor x_cast_fp16 = add(x = x_225_cast_fp16, y = var_3202_cast_fp16)[name = string("x_cast_fp16")]; + tensor input_pad_0 = const()[name = string("input_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 6, 0])]; + string input_mode_0 = const()[name = string("input_mode_0"), val = string("constant")]; + fp16 const_118_to_fp16 = const()[name = string("const_118_to_fp16"), val = fp16(0x0p+0)]; + tensor input_cast_fp16 = pad(constant_val = const_118_to_fp16, mode = input_mode_0, pad = input_pad_0, x = x_cast_fp16)[name = string("input_cast_fp16")]; + string h_1_pad_type_0 = const()[name = string("h_1_pad_type_0"), val = string("valid")]; + tensor h_1_strides_0 = const()[name = string("h_1_strides_0"), val = tensor([1, 1])]; + tensor h_1_pad_0 = const()[name = string("h_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor h_1_dilations_0 = const()[name = string("h_1_dilations_0"), val = tensor([1, 1])]; + int32 h_1_groups_0 = const()[name = string("h_1_groups_0"), val = int32(1)]; + tensor decoder_decoder_6_conv_weight_to_fp16 = const()[name = string("decoder_decoder_6_conv_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114214080)))]; + tensor decoder_decoder_6_conv_bias_to_fp16 = const()[name = string("decoder_decoder_6_conv_bias_to_fp16"), val = tensor([-0x1.1p-19])]; + tensor h_1_cast_fp16 = conv(bias = decoder_decoder_6_conv_bias_to_fp16, dilations = h_1_dilations_0, groups = h_1_groups_0, pad = h_1_pad_0, pad_type = h_1_pad_type_0, strides = h_1_strides_0, weight = decoder_decoder_6_conv_weight_to_fp16, x = input_cast_fp16)[name = string("h_1_cast_fp16")]; + fp16 var_28_promoted_to_fp16 = const()[name = string("op_28_promoted_to_fp16"), val = fp16(-0x1p+0)]; + fp16 var_32_promoted_16_to_fp16 = const()[name = string("op_32_promoted_16_to_fp16"), val = fp16(0x1p+0)]; + tensor clip_16_cast_fp16 = clip(alpha = var_28_promoted_to_fp16, beta = var_32_promoted_16_to_fp16, x = h_1_cast_fp16)[name = string("clip_16_cast_fp16")]; + tensor var_3215_begin_0 = const()[name = string("op_3215_begin_0"), val = tensor([0, 0, 0, 7125])]; + tensor var_3215_end_0 = const()[name = string("op_3215_end_0"), val = tensor([1, 1, 1, 9045])]; + tensor var_3215_end_mask_0 = const()[name = string("op_3215_end_mask_0"), val = tensor([true, true, true, true])]; + tensor audio = slice_by_index(begin = var_3215_begin_0, end = var_3215_end_0, end_mask = var_3215_end_mask_0, x = clip_16_cast_fp16)[name = string("op_3215_cast_fp16")]; + } -> (audio, key_cache_updates, value_cache_updates, hidden_context_update); +} \ No newline at end of file