program(1.0)
[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3510.2.1"}, {"coremlc-version", "3500.32.1"}})]
{
    func main<ios17>(tensor<int32, [1, 256]> input_ids) {
            tensor<fp16, [262208, 2560]> wrapped_model_language_model_embed_tokens_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [335626240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(335626368))), name = tensor<string, []>("wrapped_model_language_model_embed_tokens_weight_palettized"), shape = tensor<uint32, [2]>([262208, 2560])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_0_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(335626496))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(338248000))), name = tensor<string, []>("wrapped_model_language_model_layers_0_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_0_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(338248128))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(339558912))), name = tensor<string, []>("wrapped_model_language_model_layers_0_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_0_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(339559040))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(340869824))), name = tensor<string, []>("wrapped_model_language_model_layers_0_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_0_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(340869952))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(343491456))), name = tensor<string, []>("wrapped_model_language_model_layers_0_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_0_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(343491584))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(356598848))), name = tensor<string, []>("wrapped_model_language_model_layers_0_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_0_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(356598976))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(369706240))), name = tensor<string, []>("wrapped_model_language_model_layers_0_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_0_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(369706368))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(382813632))), name = tensor<string, []>("wrapped_model_language_model_layers_0_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_1_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(382813760))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(385435264))), name = tensor<string, []>("wrapped_model_language_model_layers_1_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_1_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(385435392))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(386746176))), name = tensor<string, []>("wrapped_model_language_model_layers_1_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_1_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(386746304))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(388057088))), name = tensor<string, []>("wrapped_model_language_model_layers_1_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_1_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(388057216))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(390678720))), name = tensor<string, []>("wrapped_model_language_model_layers_1_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_1_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(390678848))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(403786112))), name = tensor<string, []>("wrapped_model_language_model_layers_1_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_1_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(403786240))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(416893504))), name = tensor<string, []>("wrapped_model_language_model_layers_1_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_1_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(416893632))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(430000896))), name = tensor<string, []>("wrapped_model_language_model_layers_1_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_2_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(430001024))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(432622528))), name = tensor<string, []>("wrapped_model_language_model_layers_2_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_2_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(432622656))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(433933440))), name = tensor<string, []>("wrapped_model_language_model_layers_2_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_2_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(433933568))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(435244352))), name = tensor<string, []>("wrapped_model_language_model_layers_2_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_2_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(435244480))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(437865984))), name = tensor<string, []>("wrapped_model_language_model_layers_2_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_2_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(437866112))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(450973376))), name = tensor<string, []>("wrapped_model_language_model_layers_2_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_2_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(450973504))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(464080768))), name = tensor<string, []>("wrapped_model_language_model_layers_2_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_2_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(464080896))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(477188160))), name = tensor<string, []>("wrapped_model_language_model_layers_2_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_3_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(477188288))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(479809792))), name = tensor<string, []>("wrapped_model_language_model_layers_3_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_3_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(479809920))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(481120704))), name = tensor<string, []>("wrapped_model_language_model_layers_3_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_3_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(481120832))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(482431616))), name = tensor<string, []>("wrapped_model_language_model_layers_3_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_3_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(482431744))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(485053248))), name = tensor<string, []>("wrapped_model_language_model_layers_3_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_3_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(485053376))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(498160640))), name = tensor<string, []>("wrapped_model_language_model_layers_3_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_3_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(498160768))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(511268032))), name = tensor<string, []>("wrapped_model_language_model_layers_3_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_3_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(511268160))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(524375424))), name = tensor<string, []>("wrapped_model_language_model_layers_3_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_4_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(524375552))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(526997056))), name = tensor<string, []>("wrapped_model_language_model_layers_4_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_4_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(526997184))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(528307968))), name = tensor<string, []>("wrapped_model_language_model_layers_4_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_4_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(528308096))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(529618880))), name = tensor<string, []>("wrapped_model_language_model_layers_4_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_4_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(529619008))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(532240512))), name = tensor<string, []>("wrapped_model_language_model_layers_4_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_4_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(532240640))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(545347904))), name = tensor<string, []>("wrapped_model_language_model_layers_4_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_4_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(545348032))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(558455296))), name = tensor<string, []>("wrapped_model_language_model_layers_4_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_4_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(558455424))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(571562688))), name = tensor<string, []>("wrapped_model_language_model_layers_4_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_5_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(571562816))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(574184320))), name = tensor<string, []>("wrapped_model_language_model_layers_5_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_5_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(574184448))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(575495232))), name = tensor<string, []>("wrapped_model_language_model_layers_5_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_5_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(575495360))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(576806144))), name = tensor<string, []>("wrapped_model_language_model_layers_5_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_5_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(576806272))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(579427776))), name = tensor<string, []>("wrapped_model_language_model_layers_5_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_5_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(579427904))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(592535168))), name = tensor<string, []>("wrapped_model_language_model_layers_5_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_5_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(592535296))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(605642560))), name = tensor<string, []>("wrapped_model_language_model_layers_5_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_5_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(605642688))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(618749952))), name = tensor<string, []>("wrapped_model_language_model_layers_5_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_6_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(618750080))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(621371584))), name = tensor<string, []>("wrapped_model_language_model_layers_6_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_6_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(621371712))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(622682496))), name = tensor<string, []>("wrapped_model_language_model_layers_6_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_6_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(622682624))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(623993408))), name = tensor<string, []>("wrapped_model_language_model_layers_6_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_6_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(623993536))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(626615040))), name = tensor<string, []>("wrapped_model_language_model_layers_6_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_6_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(626615168))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(639722432))), name = tensor<string, []>("wrapped_model_language_model_layers_6_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_6_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(639722560))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(652829824))), name = tensor<string, []>("wrapped_model_language_model_layers_6_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_6_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(652829952))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(665937216))), name = tensor<string, []>("wrapped_model_language_model_layers_6_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_7_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(665937344))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(668558848))), name = tensor<string, []>("wrapped_model_language_model_layers_7_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_7_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(668558976))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(669869760))), name = tensor<string, []>("wrapped_model_language_model_layers_7_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_7_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(669869888))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(671180672))), name = tensor<string, []>("wrapped_model_language_model_layers_7_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_7_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(671180800))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(673802304))), name = tensor<string, []>("wrapped_model_language_model_layers_7_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_7_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(673802432))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(686909696))), name = tensor<string, []>("wrapped_model_language_model_layers_7_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_7_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(686909824))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(700017088))), name = tensor<string, []>("wrapped_model_language_model_layers_7_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_7_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(700017216))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(713124480))), name = tensor<string, []>("wrapped_model_language_model_layers_7_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_8_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(713124608))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(715746112))), name = tensor<string, []>("wrapped_model_language_model_layers_8_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_8_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(715746240))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(717057024))), name = tensor<string, []>("wrapped_model_language_model_layers_8_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_8_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(717057152))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(718367936))), name = tensor<string, []>("wrapped_model_language_model_layers_8_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_8_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(718368064))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(720989568))), name = tensor<string, []>("wrapped_model_language_model_layers_8_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_8_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(720989696))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(734096960))), name = tensor<string, []>("wrapped_model_language_model_layers_8_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_8_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(734097088))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(747204352))), name = tensor<string, []>("wrapped_model_language_model_layers_8_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_8_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(747204480))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(760311744))), name = tensor<string, []>("wrapped_model_language_model_layers_8_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_9_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(760311872))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(762933376))), name = tensor<string, []>("wrapped_model_language_model_layers_9_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_9_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(762933504))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(764244288))), name = tensor<string, []>("wrapped_model_language_model_layers_9_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_9_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(764244416))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(765555200))), name = tensor<string, []>("wrapped_model_language_model_layers_9_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_9_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(765555328))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(768176832))), name = tensor<string, []>("wrapped_model_language_model_layers_9_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_9_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(768176960))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(781284224))), name = tensor<string, []>("wrapped_model_language_model_layers_9_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_9_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(781284352))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(794391616))), name = tensor<string, []>("wrapped_model_language_model_layers_9_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_9_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(794391744))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(807499008))), name = tensor<string, []>("wrapped_model_language_model_layers_9_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_10_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(807499136))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(810120640))), name = tensor<string, []>("wrapped_model_language_model_layers_10_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_10_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(810120768))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(811431552))), name = tensor<string, []>("wrapped_model_language_model_layers_10_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_10_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(811431680))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(812742464))), name = tensor<string, []>("wrapped_model_language_model_layers_10_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_10_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(812742592))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(815364096))), name = tensor<string, []>("wrapped_model_language_model_layers_10_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_10_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(815364224))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(828471488))), name = tensor<string, []>("wrapped_model_language_model_layers_10_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_10_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(828471616))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(841578880))), name = tensor<string, []>("wrapped_model_language_model_layers_10_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_10_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(841579008))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(854686272))), name = tensor<string, []>("wrapped_model_language_model_layers_10_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_11_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(854686400))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(857307904))), name = tensor<string, []>("wrapped_model_language_model_layers_11_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_11_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(857308032))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(858618816))), name = tensor<string, []>("wrapped_model_language_model_layers_11_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_11_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(858618944))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(859929728))), name = tensor<string, []>("wrapped_model_language_model_layers_11_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_11_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(859929856))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(862551360))), name = tensor<string, []>("wrapped_model_language_model_layers_11_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_11_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(862551488))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(875658752))), name = tensor<string, []>("wrapped_model_language_model_layers_11_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_11_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(875658880))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(888766144))), name = tensor<string, []>("wrapped_model_language_model_layers_11_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_11_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(888766272))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(901873536))), name = tensor<string, []>("wrapped_model_language_model_layers_11_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_12_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(901873664))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(904495168))), name = tensor<string, []>("wrapped_model_language_model_layers_12_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_12_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(904495296))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(905806080))), name = tensor<string, []>("wrapped_model_language_model_layers_12_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_12_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(905806208))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(907116992))), name = tensor<string, []>("wrapped_model_language_model_layers_12_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_12_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(907117120))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(909738624))), name = tensor<string, []>("wrapped_model_language_model_layers_12_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_12_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(909738752))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(922846016))), name = tensor<string, []>("wrapped_model_language_model_layers_12_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_12_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(922846144))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(935953408))), name = tensor<string, []>("wrapped_model_language_model_layers_12_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_12_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(935953536))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(949060800))), name = tensor<string, []>("wrapped_model_language_model_layers_12_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_13_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(949060928))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(951682432))), name = tensor<string, []>("wrapped_model_language_model_layers_13_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_13_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(951682560))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(952993344))), name = tensor<string, []>("wrapped_model_language_model_layers_13_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_13_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(952993472))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(954304256))), name = tensor<string, []>("wrapped_model_language_model_layers_13_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_13_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(954304384))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(956925888))), name = tensor<string, []>("wrapped_model_language_model_layers_13_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_13_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(956926016))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(970033280))), name = tensor<string, []>("wrapped_model_language_model_layers_13_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_13_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(970033408))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(983140672))), name = tensor<string, []>("wrapped_model_language_model_layers_13_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_13_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(983140800))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(996248064))), name = tensor<string, []>("wrapped_model_language_model_layers_13_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_14_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(996248192))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(998869696))), name = tensor<string, []>("wrapped_model_language_model_layers_14_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_14_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(998869824))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1000180608))), name = tensor<string, []>("wrapped_model_language_model_layers_14_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_14_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1000180736))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1001491520))), name = tensor<string, []>("wrapped_model_language_model_layers_14_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_14_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1001491648))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1004113152))), name = tensor<string, []>("wrapped_model_language_model_layers_14_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_14_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1004113280))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1017220544))), name = tensor<string, []>("wrapped_model_language_model_layers_14_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_14_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1017220672))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1030327936))), name = tensor<string, []>("wrapped_model_language_model_layers_14_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_14_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1030328064))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1043435328))), name = tensor<string, []>("wrapped_model_language_model_layers_14_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_15_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1043435456))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1046056960))), name = tensor<string, []>("wrapped_model_language_model_layers_15_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_15_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1046057088))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1047367872))), name = tensor<string, []>("wrapped_model_language_model_layers_15_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_15_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1047368000))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1048678784))), name = tensor<string, []>("wrapped_model_language_model_layers_15_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_15_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1048678912))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1051300416))), name = tensor<string, []>("wrapped_model_language_model_layers_15_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_15_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1051300544))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1064407808))), name = tensor<string, []>("wrapped_model_language_model_layers_15_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_15_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1064407936))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1077515200))), name = tensor<string, []>("wrapped_model_language_model_layers_15_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_15_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1077515328))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1090622592))), name = tensor<string, []>("wrapped_model_language_model_layers_15_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_16_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1090622720))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1093244224))), name = tensor<string, []>("wrapped_model_language_model_layers_16_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_16_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1093244352))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1094555136))), name = tensor<string, []>("wrapped_model_language_model_layers_16_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_16_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1094555264))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1095866048))), name = tensor<string, []>("wrapped_model_language_model_layers_16_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_16_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1095866176))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1098487680))), name = tensor<string, []>("wrapped_model_language_model_layers_16_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_16_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1098487808))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1111595072))), name = tensor<string, []>("wrapped_model_language_model_layers_16_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_16_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1111595200))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1124702464))), name = tensor<string, []>("wrapped_model_language_model_layers_16_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_16_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1124702592))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1137809856))), name = tensor<string, []>("wrapped_model_language_model_layers_16_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_17_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1137809984))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1140431488))), name = tensor<string, []>("wrapped_model_language_model_layers_17_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_17_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1140431616))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1141742400))), name = tensor<string, []>("wrapped_model_language_model_layers_17_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_17_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1141742528))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1143053312))), name = tensor<string, []>("wrapped_model_language_model_layers_17_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_17_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1143053440))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1145674944))), name = tensor<string, []>("wrapped_model_language_model_layers_17_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_17_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1145675072))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1158782336))), name = tensor<string, []>("wrapped_model_language_model_layers_17_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_17_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1158782464))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1171889728))), name = tensor<string, []>("wrapped_model_language_model_layers_17_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_17_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1171889856))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1184997120))), name = tensor<string, []>("wrapped_model_language_model_layers_17_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_18_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1184997248))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1187618752))), name = tensor<string, []>("wrapped_model_language_model_layers_18_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_18_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1187618880))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1188929664))), name = tensor<string, []>("wrapped_model_language_model_layers_18_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_18_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1188929792))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1190240576))), name = tensor<string, []>("wrapped_model_language_model_layers_18_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_18_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1190240704))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1192862208))), name = tensor<string, []>("wrapped_model_language_model_layers_18_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_18_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1192862336))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1205969600))), name = tensor<string, []>("wrapped_model_language_model_layers_18_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_18_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1205969728))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1219076992))), name = tensor<string, []>("wrapped_model_language_model_layers_18_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_18_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1219077120))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1232184384))), name = tensor<string, []>("wrapped_model_language_model_layers_18_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_19_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1232184512))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1234806016))), name = tensor<string, []>("wrapped_model_language_model_layers_19_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_19_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1234806144))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1236116928))), name = tensor<string, []>("wrapped_model_language_model_layers_19_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_19_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1236117056))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1237427840))), name = tensor<string, []>("wrapped_model_language_model_layers_19_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_19_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1237427968))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1240049472))), name = tensor<string, []>("wrapped_model_language_model_layers_19_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_19_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1240049600))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1253156864))), name = tensor<string, []>("wrapped_model_language_model_layers_19_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_19_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1253156992))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1266264256))), name = tensor<string, []>("wrapped_model_language_model_layers_19_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_19_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1266264384))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1279371648))), name = tensor<string, []>("wrapped_model_language_model_layers_19_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_20_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1279371776))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1281993280))), name = tensor<string, []>("wrapped_model_language_model_layers_20_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_20_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1281993408))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1283304192))), name = tensor<string, []>("wrapped_model_language_model_layers_20_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_20_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1283304320))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1284615104))), name = tensor<string, []>("wrapped_model_language_model_layers_20_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_20_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1284615232))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1287236736))), name = tensor<string, []>("wrapped_model_language_model_layers_20_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_20_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1287236864))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1300344128))), name = tensor<string, []>("wrapped_model_language_model_layers_20_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_20_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1300344256))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1313451520))), name = tensor<string, []>("wrapped_model_language_model_layers_20_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_20_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1313451648))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1326558912))), name = tensor<string, []>("wrapped_model_language_model_layers_20_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_21_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1326559040))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1329180544))), name = tensor<string, []>("wrapped_model_language_model_layers_21_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_21_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1329180672))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1330491456))), name = tensor<string, []>("wrapped_model_language_model_layers_21_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_21_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1330491584))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1331802368))), name = tensor<string, []>("wrapped_model_language_model_layers_21_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_21_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1331802496))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1334424000))), name = tensor<string, []>("wrapped_model_language_model_layers_21_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_21_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1334424128))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1347531392))), name = tensor<string, []>("wrapped_model_language_model_layers_21_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_21_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1347531520))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1360638784))), name = tensor<string, []>("wrapped_model_language_model_layers_21_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_21_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1360638912))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1373746176))), name = tensor<string, []>("wrapped_model_language_model_layers_21_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_22_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1373746304))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1376367808))), name = tensor<string, []>("wrapped_model_language_model_layers_22_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_22_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1376367936))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1377678720))), name = tensor<string, []>("wrapped_model_language_model_layers_22_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_22_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1377678848))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1378989632))), name = tensor<string, []>("wrapped_model_language_model_layers_22_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_22_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1378989760))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1381611264))), name = tensor<string, []>("wrapped_model_language_model_layers_22_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_22_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1381611392))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1394718656))), name = tensor<string, []>("wrapped_model_language_model_layers_22_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_22_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1394718784))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1407826048))), name = tensor<string, []>("wrapped_model_language_model_layers_22_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_22_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1407826176))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1420933440))), name = tensor<string, []>("wrapped_model_language_model_layers_22_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_23_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1420933568))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1423555072))), name = tensor<string, []>("wrapped_model_language_model_layers_23_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_23_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1423555200))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1424865984))), name = tensor<string, []>("wrapped_model_language_model_layers_23_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_23_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1424866112))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1426176896))), name = tensor<string, []>("wrapped_model_language_model_layers_23_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_23_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1426177024))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1428798528))), name = tensor<string, []>("wrapped_model_language_model_layers_23_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_23_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1428798656))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1441905920))), name = tensor<string, []>("wrapped_model_language_model_layers_23_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_23_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1441906048))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1455013312))), name = tensor<string, []>("wrapped_model_language_model_layers_23_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_23_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1455013440))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1468120704))), name = tensor<string, []>("wrapped_model_language_model_layers_23_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_24_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1468120832))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1470742336))), name = tensor<string, []>("wrapped_model_language_model_layers_24_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_24_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1470742464))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1472053248))), name = tensor<string, []>("wrapped_model_language_model_layers_24_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_24_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1472053376))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1473364160))), name = tensor<string, []>("wrapped_model_language_model_layers_24_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_24_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1473364288))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1475985792))), name = tensor<string, []>("wrapped_model_language_model_layers_24_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_24_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1475985920))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1489093184))), name = tensor<string, []>("wrapped_model_language_model_layers_24_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_24_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1489093312))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1502200576))), name = tensor<string, []>("wrapped_model_language_model_layers_24_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_24_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1502200704))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1515307968))), name = tensor<string, []>("wrapped_model_language_model_layers_24_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_25_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1515308096))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1517929600))), name = tensor<string, []>("wrapped_model_language_model_layers_25_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_25_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1517929728))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1519240512))), name = tensor<string, []>("wrapped_model_language_model_layers_25_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_25_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1519240640))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1520551424))), name = tensor<string, []>("wrapped_model_language_model_layers_25_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_25_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1520551552))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1523173056))), name = tensor<string, []>("wrapped_model_language_model_layers_25_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_25_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1523173184))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1536280448))), name = tensor<string, []>("wrapped_model_language_model_layers_25_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_25_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1536280576))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1549387840))), name = tensor<string, []>("wrapped_model_language_model_layers_25_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_25_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1549387968))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1562495232))), name = tensor<string, []>("wrapped_model_language_model_layers_25_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_26_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1562495360))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1565116864))), name = tensor<string, []>("wrapped_model_language_model_layers_26_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_26_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1565116992))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1566427776))), name = tensor<string, []>("wrapped_model_language_model_layers_26_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_26_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1566427904))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1567738688))), name = tensor<string, []>("wrapped_model_language_model_layers_26_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_26_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1567738816))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1570360320))), name = tensor<string, []>("wrapped_model_language_model_layers_26_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_26_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1570360448))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1583467712))), name = tensor<string, []>("wrapped_model_language_model_layers_26_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_26_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1583467840))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1596575104))), name = tensor<string, []>("wrapped_model_language_model_layers_26_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_26_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1596575232))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1609682496))), name = tensor<string, []>("wrapped_model_language_model_layers_26_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_27_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1609682624))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1612304128))), name = tensor<string, []>("wrapped_model_language_model_layers_27_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_27_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1612304256))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1613615040))), name = tensor<string, []>("wrapped_model_language_model_layers_27_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_27_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1613615168))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1614925952))), name = tensor<string, []>("wrapped_model_language_model_layers_27_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_27_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1614926080))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1617547584))), name = tensor<string, []>("wrapped_model_language_model_layers_27_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_27_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1617547712))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1630654976))), name = tensor<string, []>("wrapped_model_language_model_layers_27_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_27_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1630655104))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1643762368))), name = tensor<string, []>("wrapped_model_language_model_layers_27_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_27_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1643762496))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1656869760))), name = tensor<string, []>("wrapped_model_language_model_layers_27_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_28_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1656869888))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1659491392))), name = tensor<string, []>("wrapped_model_language_model_layers_28_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_28_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1659491520))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1660802304))), name = tensor<string, []>("wrapped_model_language_model_layers_28_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_28_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1660802432))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1662113216))), name = tensor<string, []>("wrapped_model_language_model_layers_28_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_28_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1662113344))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1664734848))), name = tensor<string, []>("wrapped_model_language_model_layers_28_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_28_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1664734976))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1677842240))), name = tensor<string, []>("wrapped_model_language_model_layers_28_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_28_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1677842368))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1690949632))), name = tensor<string, []>("wrapped_model_language_model_layers_28_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_28_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1690949760))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1704057024))), name = tensor<string, []>("wrapped_model_language_model_layers_28_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_29_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1704057152))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1706678656))), name = tensor<string, []>("wrapped_model_language_model_layers_29_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_29_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1706678784))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1707989568))), name = tensor<string, []>("wrapped_model_language_model_layers_29_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_29_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1707989696))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1709300480))), name = tensor<string, []>("wrapped_model_language_model_layers_29_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_29_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1709300608))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1711922112))), name = tensor<string, []>("wrapped_model_language_model_layers_29_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_29_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1711922240))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1725029504))), name = tensor<string, []>("wrapped_model_language_model_layers_29_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_29_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1725029632))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1738136896))), name = tensor<string, []>("wrapped_model_language_model_layers_29_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_29_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1738137024))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1751244288))), name = tensor<string, []>("wrapped_model_language_model_layers_29_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_30_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1751244416))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1753865920))), name = tensor<string, []>("wrapped_model_language_model_layers_30_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_30_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1753866048))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1755176832))), name = tensor<string, []>("wrapped_model_language_model_layers_30_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_30_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1755176960))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1756487744))), name = tensor<string, []>("wrapped_model_language_model_layers_30_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_30_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1756487872))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1759109376))), name = tensor<string, []>("wrapped_model_language_model_layers_30_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_30_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1759109504))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1772216768))), name = tensor<string, []>("wrapped_model_language_model_layers_30_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_30_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1772216896))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1785324160))), name = tensor<string, []>("wrapped_model_language_model_layers_30_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_30_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1785324288))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1798431552))), name = tensor<string, []>("wrapped_model_language_model_layers_30_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_31_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1798431680))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1801053184))), name = tensor<string, []>("wrapped_model_language_model_layers_31_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_31_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1801053312))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1802364096))), name = tensor<string, []>("wrapped_model_language_model_layers_31_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_31_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1802364224))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1803675008))), name = tensor<string, []>("wrapped_model_language_model_layers_31_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_31_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1803675136))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1806296640))), name = tensor<string, []>("wrapped_model_language_model_layers_31_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_31_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1806296768))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1819404032))), name = tensor<string, []>("wrapped_model_language_model_layers_31_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_31_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1819404160))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1832511424))), name = tensor<string, []>("wrapped_model_language_model_layers_31_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_31_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1832511552))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1845618816))), name = tensor<string, []>("wrapped_model_language_model_layers_31_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_32_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1845618944))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1848240448))), name = tensor<string, []>("wrapped_model_language_model_layers_32_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_32_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1848240576))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1849551360))), name = tensor<string, []>("wrapped_model_language_model_layers_32_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_32_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1849551488))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1850862272))), name = tensor<string, []>("wrapped_model_language_model_layers_32_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_32_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1850862400))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1853483904))), name = tensor<string, []>("wrapped_model_language_model_layers_32_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_32_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1853484032))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1866591296))), name = tensor<string, []>("wrapped_model_language_model_layers_32_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_32_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1866591424))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1879698688))), name = tensor<string, []>("wrapped_model_language_model_layers_32_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_32_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1879698816))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1892806080))), name = tensor<string, []>("wrapped_model_language_model_layers_32_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<fp16, [2048, 2560]> wrapped_model_language_model_layers_33_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1892806208))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1895427712))), name = tensor<string, []>("wrapped_model_language_model_layers_33_self_attn_q_proj_weight_palettized"), shape = tensor<uint32, [2]>([2048, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_33_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1895427840))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1896738624))), name = tensor<string, []>("wrapped_model_language_model_layers_33_self_attn_k_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [1024, 2560]> wrapped_model_language_model_layers_33_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1896738752))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1898049536))), name = tensor<string, []>("wrapped_model_language_model_layers_33_self_attn_v_proj_weight_palettized"), shape = tensor<uint32, [2]>([1024, 2560])];
            tensor<fp16, [2560, 2048]> wrapped_model_language_model_layers_33_self_attn_o_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2621440]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1898049664))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1900671168))), name = tensor<string, []>("wrapped_model_language_model_layers_33_self_attn_o_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 2048])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_33_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1900671296))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1913778560))), name = tensor<string, []>("wrapped_model_language_model_layers_33_mlp_gate_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [10240, 2560]> wrapped_model_language_model_layers_33_mlp_up_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1913778688))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1926885952))), name = tensor<string, []>("wrapped_model_language_model_layers_33_mlp_up_proj_weight_palettized"), shape = tensor<uint32, [2]>([10240, 2560])];
            tensor<fp16, [2560, 10240]> wrapped_model_language_model_layers_33_mlp_down_proj_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [13107200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1926886080))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1939993344))), name = tensor<string, []>("wrapped_model_language_model_layers_33_mlp_down_proj_weight_palettized"), shape = tensor<uint32, [2]>([2560, 10240])];
            tensor<int32, []> var_24 = const()[name = tensor<string, []>("op_24"), val = tensor<int32, []>(-1)];
            tensor<int32, []> var_35_batch_dims_0 = const()[name = tensor<string, []>("op_35_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> var_35_validate_indices_0 = const()[name = tensor<string, []>("op_35_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<int32, []> greater_equal_0_y_0 = const()[name = tensor<string, []>("greater_equal_0_y_0"), val = tensor<int32, []>(0)];
            tensor<bool, [1, 256]> greater_equal_0 = greater_equal(x = input_ids, y = greater_equal_0_y_0)[name = tensor<string, []>("greater_equal_0")];
            tensor<int32, []> slice_by_index_0 = const()[name = tensor<string, []>("slice_by_index_0"), val = tensor<int32, []>(262208)];
            tensor<int32, [1, 256]> add_0 = add(x = input_ids, y = slice_by_index_0)[name = tensor<string, []>("add_0")];
            tensor<int32, [1, 256]> select_0 = select(a = input_ids, b = add_0, cond = greater_equal_0)[name = tensor<string, []>("select_0")];
            tensor<int32, []> greater_equal_0_y_0_1 = const()[name = tensor<string, []>("greater_equal_0_y_0_1"), val = tensor<int32, []>(0)];
            tensor<bool, [1, 256]> greater_equal_0_1 = greater_equal(x = select_0, y = greater_equal_0_y_0_1)[name = tensor<string, []>("greater_equal_0_1")];
            tensor<int32, []> slice_by_index_0_1 = const()[name = tensor<string, []>("slice_by_index_0_1"), val = tensor<int32, []>(262208)];
            tensor<int32, [1, 256]> add_0_1 = add(x = select_0, y = slice_by_index_0_1)[name = tensor<string, []>("add_0_1")];
            tensor<int32, [1, 256]> select_0_1 = select(a = select_0, b = add_0_1, cond = greater_equal_0_1)[name = tensor<string, []>("select_0_1")];
            tensor<int32, []> op_35_axis_0 = const()[name = tensor<string, []>("op_35_axis_0"), val = tensor<int32, []>(0)];
            tensor<fp16, [1, 256, 2560]> op_35 = gather(axis = op_35_axis_0, batch_dims = var_35_batch_dims_0, indices = select_0_1, validate_indices = var_35_validate_indices_0, x = wrapped_model_language_model_embed_tokens_weight_palettized)[name = tensor<string, []>("op_35")];
            tensor<fp16, []> const_0 = const()[name = tensor<string, []>("const_0"), val = tensor<fp16, []>(0x1.94cp+5)];
            tensor<fp16, [1, 256, 2560]> inputs_embeds = mul(x = op_35, y = const_0)[name = tensor<string, []>("inputs_embeds")];
            tensor<fp16, []> var_22_promoted_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_219_cast_fp16 = pow(x = inputs_embeds, y = var_22_promoted_to_fp16)[name = tensor<string, []>("op_219_cast_fp16")];
            tensor<int32, [1]> var_221_axes_0 = const()[name = tensor<string, []>("op_221_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_221_keep_dims_0 = const()[name = tensor<string, []>("op_221_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_221_cast_fp16 = reduce_mean(axes = var_221_axes_0, keep_dims = var_221_keep_dims_0, x = var_219_cast_fp16)[name = tensor<string, []>("op_221_cast_fp16")];
            tensor<fp16, []> var_222_to_fp16 = const()[name = tensor<string, []>("op_222_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_223_cast_fp16 = add(x = var_221_cast_fp16, y = var_222_to_fp16)[name = tensor<string, []>("op_223_cast_fp16")];
            tensor<fp32, []> var_224_epsilon_0 = const()[name = tensor<string, []>("op_224_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_224_cast_fp16 = rsqrt(epsilon = var_224_epsilon_0, x = var_223_cast_fp16)[name = tensor<string, []>("op_224_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_1_cast_fp16 = mul(x = inputs_embeds, y = var_224_cast_fp16)[name = tensor<string, []>("output_1_cast_fp16")];
            tensor<fp16, [2560]> var_228_to_fp16 = const()[name = tensor<string, []>("op_228_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1939993472)))];
            tensor<fp16, [1, 256, 2560]> output_3_cast_fp16 = mul(x = output_1_cast_fp16, y = var_228_to_fp16)[name = tensor<string, []>("output_3_cast_fp16")];
            tensor<fp16, [2048]> linear_0_bias_0 = const()[name = tensor<string, []>("linear_0_bias_0"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1939998656)))];
            tensor<fp16, [1, 256, 2048]> var_240 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_0_self_attn_q_proj_weight_palettized, x = output_3_cast_fp16)[name = tensor<string, []>("linear_0")];
            tensor<int32, [4]> var_241 = const()[name = tensor<string, []>("op_241"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_242 = reshape(shape = var_241, x = var_240)[name = tensor<string, []>("op_242")];
            tensor<int32, [4]> x_3_perm_0 = const()[name = tensor<string, []>("x_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024]> linear_1_bias_0 = const()[name = tensor<string, []>("linear_1_bias_0"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940002816)))];
            tensor<fp16, [1, 256, 1024]> var_245 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_0_self_attn_k_proj_weight_palettized, x = output_3_cast_fp16)[name = tensor<string, []>("linear_1")];
            tensor<int32, [4]> var_246 = const()[name = tensor<string, []>("op_246"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_247 = reshape(shape = var_246, x = var_245)[name = tensor<string, []>("op_247")];
            tensor<int32, [4]> x_7_perm_0 = const()[name = tensor<string, []>("x_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_250 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_0_self_attn_v_proj_weight_palettized, x = output_3_cast_fp16)[name = tensor<string, []>("linear_2")];
            tensor<int32, [4]> var_251 = const()[name = tensor<string, []>("op_251"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_252 = reshape(shape = var_251, x = var_250)[name = tensor<string, []>("op_252")];
            tensor<int32, [4]> hidden_states_7_perm_0 = const()[name = tensor<string, []>("hidden_states_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_1_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_1_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_3 = transpose(perm = x_3_perm_0, x = var_242)[name = tensor<string, []>("transpose_135")];
            tensor<fp16, [1, 8, 256, 256]> var_256_cast_fp16 = pow(x = x_3, y = var_22_promoted_1_to_fp16)[name = tensor<string, []>("op_256_cast_fp16")];
            tensor<int32, [1]> var_258_axes_0 = const()[name = tensor<string, []>("op_258_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_258_keep_dims_0 = const()[name = tensor<string, []>("op_258_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_258_cast_fp16 = reduce_mean(axes = var_258_axes_0, keep_dims = var_258_keep_dims_0, x = var_256_cast_fp16)[name = tensor<string, []>("op_258_cast_fp16")];
            tensor<fp16, []> var_259_to_fp16 = const()[name = tensor<string, []>("op_259_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_260_cast_fp16 = add(x = var_258_cast_fp16, y = var_259_to_fp16)[name = tensor<string, []>("op_260_cast_fp16")];
            tensor<fp32, []> var_261_epsilon_0 = const()[name = tensor<string, []>("op_261_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_261_cast_fp16 = rsqrt(epsilon = var_261_epsilon_0, x = var_260_cast_fp16)[name = tensor<string, []>("op_261_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_5_cast_fp16 = mul(x = x_3, y = var_261_cast_fp16)[name = tensor<string, []>("output_5_cast_fp16")];
            tensor<fp16, [256]> var_265_to_fp16 = const()[name = tensor<string, []>("op_265_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940004928)))];
            tensor<fp16, [1, 8, 256, 256]> output_7_cast_fp16 = mul(x = output_5_cast_fp16, y = var_265_to_fp16)[name = tensor<string, []>("output_7_cast_fp16")];
            tensor<fp16, []> var_22_promoted_2_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_2_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_7 = transpose(perm = x_7_perm_0, x = var_247)[name = tensor<string, []>("transpose_134")];
            tensor<fp16, [1, 4, 256, 256]> var_270_cast_fp16 = pow(x = x_7, y = var_22_promoted_2_to_fp16)[name = tensor<string, []>("op_270_cast_fp16")];
            tensor<int32, [1]> var_272_axes_0 = const()[name = tensor<string, []>("op_272_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_272_keep_dims_0 = const()[name = tensor<string, []>("op_272_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_272_cast_fp16 = reduce_mean(axes = var_272_axes_0, keep_dims = var_272_keep_dims_0, x = var_270_cast_fp16)[name = tensor<string, []>("op_272_cast_fp16")];
            tensor<fp16, []> var_273_to_fp16 = const()[name = tensor<string, []>("op_273_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_274_cast_fp16 = add(x = var_272_cast_fp16, y = var_273_to_fp16)[name = tensor<string, []>("op_274_cast_fp16")];
            tensor<fp32, []> var_275_epsilon_0 = const()[name = tensor<string, []>("op_275_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_275_cast_fp16 = rsqrt(epsilon = var_275_epsilon_0, x = var_274_cast_fp16)[name = tensor<string, []>("op_275_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_9_cast_fp16 = mul(x = x_7, y = var_275_cast_fp16)[name = tensor<string, []>("output_9_cast_fp16")];
            tensor<fp16, [256]> var_279_to_fp16 = const()[name = tensor<string, []>("op_279_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940005504)))];
            tensor<fp16, [1, 4, 256, 256]> output_11_cast_fp16 = mul(x = output_9_cast_fp16, y = var_279_to_fp16)[name = tensor<string, []>("output_11_cast_fp16")];
            tensor<fp16, [1, 1, 256, 256]> cos_7_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [32768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940006080))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940038912))), name = tensor<string, []>("cos_7_palettized"), shape = tensor<uint32, [4]>([1, 1, 256, 256])];
            tensor<fp16, [1, 1, 256, 256]> sin_7_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [32768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940039040))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940071872))), name = tensor<string, []>("sin_7_palettized"), shape = tensor<uint32, [4]>([1, 1, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> var_284 = mul(x = output_7_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_284")];
            tensor<int32, [4]> x1_1_begin_0 = const()[name = tensor<string, []>("x1_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_1_end_0 = const()[name = tensor<string, []>("x1_1_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_1_end_mask_0 = const()[name = tensor<string, []>("x1_1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_1 = slice_by_index(begin = x1_1_begin_0, end = x1_1_end_0, end_mask = x1_1_end_mask_0, x = output_7_cast_fp16)[name = tensor<string, []>("x1_1")];
            tensor<int32, [4]> x2_1_begin_0 = const()[name = tensor<string, []>("x2_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_1_end_0 = const()[name = tensor<string, []>("x2_1_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_1_end_mask_0 = const()[name = tensor<string, []>("x2_1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_1 = slice_by_index(begin = x2_1_begin_0, end = x2_1_end_0, end_mask = x2_1_end_mask_0, x = output_7_cast_fp16)[name = tensor<string, []>("x2_1")];
            tensor<fp16, []> const_35_promoted = const()[name = tensor<string, []>("const_35_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_295 = mul(x = x2_1, y = const_35_promoted)[name = tensor<string, []>("op_295")];
            tensor<bool, []> var_297_interleave_0 = const()[name = tensor<string, []>("op_297_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_297 = concat(axis = var_24, interleave = var_297_interleave_0, values = (var_295, x1_1))[name = tensor<string, []>("op_297")];
            tensor<fp16, [1, 8, 256, 256]> var_298 = mul(x = var_297, y = sin_7_palettized)[name = tensor<string, []>("op_298")];
            tensor<fp16, [1, 8, 256, 256]> query_1 = add(x = var_284, y = var_298)[name = tensor<string, []>("query_1")];
            tensor<fp16, [1, 4, 256, 256]> var_300 = mul(x = output_11_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_300")];
            tensor<int32, [4]> x1_3_begin_0 = const()[name = tensor<string, []>("x1_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_3_end_0 = const()[name = tensor<string, []>("x1_3_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_3_end_mask_0 = const()[name = tensor<string, []>("x1_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_3 = slice_by_index(begin = x1_3_begin_0, end = x1_3_end_0, end_mask = x1_3_end_mask_0, x = output_11_cast_fp16)[name = tensor<string, []>("x1_3")];
            tensor<int32, [4]> x2_3_begin_0 = const()[name = tensor<string, []>("x2_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_3_end_0 = const()[name = tensor<string, []>("x2_3_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_3_end_mask_0 = const()[name = tensor<string, []>("x2_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_3 = slice_by_index(begin = x2_3_begin_0, end = x2_3_end_0, end_mask = x2_3_end_mask_0, x = output_11_cast_fp16)[name = tensor<string, []>("x2_3")];
            tensor<fp16, []> const_38_promoted = const()[name = tensor<string, []>("const_38_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_311 = mul(x = x2_3, y = const_38_promoted)[name = tensor<string, []>("op_311")];
            tensor<bool, []> var_313_interleave_0 = const()[name = tensor<string, []>("op_313_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_313 = concat(axis = var_24, interleave = var_313_interleave_0, values = (var_311, x1_3))[name = tensor<string, []>("op_313")];
            tensor<fp16, [1, 4, 256, 256]> var_314 = mul(x = var_313, y = sin_7_palettized)[name = tensor<string, []>("op_314")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_3 = add(x = var_300, y = var_314)[name = tensor<string, []>("hidden_states_3")];
            tensor<int32, [1]> var_323_axes_0 = const()[name = tensor<string, []>("op_323_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_323 = expand_dims(axes = var_323_axes_0, x = hidden_states_3)[name = tensor<string, []>("op_323")];
            tensor<int32, [5]> hidden_states_5_reps_0 = const()[name = tensor<string, []>("hidden_states_5_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_5 = tile(reps = hidden_states_5_reps_0, x = var_323)[name = tensor<string, []>("hidden_states_5")];
            tensor<int32, [4]> var_331 = const()[name = tensor<string, []>("op_331"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_1 = reshape(shape = var_331, x = hidden_states_5)[name = tensor<string, []>("key_states_1")];
            tensor<int32, [1]> var_340_axes_0 = const()[name = tensor<string, []>("op_340_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_7 = transpose(perm = hidden_states_7_perm_0, x = var_252)[name = tensor<string, []>("transpose_133")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_340 = expand_dims(axes = var_340_axes_0, x = hidden_states_7)[name = tensor<string, []>("op_340")];
            tensor<int32, [5]> hidden_states_9_reps_0 = const()[name = tensor<string, []>("hidden_states_9_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_9 = tile(reps = hidden_states_9_reps_0, x = var_340)[name = tensor<string, []>("hidden_states_9")];
            tensor<int32, [4]> var_348 = const()[name = tensor<string, []>("op_348"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_1 = reshape(shape = var_348, x = hidden_states_9)[name = tensor<string, []>("value_states_1")];
            tensor<bool, []> var_351_transpose_x_1 = const()[name = tensor<string, []>("op_351_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_351_transpose_y_1 = const()[name = tensor<string, []>("op_351_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_351 = matmul(transpose_x = var_351_transpose_x_1, transpose_y = var_351_transpose_y_1, x = query_1, y = key_states_1)[name = tensor<string, []>("op_351")];
            tensor<fp16, []> var_352_to_fp16 = const()[name = tensor<string, []>("op_352_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_1_cast_fp16 = mul(x = var_351, y = var_352_to_fp16)[name = tensor<string, []>("attn_weights_1_cast_fp16")];
            tensor<fp16, [1, 1, 256, 256]> attention_mask_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [32768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940072000))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940104832))), name = tensor<string, []>("attention_mask_to_fp16_palettized"), shape = tensor<uint32, [4]>([1, 1, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> input_1_cast_fp16 = add(x = attn_weights_1_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_1_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_360_cast_fp16 = softmax(axis = var_24, x = input_1_cast_fp16)[name = tensor<string, []>("op_360_cast_fp16")];
            tensor<bool, []> attn_output_1_transpose_x_0 = const()[name = tensor<string, []>("attn_output_1_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_1_transpose_y_0 = const()[name = tensor<string, []>("attn_output_1_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_1 = matmul(transpose_x = attn_output_1_transpose_x_0, transpose_y = attn_output_1_transpose_y_0, x = var_360_cast_fp16, y = value_states_1)[name = tensor<string, []>("attn_output_1")];
            tensor<int32, [4]> var_364_perm_0 = const()[name = tensor<string, []>("op_364_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_366 = const()[name = tensor<string, []>("op_366"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_364 = transpose(perm = var_364_perm_0, x = attn_output_1)[name = tensor<string, []>("transpose_132")];
            tensor<fp16, [1, 256, 2048]> var_367 = reshape(shape = var_366, x = var_364)[name = tensor<string, []>("op_367")];
            tensor<fp16, [2560]> linear_3_bias_0 = const()[name = tensor<string, []>("linear_3_bias_0"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940104960)))];
            tensor<fp16, [1, 256, 2560]> x_11 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_0_self_attn_o_proj_weight_palettized, x = var_367)[name = tensor<string, []>("linear_3")];
            tensor<fp16, []> var_22_promoted_3_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_3_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_373_cast_fp16 = pow(x = x_11, y = var_22_promoted_3_to_fp16)[name = tensor<string, []>("op_373_cast_fp16")];
            tensor<int32, [1]> var_375_axes_0 = const()[name = tensor<string, []>("op_375_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_375_keep_dims_0 = const()[name = tensor<string, []>("op_375_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_375_cast_fp16 = reduce_mean(axes = var_375_axes_0, keep_dims = var_375_keep_dims_0, x = var_373_cast_fp16)[name = tensor<string, []>("op_375_cast_fp16")];
            tensor<fp16, []> var_376_to_fp16 = const()[name = tensor<string, []>("op_376_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_377_cast_fp16 = add(x = var_375_cast_fp16, y = var_376_to_fp16)[name = tensor<string, []>("op_377_cast_fp16")];
            tensor<fp32, []> var_378_epsilon_0 = const()[name = tensor<string, []>("op_378_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_378_cast_fp16 = rsqrt(epsilon = var_378_epsilon_0, x = var_377_cast_fp16)[name = tensor<string, []>("op_378_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_13_cast_fp16 = mul(x = x_11, y = var_378_cast_fp16)[name = tensor<string, []>("output_13_cast_fp16")];
            tensor<fp16, [2560]> var_382_to_fp16 = const()[name = tensor<string, []>("op_382_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940110144)))];
            tensor<fp16, [1, 256, 2560]> output_15_cast_fp16 = mul(x = output_13_cast_fp16, y = var_382_to_fp16)[name = tensor<string, []>("output_15_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_15 = add(x = inputs_embeds, y = output_15_cast_fp16)[name = tensor<string, []>("x_15")];
            tensor<fp16, []> var_22_promoted_4_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_4_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_388_cast_fp16 = pow(x = x_15, y = var_22_promoted_4_to_fp16)[name = tensor<string, []>("op_388_cast_fp16")];
            tensor<int32, [1]> var_390_axes_0 = const()[name = tensor<string, []>("op_390_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_390_keep_dims_0 = const()[name = tensor<string, []>("op_390_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_390_cast_fp16 = reduce_mean(axes = var_390_axes_0, keep_dims = var_390_keep_dims_0, x = var_388_cast_fp16)[name = tensor<string, []>("op_390_cast_fp16")];
            tensor<fp16, []> var_391_to_fp16 = const()[name = tensor<string, []>("op_391_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_392_cast_fp16 = add(x = var_390_cast_fp16, y = var_391_to_fp16)[name = tensor<string, []>("op_392_cast_fp16")];
            tensor<fp32, []> var_393_epsilon_0 = const()[name = tensor<string, []>("op_393_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_393_cast_fp16 = rsqrt(epsilon = var_393_epsilon_0, x = var_392_cast_fp16)[name = tensor<string, []>("op_393_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_17_cast_fp16 = mul(x = x_15, y = var_393_cast_fp16)[name = tensor<string, []>("output_17_cast_fp16")];
            tensor<fp16, [2560]> var_397_to_fp16 = const()[name = tensor<string, []>("op_397_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940115328)))];
            tensor<fp16, [1, 256, 2560]> output_19_cast_fp16 = mul(x = output_17_cast_fp16, y = var_397_to_fp16)[name = tensor<string, []>("output_19_cast_fp16")];
            tensor<fp16, [10240]> linear_4_bias_0 = const()[name = tensor<string, []>("linear_4_bias_0"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940120512)))];
            tensor<fp16, [1, 256, 10240]> input_9 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_0_mlp_gate_proj_weight_palettized, x = output_19_cast_fp16)[name = tensor<string, []>("linear_4")];
            tensor<string, []> var_405_mode_0 = const()[name = tensor<string, []>("op_405_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_405 = gelu(mode = var_405_mode_0, x = input_9)[name = tensor<string, []>("op_405")];
            tensor<fp16, [1, 256, 10240]> var_407 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_0_mlp_up_proj_weight_palettized, x = output_19_cast_fp16)[name = tensor<string, []>("linear_5")];
            tensor<fp16, [1, 256, 10240]> input_11 = mul(x = var_405, y = var_407)[name = tensor<string, []>("input_11")];
            tensor<fp16, [1, 256, 2560]> x_19 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_0_mlp_down_proj_weight_palettized, x = input_11)[name = tensor<string, []>("linear_6")];
            tensor<fp16, []> var_22_promoted_5_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_5_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_413_cast_fp16 = pow(x = x_19, y = var_22_promoted_5_to_fp16)[name = tensor<string, []>("op_413_cast_fp16")];
            tensor<int32, [1]> var_415_axes_0 = const()[name = tensor<string, []>("op_415_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_415_keep_dims_0 = const()[name = tensor<string, []>("op_415_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_415_cast_fp16 = reduce_mean(axes = var_415_axes_0, keep_dims = var_415_keep_dims_0, x = var_413_cast_fp16)[name = tensor<string, []>("op_415_cast_fp16")];
            tensor<fp16, []> var_416_to_fp16 = const()[name = tensor<string, []>("op_416_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_417_cast_fp16 = add(x = var_415_cast_fp16, y = var_416_to_fp16)[name = tensor<string, []>("op_417_cast_fp16")];
            tensor<fp32, []> var_418_epsilon_0 = const()[name = tensor<string, []>("op_418_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_418_cast_fp16 = rsqrt(epsilon = var_418_epsilon_0, x = var_417_cast_fp16)[name = tensor<string, []>("op_418_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_21_cast_fp16 = mul(x = x_19, y = var_418_cast_fp16)[name = tensor<string, []>("output_21_cast_fp16")];
            tensor<fp16, [2560]> var_422_to_fp16 = const()[name = tensor<string, []>("op_422_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940141056)))];
            tensor<fp16, [1, 256, 2560]> output_23_cast_fp16 = mul(x = output_21_cast_fp16, y = var_422_to_fp16)[name = tensor<string, []>("output_23_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_23 = add(x = x_15, y = output_23_cast_fp16)[name = tensor<string, []>("x_23")];
            tensor<fp16, []> var_22_promoted_6_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_6_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_434_cast_fp16 = pow(x = x_23, y = var_22_promoted_6_to_fp16)[name = tensor<string, []>("op_434_cast_fp16")];
            tensor<int32, [1]> var_436_axes_0 = const()[name = tensor<string, []>("op_436_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_436_keep_dims_0 = const()[name = tensor<string, []>("op_436_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_436_cast_fp16 = reduce_mean(axes = var_436_axes_0, keep_dims = var_436_keep_dims_0, x = var_434_cast_fp16)[name = tensor<string, []>("op_436_cast_fp16")];
            tensor<fp16, []> var_437_to_fp16 = const()[name = tensor<string, []>("op_437_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_438_cast_fp16 = add(x = var_436_cast_fp16, y = var_437_to_fp16)[name = tensor<string, []>("op_438_cast_fp16")];
            tensor<fp32, []> var_439_epsilon_0 = const()[name = tensor<string, []>("op_439_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_439_cast_fp16 = rsqrt(epsilon = var_439_epsilon_0, x = var_438_cast_fp16)[name = tensor<string, []>("op_439_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_25_cast_fp16 = mul(x = x_23, y = var_439_cast_fp16)[name = tensor<string, []>("output_25_cast_fp16")];
            tensor<fp16, [2560]> var_443_to_fp16 = const()[name = tensor<string, []>("op_443_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940146240)))];
            tensor<fp16, [1, 256, 2560]> output_27_cast_fp16 = mul(x = output_25_cast_fp16, y = var_443_to_fp16)[name = tensor<string, []>("output_27_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_455 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_1_self_attn_q_proj_weight_palettized, x = output_27_cast_fp16)[name = tensor<string, []>("linear_7")];
            tensor<int32, [4]> var_456 = const()[name = tensor<string, []>("op_456"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_457 = reshape(shape = var_456, x = var_455)[name = tensor<string, []>("op_457")];
            tensor<int32, [4]> x_27_perm_0 = const()[name = tensor<string, []>("x_27_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_460 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_1_self_attn_k_proj_weight_palettized, x = output_27_cast_fp16)[name = tensor<string, []>("linear_8")];
            tensor<int32, [4]> var_461 = const()[name = tensor<string, []>("op_461"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_462 = reshape(shape = var_461, x = var_460)[name = tensor<string, []>("op_462")];
            tensor<int32, [4]> x_31_perm_0 = const()[name = tensor<string, []>("x_31_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_465 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_1_self_attn_v_proj_weight_palettized, x = output_27_cast_fp16)[name = tensor<string, []>("linear_9")];
            tensor<int32, [4]> var_466 = const()[name = tensor<string, []>("op_466"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_467 = reshape(shape = var_466, x = var_465)[name = tensor<string, []>("op_467")];
            tensor<int32, [4]> hidden_states_21_perm_0 = const()[name = tensor<string, []>("hidden_states_21_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_7_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_7_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_27 = transpose(perm = x_27_perm_0, x = var_457)[name = tensor<string, []>("transpose_131")];
            tensor<fp16, [1, 8, 256, 256]> var_471_cast_fp16 = pow(x = x_27, y = var_22_promoted_7_to_fp16)[name = tensor<string, []>("op_471_cast_fp16")];
            tensor<int32, [1]> var_473_axes_0 = const()[name = tensor<string, []>("op_473_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_473_keep_dims_0 = const()[name = tensor<string, []>("op_473_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_473_cast_fp16 = reduce_mean(axes = var_473_axes_0, keep_dims = var_473_keep_dims_0, x = var_471_cast_fp16)[name = tensor<string, []>("op_473_cast_fp16")];
            tensor<fp16, []> var_474_to_fp16 = const()[name = tensor<string, []>("op_474_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_475_cast_fp16 = add(x = var_473_cast_fp16, y = var_474_to_fp16)[name = tensor<string, []>("op_475_cast_fp16")];
            tensor<fp32, []> var_476_epsilon_0 = const()[name = tensor<string, []>("op_476_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_476_cast_fp16 = rsqrt(epsilon = var_476_epsilon_0, x = var_475_cast_fp16)[name = tensor<string, []>("op_476_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_29_cast_fp16 = mul(x = x_27, y = var_476_cast_fp16)[name = tensor<string, []>("output_29_cast_fp16")];
            tensor<fp16, [256]> var_480_to_fp16 = const()[name = tensor<string, []>("op_480_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940151424)))];
            tensor<fp16, [1, 8, 256, 256]> output_31_cast_fp16 = mul(x = output_29_cast_fp16, y = var_480_to_fp16)[name = tensor<string, []>("output_31_cast_fp16")];
            tensor<fp16, []> var_22_promoted_8_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_8_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_31 = transpose(perm = x_31_perm_0, x = var_462)[name = tensor<string, []>("transpose_130")];
            tensor<fp16, [1, 4, 256, 256]> var_485_cast_fp16 = pow(x = x_31, y = var_22_promoted_8_to_fp16)[name = tensor<string, []>("op_485_cast_fp16")];
            tensor<int32, [1]> var_487_axes_0 = const()[name = tensor<string, []>("op_487_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_487_keep_dims_0 = const()[name = tensor<string, []>("op_487_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_487_cast_fp16 = reduce_mean(axes = var_487_axes_0, keep_dims = var_487_keep_dims_0, x = var_485_cast_fp16)[name = tensor<string, []>("op_487_cast_fp16")];
            tensor<fp16, []> var_488_to_fp16 = const()[name = tensor<string, []>("op_488_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_489_cast_fp16 = add(x = var_487_cast_fp16, y = var_488_to_fp16)[name = tensor<string, []>("op_489_cast_fp16")];
            tensor<fp32, []> var_490_epsilon_0 = const()[name = tensor<string, []>("op_490_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_490_cast_fp16 = rsqrt(epsilon = var_490_epsilon_0, x = var_489_cast_fp16)[name = tensor<string, []>("op_490_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_33_cast_fp16 = mul(x = x_31, y = var_490_cast_fp16)[name = tensor<string, []>("output_33_cast_fp16")];
            tensor<fp16, [256]> var_494_to_fp16 = const()[name = tensor<string, []>("op_494_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940152000)))];
            tensor<fp16, [1, 4, 256, 256]> output_35_cast_fp16 = mul(x = output_33_cast_fp16, y = var_494_to_fp16)[name = tensor<string, []>("output_35_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_499 = mul(x = output_31_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_499")];
            tensor<int32, [4]> x1_5_begin_0 = const()[name = tensor<string, []>("x1_5_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_5_end_0 = const()[name = tensor<string, []>("x1_5_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_5_end_mask_0 = const()[name = tensor<string, []>("x1_5_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_5 = slice_by_index(begin = x1_5_begin_0, end = x1_5_end_0, end_mask = x1_5_end_mask_0, x = output_31_cast_fp16)[name = tensor<string, []>("x1_5")];
            tensor<int32, [4]> x2_5_begin_0 = const()[name = tensor<string, []>("x2_5_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_5_end_0 = const()[name = tensor<string, []>("x2_5_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_5_end_mask_0 = const()[name = tensor<string, []>("x2_5_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_5 = slice_by_index(begin = x2_5_begin_0, end = x2_5_end_0, end_mask = x2_5_end_mask_0, x = output_31_cast_fp16)[name = tensor<string, []>("x2_5")];
            tensor<fp16, []> const_58_promoted = const()[name = tensor<string, []>("const_58_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_510 = mul(x = x2_5, y = const_58_promoted)[name = tensor<string, []>("op_510")];
            tensor<bool, []> var_512_interleave_0 = const()[name = tensor<string, []>("op_512_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_512 = concat(axis = var_24, interleave = var_512_interleave_0, values = (var_510, x1_5))[name = tensor<string, []>("op_512")];
            tensor<fp16, [1, 8, 256, 256]> var_513 = mul(x = var_512, y = sin_7_palettized)[name = tensor<string, []>("op_513")];
            tensor<fp16, [1, 8, 256, 256]> query_3 = add(x = var_499, y = var_513)[name = tensor<string, []>("query_3")];
            tensor<fp16, [1, 4, 256, 256]> var_515 = mul(x = output_35_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_515")];
            tensor<int32, [4]> x1_7_begin_0 = const()[name = tensor<string, []>("x1_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_7_end_0 = const()[name = tensor<string, []>("x1_7_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_7_end_mask_0 = const()[name = tensor<string, []>("x1_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_7 = slice_by_index(begin = x1_7_begin_0, end = x1_7_end_0, end_mask = x1_7_end_mask_0, x = output_35_cast_fp16)[name = tensor<string, []>("x1_7")];
            tensor<int32, [4]> x2_7_begin_0 = const()[name = tensor<string, []>("x2_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_7_end_0 = const()[name = tensor<string, []>("x2_7_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_7_end_mask_0 = const()[name = tensor<string, []>("x2_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_7 = slice_by_index(begin = x2_7_begin_0, end = x2_7_end_0, end_mask = x2_7_end_mask_0, x = output_35_cast_fp16)[name = tensor<string, []>("x2_7")];
            tensor<fp16, []> const_61_promoted = const()[name = tensor<string, []>("const_61_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_526 = mul(x = x2_7, y = const_61_promoted)[name = tensor<string, []>("op_526")];
            tensor<bool, []> var_528_interleave_0 = const()[name = tensor<string, []>("op_528_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_528 = concat(axis = var_24, interleave = var_528_interleave_0, values = (var_526, x1_7))[name = tensor<string, []>("op_528")];
            tensor<fp16, [1, 4, 256, 256]> var_529 = mul(x = var_528, y = sin_7_palettized)[name = tensor<string, []>("op_529")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_17 = add(x = var_515, y = var_529)[name = tensor<string, []>("hidden_states_17")];
            tensor<int32, [1]> var_538_axes_0 = const()[name = tensor<string, []>("op_538_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_538 = expand_dims(axes = var_538_axes_0, x = hidden_states_17)[name = tensor<string, []>("op_538")];
            tensor<int32, [5]> hidden_states_19_reps_0 = const()[name = tensor<string, []>("hidden_states_19_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_19 = tile(reps = hidden_states_19_reps_0, x = var_538)[name = tensor<string, []>("hidden_states_19")];
            tensor<int32, [4]> var_546 = const()[name = tensor<string, []>("op_546"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_3 = reshape(shape = var_546, x = hidden_states_19)[name = tensor<string, []>("key_states_3")];
            tensor<int32, [1]> var_555_axes_0 = const()[name = tensor<string, []>("op_555_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_21 = transpose(perm = hidden_states_21_perm_0, x = var_467)[name = tensor<string, []>("transpose_129")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_555 = expand_dims(axes = var_555_axes_0, x = hidden_states_21)[name = tensor<string, []>("op_555")];
            tensor<int32, [5]> hidden_states_23_reps_0 = const()[name = tensor<string, []>("hidden_states_23_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_23 = tile(reps = hidden_states_23_reps_0, x = var_555)[name = tensor<string, []>("hidden_states_23")];
            tensor<int32, [4]> var_563 = const()[name = tensor<string, []>("op_563"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_3 = reshape(shape = var_563, x = hidden_states_23)[name = tensor<string, []>("value_states_3")];
            tensor<bool, []> var_566_transpose_x_1 = const()[name = tensor<string, []>("op_566_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_566_transpose_y_1 = const()[name = tensor<string, []>("op_566_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_566 = matmul(transpose_x = var_566_transpose_x_1, transpose_y = var_566_transpose_y_1, x = query_3, y = key_states_3)[name = tensor<string, []>("op_566")];
            tensor<fp16, []> var_567_to_fp16 = const()[name = tensor<string, []>("op_567_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_5_cast_fp16 = mul(x = var_566, y = var_567_to_fp16)[name = tensor<string, []>("attn_weights_5_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_13_cast_fp16 = add(x = attn_weights_5_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_13_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_575_cast_fp16 = softmax(axis = var_24, x = input_13_cast_fp16)[name = tensor<string, []>("op_575_cast_fp16")];
            tensor<bool, []> attn_output_5_transpose_x_0 = const()[name = tensor<string, []>("attn_output_5_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_5_transpose_y_0 = const()[name = tensor<string, []>("attn_output_5_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_5 = matmul(transpose_x = attn_output_5_transpose_x_0, transpose_y = attn_output_5_transpose_y_0, x = var_575_cast_fp16, y = value_states_3)[name = tensor<string, []>("attn_output_5")];
            tensor<int32, [4]> var_579_perm_0 = const()[name = tensor<string, []>("op_579_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_581 = const()[name = tensor<string, []>("op_581"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_579 = transpose(perm = var_579_perm_0, x = attn_output_5)[name = tensor<string, []>("transpose_128")];
            tensor<fp16, [1, 256, 2048]> var_582 = reshape(shape = var_581, x = var_579)[name = tensor<string, []>("op_582")];
            tensor<fp16, [1, 256, 2560]> x_35 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_1_self_attn_o_proj_weight_palettized, x = var_582)[name = tensor<string, []>("linear_10")];
            tensor<fp16, []> var_22_promoted_9_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_9_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_588_cast_fp16 = pow(x = x_35, y = var_22_promoted_9_to_fp16)[name = tensor<string, []>("op_588_cast_fp16")];
            tensor<int32, [1]> var_590_axes_0 = const()[name = tensor<string, []>("op_590_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_590_keep_dims_0 = const()[name = tensor<string, []>("op_590_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_590_cast_fp16 = reduce_mean(axes = var_590_axes_0, keep_dims = var_590_keep_dims_0, x = var_588_cast_fp16)[name = tensor<string, []>("op_590_cast_fp16")];
            tensor<fp16, []> var_591_to_fp16 = const()[name = tensor<string, []>("op_591_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_592_cast_fp16 = add(x = var_590_cast_fp16, y = var_591_to_fp16)[name = tensor<string, []>("op_592_cast_fp16")];
            tensor<fp32, []> var_593_epsilon_0 = const()[name = tensor<string, []>("op_593_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_593_cast_fp16 = rsqrt(epsilon = var_593_epsilon_0, x = var_592_cast_fp16)[name = tensor<string, []>("op_593_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_37_cast_fp16 = mul(x = x_35, y = var_593_cast_fp16)[name = tensor<string, []>("output_37_cast_fp16")];
            tensor<fp16, [2560]> var_597_to_fp16 = const()[name = tensor<string, []>("op_597_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940152576)))];
            tensor<fp16, [1, 256, 2560]> output_39_cast_fp16 = mul(x = output_37_cast_fp16, y = var_597_to_fp16)[name = tensor<string, []>("output_39_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_39 = add(x = x_23, y = output_39_cast_fp16)[name = tensor<string, []>("x_39")];
            tensor<fp16, []> var_22_promoted_10_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_10_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_603_cast_fp16 = pow(x = x_39, y = var_22_promoted_10_to_fp16)[name = tensor<string, []>("op_603_cast_fp16")];
            tensor<int32, [1]> var_605_axes_0 = const()[name = tensor<string, []>("op_605_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_605_keep_dims_0 = const()[name = tensor<string, []>("op_605_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_605_cast_fp16 = reduce_mean(axes = var_605_axes_0, keep_dims = var_605_keep_dims_0, x = var_603_cast_fp16)[name = tensor<string, []>("op_605_cast_fp16")];
            tensor<fp16, []> var_606_to_fp16 = const()[name = tensor<string, []>("op_606_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_607_cast_fp16 = add(x = var_605_cast_fp16, y = var_606_to_fp16)[name = tensor<string, []>("op_607_cast_fp16")];
            tensor<fp32, []> var_608_epsilon_0 = const()[name = tensor<string, []>("op_608_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_608_cast_fp16 = rsqrt(epsilon = var_608_epsilon_0, x = var_607_cast_fp16)[name = tensor<string, []>("op_608_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_41_cast_fp16 = mul(x = x_39, y = var_608_cast_fp16)[name = tensor<string, []>("output_41_cast_fp16")];
            tensor<fp16, [2560]> var_612_to_fp16 = const()[name = tensor<string, []>("op_612_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940157760)))];
            tensor<fp16, [1, 256, 2560]> output_43_cast_fp16 = mul(x = output_41_cast_fp16, y = var_612_to_fp16)[name = tensor<string, []>("output_43_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_21 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_1_mlp_gate_proj_weight_palettized, x = output_43_cast_fp16)[name = tensor<string, []>("linear_11")];
            tensor<string, []> var_620_mode_0 = const()[name = tensor<string, []>("op_620_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_620 = gelu(mode = var_620_mode_0, x = input_21)[name = tensor<string, []>("op_620")];
            tensor<fp16, [1, 256, 10240]> var_622 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_1_mlp_up_proj_weight_palettized, x = output_43_cast_fp16)[name = tensor<string, []>("linear_12")];
            tensor<fp16, [1, 256, 10240]> input_23 = mul(x = var_620, y = var_622)[name = tensor<string, []>("input_23")];
            tensor<fp16, [1, 256, 2560]> x_43 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_1_mlp_down_proj_weight_palettized, x = input_23)[name = tensor<string, []>("linear_13")];
            tensor<fp16, []> var_22_promoted_11_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_11_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_628_cast_fp16 = pow(x = x_43, y = var_22_promoted_11_to_fp16)[name = tensor<string, []>("op_628_cast_fp16")];
            tensor<int32, [1]> var_630_axes_0 = const()[name = tensor<string, []>("op_630_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_630_keep_dims_0 = const()[name = tensor<string, []>("op_630_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_630_cast_fp16 = reduce_mean(axes = var_630_axes_0, keep_dims = var_630_keep_dims_0, x = var_628_cast_fp16)[name = tensor<string, []>("op_630_cast_fp16")];
            tensor<fp16, []> var_631_to_fp16 = const()[name = tensor<string, []>("op_631_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_632_cast_fp16 = add(x = var_630_cast_fp16, y = var_631_to_fp16)[name = tensor<string, []>("op_632_cast_fp16")];
            tensor<fp32, []> var_633_epsilon_0 = const()[name = tensor<string, []>("op_633_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_633_cast_fp16 = rsqrt(epsilon = var_633_epsilon_0, x = var_632_cast_fp16)[name = tensor<string, []>("op_633_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_45_cast_fp16 = mul(x = x_43, y = var_633_cast_fp16)[name = tensor<string, []>("output_45_cast_fp16")];
            tensor<fp16, [2560]> var_637_to_fp16 = const()[name = tensor<string, []>("op_637_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940162944)))];
            tensor<fp16, [1, 256, 2560]> output_47_cast_fp16 = mul(x = output_45_cast_fp16, y = var_637_to_fp16)[name = tensor<string, []>("output_47_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_47 = add(x = x_39, y = output_47_cast_fp16)[name = tensor<string, []>("x_47")];
            tensor<fp16, []> var_22_promoted_12_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_12_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_649_cast_fp16 = pow(x = x_47, y = var_22_promoted_12_to_fp16)[name = tensor<string, []>("op_649_cast_fp16")];
            tensor<int32, [1]> var_651_axes_0 = const()[name = tensor<string, []>("op_651_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_651_keep_dims_0 = const()[name = tensor<string, []>("op_651_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_651_cast_fp16 = reduce_mean(axes = var_651_axes_0, keep_dims = var_651_keep_dims_0, x = var_649_cast_fp16)[name = tensor<string, []>("op_651_cast_fp16")];
            tensor<fp16, []> var_652_to_fp16 = const()[name = tensor<string, []>("op_652_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_653_cast_fp16 = add(x = var_651_cast_fp16, y = var_652_to_fp16)[name = tensor<string, []>("op_653_cast_fp16")];
            tensor<fp32, []> var_654_epsilon_0 = const()[name = tensor<string, []>("op_654_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_654_cast_fp16 = rsqrt(epsilon = var_654_epsilon_0, x = var_653_cast_fp16)[name = tensor<string, []>("op_654_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_49_cast_fp16 = mul(x = x_47, y = var_654_cast_fp16)[name = tensor<string, []>("output_49_cast_fp16")];
            tensor<fp16, [2560]> var_658_to_fp16 = const()[name = tensor<string, []>("op_658_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940168128)))];
            tensor<fp16, [1, 256, 2560]> output_51_cast_fp16 = mul(x = output_49_cast_fp16, y = var_658_to_fp16)[name = tensor<string, []>("output_51_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_670 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_2_self_attn_q_proj_weight_palettized, x = output_51_cast_fp16)[name = tensor<string, []>("linear_14")];
            tensor<int32, [4]> var_671 = const()[name = tensor<string, []>("op_671"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_672 = reshape(shape = var_671, x = var_670)[name = tensor<string, []>("op_672")];
            tensor<int32, [4]> x_51_perm_0 = const()[name = tensor<string, []>("x_51_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_675 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_2_self_attn_k_proj_weight_palettized, x = output_51_cast_fp16)[name = tensor<string, []>("linear_15")];
            tensor<int32, [4]> var_676 = const()[name = tensor<string, []>("op_676"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_677 = reshape(shape = var_676, x = var_675)[name = tensor<string, []>("op_677")];
            tensor<int32, [4]> x_55_perm_0 = const()[name = tensor<string, []>("x_55_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_680 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_2_self_attn_v_proj_weight_palettized, x = output_51_cast_fp16)[name = tensor<string, []>("linear_16")];
            tensor<int32, [4]> var_681 = const()[name = tensor<string, []>("op_681"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_682 = reshape(shape = var_681, x = var_680)[name = tensor<string, []>("op_682")];
            tensor<int32, [4]> hidden_states_35_perm_0 = const()[name = tensor<string, []>("hidden_states_35_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_13_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_13_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_51 = transpose(perm = x_51_perm_0, x = var_672)[name = tensor<string, []>("transpose_127")];
            tensor<fp16, [1, 8, 256, 256]> var_686_cast_fp16 = pow(x = x_51, y = var_22_promoted_13_to_fp16)[name = tensor<string, []>("op_686_cast_fp16")];
            tensor<int32, [1]> var_688_axes_0 = const()[name = tensor<string, []>("op_688_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_688_keep_dims_0 = const()[name = tensor<string, []>("op_688_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_688_cast_fp16 = reduce_mean(axes = var_688_axes_0, keep_dims = var_688_keep_dims_0, x = var_686_cast_fp16)[name = tensor<string, []>("op_688_cast_fp16")];
            tensor<fp16, []> var_689_to_fp16 = const()[name = tensor<string, []>("op_689_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_690_cast_fp16 = add(x = var_688_cast_fp16, y = var_689_to_fp16)[name = tensor<string, []>("op_690_cast_fp16")];
            tensor<fp32, []> var_691_epsilon_0 = const()[name = tensor<string, []>("op_691_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_691_cast_fp16 = rsqrt(epsilon = var_691_epsilon_0, x = var_690_cast_fp16)[name = tensor<string, []>("op_691_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_53_cast_fp16 = mul(x = x_51, y = var_691_cast_fp16)[name = tensor<string, []>("output_53_cast_fp16")];
            tensor<fp16, [256]> var_695_to_fp16 = const()[name = tensor<string, []>("op_695_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940173312)))];
            tensor<fp16, [1, 8, 256, 256]> output_55_cast_fp16 = mul(x = output_53_cast_fp16, y = var_695_to_fp16)[name = tensor<string, []>("output_55_cast_fp16")];
            tensor<fp16, []> var_22_promoted_14_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_14_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_55 = transpose(perm = x_55_perm_0, x = var_677)[name = tensor<string, []>("transpose_126")];
            tensor<fp16, [1, 4, 256, 256]> var_700_cast_fp16 = pow(x = x_55, y = var_22_promoted_14_to_fp16)[name = tensor<string, []>("op_700_cast_fp16")];
            tensor<int32, [1]> var_702_axes_0 = const()[name = tensor<string, []>("op_702_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_702_keep_dims_0 = const()[name = tensor<string, []>("op_702_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_702_cast_fp16 = reduce_mean(axes = var_702_axes_0, keep_dims = var_702_keep_dims_0, x = var_700_cast_fp16)[name = tensor<string, []>("op_702_cast_fp16")];
            tensor<fp16, []> var_703_to_fp16 = const()[name = tensor<string, []>("op_703_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_704_cast_fp16 = add(x = var_702_cast_fp16, y = var_703_to_fp16)[name = tensor<string, []>("op_704_cast_fp16")];
            tensor<fp32, []> var_705_epsilon_0 = const()[name = tensor<string, []>("op_705_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_705_cast_fp16 = rsqrt(epsilon = var_705_epsilon_0, x = var_704_cast_fp16)[name = tensor<string, []>("op_705_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_57_cast_fp16 = mul(x = x_55, y = var_705_cast_fp16)[name = tensor<string, []>("output_57_cast_fp16")];
            tensor<fp16, [256]> var_709_to_fp16 = const()[name = tensor<string, []>("op_709_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940173888)))];
            tensor<fp16, [1, 4, 256, 256]> output_59_cast_fp16 = mul(x = output_57_cast_fp16, y = var_709_to_fp16)[name = tensor<string, []>("output_59_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_714 = mul(x = output_55_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_714")];
            tensor<int32, [4]> x1_9_begin_0 = const()[name = tensor<string, []>("x1_9_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_9_end_0 = const()[name = tensor<string, []>("x1_9_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_9_end_mask_0 = const()[name = tensor<string, []>("x1_9_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_9 = slice_by_index(begin = x1_9_begin_0, end = x1_9_end_0, end_mask = x1_9_end_mask_0, x = output_55_cast_fp16)[name = tensor<string, []>("x1_9")];
            tensor<int32, [4]> x2_9_begin_0 = const()[name = tensor<string, []>("x2_9_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_9_end_0 = const()[name = tensor<string, []>("x2_9_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_9_end_mask_0 = const()[name = tensor<string, []>("x2_9_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_9 = slice_by_index(begin = x2_9_begin_0, end = x2_9_end_0, end_mask = x2_9_end_mask_0, x = output_55_cast_fp16)[name = tensor<string, []>("x2_9")];
            tensor<fp16, []> const_81_promoted = const()[name = tensor<string, []>("const_81_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_725 = mul(x = x2_9, y = const_81_promoted)[name = tensor<string, []>("op_725")];
            tensor<bool, []> var_727_interleave_0 = const()[name = tensor<string, []>("op_727_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_727 = concat(axis = var_24, interleave = var_727_interleave_0, values = (var_725, x1_9))[name = tensor<string, []>("op_727")];
            tensor<fp16, [1, 8, 256, 256]> var_728 = mul(x = var_727, y = sin_7_palettized)[name = tensor<string, []>("op_728")];
            tensor<fp16, [1, 8, 256, 256]> query_5 = add(x = var_714, y = var_728)[name = tensor<string, []>("query_5")];
            tensor<fp16, [1, 4, 256, 256]> var_730 = mul(x = output_59_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_730")];
            tensor<int32, [4]> x1_11_begin_0 = const()[name = tensor<string, []>("x1_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_11_end_0 = const()[name = tensor<string, []>("x1_11_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_11_end_mask_0 = const()[name = tensor<string, []>("x1_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_11 = slice_by_index(begin = x1_11_begin_0, end = x1_11_end_0, end_mask = x1_11_end_mask_0, x = output_59_cast_fp16)[name = tensor<string, []>("x1_11")];
            tensor<int32, [4]> x2_11_begin_0 = const()[name = tensor<string, []>("x2_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_11_end_0 = const()[name = tensor<string, []>("x2_11_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_11_end_mask_0 = const()[name = tensor<string, []>("x2_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_11 = slice_by_index(begin = x2_11_begin_0, end = x2_11_end_0, end_mask = x2_11_end_mask_0, x = output_59_cast_fp16)[name = tensor<string, []>("x2_11")];
            tensor<fp16, []> const_84_promoted = const()[name = tensor<string, []>("const_84_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_741 = mul(x = x2_11, y = const_84_promoted)[name = tensor<string, []>("op_741")];
            tensor<bool, []> var_743_interleave_0 = const()[name = tensor<string, []>("op_743_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_743 = concat(axis = var_24, interleave = var_743_interleave_0, values = (var_741, x1_11))[name = tensor<string, []>("op_743")];
            tensor<fp16, [1, 4, 256, 256]> var_744 = mul(x = var_743, y = sin_7_palettized)[name = tensor<string, []>("op_744")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_31 = add(x = var_730, y = var_744)[name = tensor<string, []>("hidden_states_31")];
            tensor<int32, [1]> var_753_axes_0 = const()[name = tensor<string, []>("op_753_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_753 = expand_dims(axes = var_753_axes_0, x = hidden_states_31)[name = tensor<string, []>("op_753")];
            tensor<int32, [5]> hidden_states_33_reps_0 = const()[name = tensor<string, []>("hidden_states_33_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_33 = tile(reps = hidden_states_33_reps_0, x = var_753)[name = tensor<string, []>("hidden_states_33")];
            tensor<int32, [4]> var_761 = const()[name = tensor<string, []>("op_761"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_5 = reshape(shape = var_761, x = hidden_states_33)[name = tensor<string, []>("key_states_5")];
            tensor<int32, [1]> var_770_axes_0 = const()[name = tensor<string, []>("op_770_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_35 = transpose(perm = hidden_states_35_perm_0, x = var_682)[name = tensor<string, []>("transpose_125")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_770 = expand_dims(axes = var_770_axes_0, x = hidden_states_35)[name = tensor<string, []>("op_770")];
            tensor<int32, [5]> hidden_states_37_reps_0 = const()[name = tensor<string, []>("hidden_states_37_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_37 = tile(reps = hidden_states_37_reps_0, x = var_770)[name = tensor<string, []>("hidden_states_37")];
            tensor<int32, [4]> var_778 = const()[name = tensor<string, []>("op_778"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_5 = reshape(shape = var_778, x = hidden_states_37)[name = tensor<string, []>("value_states_5")];
            tensor<bool, []> var_781_transpose_x_1 = const()[name = tensor<string, []>("op_781_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_781_transpose_y_1 = const()[name = tensor<string, []>("op_781_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_781 = matmul(transpose_x = var_781_transpose_x_1, transpose_y = var_781_transpose_y_1, x = query_5, y = key_states_5)[name = tensor<string, []>("op_781")];
            tensor<fp16, []> var_782_to_fp16 = const()[name = tensor<string, []>("op_782_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_9_cast_fp16 = mul(x = var_781, y = var_782_to_fp16)[name = tensor<string, []>("attn_weights_9_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_25_cast_fp16 = add(x = attn_weights_9_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_25_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_790_cast_fp16 = softmax(axis = var_24, x = input_25_cast_fp16)[name = tensor<string, []>("op_790_cast_fp16")];
            tensor<bool, []> attn_output_9_transpose_x_0 = const()[name = tensor<string, []>("attn_output_9_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_9_transpose_y_0 = const()[name = tensor<string, []>("attn_output_9_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_9 = matmul(transpose_x = attn_output_9_transpose_x_0, transpose_y = attn_output_9_transpose_y_0, x = var_790_cast_fp16, y = value_states_5)[name = tensor<string, []>("attn_output_9")];
            tensor<int32, [4]> var_794_perm_0 = const()[name = tensor<string, []>("op_794_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_796 = const()[name = tensor<string, []>("op_796"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_794 = transpose(perm = var_794_perm_0, x = attn_output_9)[name = tensor<string, []>("transpose_124")];
            tensor<fp16, [1, 256, 2048]> var_797 = reshape(shape = var_796, x = var_794)[name = tensor<string, []>("op_797")];
            tensor<fp16, [1, 256, 2560]> x_59 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_2_self_attn_o_proj_weight_palettized, x = var_797)[name = tensor<string, []>("linear_17")];
            tensor<fp16, []> var_22_promoted_15_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_15_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_803_cast_fp16 = pow(x = x_59, y = var_22_promoted_15_to_fp16)[name = tensor<string, []>("op_803_cast_fp16")];
            tensor<int32, [1]> var_805_axes_0 = const()[name = tensor<string, []>("op_805_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_805_keep_dims_0 = const()[name = tensor<string, []>("op_805_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_805_cast_fp16 = reduce_mean(axes = var_805_axes_0, keep_dims = var_805_keep_dims_0, x = var_803_cast_fp16)[name = tensor<string, []>("op_805_cast_fp16")];
            tensor<fp16, []> var_806_to_fp16 = const()[name = tensor<string, []>("op_806_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_807_cast_fp16 = add(x = var_805_cast_fp16, y = var_806_to_fp16)[name = tensor<string, []>("op_807_cast_fp16")];
            tensor<fp32, []> var_808_epsilon_0 = const()[name = tensor<string, []>("op_808_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_808_cast_fp16 = rsqrt(epsilon = var_808_epsilon_0, x = var_807_cast_fp16)[name = tensor<string, []>("op_808_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_61_cast_fp16 = mul(x = x_59, y = var_808_cast_fp16)[name = tensor<string, []>("output_61_cast_fp16")];
            tensor<fp16, [2560]> var_812_to_fp16 = const()[name = tensor<string, []>("op_812_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940174464)))];
            tensor<fp16, [1, 256, 2560]> output_63_cast_fp16 = mul(x = output_61_cast_fp16, y = var_812_to_fp16)[name = tensor<string, []>("output_63_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_63 = add(x = x_47, y = output_63_cast_fp16)[name = tensor<string, []>("x_63")];
            tensor<fp16, []> var_22_promoted_16_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_16_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_818_cast_fp16 = pow(x = x_63, y = var_22_promoted_16_to_fp16)[name = tensor<string, []>("op_818_cast_fp16")];
            tensor<int32, [1]> var_820_axes_0 = const()[name = tensor<string, []>("op_820_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_820_keep_dims_0 = const()[name = tensor<string, []>("op_820_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_820_cast_fp16 = reduce_mean(axes = var_820_axes_0, keep_dims = var_820_keep_dims_0, x = var_818_cast_fp16)[name = tensor<string, []>("op_820_cast_fp16")];
            tensor<fp16, []> var_821_to_fp16 = const()[name = tensor<string, []>("op_821_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_822_cast_fp16 = add(x = var_820_cast_fp16, y = var_821_to_fp16)[name = tensor<string, []>("op_822_cast_fp16")];
            tensor<fp32, []> var_823_epsilon_0 = const()[name = tensor<string, []>("op_823_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_823_cast_fp16 = rsqrt(epsilon = var_823_epsilon_0, x = var_822_cast_fp16)[name = tensor<string, []>("op_823_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_65_cast_fp16 = mul(x = x_63, y = var_823_cast_fp16)[name = tensor<string, []>("output_65_cast_fp16")];
            tensor<fp16, [2560]> var_827_to_fp16 = const()[name = tensor<string, []>("op_827_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940179648)))];
            tensor<fp16, [1, 256, 2560]> output_67_cast_fp16 = mul(x = output_65_cast_fp16, y = var_827_to_fp16)[name = tensor<string, []>("output_67_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_33 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_2_mlp_gate_proj_weight_palettized, x = output_67_cast_fp16)[name = tensor<string, []>("linear_18")];
            tensor<string, []> var_835_mode_0 = const()[name = tensor<string, []>("op_835_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_835 = gelu(mode = var_835_mode_0, x = input_33)[name = tensor<string, []>("op_835")];
            tensor<fp16, [1, 256, 10240]> var_837 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_2_mlp_up_proj_weight_palettized, x = output_67_cast_fp16)[name = tensor<string, []>("linear_19")];
            tensor<fp16, [1, 256, 10240]> input_35 = mul(x = var_835, y = var_837)[name = tensor<string, []>("input_35")];
            tensor<fp16, [1, 256, 2560]> x_67 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_2_mlp_down_proj_weight_palettized, x = input_35)[name = tensor<string, []>("linear_20")];
            tensor<fp16, []> var_22_promoted_17_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_17_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_843_cast_fp16 = pow(x = x_67, y = var_22_promoted_17_to_fp16)[name = tensor<string, []>("op_843_cast_fp16")];
            tensor<int32, [1]> var_845_axes_0 = const()[name = tensor<string, []>("op_845_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_845_keep_dims_0 = const()[name = tensor<string, []>("op_845_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_845_cast_fp16 = reduce_mean(axes = var_845_axes_0, keep_dims = var_845_keep_dims_0, x = var_843_cast_fp16)[name = tensor<string, []>("op_845_cast_fp16")];
            tensor<fp16, []> var_846_to_fp16 = const()[name = tensor<string, []>("op_846_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_847_cast_fp16 = add(x = var_845_cast_fp16, y = var_846_to_fp16)[name = tensor<string, []>("op_847_cast_fp16")];
            tensor<fp32, []> var_848_epsilon_0 = const()[name = tensor<string, []>("op_848_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_848_cast_fp16 = rsqrt(epsilon = var_848_epsilon_0, x = var_847_cast_fp16)[name = tensor<string, []>("op_848_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_69_cast_fp16 = mul(x = x_67, y = var_848_cast_fp16)[name = tensor<string, []>("output_69_cast_fp16")];
            tensor<fp16, [2560]> var_852_to_fp16 = const()[name = tensor<string, []>("op_852_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940184832)))];
            tensor<fp16, [1, 256, 2560]> output_71_cast_fp16 = mul(x = output_69_cast_fp16, y = var_852_to_fp16)[name = tensor<string, []>("output_71_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_71 = add(x = x_63, y = output_71_cast_fp16)[name = tensor<string, []>("x_71")];
            tensor<fp16, []> var_22_promoted_18_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_18_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_864_cast_fp16 = pow(x = x_71, y = var_22_promoted_18_to_fp16)[name = tensor<string, []>("op_864_cast_fp16")];
            tensor<int32, [1]> var_866_axes_0 = const()[name = tensor<string, []>("op_866_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_866_keep_dims_0 = const()[name = tensor<string, []>("op_866_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_866_cast_fp16 = reduce_mean(axes = var_866_axes_0, keep_dims = var_866_keep_dims_0, x = var_864_cast_fp16)[name = tensor<string, []>("op_866_cast_fp16")];
            tensor<fp16, []> var_867_to_fp16 = const()[name = tensor<string, []>("op_867_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_868_cast_fp16 = add(x = var_866_cast_fp16, y = var_867_to_fp16)[name = tensor<string, []>("op_868_cast_fp16")];
            tensor<fp32, []> var_869_epsilon_0 = const()[name = tensor<string, []>("op_869_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_869_cast_fp16 = rsqrt(epsilon = var_869_epsilon_0, x = var_868_cast_fp16)[name = tensor<string, []>("op_869_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_73_cast_fp16 = mul(x = x_71, y = var_869_cast_fp16)[name = tensor<string, []>("output_73_cast_fp16")];
            tensor<fp16, [2560]> var_873_to_fp16 = const()[name = tensor<string, []>("op_873_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940190016)))];
            tensor<fp16, [1, 256, 2560]> output_75_cast_fp16 = mul(x = output_73_cast_fp16, y = var_873_to_fp16)[name = tensor<string, []>("output_75_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_885 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_3_self_attn_q_proj_weight_palettized, x = output_75_cast_fp16)[name = tensor<string, []>("linear_21")];
            tensor<int32, [4]> var_886 = const()[name = tensor<string, []>("op_886"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_887 = reshape(shape = var_886, x = var_885)[name = tensor<string, []>("op_887")];
            tensor<int32, [4]> x_75_perm_0 = const()[name = tensor<string, []>("x_75_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_890 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_3_self_attn_k_proj_weight_palettized, x = output_75_cast_fp16)[name = tensor<string, []>("linear_22")];
            tensor<int32, [4]> var_891 = const()[name = tensor<string, []>("op_891"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_892 = reshape(shape = var_891, x = var_890)[name = tensor<string, []>("op_892")];
            tensor<int32, [4]> x_79_perm_0 = const()[name = tensor<string, []>("x_79_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_895 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_3_self_attn_v_proj_weight_palettized, x = output_75_cast_fp16)[name = tensor<string, []>("linear_23")];
            tensor<int32, [4]> var_896 = const()[name = tensor<string, []>("op_896"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_897 = reshape(shape = var_896, x = var_895)[name = tensor<string, []>("op_897")];
            tensor<int32, [4]> hidden_states_49_perm_0 = const()[name = tensor<string, []>("hidden_states_49_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_19_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_19_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_75 = transpose(perm = x_75_perm_0, x = var_887)[name = tensor<string, []>("transpose_123")];
            tensor<fp16, [1, 8, 256, 256]> var_901_cast_fp16 = pow(x = x_75, y = var_22_promoted_19_to_fp16)[name = tensor<string, []>("op_901_cast_fp16")];
            tensor<int32, [1]> var_903_axes_0 = const()[name = tensor<string, []>("op_903_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_903_keep_dims_0 = const()[name = tensor<string, []>("op_903_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_903_cast_fp16 = reduce_mean(axes = var_903_axes_0, keep_dims = var_903_keep_dims_0, x = var_901_cast_fp16)[name = tensor<string, []>("op_903_cast_fp16")];
            tensor<fp16, []> var_904_to_fp16 = const()[name = tensor<string, []>("op_904_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_905_cast_fp16 = add(x = var_903_cast_fp16, y = var_904_to_fp16)[name = tensor<string, []>("op_905_cast_fp16")];
            tensor<fp32, []> var_906_epsilon_0 = const()[name = tensor<string, []>("op_906_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_906_cast_fp16 = rsqrt(epsilon = var_906_epsilon_0, x = var_905_cast_fp16)[name = tensor<string, []>("op_906_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_77_cast_fp16 = mul(x = x_75, y = var_906_cast_fp16)[name = tensor<string, []>("output_77_cast_fp16")];
            tensor<fp16, [256]> var_910_to_fp16 = const()[name = tensor<string, []>("op_910_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940195200)))];
            tensor<fp16, [1, 8, 256, 256]> output_79_cast_fp16 = mul(x = output_77_cast_fp16, y = var_910_to_fp16)[name = tensor<string, []>("output_79_cast_fp16")];
            tensor<fp16, []> var_22_promoted_20_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_20_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_79 = transpose(perm = x_79_perm_0, x = var_892)[name = tensor<string, []>("transpose_122")];
            tensor<fp16, [1, 4, 256, 256]> var_915_cast_fp16 = pow(x = x_79, y = var_22_promoted_20_to_fp16)[name = tensor<string, []>("op_915_cast_fp16")];
            tensor<int32, [1]> var_917_axes_0 = const()[name = tensor<string, []>("op_917_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_917_keep_dims_0 = const()[name = tensor<string, []>("op_917_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_917_cast_fp16 = reduce_mean(axes = var_917_axes_0, keep_dims = var_917_keep_dims_0, x = var_915_cast_fp16)[name = tensor<string, []>("op_917_cast_fp16")];
            tensor<fp16, []> var_918_to_fp16 = const()[name = tensor<string, []>("op_918_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_919_cast_fp16 = add(x = var_917_cast_fp16, y = var_918_to_fp16)[name = tensor<string, []>("op_919_cast_fp16")];
            tensor<fp32, []> var_920_epsilon_0 = const()[name = tensor<string, []>("op_920_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_920_cast_fp16 = rsqrt(epsilon = var_920_epsilon_0, x = var_919_cast_fp16)[name = tensor<string, []>("op_920_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_81_cast_fp16 = mul(x = x_79, y = var_920_cast_fp16)[name = tensor<string, []>("output_81_cast_fp16")];
            tensor<fp16, [256]> var_924_to_fp16 = const()[name = tensor<string, []>("op_924_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940195776)))];
            tensor<fp16, [1, 4, 256, 256]> output_83_cast_fp16 = mul(x = output_81_cast_fp16, y = var_924_to_fp16)[name = tensor<string, []>("output_83_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_929 = mul(x = output_79_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_929")];
            tensor<int32, [4]> x1_13_begin_0 = const()[name = tensor<string, []>("x1_13_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_13_end_0 = const()[name = tensor<string, []>("x1_13_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_13_end_mask_0 = const()[name = tensor<string, []>("x1_13_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_13 = slice_by_index(begin = x1_13_begin_0, end = x1_13_end_0, end_mask = x1_13_end_mask_0, x = output_79_cast_fp16)[name = tensor<string, []>("x1_13")];
            tensor<int32, [4]> x2_13_begin_0 = const()[name = tensor<string, []>("x2_13_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_13_end_0 = const()[name = tensor<string, []>("x2_13_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_13_end_mask_0 = const()[name = tensor<string, []>("x2_13_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_13 = slice_by_index(begin = x2_13_begin_0, end = x2_13_end_0, end_mask = x2_13_end_mask_0, x = output_79_cast_fp16)[name = tensor<string, []>("x2_13")];
            tensor<fp16, []> const_104_promoted = const()[name = tensor<string, []>("const_104_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_940 = mul(x = x2_13, y = const_104_promoted)[name = tensor<string, []>("op_940")];
            tensor<bool, []> var_942_interleave_0 = const()[name = tensor<string, []>("op_942_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_942 = concat(axis = var_24, interleave = var_942_interleave_0, values = (var_940, x1_13))[name = tensor<string, []>("op_942")];
            tensor<fp16, [1, 8, 256, 256]> var_943 = mul(x = var_942, y = sin_7_palettized)[name = tensor<string, []>("op_943")];
            tensor<fp16, [1, 8, 256, 256]> query_7 = add(x = var_929, y = var_943)[name = tensor<string, []>("query_7")];
            tensor<fp16, [1, 4, 256, 256]> var_945 = mul(x = output_83_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_945")];
            tensor<int32, [4]> x1_15_begin_0 = const()[name = tensor<string, []>("x1_15_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_15_end_0 = const()[name = tensor<string, []>("x1_15_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_15_end_mask_0 = const()[name = tensor<string, []>("x1_15_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_15 = slice_by_index(begin = x1_15_begin_0, end = x1_15_end_0, end_mask = x1_15_end_mask_0, x = output_83_cast_fp16)[name = tensor<string, []>("x1_15")];
            tensor<int32, [4]> x2_15_begin_0 = const()[name = tensor<string, []>("x2_15_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_15_end_0 = const()[name = tensor<string, []>("x2_15_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_15_end_mask_0 = const()[name = tensor<string, []>("x2_15_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_15 = slice_by_index(begin = x2_15_begin_0, end = x2_15_end_0, end_mask = x2_15_end_mask_0, x = output_83_cast_fp16)[name = tensor<string, []>("x2_15")];
            tensor<fp16, []> const_107_promoted = const()[name = tensor<string, []>("const_107_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_956 = mul(x = x2_15, y = const_107_promoted)[name = tensor<string, []>("op_956")];
            tensor<bool, []> var_958_interleave_0 = const()[name = tensor<string, []>("op_958_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_958 = concat(axis = var_24, interleave = var_958_interleave_0, values = (var_956, x1_15))[name = tensor<string, []>("op_958")];
            tensor<fp16, [1, 4, 256, 256]> var_959 = mul(x = var_958, y = sin_7_palettized)[name = tensor<string, []>("op_959")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_45 = add(x = var_945, y = var_959)[name = tensor<string, []>("hidden_states_45")];
            tensor<int32, [1]> var_968_axes_0 = const()[name = tensor<string, []>("op_968_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_968 = expand_dims(axes = var_968_axes_0, x = hidden_states_45)[name = tensor<string, []>("op_968")];
            tensor<int32, [5]> hidden_states_47_reps_0 = const()[name = tensor<string, []>("hidden_states_47_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_47 = tile(reps = hidden_states_47_reps_0, x = var_968)[name = tensor<string, []>("hidden_states_47")];
            tensor<int32, [4]> var_976 = const()[name = tensor<string, []>("op_976"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_7 = reshape(shape = var_976, x = hidden_states_47)[name = tensor<string, []>("key_states_7")];
            tensor<int32, [1]> var_985_axes_0 = const()[name = tensor<string, []>("op_985_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_49 = transpose(perm = hidden_states_49_perm_0, x = var_897)[name = tensor<string, []>("transpose_121")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_985 = expand_dims(axes = var_985_axes_0, x = hidden_states_49)[name = tensor<string, []>("op_985")];
            tensor<int32, [5]> hidden_states_51_reps_0 = const()[name = tensor<string, []>("hidden_states_51_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_51 = tile(reps = hidden_states_51_reps_0, x = var_985)[name = tensor<string, []>("hidden_states_51")];
            tensor<int32, [4]> var_993 = const()[name = tensor<string, []>("op_993"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_7 = reshape(shape = var_993, x = hidden_states_51)[name = tensor<string, []>("value_states_7")];
            tensor<bool, []> var_996_transpose_x_1 = const()[name = tensor<string, []>("op_996_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_996_transpose_y_1 = const()[name = tensor<string, []>("op_996_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_996 = matmul(transpose_x = var_996_transpose_x_1, transpose_y = var_996_transpose_y_1, x = query_7, y = key_states_7)[name = tensor<string, []>("op_996")];
            tensor<fp16, []> var_997_to_fp16 = const()[name = tensor<string, []>("op_997_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_13_cast_fp16 = mul(x = var_996, y = var_997_to_fp16)[name = tensor<string, []>("attn_weights_13_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_37_cast_fp16 = add(x = attn_weights_13_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_37_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_1005_cast_fp16 = softmax(axis = var_24, x = input_37_cast_fp16)[name = tensor<string, []>("op_1005_cast_fp16")];
            tensor<bool, []> attn_output_13_transpose_x_0 = const()[name = tensor<string, []>("attn_output_13_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_13_transpose_y_0 = const()[name = tensor<string, []>("attn_output_13_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_13 = matmul(transpose_x = attn_output_13_transpose_x_0, transpose_y = attn_output_13_transpose_y_0, x = var_1005_cast_fp16, y = value_states_7)[name = tensor<string, []>("attn_output_13")];
            tensor<int32, [4]> var_1009_perm_0 = const()[name = tensor<string, []>("op_1009_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1011 = const()[name = tensor<string, []>("op_1011"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_1009 = transpose(perm = var_1009_perm_0, x = attn_output_13)[name = tensor<string, []>("transpose_120")];
            tensor<fp16, [1, 256, 2048]> var_1012 = reshape(shape = var_1011, x = var_1009)[name = tensor<string, []>("op_1012")];
            tensor<fp16, [1, 256, 2560]> x_83 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_3_self_attn_o_proj_weight_palettized, x = var_1012)[name = tensor<string, []>("linear_24")];
            tensor<fp16, []> var_22_promoted_21_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_21_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_1018_cast_fp16 = pow(x = x_83, y = var_22_promoted_21_to_fp16)[name = tensor<string, []>("op_1018_cast_fp16")];
            tensor<int32, [1]> var_1020_axes_0 = const()[name = tensor<string, []>("op_1020_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1020_keep_dims_0 = const()[name = tensor<string, []>("op_1020_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_1020_cast_fp16 = reduce_mean(axes = var_1020_axes_0, keep_dims = var_1020_keep_dims_0, x = var_1018_cast_fp16)[name = tensor<string, []>("op_1020_cast_fp16")];
            tensor<fp16, []> var_1021_to_fp16 = const()[name = tensor<string, []>("op_1021_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_1022_cast_fp16 = add(x = var_1020_cast_fp16, y = var_1021_to_fp16)[name = tensor<string, []>("op_1022_cast_fp16")];
            tensor<fp32, []> var_1023_epsilon_0 = const()[name = tensor<string, []>("op_1023_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_1023_cast_fp16 = rsqrt(epsilon = var_1023_epsilon_0, x = var_1022_cast_fp16)[name = tensor<string, []>("op_1023_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_85_cast_fp16 = mul(x = x_83, y = var_1023_cast_fp16)[name = tensor<string, []>("output_85_cast_fp16")];
            tensor<fp16, [2560]> var_1027_to_fp16 = const()[name = tensor<string, []>("op_1027_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940196352)))];
            tensor<fp16, [1, 256, 2560]> output_87_cast_fp16 = mul(x = output_85_cast_fp16, y = var_1027_to_fp16)[name = tensor<string, []>("output_87_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_87 = add(x = x_71, y = output_87_cast_fp16)[name = tensor<string, []>("x_87")];
            tensor<fp16, []> var_22_promoted_22_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_22_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_1033_cast_fp16 = pow(x = x_87, y = var_22_promoted_22_to_fp16)[name = tensor<string, []>("op_1033_cast_fp16")];
            tensor<int32, [1]> var_1035_axes_0 = const()[name = tensor<string, []>("op_1035_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1035_keep_dims_0 = const()[name = tensor<string, []>("op_1035_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_1035_cast_fp16 = reduce_mean(axes = var_1035_axes_0, keep_dims = var_1035_keep_dims_0, x = var_1033_cast_fp16)[name = tensor<string, []>("op_1035_cast_fp16")];
            tensor<fp16, []> var_1036_to_fp16 = const()[name = tensor<string, []>("op_1036_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_1037_cast_fp16 = add(x = var_1035_cast_fp16, y = var_1036_to_fp16)[name = tensor<string, []>("op_1037_cast_fp16")];
            tensor<fp32, []> var_1038_epsilon_0 = const()[name = tensor<string, []>("op_1038_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_1038_cast_fp16 = rsqrt(epsilon = var_1038_epsilon_0, x = var_1037_cast_fp16)[name = tensor<string, []>("op_1038_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_89_cast_fp16 = mul(x = x_87, y = var_1038_cast_fp16)[name = tensor<string, []>("output_89_cast_fp16")];
            tensor<fp16, [2560]> var_1042_to_fp16 = const()[name = tensor<string, []>("op_1042_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940201536)))];
            tensor<fp16, [1, 256, 2560]> output_91_cast_fp16 = mul(x = output_89_cast_fp16, y = var_1042_to_fp16)[name = tensor<string, []>("output_91_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_45 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_3_mlp_gate_proj_weight_palettized, x = output_91_cast_fp16)[name = tensor<string, []>("linear_25")];
            tensor<string, []> var_1050_mode_0 = const()[name = tensor<string, []>("op_1050_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_1050 = gelu(mode = var_1050_mode_0, x = input_45)[name = tensor<string, []>("op_1050")];
            tensor<fp16, [1, 256, 10240]> var_1052 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_3_mlp_up_proj_weight_palettized, x = output_91_cast_fp16)[name = tensor<string, []>("linear_26")];
            tensor<fp16, [1, 256, 10240]> input_47 = mul(x = var_1050, y = var_1052)[name = tensor<string, []>("input_47")];
            tensor<fp16, [1, 256, 2560]> x_91 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_3_mlp_down_proj_weight_palettized, x = input_47)[name = tensor<string, []>("linear_27")];
            tensor<fp16, []> var_22_promoted_23_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_23_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_1058_cast_fp16 = pow(x = x_91, y = var_22_promoted_23_to_fp16)[name = tensor<string, []>("op_1058_cast_fp16")];
            tensor<int32, [1]> var_1060_axes_0 = const()[name = tensor<string, []>("op_1060_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1060_keep_dims_0 = const()[name = tensor<string, []>("op_1060_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_1060_cast_fp16 = reduce_mean(axes = var_1060_axes_0, keep_dims = var_1060_keep_dims_0, x = var_1058_cast_fp16)[name = tensor<string, []>("op_1060_cast_fp16")];
            tensor<fp16, []> var_1061_to_fp16 = const()[name = tensor<string, []>("op_1061_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_1062_cast_fp16 = add(x = var_1060_cast_fp16, y = var_1061_to_fp16)[name = tensor<string, []>("op_1062_cast_fp16")];
            tensor<fp32, []> var_1063_epsilon_0 = const()[name = tensor<string, []>("op_1063_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_1063_cast_fp16 = rsqrt(epsilon = var_1063_epsilon_0, x = var_1062_cast_fp16)[name = tensor<string, []>("op_1063_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_93_cast_fp16 = mul(x = x_91, y = var_1063_cast_fp16)[name = tensor<string, []>("output_93_cast_fp16")];
            tensor<fp16, [2560]> var_1067_to_fp16 = const()[name = tensor<string, []>("op_1067_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940206720)))];
            tensor<fp16, [1, 256, 2560]> output_95_cast_fp16 = mul(x = output_93_cast_fp16, y = var_1067_to_fp16)[name = tensor<string, []>("output_95_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_95 = add(x = x_87, y = output_95_cast_fp16)[name = tensor<string, []>("x_95")];
            tensor<fp16, []> var_22_promoted_24_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_24_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_1079_cast_fp16 = pow(x = x_95, y = var_22_promoted_24_to_fp16)[name = tensor<string, []>("op_1079_cast_fp16")];
            tensor<int32, [1]> var_1081_axes_0 = const()[name = tensor<string, []>("op_1081_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1081_keep_dims_0 = const()[name = tensor<string, []>("op_1081_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_1081_cast_fp16 = reduce_mean(axes = var_1081_axes_0, keep_dims = var_1081_keep_dims_0, x = var_1079_cast_fp16)[name = tensor<string, []>("op_1081_cast_fp16")];
            tensor<fp16, []> var_1082_to_fp16 = const()[name = tensor<string, []>("op_1082_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_1083_cast_fp16 = add(x = var_1081_cast_fp16, y = var_1082_to_fp16)[name = tensor<string, []>("op_1083_cast_fp16")];
            tensor<fp32, []> var_1084_epsilon_0 = const()[name = tensor<string, []>("op_1084_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_1084_cast_fp16 = rsqrt(epsilon = var_1084_epsilon_0, x = var_1083_cast_fp16)[name = tensor<string, []>("op_1084_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_97_cast_fp16 = mul(x = x_95, y = var_1084_cast_fp16)[name = tensor<string, []>("output_97_cast_fp16")];
            tensor<fp16, [2560]> var_1088_to_fp16 = const()[name = tensor<string, []>("op_1088_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940211904)))];
            tensor<fp16, [1, 256, 2560]> output_99_cast_fp16 = mul(x = output_97_cast_fp16, y = var_1088_to_fp16)[name = tensor<string, []>("output_99_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_1100 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_4_self_attn_q_proj_weight_palettized, x = output_99_cast_fp16)[name = tensor<string, []>("linear_28")];
            tensor<int32, [4]> var_1101 = const()[name = tensor<string, []>("op_1101"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_1102 = reshape(shape = var_1101, x = var_1100)[name = tensor<string, []>("op_1102")];
            tensor<int32, [4]> x_99_perm_0 = const()[name = tensor<string, []>("x_99_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_1105 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_4_self_attn_k_proj_weight_palettized, x = output_99_cast_fp16)[name = tensor<string, []>("linear_29")];
            tensor<int32, [4]> var_1106 = const()[name = tensor<string, []>("op_1106"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_1107 = reshape(shape = var_1106, x = var_1105)[name = tensor<string, []>("op_1107")];
            tensor<int32, [4]> x_103_perm_0 = const()[name = tensor<string, []>("x_103_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_1110 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_4_self_attn_v_proj_weight_palettized, x = output_99_cast_fp16)[name = tensor<string, []>("linear_30")];
            tensor<int32, [4]> var_1111 = const()[name = tensor<string, []>("op_1111"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_1112 = reshape(shape = var_1111, x = var_1110)[name = tensor<string, []>("op_1112")];
            tensor<int32, [4]> hidden_states_63_perm_0 = const()[name = tensor<string, []>("hidden_states_63_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_25_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_25_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_99 = transpose(perm = x_99_perm_0, x = var_1102)[name = tensor<string, []>("transpose_119")];
            tensor<fp16, [1, 8, 256, 256]> var_1116_cast_fp16 = pow(x = x_99, y = var_22_promoted_25_to_fp16)[name = tensor<string, []>("op_1116_cast_fp16")];
            tensor<int32, [1]> var_1118_axes_0 = const()[name = tensor<string, []>("op_1118_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1118_keep_dims_0 = const()[name = tensor<string, []>("op_1118_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_1118_cast_fp16 = reduce_mean(axes = var_1118_axes_0, keep_dims = var_1118_keep_dims_0, x = var_1116_cast_fp16)[name = tensor<string, []>("op_1118_cast_fp16")];
            tensor<fp16, []> var_1119_to_fp16 = const()[name = tensor<string, []>("op_1119_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_1120_cast_fp16 = add(x = var_1118_cast_fp16, y = var_1119_to_fp16)[name = tensor<string, []>("op_1120_cast_fp16")];
            tensor<fp32, []> var_1121_epsilon_0 = const()[name = tensor<string, []>("op_1121_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_1121_cast_fp16 = rsqrt(epsilon = var_1121_epsilon_0, x = var_1120_cast_fp16)[name = tensor<string, []>("op_1121_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_101_cast_fp16 = mul(x = x_99, y = var_1121_cast_fp16)[name = tensor<string, []>("output_101_cast_fp16")];
            tensor<fp16, [256]> var_1125_to_fp16 = const()[name = tensor<string, []>("op_1125_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940217088)))];
            tensor<fp16, [1, 8, 256, 256]> output_103_cast_fp16 = mul(x = output_101_cast_fp16, y = var_1125_to_fp16)[name = tensor<string, []>("output_103_cast_fp16")];
            tensor<fp16, []> var_22_promoted_26_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_26_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_103 = transpose(perm = x_103_perm_0, x = var_1107)[name = tensor<string, []>("transpose_118")];
            tensor<fp16, [1, 4, 256, 256]> var_1130_cast_fp16 = pow(x = x_103, y = var_22_promoted_26_to_fp16)[name = tensor<string, []>("op_1130_cast_fp16")];
            tensor<int32, [1]> var_1132_axes_0 = const()[name = tensor<string, []>("op_1132_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1132_keep_dims_0 = const()[name = tensor<string, []>("op_1132_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_1132_cast_fp16 = reduce_mean(axes = var_1132_axes_0, keep_dims = var_1132_keep_dims_0, x = var_1130_cast_fp16)[name = tensor<string, []>("op_1132_cast_fp16")];
            tensor<fp16, []> var_1133_to_fp16 = const()[name = tensor<string, []>("op_1133_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_1134_cast_fp16 = add(x = var_1132_cast_fp16, y = var_1133_to_fp16)[name = tensor<string, []>("op_1134_cast_fp16")];
            tensor<fp32, []> var_1135_epsilon_0 = const()[name = tensor<string, []>("op_1135_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_1135_cast_fp16 = rsqrt(epsilon = var_1135_epsilon_0, x = var_1134_cast_fp16)[name = tensor<string, []>("op_1135_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_105_cast_fp16 = mul(x = x_103, y = var_1135_cast_fp16)[name = tensor<string, []>("output_105_cast_fp16")];
            tensor<fp16, [256]> var_1139_to_fp16 = const()[name = tensor<string, []>("op_1139_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940217664)))];
            tensor<fp16, [1, 4, 256, 256]> output_107_cast_fp16 = mul(x = output_105_cast_fp16, y = var_1139_to_fp16)[name = tensor<string, []>("output_107_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_1144 = mul(x = output_103_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_1144")];
            tensor<int32, [4]> x1_17_begin_0 = const()[name = tensor<string, []>("x1_17_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_17_end_0 = const()[name = tensor<string, []>("x1_17_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_17_end_mask_0 = const()[name = tensor<string, []>("x1_17_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_17 = slice_by_index(begin = x1_17_begin_0, end = x1_17_end_0, end_mask = x1_17_end_mask_0, x = output_103_cast_fp16)[name = tensor<string, []>("x1_17")];
            tensor<int32, [4]> x2_17_begin_0 = const()[name = tensor<string, []>("x2_17_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_17_end_0 = const()[name = tensor<string, []>("x2_17_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_17_end_mask_0 = const()[name = tensor<string, []>("x2_17_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_17 = slice_by_index(begin = x2_17_begin_0, end = x2_17_end_0, end_mask = x2_17_end_mask_0, x = output_103_cast_fp16)[name = tensor<string, []>("x2_17")];
            tensor<fp16, []> const_127_promoted = const()[name = tensor<string, []>("const_127_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_1155 = mul(x = x2_17, y = const_127_promoted)[name = tensor<string, []>("op_1155")];
            tensor<bool, []> var_1157_interleave_0 = const()[name = tensor<string, []>("op_1157_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_1157 = concat(axis = var_24, interleave = var_1157_interleave_0, values = (var_1155, x1_17))[name = tensor<string, []>("op_1157")];
            tensor<fp16, [1, 8, 256, 256]> var_1158 = mul(x = var_1157, y = sin_7_palettized)[name = tensor<string, []>("op_1158")];
            tensor<fp16, [1, 8, 256, 256]> query_9 = add(x = var_1144, y = var_1158)[name = tensor<string, []>("query_9")];
            tensor<fp16, [1, 4, 256, 256]> var_1160 = mul(x = output_107_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_1160")];
            tensor<int32, [4]> x1_19_begin_0 = const()[name = tensor<string, []>("x1_19_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_19_end_0 = const()[name = tensor<string, []>("x1_19_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_19_end_mask_0 = const()[name = tensor<string, []>("x1_19_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_19 = slice_by_index(begin = x1_19_begin_0, end = x1_19_end_0, end_mask = x1_19_end_mask_0, x = output_107_cast_fp16)[name = tensor<string, []>("x1_19")];
            tensor<int32, [4]> x2_19_begin_0 = const()[name = tensor<string, []>("x2_19_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_19_end_0 = const()[name = tensor<string, []>("x2_19_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_19_end_mask_0 = const()[name = tensor<string, []>("x2_19_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_19 = slice_by_index(begin = x2_19_begin_0, end = x2_19_end_0, end_mask = x2_19_end_mask_0, x = output_107_cast_fp16)[name = tensor<string, []>("x2_19")];
            tensor<fp16, []> const_130_promoted = const()[name = tensor<string, []>("const_130_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_1171 = mul(x = x2_19, y = const_130_promoted)[name = tensor<string, []>("op_1171")];
            tensor<bool, []> var_1173_interleave_0 = const()[name = tensor<string, []>("op_1173_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_1173 = concat(axis = var_24, interleave = var_1173_interleave_0, values = (var_1171, x1_19))[name = tensor<string, []>("op_1173")];
            tensor<fp16, [1, 4, 256, 256]> var_1174 = mul(x = var_1173, y = sin_7_palettized)[name = tensor<string, []>("op_1174")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_59 = add(x = var_1160, y = var_1174)[name = tensor<string, []>("hidden_states_59")];
            tensor<int32, [1]> var_1183_axes_0 = const()[name = tensor<string, []>("op_1183_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_1183 = expand_dims(axes = var_1183_axes_0, x = hidden_states_59)[name = tensor<string, []>("op_1183")];
            tensor<int32, [5]> hidden_states_61_reps_0 = const()[name = tensor<string, []>("hidden_states_61_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_61 = tile(reps = hidden_states_61_reps_0, x = var_1183)[name = tensor<string, []>("hidden_states_61")];
            tensor<int32, [4]> var_1191 = const()[name = tensor<string, []>("op_1191"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_9 = reshape(shape = var_1191, x = hidden_states_61)[name = tensor<string, []>("key_states_9")];
            tensor<int32, [1]> var_1200_axes_0 = const()[name = tensor<string, []>("op_1200_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_63 = transpose(perm = hidden_states_63_perm_0, x = var_1112)[name = tensor<string, []>("transpose_117")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_1200 = expand_dims(axes = var_1200_axes_0, x = hidden_states_63)[name = tensor<string, []>("op_1200")];
            tensor<int32, [5]> hidden_states_65_reps_0 = const()[name = tensor<string, []>("hidden_states_65_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_65 = tile(reps = hidden_states_65_reps_0, x = var_1200)[name = tensor<string, []>("hidden_states_65")];
            tensor<int32, [4]> var_1208 = const()[name = tensor<string, []>("op_1208"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_9 = reshape(shape = var_1208, x = hidden_states_65)[name = tensor<string, []>("value_states_9")];
            tensor<bool, []> var_1211_transpose_x_1 = const()[name = tensor<string, []>("op_1211_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1211_transpose_y_1 = const()[name = tensor<string, []>("op_1211_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_1211 = matmul(transpose_x = var_1211_transpose_x_1, transpose_y = var_1211_transpose_y_1, x = query_9, y = key_states_9)[name = tensor<string, []>("op_1211")];
            tensor<fp16, []> var_1212_to_fp16 = const()[name = tensor<string, []>("op_1212_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_17_cast_fp16 = mul(x = var_1211, y = var_1212_to_fp16)[name = tensor<string, []>("attn_weights_17_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_49_cast_fp16 = add(x = attn_weights_17_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_49_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_1220_cast_fp16 = softmax(axis = var_24, x = input_49_cast_fp16)[name = tensor<string, []>("op_1220_cast_fp16")];
            tensor<bool, []> attn_output_17_transpose_x_0 = const()[name = tensor<string, []>("attn_output_17_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_17_transpose_y_0 = const()[name = tensor<string, []>("attn_output_17_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_17 = matmul(transpose_x = attn_output_17_transpose_x_0, transpose_y = attn_output_17_transpose_y_0, x = var_1220_cast_fp16, y = value_states_9)[name = tensor<string, []>("attn_output_17")];
            tensor<int32, [4]> var_1224_perm_0 = const()[name = tensor<string, []>("op_1224_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1226 = const()[name = tensor<string, []>("op_1226"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_1224 = transpose(perm = var_1224_perm_0, x = attn_output_17)[name = tensor<string, []>("transpose_116")];
            tensor<fp16, [1, 256, 2048]> var_1227 = reshape(shape = var_1226, x = var_1224)[name = tensor<string, []>("op_1227")];
            tensor<fp16, [1, 256, 2560]> x_107 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_4_self_attn_o_proj_weight_palettized, x = var_1227)[name = tensor<string, []>("linear_31")];
            tensor<fp16, []> var_22_promoted_27_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_27_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_1233_cast_fp16 = pow(x = x_107, y = var_22_promoted_27_to_fp16)[name = tensor<string, []>("op_1233_cast_fp16")];
            tensor<int32, [1]> var_1235_axes_0 = const()[name = tensor<string, []>("op_1235_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1235_keep_dims_0 = const()[name = tensor<string, []>("op_1235_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_1235_cast_fp16 = reduce_mean(axes = var_1235_axes_0, keep_dims = var_1235_keep_dims_0, x = var_1233_cast_fp16)[name = tensor<string, []>("op_1235_cast_fp16")];
            tensor<fp16, []> var_1236_to_fp16 = const()[name = tensor<string, []>("op_1236_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_1237_cast_fp16 = add(x = var_1235_cast_fp16, y = var_1236_to_fp16)[name = tensor<string, []>("op_1237_cast_fp16")];
            tensor<fp32, []> var_1238_epsilon_0 = const()[name = tensor<string, []>("op_1238_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_1238_cast_fp16 = rsqrt(epsilon = var_1238_epsilon_0, x = var_1237_cast_fp16)[name = tensor<string, []>("op_1238_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_109_cast_fp16 = mul(x = x_107, y = var_1238_cast_fp16)[name = tensor<string, []>("output_109_cast_fp16")];
            tensor<fp16, [2560]> var_1242_to_fp16 = const()[name = tensor<string, []>("op_1242_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940218240)))];
            tensor<fp16, [1, 256, 2560]> output_111_cast_fp16 = mul(x = output_109_cast_fp16, y = var_1242_to_fp16)[name = tensor<string, []>("output_111_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_111 = add(x = x_95, y = output_111_cast_fp16)[name = tensor<string, []>("x_111")];
            tensor<fp16, []> var_22_promoted_28_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_28_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_1248_cast_fp16 = pow(x = x_111, y = var_22_promoted_28_to_fp16)[name = tensor<string, []>("op_1248_cast_fp16")];
            tensor<int32, [1]> var_1250_axes_0 = const()[name = tensor<string, []>("op_1250_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1250_keep_dims_0 = const()[name = tensor<string, []>("op_1250_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_1250_cast_fp16 = reduce_mean(axes = var_1250_axes_0, keep_dims = var_1250_keep_dims_0, x = var_1248_cast_fp16)[name = tensor<string, []>("op_1250_cast_fp16")];
            tensor<fp16, []> var_1251_to_fp16 = const()[name = tensor<string, []>("op_1251_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_1252_cast_fp16 = add(x = var_1250_cast_fp16, y = var_1251_to_fp16)[name = tensor<string, []>("op_1252_cast_fp16")];
            tensor<fp32, []> var_1253_epsilon_0 = const()[name = tensor<string, []>("op_1253_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_1253_cast_fp16 = rsqrt(epsilon = var_1253_epsilon_0, x = var_1252_cast_fp16)[name = tensor<string, []>("op_1253_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_113_cast_fp16 = mul(x = x_111, y = var_1253_cast_fp16)[name = tensor<string, []>("output_113_cast_fp16")];
            tensor<fp16, [2560]> var_1257_to_fp16 = const()[name = tensor<string, []>("op_1257_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940223424)))];
            tensor<fp16, [1, 256, 2560]> output_115_cast_fp16 = mul(x = output_113_cast_fp16, y = var_1257_to_fp16)[name = tensor<string, []>("output_115_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_57 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_4_mlp_gate_proj_weight_palettized, x = output_115_cast_fp16)[name = tensor<string, []>("linear_32")];
            tensor<string, []> var_1265_mode_0 = const()[name = tensor<string, []>("op_1265_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_1265 = gelu(mode = var_1265_mode_0, x = input_57)[name = tensor<string, []>("op_1265")];
            tensor<fp16, [1, 256, 10240]> var_1267 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_4_mlp_up_proj_weight_palettized, x = output_115_cast_fp16)[name = tensor<string, []>("linear_33")];
            tensor<fp16, [1, 256, 10240]> input_59 = mul(x = var_1265, y = var_1267)[name = tensor<string, []>("input_59")];
            tensor<fp16, [1, 256, 2560]> x_115 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_4_mlp_down_proj_weight_palettized, x = input_59)[name = tensor<string, []>("linear_34")];
            tensor<fp16, []> var_22_promoted_29_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_29_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_1273_cast_fp16 = pow(x = x_115, y = var_22_promoted_29_to_fp16)[name = tensor<string, []>("op_1273_cast_fp16")];
            tensor<int32, [1]> var_1275_axes_0 = const()[name = tensor<string, []>("op_1275_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1275_keep_dims_0 = const()[name = tensor<string, []>("op_1275_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_1275_cast_fp16 = reduce_mean(axes = var_1275_axes_0, keep_dims = var_1275_keep_dims_0, x = var_1273_cast_fp16)[name = tensor<string, []>("op_1275_cast_fp16")];
            tensor<fp16, []> var_1276_to_fp16 = const()[name = tensor<string, []>("op_1276_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_1277_cast_fp16 = add(x = var_1275_cast_fp16, y = var_1276_to_fp16)[name = tensor<string, []>("op_1277_cast_fp16")];
            tensor<fp32, []> var_1278_epsilon_0 = const()[name = tensor<string, []>("op_1278_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_1278_cast_fp16 = rsqrt(epsilon = var_1278_epsilon_0, x = var_1277_cast_fp16)[name = tensor<string, []>("op_1278_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_117_cast_fp16 = mul(x = x_115, y = var_1278_cast_fp16)[name = tensor<string, []>("output_117_cast_fp16")];
            tensor<fp16, [2560]> var_1282_to_fp16 = const()[name = tensor<string, []>("op_1282_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940228608)))];
            tensor<fp16, [1, 256, 2560]> output_119_cast_fp16 = mul(x = output_117_cast_fp16, y = var_1282_to_fp16)[name = tensor<string, []>("output_119_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_119 = add(x = x_111, y = output_119_cast_fp16)[name = tensor<string, []>("x_119")];
            tensor<fp16, []> var_22_promoted_30_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_30_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_1294_cast_fp16 = pow(x = x_119, y = var_22_promoted_30_to_fp16)[name = tensor<string, []>("op_1294_cast_fp16")];
            tensor<int32, [1]> var_1296_axes_0 = const()[name = tensor<string, []>("op_1296_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1296_keep_dims_0 = const()[name = tensor<string, []>("op_1296_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_1296_cast_fp16 = reduce_mean(axes = var_1296_axes_0, keep_dims = var_1296_keep_dims_0, x = var_1294_cast_fp16)[name = tensor<string, []>("op_1296_cast_fp16")];
            tensor<fp16, []> var_1297_to_fp16 = const()[name = tensor<string, []>("op_1297_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_1298_cast_fp16 = add(x = var_1296_cast_fp16, y = var_1297_to_fp16)[name = tensor<string, []>("op_1298_cast_fp16")];
            tensor<fp32, []> var_1299_epsilon_0 = const()[name = tensor<string, []>("op_1299_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_1299_cast_fp16 = rsqrt(epsilon = var_1299_epsilon_0, x = var_1298_cast_fp16)[name = tensor<string, []>("op_1299_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_121_cast_fp16 = mul(x = x_119, y = var_1299_cast_fp16)[name = tensor<string, []>("output_121_cast_fp16")];
            tensor<fp16, [2560]> var_1303_to_fp16 = const()[name = tensor<string, []>("op_1303_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940233792)))];
            tensor<fp16, [1, 256, 2560]> output_123_cast_fp16 = mul(x = output_121_cast_fp16, y = var_1303_to_fp16)[name = tensor<string, []>("output_123_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_1315 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_5_self_attn_q_proj_weight_palettized, x = output_123_cast_fp16)[name = tensor<string, []>("linear_35")];
            tensor<int32, [4]> var_1316 = const()[name = tensor<string, []>("op_1316"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_1317 = reshape(shape = var_1316, x = var_1315)[name = tensor<string, []>("op_1317")];
            tensor<int32, [4]> x_123_perm_0 = const()[name = tensor<string, []>("x_123_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_1320 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_5_self_attn_k_proj_weight_palettized, x = output_123_cast_fp16)[name = tensor<string, []>("linear_36")];
            tensor<int32, [4]> var_1321 = const()[name = tensor<string, []>("op_1321"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_1322 = reshape(shape = var_1321, x = var_1320)[name = tensor<string, []>("op_1322")];
            tensor<int32, [4]> x_127_perm_0 = const()[name = tensor<string, []>("x_127_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_1325 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_5_self_attn_v_proj_weight_palettized, x = output_123_cast_fp16)[name = tensor<string, []>("linear_37")];
            tensor<int32, [4]> var_1326 = const()[name = tensor<string, []>("op_1326"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_1327 = reshape(shape = var_1326, x = var_1325)[name = tensor<string, []>("op_1327")];
            tensor<int32, [4]> hidden_states_77_perm_0 = const()[name = tensor<string, []>("hidden_states_77_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_31_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_31_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_123 = transpose(perm = x_123_perm_0, x = var_1317)[name = tensor<string, []>("transpose_115")];
            tensor<fp16, [1, 8, 256, 256]> var_1331_cast_fp16 = pow(x = x_123, y = var_22_promoted_31_to_fp16)[name = tensor<string, []>("op_1331_cast_fp16")];
            tensor<int32, [1]> var_1333_axes_0 = const()[name = tensor<string, []>("op_1333_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1333_keep_dims_0 = const()[name = tensor<string, []>("op_1333_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_1333_cast_fp16 = reduce_mean(axes = var_1333_axes_0, keep_dims = var_1333_keep_dims_0, x = var_1331_cast_fp16)[name = tensor<string, []>("op_1333_cast_fp16")];
            tensor<fp16, []> var_1334_to_fp16 = const()[name = tensor<string, []>("op_1334_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_1335_cast_fp16 = add(x = var_1333_cast_fp16, y = var_1334_to_fp16)[name = tensor<string, []>("op_1335_cast_fp16")];
            tensor<fp32, []> var_1336_epsilon_0 = const()[name = tensor<string, []>("op_1336_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_1336_cast_fp16 = rsqrt(epsilon = var_1336_epsilon_0, x = var_1335_cast_fp16)[name = tensor<string, []>("op_1336_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_125_cast_fp16 = mul(x = x_123, y = var_1336_cast_fp16)[name = tensor<string, []>("output_125_cast_fp16")];
            tensor<fp16, [256]> var_1340_to_fp16 = const()[name = tensor<string, []>("op_1340_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940238976)))];
            tensor<fp16, [1, 8, 256, 256]> output_127_cast_fp16 = mul(x = output_125_cast_fp16, y = var_1340_to_fp16)[name = tensor<string, []>("output_127_cast_fp16")];
            tensor<fp16, []> var_22_promoted_32_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_32_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_127 = transpose(perm = x_127_perm_0, x = var_1322)[name = tensor<string, []>("transpose_114")];
            tensor<fp16, [1, 4, 256, 256]> var_1345_cast_fp16 = pow(x = x_127, y = var_22_promoted_32_to_fp16)[name = tensor<string, []>("op_1345_cast_fp16")];
            tensor<int32, [1]> var_1347_axes_0 = const()[name = tensor<string, []>("op_1347_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1347_keep_dims_0 = const()[name = tensor<string, []>("op_1347_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_1347_cast_fp16 = reduce_mean(axes = var_1347_axes_0, keep_dims = var_1347_keep_dims_0, x = var_1345_cast_fp16)[name = tensor<string, []>("op_1347_cast_fp16")];
            tensor<fp16, []> var_1348_to_fp16 = const()[name = tensor<string, []>("op_1348_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_1349_cast_fp16 = add(x = var_1347_cast_fp16, y = var_1348_to_fp16)[name = tensor<string, []>("op_1349_cast_fp16")];
            tensor<fp32, []> var_1350_epsilon_0 = const()[name = tensor<string, []>("op_1350_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_1350_cast_fp16 = rsqrt(epsilon = var_1350_epsilon_0, x = var_1349_cast_fp16)[name = tensor<string, []>("op_1350_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_129_cast_fp16 = mul(x = x_127, y = var_1350_cast_fp16)[name = tensor<string, []>("output_129_cast_fp16")];
            tensor<fp16, [256]> var_1354_to_fp16 = const()[name = tensor<string, []>("op_1354_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940239552)))];
            tensor<fp16, [1, 4, 256, 256]> output_131_cast_fp16 = mul(x = output_129_cast_fp16, y = var_1354_to_fp16)[name = tensor<string, []>("output_131_cast_fp16")];
            tensor<fp16, [1, 1, 256, 256]> cos_19_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [32768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940240128))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940272960))), name = tensor<string, []>("cos_19_palettized"), shape = tensor<uint32, [4]>([1, 1, 256, 256])];
            tensor<fp16, [1, 1, 256, 256]> sin_19_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [32768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940273088))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940305920))), name = tensor<string, []>("sin_19_palettized"), shape = tensor<uint32, [4]>([1, 1, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> var_1359 = mul(x = output_127_cast_fp16, y = cos_19_palettized)[name = tensor<string, []>("op_1359")];
            tensor<int32, [4]> x1_21_begin_0 = const()[name = tensor<string, []>("x1_21_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_21_end_0 = const()[name = tensor<string, []>("x1_21_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_21_end_mask_0 = const()[name = tensor<string, []>("x1_21_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_21 = slice_by_index(begin = x1_21_begin_0, end = x1_21_end_0, end_mask = x1_21_end_mask_0, x = output_127_cast_fp16)[name = tensor<string, []>("x1_21")];
            tensor<int32, [4]> x2_21_begin_0 = const()[name = tensor<string, []>("x2_21_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_21_end_0 = const()[name = tensor<string, []>("x2_21_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_21_end_mask_0 = const()[name = tensor<string, []>("x2_21_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_21 = slice_by_index(begin = x2_21_begin_0, end = x2_21_end_0, end_mask = x2_21_end_mask_0, x = output_127_cast_fp16)[name = tensor<string, []>("x2_21")];
            tensor<fp16, []> const_150_promoted = const()[name = tensor<string, []>("const_150_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_1370 = mul(x = x2_21, y = const_150_promoted)[name = tensor<string, []>("op_1370")];
            tensor<bool, []> var_1372_interleave_0 = const()[name = tensor<string, []>("op_1372_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_1372 = concat(axis = var_24, interleave = var_1372_interleave_0, values = (var_1370, x1_21))[name = tensor<string, []>("op_1372")];
            tensor<fp16, [1, 8, 256, 256]> var_1373 = mul(x = var_1372, y = sin_19_palettized)[name = tensor<string, []>("op_1373")];
            tensor<fp16, [1, 8, 256, 256]> query_11 = add(x = var_1359, y = var_1373)[name = tensor<string, []>("query_11")];
            tensor<fp16, [1, 4, 256, 256]> var_1375 = mul(x = output_131_cast_fp16, y = cos_19_palettized)[name = tensor<string, []>("op_1375")];
            tensor<int32, [4]> x1_23_begin_0 = const()[name = tensor<string, []>("x1_23_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_23_end_0 = const()[name = tensor<string, []>("x1_23_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_23_end_mask_0 = const()[name = tensor<string, []>("x1_23_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_23 = slice_by_index(begin = x1_23_begin_0, end = x1_23_end_0, end_mask = x1_23_end_mask_0, x = output_131_cast_fp16)[name = tensor<string, []>("x1_23")];
            tensor<int32, [4]> x2_23_begin_0 = const()[name = tensor<string, []>("x2_23_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_23_end_0 = const()[name = tensor<string, []>("x2_23_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_23_end_mask_0 = const()[name = tensor<string, []>("x2_23_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_23 = slice_by_index(begin = x2_23_begin_0, end = x2_23_end_0, end_mask = x2_23_end_mask_0, x = output_131_cast_fp16)[name = tensor<string, []>("x2_23")];
            tensor<fp16, []> const_153_promoted = const()[name = tensor<string, []>("const_153_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_1386 = mul(x = x2_23, y = const_153_promoted)[name = tensor<string, []>("op_1386")];
            tensor<bool, []> var_1388_interleave_0 = const()[name = tensor<string, []>("op_1388_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_1388 = concat(axis = var_24, interleave = var_1388_interleave_0, values = (var_1386, x1_23))[name = tensor<string, []>("op_1388")];
            tensor<fp16, [1, 4, 256, 256]> var_1389 = mul(x = var_1388, y = sin_19_palettized)[name = tensor<string, []>("op_1389")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_73 = add(x = var_1375, y = var_1389)[name = tensor<string, []>("hidden_states_73")];
            tensor<int32, [1]> var_1398_axes_0 = const()[name = tensor<string, []>("op_1398_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_1398 = expand_dims(axes = var_1398_axes_0, x = hidden_states_73)[name = tensor<string, []>("op_1398")];
            tensor<int32, [5]> hidden_states_75_reps_0 = const()[name = tensor<string, []>("hidden_states_75_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_75 = tile(reps = hidden_states_75_reps_0, x = var_1398)[name = tensor<string, []>("hidden_states_75")];
            tensor<int32, [4]> var_1406 = const()[name = tensor<string, []>("op_1406"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_11 = reshape(shape = var_1406, x = hidden_states_75)[name = tensor<string, []>("key_states_11")];
            tensor<int32, [1]> var_1415_axes_0 = const()[name = tensor<string, []>("op_1415_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_77 = transpose(perm = hidden_states_77_perm_0, x = var_1327)[name = tensor<string, []>("transpose_113")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_1415 = expand_dims(axes = var_1415_axes_0, x = hidden_states_77)[name = tensor<string, []>("op_1415")];
            tensor<int32, [5]> hidden_states_79_reps_0 = const()[name = tensor<string, []>("hidden_states_79_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_79 = tile(reps = hidden_states_79_reps_0, x = var_1415)[name = tensor<string, []>("hidden_states_79")];
            tensor<int32, [4]> var_1423 = const()[name = tensor<string, []>("op_1423"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_11 = reshape(shape = var_1423, x = hidden_states_79)[name = tensor<string, []>("value_states_11")];
            tensor<bool, []> var_1426_transpose_x_1 = const()[name = tensor<string, []>("op_1426_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1426_transpose_y_1 = const()[name = tensor<string, []>("op_1426_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_1426 = matmul(transpose_x = var_1426_transpose_x_1, transpose_y = var_1426_transpose_y_1, x = query_11, y = key_states_11)[name = tensor<string, []>("op_1426")];
            tensor<fp16, []> var_1427_to_fp16 = const()[name = tensor<string, []>("op_1427_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_21_cast_fp16 = mul(x = var_1426, y = var_1427_to_fp16)[name = tensor<string, []>("attn_weights_21_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_61_cast_fp16 = add(x = attn_weights_21_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_61_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_1435_cast_fp16 = softmax(axis = var_24, x = input_61_cast_fp16)[name = tensor<string, []>("op_1435_cast_fp16")];
            tensor<bool, []> attn_output_21_transpose_x_0 = const()[name = tensor<string, []>("attn_output_21_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_21_transpose_y_0 = const()[name = tensor<string, []>("attn_output_21_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_21 = matmul(transpose_x = attn_output_21_transpose_x_0, transpose_y = attn_output_21_transpose_y_0, x = var_1435_cast_fp16, y = value_states_11)[name = tensor<string, []>("attn_output_21")];
            tensor<int32, [4]> var_1439_perm_0 = const()[name = tensor<string, []>("op_1439_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1441 = const()[name = tensor<string, []>("op_1441"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_1439 = transpose(perm = var_1439_perm_0, x = attn_output_21)[name = tensor<string, []>("transpose_112")];
            tensor<fp16, [1, 256, 2048]> var_1442 = reshape(shape = var_1441, x = var_1439)[name = tensor<string, []>("op_1442")];
            tensor<fp16, [1, 256, 2560]> x_131 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_5_self_attn_o_proj_weight_palettized, x = var_1442)[name = tensor<string, []>("linear_38")];
            tensor<fp16, []> var_22_promoted_33_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_33_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_1448_cast_fp16 = pow(x = x_131, y = var_22_promoted_33_to_fp16)[name = tensor<string, []>("op_1448_cast_fp16")];
            tensor<int32, [1]> var_1450_axes_0 = const()[name = tensor<string, []>("op_1450_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1450_keep_dims_0 = const()[name = tensor<string, []>("op_1450_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_1450_cast_fp16 = reduce_mean(axes = var_1450_axes_0, keep_dims = var_1450_keep_dims_0, x = var_1448_cast_fp16)[name = tensor<string, []>("op_1450_cast_fp16")];
            tensor<fp16, []> var_1451_to_fp16 = const()[name = tensor<string, []>("op_1451_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_1452_cast_fp16 = add(x = var_1450_cast_fp16, y = var_1451_to_fp16)[name = tensor<string, []>("op_1452_cast_fp16")];
            tensor<fp32, []> var_1453_epsilon_0 = const()[name = tensor<string, []>("op_1453_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_1453_cast_fp16 = rsqrt(epsilon = var_1453_epsilon_0, x = var_1452_cast_fp16)[name = tensor<string, []>("op_1453_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_133_cast_fp16 = mul(x = x_131, y = var_1453_cast_fp16)[name = tensor<string, []>("output_133_cast_fp16")];
            tensor<fp16, [2560]> var_1457_to_fp16 = const()[name = tensor<string, []>("op_1457_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940306048)))];
            tensor<fp16, [1, 256, 2560]> output_135_cast_fp16 = mul(x = output_133_cast_fp16, y = var_1457_to_fp16)[name = tensor<string, []>("output_135_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_135 = add(x = x_119, y = output_135_cast_fp16)[name = tensor<string, []>("x_135")];
            tensor<fp16, []> var_22_promoted_34_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_34_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_1463_cast_fp16 = pow(x = x_135, y = var_22_promoted_34_to_fp16)[name = tensor<string, []>("op_1463_cast_fp16")];
            tensor<int32, [1]> var_1465_axes_0 = const()[name = tensor<string, []>("op_1465_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1465_keep_dims_0 = const()[name = tensor<string, []>("op_1465_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_1465_cast_fp16 = reduce_mean(axes = var_1465_axes_0, keep_dims = var_1465_keep_dims_0, x = var_1463_cast_fp16)[name = tensor<string, []>("op_1465_cast_fp16")];
            tensor<fp16, []> var_1466_to_fp16 = const()[name = tensor<string, []>("op_1466_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_1467_cast_fp16 = add(x = var_1465_cast_fp16, y = var_1466_to_fp16)[name = tensor<string, []>("op_1467_cast_fp16")];
            tensor<fp32, []> var_1468_epsilon_0 = const()[name = tensor<string, []>("op_1468_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_1468_cast_fp16 = rsqrt(epsilon = var_1468_epsilon_0, x = var_1467_cast_fp16)[name = tensor<string, []>("op_1468_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_137_cast_fp16 = mul(x = x_135, y = var_1468_cast_fp16)[name = tensor<string, []>("output_137_cast_fp16")];
            tensor<fp16, [2560]> var_1472_to_fp16 = const()[name = tensor<string, []>("op_1472_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940311232)))];
            tensor<fp16, [1, 256, 2560]> output_139_cast_fp16 = mul(x = output_137_cast_fp16, y = var_1472_to_fp16)[name = tensor<string, []>("output_139_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_69 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_5_mlp_gate_proj_weight_palettized, x = output_139_cast_fp16)[name = tensor<string, []>("linear_39")];
            tensor<string, []> var_1480_mode_0 = const()[name = tensor<string, []>("op_1480_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_1480 = gelu(mode = var_1480_mode_0, x = input_69)[name = tensor<string, []>("op_1480")];
            tensor<fp16, [1, 256, 10240]> var_1482 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_5_mlp_up_proj_weight_palettized, x = output_139_cast_fp16)[name = tensor<string, []>("linear_40")];
            tensor<fp16, [1, 256, 10240]> input_71 = mul(x = var_1480, y = var_1482)[name = tensor<string, []>("input_71")];
            tensor<fp16, [1, 256, 2560]> x_139 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_5_mlp_down_proj_weight_palettized, x = input_71)[name = tensor<string, []>("linear_41")];
            tensor<fp16, []> var_22_promoted_35_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_35_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_1488_cast_fp16 = pow(x = x_139, y = var_22_promoted_35_to_fp16)[name = tensor<string, []>("op_1488_cast_fp16")];
            tensor<int32, [1]> var_1490_axes_0 = const()[name = tensor<string, []>("op_1490_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1490_keep_dims_0 = const()[name = tensor<string, []>("op_1490_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_1490_cast_fp16 = reduce_mean(axes = var_1490_axes_0, keep_dims = var_1490_keep_dims_0, x = var_1488_cast_fp16)[name = tensor<string, []>("op_1490_cast_fp16")];
            tensor<fp16, []> var_1491_to_fp16 = const()[name = tensor<string, []>("op_1491_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_1492_cast_fp16 = add(x = var_1490_cast_fp16, y = var_1491_to_fp16)[name = tensor<string, []>("op_1492_cast_fp16")];
            tensor<fp32, []> var_1493_epsilon_0 = const()[name = tensor<string, []>("op_1493_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_1493_cast_fp16 = rsqrt(epsilon = var_1493_epsilon_0, x = var_1492_cast_fp16)[name = tensor<string, []>("op_1493_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_141_cast_fp16 = mul(x = x_139, y = var_1493_cast_fp16)[name = tensor<string, []>("output_141_cast_fp16")];
            tensor<fp16, [2560]> var_1497_to_fp16 = const()[name = tensor<string, []>("op_1497_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940316416)))];
            tensor<fp16, [1, 256, 2560]> output_143_cast_fp16 = mul(x = output_141_cast_fp16, y = var_1497_to_fp16)[name = tensor<string, []>("output_143_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_143 = add(x = x_135, y = output_143_cast_fp16)[name = tensor<string, []>("x_143")];
            tensor<fp16, []> var_22_promoted_36_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_36_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_1509_cast_fp16 = pow(x = x_143, y = var_22_promoted_36_to_fp16)[name = tensor<string, []>("op_1509_cast_fp16")];
            tensor<int32, [1]> var_1511_axes_0 = const()[name = tensor<string, []>("op_1511_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1511_keep_dims_0 = const()[name = tensor<string, []>("op_1511_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_1511_cast_fp16 = reduce_mean(axes = var_1511_axes_0, keep_dims = var_1511_keep_dims_0, x = var_1509_cast_fp16)[name = tensor<string, []>("op_1511_cast_fp16")];
            tensor<fp16, []> var_1512_to_fp16 = const()[name = tensor<string, []>("op_1512_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_1513_cast_fp16 = add(x = var_1511_cast_fp16, y = var_1512_to_fp16)[name = tensor<string, []>("op_1513_cast_fp16")];
            tensor<fp32, []> var_1514_epsilon_0 = const()[name = tensor<string, []>("op_1514_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_1514_cast_fp16 = rsqrt(epsilon = var_1514_epsilon_0, x = var_1513_cast_fp16)[name = tensor<string, []>("op_1514_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_145_cast_fp16 = mul(x = x_143, y = var_1514_cast_fp16)[name = tensor<string, []>("output_145_cast_fp16")];
            tensor<fp16, [2560]> var_1518_to_fp16 = const()[name = tensor<string, []>("op_1518_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940321600)))];
            tensor<fp16, [1, 256, 2560]> output_147_cast_fp16 = mul(x = output_145_cast_fp16, y = var_1518_to_fp16)[name = tensor<string, []>("output_147_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_1530 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_6_self_attn_q_proj_weight_palettized, x = output_147_cast_fp16)[name = tensor<string, []>("linear_42")];
            tensor<int32, [4]> var_1531 = const()[name = tensor<string, []>("op_1531"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_1532 = reshape(shape = var_1531, x = var_1530)[name = tensor<string, []>("op_1532")];
            tensor<int32, [4]> x_147_perm_0 = const()[name = tensor<string, []>("x_147_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_1535 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_6_self_attn_k_proj_weight_palettized, x = output_147_cast_fp16)[name = tensor<string, []>("linear_43")];
            tensor<int32, [4]> var_1536 = const()[name = tensor<string, []>("op_1536"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_1537 = reshape(shape = var_1536, x = var_1535)[name = tensor<string, []>("op_1537")];
            tensor<int32, [4]> x_151_perm_0 = const()[name = tensor<string, []>("x_151_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_1540 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_6_self_attn_v_proj_weight_palettized, x = output_147_cast_fp16)[name = tensor<string, []>("linear_44")];
            tensor<int32, [4]> var_1541 = const()[name = tensor<string, []>("op_1541"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_1542 = reshape(shape = var_1541, x = var_1540)[name = tensor<string, []>("op_1542")];
            tensor<int32, [4]> hidden_states_91_perm_0 = const()[name = tensor<string, []>("hidden_states_91_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_37_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_37_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_147 = transpose(perm = x_147_perm_0, x = var_1532)[name = tensor<string, []>("transpose_111")];
            tensor<fp16, [1, 8, 256, 256]> var_1546_cast_fp16 = pow(x = x_147, y = var_22_promoted_37_to_fp16)[name = tensor<string, []>("op_1546_cast_fp16")];
            tensor<int32, [1]> var_1548_axes_0 = const()[name = tensor<string, []>("op_1548_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1548_keep_dims_0 = const()[name = tensor<string, []>("op_1548_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_1548_cast_fp16 = reduce_mean(axes = var_1548_axes_0, keep_dims = var_1548_keep_dims_0, x = var_1546_cast_fp16)[name = tensor<string, []>("op_1548_cast_fp16")];
            tensor<fp16, []> var_1549_to_fp16 = const()[name = tensor<string, []>("op_1549_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_1550_cast_fp16 = add(x = var_1548_cast_fp16, y = var_1549_to_fp16)[name = tensor<string, []>("op_1550_cast_fp16")];
            tensor<fp32, []> var_1551_epsilon_0 = const()[name = tensor<string, []>("op_1551_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_1551_cast_fp16 = rsqrt(epsilon = var_1551_epsilon_0, x = var_1550_cast_fp16)[name = tensor<string, []>("op_1551_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_149_cast_fp16 = mul(x = x_147, y = var_1551_cast_fp16)[name = tensor<string, []>("output_149_cast_fp16")];
            tensor<fp16, [256]> var_1555_to_fp16 = const()[name = tensor<string, []>("op_1555_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940326784)))];
            tensor<fp16, [1, 8, 256, 256]> output_151_cast_fp16 = mul(x = output_149_cast_fp16, y = var_1555_to_fp16)[name = tensor<string, []>("output_151_cast_fp16")];
            tensor<fp16, []> var_22_promoted_38_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_38_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_151 = transpose(perm = x_151_perm_0, x = var_1537)[name = tensor<string, []>("transpose_110")];
            tensor<fp16, [1, 4, 256, 256]> var_1560_cast_fp16 = pow(x = x_151, y = var_22_promoted_38_to_fp16)[name = tensor<string, []>("op_1560_cast_fp16")];
            tensor<int32, [1]> var_1562_axes_0 = const()[name = tensor<string, []>("op_1562_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1562_keep_dims_0 = const()[name = tensor<string, []>("op_1562_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_1562_cast_fp16 = reduce_mean(axes = var_1562_axes_0, keep_dims = var_1562_keep_dims_0, x = var_1560_cast_fp16)[name = tensor<string, []>("op_1562_cast_fp16")];
            tensor<fp16, []> var_1563_to_fp16 = const()[name = tensor<string, []>("op_1563_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_1564_cast_fp16 = add(x = var_1562_cast_fp16, y = var_1563_to_fp16)[name = tensor<string, []>("op_1564_cast_fp16")];
            tensor<fp32, []> var_1565_epsilon_0 = const()[name = tensor<string, []>("op_1565_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_1565_cast_fp16 = rsqrt(epsilon = var_1565_epsilon_0, x = var_1564_cast_fp16)[name = tensor<string, []>("op_1565_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_153_cast_fp16 = mul(x = x_151, y = var_1565_cast_fp16)[name = tensor<string, []>("output_153_cast_fp16")];
            tensor<fp16, [256]> var_1569_to_fp16 = const()[name = tensor<string, []>("op_1569_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940327360)))];
            tensor<fp16, [1, 4, 256, 256]> output_155_cast_fp16 = mul(x = output_153_cast_fp16, y = var_1569_to_fp16)[name = tensor<string, []>("output_155_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_1574 = mul(x = output_151_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_1574")];
            tensor<int32, [4]> x1_25_begin_0 = const()[name = tensor<string, []>("x1_25_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_25_end_0 = const()[name = tensor<string, []>("x1_25_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_25_end_mask_0 = const()[name = tensor<string, []>("x1_25_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_25 = slice_by_index(begin = x1_25_begin_0, end = x1_25_end_0, end_mask = x1_25_end_mask_0, x = output_151_cast_fp16)[name = tensor<string, []>("x1_25")];
            tensor<int32, [4]> x2_25_begin_0 = const()[name = tensor<string, []>("x2_25_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_25_end_0 = const()[name = tensor<string, []>("x2_25_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_25_end_mask_0 = const()[name = tensor<string, []>("x2_25_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_25 = slice_by_index(begin = x2_25_begin_0, end = x2_25_end_0, end_mask = x2_25_end_mask_0, x = output_151_cast_fp16)[name = tensor<string, []>("x2_25")];
            tensor<fp16, []> const_173_promoted = const()[name = tensor<string, []>("const_173_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_1585 = mul(x = x2_25, y = const_173_promoted)[name = tensor<string, []>("op_1585")];
            tensor<bool, []> var_1587_interleave_0 = const()[name = tensor<string, []>("op_1587_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_1587 = concat(axis = var_24, interleave = var_1587_interleave_0, values = (var_1585, x1_25))[name = tensor<string, []>("op_1587")];
            tensor<fp16, [1, 8, 256, 256]> var_1588 = mul(x = var_1587, y = sin_7_palettized)[name = tensor<string, []>("op_1588")];
            tensor<fp16, [1, 8, 256, 256]> query_13 = add(x = var_1574, y = var_1588)[name = tensor<string, []>("query_13")];
            tensor<fp16, [1, 4, 256, 256]> var_1590 = mul(x = output_155_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_1590")];
            tensor<int32, [4]> x1_27_begin_0 = const()[name = tensor<string, []>("x1_27_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_27_end_0 = const()[name = tensor<string, []>("x1_27_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_27_end_mask_0 = const()[name = tensor<string, []>("x1_27_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_27 = slice_by_index(begin = x1_27_begin_0, end = x1_27_end_0, end_mask = x1_27_end_mask_0, x = output_155_cast_fp16)[name = tensor<string, []>("x1_27")];
            tensor<int32, [4]> x2_27_begin_0 = const()[name = tensor<string, []>("x2_27_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_27_end_0 = const()[name = tensor<string, []>("x2_27_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_27_end_mask_0 = const()[name = tensor<string, []>("x2_27_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_27 = slice_by_index(begin = x2_27_begin_0, end = x2_27_end_0, end_mask = x2_27_end_mask_0, x = output_155_cast_fp16)[name = tensor<string, []>("x2_27")];
            tensor<fp16, []> const_176_promoted = const()[name = tensor<string, []>("const_176_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_1601 = mul(x = x2_27, y = const_176_promoted)[name = tensor<string, []>("op_1601")];
            tensor<bool, []> var_1603_interleave_0 = const()[name = tensor<string, []>("op_1603_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_1603 = concat(axis = var_24, interleave = var_1603_interleave_0, values = (var_1601, x1_27))[name = tensor<string, []>("op_1603")];
            tensor<fp16, [1, 4, 256, 256]> var_1604 = mul(x = var_1603, y = sin_7_palettized)[name = tensor<string, []>("op_1604")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_87 = add(x = var_1590, y = var_1604)[name = tensor<string, []>("hidden_states_87")];
            tensor<int32, [1]> var_1613_axes_0 = const()[name = tensor<string, []>("op_1613_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_1613 = expand_dims(axes = var_1613_axes_0, x = hidden_states_87)[name = tensor<string, []>("op_1613")];
            tensor<int32, [5]> hidden_states_89_reps_0 = const()[name = tensor<string, []>("hidden_states_89_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_89 = tile(reps = hidden_states_89_reps_0, x = var_1613)[name = tensor<string, []>("hidden_states_89")];
            tensor<int32, [4]> var_1621 = const()[name = tensor<string, []>("op_1621"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_13 = reshape(shape = var_1621, x = hidden_states_89)[name = tensor<string, []>("key_states_13")];
            tensor<int32, [1]> var_1630_axes_0 = const()[name = tensor<string, []>("op_1630_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_91 = transpose(perm = hidden_states_91_perm_0, x = var_1542)[name = tensor<string, []>("transpose_109")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_1630 = expand_dims(axes = var_1630_axes_0, x = hidden_states_91)[name = tensor<string, []>("op_1630")];
            tensor<int32, [5]> hidden_states_93_reps_0 = const()[name = tensor<string, []>("hidden_states_93_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_93 = tile(reps = hidden_states_93_reps_0, x = var_1630)[name = tensor<string, []>("hidden_states_93")];
            tensor<int32, [4]> var_1638 = const()[name = tensor<string, []>("op_1638"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_13 = reshape(shape = var_1638, x = hidden_states_93)[name = tensor<string, []>("value_states_13")];
            tensor<bool, []> var_1641_transpose_x_1 = const()[name = tensor<string, []>("op_1641_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1641_transpose_y_1 = const()[name = tensor<string, []>("op_1641_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_1641 = matmul(transpose_x = var_1641_transpose_x_1, transpose_y = var_1641_transpose_y_1, x = query_13, y = key_states_13)[name = tensor<string, []>("op_1641")];
            tensor<fp16, []> var_1642_to_fp16 = const()[name = tensor<string, []>("op_1642_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_25_cast_fp16 = mul(x = var_1641, y = var_1642_to_fp16)[name = tensor<string, []>("attn_weights_25_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_73_cast_fp16 = add(x = attn_weights_25_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_73_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_1650_cast_fp16 = softmax(axis = var_24, x = input_73_cast_fp16)[name = tensor<string, []>("op_1650_cast_fp16")];
            tensor<bool, []> attn_output_25_transpose_x_0 = const()[name = tensor<string, []>("attn_output_25_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_25_transpose_y_0 = const()[name = tensor<string, []>("attn_output_25_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_25 = matmul(transpose_x = attn_output_25_transpose_x_0, transpose_y = attn_output_25_transpose_y_0, x = var_1650_cast_fp16, y = value_states_13)[name = tensor<string, []>("attn_output_25")];
            tensor<int32, [4]> var_1654_perm_0 = const()[name = tensor<string, []>("op_1654_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1656 = const()[name = tensor<string, []>("op_1656"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_1654 = transpose(perm = var_1654_perm_0, x = attn_output_25)[name = tensor<string, []>("transpose_108")];
            tensor<fp16, [1, 256, 2048]> var_1657 = reshape(shape = var_1656, x = var_1654)[name = tensor<string, []>("op_1657")];
            tensor<fp16, [1, 256, 2560]> x_155 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_6_self_attn_o_proj_weight_palettized, x = var_1657)[name = tensor<string, []>("linear_45")];
            tensor<fp16, []> var_22_promoted_39_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_39_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_1663_cast_fp16 = pow(x = x_155, y = var_22_promoted_39_to_fp16)[name = tensor<string, []>("op_1663_cast_fp16")];
            tensor<int32, [1]> var_1665_axes_0 = const()[name = tensor<string, []>("op_1665_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1665_keep_dims_0 = const()[name = tensor<string, []>("op_1665_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_1665_cast_fp16 = reduce_mean(axes = var_1665_axes_0, keep_dims = var_1665_keep_dims_0, x = var_1663_cast_fp16)[name = tensor<string, []>("op_1665_cast_fp16")];
            tensor<fp16, []> var_1666_to_fp16 = const()[name = tensor<string, []>("op_1666_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_1667_cast_fp16 = add(x = var_1665_cast_fp16, y = var_1666_to_fp16)[name = tensor<string, []>("op_1667_cast_fp16")];
            tensor<fp32, []> var_1668_epsilon_0 = const()[name = tensor<string, []>("op_1668_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_1668_cast_fp16 = rsqrt(epsilon = var_1668_epsilon_0, x = var_1667_cast_fp16)[name = tensor<string, []>("op_1668_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_157_cast_fp16 = mul(x = x_155, y = var_1668_cast_fp16)[name = tensor<string, []>("output_157_cast_fp16")];
            tensor<fp16, [2560]> var_1672_to_fp16 = const()[name = tensor<string, []>("op_1672_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940327936)))];
            tensor<fp16, [1, 256, 2560]> output_159_cast_fp16 = mul(x = output_157_cast_fp16, y = var_1672_to_fp16)[name = tensor<string, []>("output_159_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_159 = add(x = x_143, y = output_159_cast_fp16)[name = tensor<string, []>("x_159")];
            tensor<fp16, []> var_22_promoted_40_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_40_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_1678_cast_fp16 = pow(x = x_159, y = var_22_promoted_40_to_fp16)[name = tensor<string, []>("op_1678_cast_fp16")];
            tensor<int32, [1]> var_1680_axes_0 = const()[name = tensor<string, []>("op_1680_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1680_keep_dims_0 = const()[name = tensor<string, []>("op_1680_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_1680_cast_fp16 = reduce_mean(axes = var_1680_axes_0, keep_dims = var_1680_keep_dims_0, x = var_1678_cast_fp16)[name = tensor<string, []>("op_1680_cast_fp16")];
            tensor<fp16, []> var_1681_to_fp16 = const()[name = tensor<string, []>("op_1681_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_1682_cast_fp16 = add(x = var_1680_cast_fp16, y = var_1681_to_fp16)[name = tensor<string, []>("op_1682_cast_fp16")];
            tensor<fp32, []> var_1683_epsilon_0 = const()[name = tensor<string, []>("op_1683_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_1683_cast_fp16 = rsqrt(epsilon = var_1683_epsilon_0, x = var_1682_cast_fp16)[name = tensor<string, []>("op_1683_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_161_cast_fp16 = mul(x = x_159, y = var_1683_cast_fp16)[name = tensor<string, []>("output_161_cast_fp16")];
            tensor<fp16, [2560]> var_1687_to_fp16 = const()[name = tensor<string, []>("op_1687_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940333120)))];
            tensor<fp16, [1, 256, 2560]> output_163_cast_fp16 = mul(x = output_161_cast_fp16, y = var_1687_to_fp16)[name = tensor<string, []>("output_163_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_81 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_6_mlp_gate_proj_weight_palettized, x = output_163_cast_fp16)[name = tensor<string, []>("linear_46")];
            tensor<string, []> var_1695_mode_0 = const()[name = tensor<string, []>("op_1695_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_1695 = gelu(mode = var_1695_mode_0, x = input_81)[name = tensor<string, []>("op_1695")];
            tensor<fp16, [1, 256, 10240]> var_1697 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_6_mlp_up_proj_weight_palettized, x = output_163_cast_fp16)[name = tensor<string, []>("linear_47")];
            tensor<fp16, [1, 256, 10240]> input_83 = mul(x = var_1695, y = var_1697)[name = tensor<string, []>("input_83")];
            tensor<fp16, [1, 256, 2560]> x_163 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_6_mlp_down_proj_weight_palettized, x = input_83)[name = tensor<string, []>("linear_48")];
            tensor<fp16, []> var_22_promoted_41_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_41_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_1703_cast_fp16 = pow(x = x_163, y = var_22_promoted_41_to_fp16)[name = tensor<string, []>("op_1703_cast_fp16")];
            tensor<int32, [1]> var_1705_axes_0 = const()[name = tensor<string, []>("op_1705_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1705_keep_dims_0 = const()[name = tensor<string, []>("op_1705_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_1705_cast_fp16 = reduce_mean(axes = var_1705_axes_0, keep_dims = var_1705_keep_dims_0, x = var_1703_cast_fp16)[name = tensor<string, []>("op_1705_cast_fp16")];
            tensor<fp16, []> var_1706_to_fp16 = const()[name = tensor<string, []>("op_1706_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_1707_cast_fp16 = add(x = var_1705_cast_fp16, y = var_1706_to_fp16)[name = tensor<string, []>("op_1707_cast_fp16")];
            tensor<fp32, []> var_1708_epsilon_0 = const()[name = tensor<string, []>("op_1708_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_1708_cast_fp16 = rsqrt(epsilon = var_1708_epsilon_0, x = var_1707_cast_fp16)[name = tensor<string, []>("op_1708_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_165_cast_fp16 = mul(x = x_163, y = var_1708_cast_fp16)[name = tensor<string, []>("output_165_cast_fp16")];
            tensor<fp16, [2560]> var_1712_to_fp16 = const()[name = tensor<string, []>("op_1712_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940338304)))];
            tensor<fp16, [1, 256, 2560]> output_167_cast_fp16 = mul(x = output_165_cast_fp16, y = var_1712_to_fp16)[name = tensor<string, []>("output_167_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_167 = add(x = x_159, y = output_167_cast_fp16)[name = tensor<string, []>("x_167")];
            tensor<fp16, []> var_22_promoted_42_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_42_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_1724_cast_fp16 = pow(x = x_167, y = var_22_promoted_42_to_fp16)[name = tensor<string, []>("op_1724_cast_fp16")];
            tensor<int32, [1]> var_1726_axes_0 = const()[name = tensor<string, []>("op_1726_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1726_keep_dims_0 = const()[name = tensor<string, []>("op_1726_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_1726_cast_fp16 = reduce_mean(axes = var_1726_axes_0, keep_dims = var_1726_keep_dims_0, x = var_1724_cast_fp16)[name = tensor<string, []>("op_1726_cast_fp16")];
            tensor<fp16, []> var_1727_to_fp16 = const()[name = tensor<string, []>("op_1727_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_1728_cast_fp16 = add(x = var_1726_cast_fp16, y = var_1727_to_fp16)[name = tensor<string, []>("op_1728_cast_fp16")];
            tensor<fp32, []> var_1729_epsilon_0 = const()[name = tensor<string, []>("op_1729_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_1729_cast_fp16 = rsqrt(epsilon = var_1729_epsilon_0, x = var_1728_cast_fp16)[name = tensor<string, []>("op_1729_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_169_cast_fp16 = mul(x = x_167, y = var_1729_cast_fp16)[name = tensor<string, []>("output_169_cast_fp16")];
            tensor<fp16, [2560]> var_1733_to_fp16 = const()[name = tensor<string, []>("op_1733_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940343488)))];
            tensor<fp16, [1, 256, 2560]> output_171_cast_fp16 = mul(x = output_169_cast_fp16, y = var_1733_to_fp16)[name = tensor<string, []>("output_171_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_1745 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_7_self_attn_q_proj_weight_palettized, x = output_171_cast_fp16)[name = tensor<string, []>("linear_49")];
            tensor<int32, [4]> var_1746 = const()[name = tensor<string, []>("op_1746"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_1747 = reshape(shape = var_1746, x = var_1745)[name = tensor<string, []>("op_1747")];
            tensor<int32, [4]> x_171_perm_0 = const()[name = tensor<string, []>("x_171_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_1750 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_7_self_attn_k_proj_weight_palettized, x = output_171_cast_fp16)[name = tensor<string, []>("linear_50")];
            tensor<int32, [4]> var_1751 = const()[name = tensor<string, []>("op_1751"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_1752 = reshape(shape = var_1751, x = var_1750)[name = tensor<string, []>("op_1752")];
            tensor<int32, [4]> x_175_perm_0 = const()[name = tensor<string, []>("x_175_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_1755 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_7_self_attn_v_proj_weight_palettized, x = output_171_cast_fp16)[name = tensor<string, []>("linear_51")];
            tensor<int32, [4]> var_1756 = const()[name = tensor<string, []>("op_1756"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_1757 = reshape(shape = var_1756, x = var_1755)[name = tensor<string, []>("op_1757")];
            tensor<int32, [4]> hidden_states_105_perm_0 = const()[name = tensor<string, []>("hidden_states_105_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_43_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_43_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_171 = transpose(perm = x_171_perm_0, x = var_1747)[name = tensor<string, []>("transpose_107")];
            tensor<fp16, [1, 8, 256, 256]> var_1761_cast_fp16 = pow(x = x_171, y = var_22_promoted_43_to_fp16)[name = tensor<string, []>("op_1761_cast_fp16")];
            tensor<int32, [1]> var_1763_axes_0 = const()[name = tensor<string, []>("op_1763_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1763_keep_dims_0 = const()[name = tensor<string, []>("op_1763_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_1763_cast_fp16 = reduce_mean(axes = var_1763_axes_0, keep_dims = var_1763_keep_dims_0, x = var_1761_cast_fp16)[name = tensor<string, []>("op_1763_cast_fp16")];
            tensor<fp16, []> var_1764_to_fp16 = const()[name = tensor<string, []>("op_1764_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_1765_cast_fp16 = add(x = var_1763_cast_fp16, y = var_1764_to_fp16)[name = tensor<string, []>("op_1765_cast_fp16")];
            tensor<fp32, []> var_1766_epsilon_0 = const()[name = tensor<string, []>("op_1766_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_1766_cast_fp16 = rsqrt(epsilon = var_1766_epsilon_0, x = var_1765_cast_fp16)[name = tensor<string, []>("op_1766_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_173_cast_fp16 = mul(x = x_171, y = var_1766_cast_fp16)[name = tensor<string, []>("output_173_cast_fp16")];
            tensor<fp16, [256]> var_1770_to_fp16 = const()[name = tensor<string, []>("op_1770_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940348672)))];
            tensor<fp16, [1, 8, 256, 256]> output_175_cast_fp16 = mul(x = output_173_cast_fp16, y = var_1770_to_fp16)[name = tensor<string, []>("output_175_cast_fp16")];
            tensor<fp16, []> var_22_promoted_44_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_44_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_175 = transpose(perm = x_175_perm_0, x = var_1752)[name = tensor<string, []>("transpose_106")];
            tensor<fp16, [1, 4, 256, 256]> var_1775_cast_fp16 = pow(x = x_175, y = var_22_promoted_44_to_fp16)[name = tensor<string, []>("op_1775_cast_fp16")];
            tensor<int32, [1]> var_1777_axes_0 = const()[name = tensor<string, []>("op_1777_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1777_keep_dims_0 = const()[name = tensor<string, []>("op_1777_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_1777_cast_fp16 = reduce_mean(axes = var_1777_axes_0, keep_dims = var_1777_keep_dims_0, x = var_1775_cast_fp16)[name = tensor<string, []>("op_1777_cast_fp16")];
            tensor<fp16, []> var_1778_to_fp16 = const()[name = tensor<string, []>("op_1778_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_1779_cast_fp16 = add(x = var_1777_cast_fp16, y = var_1778_to_fp16)[name = tensor<string, []>("op_1779_cast_fp16")];
            tensor<fp32, []> var_1780_epsilon_0 = const()[name = tensor<string, []>("op_1780_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_1780_cast_fp16 = rsqrt(epsilon = var_1780_epsilon_0, x = var_1779_cast_fp16)[name = tensor<string, []>("op_1780_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_177_cast_fp16 = mul(x = x_175, y = var_1780_cast_fp16)[name = tensor<string, []>("output_177_cast_fp16")];
            tensor<fp16, [256]> var_1784_to_fp16 = const()[name = tensor<string, []>("op_1784_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940349248)))];
            tensor<fp16, [1, 4, 256, 256]> output_179_cast_fp16 = mul(x = output_177_cast_fp16, y = var_1784_to_fp16)[name = tensor<string, []>("output_179_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_1789 = mul(x = output_175_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_1789")];
            tensor<int32, [4]> x1_29_begin_0 = const()[name = tensor<string, []>("x1_29_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_29_end_0 = const()[name = tensor<string, []>("x1_29_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_29_end_mask_0 = const()[name = tensor<string, []>("x1_29_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_29 = slice_by_index(begin = x1_29_begin_0, end = x1_29_end_0, end_mask = x1_29_end_mask_0, x = output_175_cast_fp16)[name = tensor<string, []>("x1_29")];
            tensor<int32, [4]> x2_29_begin_0 = const()[name = tensor<string, []>("x2_29_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_29_end_0 = const()[name = tensor<string, []>("x2_29_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_29_end_mask_0 = const()[name = tensor<string, []>("x2_29_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_29 = slice_by_index(begin = x2_29_begin_0, end = x2_29_end_0, end_mask = x2_29_end_mask_0, x = output_175_cast_fp16)[name = tensor<string, []>("x2_29")];
            tensor<fp16, []> const_196_promoted = const()[name = tensor<string, []>("const_196_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_1800 = mul(x = x2_29, y = const_196_promoted)[name = tensor<string, []>("op_1800")];
            tensor<bool, []> var_1802_interleave_0 = const()[name = tensor<string, []>("op_1802_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_1802 = concat(axis = var_24, interleave = var_1802_interleave_0, values = (var_1800, x1_29))[name = tensor<string, []>("op_1802")];
            tensor<fp16, [1, 8, 256, 256]> var_1803 = mul(x = var_1802, y = sin_7_palettized)[name = tensor<string, []>("op_1803")];
            tensor<fp16, [1, 8, 256, 256]> query_15 = add(x = var_1789, y = var_1803)[name = tensor<string, []>("query_15")];
            tensor<fp16, [1, 4, 256, 256]> var_1805 = mul(x = output_179_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_1805")];
            tensor<int32, [4]> x1_31_begin_0 = const()[name = tensor<string, []>("x1_31_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_31_end_0 = const()[name = tensor<string, []>("x1_31_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_31_end_mask_0 = const()[name = tensor<string, []>("x1_31_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_31 = slice_by_index(begin = x1_31_begin_0, end = x1_31_end_0, end_mask = x1_31_end_mask_0, x = output_179_cast_fp16)[name = tensor<string, []>("x1_31")];
            tensor<int32, [4]> x2_31_begin_0 = const()[name = tensor<string, []>("x2_31_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_31_end_0 = const()[name = tensor<string, []>("x2_31_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_31_end_mask_0 = const()[name = tensor<string, []>("x2_31_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_31 = slice_by_index(begin = x2_31_begin_0, end = x2_31_end_0, end_mask = x2_31_end_mask_0, x = output_179_cast_fp16)[name = tensor<string, []>("x2_31")];
            tensor<fp16, []> const_199_promoted = const()[name = tensor<string, []>("const_199_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_1816 = mul(x = x2_31, y = const_199_promoted)[name = tensor<string, []>("op_1816")];
            tensor<bool, []> var_1818_interleave_0 = const()[name = tensor<string, []>("op_1818_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_1818 = concat(axis = var_24, interleave = var_1818_interleave_0, values = (var_1816, x1_31))[name = tensor<string, []>("op_1818")];
            tensor<fp16, [1, 4, 256, 256]> var_1819 = mul(x = var_1818, y = sin_7_palettized)[name = tensor<string, []>("op_1819")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_101 = add(x = var_1805, y = var_1819)[name = tensor<string, []>("hidden_states_101")];
            tensor<int32, [1]> var_1828_axes_0 = const()[name = tensor<string, []>("op_1828_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_1828 = expand_dims(axes = var_1828_axes_0, x = hidden_states_101)[name = tensor<string, []>("op_1828")];
            tensor<int32, [5]> hidden_states_103_reps_0 = const()[name = tensor<string, []>("hidden_states_103_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_103 = tile(reps = hidden_states_103_reps_0, x = var_1828)[name = tensor<string, []>("hidden_states_103")];
            tensor<int32, [4]> var_1836 = const()[name = tensor<string, []>("op_1836"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_15 = reshape(shape = var_1836, x = hidden_states_103)[name = tensor<string, []>("key_states_15")];
            tensor<int32, [1]> var_1845_axes_0 = const()[name = tensor<string, []>("op_1845_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_105 = transpose(perm = hidden_states_105_perm_0, x = var_1757)[name = tensor<string, []>("transpose_105")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_1845 = expand_dims(axes = var_1845_axes_0, x = hidden_states_105)[name = tensor<string, []>("op_1845")];
            tensor<int32, [5]> hidden_states_107_reps_0 = const()[name = tensor<string, []>("hidden_states_107_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_107 = tile(reps = hidden_states_107_reps_0, x = var_1845)[name = tensor<string, []>("hidden_states_107")];
            tensor<int32, [4]> var_1853 = const()[name = tensor<string, []>("op_1853"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_15 = reshape(shape = var_1853, x = hidden_states_107)[name = tensor<string, []>("value_states_15")];
            tensor<bool, []> var_1856_transpose_x_1 = const()[name = tensor<string, []>("op_1856_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1856_transpose_y_1 = const()[name = tensor<string, []>("op_1856_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_1856 = matmul(transpose_x = var_1856_transpose_x_1, transpose_y = var_1856_transpose_y_1, x = query_15, y = key_states_15)[name = tensor<string, []>("op_1856")];
            tensor<fp16, []> var_1857_to_fp16 = const()[name = tensor<string, []>("op_1857_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_29_cast_fp16 = mul(x = var_1856, y = var_1857_to_fp16)[name = tensor<string, []>("attn_weights_29_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_85_cast_fp16 = add(x = attn_weights_29_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_85_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_1865_cast_fp16 = softmax(axis = var_24, x = input_85_cast_fp16)[name = tensor<string, []>("op_1865_cast_fp16")];
            tensor<bool, []> attn_output_29_transpose_x_0 = const()[name = tensor<string, []>("attn_output_29_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_29_transpose_y_0 = const()[name = tensor<string, []>("attn_output_29_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_29 = matmul(transpose_x = attn_output_29_transpose_x_0, transpose_y = attn_output_29_transpose_y_0, x = var_1865_cast_fp16, y = value_states_15)[name = tensor<string, []>("attn_output_29")];
            tensor<int32, [4]> var_1869_perm_0 = const()[name = tensor<string, []>("op_1869_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1871 = const()[name = tensor<string, []>("op_1871"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_1869 = transpose(perm = var_1869_perm_0, x = attn_output_29)[name = tensor<string, []>("transpose_104")];
            tensor<fp16, [1, 256, 2048]> var_1872 = reshape(shape = var_1871, x = var_1869)[name = tensor<string, []>("op_1872")];
            tensor<fp16, [1, 256, 2560]> x_179 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_7_self_attn_o_proj_weight_palettized, x = var_1872)[name = tensor<string, []>("linear_52")];
            tensor<fp16, []> var_22_promoted_45_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_45_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_1878_cast_fp16 = pow(x = x_179, y = var_22_promoted_45_to_fp16)[name = tensor<string, []>("op_1878_cast_fp16")];
            tensor<int32, [1]> var_1880_axes_0 = const()[name = tensor<string, []>("op_1880_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1880_keep_dims_0 = const()[name = tensor<string, []>("op_1880_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_1880_cast_fp16 = reduce_mean(axes = var_1880_axes_0, keep_dims = var_1880_keep_dims_0, x = var_1878_cast_fp16)[name = tensor<string, []>("op_1880_cast_fp16")];
            tensor<fp16, []> var_1881_to_fp16 = const()[name = tensor<string, []>("op_1881_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_1882_cast_fp16 = add(x = var_1880_cast_fp16, y = var_1881_to_fp16)[name = tensor<string, []>("op_1882_cast_fp16")];
            tensor<fp32, []> var_1883_epsilon_0 = const()[name = tensor<string, []>("op_1883_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_1883_cast_fp16 = rsqrt(epsilon = var_1883_epsilon_0, x = var_1882_cast_fp16)[name = tensor<string, []>("op_1883_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_181_cast_fp16 = mul(x = x_179, y = var_1883_cast_fp16)[name = tensor<string, []>("output_181_cast_fp16")];
            tensor<fp16, [2560]> var_1887_to_fp16 = const()[name = tensor<string, []>("op_1887_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940349824)))];
            tensor<fp16, [1, 256, 2560]> output_183_cast_fp16 = mul(x = output_181_cast_fp16, y = var_1887_to_fp16)[name = tensor<string, []>("output_183_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_183 = add(x = x_167, y = output_183_cast_fp16)[name = tensor<string, []>("x_183")];
            tensor<fp16, []> var_22_promoted_46_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_46_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_1893_cast_fp16 = pow(x = x_183, y = var_22_promoted_46_to_fp16)[name = tensor<string, []>("op_1893_cast_fp16")];
            tensor<int32, [1]> var_1895_axes_0 = const()[name = tensor<string, []>("op_1895_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1895_keep_dims_0 = const()[name = tensor<string, []>("op_1895_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_1895_cast_fp16 = reduce_mean(axes = var_1895_axes_0, keep_dims = var_1895_keep_dims_0, x = var_1893_cast_fp16)[name = tensor<string, []>("op_1895_cast_fp16")];
            tensor<fp16, []> var_1896_to_fp16 = const()[name = tensor<string, []>("op_1896_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_1897_cast_fp16 = add(x = var_1895_cast_fp16, y = var_1896_to_fp16)[name = tensor<string, []>("op_1897_cast_fp16")];
            tensor<fp32, []> var_1898_epsilon_0 = const()[name = tensor<string, []>("op_1898_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_1898_cast_fp16 = rsqrt(epsilon = var_1898_epsilon_0, x = var_1897_cast_fp16)[name = tensor<string, []>("op_1898_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_185_cast_fp16 = mul(x = x_183, y = var_1898_cast_fp16)[name = tensor<string, []>("output_185_cast_fp16")];
            tensor<fp16, [2560]> var_1902_to_fp16 = const()[name = tensor<string, []>("op_1902_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940355008)))];
            tensor<fp16, [1, 256, 2560]> output_187_cast_fp16 = mul(x = output_185_cast_fp16, y = var_1902_to_fp16)[name = tensor<string, []>("output_187_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_93 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_7_mlp_gate_proj_weight_palettized, x = output_187_cast_fp16)[name = tensor<string, []>("linear_53")];
            tensor<string, []> var_1910_mode_0 = const()[name = tensor<string, []>("op_1910_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_1910 = gelu(mode = var_1910_mode_0, x = input_93)[name = tensor<string, []>("op_1910")];
            tensor<fp16, [1, 256, 10240]> var_1912 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_7_mlp_up_proj_weight_palettized, x = output_187_cast_fp16)[name = tensor<string, []>("linear_54")];
            tensor<fp16, [1, 256, 10240]> input_95 = mul(x = var_1910, y = var_1912)[name = tensor<string, []>("input_95")];
            tensor<fp16, [1, 256, 2560]> x_187 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_7_mlp_down_proj_weight_palettized, x = input_95)[name = tensor<string, []>("linear_55")];
            tensor<fp16, []> var_22_promoted_47_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_47_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_1918_cast_fp16 = pow(x = x_187, y = var_22_promoted_47_to_fp16)[name = tensor<string, []>("op_1918_cast_fp16")];
            tensor<int32, [1]> var_1920_axes_0 = const()[name = tensor<string, []>("op_1920_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1920_keep_dims_0 = const()[name = tensor<string, []>("op_1920_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_1920_cast_fp16 = reduce_mean(axes = var_1920_axes_0, keep_dims = var_1920_keep_dims_0, x = var_1918_cast_fp16)[name = tensor<string, []>("op_1920_cast_fp16")];
            tensor<fp16, []> var_1921_to_fp16 = const()[name = tensor<string, []>("op_1921_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_1922_cast_fp16 = add(x = var_1920_cast_fp16, y = var_1921_to_fp16)[name = tensor<string, []>("op_1922_cast_fp16")];
            tensor<fp32, []> var_1923_epsilon_0 = const()[name = tensor<string, []>("op_1923_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_1923_cast_fp16 = rsqrt(epsilon = var_1923_epsilon_0, x = var_1922_cast_fp16)[name = tensor<string, []>("op_1923_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_189_cast_fp16 = mul(x = x_187, y = var_1923_cast_fp16)[name = tensor<string, []>("output_189_cast_fp16")];
            tensor<fp16, [2560]> var_1927_to_fp16 = const()[name = tensor<string, []>("op_1927_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940360192)))];
            tensor<fp16, [1, 256, 2560]> output_191_cast_fp16 = mul(x = output_189_cast_fp16, y = var_1927_to_fp16)[name = tensor<string, []>("output_191_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_191 = add(x = x_183, y = output_191_cast_fp16)[name = tensor<string, []>("x_191")];
            tensor<fp16, []> var_22_promoted_48_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_48_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_1939_cast_fp16 = pow(x = x_191, y = var_22_promoted_48_to_fp16)[name = tensor<string, []>("op_1939_cast_fp16")];
            tensor<int32, [1]> var_1941_axes_0 = const()[name = tensor<string, []>("op_1941_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1941_keep_dims_0 = const()[name = tensor<string, []>("op_1941_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_1941_cast_fp16 = reduce_mean(axes = var_1941_axes_0, keep_dims = var_1941_keep_dims_0, x = var_1939_cast_fp16)[name = tensor<string, []>("op_1941_cast_fp16")];
            tensor<fp16, []> var_1942_to_fp16 = const()[name = tensor<string, []>("op_1942_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_1943_cast_fp16 = add(x = var_1941_cast_fp16, y = var_1942_to_fp16)[name = tensor<string, []>("op_1943_cast_fp16")];
            tensor<fp32, []> var_1944_epsilon_0 = const()[name = tensor<string, []>("op_1944_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_1944_cast_fp16 = rsqrt(epsilon = var_1944_epsilon_0, x = var_1943_cast_fp16)[name = tensor<string, []>("op_1944_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_193_cast_fp16 = mul(x = x_191, y = var_1944_cast_fp16)[name = tensor<string, []>("output_193_cast_fp16")];
            tensor<fp16, [2560]> var_1948_to_fp16 = const()[name = tensor<string, []>("op_1948_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940365376)))];
            tensor<fp16, [1, 256, 2560]> output_195_cast_fp16 = mul(x = output_193_cast_fp16, y = var_1948_to_fp16)[name = tensor<string, []>("output_195_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_1960 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_8_self_attn_q_proj_weight_palettized, x = output_195_cast_fp16)[name = tensor<string, []>("linear_56")];
            tensor<int32, [4]> var_1961 = const()[name = tensor<string, []>("op_1961"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_1962 = reshape(shape = var_1961, x = var_1960)[name = tensor<string, []>("op_1962")];
            tensor<int32, [4]> x_195_perm_0 = const()[name = tensor<string, []>("x_195_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_1965 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_8_self_attn_k_proj_weight_palettized, x = output_195_cast_fp16)[name = tensor<string, []>("linear_57")];
            tensor<int32, [4]> var_1966 = const()[name = tensor<string, []>("op_1966"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_1967 = reshape(shape = var_1966, x = var_1965)[name = tensor<string, []>("op_1967")];
            tensor<int32, [4]> x_199_perm_0 = const()[name = tensor<string, []>("x_199_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_1970 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_8_self_attn_v_proj_weight_palettized, x = output_195_cast_fp16)[name = tensor<string, []>("linear_58")];
            tensor<int32, [4]> var_1971 = const()[name = tensor<string, []>("op_1971"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_1972 = reshape(shape = var_1971, x = var_1970)[name = tensor<string, []>("op_1972")];
            tensor<int32, [4]> hidden_states_119_perm_0 = const()[name = tensor<string, []>("hidden_states_119_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_49_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_49_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_195 = transpose(perm = x_195_perm_0, x = var_1962)[name = tensor<string, []>("transpose_103")];
            tensor<fp16, [1, 8, 256, 256]> var_1976_cast_fp16 = pow(x = x_195, y = var_22_promoted_49_to_fp16)[name = tensor<string, []>("op_1976_cast_fp16")];
            tensor<int32, [1]> var_1978_axes_0 = const()[name = tensor<string, []>("op_1978_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1978_keep_dims_0 = const()[name = tensor<string, []>("op_1978_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_1978_cast_fp16 = reduce_mean(axes = var_1978_axes_0, keep_dims = var_1978_keep_dims_0, x = var_1976_cast_fp16)[name = tensor<string, []>("op_1978_cast_fp16")];
            tensor<fp16, []> var_1979_to_fp16 = const()[name = tensor<string, []>("op_1979_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_1980_cast_fp16 = add(x = var_1978_cast_fp16, y = var_1979_to_fp16)[name = tensor<string, []>("op_1980_cast_fp16")];
            tensor<fp32, []> var_1981_epsilon_0 = const()[name = tensor<string, []>("op_1981_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_1981_cast_fp16 = rsqrt(epsilon = var_1981_epsilon_0, x = var_1980_cast_fp16)[name = tensor<string, []>("op_1981_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_197_cast_fp16 = mul(x = x_195, y = var_1981_cast_fp16)[name = tensor<string, []>("output_197_cast_fp16")];
            tensor<fp16, [256]> var_1985_to_fp16 = const()[name = tensor<string, []>("op_1985_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940370560)))];
            tensor<fp16, [1, 8, 256, 256]> output_199_cast_fp16 = mul(x = output_197_cast_fp16, y = var_1985_to_fp16)[name = tensor<string, []>("output_199_cast_fp16")];
            tensor<fp16, []> var_22_promoted_50_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_50_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_199 = transpose(perm = x_199_perm_0, x = var_1967)[name = tensor<string, []>("transpose_102")];
            tensor<fp16, [1, 4, 256, 256]> var_1990_cast_fp16 = pow(x = x_199, y = var_22_promoted_50_to_fp16)[name = tensor<string, []>("op_1990_cast_fp16")];
            tensor<int32, [1]> var_1992_axes_0 = const()[name = tensor<string, []>("op_1992_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1992_keep_dims_0 = const()[name = tensor<string, []>("op_1992_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_1992_cast_fp16 = reduce_mean(axes = var_1992_axes_0, keep_dims = var_1992_keep_dims_0, x = var_1990_cast_fp16)[name = tensor<string, []>("op_1992_cast_fp16")];
            tensor<fp16, []> var_1993_to_fp16 = const()[name = tensor<string, []>("op_1993_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_1994_cast_fp16 = add(x = var_1992_cast_fp16, y = var_1993_to_fp16)[name = tensor<string, []>("op_1994_cast_fp16")];
            tensor<fp32, []> var_1995_epsilon_0 = const()[name = tensor<string, []>("op_1995_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_1995_cast_fp16 = rsqrt(epsilon = var_1995_epsilon_0, x = var_1994_cast_fp16)[name = tensor<string, []>("op_1995_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_201_cast_fp16 = mul(x = x_199, y = var_1995_cast_fp16)[name = tensor<string, []>("output_201_cast_fp16")];
            tensor<fp16, [256]> var_1999_to_fp16 = const()[name = tensor<string, []>("op_1999_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940371136)))];
            tensor<fp16, [1, 4, 256, 256]> output_203_cast_fp16 = mul(x = output_201_cast_fp16, y = var_1999_to_fp16)[name = tensor<string, []>("output_203_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_2004 = mul(x = output_199_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_2004")];
            tensor<int32, [4]> x1_33_begin_0 = const()[name = tensor<string, []>("x1_33_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_33_end_0 = const()[name = tensor<string, []>("x1_33_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_33_end_mask_0 = const()[name = tensor<string, []>("x1_33_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_33 = slice_by_index(begin = x1_33_begin_0, end = x1_33_end_0, end_mask = x1_33_end_mask_0, x = output_199_cast_fp16)[name = tensor<string, []>("x1_33")];
            tensor<int32, [4]> x2_33_begin_0 = const()[name = tensor<string, []>("x2_33_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_33_end_0 = const()[name = tensor<string, []>("x2_33_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_33_end_mask_0 = const()[name = tensor<string, []>("x2_33_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_33 = slice_by_index(begin = x2_33_begin_0, end = x2_33_end_0, end_mask = x2_33_end_mask_0, x = output_199_cast_fp16)[name = tensor<string, []>("x2_33")];
            tensor<fp16, []> const_219_promoted = const()[name = tensor<string, []>("const_219_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_2015 = mul(x = x2_33, y = const_219_promoted)[name = tensor<string, []>("op_2015")];
            tensor<bool, []> var_2017_interleave_0 = const()[name = tensor<string, []>("op_2017_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_2017 = concat(axis = var_24, interleave = var_2017_interleave_0, values = (var_2015, x1_33))[name = tensor<string, []>("op_2017")];
            tensor<fp16, [1, 8, 256, 256]> var_2018 = mul(x = var_2017, y = sin_7_palettized)[name = tensor<string, []>("op_2018")];
            tensor<fp16, [1, 8, 256, 256]> query_17 = add(x = var_2004, y = var_2018)[name = tensor<string, []>("query_17")];
            tensor<fp16, [1, 4, 256, 256]> var_2020 = mul(x = output_203_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_2020")];
            tensor<int32, [4]> x1_35_begin_0 = const()[name = tensor<string, []>("x1_35_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_35_end_0 = const()[name = tensor<string, []>("x1_35_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_35_end_mask_0 = const()[name = tensor<string, []>("x1_35_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_35 = slice_by_index(begin = x1_35_begin_0, end = x1_35_end_0, end_mask = x1_35_end_mask_0, x = output_203_cast_fp16)[name = tensor<string, []>("x1_35")];
            tensor<int32, [4]> x2_35_begin_0 = const()[name = tensor<string, []>("x2_35_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_35_end_0 = const()[name = tensor<string, []>("x2_35_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_35_end_mask_0 = const()[name = tensor<string, []>("x2_35_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_35 = slice_by_index(begin = x2_35_begin_0, end = x2_35_end_0, end_mask = x2_35_end_mask_0, x = output_203_cast_fp16)[name = tensor<string, []>("x2_35")];
            tensor<fp16, []> const_222_promoted = const()[name = tensor<string, []>("const_222_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_2031 = mul(x = x2_35, y = const_222_promoted)[name = tensor<string, []>("op_2031")];
            tensor<bool, []> var_2033_interleave_0 = const()[name = tensor<string, []>("op_2033_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_2033 = concat(axis = var_24, interleave = var_2033_interleave_0, values = (var_2031, x1_35))[name = tensor<string, []>("op_2033")];
            tensor<fp16, [1, 4, 256, 256]> var_2034 = mul(x = var_2033, y = sin_7_palettized)[name = tensor<string, []>("op_2034")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_115 = add(x = var_2020, y = var_2034)[name = tensor<string, []>("hidden_states_115")];
            tensor<int32, [1]> var_2043_axes_0 = const()[name = tensor<string, []>("op_2043_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_2043 = expand_dims(axes = var_2043_axes_0, x = hidden_states_115)[name = tensor<string, []>("op_2043")];
            tensor<int32, [5]> hidden_states_117_reps_0 = const()[name = tensor<string, []>("hidden_states_117_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_117 = tile(reps = hidden_states_117_reps_0, x = var_2043)[name = tensor<string, []>("hidden_states_117")];
            tensor<int32, [4]> var_2051 = const()[name = tensor<string, []>("op_2051"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_17 = reshape(shape = var_2051, x = hidden_states_117)[name = tensor<string, []>("key_states_17")];
            tensor<int32, [1]> var_2060_axes_0 = const()[name = tensor<string, []>("op_2060_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_119 = transpose(perm = hidden_states_119_perm_0, x = var_1972)[name = tensor<string, []>("transpose_101")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_2060 = expand_dims(axes = var_2060_axes_0, x = hidden_states_119)[name = tensor<string, []>("op_2060")];
            tensor<int32, [5]> hidden_states_121_reps_0 = const()[name = tensor<string, []>("hidden_states_121_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_121 = tile(reps = hidden_states_121_reps_0, x = var_2060)[name = tensor<string, []>("hidden_states_121")];
            tensor<int32, [4]> var_2068 = const()[name = tensor<string, []>("op_2068"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_17 = reshape(shape = var_2068, x = hidden_states_121)[name = tensor<string, []>("value_states_17")];
            tensor<bool, []> var_2071_transpose_x_1 = const()[name = tensor<string, []>("op_2071_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2071_transpose_y_1 = const()[name = tensor<string, []>("op_2071_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_2071 = matmul(transpose_x = var_2071_transpose_x_1, transpose_y = var_2071_transpose_y_1, x = query_17, y = key_states_17)[name = tensor<string, []>("op_2071")];
            tensor<fp16, []> var_2072_to_fp16 = const()[name = tensor<string, []>("op_2072_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_33_cast_fp16 = mul(x = var_2071, y = var_2072_to_fp16)[name = tensor<string, []>("attn_weights_33_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_97_cast_fp16 = add(x = attn_weights_33_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_97_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_2080_cast_fp16 = softmax(axis = var_24, x = input_97_cast_fp16)[name = tensor<string, []>("op_2080_cast_fp16")];
            tensor<bool, []> attn_output_33_transpose_x_0 = const()[name = tensor<string, []>("attn_output_33_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_33_transpose_y_0 = const()[name = tensor<string, []>("attn_output_33_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_33 = matmul(transpose_x = attn_output_33_transpose_x_0, transpose_y = attn_output_33_transpose_y_0, x = var_2080_cast_fp16, y = value_states_17)[name = tensor<string, []>("attn_output_33")];
            tensor<int32, [4]> var_2084_perm_0 = const()[name = tensor<string, []>("op_2084_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2086 = const()[name = tensor<string, []>("op_2086"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_2084 = transpose(perm = var_2084_perm_0, x = attn_output_33)[name = tensor<string, []>("transpose_100")];
            tensor<fp16, [1, 256, 2048]> var_2087 = reshape(shape = var_2086, x = var_2084)[name = tensor<string, []>("op_2087")];
            tensor<fp16, [1, 256, 2560]> x_203 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_8_self_attn_o_proj_weight_palettized, x = var_2087)[name = tensor<string, []>("linear_59")];
            tensor<fp16, []> var_22_promoted_51_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_51_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_2093_cast_fp16 = pow(x = x_203, y = var_22_promoted_51_to_fp16)[name = tensor<string, []>("op_2093_cast_fp16")];
            tensor<int32, [1]> var_2095_axes_0 = const()[name = tensor<string, []>("op_2095_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2095_keep_dims_0 = const()[name = tensor<string, []>("op_2095_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_2095_cast_fp16 = reduce_mean(axes = var_2095_axes_0, keep_dims = var_2095_keep_dims_0, x = var_2093_cast_fp16)[name = tensor<string, []>("op_2095_cast_fp16")];
            tensor<fp16, []> var_2096_to_fp16 = const()[name = tensor<string, []>("op_2096_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_2097_cast_fp16 = add(x = var_2095_cast_fp16, y = var_2096_to_fp16)[name = tensor<string, []>("op_2097_cast_fp16")];
            tensor<fp32, []> var_2098_epsilon_0 = const()[name = tensor<string, []>("op_2098_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_2098_cast_fp16 = rsqrt(epsilon = var_2098_epsilon_0, x = var_2097_cast_fp16)[name = tensor<string, []>("op_2098_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_205_cast_fp16 = mul(x = x_203, y = var_2098_cast_fp16)[name = tensor<string, []>("output_205_cast_fp16")];
            tensor<fp16, [2560]> var_2102_to_fp16 = const()[name = tensor<string, []>("op_2102_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940371712)))];
            tensor<fp16, [1, 256, 2560]> output_207_cast_fp16 = mul(x = output_205_cast_fp16, y = var_2102_to_fp16)[name = tensor<string, []>("output_207_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_207 = add(x = x_191, y = output_207_cast_fp16)[name = tensor<string, []>("x_207")];
            tensor<fp16, []> var_22_promoted_52_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_52_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_2108_cast_fp16 = pow(x = x_207, y = var_22_promoted_52_to_fp16)[name = tensor<string, []>("op_2108_cast_fp16")];
            tensor<int32, [1]> var_2110_axes_0 = const()[name = tensor<string, []>("op_2110_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2110_keep_dims_0 = const()[name = tensor<string, []>("op_2110_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_2110_cast_fp16 = reduce_mean(axes = var_2110_axes_0, keep_dims = var_2110_keep_dims_0, x = var_2108_cast_fp16)[name = tensor<string, []>("op_2110_cast_fp16")];
            tensor<fp16, []> var_2111_to_fp16 = const()[name = tensor<string, []>("op_2111_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_2112_cast_fp16 = add(x = var_2110_cast_fp16, y = var_2111_to_fp16)[name = tensor<string, []>("op_2112_cast_fp16")];
            tensor<fp32, []> var_2113_epsilon_0 = const()[name = tensor<string, []>("op_2113_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_2113_cast_fp16 = rsqrt(epsilon = var_2113_epsilon_0, x = var_2112_cast_fp16)[name = tensor<string, []>("op_2113_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_209_cast_fp16 = mul(x = x_207, y = var_2113_cast_fp16)[name = tensor<string, []>("output_209_cast_fp16")];
            tensor<fp16, [2560]> var_2117_to_fp16 = const()[name = tensor<string, []>("op_2117_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940376896)))];
            tensor<fp16, [1, 256, 2560]> output_211_cast_fp16 = mul(x = output_209_cast_fp16, y = var_2117_to_fp16)[name = tensor<string, []>("output_211_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_105 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_8_mlp_gate_proj_weight_palettized, x = output_211_cast_fp16)[name = tensor<string, []>("linear_60")];
            tensor<string, []> var_2125_mode_0 = const()[name = tensor<string, []>("op_2125_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_2125 = gelu(mode = var_2125_mode_0, x = input_105)[name = tensor<string, []>("op_2125")];
            tensor<fp16, [1, 256, 10240]> var_2127 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_8_mlp_up_proj_weight_palettized, x = output_211_cast_fp16)[name = tensor<string, []>("linear_61")];
            tensor<fp16, [1, 256, 10240]> input_107 = mul(x = var_2125, y = var_2127)[name = tensor<string, []>("input_107")];
            tensor<fp16, [1, 256, 2560]> x_211 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_8_mlp_down_proj_weight_palettized, x = input_107)[name = tensor<string, []>("linear_62")];
            tensor<fp16, []> var_22_promoted_53_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_53_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_2133_cast_fp16 = pow(x = x_211, y = var_22_promoted_53_to_fp16)[name = tensor<string, []>("op_2133_cast_fp16")];
            tensor<int32, [1]> var_2135_axes_0 = const()[name = tensor<string, []>("op_2135_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2135_keep_dims_0 = const()[name = tensor<string, []>("op_2135_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_2135_cast_fp16 = reduce_mean(axes = var_2135_axes_0, keep_dims = var_2135_keep_dims_0, x = var_2133_cast_fp16)[name = tensor<string, []>("op_2135_cast_fp16")];
            tensor<fp16, []> var_2136_to_fp16 = const()[name = tensor<string, []>("op_2136_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_2137_cast_fp16 = add(x = var_2135_cast_fp16, y = var_2136_to_fp16)[name = tensor<string, []>("op_2137_cast_fp16")];
            tensor<fp32, []> var_2138_epsilon_0 = const()[name = tensor<string, []>("op_2138_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_2138_cast_fp16 = rsqrt(epsilon = var_2138_epsilon_0, x = var_2137_cast_fp16)[name = tensor<string, []>("op_2138_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_213_cast_fp16 = mul(x = x_211, y = var_2138_cast_fp16)[name = tensor<string, []>("output_213_cast_fp16")];
            tensor<fp16, [2560]> var_2142_to_fp16 = const()[name = tensor<string, []>("op_2142_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940382080)))];
            tensor<fp16, [1, 256, 2560]> output_215_cast_fp16 = mul(x = output_213_cast_fp16, y = var_2142_to_fp16)[name = tensor<string, []>("output_215_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_215 = add(x = x_207, y = output_215_cast_fp16)[name = tensor<string, []>("x_215")];
            tensor<fp16, []> var_22_promoted_54_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_54_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_2154_cast_fp16 = pow(x = x_215, y = var_22_promoted_54_to_fp16)[name = tensor<string, []>("op_2154_cast_fp16")];
            tensor<int32, [1]> var_2156_axes_0 = const()[name = tensor<string, []>("op_2156_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2156_keep_dims_0 = const()[name = tensor<string, []>("op_2156_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_2156_cast_fp16 = reduce_mean(axes = var_2156_axes_0, keep_dims = var_2156_keep_dims_0, x = var_2154_cast_fp16)[name = tensor<string, []>("op_2156_cast_fp16")];
            tensor<fp16, []> var_2157_to_fp16 = const()[name = tensor<string, []>("op_2157_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_2158_cast_fp16 = add(x = var_2156_cast_fp16, y = var_2157_to_fp16)[name = tensor<string, []>("op_2158_cast_fp16")];
            tensor<fp32, []> var_2159_epsilon_0 = const()[name = tensor<string, []>("op_2159_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_2159_cast_fp16 = rsqrt(epsilon = var_2159_epsilon_0, x = var_2158_cast_fp16)[name = tensor<string, []>("op_2159_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_217_cast_fp16 = mul(x = x_215, y = var_2159_cast_fp16)[name = tensor<string, []>("output_217_cast_fp16")];
            tensor<fp16, [2560]> var_2163_to_fp16 = const()[name = tensor<string, []>("op_2163_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940387264)))];
            tensor<fp16, [1, 256, 2560]> output_219_cast_fp16 = mul(x = output_217_cast_fp16, y = var_2163_to_fp16)[name = tensor<string, []>("output_219_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_2175 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_9_self_attn_q_proj_weight_palettized, x = output_219_cast_fp16)[name = tensor<string, []>("linear_63")];
            tensor<int32, [4]> var_2176 = const()[name = tensor<string, []>("op_2176"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_2177 = reshape(shape = var_2176, x = var_2175)[name = tensor<string, []>("op_2177")];
            tensor<int32, [4]> x_219_perm_0 = const()[name = tensor<string, []>("x_219_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_2180 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_9_self_attn_k_proj_weight_palettized, x = output_219_cast_fp16)[name = tensor<string, []>("linear_64")];
            tensor<int32, [4]> var_2181 = const()[name = tensor<string, []>("op_2181"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_2182 = reshape(shape = var_2181, x = var_2180)[name = tensor<string, []>("op_2182")];
            tensor<int32, [4]> x_223_perm_0 = const()[name = tensor<string, []>("x_223_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_2185 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_9_self_attn_v_proj_weight_palettized, x = output_219_cast_fp16)[name = tensor<string, []>("linear_65")];
            tensor<int32, [4]> var_2186 = const()[name = tensor<string, []>("op_2186"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_2187 = reshape(shape = var_2186, x = var_2185)[name = tensor<string, []>("op_2187")];
            tensor<int32, [4]> hidden_states_133_perm_0 = const()[name = tensor<string, []>("hidden_states_133_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_55_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_55_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_219 = transpose(perm = x_219_perm_0, x = var_2177)[name = tensor<string, []>("transpose_99")];
            tensor<fp16, [1, 8, 256, 256]> var_2191_cast_fp16 = pow(x = x_219, y = var_22_promoted_55_to_fp16)[name = tensor<string, []>("op_2191_cast_fp16")];
            tensor<int32, [1]> var_2193_axes_0 = const()[name = tensor<string, []>("op_2193_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2193_keep_dims_0 = const()[name = tensor<string, []>("op_2193_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_2193_cast_fp16 = reduce_mean(axes = var_2193_axes_0, keep_dims = var_2193_keep_dims_0, x = var_2191_cast_fp16)[name = tensor<string, []>("op_2193_cast_fp16")];
            tensor<fp16, []> var_2194_to_fp16 = const()[name = tensor<string, []>("op_2194_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_2195_cast_fp16 = add(x = var_2193_cast_fp16, y = var_2194_to_fp16)[name = tensor<string, []>("op_2195_cast_fp16")];
            tensor<fp32, []> var_2196_epsilon_0 = const()[name = tensor<string, []>("op_2196_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_2196_cast_fp16 = rsqrt(epsilon = var_2196_epsilon_0, x = var_2195_cast_fp16)[name = tensor<string, []>("op_2196_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_221_cast_fp16 = mul(x = x_219, y = var_2196_cast_fp16)[name = tensor<string, []>("output_221_cast_fp16")];
            tensor<fp16, [256]> var_2200_to_fp16 = const()[name = tensor<string, []>("op_2200_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940392448)))];
            tensor<fp16, [1, 8, 256, 256]> output_223_cast_fp16 = mul(x = output_221_cast_fp16, y = var_2200_to_fp16)[name = tensor<string, []>("output_223_cast_fp16")];
            tensor<fp16, []> var_22_promoted_56_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_56_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_223 = transpose(perm = x_223_perm_0, x = var_2182)[name = tensor<string, []>("transpose_98")];
            tensor<fp16, [1, 4, 256, 256]> var_2205_cast_fp16 = pow(x = x_223, y = var_22_promoted_56_to_fp16)[name = tensor<string, []>("op_2205_cast_fp16")];
            tensor<int32, [1]> var_2207_axes_0 = const()[name = tensor<string, []>("op_2207_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2207_keep_dims_0 = const()[name = tensor<string, []>("op_2207_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_2207_cast_fp16 = reduce_mean(axes = var_2207_axes_0, keep_dims = var_2207_keep_dims_0, x = var_2205_cast_fp16)[name = tensor<string, []>("op_2207_cast_fp16")];
            tensor<fp16, []> var_2208_to_fp16 = const()[name = tensor<string, []>("op_2208_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_2209_cast_fp16 = add(x = var_2207_cast_fp16, y = var_2208_to_fp16)[name = tensor<string, []>("op_2209_cast_fp16")];
            tensor<fp32, []> var_2210_epsilon_0 = const()[name = tensor<string, []>("op_2210_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_2210_cast_fp16 = rsqrt(epsilon = var_2210_epsilon_0, x = var_2209_cast_fp16)[name = tensor<string, []>("op_2210_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_225_cast_fp16 = mul(x = x_223, y = var_2210_cast_fp16)[name = tensor<string, []>("output_225_cast_fp16")];
            tensor<fp16, [256]> var_2214_to_fp16 = const()[name = tensor<string, []>("op_2214_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940393024)))];
            tensor<fp16, [1, 4, 256, 256]> output_227_cast_fp16 = mul(x = output_225_cast_fp16, y = var_2214_to_fp16)[name = tensor<string, []>("output_227_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_2219 = mul(x = output_223_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_2219")];
            tensor<int32, [4]> x1_37_begin_0 = const()[name = tensor<string, []>("x1_37_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_37_end_0 = const()[name = tensor<string, []>("x1_37_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_37_end_mask_0 = const()[name = tensor<string, []>("x1_37_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_37 = slice_by_index(begin = x1_37_begin_0, end = x1_37_end_0, end_mask = x1_37_end_mask_0, x = output_223_cast_fp16)[name = tensor<string, []>("x1_37")];
            tensor<int32, [4]> x2_37_begin_0 = const()[name = tensor<string, []>("x2_37_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_37_end_0 = const()[name = tensor<string, []>("x2_37_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_37_end_mask_0 = const()[name = tensor<string, []>("x2_37_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_37 = slice_by_index(begin = x2_37_begin_0, end = x2_37_end_0, end_mask = x2_37_end_mask_0, x = output_223_cast_fp16)[name = tensor<string, []>("x2_37")];
            tensor<fp16, []> const_242_promoted = const()[name = tensor<string, []>("const_242_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_2230 = mul(x = x2_37, y = const_242_promoted)[name = tensor<string, []>("op_2230")];
            tensor<bool, []> var_2232_interleave_0 = const()[name = tensor<string, []>("op_2232_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_2232 = concat(axis = var_24, interleave = var_2232_interleave_0, values = (var_2230, x1_37))[name = tensor<string, []>("op_2232")];
            tensor<fp16, [1, 8, 256, 256]> var_2233 = mul(x = var_2232, y = sin_7_palettized)[name = tensor<string, []>("op_2233")];
            tensor<fp16, [1, 8, 256, 256]> query_19 = add(x = var_2219, y = var_2233)[name = tensor<string, []>("query_19")];
            tensor<fp16, [1, 4, 256, 256]> var_2235 = mul(x = output_227_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_2235")];
            tensor<int32, [4]> x1_39_begin_0 = const()[name = tensor<string, []>("x1_39_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_39_end_0 = const()[name = tensor<string, []>("x1_39_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_39_end_mask_0 = const()[name = tensor<string, []>("x1_39_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_39 = slice_by_index(begin = x1_39_begin_0, end = x1_39_end_0, end_mask = x1_39_end_mask_0, x = output_227_cast_fp16)[name = tensor<string, []>("x1_39")];
            tensor<int32, [4]> x2_39_begin_0 = const()[name = tensor<string, []>("x2_39_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_39_end_0 = const()[name = tensor<string, []>("x2_39_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_39_end_mask_0 = const()[name = tensor<string, []>("x2_39_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_39 = slice_by_index(begin = x2_39_begin_0, end = x2_39_end_0, end_mask = x2_39_end_mask_0, x = output_227_cast_fp16)[name = tensor<string, []>("x2_39")];
            tensor<fp16, []> const_245_promoted = const()[name = tensor<string, []>("const_245_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_2246 = mul(x = x2_39, y = const_245_promoted)[name = tensor<string, []>("op_2246")];
            tensor<bool, []> var_2248_interleave_0 = const()[name = tensor<string, []>("op_2248_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_2248 = concat(axis = var_24, interleave = var_2248_interleave_0, values = (var_2246, x1_39))[name = tensor<string, []>("op_2248")];
            tensor<fp16, [1, 4, 256, 256]> var_2249 = mul(x = var_2248, y = sin_7_palettized)[name = tensor<string, []>("op_2249")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_129 = add(x = var_2235, y = var_2249)[name = tensor<string, []>("hidden_states_129")];
            tensor<int32, [1]> var_2258_axes_0 = const()[name = tensor<string, []>("op_2258_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_2258 = expand_dims(axes = var_2258_axes_0, x = hidden_states_129)[name = tensor<string, []>("op_2258")];
            tensor<int32, [5]> hidden_states_131_reps_0 = const()[name = tensor<string, []>("hidden_states_131_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_131 = tile(reps = hidden_states_131_reps_0, x = var_2258)[name = tensor<string, []>("hidden_states_131")];
            tensor<int32, [4]> var_2266 = const()[name = tensor<string, []>("op_2266"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_19 = reshape(shape = var_2266, x = hidden_states_131)[name = tensor<string, []>("key_states_19")];
            tensor<int32, [1]> var_2275_axes_0 = const()[name = tensor<string, []>("op_2275_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_133 = transpose(perm = hidden_states_133_perm_0, x = var_2187)[name = tensor<string, []>("transpose_97")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_2275 = expand_dims(axes = var_2275_axes_0, x = hidden_states_133)[name = tensor<string, []>("op_2275")];
            tensor<int32, [5]> hidden_states_135_reps_0 = const()[name = tensor<string, []>("hidden_states_135_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_135 = tile(reps = hidden_states_135_reps_0, x = var_2275)[name = tensor<string, []>("hidden_states_135")];
            tensor<int32, [4]> var_2283 = const()[name = tensor<string, []>("op_2283"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_19 = reshape(shape = var_2283, x = hidden_states_135)[name = tensor<string, []>("value_states_19")];
            tensor<bool, []> var_2286_transpose_x_1 = const()[name = tensor<string, []>("op_2286_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2286_transpose_y_1 = const()[name = tensor<string, []>("op_2286_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_2286 = matmul(transpose_x = var_2286_transpose_x_1, transpose_y = var_2286_transpose_y_1, x = query_19, y = key_states_19)[name = tensor<string, []>("op_2286")];
            tensor<fp16, []> var_2287_to_fp16 = const()[name = tensor<string, []>("op_2287_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_37_cast_fp16 = mul(x = var_2286, y = var_2287_to_fp16)[name = tensor<string, []>("attn_weights_37_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_109_cast_fp16 = add(x = attn_weights_37_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_109_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_2295_cast_fp16 = softmax(axis = var_24, x = input_109_cast_fp16)[name = tensor<string, []>("op_2295_cast_fp16")];
            tensor<bool, []> attn_output_37_transpose_x_0 = const()[name = tensor<string, []>("attn_output_37_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_37_transpose_y_0 = const()[name = tensor<string, []>("attn_output_37_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_37 = matmul(transpose_x = attn_output_37_transpose_x_0, transpose_y = attn_output_37_transpose_y_0, x = var_2295_cast_fp16, y = value_states_19)[name = tensor<string, []>("attn_output_37")];
            tensor<int32, [4]> var_2299_perm_0 = const()[name = tensor<string, []>("op_2299_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2301 = const()[name = tensor<string, []>("op_2301"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_2299 = transpose(perm = var_2299_perm_0, x = attn_output_37)[name = tensor<string, []>("transpose_96")];
            tensor<fp16, [1, 256, 2048]> var_2302 = reshape(shape = var_2301, x = var_2299)[name = tensor<string, []>("op_2302")];
            tensor<fp16, [1, 256, 2560]> x_227 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_9_self_attn_o_proj_weight_palettized, x = var_2302)[name = tensor<string, []>("linear_66")];
            tensor<fp16, []> var_22_promoted_57_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_57_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_2308_cast_fp16 = pow(x = x_227, y = var_22_promoted_57_to_fp16)[name = tensor<string, []>("op_2308_cast_fp16")];
            tensor<int32, [1]> var_2310_axes_0 = const()[name = tensor<string, []>("op_2310_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2310_keep_dims_0 = const()[name = tensor<string, []>("op_2310_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_2310_cast_fp16 = reduce_mean(axes = var_2310_axes_0, keep_dims = var_2310_keep_dims_0, x = var_2308_cast_fp16)[name = tensor<string, []>("op_2310_cast_fp16")];
            tensor<fp16, []> var_2311_to_fp16 = const()[name = tensor<string, []>("op_2311_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_2312_cast_fp16 = add(x = var_2310_cast_fp16, y = var_2311_to_fp16)[name = tensor<string, []>("op_2312_cast_fp16")];
            tensor<fp32, []> var_2313_epsilon_0 = const()[name = tensor<string, []>("op_2313_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_2313_cast_fp16 = rsqrt(epsilon = var_2313_epsilon_0, x = var_2312_cast_fp16)[name = tensor<string, []>("op_2313_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_229_cast_fp16 = mul(x = x_227, y = var_2313_cast_fp16)[name = tensor<string, []>("output_229_cast_fp16")];
            tensor<fp16, [2560]> var_2317_to_fp16 = const()[name = tensor<string, []>("op_2317_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940393600)))];
            tensor<fp16, [1, 256, 2560]> output_231_cast_fp16 = mul(x = output_229_cast_fp16, y = var_2317_to_fp16)[name = tensor<string, []>("output_231_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_231 = add(x = x_215, y = output_231_cast_fp16)[name = tensor<string, []>("x_231")];
            tensor<fp16, []> var_22_promoted_58_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_58_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_2323_cast_fp16 = pow(x = x_231, y = var_22_promoted_58_to_fp16)[name = tensor<string, []>("op_2323_cast_fp16")];
            tensor<int32, [1]> var_2325_axes_0 = const()[name = tensor<string, []>("op_2325_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2325_keep_dims_0 = const()[name = tensor<string, []>("op_2325_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_2325_cast_fp16 = reduce_mean(axes = var_2325_axes_0, keep_dims = var_2325_keep_dims_0, x = var_2323_cast_fp16)[name = tensor<string, []>("op_2325_cast_fp16")];
            tensor<fp16, []> var_2326_to_fp16 = const()[name = tensor<string, []>("op_2326_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_2327_cast_fp16 = add(x = var_2325_cast_fp16, y = var_2326_to_fp16)[name = tensor<string, []>("op_2327_cast_fp16")];
            tensor<fp32, []> var_2328_epsilon_0 = const()[name = tensor<string, []>("op_2328_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_2328_cast_fp16 = rsqrt(epsilon = var_2328_epsilon_0, x = var_2327_cast_fp16)[name = tensor<string, []>("op_2328_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_233_cast_fp16 = mul(x = x_231, y = var_2328_cast_fp16)[name = tensor<string, []>("output_233_cast_fp16")];
            tensor<fp16, [2560]> var_2332_to_fp16 = const()[name = tensor<string, []>("op_2332_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940398784)))];
            tensor<fp16, [1, 256, 2560]> output_235_cast_fp16 = mul(x = output_233_cast_fp16, y = var_2332_to_fp16)[name = tensor<string, []>("output_235_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_117 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_9_mlp_gate_proj_weight_palettized, x = output_235_cast_fp16)[name = tensor<string, []>("linear_67")];
            tensor<string, []> var_2340_mode_0 = const()[name = tensor<string, []>("op_2340_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_2340 = gelu(mode = var_2340_mode_0, x = input_117)[name = tensor<string, []>("op_2340")];
            tensor<fp16, [1, 256, 10240]> var_2342 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_9_mlp_up_proj_weight_palettized, x = output_235_cast_fp16)[name = tensor<string, []>("linear_68")];
            tensor<fp16, [1, 256, 10240]> input_119 = mul(x = var_2340, y = var_2342)[name = tensor<string, []>("input_119")];
            tensor<fp16, [1, 256, 2560]> x_235 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_9_mlp_down_proj_weight_palettized, x = input_119)[name = tensor<string, []>("linear_69")];
            tensor<fp16, []> var_22_promoted_59_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_59_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_2348_cast_fp16 = pow(x = x_235, y = var_22_promoted_59_to_fp16)[name = tensor<string, []>("op_2348_cast_fp16")];
            tensor<int32, [1]> var_2350_axes_0 = const()[name = tensor<string, []>("op_2350_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2350_keep_dims_0 = const()[name = tensor<string, []>("op_2350_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_2350_cast_fp16 = reduce_mean(axes = var_2350_axes_0, keep_dims = var_2350_keep_dims_0, x = var_2348_cast_fp16)[name = tensor<string, []>("op_2350_cast_fp16")];
            tensor<fp16, []> var_2351_to_fp16 = const()[name = tensor<string, []>("op_2351_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_2352_cast_fp16 = add(x = var_2350_cast_fp16, y = var_2351_to_fp16)[name = tensor<string, []>("op_2352_cast_fp16")];
            tensor<fp32, []> var_2353_epsilon_0 = const()[name = tensor<string, []>("op_2353_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_2353_cast_fp16 = rsqrt(epsilon = var_2353_epsilon_0, x = var_2352_cast_fp16)[name = tensor<string, []>("op_2353_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_237_cast_fp16 = mul(x = x_235, y = var_2353_cast_fp16)[name = tensor<string, []>("output_237_cast_fp16")];
            tensor<fp16, [2560]> var_2357_to_fp16 = const()[name = tensor<string, []>("op_2357_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940403968)))];
            tensor<fp16, [1, 256, 2560]> output_239_cast_fp16 = mul(x = output_237_cast_fp16, y = var_2357_to_fp16)[name = tensor<string, []>("output_239_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_239 = add(x = x_231, y = output_239_cast_fp16)[name = tensor<string, []>("x_239")];
            tensor<fp16, []> var_22_promoted_60_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_60_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_2369_cast_fp16 = pow(x = x_239, y = var_22_promoted_60_to_fp16)[name = tensor<string, []>("op_2369_cast_fp16")];
            tensor<int32, [1]> var_2371_axes_0 = const()[name = tensor<string, []>("op_2371_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2371_keep_dims_0 = const()[name = tensor<string, []>("op_2371_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_2371_cast_fp16 = reduce_mean(axes = var_2371_axes_0, keep_dims = var_2371_keep_dims_0, x = var_2369_cast_fp16)[name = tensor<string, []>("op_2371_cast_fp16")];
            tensor<fp16, []> var_2372_to_fp16 = const()[name = tensor<string, []>("op_2372_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_2373_cast_fp16 = add(x = var_2371_cast_fp16, y = var_2372_to_fp16)[name = tensor<string, []>("op_2373_cast_fp16")];
            tensor<fp32, []> var_2374_epsilon_0 = const()[name = tensor<string, []>("op_2374_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_2374_cast_fp16 = rsqrt(epsilon = var_2374_epsilon_0, x = var_2373_cast_fp16)[name = tensor<string, []>("op_2374_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_241_cast_fp16 = mul(x = x_239, y = var_2374_cast_fp16)[name = tensor<string, []>("output_241_cast_fp16")];
            tensor<fp16, [2560]> var_2378_to_fp16 = const()[name = tensor<string, []>("op_2378_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940409152)))];
            tensor<fp16, [1, 256, 2560]> output_243_cast_fp16 = mul(x = output_241_cast_fp16, y = var_2378_to_fp16)[name = tensor<string, []>("output_243_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_2390 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_10_self_attn_q_proj_weight_palettized, x = output_243_cast_fp16)[name = tensor<string, []>("linear_70")];
            tensor<int32, [4]> var_2391 = const()[name = tensor<string, []>("op_2391"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_2392 = reshape(shape = var_2391, x = var_2390)[name = tensor<string, []>("op_2392")];
            tensor<int32, [4]> x_243_perm_0 = const()[name = tensor<string, []>("x_243_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_2395 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_10_self_attn_k_proj_weight_palettized, x = output_243_cast_fp16)[name = tensor<string, []>("linear_71")];
            tensor<int32, [4]> var_2396 = const()[name = tensor<string, []>("op_2396"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_2397 = reshape(shape = var_2396, x = var_2395)[name = tensor<string, []>("op_2397")];
            tensor<int32, [4]> x_247_perm_0 = const()[name = tensor<string, []>("x_247_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_2400 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_10_self_attn_v_proj_weight_palettized, x = output_243_cast_fp16)[name = tensor<string, []>("linear_72")];
            tensor<int32, [4]> var_2401 = const()[name = tensor<string, []>("op_2401"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_2402 = reshape(shape = var_2401, x = var_2400)[name = tensor<string, []>("op_2402")];
            tensor<int32, [4]> hidden_states_147_perm_0 = const()[name = tensor<string, []>("hidden_states_147_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_61_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_61_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_243 = transpose(perm = x_243_perm_0, x = var_2392)[name = tensor<string, []>("transpose_95")];
            tensor<fp16, [1, 8, 256, 256]> var_2406_cast_fp16 = pow(x = x_243, y = var_22_promoted_61_to_fp16)[name = tensor<string, []>("op_2406_cast_fp16")];
            tensor<int32, [1]> var_2408_axes_0 = const()[name = tensor<string, []>("op_2408_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2408_keep_dims_0 = const()[name = tensor<string, []>("op_2408_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_2408_cast_fp16 = reduce_mean(axes = var_2408_axes_0, keep_dims = var_2408_keep_dims_0, x = var_2406_cast_fp16)[name = tensor<string, []>("op_2408_cast_fp16")];
            tensor<fp16, []> var_2409_to_fp16 = const()[name = tensor<string, []>("op_2409_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_2410_cast_fp16 = add(x = var_2408_cast_fp16, y = var_2409_to_fp16)[name = tensor<string, []>("op_2410_cast_fp16")];
            tensor<fp32, []> var_2411_epsilon_0 = const()[name = tensor<string, []>("op_2411_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_2411_cast_fp16 = rsqrt(epsilon = var_2411_epsilon_0, x = var_2410_cast_fp16)[name = tensor<string, []>("op_2411_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_245_cast_fp16 = mul(x = x_243, y = var_2411_cast_fp16)[name = tensor<string, []>("output_245_cast_fp16")];
            tensor<fp16, [256]> var_2415_to_fp16 = const()[name = tensor<string, []>("op_2415_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940414336)))];
            tensor<fp16, [1, 8, 256, 256]> output_247_cast_fp16 = mul(x = output_245_cast_fp16, y = var_2415_to_fp16)[name = tensor<string, []>("output_247_cast_fp16")];
            tensor<fp16, []> var_22_promoted_62_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_62_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_247 = transpose(perm = x_247_perm_0, x = var_2397)[name = tensor<string, []>("transpose_94")];
            tensor<fp16, [1, 4, 256, 256]> var_2420_cast_fp16 = pow(x = x_247, y = var_22_promoted_62_to_fp16)[name = tensor<string, []>("op_2420_cast_fp16")];
            tensor<int32, [1]> var_2422_axes_0 = const()[name = tensor<string, []>("op_2422_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2422_keep_dims_0 = const()[name = tensor<string, []>("op_2422_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_2422_cast_fp16 = reduce_mean(axes = var_2422_axes_0, keep_dims = var_2422_keep_dims_0, x = var_2420_cast_fp16)[name = tensor<string, []>("op_2422_cast_fp16")];
            tensor<fp16, []> var_2423_to_fp16 = const()[name = tensor<string, []>("op_2423_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_2424_cast_fp16 = add(x = var_2422_cast_fp16, y = var_2423_to_fp16)[name = tensor<string, []>("op_2424_cast_fp16")];
            tensor<fp32, []> var_2425_epsilon_0 = const()[name = tensor<string, []>("op_2425_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_2425_cast_fp16 = rsqrt(epsilon = var_2425_epsilon_0, x = var_2424_cast_fp16)[name = tensor<string, []>("op_2425_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_249_cast_fp16 = mul(x = x_247, y = var_2425_cast_fp16)[name = tensor<string, []>("output_249_cast_fp16")];
            tensor<fp16, [256]> var_2429_to_fp16 = const()[name = tensor<string, []>("op_2429_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940414912)))];
            tensor<fp16, [1, 4, 256, 256]> output_251_cast_fp16 = mul(x = output_249_cast_fp16, y = var_2429_to_fp16)[name = tensor<string, []>("output_251_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_2434 = mul(x = output_247_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_2434")];
            tensor<int32, [4]> x1_41_begin_0 = const()[name = tensor<string, []>("x1_41_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_41_end_0 = const()[name = tensor<string, []>("x1_41_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_41_end_mask_0 = const()[name = tensor<string, []>("x1_41_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_41 = slice_by_index(begin = x1_41_begin_0, end = x1_41_end_0, end_mask = x1_41_end_mask_0, x = output_247_cast_fp16)[name = tensor<string, []>("x1_41")];
            tensor<int32, [4]> x2_41_begin_0 = const()[name = tensor<string, []>("x2_41_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_41_end_0 = const()[name = tensor<string, []>("x2_41_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_41_end_mask_0 = const()[name = tensor<string, []>("x2_41_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_41 = slice_by_index(begin = x2_41_begin_0, end = x2_41_end_0, end_mask = x2_41_end_mask_0, x = output_247_cast_fp16)[name = tensor<string, []>("x2_41")];
            tensor<fp16, []> const_265_promoted = const()[name = tensor<string, []>("const_265_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_2445 = mul(x = x2_41, y = const_265_promoted)[name = tensor<string, []>("op_2445")];
            tensor<bool, []> var_2447_interleave_0 = const()[name = tensor<string, []>("op_2447_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_2447 = concat(axis = var_24, interleave = var_2447_interleave_0, values = (var_2445, x1_41))[name = tensor<string, []>("op_2447")];
            tensor<fp16, [1, 8, 256, 256]> var_2448 = mul(x = var_2447, y = sin_7_palettized)[name = tensor<string, []>("op_2448")];
            tensor<fp16, [1, 8, 256, 256]> query_21 = add(x = var_2434, y = var_2448)[name = tensor<string, []>("query_21")];
            tensor<fp16, [1, 4, 256, 256]> var_2450 = mul(x = output_251_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_2450")];
            tensor<int32, [4]> x1_43_begin_0 = const()[name = tensor<string, []>("x1_43_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_43_end_0 = const()[name = tensor<string, []>("x1_43_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_43_end_mask_0 = const()[name = tensor<string, []>("x1_43_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_43 = slice_by_index(begin = x1_43_begin_0, end = x1_43_end_0, end_mask = x1_43_end_mask_0, x = output_251_cast_fp16)[name = tensor<string, []>("x1_43")];
            tensor<int32, [4]> x2_43_begin_0 = const()[name = tensor<string, []>("x2_43_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_43_end_0 = const()[name = tensor<string, []>("x2_43_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_43_end_mask_0 = const()[name = tensor<string, []>("x2_43_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_43 = slice_by_index(begin = x2_43_begin_0, end = x2_43_end_0, end_mask = x2_43_end_mask_0, x = output_251_cast_fp16)[name = tensor<string, []>("x2_43")];
            tensor<fp16, []> const_268_promoted = const()[name = tensor<string, []>("const_268_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_2461 = mul(x = x2_43, y = const_268_promoted)[name = tensor<string, []>("op_2461")];
            tensor<bool, []> var_2463_interleave_0 = const()[name = tensor<string, []>("op_2463_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_2463 = concat(axis = var_24, interleave = var_2463_interleave_0, values = (var_2461, x1_43))[name = tensor<string, []>("op_2463")];
            tensor<fp16, [1, 4, 256, 256]> var_2464 = mul(x = var_2463, y = sin_7_palettized)[name = tensor<string, []>("op_2464")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_143 = add(x = var_2450, y = var_2464)[name = tensor<string, []>("hidden_states_143")];
            tensor<int32, [1]> var_2473_axes_0 = const()[name = tensor<string, []>("op_2473_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_2473 = expand_dims(axes = var_2473_axes_0, x = hidden_states_143)[name = tensor<string, []>("op_2473")];
            tensor<int32, [5]> hidden_states_145_reps_0 = const()[name = tensor<string, []>("hidden_states_145_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_145 = tile(reps = hidden_states_145_reps_0, x = var_2473)[name = tensor<string, []>("hidden_states_145")];
            tensor<int32, [4]> var_2481 = const()[name = tensor<string, []>("op_2481"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_21 = reshape(shape = var_2481, x = hidden_states_145)[name = tensor<string, []>("key_states_21")];
            tensor<int32, [1]> var_2490_axes_0 = const()[name = tensor<string, []>("op_2490_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_147 = transpose(perm = hidden_states_147_perm_0, x = var_2402)[name = tensor<string, []>("transpose_93")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_2490 = expand_dims(axes = var_2490_axes_0, x = hidden_states_147)[name = tensor<string, []>("op_2490")];
            tensor<int32, [5]> hidden_states_149_reps_0 = const()[name = tensor<string, []>("hidden_states_149_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_149 = tile(reps = hidden_states_149_reps_0, x = var_2490)[name = tensor<string, []>("hidden_states_149")];
            tensor<int32, [4]> var_2498 = const()[name = tensor<string, []>("op_2498"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_21 = reshape(shape = var_2498, x = hidden_states_149)[name = tensor<string, []>("value_states_21")];
            tensor<bool, []> var_2501_transpose_x_1 = const()[name = tensor<string, []>("op_2501_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2501_transpose_y_1 = const()[name = tensor<string, []>("op_2501_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_2501 = matmul(transpose_x = var_2501_transpose_x_1, transpose_y = var_2501_transpose_y_1, x = query_21, y = key_states_21)[name = tensor<string, []>("op_2501")];
            tensor<fp16, []> var_2502_to_fp16 = const()[name = tensor<string, []>("op_2502_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_41_cast_fp16 = mul(x = var_2501, y = var_2502_to_fp16)[name = tensor<string, []>("attn_weights_41_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_121_cast_fp16 = add(x = attn_weights_41_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_121_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_2510_cast_fp16 = softmax(axis = var_24, x = input_121_cast_fp16)[name = tensor<string, []>("op_2510_cast_fp16")];
            tensor<bool, []> attn_output_41_transpose_x_0 = const()[name = tensor<string, []>("attn_output_41_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_41_transpose_y_0 = const()[name = tensor<string, []>("attn_output_41_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_41 = matmul(transpose_x = attn_output_41_transpose_x_0, transpose_y = attn_output_41_transpose_y_0, x = var_2510_cast_fp16, y = value_states_21)[name = tensor<string, []>("attn_output_41")];
            tensor<int32, [4]> var_2514_perm_0 = const()[name = tensor<string, []>("op_2514_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2516 = const()[name = tensor<string, []>("op_2516"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_2514 = transpose(perm = var_2514_perm_0, x = attn_output_41)[name = tensor<string, []>("transpose_92")];
            tensor<fp16, [1, 256, 2048]> var_2517 = reshape(shape = var_2516, x = var_2514)[name = tensor<string, []>("op_2517")];
            tensor<fp16, [1, 256, 2560]> x_251 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_10_self_attn_o_proj_weight_palettized, x = var_2517)[name = tensor<string, []>("linear_73")];
            tensor<fp16, []> var_22_promoted_63_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_63_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_2523_cast_fp16 = pow(x = x_251, y = var_22_promoted_63_to_fp16)[name = tensor<string, []>("op_2523_cast_fp16")];
            tensor<int32, [1]> var_2525_axes_0 = const()[name = tensor<string, []>("op_2525_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2525_keep_dims_0 = const()[name = tensor<string, []>("op_2525_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_2525_cast_fp16 = reduce_mean(axes = var_2525_axes_0, keep_dims = var_2525_keep_dims_0, x = var_2523_cast_fp16)[name = tensor<string, []>("op_2525_cast_fp16")];
            tensor<fp16, []> var_2526_to_fp16 = const()[name = tensor<string, []>("op_2526_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_2527_cast_fp16 = add(x = var_2525_cast_fp16, y = var_2526_to_fp16)[name = tensor<string, []>("op_2527_cast_fp16")];
            tensor<fp32, []> var_2528_epsilon_0 = const()[name = tensor<string, []>("op_2528_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_2528_cast_fp16 = rsqrt(epsilon = var_2528_epsilon_0, x = var_2527_cast_fp16)[name = tensor<string, []>("op_2528_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_253_cast_fp16 = mul(x = x_251, y = var_2528_cast_fp16)[name = tensor<string, []>("output_253_cast_fp16")];
            tensor<fp16, [2560]> var_2532_to_fp16 = const()[name = tensor<string, []>("op_2532_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940415488)))];
            tensor<fp16, [1, 256, 2560]> output_255_cast_fp16 = mul(x = output_253_cast_fp16, y = var_2532_to_fp16)[name = tensor<string, []>("output_255_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_255 = add(x = x_239, y = output_255_cast_fp16)[name = tensor<string, []>("x_255")];
            tensor<fp16, []> var_22_promoted_64_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_64_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_2538_cast_fp16 = pow(x = x_255, y = var_22_promoted_64_to_fp16)[name = tensor<string, []>("op_2538_cast_fp16")];
            tensor<int32, [1]> var_2540_axes_0 = const()[name = tensor<string, []>("op_2540_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2540_keep_dims_0 = const()[name = tensor<string, []>("op_2540_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_2540_cast_fp16 = reduce_mean(axes = var_2540_axes_0, keep_dims = var_2540_keep_dims_0, x = var_2538_cast_fp16)[name = tensor<string, []>("op_2540_cast_fp16")];
            tensor<fp16, []> var_2541_to_fp16 = const()[name = tensor<string, []>("op_2541_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_2542_cast_fp16 = add(x = var_2540_cast_fp16, y = var_2541_to_fp16)[name = tensor<string, []>("op_2542_cast_fp16")];
            tensor<fp32, []> var_2543_epsilon_0 = const()[name = tensor<string, []>("op_2543_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_2543_cast_fp16 = rsqrt(epsilon = var_2543_epsilon_0, x = var_2542_cast_fp16)[name = tensor<string, []>("op_2543_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_257_cast_fp16 = mul(x = x_255, y = var_2543_cast_fp16)[name = tensor<string, []>("output_257_cast_fp16")];
            tensor<fp16, [2560]> var_2547_to_fp16 = const()[name = tensor<string, []>("op_2547_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940420672)))];
            tensor<fp16, [1, 256, 2560]> output_259_cast_fp16 = mul(x = output_257_cast_fp16, y = var_2547_to_fp16)[name = tensor<string, []>("output_259_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_129 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_10_mlp_gate_proj_weight_palettized, x = output_259_cast_fp16)[name = tensor<string, []>("linear_74")];
            tensor<string, []> var_2555_mode_0 = const()[name = tensor<string, []>("op_2555_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_2555 = gelu(mode = var_2555_mode_0, x = input_129)[name = tensor<string, []>("op_2555")];
            tensor<fp16, [1, 256, 10240]> var_2557 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_10_mlp_up_proj_weight_palettized, x = output_259_cast_fp16)[name = tensor<string, []>("linear_75")];
            tensor<fp16, [1, 256, 10240]> input_131 = mul(x = var_2555, y = var_2557)[name = tensor<string, []>("input_131")];
            tensor<fp16, [1, 256, 2560]> x_259 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_10_mlp_down_proj_weight_palettized, x = input_131)[name = tensor<string, []>("linear_76")];
            tensor<fp16, []> var_22_promoted_65_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_65_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_2563_cast_fp16 = pow(x = x_259, y = var_22_promoted_65_to_fp16)[name = tensor<string, []>("op_2563_cast_fp16")];
            tensor<int32, [1]> var_2565_axes_0 = const()[name = tensor<string, []>("op_2565_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2565_keep_dims_0 = const()[name = tensor<string, []>("op_2565_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_2565_cast_fp16 = reduce_mean(axes = var_2565_axes_0, keep_dims = var_2565_keep_dims_0, x = var_2563_cast_fp16)[name = tensor<string, []>("op_2565_cast_fp16")];
            tensor<fp16, []> var_2566_to_fp16 = const()[name = tensor<string, []>("op_2566_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_2567_cast_fp16 = add(x = var_2565_cast_fp16, y = var_2566_to_fp16)[name = tensor<string, []>("op_2567_cast_fp16")];
            tensor<fp32, []> var_2568_epsilon_0 = const()[name = tensor<string, []>("op_2568_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_2568_cast_fp16 = rsqrt(epsilon = var_2568_epsilon_0, x = var_2567_cast_fp16)[name = tensor<string, []>("op_2568_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_261_cast_fp16 = mul(x = x_259, y = var_2568_cast_fp16)[name = tensor<string, []>("output_261_cast_fp16")];
            tensor<fp16, [2560]> var_2572_to_fp16 = const()[name = tensor<string, []>("op_2572_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940425856)))];
            tensor<fp16, [1, 256, 2560]> output_263_cast_fp16 = mul(x = output_261_cast_fp16, y = var_2572_to_fp16)[name = tensor<string, []>("output_263_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_263 = add(x = x_255, y = output_263_cast_fp16)[name = tensor<string, []>("x_263")];
            tensor<fp16, []> var_22_promoted_66_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_66_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_2584_cast_fp16 = pow(x = x_263, y = var_22_promoted_66_to_fp16)[name = tensor<string, []>("op_2584_cast_fp16")];
            tensor<int32, [1]> var_2586_axes_0 = const()[name = tensor<string, []>("op_2586_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2586_keep_dims_0 = const()[name = tensor<string, []>("op_2586_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_2586_cast_fp16 = reduce_mean(axes = var_2586_axes_0, keep_dims = var_2586_keep_dims_0, x = var_2584_cast_fp16)[name = tensor<string, []>("op_2586_cast_fp16")];
            tensor<fp16, []> var_2587_to_fp16 = const()[name = tensor<string, []>("op_2587_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_2588_cast_fp16 = add(x = var_2586_cast_fp16, y = var_2587_to_fp16)[name = tensor<string, []>("op_2588_cast_fp16")];
            tensor<fp32, []> var_2589_epsilon_0 = const()[name = tensor<string, []>("op_2589_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_2589_cast_fp16 = rsqrt(epsilon = var_2589_epsilon_0, x = var_2588_cast_fp16)[name = tensor<string, []>("op_2589_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_265_cast_fp16 = mul(x = x_263, y = var_2589_cast_fp16)[name = tensor<string, []>("output_265_cast_fp16")];
            tensor<fp16, [2560]> var_2593_to_fp16 = const()[name = tensor<string, []>("op_2593_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940431040)))];
            tensor<fp16, [1, 256, 2560]> output_267_cast_fp16 = mul(x = output_265_cast_fp16, y = var_2593_to_fp16)[name = tensor<string, []>("output_267_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_2605 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_11_self_attn_q_proj_weight_palettized, x = output_267_cast_fp16)[name = tensor<string, []>("linear_77")];
            tensor<int32, [4]> var_2606 = const()[name = tensor<string, []>("op_2606"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_2607 = reshape(shape = var_2606, x = var_2605)[name = tensor<string, []>("op_2607")];
            tensor<int32, [4]> x_267_perm_0 = const()[name = tensor<string, []>("x_267_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_2610 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_11_self_attn_k_proj_weight_palettized, x = output_267_cast_fp16)[name = tensor<string, []>("linear_78")];
            tensor<int32, [4]> var_2611 = const()[name = tensor<string, []>("op_2611"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_2612 = reshape(shape = var_2611, x = var_2610)[name = tensor<string, []>("op_2612")];
            tensor<int32, [4]> x_271_perm_0 = const()[name = tensor<string, []>("x_271_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_2615 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_11_self_attn_v_proj_weight_palettized, x = output_267_cast_fp16)[name = tensor<string, []>("linear_79")];
            tensor<int32, [4]> var_2616 = const()[name = tensor<string, []>("op_2616"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_2617 = reshape(shape = var_2616, x = var_2615)[name = tensor<string, []>("op_2617")];
            tensor<int32, [4]> hidden_states_161_perm_0 = const()[name = tensor<string, []>("hidden_states_161_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_67_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_67_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_267 = transpose(perm = x_267_perm_0, x = var_2607)[name = tensor<string, []>("transpose_91")];
            tensor<fp16, [1, 8, 256, 256]> var_2621_cast_fp16 = pow(x = x_267, y = var_22_promoted_67_to_fp16)[name = tensor<string, []>("op_2621_cast_fp16")];
            tensor<int32, [1]> var_2623_axes_0 = const()[name = tensor<string, []>("op_2623_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2623_keep_dims_0 = const()[name = tensor<string, []>("op_2623_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_2623_cast_fp16 = reduce_mean(axes = var_2623_axes_0, keep_dims = var_2623_keep_dims_0, x = var_2621_cast_fp16)[name = tensor<string, []>("op_2623_cast_fp16")];
            tensor<fp16, []> var_2624_to_fp16 = const()[name = tensor<string, []>("op_2624_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_2625_cast_fp16 = add(x = var_2623_cast_fp16, y = var_2624_to_fp16)[name = tensor<string, []>("op_2625_cast_fp16")];
            tensor<fp32, []> var_2626_epsilon_0 = const()[name = tensor<string, []>("op_2626_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_2626_cast_fp16 = rsqrt(epsilon = var_2626_epsilon_0, x = var_2625_cast_fp16)[name = tensor<string, []>("op_2626_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_269_cast_fp16 = mul(x = x_267, y = var_2626_cast_fp16)[name = tensor<string, []>("output_269_cast_fp16")];
            tensor<fp16, [256]> var_2630_to_fp16 = const()[name = tensor<string, []>("op_2630_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940436224)))];
            tensor<fp16, [1, 8, 256, 256]> output_271_cast_fp16 = mul(x = output_269_cast_fp16, y = var_2630_to_fp16)[name = tensor<string, []>("output_271_cast_fp16")];
            tensor<fp16, []> var_22_promoted_68_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_68_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_271 = transpose(perm = x_271_perm_0, x = var_2612)[name = tensor<string, []>("transpose_90")];
            tensor<fp16, [1, 4, 256, 256]> var_2635_cast_fp16 = pow(x = x_271, y = var_22_promoted_68_to_fp16)[name = tensor<string, []>("op_2635_cast_fp16")];
            tensor<int32, [1]> var_2637_axes_0 = const()[name = tensor<string, []>("op_2637_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2637_keep_dims_0 = const()[name = tensor<string, []>("op_2637_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_2637_cast_fp16 = reduce_mean(axes = var_2637_axes_0, keep_dims = var_2637_keep_dims_0, x = var_2635_cast_fp16)[name = tensor<string, []>("op_2637_cast_fp16")];
            tensor<fp16, []> var_2638_to_fp16 = const()[name = tensor<string, []>("op_2638_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_2639_cast_fp16 = add(x = var_2637_cast_fp16, y = var_2638_to_fp16)[name = tensor<string, []>("op_2639_cast_fp16")];
            tensor<fp32, []> var_2640_epsilon_0 = const()[name = tensor<string, []>("op_2640_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_2640_cast_fp16 = rsqrt(epsilon = var_2640_epsilon_0, x = var_2639_cast_fp16)[name = tensor<string, []>("op_2640_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_273_cast_fp16 = mul(x = x_271, y = var_2640_cast_fp16)[name = tensor<string, []>("output_273_cast_fp16")];
            tensor<fp16, [256]> var_2644_to_fp16 = const()[name = tensor<string, []>("op_2644_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940436800)))];
            tensor<fp16, [1, 4, 256, 256]> output_275_cast_fp16 = mul(x = output_273_cast_fp16, y = var_2644_to_fp16)[name = tensor<string, []>("output_275_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_2649 = mul(x = output_271_cast_fp16, y = cos_19_palettized)[name = tensor<string, []>("op_2649")];
            tensor<int32, [4]> x1_45_begin_0 = const()[name = tensor<string, []>("x1_45_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_45_end_0 = const()[name = tensor<string, []>("x1_45_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_45_end_mask_0 = const()[name = tensor<string, []>("x1_45_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_45 = slice_by_index(begin = x1_45_begin_0, end = x1_45_end_0, end_mask = x1_45_end_mask_0, x = output_271_cast_fp16)[name = tensor<string, []>("x1_45")];
            tensor<int32, [4]> x2_45_begin_0 = const()[name = tensor<string, []>("x2_45_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_45_end_0 = const()[name = tensor<string, []>("x2_45_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_45_end_mask_0 = const()[name = tensor<string, []>("x2_45_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_45 = slice_by_index(begin = x2_45_begin_0, end = x2_45_end_0, end_mask = x2_45_end_mask_0, x = output_271_cast_fp16)[name = tensor<string, []>("x2_45")];
            tensor<fp16, []> const_288_promoted = const()[name = tensor<string, []>("const_288_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_2660 = mul(x = x2_45, y = const_288_promoted)[name = tensor<string, []>("op_2660")];
            tensor<bool, []> var_2662_interleave_0 = const()[name = tensor<string, []>("op_2662_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_2662 = concat(axis = var_24, interleave = var_2662_interleave_0, values = (var_2660, x1_45))[name = tensor<string, []>("op_2662")];
            tensor<fp16, [1, 8, 256, 256]> var_2663 = mul(x = var_2662, y = sin_19_palettized)[name = tensor<string, []>("op_2663")];
            tensor<fp16, [1, 8, 256, 256]> query_23 = add(x = var_2649, y = var_2663)[name = tensor<string, []>("query_23")];
            tensor<fp16, [1, 4, 256, 256]> var_2665 = mul(x = output_275_cast_fp16, y = cos_19_palettized)[name = tensor<string, []>("op_2665")];
            tensor<int32, [4]> x1_47_begin_0 = const()[name = tensor<string, []>("x1_47_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_47_end_0 = const()[name = tensor<string, []>("x1_47_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_47_end_mask_0 = const()[name = tensor<string, []>("x1_47_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_47 = slice_by_index(begin = x1_47_begin_0, end = x1_47_end_0, end_mask = x1_47_end_mask_0, x = output_275_cast_fp16)[name = tensor<string, []>("x1_47")];
            tensor<int32, [4]> x2_47_begin_0 = const()[name = tensor<string, []>("x2_47_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_47_end_0 = const()[name = tensor<string, []>("x2_47_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_47_end_mask_0 = const()[name = tensor<string, []>("x2_47_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_47 = slice_by_index(begin = x2_47_begin_0, end = x2_47_end_0, end_mask = x2_47_end_mask_0, x = output_275_cast_fp16)[name = tensor<string, []>("x2_47")];
            tensor<fp16, []> const_291_promoted = const()[name = tensor<string, []>("const_291_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_2676 = mul(x = x2_47, y = const_291_promoted)[name = tensor<string, []>("op_2676")];
            tensor<bool, []> var_2678_interleave_0 = const()[name = tensor<string, []>("op_2678_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_2678 = concat(axis = var_24, interleave = var_2678_interleave_0, values = (var_2676, x1_47))[name = tensor<string, []>("op_2678")];
            tensor<fp16, [1, 4, 256, 256]> var_2679 = mul(x = var_2678, y = sin_19_palettized)[name = tensor<string, []>("op_2679")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_157 = add(x = var_2665, y = var_2679)[name = tensor<string, []>("hidden_states_157")];
            tensor<int32, [1]> var_2688_axes_0 = const()[name = tensor<string, []>("op_2688_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_2688 = expand_dims(axes = var_2688_axes_0, x = hidden_states_157)[name = tensor<string, []>("op_2688")];
            tensor<int32, [5]> hidden_states_159_reps_0 = const()[name = tensor<string, []>("hidden_states_159_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_159 = tile(reps = hidden_states_159_reps_0, x = var_2688)[name = tensor<string, []>("hidden_states_159")];
            tensor<int32, [4]> var_2696 = const()[name = tensor<string, []>("op_2696"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_23 = reshape(shape = var_2696, x = hidden_states_159)[name = tensor<string, []>("key_states_23")];
            tensor<int32, [1]> var_2705_axes_0 = const()[name = tensor<string, []>("op_2705_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_161 = transpose(perm = hidden_states_161_perm_0, x = var_2617)[name = tensor<string, []>("transpose_89")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_2705 = expand_dims(axes = var_2705_axes_0, x = hidden_states_161)[name = tensor<string, []>("op_2705")];
            tensor<int32, [5]> hidden_states_163_reps_0 = const()[name = tensor<string, []>("hidden_states_163_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_163 = tile(reps = hidden_states_163_reps_0, x = var_2705)[name = tensor<string, []>("hidden_states_163")];
            tensor<int32, [4]> var_2713 = const()[name = tensor<string, []>("op_2713"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_23 = reshape(shape = var_2713, x = hidden_states_163)[name = tensor<string, []>("value_states_23")];
            tensor<bool, []> var_2716_transpose_x_1 = const()[name = tensor<string, []>("op_2716_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2716_transpose_y_1 = const()[name = tensor<string, []>("op_2716_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_2716 = matmul(transpose_x = var_2716_transpose_x_1, transpose_y = var_2716_transpose_y_1, x = query_23, y = key_states_23)[name = tensor<string, []>("op_2716")];
            tensor<fp16, []> var_2717_to_fp16 = const()[name = tensor<string, []>("op_2717_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_45_cast_fp16 = mul(x = var_2716, y = var_2717_to_fp16)[name = tensor<string, []>("attn_weights_45_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_133_cast_fp16 = add(x = attn_weights_45_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_133_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_2725_cast_fp16 = softmax(axis = var_24, x = input_133_cast_fp16)[name = tensor<string, []>("op_2725_cast_fp16")];
            tensor<bool, []> attn_output_45_transpose_x_0 = const()[name = tensor<string, []>("attn_output_45_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_45_transpose_y_0 = const()[name = tensor<string, []>("attn_output_45_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_45 = matmul(transpose_x = attn_output_45_transpose_x_0, transpose_y = attn_output_45_transpose_y_0, x = var_2725_cast_fp16, y = value_states_23)[name = tensor<string, []>("attn_output_45")];
            tensor<int32, [4]> var_2729_perm_0 = const()[name = tensor<string, []>("op_2729_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2731 = const()[name = tensor<string, []>("op_2731"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_2729 = transpose(perm = var_2729_perm_0, x = attn_output_45)[name = tensor<string, []>("transpose_88")];
            tensor<fp16, [1, 256, 2048]> var_2732 = reshape(shape = var_2731, x = var_2729)[name = tensor<string, []>("op_2732")];
            tensor<fp16, [1, 256, 2560]> x_275 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_11_self_attn_o_proj_weight_palettized, x = var_2732)[name = tensor<string, []>("linear_80")];
            tensor<fp16, []> var_22_promoted_69_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_69_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_2738_cast_fp16 = pow(x = x_275, y = var_22_promoted_69_to_fp16)[name = tensor<string, []>("op_2738_cast_fp16")];
            tensor<int32, [1]> var_2740_axes_0 = const()[name = tensor<string, []>("op_2740_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2740_keep_dims_0 = const()[name = tensor<string, []>("op_2740_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_2740_cast_fp16 = reduce_mean(axes = var_2740_axes_0, keep_dims = var_2740_keep_dims_0, x = var_2738_cast_fp16)[name = tensor<string, []>("op_2740_cast_fp16")];
            tensor<fp16, []> var_2741_to_fp16 = const()[name = tensor<string, []>("op_2741_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_2742_cast_fp16 = add(x = var_2740_cast_fp16, y = var_2741_to_fp16)[name = tensor<string, []>("op_2742_cast_fp16")];
            tensor<fp32, []> var_2743_epsilon_0 = const()[name = tensor<string, []>("op_2743_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_2743_cast_fp16 = rsqrt(epsilon = var_2743_epsilon_0, x = var_2742_cast_fp16)[name = tensor<string, []>("op_2743_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_277_cast_fp16 = mul(x = x_275, y = var_2743_cast_fp16)[name = tensor<string, []>("output_277_cast_fp16")];
            tensor<fp16, [2560]> var_2747_to_fp16 = const()[name = tensor<string, []>("op_2747_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940437376)))];
            tensor<fp16, [1, 256, 2560]> output_279_cast_fp16 = mul(x = output_277_cast_fp16, y = var_2747_to_fp16)[name = tensor<string, []>("output_279_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_279 = add(x = x_263, y = output_279_cast_fp16)[name = tensor<string, []>("x_279")];
            tensor<fp16, []> var_22_promoted_70_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_70_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_2753_cast_fp16 = pow(x = x_279, y = var_22_promoted_70_to_fp16)[name = tensor<string, []>("op_2753_cast_fp16")];
            tensor<int32, [1]> var_2755_axes_0 = const()[name = tensor<string, []>("op_2755_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2755_keep_dims_0 = const()[name = tensor<string, []>("op_2755_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_2755_cast_fp16 = reduce_mean(axes = var_2755_axes_0, keep_dims = var_2755_keep_dims_0, x = var_2753_cast_fp16)[name = tensor<string, []>("op_2755_cast_fp16")];
            tensor<fp16, []> var_2756_to_fp16 = const()[name = tensor<string, []>("op_2756_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_2757_cast_fp16 = add(x = var_2755_cast_fp16, y = var_2756_to_fp16)[name = tensor<string, []>("op_2757_cast_fp16")];
            tensor<fp32, []> var_2758_epsilon_0 = const()[name = tensor<string, []>("op_2758_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_2758_cast_fp16 = rsqrt(epsilon = var_2758_epsilon_0, x = var_2757_cast_fp16)[name = tensor<string, []>("op_2758_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_281_cast_fp16 = mul(x = x_279, y = var_2758_cast_fp16)[name = tensor<string, []>("output_281_cast_fp16")];
            tensor<fp16, [2560]> var_2762_to_fp16 = const()[name = tensor<string, []>("op_2762_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940442560)))];
            tensor<fp16, [1, 256, 2560]> output_283_cast_fp16 = mul(x = output_281_cast_fp16, y = var_2762_to_fp16)[name = tensor<string, []>("output_283_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_141 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_11_mlp_gate_proj_weight_palettized, x = output_283_cast_fp16)[name = tensor<string, []>("linear_81")];
            tensor<string, []> var_2770_mode_0 = const()[name = tensor<string, []>("op_2770_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_2770 = gelu(mode = var_2770_mode_0, x = input_141)[name = tensor<string, []>("op_2770")];
            tensor<fp16, [1, 256, 10240]> var_2772 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_11_mlp_up_proj_weight_palettized, x = output_283_cast_fp16)[name = tensor<string, []>("linear_82")];
            tensor<fp16, [1, 256, 10240]> input_143 = mul(x = var_2770, y = var_2772)[name = tensor<string, []>("input_143")];
            tensor<fp16, [1, 256, 2560]> x_283 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_11_mlp_down_proj_weight_palettized, x = input_143)[name = tensor<string, []>("linear_83")];
            tensor<fp16, []> var_22_promoted_71_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_71_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_2778_cast_fp16 = pow(x = x_283, y = var_22_promoted_71_to_fp16)[name = tensor<string, []>("op_2778_cast_fp16")];
            tensor<int32, [1]> var_2780_axes_0 = const()[name = tensor<string, []>("op_2780_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2780_keep_dims_0 = const()[name = tensor<string, []>("op_2780_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_2780_cast_fp16 = reduce_mean(axes = var_2780_axes_0, keep_dims = var_2780_keep_dims_0, x = var_2778_cast_fp16)[name = tensor<string, []>("op_2780_cast_fp16")];
            tensor<fp16, []> var_2781_to_fp16 = const()[name = tensor<string, []>("op_2781_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_2782_cast_fp16 = add(x = var_2780_cast_fp16, y = var_2781_to_fp16)[name = tensor<string, []>("op_2782_cast_fp16")];
            tensor<fp32, []> var_2783_epsilon_0 = const()[name = tensor<string, []>("op_2783_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_2783_cast_fp16 = rsqrt(epsilon = var_2783_epsilon_0, x = var_2782_cast_fp16)[name = tensor<string, []>("op_2783_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_285_cast_fp16 = mul(x = x_283, y = var_2783_cast_fp16)[name = tensor<string, []>("output_285_cast_fp16")];
            tensor<fp16, [2560]> var_2787_to_fp16 = const()[name = tensor<string, []>("op_2787_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940447744)))];
            tensor<fp16, [1, 256, 2560]> output_287_cast_fp16 = mul(x = output_285_cast_fp16, y = var_2787_to_fp16)[name = tensor<string, []>("output_287_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_287 = add(x = x_279, y = output_287_cast_fp16)[name = tensor<string, []>("x_287")];
            tensor<fp16, []> var_22_promoted_72_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_72_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_2799_cast_fp16 = pow(x = x_287, y = var_22_promoted_72_to_fp16)[name = tensor<string, []>("op_2799_cast_fp16")];
            tensor<int32, [1]> var_2801_axes_0 = const()[name = tensor<string, []>("op_2801_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2801_keep_dims_0 = const()[name = tensor<string, []>("op_2801_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_2801_cast_fp16 = reduce_mean(axes = var_2801_axes_0, keep_dims = var_2801_keep_dims_0, x = var_2799_cast_fp16)[name = tensor<string, []>("op_2801_cast_fp16")];
            tensor<fp16, []> var_2802_to_fp16 = const()[name = tensor<string, []>("op_2802_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_2803_cast_fp16 = add(x = var_2801_cast_fp16, y = var_2802_to_fp16)[name = tensor<string, []>("op_2803_cast_fp16")];
            tensor<fp32, []> var_2804_epsilon_0 = const()[name = tensor<string, []>("op_2804_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_2804_cast_fp16 = rsqrt(epsilon = var_2804_epsilon_0, x = var_2803_cast_fp16)[name = tensor<string, []>("op_2804_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_289_cast_fp16 = mul(x = x_287, y = var_2804_cast_fp16)[name = tensor<string, []>("output_289_cast_fp16")];
            tensor<fp16, [2560]> var_2808_to_fp16 = const()[name = tensor<string, []>("op_2808_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940452928)))];
            tensor<fp16, [1, 256, 2560]> output_291_cast_fp16 = mul(x = output_289_cast_fp16, y = var_2808_to_fp16)[name = tensor<string, []>("output_291_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_2820 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_12_self_attn_q_proj_weight_palettized, x = output_291_cast_fp16)[name = tensor<string, []>("linear_84")];
            tensor<int32, [4]> var_2821 = const()[name = tensor<string, []>("op_2821"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_2822 = reshape(shape = var_2821, x = var_2820)[name = tensor<string, []>("op_2822")];
            tensor<int32, [4]> x_291_perm_0 = const()[name = tensor<string, []>("x_291_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_2825 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_12_self_attn_k_proj_weight_palettized, x = output_291_cast_fp16)[name = tensor<string, []>("linear_85")];
            tensor<int32, [4]> var_2826 = const()[name = tensor<string, []>("op_2826"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_2827 = reshape(shape = var_2826, x = var_2825)[name = tensor<string, []>("op_2827")];
            tensor<int32, [4]> x_295_perm_0 = const()[name = tensor<string, []>("x_295_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_2830 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_12_self_attn_v_proj_weight_palettized, x = output_291_cast_fp16)[name = tensor<string, []>("linear_86")];
            tensor<int32, [4]> var_2831 = const()[name = tensor<string, []>("op_2831"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_2832 = reshape(shape = var_2831, x = var_2830)[name = tensor<string, []>("op_2832")];
            tensor<int32, [4]> hidden_states_175_perm_0 = const()[name = tensor<string, []>("hidden_states_175_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_73_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_73_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_291 = transpose(perm = x_291_perm_0, x = var_2822)[name = tensor<string, []>("transpose_87")];
            tensor<fp16, [1, 8, 256, 256]> var_2836_cast_fp16 = pow(x = x_291, y = var_22_promoted_73_to_fp16)[name = tensor<string, []>("op_2836_cast_fp16")];
            tensor<int32, [1]> var_2838_axes_0 = const()[name = tensor<string, []>("op_2838_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2838_keep_dims_0 = const()[name = tensor<string, []>("op_2838_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_2838_cast_fp16 = reduce_mean(axes = var_2838_axes_0, keep_dims = var_2838_keep_dims_0, x = var_2836_cast_fp16)[name = tensor<string, []>("op_2838_cast_fp16")];
            tensor<fp16, []> var_2839_to_fp16 = const()[name = tensor<string, []>("op_2839_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_2840_cast_fp16 = add(x = var_2838_cast_fp16, y = var_2839_to_fp16)[name = tensor<string, []>("op_2840_cast_fp16")];
            tensor<fp32, []> var_2841_epsilon_0 = const()[name = tensor<string, []>("op_2841_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_2841_cast_fp16 = rsqrt(epsilon = var_2841_epsilon_0, x = var_2840_cast_fp16)[name = tensor<string, []>("op_2841_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_293_cast_fp16 = mul(x = x_291, y = var_2841_cast_fp16)[name = tensor<string, []>("output_293_cast_fp16")];
            tensor<fp16, [256]> var_2845_to_fp16 = const()[name = tensor<string, []>("op_2845_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940458112)))];
            tensor<fp16, [1, 8, 256, 256]> output_295_cast_fp16 = mul(x = output_293_cast_fp16, y = var_2845_to_fp16)[name = tensor<string, []>("output_295_cast_fp16")];
            tensor<fp16, []> var_22_promoted_74_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_74_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_295 = transpose(perm = x_295_perm_0, x = var_2827)[name = tensor<string, []>("transpose_86")];
            tensor<fp16, [1, 4, 256, 256]> var_2850_cast_fp16 = pow(x = x_295, y = var_22_promoted_74_to_fp16)[name = tensor<string, []>("op_2850_cast_fp16")];
            tensor<int32, [1]> var_2852_axes_0 = const()[name = tensor<string, []>("op_2852_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2852_keep_dims_0 = const()[name = tensor<string, []>("op_2852_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_2852_cast_fp16 = reduce_mean(axes = var_2852_axes_0, keep_dims = var_2852_keep_dims_0, x = var_2850_cast_fp16)[name = tensor<string, []>("op_2852_cast_fp16")];
            tensor<fp16, []> var_2853_to_fp16 = const()[name = tensor<string, []>("op_2853_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_2854_cast_fp16 = add(x = var_2852_cast_fp16, y = var_2853_to_fp16)[name = tensor<string, []>("op_2854_cast_fp16")];
            tensor<fp32, []> var_2855_epsilon_0 = const()[name = tensor<string, []>("op_2855_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_2855_cast_fp16 = rsqrt(epsilon = var_2855_epsilon_0, x = var_2854_cast_fp16)[name = tensor<string, []>("op_2855_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_297_cast_fp16 = mul(x = x_295, y = var_2855_cast_fp16)[name = tensor<string, []>("output_297_cast_fp16")];
            tensor<fp16, [256]> var_2859_to_fp16 = const()[name = tensor<string, []>("op_2859_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940458688)))];
            tensor<fp16, [1, 4, 256, 256]> output_299_cast_fp16 = mul(x = output_297_cast_fp16, y = var_2859_to_fp16)[name = tensor<string, []>("output_299_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_2864 = mul(x = output_295_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_2864")];
            tensor<int32, [4]> x1_49_begin_0 = const()[name = tensor<string, []>("x1_49_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_49_end_0 = const()[name = tensor<string, []>("x1_49_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_49_end_mask_0 = const()[name = tensor<string, []>("x1_49_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_49 = slice_by_index(begin = x1_49_begin_0, end = x1_49_end_0, end_mask = x1_49_end_mask_0, x = output_295_cast_fp16)[name = tensor<string, []>("x1_49")];
            tensor<int32, [4]> x2_49_begin_0 = const()[name = tensor<string, []>("x2_49_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_49_end_0 = const()[name = tensor<string, []>("x2_49_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_49_end_mask_0 = const()[name = tensor<string, []>("x2_49_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_49 = slice_by_index(begin = x2_49_begin_0, end = x2_49_end_0, end_mask = x2_49_end_mask_0, x = output_295_cast_fp16)[name = tensor<string, []>("x2_49")];
            tensor<fp16, []> const_311_promoted = const()[name = tensor<string, []>("const_311_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_2875 = mul(x = x2_49, y = const_311_promoted)[name = tensor<string, []>("op_2875")];
            tensor<bool, []> var_2877_interleave_0 = const()[name = tensor<string, []>("op_2877_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_2877 = concat(axis = var_24, interleave = var_2877_interleave_0, values = (var_2875, x1_49))[name = tensor<string, []>("op_2877")];
            tensor<fp16, [1, 8, 256, 256]> var_2878 = mul(x = var_2877, y = sin_7_palettized)[name = tensor<string, []>("op_2878")];
            tensor<fp16, [1, 8, 256, 256]> query_25 = add(x = var_2864, y = var_2878)[name = tensor<string, []>("query_25")];
            tensor<fp16, [1, 4, 256, 256]> var_2880 = mul(x = output_299_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_2880")];
            tensor<int32, [4]> x1_51_begin_0 = const()[name = tensor<string, []>("x1_51_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_51_end_0 = const()[name = tensor<string, []>("x1_51_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_51_end_mask_0 = const()[name = tensor<string, []>("x1_51_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_51 = slice_by_index(begin = x1_51_begin_0, end = x1_51_end_0, end_mask = x1_51_end_mask_0, x = output_299_cast_fp16)[name = tensor<string, []>("x1_51")];
            tensor<int32, [4]> x2_51_begin_0 = const()[name = tensor<string, []>("x2_51_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_51_end_0 = const()[name = tensor<string, []>("x2_51_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_51_end_mask_0 = const()[name = tensor<string, []>("x2_51_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_51 = slice_by_index(begin = x2_51_begin_0, end = x2_51_end_0, end_mask = x2_51_end_mask_0, x = output_299_cast_fp16)[name = tensor<string, []>("x2_51")];
            tensor<fp16, []> const_314_promoted = const()[name = tensor<string, []>("const_314_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_2891 = mul(x = x2_51, y = const_314_promoted)[name = tensor<string, []>("op_2891")];
            tensor<bool, []> var_2893_interleave_0 = const()[name = tensor<string, []>("op_2893_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_2893 = concat(axis = var_24, interleave = var_2893_interleave_0, values = (var_2891, x1_51))[name = tensor<string, []>("op_2893")];
            tensor<fp16, [1, 4, 256, 256]> var_2894 = mul(x = var_2893, y = sin_7_palettized)[name = tensor<string, []>("op_2894")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_171 = add(x = var_2880, y = var_2894)[name = tensor<string, []>("hidden_states_171")];
            tensor<int32, [1]> var_2903_axes_0 = const()[name = tensor<string, []>("op_2903_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_2903 = expand_dims(axes = var_2903_axes_0, x = hidden_states_171)[name = tensor<string, []>("op_2903")];
            tensor<int32, [5]> hidden_states_173_reps_0 = const()[name = tensor<string, []>("hidden_states_173_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_173 = tile(reps = hidden_states_173_reps_0, x = var_2903)[name = tensor<string, []>("hidden_states_173")];
            tensor<int32, [4]> var_2911 = const()[name = tensor<string, []>("op_2911"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_25 = reshape(shape = var_2911, x = hidden_states_173)[name = tensor<string, []>("key_states_25")];
            tensor<int32, [1]> var_2920_axes_0 = const()[name = tensor<string, []>("op_2920_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_175 = transpose(perm = hidden_states_175_perm_0, x = var_2832)[name = tensor<string, []>("transpose_85")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_2920 = expand_dims(axes = var_2920_axes_0, x = hidden_states_175)[name = tensor<string, []>("op_2920")];
            tensor<int32, [5]> hidden_states_177_reps_0 = const()[name = tensor<string, []>("hidden_states_177_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_177 = tile(reps = hidden_states_177_reps_0, x = var_2920)[name = tensor<string, []>("hidden_states_177")];
            tensor<int32, [4]> var_2928 = const()[name = tensor<string, []>("op_2928"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_25 = reshape(shape = var_2928, x = hidden_states_177)[name = tensor<string, []>("value_states_25")];
            tensor<bool, []> var_2931_transpose_x_1 = const()[name = tensor<string, []>("op_2931_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2931_transpose_y_1 = const()[name = tensor<string, []>("op_2931_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_2931 = matmul(transpose_x = var_2931_transpose_x_1, transpose_y = var_2931_transpose_y_1, x = query_25, y = key_states_25)[name = tensor<string, []>("op_2931")];
            tensor<fp16, []> var_2932_to_fp16 = const()[name = tensor<string, []>("op_2932_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_49_cast_fp16 = mul(x = var_2931, y = var_2932_to_fp16)[name = tensor<string, []>("attn_weights_49_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_145_cast_fp16 = add(x = attn_weights_49_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_145_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_2940_cast_fp16 = softmax(axis = var_24, x = input_145_cast_fp16)[name = tensor<string, []>("op_2940_cast_fp16")];
            tensor<bool, []> attn_output_49_transpose_x_0 = const()[name = tensor<string, []>("attn_output_49_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_49_transpose_y_0 = const()[name = tensor<string, []>("attn_output_49_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_49 = matmul(transpose_x = attn_output_49_transpose_x_0, transpose_y = attn_output_49_transpose_y_0, x = var_2940_cast_fp16, y = value_states_25)[name = tensor<string, []>("attn_output_49")];
            tensor<int32, [4]> var_2944_perm_0 = const()[name = tensor<string, []>("op_2944_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2946 = const()[name = tensor<string, []>("op_2946"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_2944 = transpose(perm = var_2944_perm_0, x = attn_output_49)[name = tensor<string, []>("transpose_84")];
            tensor<fp16, [1, 256, 2048]> var_2947 = reshape(shape = var_2946, x = var_2944)[name = tensor<string, []>("op_2947")];
            tensor<fp16, [1, 256, 2560]> x_299 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_12_self_attn_o_proj_weight_palettized, x = var_2947)[name = tensor<string, []>("linear_87")];
            tensor<fp16, []> var_22_promoted_75_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_75_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_2953_cast_fp16 = pow(x = x_299, y = var_22_promoted_75_to_fp16)[name = tensor<string, []>("op_2953_cast_fp16")];
            tensor<int32, [1]> var_2955_axes_0 = const()[name = tensor<string, []>("op_2955_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2955_keep_dims_0 = const()[name = tensor<string, []>("op_2955_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_2955_cast_fp16 = reduce_mean(axes = var_2955_axes_0, keep_dims = var_2955_keep_dims_0, x = var_2953_cast_fp16)[name = tensor<string, []>("op_2955_cast_fp16")];
            tensor<fp16, []> var_2956_to_fp16 = const()[name = tensor<string, []>("op_2956_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_2957_cast_fp16 = add(x = var_2955_cast_fp16, y = var_2956_to_fp16)[name = tensor<string, []>("op_2957_cast_fp16")];
            tensor<fp32, []> var_2958_epsilon_0 = const()[name = tensor<string, []>("op_2958_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_2958_cast_fp16 = rsqrt(epsilon = var_2958_epsilon_0, x = var_2957_cast_fp16)[name = tensor<string, []>("op_2958_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_301_cast_fp16 = mul(x = x_299, y = var_2958_cast_fp16)[name = tensor<string, []>("output_301_cast_fp16")];
            tensor<fp16, [2560]> var_2962_to_fp16 = const()[name = tensor<string, []>("op_2962_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940459264)))];
            tensor<fp16, [1, 256, 2560]> output_303_cast_fp16 = mul(x = output_301_cast_fp16, y = var_2962_to_fp16)[name = tensor<string, []>("output_303_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_303 = add(x = x_287, y = output_303_cast_fp16)[name = tensor<string, []>("x_303")];
            tensor<fp16, []> var_22_promoted_76_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_76_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_2968_cast_fp16 = pow(x = x_303, y = var_22_promoted_76_to_fp16)[name = tensor<string, []>("op_2968_cast_fp16")];
            tensor<int32, [1]> var_2970_axes_0 = const()[name = tensor<string, []>("op_2970_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2970_keep_dims_0 = const()[name = tensor<string, []>("op_2970_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_2970_cast_fp16 = reduce_mean(axes = var_2970_axes_0, keep_dims = var_2970_keep_dims_0, x = var_2968_cast_fp16)[name = tensor<string, []>("op_2970_cast_fp16")];
            tensor<fp16, []> var_2971_to_fp16 = const()[name = tensor<string, []>("op_2971_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_2972_cast_fp16 = add(x = var_2970_cast_fp16, y = var_2971_to_fp16)[name = tensor<string, []>("op_2972_cast_fp16")];
            tensor<fp32, []> var_2973_epsilon_0 = const()[name = tensor<string, []>("op_2973_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_2973_cast_fp16 = rsqrt(epsilon = var_2973_epsilon_0, x = var_2972_cast_fp16)[name = tensor<string, []>("op_2973_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_305_cast_fp16 = mul(x = x_303, y = var_2973_cast_fp16)[name = tensor<string, []>("output_305_cast_fp16")];
            tensor<fp16, [2560]> var_2977_to_fp16 = const()[name = tensor<string, []>("op_2977_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940464448)))];
            tensor<fp16, [1, 256, 2560]> output_307_cast_fp16 = mul(x = output_305_cast_fp16, y = var_2977_to_fp16)[name = tensor<string, []>("output_307_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_153 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_12_mlp_gate_proj_weight_palettized, x = output_307_cast_fp16)[name = tensor<string, []>("linear_88")];
            tensor<string, []> var_2985_mode_0 = const()[name = tensor<string, []>("op_2985_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_2985 = gelu(mode = var_2985_mode_0, x = input_153)[name = tensor<string, []>("op_2985")];
            tensor<fp16, [1, 256, 10240]> var_2987 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_12_mlp_up_proj_weight_palettized, x = output_307_cast_fp16)[name = tensor<string, []>("linear_89")];
            tensor<fp16, [1, 256, 10240]> input_155 = mul(x = var_2985, y = var_2987)[name = tensor<string, []>("input_155")];
            tensor<fp16, [1, 256, 2560]> x_307 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_12_mlp_down_proj_weight_palettized, x = input_155)[name = tensor<string, []>("linear_90")];
            tensor<fp16, []> var_22_promoted_77_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_77_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_2993_cast_fp16 = pow(x = x_307, y = var_22_promoted_77_to_fp16)[name = tensor<string, []>("op_2993_cast_fp16")];
            tensor<int32, [1]> var_2995_axes_0 = const()[name = tensor<string, []>("op_2995_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_2995_keep_dims_0 = const()[name = tensor<string, []>("op_2995_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_2995_cast_fp16 = reduce_mean(axes = var_2995_axes_0, keep_dims = var_2995_keep_dims_0, x = var_2993_cast_fp16)[name = tensor<string, []>("op_2995_cast_fp16")];
            tensor<fp16, []> var_2996_to_fp16 = const()[name = tensor<string, []>("op_2996_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_2997_cast_fp16 = add(x = var_2995_cast_fp16, y = var_2996_to_fp16)[name = tensor<string, []>("op_2997_cast_fp16")];
            tensor<fp32, []> var_2998_epsilon_0 = const()[name = tensor<string, []>("op_2998_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_2998_cast_fp16 = rsqrt(epsilon = var_2998_epsilon_0, x = var_2997_cast_fp16)[name = tensor<string, []>("op_2998_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_309_cast_fp16 = mul(x = x_307, y = var_2998_cast_fp16)[name = tensor<string, []>("output_309_cast_fp16")];
            tensor<fp16, [2560]> var_3002_to_fp16 = const()[name = tensor<string, []>("op_3002_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940469632)))];
            tensor<fp16, [1, 256, 2560]> output_311_cast_fp16 = mul(x = output_309_cast_fp16, y = var_3002_to_fp16)[name = tensor<string, []>("output_311_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_311 = add(x = x_303, y = output_311_cast_fp16)[name = tensor<string, []>("x_311")];
            tensor<fp16, []> var_22_promoted_78_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_78_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_3014_cast_fp16 = pow(x = x_311, y = var_22_promoted_78_to_fp16)[name = tensor<string, []>("op_3014_cast_fp16")];
            tensor<int32, [1]> var_3016_axes_0 = const()[name = tensor<string, []>("op_3016_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3016_keep_dims_0 = const()[name = tensor<string, []>("op_3016_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_3016_cast_fp16 = reduce_mean(axes = var_3016_axes_0, keep_dims = var_3016_keep_dims_0, x = var_3014_cast_fp16)[name = tensor<string, []>("op_3016_cast_fp16")];
            tensor<fp16, []> var_3017_to_fp16 = const()[name = tensor<string, []>("op_3017_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_3018_cast_fp16 = add(x = var_3016_cast_fp16, y = var_3017_to_fp16)[name = tensor<string, []>("op_3018_cast_fp16")];
            tensor<fp32, []> var_3019_epsilon_0 = const()[name = tensor<string, []>("op_3019_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_3019_cast_fp16 = rsqrt(epsilon = var_3019_epsilon_0, x = var_3018_cast_fp16)[name = tensor<string, []>("op_3019_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_313_cast_fp16 = mul(x = x_311, y = var_3019_cast_fp16)[name = tensor<string, []>("output_313_cast_fp16")];
            tensor<fp16, [2560]> var_3023_to_fp16 = const()[name = tensor<string, []>("op_3023_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940474816)))];
            tensor<fp16, [1, 256, 2560]> output_315_cast_fp16 = mul(x = output_313_cast_fp16, y = var_3023_to_fp16)[name = tensor<string, []>("output_315_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_3035 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_13_self_attn_q_proj_weight_palettized, x = output_315_cast_fp16)[name = tensor<string, []>("linear_91")];
            tensor<int32, [4]> var_3036 = const()[name = tensor<string, []>("op_3036"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_3037 = reshape(shape = var_3036, x = var_3035)[name = tensor<string, []>("op_3037")];
            tensor<int32, [4]> x_315_perm_0 = const()[name = tensor<string, []>("x_315_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_3040 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_13_self_attn_k_proj_weight_palettized, x = output_315_cast_fp16)[name = tensor<string, []>("linear_92")];
            tensor<int32, [4]> var_3041 = const()[name = tensor<string, []>("op_3041"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_3042 = reshape(shape = var_3041, x = var_3040)[name = tensor<string, []>("op_3042")];
            tensor<int32, [4]> x_319_perm_0 = const()[name = tensor<string, []>("x_319_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_3045 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_13_self_attn_v_proj_weight_palettized, x = output_315_cast_fp16)[name = tensor<string, []>("linear_93")];
            tensor<int32, [4]> var_3046 = const()[name = tensor<string, []>("op_3046"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_3047 = reshape(shape = var_3046, x = var_3045)[name = tensor<string, []>("op_3047")];
            tensor<int32, [4]> hidden_states_189_perm_0 = const()[name = tensor<string, []>("hidden_states_189_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_79_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_79_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_315 = transpose(perm = x_315_perm_0, x = var_3037)[name = tensor<string, []>("transpose_83")];
            tensor<fp16, [1, 8, 256, 256]> var_3051_cast_fp16 = pow(x = x_315, y = var_22_promoted_79_to_fp16)[name = tensor<string, []>("op_3051_cast_fp16")];
            tensor<int32, [1]> var_3053_axes_0 = const()[name = tensor<string, []>("op_3053_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3053_keep_dims_0 = const()[name = tensor<string, []>("op_3053_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_3053_cast_fp16 = reduce_mean(axes = var_3053_axes_0, keep_dims = var_3053_keep_dims_0, x = var_3051_cast_fp16)[name = tensor<string, []>("op_3053_cast_fp16")];
            tensor<fp16, []> var_3054_to_fp16 = const()[name = tensor<string, []>("op_3054_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_3055_cast_fp16 = add(x = var_3053_cast_fp16, y = var_3054_to_fp16)[name = tensor<string, []>("op_3055_cast_fp16")];
            tensor<fp32, []> var_3056_epsilon_0 = const()[name = tensor<string, []>("op_3056_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_3056_cast_fp16 = rsqrt(epsilon = var_3056_epsilon_0, x = var_3055_cast_fp16)[name = tensor<string, []>("op_3056_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_317_cast_fp16 = mul(x = x_315, y = var_3056_cast_fp16)[name = tensor<string, []>("output_317_cast_fp16")];
            tensor<fp16, [256]> var_3060_to_fp16 = const()[name = tensor<string, []>("op_3060_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940480000)))];
            tensor<fp16, [1, 8, 256, 256]> output_319_cast_fp16 = mul(x = output_317_cast_fp16, y = var_3060_to_fp16)[name = tensor<string, []>("output_319_cast_fp16")];
            tensor<fp16, []> var_22_promoted_80_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_80_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_319 = transpose(perm = x_319_perm_0, x = var_3042)[name = tensor<string, []>("transpose_82")];
            tensor<fp16, [1, 4, 256, 256]> var_3065_cast_fp16 = pow(x = x_319, y = var_22_promoted_80_to_fp16)[name = tensor<string, []>("op_3065_cast_fp16")];
            tensor<int32, [1]> var_3067_axes_0 = const()[name = tensor<string, []>("op_3067_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3067_keep_dims_0 = const()[name = tensor<string, []>("op_3067_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_3067_cast_fp16 = reduce_mean(axes = var_3067_axes_0, keep_dims = var_3067_keep_dims_0, x = var_3065_cast_fp16)[name = tensor<string, []>("op_3067_cast_fp16")];
            tensor<fp16, []> var_3068_to_fp16 = const()[name = tensor<string, []>("op_3068_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_3069_cast_fp16 = add(x = var_3067_cast_fp16, y = var_3068_to_fp16)[name = tensor<string, []>("op_3069_cast_fp16")];
            tensor<fp32, []> var_3070_epsilon_0 = const()[name = tensor<string, []>("op_3070_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_3070_cast_fp16 = rsqrt(epsilon = var_3070_epsilon_0, x = var_3069_cast_fp16)[name = tensor<string, []>("op_3070_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_321_cast_fp16 = mul(x = x_319, y = var_3070_cast_fp16)[name = tensor<string, []>("output_321_cast_fp16")];
            tensor<fp16, [256]> var_3074_to_fp16 = const()[name = tensor<string, []>("op_3074_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940480576)))];
            tensor<fp16, [1, 4, 256, 256]> output_323_cast_fp16 = mul(x = output_321_cast_fp16, y = var_3074_to_fp16)[name = tensor<string, []>("output_323_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_3079 = mul(x = output_319_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_3079")];
            tensor<int32, [4]> x1_53_begin_0 = const()[name = tensor<string, []>("x1_53_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_53_end_0 = const()[name = tensor<string, []>("x1_53_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_53_end_mask_0 = const()[name = tensor<string, []>("x1_53_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_53 = slice_by_index(begin = x1_53_begin_0, end = x1_53_end_0, end_mask = x1_53_end_mask_0, x = output_319_cast_fp16)[name = tensor<string, []>("x1_53")];
            tensor<int32, [4]> x2_53_begin_0 = const()[name = tensor<string, []>("x2_53_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_53_end_0 = const()[name = tensor<string, []>("x2_53_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_53_end_mask_0 = const()[name = tensor<string, []>("x2_53_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_53 = slice_by_index(begin = x2_53_begin_0, end = x2_53_end_0, end_mask = x2_53_end_mask_0, x = output_319_cast_fp16)[name = tensor<string, []>("x2_53")];
            tensor<fp16, []> const_334_promoted = const()[name = tensor<string, []>("const_334_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_3090 = mul(x = x2_53, y = const_334_promoted)[name = tensor<string, []>("op_3090")];
            tensor<bool, []> var_3092_interleave_0 = const()[name = tensor<string, []>("op_3092_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_3092 = concat(axis = var_24, interleave = var_3092_interleave_0, values = (var_3090, x1_53))[name = tensor<string, []>("op_3092")];
            tensor<fp16, [1, 8, 256, 256]> var_3093 = mul(x = var_3092, y = sin_7_palettized)[name = tensor<string, []>("op_3093")];
            tensor<fp16, [1, 8, 256, 256]> query_27 = add(x = var_3079, y = var_3093)[name = tensor<string, []>("query_27")];
            tensor<fp16, [1, 4, 256, 256]> var_3095 = mul(x = output_323_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_3095")];
            tensor<int32, [4]> x1_55_begin_0 = const()[name = tensor<string, []>("x1_55_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_55_end_0 = const()[name = tensor<string, []>("x1_55_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_55_end_mask_0 = const()[name = tensor<string, []>("x1_55_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_55 = slice_by_index(begin = x1_55_begin_0, end = x1_55_end_0, end_mask = x1_55_end_mask_0, x = output_323_cast_fp16)[name = tensor<string, []>("x1_55")];
            tensor<int32, [4]> x2_55_begin_0 = const()[name = tensor<string, []>("x2_55_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_55_end_0 = const()[name = tensor<string, []>("x2_55_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_55_end_mask_0 = const()[name = tensor<string, []>("x2_55_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_55 = slice_by_index(begin = x2_55_begin_0, end = x2_55_end_0, end_mask = x2_55_end_mask_0, x = output_323_cast_fp16)[name = tensor<string, []>("x2_55")];
            tensor<fp16, []> const_337_promoted = const()[name = tensor<string, []>("const_337_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_3106 = mul(x = x2_55, y = const_337_promoted)[name = tensor<string, []>("op_3106")];
            tensor<bool, []> var_3108_interleave_0 = const()[name = tensor<string, []>("op_3108_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_3108 = concat(axis = var_24, interleave = var_3108_interleave_0, values = (var_3106, x1_55))[name = tensor<string, []>("op_3108")];
            tensor<fp16, [1, 4, 256, 256]> var_3109 = mul(x = var_3108, y = sin_7_palettized)[name = tensor<string, []>("op_3109")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_185 = add(x = var_3095, y = var_3109)[name = tensor<string, []>("hidden_states_185")];
            tensor<int32, [1]> var_3118_axes_0 = const()[name = tensor<string, []>("op_3118_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_3118 = expand_dims(axes = var_3118_axes_0, x = hidden_states_185)[name = tensor<string, []>("op_3118")];
            tensor<int32, [5]> hidden_states_187_reps_0 = const()[name = tensor<string, []>("hidden_states_187_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_187 = tile(reps = hidden_states_187_reps_0, x = var_3118)[name = tensor<string, []>("hidden_states_187")];
            tensor<int32, [4]> var_3126 = const()[name = tensor<string, []>("op_3126"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_27 = reshape(shape = var_3126, x = hidden_states_187)[name = tensor<string, []>("key_states_27")];
            tensor<int32, [1]> var_3135_axes_0 = const()[name = tensor<string, []>("op_3135_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_189 = transpose(perm = hidden_states_189_perm_0, x = var_3047)[name = tensor<string, []>("transpose_81")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_3135 = expand_dims(axes = var_3135_axes_0, x = hidden_states_189)[name = tensor<string, []>("op_3135")];
            tensor<int32, [5]> hidden_states_191_reps_0 = const()[name = tensor<string, []>("hidden_states_191_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_191 = tile(reps = hidden_states_191_reps_0, x = var_3135)[name = tensor<string, []>("hidden_states_191")];
            tensor<int32, [4]> var_3143 = const()[name = tensor<string, []>("op_3143"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_27 = reshape(shape = var_3143, x = hidden_states_191)[name = tensor<string, []>("value_states_27")];
            tensor<bool, []> var_3146_transpose_x_1 = const()[name = tensor<string, []>("op_3146_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_3146_transpose_y_1 = const()[name = tensor<string, []>("op_3146_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_3146 = matmul(transpose_x = var_3146_transpose_x_1, transpose_y = var_3146_transpose_y_1, x = query_27, y = key_states_27)[name = tensor<string, []>("op_3146")];
            tensor<fp16, []> var_3147_to_fp16 = const()[name = tensor<string, []>("op_3147_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_53_cast_fp16 = mul(x = var_3146, y = var_3147_to_fp16)[name = tensor<string, []>("attn_weights_53_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_157_cast_fp16 = add(x = attn_weights_53_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_157_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_3155_cast_fp16 = softmax(axis = var_24, x = input_157_cast_fp16)[name = tensor<string, []>("op_3155_cast_fp16")];
            tensor<bool, []> attn_output_53_transpose_x_0 = const()[name = tensor<string, []>("attn_output_53_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_53_transpose_y_0 = const()[name = tensor<string, []>("attn_output_53_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_53 = matmul(transpose_x = attn_output_53_transpose_x_0, transpose_y = attn_output_53_transpose_y_0, x = var_3155_cast_fp16, y = value_states_27)[name = tensor<string, []>("attn_output_53")];
            tensor<int32, [4]> var_3159_perm_0 = const()[name = tensor<string, []>("op_3159_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3161 = const()[name = tensor<string, []>("op_3161"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_3159 = transpose(perm = var_3159_perm_0, x = attn_output_53)[name = tensor<string, []>("transpose_80")];
            tensor<fp16, [1, 256, 2048]> var_3162 = reshape(shape = var_3161, x = var_3159)[name = tensor<string, []>("op_3162")];
            tensor<fp16, [1, 256, 2560]> x_323 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_13_self_attn_o_proj_weight_palettized, x = var_3162)[name = tensor<string, []>("linear_94")];
            tensor<fp16, []> var_22_promoted_81_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_81_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_3168_cast_fp16 = pow(x = x_323, y = var_22_promoted_81_to_fp16)[name = tensor<string, []>("op_3168_cast_fp16")];
            tensor<int32, [1]> var_3170_axes_0 = const()[name = tensor<string, []>("op_3170_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3170_keep_dims_0 = const()[name = tensor<string, []>("op_3170_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_3170_cast_fp16 = reduce_mean(axes = var_3170_axes_0, keep_dims = var_3170_keep_dims_0, x = var_3168_cast_fp16)[name = tensor<string, []>("op_3170_cast_fp16")];
            tensor<fp16, []> var_3171_to_fp16 = const()[name = tensor<string, []>("op_3171_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_3172_cast_fp16 = add(x = var_3170_cast_fp16, y = var_3171_to_fp16)[name = tensor<string, []>("op_3172_cast_fp16")];
            tensor<fp32, []> var_3173_epsilon_0 = const()[name = tensor<string, []>("op_3173_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_3173_cast_fp16 = rsqrt(epsilon = var_3173_epsilon_0, x = var_3172_cast_fp16)[name = tensor<string, []>("op_3173_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_325_cast_fp16 = mul(x = x_323, y = var_3173_cast_fp16)[name = tensor<string, []>("output_325_cast_fp16")];
            tensor<fp16, [2560]> var_3177_to_fp16 = const()[name = tensor<string, []>("op_3177_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940481152)))];
            tensor<fp16, [1, 256, 2560]> output_327_cast_fp16 = mul(x = output_325_cast_fp16, y = var_3177_to_fp16)[name = tensor<string, []>("output_327_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_327 = add(x = x_311, y = output_327_cast_fp16)[name = tensor<string, []>("x_327")];
            tensor<fp16, []> var_22_promoted_82_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_82_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_3183_cast_fp16 = pow(x = x_327, y = var_22_promoted_82_to_fp16)[name = tensor<string, []>("op_3183_cast_fp16")];
            tensor<int32, [1]> var_3185_axes_0 = const()[name = tensor<string, []>("op_3185_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3185_keep_dims_0 = const()[name = tensor<string, []>("op_3185_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_3185_cast_fp16 = reduce_mean(axes = var_3185_axes_0, keep_dims = var_3185_keep_dims_0, x = var_3183_cast_fp16)[name = tensor<string, []>("op_3185_cast_fp16")];
            tensor<fp16, []> var_3186_to_fp16 = const()[name = tensor<string, []>("op_3186_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_3187_cast_fp16 = add(x = var_3185_cast_fp16, y = var_3186_to_fp16)[name = tensor<string, []>("op_3187_cast_fp16")];
            tensor<fp32, []> var_3188_epsilon_0 = const()[name = tensor<string, []>("op_3188_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_3188_cast_fp16 = rsqrt(epsilon = var_3188_epsilon_0, x = var_3187_cast_fp16)[name = tensor<string, []>("op_3188_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_329_cast_fp16 = mul(x = x_327, y = var_3188_cast_fp16)[name = tensor<string, []>("output_329_cast_fp16")];
            tensor<fp16, [2560]> var_3192_to_fp16 = const()[name = tensor<string, []>("op_3192_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940486336)))];
            tensor<fp16, [1, 256, 2560]> output_331_cast_fp16 = mul(x = output_329_cast_fp16, y = var_3192_to_fp16)[name = tensor<string, []>("output_331_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_165 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_13_mlp_gate_proj_weight_palettized, x = output_331_cast_fp16)[name = tensor<string, []>("linear_95")];
            tensor<string, []> var_3200_mode_0 = const()[name = tensor<string, []>("op_3200_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_3200 = gelu(mode = var_3200_mode_0, x = input_165)[name = tensor<string, []>("op_3200")];
            tensor<fp16, [1, 256, 10240]> var_3202 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_13_mlp_up_proj_weight_palettized, x = output_331_cast_fp16)[name = tensor<string, []>("linear_96")];
            tensor<fp16, [1, 256, 10240]> input_167 = mul(x = var_3200, y = var_3202)[name = tensor<string, []>("input_167")];
            tensor<fp16, [1, 256, 2560]> x_331 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_13_mlp_down_proj_weight_palettized, x = input_167)[name = tensor<string, []>("linear_97")];
            tensor<fp16, []> var_22_promoted_83_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_83_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_3208_cast_fp16 = pow(x = x_331, y = var_22_promoted_83_to_fp16)[name = tensor<string, []>("op_3208_cast_fp16")];
            tensor<int32, [1]> var_3210_axes_0 = const()[name = tensor<string, []>("op_3210_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3210_keep_dims_0 = const()[name = tensor<string, []>("op_3210_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_3210_cast_fp16 = reduce_mean(axes = var_3210_axes_0, keep_dims = var_3210_keep_dims_0, x = var_3208_cast_fp16)[name = tensor<string, []>("op_3210_cast_fp16")];
            tensor<fp16, []> var_3211_to_fp16 = const()[name = tensor<string, []>("op_3211_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_3212_cast_fp16 = add(x = var_3210_cast_fp16, y = var_3211_to_fp16)[name = tensor<string, []>("op_3212_cast_fp16")];
            tensor<fp32, []> var_3213_epsilon_0 = const()[name = tensor<string, []>("op_3213_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_3213_cast_fp16 = rsqrt(epsilon = var_3213_epsilon_0, x = var_3212_cast_fp16)[name = tensor<string, []>("op_3213_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_333_cast_fp16 = mul(x = x_331, y = var_3213_cast_fp16)[name = tensor<string, []>("output_333_cast_fp16")];
            tensor<fp16, [2560]> var_3217_to_fp16 = const()[name = tensor<string, []>("op_3217_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940491520)))];
            tensor<fp16, [1, 256, 2560]> output_335_cast_fp16 = mul(x = output_333_cast_fp16, y = var_3217_to_fp16)[name = tensor<string, []>("output_335_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_335 = add(x = x_327, y = output_335_cast_fp16)[name = tensor<string, []>("x_335")];
            tensor<fp16, []> var_22_promoted_84_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_84_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_3229_cast_fp16 = pow(x = x_335, y = var_22_promoted_84_to_fp16)[name = tensor<string, []>("op_3229_cast_fp16")];
            tensor<int32, [1]> var_3231_axes_0 = const()[name = tensor<string, []>("op_3231_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3231_keep_dims_0 = const()[name = tensor<string, []>("op_3231_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_3231_cast_fp16 = reduce_mean(axes = var_3231_axes_0, keep_dims = var_3231_keep_dims_0, x = var_3229_cast_fp16)[name = tensor<string, []>("op_3231_cast_fp16")];
            tensor<fp16, []> var_3232_to_fp16 = const()[name = tensor<string, []>("op_3232_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_3233_cast_fp16 = add(x = var_3231_cast_fp16, y = var_3232_to_fp16)[name = tensor<string, []>("op_3233_cast_fp16")];
            tensor<fp32, []> var_3234_epsilon_0 = const()[name = tensor<string, []>("op_3234_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_3234_cast_fp16 = rsqrt(epsilon = var_3234_epsilon_0, x = var_3233_cast_fp16)[name = tensor<string, []>("op_3234_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_337_cast_fp16 = mul(x = x_335, y = var_3234_cast_fp16)[name = tensor<string, []>("output_337_cast_fp16")];
            tensor<fp16, [2560]> var_3238_to_fp16 = const()[name = tensor<string, []>("op_3238_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940496704)))];
            tensor<fp16, [1, 256, 2560]> output_339_cast_fp16 = mul(x = output_337_cast_fp16, y = var_3238_to_fp16)[name = tensor<string, []>("output_339_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_3250 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_14_self_attn_q_proj_weight_palettized, x = output_339_cast_fp16)[name = tensor<string, []>("linear_98")];
            tensor<int32, [4]> var_3251 = const()[name = tensor<string, []>("op_3251"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_3252 = reshape(shape = var_3251, x = var_3250)[name = tensor<string, []>("op_3252")];
            tensor<int32, [4]> x_339_perm_0 = const()[name = tensor<string, []>("x_339_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_3255 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_14_self_attn_k_proj_weight_palettized, x = output_339_cast_fp16)[name = tensor<string, []>("linear_99")];
            tensor<int32, [4]> var_3256 = const()[name = tensor<string, []>("op_3256"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_3257 = reshape(shape = var_3256, x = var_3255)[name = tensor<string, []>("op_3257")];
            tensor<int32, [4]> x_343_perm_0 = const()[name = tensor<string, []>("x_343_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_3260 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_14_self_attn_v_proj_weight_palettized, x = output_339_cast_fp16)[name = tensor<string, []>("linear_100")];
            tensor<int32, [4]> var_3261 = const()[name = tensor<string, []>("op_3261"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_3262 = reshape(shape = var_3261, x = var_3260)[name = tensor<string, []>("op_3262")];
            tensor<int32, [4]> hidden_states_203_perm_0 = const()[name = tensor<string, []>("hidden_states_203_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_85_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_85_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_339 = transpose(perm = x_339_perm_0, x = var_3252)[name = tensor<string, []>("transpose_79")];
            tensor<fp16, [1, 8, 256, 256]> var_3266_cast_fp16 = pow(x = x_339, y = var_22_promoted_85_to_fp16)[name = tensor<string, []>("op_3266_cast_fp16")];
            tensor<int32, [1]> var_3268_axes_0 = const()[name = tensor<string, []>("op_3268_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3268_keep_dims_0 = const()[name = tensor<string, []>("op_3268_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_3268_cast_fp16 = reduce_mean(axes = var_3268_axes_0, keep_dims = var_3268_keep_dims_0, x = var_3266_cast_fp16)[name = tensor<string, []>("op_3268_cast_fp16")];
            tensor<fp16, []> var_3269_to_fp16 = const()[name = tensor<string, []>("op_3269_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_3270_cast_fp16 = add(x = var_3268_cast_fp16, y = var_3269_to_fp16)[name = tensor<string, []>("op_3270_cast_fp16")];
            tensor<fp32, []> var_3271_epsilon_0 = const()[name = tensor<string, []>("op_3271_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_3271_cast_fp16 = rsqrt(epsilon = var_3271_epsilon_0, x = var_3270_cast_fp16)[name = tensor<string, []>("op_3271_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_341_cast_fp16 = mul(x = x_339, y = var_3271_cast_fp16)[name = tensor<string, []>("output_341_cast_fp16")];
            tensor<fp16, [256]> var_3275_to_fp16 = const()[name = tensor<string, []>("op_3275_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940501888)))];
            tensor<fp16, [1, 8, 256, 256]> output_343_cast_fp16 = mul(x = output_341_cast_fp16, y = var_3275_to_fp16)[name = tensor<string, []>("output_343_cast_fp16")];
            tensor<fp16, []> var_22_promoted_86_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_86_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_343 = transpose(perm = x_343_perm_0, x = var_3257)[name = tensor<string, []>("transpose_78")];
            tensor<fp16, [1, 4, 256, 256]> var_3280_cast_fp16 = pow(x = x_343, y = var_22_promoted_86_to_fp16)[name = tensor<string, []>("op_3280_cast_fp16")];
            tensor<int32, [1]> var_3282_axes_0 = const()[name = tensor<string, []>("op_3282_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3282_keep_dims_0 = const()[name = tensor<string, []>("op_3282_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_3282_cast_fp16 = reduce_mean(axes = var_3282_axes_0, keep_dims = var_3282_keep_dims_0, x = var_3280_cast_fp16)[name = tensor<string, []>("op_3282_cast_fp16")];
            tensor<fp16, []> var_3283_to_fp16 = const()[name = tensor<string, []>("op_3283_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_3284_cast_fp16 = add(x = var_3282_cast_fp16, y = var_3283_to_fp16)[name = tensor<string, []>("op_3284_cast_fp16")];
            tensor<fp32, []> var_3285_epsilon_0 = const()[name = tensor<string, []>("op_3285_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_3285_cast_fp16 = rsqrt(epsilon = var_3285_epsilon_0, x = var_3284_cast_fp16)[name = tensor<string, []>("op_3285_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_345_cast_fp16 = mul(x = x_343, y = var_3285_cast_fp16)[name = tensor<string, []>("output_345_cast_fp16")];
            tensor<fp16, [256]> var_3289_to_fp16 = const()[name = tensor<string, []>("op_3289_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940502464)))];
            tensor<fp16, [1, 4, 256, 256]> output_347_cast_fp16 = mul(x = output_345_cast_fp16, y = var_3289_to_fp16)[name = tensor<string, []>("output_347_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_3294 = mul(x = output_343_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_3294")];
            tensor<int32, [4]> x1_57_begin_0 = const()[name = tensor<string, []>("x1_57_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_57_end_0 = const()[name = tensor<string, []>("x1_57_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_57_end_mask_0 = const()[name = tensor<string, []>("x1_57_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_57 = slice_by_index(begin = x1_57_begin_0, end = x1_57_end_0, end_mask = x1_57_end_mask_0, x = output_343_cast_fp16)[name = tensor<string, []>("x1_57")];
            tensor<int32, [4]> x2_57_begin_0 = const()[name = tensor<string, []>("x2_57_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_57_end_0 = const()[name = tensor<string, []>("x2_57_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_57_end_mask_0 = const()[name = tensor<string, []>("x2_57_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_57 = slice_by_index(begin = x2_57_begin_0, end = x2_57_end_0, end_mask = x2_57_end_mask_0, x = output_343_cast_fp16)[name = tensor<string, []>("x2_57")];
            tensor<fp16, []> const_357_promoted = const()[name = tensor<string, []>("const_357_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_3305 = mul(x = x2_57, y = const_357_promoted)[name = tensor<string, []>("op_3305")];
            tensor<bool, []> var_3307_interleave_0 = const()[name = tensor<string, []>("op_3307_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_3307 = concat(axis = var_24, interleave = var_3307_interleave_0, values = (var_3305, x1_57))[name = tensor<string, []>("op_3307")];
            tensor<fp16, [1, 8, 256, 256]> var_3308 = mul(x = var_3307, y = sin_7_palettized)[name = tensor<string, []>("op_3308")];
            tensor<fp16, [1, 8, 256, 256]> query_29 = add(x = var_3294, y = var_3308)[name = tensor<string, []>("query_29")];
            tensor<fp16, [1, 4, 256, 256]> var_3310 = mul(x = output_347_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_3310")];
            tensor<int32, [4]> x1_59_begin_0 = const()[name = tensor<string, []>("x1_59_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_59_end_0 = const()[name = tensor<string, []>("x1_59_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_59_end_mask_0 = const()[name = tensor<string, []>("x1_59_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_59 = slice_by_index(begin = x1_59_begin_0, end = x1_59_end_0, end_mask = x1_59_end_mask_0, x = output_347_cast_fp16)[name = tensor<string, []>("x1_59")];
            tensor<int32, [4]> x2_59_begin_0 = const()[name = tensor<string, []>("x2_59_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_59_end_0 = const()[name = tensor<string, []>("x2_59_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_59_end_mask_0 = const()[name = tensor<string, []>("x2_59_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_59 = slice_by_index(begin = x2_59_begin_0, end = x2_59_end_0, end_mask = x2_59_end_mask_0, x = output_347_cast_fp16)[name = tensor<string, []>("x2_59")];
            tensor<fp16, []> const_360_promoted = const()[name = tensor<string, []>("const_360_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_3321 = mul(x = x2_59, y = const_360_promoted)[name = tensor<string, []>("op_3321")];
            tensor<bool, []> var_3323_interleave_0 = const()[name = tensor<string, []>("op_3323_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_3323 = concat(axis = var_24, interleave = var_3323_interleave_0, values = (var_3321, x1_59))[name = tensor<string, []>("op_3323")];
            tensor<fp16, [1, 4, 256, 256]> var_3324 = mul(x = var_3323, y = sin_7_palettized)[name = tensor<string, []>("op_3324")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_199 = add(x = var_3310, y = var_3324)[name = tensor<string, []>("hidden_states_199")];
            tensor<int32, [1]> var_3333_axes_0 = const()[name = tensor<string, []>("op_3333_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_3333 = expand_dims(axes = var_3333_axes_0, x = hidden_states_199)[name = tensor<string, []>("op_3333")];
            tensor<int32, [5]> hidden_states_201_reps_0 = const()[name = tensor<string, []>("hidden_states_201_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_201 = tile(reps = hidden_states_201_reps_0, x = var_3333)[name = tensor<string, []>("hidden_states_201")];
            tensor<int32, [4]> var_3341 = const()[name = tensor<string, []>("op_3341"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_29 = reshape(shape = var_3341, x = hidden_states_201)[name = tensor<string, []>("key_states_29")];
            tensor<int32, [1]> var_3350_axes_0 = const()[name = tensor<string, []>("op_3350_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_203 = transpose(perm = hidden_states_203_perm_0, x = var_3262)[name = tensor<string, []>("transpose_77")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_3350 = expand_dims(axes = var_3350_axes_0, x = hidden_states_203)[name = tensor<string, []>("op_3350")];
            tensor<int32, [5]> hidden_states_205_reps_0 = const()[name = tensor<string, []>("hidden_states_205_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_205 = tile(reps = hidden_states_205_reps_0, x = var_3350)[name = tensor<string, []>("hidden_states_205")];
            tensor<int32, [4]> var_3358 = const()[name = tensor<string, []>("op_3358"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_29 = reshape(shape = var_3358, x = hidden_states_205)[name = tensor<string, []>("value_states_29")];
            tensor<bool, []> var_3361_transpose_x_1 = const()[name = tensor<string, []>("op_3361_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_3361_transpose_y_1 = const()[name = tensor<string, []>("op_3361_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_3361 = matmul(transpose_x = var_3361_transpose_x_1, transpose_y = var_3361_transpose_y_1, x = query_29, y = key_states_29)[name = tensor<string, []>("op_3361")];
            tensor<fp16, []> var_3362_to_fp16 = const()[name = tensor<string, []>("op_3362_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_57_cast_fp16 = mul(x = var_3361, y = var_3362_to_fp16)[name = tensor<string, []>("attn_weights_57_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_169_cast_fp16 = add(x = attn_weights_57_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_169_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_3370_cast_fp16 = softmax(axis = var_24, x = input_169_cast_fp16)[name = tensor<string, []>("op_3370_cast_fp16")];
            tensor<bool, []> attn_output_57_transpose_x_0 = const()[name = tensor<string, []>("attn_output_57_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_57_transpose_y_0 = const()[name = tensor<string, []>("attn_output_57_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_57 = matmul(transpose_x = attn_output_57_transpose_x_0, transpose_y = attn_output_57_transpose_y_0, x = var_3370_cast_fp16, y = value_states_29)[name = tensor<string, []>("attn_output_57")];
            tensor<int32, [4]> var_3374_perm_0 = const()[name = tensor<string, []>("op_3374_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3376 = const()[name = tensor<string, []>("op_3376"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_3374 = transpose(perm = var_3374_perm_0, x = attn_output_57)[name = tensor<string, []>("transpose_76")];
            tensor<fp16, [1, 256, 2048]> var_3377 = reshape(shape = var_3376, x = var_3374)[name = tensor<string, []>("op_3377")];
            tensor<fp16, [1, 256, 2560]> x_347 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_14_self_attn_o_proj_weight_palettized, x = var_3377)[name = tensor<string, []>("linear_101")];
            tensor<fp16, []> var_22_promoted_87_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_87_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_3383_cast_fp16 = pow(x = x_347, y = var_22_promoted_87_to_fp16)[name = tensor<string, []>("op_3383_cast_fp16")];
            tensor<int32, [1]> var_3385_axes_0 = const()[name = tensor<string, []>("op_3385_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3385_keep_dims_0 = const()[name = tensor<string, []>("op_3385_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_3385_cast_fp16 = reduce_mean(axes = var_3385_axes_0, keep_dims = var_3385_keep_dims_0, x = var_3383_cast_fp16)[name = tensor<string, []>("op_3385_cast_fp16")];
            tensor<fp16, []> var_3386_to_fp16 = const()[name = tensor<string, []>("op_3386_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_3387_cast_fp16 = add(x = var_3385_cast_fp16, y = var_3386_to_fp16)[name = tensor<string, []>("op_3387_cast_fp16")];
            tensor<fp32, []> var_3388_epsilon_0 = const()[name = tensor<string, []>("op_3388_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_3388_cast_fp16 = rsqrt(epsilon = var_3388_epsilon_0, x = var_3387_cast_fp16)[name = tensor<string, []>("op_3388_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_349_cast_fp16 = mul(x = x_347, y = var_3388_cast_fp16)[name = tensor<string, []>("output_349_cast_fp16")];
            tensor<fp16, [2560]> var_3392_to_fp16 = const()[name = tensor<string, []>("op_3392_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940503040)))];
            tensor<fp16, [1, 256, 2560]> output_351_cast_fp16 = mul(x = output_349_cast_fp16, y = var_3392_to_fp16)[name = tensor<string, []>("output_351_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_351 = add(x = x_335, y = output_351_cast_fp16)[name = tensor<string, []>("x_351")];
            tensor<fp16, []> var_22_promoted_88_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_88_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_3398_cast_fp16 = pow(x = x_351, y = var_22_promoted_88_to_fp16)[name = tensor<string, []>("op_3398_cast_fp16")];
            tensor<int32, [1]> var_3400_axes_0 = const()[name = tensor<string, []>("op_3400_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3400_keep_dims_0 = const()[name = tensor<string, []>("op_3400_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_3400_cast_fp16 = reduce_mean(axes = var_3400_axes_0, keep_dims = var_3400_keep_dims_0, x = var_3398_cast_fp16)[name = tensor<string, []>("op_3400_cast_fp16")];
            tensor<fp16, []> var_3401_to_fp16 = const()[name = tensor<string, []>("op_3401_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_3402_cast_fp16 = add(x = var_3400_cast_fp16, y = var_3401_to_fp16)[name = tensor<string, []>("op_3402_cast_fp16")];
            tensor<fp32, []> var_3403_epsilon_0 = const()[name = tensor<string, []>("op_3403_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_3403_cast_fp16 = rsqrt(epsilon = var_3403_epsilon_0, x = var_3402_cast_fp16)[name = tensor<string, []>("op_3403_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_353_cast_fp16 = mul(x = x_351, y = var_3403_cast_fp16)[name = tensor<string, []>("output_353_cast_fp16")];
            tensor<fp16, [2560]> var_3407_to_fp16 = const()[name = tensor<string, []>("op_3407_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940508224)))];
            tensor<fp16, [1, 256, 2560]> output_355_cast_fp16 = mul(x = output_353_cast_fp16, y = var_3407_to_fp16)[name = tensor<string, []>("output_355_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_177 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_14_mlp_gate_proj_weight_palettized, x = output_355_cast_fp16)[name = tensor<string, []>("linear_102")];
            tensor<string, []> var_3415_mode_0 = const()[name = tensor<string, []>("op_3415_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_3415 = gelu(mode = var_3415_mode_0, x = input_177)[name = tensor<string, []>("op_3415")];
            tensor<fp16, [1, 256, 10240]> var_3417 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_14_mlp_up_proj_weight_palettized, x = output_355_cast_fp16)[name = tensor<string, []>("linear_103")];
            tensor<fp16, [1, 256, 10240]> input_179 = mul(x = var_3415, y = var_3417)[name = tensor<string, []>("input_179")];
            tensor<fp16, [1, 256, 2560]> x_355 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_14_mlp_down_proj_weight_palettized, x = input_179)[name = tensor<string, []>("linear_104")];
            tensor<fp16, []> var_22_promoted_89_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_89_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_3423_cast_fp16 = pow(x = x_355, y = var_22_promoted_89_to_fp16)[name = tensor<string, []>("op_3423_cast_fp16")];
            tensor<int32, [1]> var_3425_axes_0 = const()[name = tensor<string, []>("op_3425_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3425_keep_dims_0 = const()[name = tensor<string, []>("op_3425_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_3425_cast_fp16 = reduce_mean(axes = var_3425_axes_0, keep_dims = var_3425_keep_dims_0, x = var_3423_cast_fp16)[name = tensor<string, []>("op_3425_cast_fp16")];
            tensor<fp16, []> var_3426_to_fp16 = const()[name = tensor<string, []>("op_3426_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_3427_cast_fp16 = add(x = var_3425_cast_fp16, y = var_3426_to_fp16)[name = tensor<string, []>("op_3427_cast_fp16")];
            tensor<fp32, []> var_3428_epsilon_0 = const()[name = tensor<string, []>("op_3428_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_3428_cast_fp16 = rsqrt(epsilon = var_3428_epsilon_0, x = var_3427_cast_fp16)[name = tensor<string, []>("op_3428_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_357_cast_fp16 = mul(x = x_355, y = var_3428_cast_fp16)[name = tensor<string, []>("output_357_cast_fp16")];
            tensor<fp16, [2560]> var_3432_to_fp16 = const()[name = tensor<string, []>("op_3432_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940513408)))];
            tensor<fp16, [1, 256, 2560]> output_359_cast_fp16 = mul(x = output_357_cast_fp16, y = var_3432_to_fp16)[name = tensor<string, []>("output_359_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_359 = add(x = x_351, y = output_359_cast_fp16)[name = tensor<string, []>("x_359")];
            tensor<fp16, []> var_22_promoted_90_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_90_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_3444_cast_fp16 = pow(x = x_359, y = var_22_promoted_90_to_fp16)[name = tensor<string, []>("op_3444_cast_fp16")];
            tensor<int32, [1]> var_3446_axes_0 = const()[name = tensor<string, []>("op_3446_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3446_keep_dims_0 = const()[name = tensor<string, []>("op_3446_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_3446_cast_fp16 = reduce_mean(axes = var_3446_axes_0, keep_dims = var_3446_keep_dims_0, x = var_3444_cast_fp16)[name = tensor<string, []>("op_3446_cast_fp16")];
            tensor<fp16, []> var_3447_to_fp16 = const()[name = tensor<string, []>("op_3447_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_3448_cast_fp16 = add(x = var_3446_cast_fp16, y = var_3447_to_fp16)[name = tensor<string, []>("op_3448_cast_fp16")];
            tensor<fp32, []> var_3449_epsilon_0 = const()[name = tensor<string, []>("op_3449_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_3449_cast_fp16 = rsqrt(epsilon = var_3449_epsilon_0, x = var_3448_cast_fp16)[name = tensor<string, []>("op_3449_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_361_cast_fp16 = mul(x = x_359, y = var_3449_cast_fp16)[name = tensor<string, []>("output_361_cast_fp16")];
            tensor<fp16, [2560]> var_3453_to_fp16 = const()[name = tensor<string, []>("op_3453_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940518592)))];
            tensor<fp16, [1, 256, 2560]> output_363_cast_fp16 = mul(x = output_361_cast_fp16, y = var_3453_to_fp16)[name = tensor<string, []>("output_363_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_3465 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_15_self_attn_q_proj_weight_palettized, x = output_363_cast_fp16)[name = tensor<string, []>("linear_105")];
            tensor<int32, [4]> var_3466 = const()[name = tensor<string, []>("op_3466"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_3467 = reshape(shape = var_3466, x = var_3465)[name = tensor<string, []>("op_3467")];
            tensor<int32, [4]> x_363_perm_0 = const()[name = tensor<string, []>("x_363_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_3470 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_15_self_attn_k_proj_weight_palettized, x = output_363_cast_fp16)[name = tensor<string, []>("linear_106")];
            tensor<int32, [4]> var_3471 = const()[name = tensor<string, []>("op_3471"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_3472 = reshape(shape = var_3471, x = var_3470)[name = tensor<string, []>("op_3472")];
            tensor<int32, [4]> x_367_perm_0 = const()[name = tensor<string, []>("x_367_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_3475 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_15_self_attn_v_proj_weight_palettized, x = output_363_cast_fp16)[name = tensor<string, []>("linear_107")];
            tensor<int32, [4]> var_3476 = const()[name = tensor<string, []>("op_3476"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_3477 = reshape(shape = var_3476, x = var_3475)[name = tensor<string, []>("op_3477")];
            tensor<int32, [4]> hidden_states_217_perm_0 = const()[name = tensor<string, []>("hidden_states_217_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_91_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_91_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_363 = transpose(perm = x_363_perm_0, x = var_3467)[name = tensor<string, []>("transpose_75")];
            tensor<fp16, [1, 8, 256, 256]> var_3481_cast_fp16 = pow(x = x_363, y = var_22_promoted_91_to_fp16)[name = tensor<string, []>("op_3481_cast_fp16")];
            tensor<int32, [1]> var_3483_axes_0 = const()[name = tensor<string, []>("op_3483_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3483_keep_dims_0 = const()[name = tensor<string, []>("op_3483_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_3483_cast_fp16 = reduce_mean(axes = var_3483_axes_0, keep_dims = var_3483_keep_dims_0, x = var_3481_cast_fp16)[name = tensor<string, []>("op_3483_cast_fp16")];
            tensor<fp16, []> var_3484_to_fp16 = const()[name = tensor<string, []>("op_3484_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_3485_cast_fp16 = add(x = var_3483_cast_fp16, y = var_3484_to_fp16)[name = tensor<string, []>("op_3485_cast_fp16")];
            tensor<fp32, []> var_3486_epsilon_0 = const()[name = tensor<string, []>("op_3486_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_3486_cast_fp16 = rsqrt(epsilon = var_3486_epsilon_0, x = var_3485_cast_fp16)[name = tensor<string, []>("op_3486_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_365_cast_fp16 = mul(x = x_363, y = var_3486_cast_fp16)[name = tensor<string, []>("output_365_cast_fp16")];
            tensor<fp16, [256]> var_3490_to_fp16 = const()[name = tensor<string, []>("op_3490_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940523776)))];
            tensor<fp16, [1, 8, 256, 256]> output_367_cast_fp16 = mul(x = output_365_cast_fp16, y = var_3490_to_fp16)[name = tensor<string, []>("output_367_cast_fp16")];
            tensor<fp16, []> var_22_promoted_92_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_92_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_367 = transpose(perm = x_367_perm_0, x = var_3472)[name = tensor<string, []>("transpose_74")];
            tensor<fp16, [1, 4, 256, 256]> var_3495_cast_fp16 = pow(x = x_367, y = var_22_promoted_92_to_fp16)[name = tensor<string, []>("op_3495_cast_fp16")];
            tensor<int32, [1]> var_3497_axes_0 = const()[name = tensor<string, []>("op_3497_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3497_keep_dims_0 = const()[name = tensor<string, []>("op_3497_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_3497_cast_fp16 = reduce_mean(axes = var_3497_axes_0, keep_dims = var_3497_keep_dims_0, x = var_3495_cast_fp16)[name = tensor<string, []>("op_3497_cast_fp16")];
            tensor<fp16, []> var_3498_to_fp16 = const()[name = tensor<string, []>("op_3498_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_3499_cast_fp16 = add(x = var_3497_cast_fp16, y = var_3498_to_fp16)[name = tensor<string, []>("op_3499_cast_fp16")];
            tensor<fp32, []> var_3500_epsilon_0 = const()[name = tensor<string, []>("op_3500_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_3500_cast_fp16 = rsqrt(epsilon = var_3500_epsilon_0, x = var_3499_cast_fp16)[name = tensor<string, []>("op_3500_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_369_cast_fp16 = mul(x = x_367, y = var_3500_cast_fp16)[name = tensor<string, []>("output_369_cast_fp16")];
            tensor<fp16, [256]> var_3504_to_fp16 = const()[name = tensor<string, []>("op_3504_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940524352)))];
            tensor<fp16, [1, 4, 256, 256]> output_371_cast_fp16 = mul(x = output_369_cast_fp16, y = var_3504_to_fp16)[name = tensor<string, []>("output_371_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_3509 = mul(x = output_367_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_3509")];
            tensor<int32, [4]> x1_61_begin_0 = const()[name = tensor<string, []>("x1_61_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_61_end_0 = const()[name = tensor<string, []>("x1_61_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_61_end_mask_0 = const()[name = tensor<string, []>("x1_61_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_61 = slice_by_index(begin = x1_61_begin_0, end = x1_61_end_0, end_mask = x1_61_end_mask_0, x = output_367_cast_fp16)[name = tensor<string, []>("x1_61")];
            tensor<int32, [4]> x2_61_begin_0 = const()[name = tensor<string, []>("x2_61_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_61_end_0 = const()[name = tensor<string, []>("x2_61_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_61_end_mask_0 = const()[name = tensor<string, []>("x2_61_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_61 = slice_by_index(begin = x2_61_begin_0, end = x2_61_end_0, end_mask = x2_61_end_mask_0, x = output_367_cast_fp16)[name = tensor<string, []>("x2_61")];
            tensor<fp16, []> const_380_promoted = const()[name = tensor<string, []>("const_380_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_3520 = mul(x = x2_61, y = const_380_promoted)[name = tensor<string, []>("op_3520")];
            tensor<bool, []> var_3522_interleave_0 = const()[name = tensor<string, []>("op_3522_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_3522 = concat(axis = var_24, interleave = var_3522_interleave_0, values = (var_3520, x1_61))[name = tensor<string, []>("op_3522")];
            tensor<fp16, [1, 8, 256, 256]> var_3523 = mul(x = var_3522, y = sin_7_palettized)[name = tensor<string, []>("op_3523")];
            tensor<fp16, [1, 8, 256, 256]> query_31 = add(x = var_3509, y = var_3523)[name = tensor<string, []>("query_31")];
            tensor<fp16, [1, 4, 256, 256]> var_3525 = mul(x = output_371_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_3525")];
            tensor<int32, [4]> x1_63_begin_0 = const()[name = tensor<string, []>("x1_63_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_63_end_0 = const()[name = tensor<string, []>("x1_63_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_63_end_mask_0 = const()[name = tensor<string, []>("x1_63_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_63 = slice_by_index(begin = x1_63_begin_0, end = x1_63_end_0, end_mask = x1_63_end_mask_0, x = output_371_cast_fp16)[name = tensor<string, []>("x1_63")];
            tensor<int32, [4]> x2_63_begin_0 = const()[name = tensor<string, []>("x2_63_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_63_end_0 = const()[name = tensor<string, []>("x2_63_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_63_end_mask_0 = const()[name = tensor<string, []>("x2_63_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_63 = slice_by_index(begin = x2_63_begin_0, end = x2_63_end_0, end_mask = x2_63_end_mask_0, x = output_371_cast_fp16)[name = tensor<string, []>("x2_63")];
            tensor<fp16, []> const_383_promoted = const()[name = tensor<string, []>("const_383_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_3536 = mul(x = x2_63, y = const_383_promoted)[name = tensor<string, []>("op_3536")];
            tensor<bool, []> var_3538_interleave_0 = const()[name = tensor<string, []>("op_3538_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_3538 = concat(axis = var_24, interleave = var_3538_interleave_0, values = (var_3536, x1_63))[name = tensor<string, []>("op_3538")];
            tensor<fp16, [1, 4, 256, 256]> var_3539 = mul(x = var_3538, y = sin_7_palettized)[name = tensor<string, []>("op_3539")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_213 = add(x = var_3525, y = var_3539)[name = tensor<string, []>("hidden_states_213")];
            tensor<int32, [1]> var_3548_axes_0 = const()[name = tensor<string, []>("op_3548_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_3548 = expand_dims(axes = var_3548_axes_0, x = hidden_states_213)[name = tensor<string, []>("op_3548")];
            tensor<int32, [5]> hidden_states_215_reps_0 = const()[name = tensor<string, []>("hidden_states_215_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_215 = tile(reps = hidden_states_215_reps_0, x = var_3548)[name = tensor<string, []>("hidden_states_215")];
            tensor<int32, [4]> var_3556 = const()[name = tensor<string, []>("op_3556"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_31 = reshape(shape = var_3556, x = hidden_states_215)[name = tensor<string, []>("key_states_31")];
            tensor<int32, [1]> var_3565_axes_0 = const()[name = tensor<string, []>("op_3565_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_217 = transpose(perm = hidden_states_217_perm_0, x = var_3477)[name = tensor<string, []>("transpose_73")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_3565 = expand_dims(axes = var_3565_axes_0, x = hidden_states_217)[name = tensor<string, []>("op_3565")];
            tensor<int32, [5]> hidden_states_219_reps_0 = const()[name = tensor<string, []>("hidden_states_219_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_219 = tile(reps = hidden_states_219_reps_0, x = var_3565)[name = tensor<string, []>("hidden_states_219")];
            tensor<int32, [4]> var_3573 = const()[name = tensor<string, []>("op_3573"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_31 = reshape(shape = var_3573, x = hidden_states_219)[name = tensor<string, []>("value_states_31")];
            tensor<bool, []> var_3576_transpose_x_1 = const()[name = tensor<string, []>("op_3576_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_3576_transpose_y_1 = const()[name = tensor<string, []>("op_3576_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_3576 = matmul(transpose_x = var_3576_transpose_x_1, transpose_y = var_3576_transpose_y_1, x = query_31, y = key_states_31)[name = tensor<string, []>("op_3576")];
            tensor<fp16, []> var_3577_to_fp16 = const()[name = tensor<string, []>("op_3577_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_61_cast_fp16 = mul(x = var_3576, y = var_3577_to_fp16)[name = tensor<string, []>("attn_weights_61_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_181_cast_fp16 = add(x = attn_weights_61_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_181_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_3585_cast_fp16 = softmax(axis = var_24, x = input_181_cast_fp16)[name = tensor<string, []>("op_3585_cast_fp16")];
            tensor<bool, []> attn_output_61_transpose_x_0 = const()[name = tensor<string, []>("attn_output_61_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_61_transpose_y_0 = const()[name = tensor<string, []>("attn_output_61_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_61 = matmul(transpose_x = attn_output_61_transpose_x_0, transpose_y = attn_output_61_transpose_y_0, x = var_3585_cast_fp16, y = value_states_31)[name = tensor<string, []>("attn_output_61")];
            tensor<int32, [4]> var_3589_perm_0 = const()[name = tensor<string, []>("op_3589_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3591 = const()[name = tensor<string, []>("op_3591"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_3589 = transpose(perm = var_3589_perm_0, x = attn_output_61)[name = tensor<string, []>("transpose_72")];
            tensor<fp16, [1, 256, 2048]> var_3592 = reshape(shape = var_3591, x = var_3589)[name = tensor<string, []>("op_3592")];
            tensor<fp16, [1, 256, 2560]> x_371 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_15_self_attn_o_proj_weight_palettized, x = var_3592)[name = tensor<string, []>("linear_108")];
            tensor<fp16, []> var_22_promoted_93_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_93_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_3598_cast_fp16 = pow(x = x_371, y = var_22_promoted_93_to_fp16)[name = tensor<string, []>("op_3598_cast_fp16")];
            tensor<int32, [1]> var_3600_axes_0 = const()[name = tensor<string, []>("op_3600_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3600_keep_dims_0 = const()[name = tensor<string, []>("op_3600_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_3600_cast_fp16 = reduce_mean(axes = var_3600_axes_0, keep_dims = var_3600_keep_dims_0, x = var_3598_cast_fp16)[name = tensor<string, []>("op_3600_cast_fp16")];
            tensor<fp16, []> var_3601_to_fp16 = const()[name = tensor<string, []>("op_3601_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_3602_cast_fp16 = add(x = var_3600_cast_fp16, y = var_3601_to_fp16)[name = tensor<string, []>("op_3602_cast_fp16")];
            tensor<fp32, []> var_3603_epsilon_0 = const()[name = tensor<string, []>("op_3603_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_3603_cast_fp16 = rsqrt(epsilon = var_3603_epsilon_0, x = var_3602_cast_fp16)[name = tensor<string, []>("op_3603_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_373_cast_fp16 = mul(x = x_371, y = var_3603_cast_fp16)[name = tensor<string, []>("output_373_cast_fp16")];
            tensor<fp16, [2560]> var_3607_to_fp16 = const()[name = tensor<string, []>("op_3607_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940524928)))];
            tensor<fp16, [1, 256, 2560]> output_375_cast_fp16 = mul(x = output_373_cast_fp16, y = var_3607_to_fp16)[name = tensor<string, []>("output_375_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_375 = add(x = x_359, y = output_375_cast_fp16)[name = tensor<string, []>("x_375")];
            tensor<fp16, []> var_22_promoted_94_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_94_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_3613_cast_fp16 = pow(x = x_375, y = var_22_promoted_94_to_fp16)[name = tensor<string, []>("op_3613_cast_fp16")];
            tensor<int32, [1]> var_3615_axes_0 = const()[name = tensor<string, []>("op_3615_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3615_keep_dims_0 = const()[name = tensor<string, []>("op_3615_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_3615_cast_fp16 = reduce_mean(axes = var_3615_axes_0, keep_dims = var_3615_keep_dims_0, x = var_3613_cast_fp16)[name = tensor<string, []>("op_3615_cast_fp16")];
            tensor<fp16, []> var_3616_to_fp16 = const()[name = tensor<string, []>("op_3616_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_3617_cast_fp16 = add(x = var_3615_cast_fp16, y = var_3616_to_fp16)[name = tensor<string, []>("op_3617_cast_fp16")];
            tensor<fp32, []> var_3618_epsilon_0 = const()[name = tensor<string, []>("op_3618_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_3618_cast_fp16 = rsqrt(epsilon = var_3618_epsilon_0, x = var_3617_cast_fp16)[name = tensor<string, []>("op_3618_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_377_cast_fp16 = mul(x = x_375, y = var_3618_cast_fp16)[name = tensor<string, []>("output_377_cast_fp16")];
            tensor<fp16, [2560]> var_3622_to_fp16 = const()[name = tensor<string, []>("op_3622_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940530112)))];
            tensor<fp16, [1, 256, 2560]> output_379_cast_fp16 = mul(x = output_377_cast_fp16, y = var_3622_to_fp16)[name = tensor<string, []>("output_379_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_189 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_15_mlp_gate_proj_weight_palettized, x = output_379_cast_fp16)[name = tensor<string, []>("linear_109")];
            tensor<string, []> var_3630_mode_0 = const()[name = tensor<string, []>("op_3630_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_3630 = gelu(mode = var_3630_mode_0, x = input_189)[name = tensor<string, []>("op_3630")];
            tensor<fp16, [1, 256, 10240]> var_3632 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_15_mlp_up_proj_weight_palettized, x = output_379_cast_fp16)[name = tensor<string, []>("linear_110")];
            tensor<fp16, [1, 256, 10240]> input_191 = mul(x = var_3630, y = var_3632)[name = tensor<string, []>("input_191")];
            tensor<fp16, [1, 256, 2560]> x_379 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_15_mlp_down_proj_weight_palettized, x = input_191)[name = tensor<string, []>("linear_111")];
            tensor<fp16, []> var_22_promoted_95_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_95_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_3638_cast_fp16 = pow(x = x_379, y = var_22_promoted_95_to_fp16)[name = tensor<string, []>("op_3638_cast_fp16")];
            tensor<int32, [1]> var_3640_axes_0 = const()[name = tensor<string, []>("op_3640_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3640_keep_dims_0 = const()[name = tensor<string, []>("op_3640_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_3640_cast_fp16 = reduce_mean(axes = var_3640_axes_0, keep_dims = var_3640_keep_dims_0, x = var_3638_cast_fp16)[name = tensor<string, []>("op_3640_cast_fp16")];
            tensor<fp16, []> var_3641_to_fp16 = const()[name = tensor<string, []>("op_3641_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_3642_cast_fp16 = add(x = var_3640_cast_fp16, y = var_3641_to_fp16)[name = tensor<string, []>("op_3642_cast_fp16")];
            tensor<fp32, []> var_3643_epsilon_0 = const()[name = tensor<string, []>("op_3643_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_3643_cast_fp16 = rsqrt(epsilon = var_3643_epsilon_0, x = var_3642_cast_fp16)[name = tensor<string, []>("op_3643_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_381_cast_fp16 = mul(x = x_379, y = var_3643_cast_fp16)[name = tensor<string, []>("output_381_cast_fp16")];
            tensor<fp16, [2560]> var_3647_to_fp16 = const()[name = tensor<string, []>("op_3647_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940535296)))];
            tensor<fp16, [1, 256, 2560]> output_383_cast_fp16 = mul(x = output_381_cast_fp16, y = var_3647_to_fp16)[name = tensor<string, []>("output_383_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_383 = add(x = x_375, y = output_383_cast_fp16)[name = tensor<string, []>("x_383")];
            tensor<fp16, []> var_22_promoted_96_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_96_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_3659_cast_fp16 = pow(x = x_383, y = var_22_promoted_96_to_fp16)[name = tensor<string, []>("op_3659_cast_fp16")];
            tensor<int32, [1]> var_3661_axes_0 = const()[name = tensor<string, []>("op_3661_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3661_keep_dims_0 = const()[name = tensor<string, []>("op_3661_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_3661_cast_fp16 = reduce_mean(axes = var_3661_axes_0, keep_dims = var_3661_keep_dims_0, x = var_3659_cast_fp16)[name = tensor<string, []>("op_3661_cast_fp16")];
            tensor<fp16, []> var_3662_to_fp16 = const()[name = tensor<string, []>("op_3662_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_3663_cast_fp16 = add(x = var_3661_cast_fp16, y = var_3662_to_fp16)[name = tensor<string, []>("op_3663_cast_fp16")];
            tensor<fp32, []> var_3664_epsilon_0 = const()[name = tensor<string, []>("op_3664_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_3664_cast_fp16 = rsqrt(epsilon = var_3664_epsilon_0, x = var_3663_cast_fp16)[name = tensor<string, []>("op_3664_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_385_cast_fp16 = mul(x = x_383, y = var_3664_cast_fp16)[name = tensor<string, []>("output_385_cast_fp16")];
            tensor<fp16, [2560]> var_3668_to_fp16 = const()[name = tensor<string, []>("op_3668_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940540480)))];
            tensor<fp16, [1, 256, 2560]> output_387_cast_fp16 = mul(x = output_385_cast_fp16, y = var_3668_to_fp16)[name = tensor<string, []>("output_387_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_3680 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_16_self_attn_q_proj_weight_palettized, x = output_387_cast_fp16)[name = tensor<string, []>("linear_112")];
            tensor<int32, [4]> var_3681 = const()[name = tensor<string, []>("op_3681"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_3682 = reshape(shape = var_3681, x = var_3680)[name = tensor<string, []>("op_3682")];
            tensor<int32, [4]> x_387_perm_0 = const()[name = tensor<string, []>("x_387_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_3685 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_16_self_attn_k_proj_weight_palettized, x = output_387_cast_fp16)[name = tensor<string, []>("linear_113")];
            tensor<int32, [4]> var_3686 = const()[name = tensor<string, []>("op_3686"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_3687 = reshape(shape = var_3686, x = var_3685)[name = tensor<string, []>("op_3687")];
            tensor<int32, [4]> x_391_perm_0 = const()[name = tensor<string, []>("x_391_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_3690 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_16_self_attn_v_proj_weight_palettized, x = output_387_cast_fp16)[name = tensor<string, []>("linear_114")];
            tensor<int32, [4]> var_3691 = const()[name = tensor<string, []>("op_3691"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_3692 = reshape(shape = var_3691, x = var_3690)[name = tensor<string, []>("op_3692")];
            tensor<int32, [4]> hidden_states_231_perm_0 = const()[name = tensor<string, []>("hidden_states_231_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_97_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_97_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_387 = transpose(perm = x_387_perm_0, x = var_3682)[name = tensor<string, []>("transpose_71")];
            tensor<fp16, [1, 8, 256, 256]> var_3696_cast_fp16 = pow(x = x_387, y = var_22_promoted_97_to_fp16)[name = tensor<string, []>("op_3696_cast_fp16")];
            tensor<int32, [1]> var_3698_axes_0 = const()[name = tensor<string, []>("op_3698_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3698_keep_dims_0 = const()[name = tensor<string, []>("op_3698_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_3698_cast_fp16 = reduce_mean(axes = var_3698_axes_0, keep_dims = var_3698_keep_dims_0, x = var_3696_cast_fp16)[name = tensor<string, []>("op_3698_cast_fp16")];
            tensor<fp16, []> var_3699_to_fp16 = const()[name = tensor<string, []>("op_3699_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_3700_cast_fp16 = add(x = var_3698_cast_fp16, y = var_3699_to_fp16)[name = tensor<string, []>("op_3700_cast_fp16")];
            tensor<fp32, []> var_3701_epsilon_0 = const()[name = tensor<string, []>("op_3701_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_3701_cast_fp16 = rsqrt(epsilon = var_3701_epsilon_0, x = var_3700_cast_fp16)[name = tensor<string, []>("op_3701_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_389_cast_fp16 = mul(x = x_387, y = var_3701_cast_fp16)[name = tensor<string, []>("output_389_cast_fp16")];
            tensor<fp16, [256]> var_3705_to_fp16 = const()[name = tensor<string, []>("op_3705_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940545664)))];
            tensor<fp16, [1, 8, 256, 256]> output_391_cast_fp16 = mul(x = output_389_cast_fp16, y = var_3705_to_fp16)[name = tensor<string, []>("output_391_cast_fp16")];
            tensor<fp16, []> var_22_promoted_98_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_98_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_391 = transpose(perm = x_391_perm_0, x = var_3687)[name = tensor<string, []>("transpose_70")];
            tensor<fp16, [1, 4, 256, 256]> var_3710_cast_fp16 = pow(x = x_391, y = var_22_promoted_98_to_fp16)[name = tensor<string, []>("op_3710_cast_fp16")];
            tensor<int32, [1]> var_3712_axes_0 = const()[name = tensor<string, []>("op_3712_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3712_keep_dims_0 = const()[name = tensor<string, []>("op_3712_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_3712_cast_fp16 = reduce_mean(axes = var_3712_axes_0, keep_dims = var_3712_keep_dims_0, x = var_3710_cast_fp16)[name = tensor<string, []>("op_3712_cast_fp16")];
            tensor<fp16, []> var_3713_to_fp16 = const()[name = tensor<string, []>("op_3713_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_3714_cast_fp16 = add(x = var_3712_cast_fp16, y = var_3713_to_fp16)[name = tensor<string, []>("op_3714_cast_fp16")];
            tensor<fp32, []> var_3715_epsilon_0 = const()[name = tensor<string, []>("op_3715_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_3715_cast_fp16 = rsqrt(epsilon = var_3715_epsilon_0, x = var_3714_cast_fp16)[name = tensor<string, []>("op_3715_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_393_cast_fp16 = mul(x = x_391, y = var_3715_cast_fp16)[name = tensor<string, []>("output_393_cast_fp16")];
            tensor<fp16, [256]> var_3719_to_fp16 = const()[name = tensor<string, []>("op_3719_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940546240)))];
            tensor<fp16, [1, 4, 256, 256]> output_395_cast_fp16 = mul(x = output_393_cast_fp16, y = var_3719_to_fp16)[name = tensor<string, []>("output_395_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_3724 = mul(x = output_391_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_3724")];
            tensor<int32, [4]> x1_65_begin_0 = const()[name = tensor<string, []>("x1_65_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_65_end_0 = const()[name = tensor<string, []>("x1_65_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_65_end_mask_0 = const()[name = tensor<string, []>("x1_65_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_65 = slice_by_index(begin = x1_65_begin_0, end = x1_65_end_0, end_mask = x1_65_end_mask_0, x = output_391_cast_fp16)[name = tensor<string, []>("x1_65")];
            tensor<int32, [4]> x2_65_begin_0 = const()[name = tensor<string, []>("x2_65_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_65_end_0 = const()[name = tensor<string, []>("x2_65_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_65_end_mask_0 = const()[name = tensor<string, []>("x2_65_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_65 = slice_by_index(begin = x2_65_begin_0, end = x2_65_end_0, end_mask = x2_65_end_mask_0, x = output_391_cast_fp16)[name = tensor<string, []>("x2_65")];
            tensor<fp16, []> const_403_promoted = const()[name = tensor<string, []>("const_403_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_3735 = mul(x = x2_65, y = const_403_promoted)[name = tensor<string, []>("op_3735")];
            tensor<bool, []> var_3737_interleave_0 = const()[name = tensor<string, []>("op_3737_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_3737 = concat(axis = var_24, interleave = var_3737_interleave_0, values = (var_3735, x1_65))[name = tensor<string, []>("op_3737")];
            tensor<fp16, [1, 8, 256, 256]> var_3738 = mul(x = var_3737, y = sin_7_palettized)[name = tensor<string, []>("op_3738")];
            tensor<fp16, [1, 8, 256, 256]> query_33 = add(x = var_3724, y = var_3738)[name = tensor<string, []>("query_33")];
            tensor<fp16, [1, 4, 256, 256]> var_3740 = mul(x = output_395_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_3740")];
            tensor<int32, [4]> x1_67_begin_0 = const()[name = tensor<string, []>("x1_67_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_67_end_0 = const()[name = tensor<string, []>("x1_67_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_67_end_mask_0 = const()[name = tensor<string, []>("x1_67_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_67 = slice_by_index(begin = x1_67_begin_0, end = x1_67_end_0, end_mask = x1_67_end_mask_0, x = output_395_cast_fp16)[name = tensor<string, []>("x1_67")];
            tensor<int32, [4]> x2_67_begin_0 = const()[name = tensor<string, []>("x2_67_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_67_end_0 = const()[name = tensor<string, []>("x2_67_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_67_end_mask_0 = const()[name = tensor<string, []>("x2_67_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_67 = slice_by_index(begin = x2_67_begin_0, end = x2_67_end_0, end_mask = x2_67_end_mask_0, x = output_395_cast_fp16)[name = tensor<string, []>("x2_67")];
            tensor<fp16, []> const_406_promoted = const()[name = tensor<string, []>("const_406_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_3751 = mul(x = x2_67, y = const_406_promoted)[name = tensor<string, []>("op_3751")];
            tensor<bool, []> var_3753_interleave_0 = const()[name = tensor<string, []>("op_3753_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_3753 = concat(axis = var_24, interleave = var_3753_interleave_0, values = (var_3751, x1_67))[name = tensor<string, []>("op_3753")];
            tensor<fp16, [1, 4, 256, 256]> var_3754 = mul(x = var_3753, y = sin_7_palettized)[name = tensor<string, []>("op_3754")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_227 = add(x = var_3740, y = var_3754)[name = tensor<string, []>("hidden_states_227")];
            tensor<int32, [1]> var_3763_axes_0 = const()[name = tensor<string, []>("op_3763_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_3763 = expand_dims(axes = var_3763_axes_0, x = hidden_states_227)[name = tensor<string, []>("op_3763")];
            tensor<int32, [5]> hidden_states_229_reps_0 = const()[name = tensor<string, []>("hidden_states_229_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_229 = tile(reps = hidden_states_229_reps_0, x = var_3763)[name = tensor<string, []>("hidden_states_229")];
            tensor<int32, [4]> var_3771 = const()[name = tensor<string, []>("op_3771"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_33 = reshape(shape = var_3771, x = hidden_states_229)[name = tensor<string, []>("key_states_33")];
            tensor<int32, [1]> var_3780_axes_0 = const()[name = tensor<string, []>("op_3780_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_231 = transpose(perm = hidden_states_231_perm_0, x = var_3692)[name = tensor<string, []>("transpose_69")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_3780 = expand_dims(axes = var_3780_axes_0, x = hidden_states_231)[name = tensor<string, []>("op_3780")];
            tensor<int32, [5]> hidden_states_233_reps_0 = const()[name = tensor<string, []>("hidden_states_233_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_233 = tile(reps = hidden_states_233_reps_0, x = var_3780)[name = tensor<string, []>("hidden_states_233")];
            tensor<int32, [4]> var_3788 = const()[name = tensor<string, []>("op_3788"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_33 = reshape(shape = var_3788, x = hidden_states_233)[name = tensor<string, []>("value_states_33")];
            tensor<bool, []> var_3791_transpose_x_1 = const()[name = tensor<string, []>("op_3791_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_3791_transpose_y_1 = const()[name = tensor<string, []>("op_3791_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_3791 = matmul(transpose_x = var_3791_transpose_x_1, transpose_y = var_3791_transpose_y_1, x = query_33, y = key_states_33)[name = tensor<string, []>("op_3791")];
            tensor<fp16, []> var_3792_to_fp16 = const()[name = tensor<string, []>("op_3792_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_65_cast_fp16 = mul(x = var_3791, y = var_3792_to_fp16)[name = tensor<string, []>("attn_weights_65_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_193_cast_fp16 = add(x = attn_weights_65_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_193_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_3800_cast_fp16 = softmax(axis = var_24, x = input_193_cast_fp16)[name = tensor<string, []>("op_3800_cast_fp16")];
            tensor<bool, []> attn_output_65_transpose_x_0 = const()[name = tensor<string, []>("attn_output_65_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_65_transpose_y_0 = const()[name = tensor<string, []>("attn_output_65_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_65 = matmul(transpose_x = attn_output_65_transpose_x_0, transpose_y = attn_output_65_transpose_y_0, x = var_3800_cast_fp16, y = value_states_33)[name = tensor<string, []>("attn_output_65")];
            tensor<int32, [4]> var_3804_perm_0 = const()[name = tensor<string, []>("op_3804_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3806 = const()[name = tensor<string, []>("op_3806"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_3804 = transpose(perm = var_3804_perm_0, x = attn_output_65)[name = tensor<string, []>("transpose_68")];
            tensor<fp16, [1, 256, 2048]> var_3807 = reshape(shape = var_3806, x = var_3804)[name = tensor<string, []>("op_3807")];
            tensor<fp16, [1, 256, 2560]> x_395 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_16_self_attn_o_proj_weight_palettized, x = var_3807)[name = tensor<string, []>("linear_115")];
            tensor<fp16, []> var_22_promoted_99_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_99_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_3813_cast_fp16 = pow(x = x_395, y = var_22_promoted_99_to_fp16)[name = tensor<string, []>("op_3813_cast_fp16")];
            tensor<int32, [1]> var_3815_axes_0 = const()[name = tensor<string, []>("op_3815_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3815_keep_dims_0 = const()[name = tensor<string, []>("op_3815_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_3815_cast_fp16 = reduce_mean(axes = var_3815_axes_0, keep_dims = var_3815_keep_dims_0, x = var_3813_cast_fp16)[name = tensor<string, []>("op_3815_cast_fp16")];
            tensor<fp16, []> var_3816_to_fp16 = const()[name = tensor<string, []>("op_3816_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_3817_cast_fp16 = add(x = var_3815_cast_fp16, y = var_3816_to_fp16)[name = tensor<string, []>("op_3817_cast_fp16")];
            tensor<fp32, []> var_3818_epsilon_0 = const()[name = tensor<string, []>("op_3818_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_3818_cast_fp16 = rsqrt(epsilon = var_3818_epsilon_0, x = var_3817_cast_fp16)[name = tensor<string, []>("op_3818_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_397_cast_fp16 = mul(x = x_395, y = var_3818_cast_fp16)[name = tensor<string, []>("output_397_cast_fp16")];
            tensor<fp16, [2560]> var_3822_to_fp16 = const()[name = tensor<string, []>("op_3822_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940546816)))];
            tensor<fp16, [1, 256, 2560]> output_399_cast_fp16 = mul(x = output_397_cast_fp16, y = var_3822_to_fp16)[name = tensor<string, []>("output_399_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_399 = add(x = x_383, y = output_399_cast_fp16)[name = tensor<string, []>("x_399")];
            tensor<fp16, []> var_22_promoted_100_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_100_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_3828_cast_fp16 = pow(x = x_399, y = var_22_promoted_100_to_fp16)[name = tensor<string, []>("op_3828_cast_fp16")];
            tensor<int32, [1]> var_3830_axes_0 = const()[name = tensor<string, []>("op_3830_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3830_keep_dims_0 = const()[name = tensor<string, []>("op_3830_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_3830_cast_fp16 = reduce_mean(axes = var_3830_axes_0, keep_dims = var_3830_keep_dims_0, x = var_3828_cast_fp16)[name = tensor<string, []>("op_3830_cast_fp16")];
            tensor<fp16, []> var_3831_to_fp16 = const()[name = tensor<string, []>("op_3831_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_3832_cast_fp16 = add(x = var_3830_cast_fp16, y = var_3831_to_fp16)[name = tensor<string, []>("op_3832_cast_fp16")];
            tensor<fp32, []> var_3833_epsilon_0 = const()[name = tensor<string, []>("op_3833_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_3833_cast_fp16 = rsqrt(epsilon = var_3833_epsilon_0, x = var_3832_cast_fp16)[name = tensor<string, []>("op_3833_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_401_cast_fp16 = mul(x = x_399, y = var_3833_cast_fp16)[name = tensor<string, []>("output_401_cast_fp16")];
            tensor<fp16, [2560]> var_3837_to_fp16 = const()[name = tensor<string, []>("op_3837_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940552000)))];
            tensor<fp16, [1, 256, 2560]> output_403_cast_fp16 = mul(x = output_401_cast_fp16, y = var_3837_to_fp16)[name = tensor<string, []>("output_403_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_201 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_16_mlp_gate_proj_weight_palettized, x = output_403_cast_fp16)[name = tensor<string, []>("linear_116")];
            tensor<string, []> var_3845_mode_0 = const()[name = tensor<string, []>("op_3845_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_3845 = gelu(mode = var_3845_mode_0, x = input_201)[name = tensor<string, []>("op_3845")];
            tensor<fp16, [1, 256, 10240]> var_3847 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_16_mlp_up_proj_weight_palettized, x = output_403_cast_fp16)[name = tensor<string, []>("linear_117")];
            tensor<fp16, [1, 256, 10240]> input_203 = mul(x = var_3845, y = var_3847)[name = tensor<string, []>("input_203")];
            tensor<fp16, [1, 256, 2560]> x_403 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_16_mlp_down_proj_weight_palettized, x = input_203)[name = tensor<string, []>("linear_118")];
            tensor<fp16, []> var_22_promoted_101_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_101_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_3853_cast_fp16 = pow(x = x_403, y = var_22_promoted_101_to_fp16)[name = tensor<string, []>("op_3853_cast_fp16")];
            tensor<int32, [1]> var_3855_axes_0 = const()[name = tensor<string, []>("op_3855_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3855_keep_dims_0 = const()[name = tensor<string, []>("op_3855_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_3855_cast_fp16 = reduce_mean(axes = var_3855_axes_0, keep_dims = var_3855_keep_dims_0, x = var_3853_cast_fp16)[name = tensor<string, []>("op_3855_cast_fp16")];
            tensor<fp16, []> var_3856_to_fp16 = const()[name = tensor<string, []>("op_3856_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_3857_cast_fp16 = add(x = var_3855_cast_fp16, y = var_3856_to_fp16)[name = tensor<string, []>("op_3857_cast_fp16")];
            tensor<fp32, []> var_3858_epsilon_0 = const()[name = tensor<string, []>("op_3858_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_3858_cast_fp16 = rsqrt(epsilon = var_3858_epsilon_0, x = var_3857_cast_fp16)[name = tensor<string, []>("op_3858_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_405_cast_fp16 = mul(x = x_403, y = var_3858_cast_fp16)[name = tensor<string, []>("output_405_cast_fp16")];
            tensor<fp16, [2560]> var_3862_to_fp16 = const()[name = tensor<string, []>("op_3862_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940557184)))];
            tensor<fp16, [1, 256, 2560]> output_407_cast_fp16 = mul(x = output_405_cast_fp16, y = var_3862_to_fp16)[name = tensor<string, []>("output_407_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_407 = add(x = x_399, y = output_407_cast_fp16)[name = tensor<string, []>("x_407")];
            tensor<fp16, []> var_22_promoted_102_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_102_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_3874_cast_fp16 = pow(x = x_407, y = var_22_promoted_102_to_fp16)[name = tensor<string, []>("op_3874_cast_fp16")];
            tensor<int32, [1]> var_3876_axes_0 = const()[name = tensor<string, []>("op_3876_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3876_keep_dims_0 = const()[name = tensor<string, []>("op_3876_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_3876_cast_fp16 = reduce_mean(axes = var_3876_axes_0, keep_dims = var_3876_keep_dims_0, x = var_3874_cast_fp16)[name = tensor<string, []>("op_3876_cast_fp16")];
            tensor<fp16, []> var_3877_to_fp16 = const()[name = tensor<string, []>("op_3877_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_3878_cast_fp16 = add(x = var_3876_cast_fp16, y = var_3877_to_fp16)[name = tensor<string, []>("op_3878_cast_fp16")];
            tensor<fp32, []> var_3879_epsilon_0 = const()[name = tensor<string, []>("op_3879_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_3879_cast_fp16 = rsqrt(epsilon = var_3879_epsilon_0, x = var_3878_cast_fp16)[name = tensor<string, []>("op_3879_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_409_cast_fp16 = mul(x = x_407, y = var_3879_cast_fp16)[name = tensor<string, []>("output_409_cast_fp16")];
            tensor<fp16, [2560]> var_3883_to_fp16 = const()[name = tensor<string, []>("op_3883_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940562368)))];
            tensor<fp16, [1, 256, 2560]> output_411_cast_fp16 = mul(x = output_409_cast_fp16, y = var_3883_to_fp16)[name = tensor<string, []>("output_411_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_3895 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_17_self_attn_q_proj_weight_palettized, x = output_411_cast_fp16)[name = tensor<string, []>("linear_119")];
            tensor<int32, [4]> var_3896 = const()[name = tensor<string, []>("op_3896"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_3897 = reshape(shape = var_3896, x = var_3895)[name = tensor<string, []>("op_3897")];
            tensor<int32, [4]> x_411_perm_0 = const()[name = tensor<string, []>("x_411_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_3900 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_17_self_attn_k_proj_weight_palettized, x = output_411_cast_fp16)[name = tensor<string, []>("linear_120")];
            tensor<int32, [4]> var_3901 = const()[name = tensor<string, []>("op_3901"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_3902 = reshape(shape = var_3901, x = var_3900)[name = tensor<string, []>("op_3902")];
            tensor<int32, [4]> x_415_perm_0 = const()[name = tensor<string, []>("x_415_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_3905 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_17_self_attn_v_proj_weight_palettized, x = output_411_cast_fp16)[name = tensor<string, []>("linear_121")];
            tensor<int32, [4]> var_3906 = const()[name = tensor<string, []>("op_3906"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_3907 = reshape(shape = var_3906, x = var_3905)[name = tensor<string, []>("op_3907")];
            tensor<int32, [4]> hidden_states_245_perm_0 = const()[name = tensor<string, []>("hidden_states_245_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_103_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_103_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_411 = transpose(perm = x_411_perm_0, x = var_3897)[name = tensor<string, []>("transpose_67")];
            tensor<fp16, [1, 8, 256, 256]> var_3911_cast_fp16 = pow(x = x_411, y = var_22_promoted_103_to_fp16)[name = tensor<string, []>("op_3911_cast_fp16")];
            tensor<int32, [1]> var_3913_axes_0 = const()[name = tensor<string, []>("op_3913_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3913_keep_dims_0 = const()[name = tensor<string, []>("op_3913_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_3913_cast_fp16 = reduce_mean(axes = var_3913_axes_0, keep_dims = var_3913_keep_dims_0, x = var_3911_cast_fp16)[name = tensor<string, []>("op_3913_cast_fp16")];
            tensor<fp16, []> var_3914_to_fp16 = const()[name = tensor<string, []>("op_3914_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_3915_cast_fp16 = add(x = var_3913_cast_fp16, y = var_3914_to_fp16)[name = tensor<string, []>("op_3915_cast_fp16")];
            tensor<fp32, []> var_3916_epsilon_0 = const()[name = tensor<string, []>("op_3916_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_3916_cast_fp16 = rsqrt(epsilon = var_3916_epsilon_0, x = var_3915_cast_fp16)[name = tensor<string, []>("op_3916_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_413_cast_fp16 = mul(x = x_411, y = var_3916_cast_fp16)[name = tensor<string, []>("output_413_cast_fp16")];
            tensor<fp16, [256]> var_3920_to_fp16 = const()[name = tensor<string, []>("op_3920_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940567552)))];
            tensor<fp16, [1, 8, 256, 256]> output_415_cast_fp16 = mul(x = output_413_cast_fp16, y = var_3920_to_fp16)[name = tensor<string, []>("output_415_cast_fp16")];
            tensor<fp16, []> var_22_promoted_104_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_104_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_415 = transpose(perm = x_415_perm_0, x = var_3902)[name = tensor<string, []>("transpose_66")];
            tensor<fp16, [1, 4, 256, 256]> var_3925_cast_fp16 = pow(x = x_415, y = var_22_promoted_104_to_fp16)[name = tensor<string, []>("op_3925_cast_fp16")];
            tensor<int32, [1]> var_3927_axes_0 = const()[name = tensor<string, []>("op_3927_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_3927_keep_dims_0 = const()[name = tensor<string, []>("op_3927_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_3927_cast_fp16 = reduce_mean(axes = var_3927_axes_0, keep_dims = var_3927_keep_dims_0, x = var_3925_cast_fp16)[name = tensor<string, []>("op_3927_cast_fp16")];
            tensor<fp16, []> var_3928_to_fp16 = const()[name = tensor<string, []>("op_3928_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_3929_cast_fp16 = add(x = var_3927_cast_fp16, y = var_3928_to_fp16)[name = tensor<string, []>("op_3929_cast_fp16")];
            tensor<fp32, []> var_3930_epsilon_0 = const()[name = tensor<string, []>("op_3930_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_3930_cast_fp16 = rsqrt(epsilon = var_3930_epsilon_0, x = var_3929_cast_fp16)[name = tensor<string, []>("op_3930_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_417_cast_fp16 = mul(x = x_415, y = var_3930_cast_fp16)[name = tensor<string, []>("output_417_cast_fp16")];
            tensor<fp16, [256]> var_3934_to_fp16 = const()[name = tensor<string, []>("op_3934_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940568128)))];
            tensor<fp16, [1, 4, 256, 256]> output_419_cast_fp16 = mul(x = output_417_cast_fp16, y = var_3934_to_fp16)[name = tensor<string, []>("output_419_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_3939 = mul(x = output_415_cast_fp16, y = cos_19_palettized)[name = tensor<string, []>("op_3939")];
            tensor<int32, [4]> x1_69_begin_0 = const()[name = tensor<string, []>("x1_69_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_69_end_0 = const()[name = tensor<string, []>("x1_69_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_69_end_mask_0 = const()[name = tensor<string, []>("x1_69_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_69 = slice_by_index(begin = x1_69_begin_0, end = x1_69_end_0, end_mask = x1_69_end_mask_0, x = output_415_cast_fp16)[name = tensor<string, []>("x1_69")];
            tensor<int32, [4]> x2_69_begin_0 = const()[name = tensor<string, []>("x2_69_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_69_end_0 = const()[name = tensor<string, []>("x2_69_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_69_end_mask_0 = const()[name = tensor<string, []>("x2_69_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_69 = slice_by_index(begin = x2_69_begin_0, end = x2_69_end_0, end_mask = x2_69_end_mask_0, x = output_415_cast_fp16)[name = tensor<string, []>("x2_69")];
            tensor<fp16, []> const_426_promoted = const()[name = tensor<string, []>("const_426_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_3950 = mul(x = x2_69, y = const_426_promoted)[name = tensor<string, []>("op_3950")];
            tensor<bool, []> var_3952_interleave_0 = const()[name = tensor<string, []>("op_3952_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_3952 = concat(axis = var_24, interleave = var_3952_interleave_0, values = (var_3950, x1_69))[name = tensor<string, []>("op_3952")];
            tensor<fp16, [1, 8, 256, 256]> var_3953 = mul(x = var_3952, y = sin_19_palettized)[name = tensor<string, []>("op_3953")];
            tensor<fp16, [1, 8, 256, 256]> query_35 = add(x = var_3939, y = var_3953)[name = tensor<string, []>("query_35")];
            tensor<fp16, [1, 4, 256, 256]> var_3955 = mul(x = output_419_cast_fp16, y = cos_19_palettized)[name = tensor<string, []>("op_3955")];
            tensor<int32, [4]> x1_71_begin_0 = const()[name = tensor<string, []>("x1_71_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_71_end_0 = const()[name = tensor<string, []>("x1_71_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_71_end_mask_0 = const()[name = tensor<string, []>("x1_71_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_71 = slice_by_index(begin = x1_71_begin_0, end = x1_71_end_0, end_mask = x1_71_end_mask_0, x = output_419_cast_fp16)[name = tensor<string, []>("x1_71")];
            tensor<int32, [4]> x2_71_begin_0 = const()[name = tensor<string, []>("x2_71_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_71_end_0 = const()[name = tensor<string, []>("x2_71_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_71_end_mask_0 = const()[name = tensor<string, []>("x2_71_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_71 = slice_by_index(begin = x2_71_begin_0, end = x2_71_end_0, end_mask = x2_71_end_mask_0, x = output_419_cast_fp16)[name = tensor<string, []>("x2_71")];
            tensor<fp16, []> const_429_promoted = const()[name = tensor<string, []>("const_429_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_3966 = mul(x = x2_71, y = const_429_promoted)[name = tensor<string, []>("op_3966")];
            tensor<bool, []> var_3968_interleave_0 = const()[name = tensor<string, []>("op_3968_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_3968 = concat(axis = var_24, interleave = var_3968_interleave_0, values = (var_3966, x1_71))[name = tensor<string, []>("op_3968")];
            tensor<fp16, [1, 4, 256, 256]> var_3969 = mul(x = var_3968, y = sin_19_palettized)[name = tensor<string, []>("op_3969")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_241 = add(x = var_3955, y = var_3969)[name = tensor<string, []>("hidden_states_241")];
            tensor<int32, [1]> var_3978_axes_0 = const()[name = tensor<string, []>("op_3978_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_3978 = expand_dims(axes = var_3978_axes_0, x = hidden_states_241)[name = tensor<string, []>("op_3978")];
            tensor<int32, [5]> hidden_states_243_reps_0 = const()[name = tensor<string, []>("hidden_states_243_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_243 = tile(reps = hidden_states_243_reps_0, x = var_3978)[name = tensor<string, []>("hidden_states_243")];
            tensor<int32, [4]> var_3986 = const()[name = tensor<string, []>("op_3986"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_35 = reshape(shape = var_3986, x = hidden_states_243)[name = tensor<string, []>("key_states_35")];
            tensor<int32, [1]> var_3995_axes_0 = const()[name = tensor<string, []>("op_3995_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_245 = transpose(perm = hidden_states_245_perm_0, x = var_3907)[name = tensor<string, []>("transpose_65")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_3995 = expand_dims(axes = var_3995_axes_0, x = hidden_states_245)[name = tensor<string, []>("op_3995")];
            tensor<int32, [5]> hidden_states_247_reps_0 = const()[name = tensor<string, []>("hidden_states_247_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_247 = tile(reps = hidden_states_247_reps_0, x = var_3995)[name = tensor<string, []>("hidden_states_247")];
            tensor<int32, [4]> var_4003 = const()[name = tensor<string, []>("op_4003"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_35 = reshape(shape = var_4003, x = hidden_states_247)[name = tensor<string, []>("value_states_35")];
            tensor<bool, []> var_4006_transpose_x_1 = const()[name = tensor<string, []>("op_4006_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_4006_transpose_y_1 = const()[name = tensor<string, []>("op_4006_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_4006 = matmul(transpose_x = var_4006_transpose_x_1, transpose_y = var_4006_transpose_y_1, x = query_35, y = key_states_35)[name = tensor<string, []>("op_4006")];
            tensor<fp16, []> var_4007_to_fp16 = const()[name = tensor<string, []>("op_4007_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_69_cast_fp16 = mul(x = var_4006, y = var_4007_to_fp16)[name = tensor<string, []>("attn_weights_69_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_205_cast_fp16 = add(x = attn_weights_69_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_205_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_4015_cast_fp16 = softmax(axis = var_24, x = input_205_cast_fp16)[name = tensor<string, []>("op_4015_cast_fp16")];
            tensor<bool, []> attn_output_69_transpose_x_0 = const()[name = tensor<string, []>("attn_output_69_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_69_transpose_y_0 = const()[name = tensor<string, []>("attn_output_69_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_69 = matmul(transpose_x = attn_output_69_transpose_x_0, transpose_y = attn_output_69_transpose_y_0, x = var_4015_cast_fp16, y = value_states_35)[name = tensor<string, []>("attn_output_69")];
            tensor<int32, [4]> var_4019_perm_0 = const()[name = tensor<string, []>("op_4019_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_4021 = const()[name = tensor<string, []>("op_4021"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_4019 = transpose(perm = var_4019_perm_0, x = attn_output_69)[name = tensor<string, []>("transpose_64")];
            tensor<fp16, [1, 256, 2048]> var_4022 = reshape(shape = var_4021, x = var_4019)[name = tensor<string, []>("op_4022")];
            tensor<fp16, [1, 256, 2560]> x_419 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_17_self_attn_o_proj_weight_palettized, x = var_4022)[name = tensor<string, []>("linear_122")];
            tensor<fp16, []> var_22_promoted_105_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_105_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_4028_cast_fp16 = pow(x = x_419, y = var_22_promoted_105_to_fp16)[name = tensor<string, []>("op_4028_cast_fp16")];
            tensor<int32, [1]> var_4030_axes_0 = const()[name = tensor<string, []>("op_4030_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4030_keep_dims_0 = const()[name = tensor<string, []>("op_4030_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_4030_cast_fp16 = reduce_mean(axes = var_4030_axes_0, keep_dims = var_4030_keep_dims_0, x = var_4028_cast_fp16)[name = tensor<string, []>("op_4030_cast_fp16")];
            tensor<fp16, []> var_4031_to_fp16 = const()[name = tensor<string, []>("op_4031_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_4032_cast_fp16 = add(x = var_4030_cast_fp16, y = var_4031_to_fp16)[name = tensor<string, []>("op_4032_cast_fp16")];
            tensor<fp32, []> var_4033_epsilon_0 = const()[name = tensor<string, []>("op_4033_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_4033_cast_fp16 = rsqrt(epsilon = var_4033_epsilon_0, x = var_4032_cast_fp16)[name = tensor<string, []>("op_4033_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_421_cast_fp16 = mul(x = x_419, y = var_4033_cast_fp16)[name = tensor<string, []>("output_421_cast_fp16")];
            tensor<fp16, [2560]> var_4037_to_fp16 = const()[name = tensor<string, []>("op_4037_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940568704)))];
            tensor<fp16, [1, 256, 2560]> output_423_cast_fp16 = mul(x = output_421_cast_fp16, y = var_4037_to_fp16)[name = tensor<string, []>("output_423_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_423 = add(x = x_407, y = output_423_cast_fp16)[name = tensor<string, []>("x_423")];
            tensor<fp16, []> var_22_promoted_106_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_106_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_4043_cast_fp16 = pow(x = x_423, y = var_22_promoted_106_to_fp16)[name = tensor<string, []>("op_4043_cast_fp16")];
            tensor<int32, [1]> var_4045_axes_0 = const()[name = tensor<string, []>("op_4045_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4045_keep_dims_0 = const()[name = tensor<string, []>("op_4045_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_4045_cast_fp16 = reduce_mean(axes = var_4045_axes_0, keep_dims = var_4045_keep_dims_0, x = var_4043_cast_fp16)[name = tensor<string, []>("op_4045_cast_fp16")];
            tensor<fp16, []> var_4046_to_fp16 = const()[name = tensor<string, []>("op_4046_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_4047_cast_fp16 = add(x = var_4045_cast_fp16, y = var_4046_to_fp16)[name = tensor<string, []>("op_4047_cast_fp16")];
            tensor<fp32, []> var_4048_epsilon_0 = const()[name = tensor<string, []>("op_4048_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_4048_cast_fp16 = rsqrt(epsilon = var_4048_epsilon_0, x = var_4047_cast_fp16)[name = tensor<string, []>("op_4048_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_425_cast_fp16 = mul(x = x_423, y = var_4048_cast_fp16)[name = tensor<string, []>("output_425_cast_fp16")];
            tensor<fp16, [2560]> var_4052_to_fp16 = const()[name = tensor<string, []>("op_4052_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940573888)))];
            tensor<fp16, [1, 256, 2560]> output_427_cast_fp16 = mul(x = output_425_cast_fp16, y = var_4052_to_fp16)[name = tensor<string, []>("output_427_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_213 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_17_mlp_gate_proj_weight_palettized, x = output_427_cast_fp16)[name = tensor<string, []>("linear_123")];
            tensor<string, []> var_4060_mode_0 = const()[name = tensor<string, []>("op_4060_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_4060 = gelu(mode = var_4060_mode_0, x = input_213)[name = tensor<string, []>("op_4060")];
            tensor<fp16, [1, 256, 10240]> var_4062 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_17_mlp_up_proj_weight_palettized, x = output_427_cast_fp16)[name = tensor<string, []>("linear_124")];
            tensor<fp16, [1, 256, 10240]> input_215 = mul(x = var_4060, y = var_4062)[name = tensor<string, []>("input_215")];
            tensor<fp16, [1, 256, 2560]> x_427 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_17_mlp_down_proj_weight_palettized, x = input_215)[name = tensor<string, []>("linear_125")];
            tensor<fp16, []> var_22_promoted_107_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_107_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_4068_cast_fp16 = pow(x = x_427, y = var_22_promoted_107_to_fp16)[name = tensor<string, []>("op_4068_cast_fp16")];
            tensor<int32, [1]> var_4070_axes_0 = const()[name = tensor<string, []>("op_4070_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4070_keep_dims_0 = const()[name = tensor<string, []>("op_4070_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_4070_cast_fp16 = reduce_mean(axes = var_4070_axes_0, keep_dims = var_4070_keep_dims_0, x = var_4068_cast_fp16)[name = tensor<string, []>("op_4070_cast_fp16")];
            tensor<fp16, []> var_4071_to_fp16 = const()[name = tensor<string, []>("op_4071_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_4072_cast_fp16 = add(x = var_4070_cast_fp16, y = var_4071_to_fp16)[name = tensor<string, []>("op_4072_cast_fp16")];
            tensor<fp32, []> var_4073_epsilon_0 = const()[name = tensor<string, []>("op_4073_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_4073_cast_fp16 = rsqrt(epsilon = var_4073_epsilon_0, x = var_4072_cast_fp16)[name = tensor<string, []>("op_4073_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_429_cast_fp16 = mul(x = x_427, y = var_4073_cast_fp16)[name = tensor<string, []>("output_429_cast_fp16")];
            tensor<fp16, [2560]> var_4077_to_fp16 = const()[name = tensor<string, []>("op_4077_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940579072)))];
            tensor<fp16, [1, 256, 2560]> output_431_cast_fp16 = mul(x = output_429_cast_fp16, y = var_4077_to_fp16)[name = tensor<string, []>("output_431_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_431 = add(x = x_423, y = output_431_cast_fp16)[name = tensor<string, []>("x_431")];
            tensor<fp16, []> var_22_promoted_108_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_108_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_4089_cast_fp16 = pow(x = x_431, y = var_22_promoted_108_to_fp16)[name = tensor<string, []>("op_4089_cast_fp16")];
            tensor<int32, [1]> var_4091_axes_0 = const()[name = tensor<string, []>("op_4091_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4091_keep_dims_0 = const()[name = tensor<string, []>("op_4091_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_4091_cast_fp16 = reduce_mean(axes = var_4091_axes_0, keep_dims = var_4091_keep_dims_0, x = var_4089_cast_fp16)[name = tensor<string, []>("op_4091_cast_fp16")];
            tensor<fp16, []> var_4092_to_fp16 = const()[name = tensor<string, []>("op_4092_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_4093_cast_fp16 = add(x = var_4091_cast_fp16, y = var_4092_to_fp16)[name = tensor<string, []>("op_4093_cast_fp16")];
            tensor<fp32, []> var_4094_epsilon_0 = const()[name = tensor<string, []>("op_4094_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_4094_cast_fp16 = rsqrt(epsilon = var_4094_epsilon_0, x = var_4093_cast_fp16)[name = tensor<string, []>("op_4094_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_433_cast_fp16 = mul(x = x_431, y = var_4094_cast_fp16)[name = tensor<string, []>("output_433_cast_fp16")];
            tensor<fp16, [2560]> var_4098_to_fp16 = const()[name = tensor<string, []>("op_4098_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940584256)))];
            tensor<fp16, [1, 256, 2560]> output_435_cast_fp16 = mul(x = output_433_cast_fp16, y = var_4098_to_fp16)[name = tensor<string, []>("output_435_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_4110 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_18_self_attn_q_proj_weight_palettized, x = output_435_cast_fp16)[name = tensor<string, []>("linear_126")];
            tensor<int32, [4]> var_4111 = const()[name = tensor<string, []>("op_4111"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_4112 = reshape(shape = var_4111, x = var_4110)[name = tensor<string, []>("op_4112")];
            tensor<int32, [4]> x_435_perm_0 = const()[name = tensor<string, []>("x_435_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_4115 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_18_self_attn_k_proj_weight_palettized, x = output_435_cast_fp16)[name = tensor<string, []>("linear_127")];
            tensor<int32, [4]> var_4116 = const()[name = tensor<string, []>("op_4116"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_4117 = reshape(shape = var_4116, x = var_4115)[name = tensor<string, []>("op_4117")];
            tensor<int32, [4]> x_439_perm_0 = const()[name = tensor<string, []>("x_439_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_4120 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_18_self_attn_v_proj_weight_palettized, x = output_435_cast_fp16)[name = tensor<string, []>("linear_128")];
            tensor<int32, [4]> var_4121 = const()[name = tensor<string, []>("op_4121"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_4122 = reshape(shape = var_4121, x = var_4120)[name = tensor<string, []>("op_4122")];
            tensor<int32, [4]> hidden_states_259_perm_0 = const()[name = tensor<string, []>("hidden_states_259_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_109_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_109_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_435 = transpose(perm = x_435_perm_0, x = var_4112)[name = tensor<string, []>("transpose_63")];
            tensor<fp16, [1, 8, 256, 256]> var_4126_cast_fp16 = pow(x = x_435, y = var_22_promoted_109_to_fp16)[name = tensor<string, []>("op_4126_cast_fp16")];
            tensor<int32, [1]> var_4128_axes_0 = const()[name = tensor<string, []>("op_4128_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4128_keep_dims_0 = const()[name = tensor<string, []>("op_4128_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_4128_cast_fp16 = reduce_mean(axes = var_4128_axes_0, keep_dims = var_4128_keep_dims_0, x = var_4126_cast_fp16)[name = tensor<string, []>("op_4128_cast_fp16")];
            tensor<fp16, []> var_4129_to_fp16 = const()[name = tensor<string, []>("op_4129_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_4130_cast_fp16 = add(x = var_4128_cast_fp16, y = var_4129_to_fp16)[name = tensor<string, []>("op_4130_cast_fp16")];
            tensor<fp32, []> var_4131_epsilon_0 = const()[name = tensor<string, []>("op_4131_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_4131_cast_fp16 = rsqrt(epsilon = var_4131_epsilon_0, x = var_4130_cast_fp16)[name = tensor<string, []>("op_4131_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_437_cast_fp16 = mul(x = x_435, y = var_4131_cast_fp16)[name = tensor<string, []>("output_437_cast_fp16")];
            tensor<fp16, [256]> var_4135_to_fp16 = const()[name = tensor<string, []>("op_4135_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940589440)))];
            tensor<fp16, [1, 8, 256, 256]> output_439_cast_fp16 = mul(x = output_437_cast_fp16, y = var_4135_to_fp16)[name = tensor<string, []>("output_439_cast_fp16")];
            tensor<fp16, []> var_22_promoted_110_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_110_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_439 = transpose(perm = x_439_perm_0, x = var_4117)[name = tensor<string, []>("transpose_62")];
            tensor<fp16, [1, 4, 256, 256]> var_4140_cast_fp16 = pow(x = x_439, y = var_22_promoted_110_to_fp16)[name = tensor<string, []>("op_4140_cast_fp16")];
            tensor<int32, [1]> var_4142_axes_0 = const()[name = tensor<string, []>("op_4142_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4142_keep_dims_0 = const()[name = tensor<string, []>("op_4142_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_4142_cast_fp16 = reduce_mean(axes = var_4142_axes_0, keep_dims = var_4142_keep_dims_0, x = var_4140_cast_fp16)[name = tensor<string, []>("op_4142_cast_fp16")];
            tensor<fp16, []> var_4143_to_fp16 = const()[name = tensor<string, []>("op_4143_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_4144_cast_fp16 = add(x = var_4142_cast_fp16, y = var_4143_to_fp16)[name = tensor<string, []>("op_4144_cast_fp16")];
            tensor<fp32, []> var_4145_epsilon_0 = const()[name = tensor<string, []>("op_4145_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_4145_cast_fp16 = rsqrt(epsilon = var_4145_epsilon_0, x = var_4144_cast_fp16)[name = tensor<string, []>("op_4145_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_441_cast_fp16 = mul(x = x_439, y = var_4145_cast_fp16)[name = tensor<string, []>("output_441_cast_fp16")];
            tensor<fp16, [256]> var_4149_to_fp16 = const()[name = tensor<string, []>("op_4149_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940590016)))];
            tensor<fp16, [1, 4, 256, 256]> output_443_cast_fp16 = mul(x = output_441_cast_fp16, y = var_4149_to_fp16)[name = tensor<string, []>("output_443_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_4154 = mul(x = output_439_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_4154")];
            tensor<int32, [4]> x1_73_begin_0 = const()[name = tensor<string, []>("x1_73_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_73_end_0 = const()[name = tensor<string, []>("x1_73_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_73_end_mask_0 = const()[name = tensor<string, []>("x1_73_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_73 = slice_by_index(begin = x1_73_begin_0, end = x1_73_end_0, end_mask = x1_73_end_mask_0, x = output_439_cast_fp16)[name = tensor<string, []>("x1_73")];
            tensor<int32, [4]> x2_73_begin_0 = const()[name = tensor<string, []>("x2_73_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_73_end_0 = const()[name = tensor<string, []>("x2_73_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_73_end_mask_0 = const()[name = tensor<string, []>("x2_73_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_73 = slice_by_index(begin = x2_73_begin_0, end = x2_73_end_0, end_mask = x2_73_end_mask_0, x = output_439_cast_fp16)[name = tensor<string, []>("x2_73")];
            tensor<fp16, []> const_449_promoted = const()[name = tensor<string, []>("const_449_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_4165 = mul(x = x2_73, y = const_449_promoted)[name = tensor<string, []>("op_4165")];
            tensor<bool, []> var_4167_interleave_0 = const()[name = tensor<string, []>("op_4167_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_4167 = concat(axis = var_24, interleave = var_4167_interleave_0, values = (var_4165, x1_73))[name = tensor<string, []>("op_4167")];
            tensor<fp16, [1, 8, 256, 256]> var_4168 = mul(x = var_4167, y = sin_7_palettized)[name = tensor<string, []>("op_4168")];
            tensor<fp16, [1, 8, 256, 256]> query_37 = add(x = var_4154, y = var_4168)[name = tensor<string, []>("query_37")];
            tensor<fp16, [1, 4, 256, 256]> var_4170 = mul(x = output_443_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_4170")];
            tensor<int32, [4]> x1_75_begin_0 = const()[name = tensor<string, []>("x1_75_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_75_end_0 = const()[name = tensor<string, []>("x1_75_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_75_end_mask_0 = const()[name = tensor<string, []>("x1_75_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_75 = slice_by_index(begin = x1_75_begin_0, end = x1_75_end_0, end_mask = x1_75_end_mask_0, x = output_443_cast_fp16)[name = tensor<string, []>("x1_75")];
            tensor<int32, [4]> x2_75_begin_0 = const()[name = tensor<string, []>("x2_75_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_75_end_0 = const()[name = tensor<string, []>("x2_75_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_75_end_mask_0 = const()[name = tensor<string, []>("x2_75_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_75 = slice_by_index(begin = x2_75_begin_0, end = x2_75_end_0, end_mask = x2_75_end_mask_0, x = output_443_cast_fp16)[name = tensor<string, []>("x2_75")];
            tensor<fp16, []> const_452_promoted = const()[name = tensor<string, []>("const_452_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_4181 = mul(x = x2_75, y = const_452_promoted)[name = tensor<string, []>("op_4181")];
            tensor<bool, []> var_4183_interleave_0 = const()[name = tensor<string, []>("op_4183_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_4183 = concat(axis = var_24, interleave = var_4183_interleave_0, values = (var_4181, x1_75))[name = tensor<string, []>("op_4183")];
            tensor<fp16, [1, 4, 256, 256]> var_4184 = mul(x = var_4183, y = sin_7_palettized)[name = tensor<string, []>("op_4184")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_255 = add(x = var_4170, y = var_4184)[name = tensor<string, []>("hidden_states_255")];
            tensor<int32, [1]> var_4193_axes_0 = const()[name = tensor<string, []>("op_4193_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_4193 = expand_dims(axes = var_4193_axes_0, x = hidden_states_255)[name = tensor<string, []>("op_4193")];
            tensor<int32, [5]> hidden_states_257_reps_0 = const()[name = tensor<string, []>("hidden_states_257_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_257 = tile(reps = hidden_states_257_reps_0, x = var_4193)[name = tensor<string, []>("hidden_states_257")];
            tensor<int32, [4]> var_4201 = const()[name = tensor<string, []>("op_4201"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_37 = reshape(shape = var_4201, x = hidden_states_257)[name = tensor<string, []>("key_states_37")];
            tensor<int32, [1]> var_4210_axes_0 = const()[name = tensor<string, []>("op_4210_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_259 = transpose(perm = hidden_states_259_perm_0, x = var_4122)[name = tensor<string, []>("transpose_61")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_4210 = expand_dims(axes = var_4210_axes_0, x = hidden_states_259)[name = tensor<string, []>("op_4210")];
            tensor<int32, [5]> hidden_states_261_reps_0 = const()[name = tensor<string, []>("hidden_states_261_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_261 = tile(reps = hidden_states_261_reps_0, x = var_4210)[name = tensor<string, []>("hidden_states_261")];
            tensor<int32, [4]> var_4218 = const()[name = tensor<string, []>("op_4218"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_37 = reshape(shape = var_4218, x = hidden_states_261)[name = tensor<string, []>("value_states_37")];
            tensor<bool, []> var_4221_transpose_x_1 = const()[name = tensor<string, []>("op_4221_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_4221_transpose_y_1 = const()[name = tensor<string, []>("op_4221_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_4221 = matmul(transpose_x = var_4221_transpose_x_1, transpose_y = var_4221_transpose_y_1, x = query_37, y = key_states_37)[name = tensor<string, []>("op_4221")];
            tensor<fp16, []> var_4222_to_fp16 = const()[name = tensor<string, []>("op_4222_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_73_cast_fp16 = mul(x = var_4221, y = var_4222_to_fp16)[name = tensor<string, []>("attn_weights_73_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_217_cast_fp16 = add(x = attn_weights_73_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_217_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_4230_cast_fp16 = softmax(axis = var_24, x = input_217_cast_fp16)[name = tensor<string, []>("op_4230_cast_fp16")];
            tensor<bool, []> attn_output_73_transpose_x_0 = const()[name = tensor<string, []>("attn_output_73_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_73_transpose_y_0 = const()[name = tensor<string, []>("attn_output_73_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_73 = matmul(transpose_x = attn_output_73_transpose_x_0, transpose_y = attn_output_73_transpose_y_0, x = var_4230_cast_fp16, y = value_states_37)[name = tensor<string, []>("attn_output_73")];
            tensor<int32, [4]> var_4234_perm_0 = const()[name = tensor<string, []>("op_4234_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_4236 = const()[name = tensor<string, []>("op_4236"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_4234 = transpose(perm = var_4234_perm_0, x = attn_output_73)[name = tensor<string, []>("transpose_60")];
            tensor<fp16, [1, 256, 2048]> var_4237 = reshape(shape = var_4236, x = var_4234)[name = tensor<string, []>("op_4237")];
            tensor<fp16, [1, 256, 2560]> x_443 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_18_self_attn_o_proj_weight_palettized, x = var_4237)[name = tensor<string, []>("linear_129")];
            tensor<fp16, []> var_22_promoted_111_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_111_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_4243_cast_fp16 = pow(x = x_443, y = var_22_promoted_111_to_fp16)[name = tensor<string, []>("op_4243_cast_fp16")];
            tensor<int32, [1]> var_4245_axes_0 = const()[name = tensor<string, []>("op_4245_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4245_keep_dims_0 = const()[name = tensor<string, []>("op_4245_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_4245_cast_fp16 = reduce_mean(axes = var_4245_axes_0, keep_dims = var_4245_keep_dims_0, x = var_4243_cast_fp16)[name = tensor<string, []>("op_4245_cast_fp16")];
            tensor<fp16, []> var_4246_to_fp16 = const()[name = tensor<string, []>("op_4246_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_4247_cast_fp16 = add(x = var_4245_cast_fp16, y = var_4246_to_fp16)[name = tensor<string, []>("op_4247_cast_fp16")];
            tensor<fp32, []> var_4248_epsilon_0 = const()[name = tensor<string, []>("op_4248_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_4248_cast_fp16 = rsqrt(epsilon = var_4248_epsilon_0, x = var_4247_cast_fp16)[name = tensor<string, []>("op_4248_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_445_cast_fp16 = mul(x = x_443, y = var_4248_cast_fp16)[name = tensor<string, []>("output_445_cast_fp16")];
            tensor<fp16, [2560]> var_4252_to_fp16 = const()[name = tensor<string, []>("op_4252_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940590592)))];
            tensor<fp16, [1, 256, 2560]> output_447_cast_fp16 = mul(x = output_445_cast_fp16, y = var_4252_to_fp16)[name = tensor<string, []>("output_447_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_447 = add(x = x_431, y = output_447_cast_fp16)[name = tensor<string, []>("x_447")];
            tensor<fp16, []> var_22_promoted_112_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_112_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_4258_cast_fp16 = pow(x = x_447, y = var_22_promoted_112_to_fp16)[name = tensor<string, []>("op_4258_cast_fp16")];
            tensor<int32, [1]> var_4260_axes_0 = const()[name = tensor<string, []>("op_4260_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4260_keep_dims_0 = const()[name = tensor<string, []>("op_4260_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_4260_cast_fp16 = reduce_mean(axes = var_4260_axes_0, keep_dims = var_4260_keep_dims_0, x = var_4258_cast_fp16)[name = tensor<string, []>("op_4260_cast_fp16")];
            tensor<fp16, []> var_4261_to_fp16 = const()[name = tensor<string, []>("op_4261_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_4262_cast_fp16 = add(x = var_4260_cast_fp16, y = var_4261_to_fp16)[name = tensor<string, []>("op_4262_cast_fp16")];
            tensor<fp32, []> var_4263_epsilon_0 = const()[name = tensor<string, []>("op_4263_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_4263_cast_fp16 = rsqrt(epsilon = var_4263_epsilon_0, x = var_4262_cast_fp16)[name = tensor<string, []>("op_4263_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_449_cast_fp16 = mul(x = x_447, y = var_4263_cast_fp16)[name = tensor<string, []>("output_449_cast_fp16")];
            tensor<fp16, [2560]> var_4267_to_fp16 = const()[name = tensor<string, []>("op_4267_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940595776)))];
            tensor<fp16, [1, 256, 2560]> output_451_cast_fp16 = mul(x = output_449_cast_fp16, y = var_4267_to_fp16)[name = tensor<string, []>("output_451_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_225 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_18_mlp_gate_proj_weight_palettized, x = output_451_cast_fp16)[name = tensor<string, []>("linear_130")];
            tensor<string, []> var_4275_mode_0 = const()[name = tensor<string, []>("op_4275_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_4275 = gelu(mode = var_4275_mode_0, x = input_225)[name = tensor<string, []>("op_4275")];
            tensor<fp16, [1, 256, 10240]> var_4277 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_18_mlp_up_proj_weight_palettized, x = output_451_cast_fp16)[name = tensor<string, []>("linear_131")];
            tensor<fp16, [1, 256, 10240]> input_227 = mul(x = var_4275, y = var_4277)[name = tensor<string, []>("input_227")];
            tensor<fp16, [1, 256, 2560]> x_451 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_18_mlp_down_proj_weight_palettized, x = input_227)[name = tensor<string, []>("linear_132")];
            tensor<fp16, []> var_22_promoted_113_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_113_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_4283_cast_fp16 = pow(x = x_451, y = var_22_promoted_113_to_fp16)[name = tensor<string, []>("op_4283_cast_fp16")];
            tensor<int32, [1]> var_4285_axes_0 = const()[name = tensor<string, []>("op_4285_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4285_keep_dims_0 = const()[name = tensor<string, []>("op_4285_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_4285_cast_fp16 = reduce_mean(axes = var_4285_axes_0, keep_dims = var_4285_keep_dims_0, x = var_4283_cast_fp16)[name = tensor<string, []>("op_4285_cast_fp16")];
            tensor<fp16, []> var_4286_to_fp16 = const()[name = tensor<string, []>("op_4286_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_4287_cast_fp16 = add(x = var_4285_cast_fp16, y = var_4286_to_fp16)[name = tensor<string, []>("op_4287_cast_fp16")];
            tensor<fp32, []> var_4288_epsilon_0 = const()[name = tensor<string, []>("op_4288_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_4288_cast_fp16 = rsqrt(epsilon = var_4288_epsilon_0, x = var_4287_cast_fp16)[name = tensor<string, []>("op_4288_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_453_cast_fp16 = mul(x = x_451, y = var_4288_cast_fp16)[name = tensor<string, []>("output_453_cast_fp16")];
            tensor<fp16, [2560]> var_4292_to_fp16 = const()[name = tensor<string, []>("op_4292_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940600960)))];
            tensor<fp16, [1, 256, 2560]> output_455_cast_fp16 = mul(x = output_453_cast_fp16, y = var_4292_to_fp16)[name = tensor<string, []>("output_455_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_455 = add(x = x_447, y = output_455_cast_fp16)[name = tensor<string, []>("x_455")];
            tensor<fp16, []> var_22_promoted_114_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_114_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_4304_cast_fp16 = pow(x = x_455, y = var_22_promoted_114_to_fp16)[name = tensor<string, []>("op_4304_cast_fp16")];
            tensor<int32, [1]> var_4306_axes_0 = const()[name = tensor<string, []>("op_4306_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4306_keep_dims_0 = const()[name = tensor<string, []>("op_4306_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_4306_cast_fp16 = reduce_mean(axes = var_4306_axes_0, keep_dims = var_4306_keep_dims_0, x = var_4304_cast_fp16)[name = tensor<string, []>("op_4306_cast_fp16")];
            tensor<fp16, []> var_4307_to_fp16 = const()[name = tensor<string, []>("op_4307_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_4308_cast_fp16 = add(x = var_4306_cast_fp16, y = var_4307_to_fp16)[name = tensor<string, []>("op_4308_cast_fp16")];
            tensor<fp32, []> var_4309_epsilon_0 = const()[name = tensor<string, []>("op_4309_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_4309_cast_fp16 = rsqrt(epsilon = var_4309_epsilon_0, x = var_4308_cast_fp16)[name = tensor<string, []>("op_4309_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_457_cast_fp16 = mul(x = x_455, y = var_4309_cast_fp16)[name = tensor<string, []>("output_457_cast_fp16")];
            tensor<fp16, [2560]> var_4313_to_fp16 = const()[name = tensor<string, []>("op_4313_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940606144)))];
            tensor<fp16, [1, 256, 2560]> output_459_cast_fp16 = mul(x = output_457_cast_fp16, y = var_4313_to_fp16)[name = tensor<string, []>("output_459_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_4325 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_19_self_attn_q_proj_weight_palettized, x = output_459_cast_fp16)[name = tensor<string, []>("linear_133")];
            tensor<int32, [4]> var_4326 = const()[name = tensor<string, []>("op_4326"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_4327 = reshape(shape = var_4326, x = var_4325)[name = tensor<string, []>("op_4327")];
            tensor<int32, [4]> x_459_perm_0 = const()[name = tensor<string, []>("x_459_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_4330 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_19_self_attn_k_proj_weight_palettized, x = output_459_cast_fp16)[name = tensor<string, []>("linear_134")];
            tensor<int32, [4]> var_4331 = const()[name = tensor<string, []>("op_4331"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_4332 = reshape(shape = var_4331, x = var_4330)[name = tensor<string, []>("op_4332")];
            tensor<int32, [4]> x_463_perm_0 = const()[name = tensor<string, []>("x_463_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_4335 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_19_self_attn_v_proj_weight_palettized, x = output_459_cast_fp16)[name = tensor<string, []>("linear_135")];
            tensor<int32, [4]> var_4336 = const()[name = tensor<string, []>("op_4336"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_4337 = reshape(shape = var_4336, x = var_4335)[name = tensor<string, []>("op_4337")];
            tensor<int32, [4]> hidden_states_273_perm_0 = const()[name = tensor<string, []>("hidden_states_273_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_115_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_115_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_459 = transpose(perm = x_459_perm_0, x = var_4327)[name = tensor<string, []>("transpose_59")];
            tensor<fp16, [1, 8, 256, 256]> var_4341_cast_fp16 = pow(x = x_459, y = var_22_promoted_115_to_fp16)[name = tensor<string, []>("op_4341_cast_fp16")];
            tensor<int32, [1]> var_4343_axes_0 = const()[name = tensor<string, []>("op_4343_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4343_keep_dims_0 = const()[name = tensor<string, []>("op_4343_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_4343_cast_fp16 = reduce_mean(axes = var_4343_axes_0, keep_dims = var_4343_keep_dims_0, x = var_4341_cast_fp16)[name = tensor<string, []>("op_4343_cast_fp16")];
            tensor<fp16, []> var_4344_to_fp16 = const()[name = tensor<string, []>("op_4344_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_4345_cast_fp16 = add(x = var_4343_cast_fp16, y = var_4344_to_fp16)[name = tensor<string, []>("op_4345_cast_fp16")];
            tensor<fp32, []> var_4346_epsilon_0 = const()[name = tensor<string, []>("op_4346_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_4346_cast_fp16 = rsqrt(epsilon = var_4346_epsilon_0, x = var_4345_cast_fp16)[name = tensor<string, []>("op_4346_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_461_cast_fp16 = mul(x = x_459, y = var_4346_cast_fp16)[name = tensor<string, []>("output_461_cast_fp16")];
            tensor<fp16, [256]> var_4350_to_fp16 = const()[name = tensor<string, []>("op_4350_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940611328)))];
            tensor<fp16, [1, 8, 256, 256]> output_463_cast_fp16 = mul(x = output_461_cast_fp16, y = var_4350_to_fp16)[name = tensor<string, []>("output_463_cast_fp16")];
            tensor<fp16, []> var_22_promoted_116_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_116_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_463 = transpose(perm = x_463_perm_0, x = var_4332)[name = tensor<string, []>("transpose_58")];
            tensor<fp16, [1, 4, 256, 256]> var_4355_cast_fp16 = pow(x = x_463, y = var_22_promoted_116_to_fp16)[name = tensor<string, []>("op_4355_cast_fp16")];
            tensor<int32, [1]> var_4357_axes_0 = const()[name = tensor<string, []>("op_4357_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4357_keep_dims_0 = const()[name = tensor<string, []>("op_4357_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_4357_cast_fp16 = reduce_mean(axes = var_4357_axes_0, keep_dims = var_4357_keep_dims_0, x = var_4355_cast_fp16)[name = tensor<string, []>("op_4357_cast_fp16")];
            tensor<fp16, []> var_4358_to_fp16 = const()[name = tensor<string, []>("op_4358_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_4359_cast_fp16 = add(x = var_4357_cast_fp16, y = var_4358_to_fp16)[name = tensor<string, []>("op_4359_cast_fp16")];
            tensor<fp32, []> var_4360_epsilon_0 = const()[name = tensor<string, []>("op_4360_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_4360_cast_fp16 = rsqrt(epsilon = var_4360_epsilon_0, x = var_4359_cast_fp16)[name = tensor<string, []>("op_4360_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_465_cast_fp16 = mul(x = x_463, y = var_4360_cast_fp16)[name = tensor<string, []>("output_465_cast_fp16")];
            tensor<fp16, [256]> var_4364_to_fp16 = const()[name = tensor<string, []>("op_4364_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940611904)))];
            tensor<fp16, [1, 4, 256, 256]> output_467_cast_fp16 = mul(x = output_465_cast_fp16, y = var_4364_to_fp16)[name = tensor<string, []>("output_467_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_4369 = mul(x = output_463_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_4369")];
            tensor<int32, [4]> x1_77_begin_0 = const()[name = tensor<string, []>("x1_77_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_77_end_0 = const()[name = tensor<string, []>("x1_77_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_77_end_mask_0 = const()[name = tensor<string, []>("x1_77_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_77 = slice_by_index(begin = x1_77_begin_0, end = x1_77_end_0, end_mask = x1_77_end_mask_0, x = output_463_cast_fp16)[name = tensor<string, []>("x1_77")];
            tensor<int32, [4]> x2_77_begin_0 = const()[name = tensor<string, []>("x2_77_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_77_end_0 = const()[name = tensor<string, []>("x2_77_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_77_end_mask_0 = const()[name = tensor<string, []>("x2_77_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_77 = slice_by_index(begin = x2_77_begin_0, end = x2_77_end_0, end_mask = x2_77_end_mask_0, x = output_463_cast_fp16)[name = tensor<string, []>("x2_77")];
            tensor<fp16, []> const_472_promoted = const()[name = tensor<string, []>("const_472_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_4380 = mul(x = x2_77, y = const_472_promoted)[name = tensor<string, []>("op_4380")];
            tensor<bool, []> var_4382_interleave_0 = const()[name = tensor<string, []>("op_4382_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_4382 = concat(axis = var_24, interleave = var_4382_interleave_0, values = (var_4380, x1_77))[name = tensor<string, []>("op_4382")];
            tensor<fp16, [1, 8, 256, 256]> var_4383 = mul(x = var_4382, y = sin_7_palettized)[name = tensor<string, []>("op_4383")];
            tensor<fp16, [1, 8, 256, 256]> query_39 = add(x = var_4369, y = var_4383)[name = tensor<string, []>("query_39")];
            tensor<fp16, [1, 4, 256, 256]> var_4385 = mul(x = output_467_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_4385")];
            tensor<int32, [4]> x1_79_begin_0 = const()[name = tensor<string, []>("x1_79_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_79_end_0 = const()[name = tensor<string, []>("x1_79_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_79_end_mask_0 = const()[name = tensor<string, []>("x1_79_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_79 = slice_by_index(begin = x1_79_begin_0, end = x1_79_end_0, end_mask = x1_79_end_mask_0, x = output_467_cast_fp16)[name = tensor<string, []>("x1_79")];
            tensor<int32, [4]> x2_79_begin_0 = const()[name = tensor<string, []>("x2_79_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_79_end_0 = const()[name = tensor<string, []>("x2_79_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_79_end_mask_0 = const()[name = tensor<string, []>("x2_79_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_79 = slice_by_index(begin = x2_79_begin_0, end = x2_79_end_0, end_mask = x2_79_end_mask_0, x = output_467_cast_fp16)[name = tensor<string, []>("x2_79")];
            tensor<fp16, []> const_475_promoted = const()[name = tensor<string, []>("const_475_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_4396 = mul(x = x2_79, y = const_475_promoted)[name = tensor<string, []>("op_4396")];
            tensor<bool, []> var_4398_interleave_0 = const()[name = tensor<string, []>("op_4398_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_4398 = concat(axis = var_24, interleave = var_4398_interleave_0, values = (var_4396, x1_79))[name = tensor<string, []>("op_4398")];
            tensor<fp16, [1, 4, 256, 256]> var_4399 = mul(x = var_4398, y = sin_7_palettized)[name = tensor<string, []>("op_4399")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_269 = add(x = var_4385, y = var_4399)[name = tensor<string, []>("hidden_states_269")];
            tensor<int32, [1]> var_4408_axes_0 = const()[name = tensor<string, []>("op_4408_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_4408 = expand_dims(axes = var_4408_axes_0, x = hidden_states_269)[name = tensor<string, []>("op_4408")];
            tensor<int32, [5]> hidden_states_271_reps_0 = const()[name = tensor<string, []>("hidden_states_271_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_271 = tile(reps = hidden_states_271_reps_0, x = var_4408)[name = tensor<string, []>("hidden_states_271")];
            tensor<int32, [4]> var_4416 = const()[name = tensor<string, []>("op_4416"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_39 = reshape(shape = var_4416, x = hidden_states_271)[name = tensor<string, []>("key_states_39")];
            tensor<int32, [1]> var_4425_axes_0 = const()[name = tensor<string, []>("op_4425_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_273 = transpose(perm = hidden_states_273_perm_0, x = var_4337)[name = tensor<string, []>("transpose_57")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_4425 = expand_dims(axes = var_4425_axes_0, x = hidden_states_273)[name = tensor<string, []>("op_4425")];
            tensor<int32, [5]> hidden_states_275_reps_0 = const()[name = tensor<string, []>("hidden_states_275_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_275 = tile(reps = hidden_states_275_reps_0, x = var_4425)[name = tensor<string, []>("hidden_states_275")];
            tensor<int32, [4]> var_4433 = const()[name = tensor<string, []>("op_4433"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_39 = reshape(shape = var_4433, x = hidden_states_275)[name = tensor<string, []>("value_states_39")];
            tensor<bool, []> var_4436_transpose_x_1 = const()[name = tensor<string, []>("op_4436_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_4436_transpose_y_1 = const()[name = tensor<string, []>("op_4436_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_4436 = matmul(transpose_x = var_4436_transpose_x_1, transpose_y = var_4436_transpose_y_1, x = query_39, y = key_states_39)[name = tensor<string, []>("op_4436")];
            tensor<fp16, []> var_4437_to_fp16 = const()[name = tensor<string, []>("op_4437_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_77_cast_fp16 = mul(x = var_4436, y = var_4437_to_fp16)[name = tensor<string, []>("attn_weights_77_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_229_cast_fp16 = add(x = attn_weights_77_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_229_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_4445_cast_fp16 = softmax(axis = var_24, x = input_229_cast_fp16)[name = tensor<string, []>("op_4445_cast_fp16")];
            tensor<bool, []> attn_output_77_transpose_x_0 = const()[name = tensor<string, []>("attn_output_77_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_77_transpose_y_0 = const()[name = tensor<string, []>("attn_output_77_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_77 = matmul(transpose_x = attn_output_77_transpose_x_0, transpose_y = attn_output_77_transpose_y_0, x = var_4445_cast_fp16, y = value_states_39)[name = tensor<string, []>("attn_output_77")];
            tensor<int32, [4]> var_4449_perm_0 = const()[name = tensor<string, []>("op_4449_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_4451 = const()[name = tensor<string, []>("op_4451"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_4449 = transpose(perm = var_4449_perm_0, x = attn_output_77)[name = tensor<string, []>("transpose_56")];
            tensor<fp16, [1, 256, 2048]> var_4452 = reshape(shape = var_4451, x = var_4449)[name = tensor<string, []>("op_4452")];
            tensor<fp16, [1, 256, 2560]> x_467 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_19_self_attn_o_proj_weight_palettized, x = var_4452)[name = tensor<string, []>("linear_136")];
            tensor<fp16, []> var_22_promoted_117_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_117_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_4458_cast_fp16 = pow(x = x_467, y = var_22_promoted_117_to_fp16)[name = tensor<string, []>("op_4458_cast_fp16")];
            tensor<int32, [1]> var_4460_axes_0 = const()[name = tensor<string, []>("op_4460_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4460_keep_dims_0 = const()[name = tensor<string, []>("op_4460_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_4460_cast_fp16 = reduce_mean(axes = var_4460_axes_0, keep_dims = var_4460_keep_dims_0, x = var_4458_cast_fp16)[name = tensor<string, []>("op_4460_cast_fp16")];
            tensor<fp16, []> var_4461_to_fp16 = const()[name = tensor<string, []>("op_4461_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_4462_cast_fp16 = add(x = var_4460_cast_fp16, y = var_4461_to_fp16)[name = tensor<string, []>("op_4462_cast_fp16")];
            tensor<fp32, []> var_4463_epsilon_0 = const()[name = tensor<string, []>("op_4463_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_4463_cast_fp16 = rsqrt(epsilon = var_4463_epsilon_0, x = var_4462_cast_fp16)[name = tensor<string, []>("op_4463_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_469_cast_fp16 = mul(x = x_467, y = var_4463_cast_fp16)[name = tensor<string, []>("output_469_cast_fp16")];
            tensor<fp16, [2560]> var_4467_to_fp16 = const()[name = tensor<string, []>("op_4467_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940612480)))];
            tensor<fp16, [1, 256, 2560]> output_471_cast_fp16 = mul(x = output_469_cast_fp16, y = var_4467_to_fp16)[name = tensor<string, []>("output_471_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_471 = add(x = x_455, y = output_471_cast_fp16)[name = tensor<string, []>("x_471")];
            tensor<fp16, []> var_22_promoted_118_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_118_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_4473_cast_fp16 = pow(x = x_471, y = var_22_promoted_118_to_fp16)[name = tensor<string, []>("op_4473_cast_fp16")];
            tensor<int32, [1]> var_4475_axes_0 = const()[name = tensor<string, []>("op_4475_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4475_keep_dims_0 = const()[name = tensor<string, []>("op_4475_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_4475_cast_fp16 = reduce_mean(axes = var_4475_axes_0, keep_dims = var_4475_keep_dims_0, x = var_4473_cast_fp16)[name = tensor<string, []>("op_4475_cast_fp16")];
            tensor<fp16, []> var_4476_to_fp16 = const()[name = tensor<string, []>("op_4476_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_4477_cast_fp16 = add(x = var_4475_cast_fp16, y = var_4476_to_fp16)[name = tensor<string, []>("op_4477_cast_fp16")];
            tensor<fp32, []> var_4478_epsilon_0 = const()[name = tensor<string, []>("op_4478_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_4478_cast_fp16 = rsqrt(epsilon = var_4478_epsilon_0, x = var_4477_cast_fp16)[name = tensor<string, []>("op_4478_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_473_cast_fp16 = mul(x = x_471, y = var_4478_cast_fp16)[name = tensor<string, []>("output_473_cast_fp16")];
            tensor<fp16, [2560]> var_4482_to_fp16 = const()[name = tensor<string, []>("op_4482_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940617664)))];
            tensor<fp16, [1, 256, 2560]> output_475_cast_fp16 = mul(x = output_473_cast_fp16, y = var_4482_to_fp16)[name = tensor<string, []>("output_475_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_237 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_19_mlp_gate_proj_weight_palettized, x = output_475_cast_fp16)[name = tensor<string, []>("linear_137")];
            tensor<string, []> var_4490_mode_0 = const()[name = tensor<string, []>("op_4490_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_4490 = gelu(mode = var_4490_mode_0, x = input_237)[name = tensor<string, []>("op_4490")];
            tensor<fp16, [1, 256, 10240]> var_4492 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_19_mlp_up_proj_weight_palettized, x = output_475_cast_fp16)[name = tensor<string, []>("linear_138")];
            tensor<fp16, [1, 256, 10240]> input_239 = mul(x = var_4490, y = var_4492)[name = tensor<string, []>("input_239")];
            tensor<fp16, [1, 256, 2560]> x_475 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_19_mlp_down_proj_weight_palettized, x = input_239)[name = tensor<string, []>("linear_139")];
            tensor<fp16, []> var_22_promoted_119_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_119_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_4498_cast_fp16 = pow(x = x_475, y = var_22_promoted_119_to_fp16)[name = tensor<string, []>("op_4498_cast_fp16")];
            tensor<int32, [1]> var_4500_axes_0 = const()[name = tensor<string, []>("op_4500_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4500_keep_dims_0 = const()[name = tensor<string, []>("op_4500_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_4500_cast_fp16 = reduce_mean(axes = var_4500_axes_0, keep_dims = var_4500_keep_dims_0, x = var_4498_cast_fp16)[name = tensor<string, []>("op_4500_cast_fp16")];
            tensor<fp16, []> var_4501_to_fp16 = const()[name = tensor<string, []>("op_4501_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_4502_cast_fp16 = add(x = var_4500_cast_fp16, y = var_4501_to_fp16)[name = tensor<string, []>("op_4502_cast_fp16")];
            tensor<fp32, []> var_4503_epsilon_0 = const()[name = tensor<string, []>("op_4503_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_4503_cast_fp16 = rsqrt(epsilon = var_4503_epsilon_0, x = var_4502_cast_fp16)[name = tensor<string, []>("op_4503_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_477_cast_fp16 = mul(x = x_475, y = var_4503_cast_fp16)[name = tensor<string, []>("output_477_cast_fp16")];
            tensor<fp16, [2560]> var_4507_to_fp16 = const()[name = tensor<string, []>("op_4507_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940622848)))];
            tensor<fp16, [1, 256, 2560]> output_479_cast_fp16 = mul(x = output_477_cast_fp16, y = var_4507_to_fp16)[name = tensor<string, []>("output_479_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_479 = add(x = x_471, y = output_479_cast_fp16)[name = tensor<string, []>("x_479")];
            tensor<fp16, []> var_22_promoted_120_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_120_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_4519_cast_fp16 = pow(x = x_479, y = var_22_promoted_120_to_fp16)[name = tensor<string, []>("op_4519_cast_fp16")];
            tensor<int32, [1]> var_4521_axes_0 = const()[name = tensor<string, []>("op_4521_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4521_keep_dims_0 = const()[name = tensor<string, []>("op_4521_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_4521_cast_fp16 = reduce_mean(axes = var_4521_axes_0, keep_dims = var_4521_keep_dims_0, x = var_4519_cast_fp16)[name = tensor<string, []>("op_4521_cast_fp16")];
            tensor<fp16, []> var_4522_to_fp16 = const()[name = tensor<string, []>("op_4522_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_4523_cast_fp16 = add(x = var_4521_cast_fp16, y = var_4522_to_fp16)[name = tensor<string, []>("op_4523_cast_fp16")];
            tensor<fp32, []> var_4524_epsilon_0 = const()[name = tensor<string, []>("op_4524_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_4524_cast_fp16 = rsqrt(epsilon = var_4524_epsilon_0, x = var_4523_cast_fp16)[name = tensor<string, []>("op_4524_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_481_cast_fp16 = mul(x = x_479, y = var_4524_cast_fp16)[name = tensor<string, []>("output_481_cast_fp16")];
            tensor<fp16, [2560]> var_4528_to_fp16 = const()[name = tensor<string, []>("op_4528_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940628032)))];
            tensor<fp16, [1, 256, 2560]> output_483_cast_fp16 = mul(x = output_481_cast_fp16, y = var_4528_to_fp16)[name = tensor<string, []>("output_483_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_4540 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_20_self_attn_q_proj_weight_palettized, x = output_483_cast_fp16)[name = tensor<string, []>("linear_140")];
            tensor<int32, [4]> var_4541 = const()[name = tensor<string, []>("op_4541"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_4542 = reshape(shape = var_4541, x = var_4540)[name = tensor<string, []>("op_4542")];
            tensor<int32, [4]> x_483_perm_0 = const()[name = tensor<string, []>("x_483_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_4545 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_20_self_attn_k_proj_weight_palettized, x = output_483_cast_fp16)[name = tensor<string, []>("linear_141")];
            tensor<int32, [4]> var_4546 = const()[name = tensor<string, []>("op_4546"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_4547 = reshape(shape = var_4546, x = var_4545)[name = tensor<string, []>("op_4547")];
            tensor<int32, [4]> x_487_perm_0 = const()[name = tensor<string, []>("x_487_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_4550 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_20_self_attn_v_proj_weight_palettized, x = output_483_cast_fp16)[name = tensor<string, []>("linear_142")];
            tensor<int32, [4]> var_4551 = const()[name = tensor<string, []>("op_4551"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_4552 = reshape(shape = var_4551, x = var_4550)[name = tensor<string, []>("op_4552")];
            tensor<int32, [4]> hidden_states_287_perm_0 = const()[name = tensor<string, []>("hidden_states_287_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_121_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_121_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_483 = transpose(perm = x_483_perm_0, x = var_4542)[name = tensor<string, []>("transpose_55")];
            tensor<fp16, [1, 8, 256, 256]> var_4556_cast_fp16 = pow(x = x_483, y = var_22_promoted_121_to_fp16)[name = tensor<string, []>("op_4556_cast_fp16")];
            tensor<int32, [1]> var_4558_axes_0 = const()[name = tensor<string, []>("op_4558_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4558_keep_dims_0 = const()[name = tensor<string, []>("op_4558_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_4558_cast_fp16 = reduce_mean(axes = var_4558_axes_0, keep_dims = var_4558_keep_dims_0, x = var_4556_cast_fp16)[name = tensor<string, []>("op_4558_cast_fp16")];
            tensor<fp16, []> var_4559_to_fp16 = const()[name = tensor<string, []>("op_4559_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_4560_cast_fp16 = add(x = var_4558_cast_fp16, y = var_4559_to_fp16)[name = tensor<string, []>("op_4560_cast_fp16")];
            tensor<fp32, []> var_4561_epsilon_0 = const()[name = tensor<string, []>("op_4561_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_4561_cast_fp16 = rsqrt(epsilon = var_4561_epsilon_0, x = var_4560_cast_fp16)[name = tensor<string, []>("op_4561_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_485_cast_fp16 = mul(x = x_483, y = var_4561_cast_fp16)[name = tensor<string, []>("output_485_cast_fp16")];
            tensor<fp16, [256]> var_4565_to_fp16 = const()[name = tensor<string, []>("op_4565_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940633216)))];
            tensor<fp16, [1, 8, 256, 256]> output_487_cast_fp16 = mul(x = output_485_cast_fp16, y = var_4565_to_fp16)[name = tensor<string, []>("output_487_cast_fp16")];
            tensor<fp16, []> var_22_promoted_122_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_122_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_487 = transpose(perm = x_487_perm_0, x = var_4547)[name = tensor<string, []>("transpose_54")];
            tensor<fp16, [1, 4, 256, 256]> var_4570_cast_fp16 = pow(x = x_487, y = var_22_promoted_122_to_fp16)[name = tensor<string, []>("op_4570_cast_fp16")];
            tensor<int32, [1]> var_4572_axes_0 = const()[name = tensor<string, []>("op_4572_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4572_keep_dims_0 = const()[name = tensor<string, []>("op_4572_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_4572_cast_fp16 = reduce_mean(axes = var_4572_axes_0, keep_dims = var_4572_keep_dims_0, x = var_4570_cast_fp16)[name = tensor<string, []>("op_4572_cast_fp16")];
            tensor<fp16, []> var_4573_to_fp16 = const()[name = tensor<string, []>("op_4573_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_4574_cast_fp16 = add(x = var_4572_cast_fp16, y = var_4573_to_fp16)[name = tensor<string, []>("op_4574_cast_fp16")];
            tensor<fp32, []> var_4575_epsilon_0 = const()[name = tensor<string, []>("op_4575_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_4575_cast_fp16 = rsqrt(epsilon = var_4575_epsilon_0, x = var_4574_cast_fp16)[name = tensor<string, []>("op_4575_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_489_cast_fp16 = mul(x = x_487, y = var_4575_cast_fp16)[name = tensor<string, []>("output_489_cast_fp16")];
            tensor<fp16, [256]> var_4579_to_fp16 = const()[name = tensor<string, []>("op_4579_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940633792)))];
            tensor<fp16, [1, 4, 256, 256]> output_491_cast_fp16 = mul(x = output_489_cast_fp16, y = var_4579_to_fp16)[name = tensor<string, []>("output_491_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_4584 = mul(x = output_487_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_4584")];
            tensor<int32, [4]> x1_81_begin_0 = const()[name = tensor<string, []>("x1_81_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_81_end_0 = const()[name = tensor<string, []>("x1_81_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_81_end_mask_0 = const()[name = tensor<string, []>("x1_81_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_81 = slice_by_index(begin = x1_81_begin_0, end = x1_81_end_0, end_mask = x1_81_end_mask_0, x = output_487_cast_fp16)[name = tensor<string, []>("x1_81")];
            tensor<int32, [4]> x2_81_begin_0 = const()[name = tensor<string, []>("x2_81_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_81_end_0 = const()[name = tensor<string, []>("x2_81_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_81_end_mask_0 = const()[name = tensor<string, []>("x2_81_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_81 = slice_by_index(begin = x2_81_begin_0, end = x2_81_end_0, end_mask = x2_81_end_mask_0, x = output_487_cast_fp16)[name = tensor<string, []>("x2_81")];
            tensor<fp16, []> const_495_promoted = const()[name = tensor<string, []>("const_495_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_4595 = mul(x = x2_81, y = const_495_promoted)[name = tensor<string, []>("op_4595")];
            tensor<bool, []> var_4597_interleave_0 = const()[name = tensor<string, []>("op_4597_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_4597 = concat(axis = var_24, interleave = var_4597_interleave_0, values = (var_4595, x1_81))[name = tensor<string, []>("op_4597")];
            tensor<fp16, [1, 8, 256, 256]> var_4598 = mul(x = var_4597, y = sin_7_palettized)[name = tensor<string, []>("op_4598")];
            tensor<fp16, [1, 8, 256, 256]> query_41 = add(x = var_4584, y = var_4598)[name = tensor<string, []>("query_41")];
            tensor<fp16, [1, 4, 256, 256]> var_4600 = mul(x = output_491_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_4600")];
            tensor<int32, [4]> x1_83_begin_0 = const()[name = tensor<string, []>("x1_83_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_83_end_0 = const()[name = tensor<string, []>("x1_83_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_83_end_mask_0 = const()[name = tensor<string, []>("x1_83_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_83 = slice_by_index(begin = x1_83_begin_0, end = x1_83_end_0, end_mask = x1_83_end_mask_0, x = output_491_cast_fp16)[name = tensor<string, []>("x1_83")];
            tensor<int32, [4]> x2_83_begin_0 = const()[name = tensor<string, []>("x2_83_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_83_end_0 = const()[name = tensor<string, []>("x2_83_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_83_end_mask_0 = const()[name = tensor<string, []>("x2_83_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_83 = slice_by_index(begin = x2_83_begin_0, end = x2_83_end_0, end_mask = x2_83_end_mask_0, x = output_491_cast_fp16)[name = tensor<string, []>("x2_83")];
            tensor<fp16, []> const_498_promoted = const()[name = tensor<string, []>("const_498_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_4611 = mul(x = x2_83, y = const_498_promoted)[name = tensor<string, []>("op_4611")];
            tensor<bool, []> var_4613_interleave_0 = const()[name = tensor<string, []>("op_4613_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_4613 = concat(axis = var_24, interleave = var_4613_interleave_0, values = (var_4611, x1_83))[name = tensor<string, []>("op_4613")];
            tensor<fp16, [1, 4, 256, 256]> var_4614 = mul(x = var_4613, y = sin_7_palettized)[name = tensor<string, []>("op_4614")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_283 = add(x = var_4600, y = var_4614)[name = tensor<string, []>("hidden_states_283")];
            tensor<int32, [1]> var_4623_axes_0 = const()[name = tensor<string, []>("op_4623_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_4623 = expand_dims(axes = var_4623_axes_0, x = hidden_states_283)[name = tensor<string, []>("op_4623")];
            tensor<int32, [5]> hidden_states_285_reps_0 = const()[name = tensor<string, []>("hidden_states_285_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_285 = tile(reps = hidden_states_285_reps_0, x = var_4623)[name = tensor<string, []>("hidden_states_285")];
            tensor<int32, [4]> var_4631 = const()[name = tensor<string, []>("op_4631"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_41 = reshape(shape = var_4631, x = hidden_states_285)[name = tensor<string, []>("key_states_41")];
            tensor<int32, [1]> var_4640_axes_0 = const()[name = tensor<string, []>("op_4640_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_287 = transpose(perm = hidden_states_287_perm_0, x = var_4552)[name = tensor<string, []>("transpose_53")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_4640 = expand_dims(axes = var_4640_axes_0, x = hidden_states_287)[name = tensor<string, []>("op_4640")];
            tensor<int32, [5]> hidden_states_289_reps_0 = const()[name = tensor<string, []>("hidden_states_289_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_289 = tile(reps = hidden_states_289_reps_0, x = var_4640)[name = tensor<string, []>("hidden_states_289")];
            tensor<int32, [4]> var_4648 = const()[name = tensor<string, []>("op_4648"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_41 = reshape(shape = var_4648, x = hidden_states_289)[name = tensor<string, []>("value_states_41")];
            tensor<bool, []> var_4651_transpose_x_1 = const()[name = tensor<string, []>("op_4651_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_4651_transpose_y_1 = const()[name = tensor<string, []>("op_4651_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_4651 = matmul(transpose_x = var_4651_transpose_x_1, transpose_y = var_4651_transpose_y_1, x = query_41, y = key_states_41)[name = tensor<string, []>("op_4651")];
            tensor<fp16, []> var_4652_to_fp16 = const()[name = tensor<string, []>("op_4652_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_81_cast_fp16 = mul(x = var_4651, y = var_4652_to_fp16)[name = tensor<string, []>("attn_weights_81_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_241_cast_fp16 = add(x = attn_weights_81_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_241_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_4660_cast_fp16 = softmax(axis = var_24, x = input_241_cast_fp16)[name = tensor<string, []>("op_4660_cast_fp16")];
            tensor<bool, []> attn_output_81_transpose_x_0 = const()[name = tensor<string, []>("attn_output_81_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_81_transpose_y_0 = const()[name = tensor<string, []>("attn_output_81_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_81 = matmul(transpose_x = attn_output_81_transpose_x_0, transpose_y = attn_output_81_transpose_y_0, x = var_4660_cast_fp16, y = value_states_41)[name = tensor<string, []>("attn_output_81")];
            tensor<int32, [4]> var_4664_perm_0 = const()[name = tensor<string, []>("op_4664_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_4666 = const()[name = tensor<string, []>("op_4666"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_4664 = transpose(perm = var_4664_perm_0, x = attn_output_81)[name = tensor<string, []>("transpose_52")];
            tensor<fp16, [1, 256, 2048]> var_4667 = reshape(shape = var_4666, x = var_4664)[name = tensor<string, []>("op_4667")];
            tensor<fp16, [1, 256, 2560]> x_491 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_20_self_attn_o_proj_weight_palettized, x = var_4667)[name = tensor<string, []>("linear_143")];
            tensor<fp16, []> var_22_promoted_123_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_123_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_4673_cast_fp16 = pow(x = x_491, y = var_22_promoted_123_to_fp16)[name = tensor<string, []>("op_4673_cast_fp16")];
            tensor<int32, [1]> var_4675_axes_0 = const()[name = tensor<string, []>("op_4675_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4675_keep_dims_0 = const()[name = tensor<string, []>("op_4675_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_4675_cast_fp16 = reduce_mean(axes = var_4675_axes_0, keep_dims = var_4675_keep_dims_0, x = var_4673_cast_fp16)[name = tensor<string, []>("op_4675_cast_fp16")];
            tensor<fp16, []> var_4676_to_fp16 = const()[name = tensor<string, []>("op_4676_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_4677_cast_fp16 = add(x = var_4675_cast_fp16, y = var_4676_to_fp16)[name = tensor<string, []>("op_4677_cast_fp16")];
            tensor<fp32, []> var_4678_epsilon_0 = const()[name = tensor<string, []>("op_4678_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_4678_cast_fp16 = rsqrt(epsilon = var_4678_epsilon_0, x = var_4677_cast_fp16)[name = tensor<string, []>("op_4678_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_493_cast_fp16 = mul(x = x_491, y = var_4678_cast_fp16)[name = tensor<string, []>("output_493_cast_fp16")];
            tensor<fp16, [2560]> var_4682_to_fp16 = const()[name = tensor<string, []>("op_4682_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940634368)))];
            tensor<fp16, [1, 256, 2560]> output_495_cast_fp16 = mul(x = output_493_cast_fp16, y = var_4682_to_fp16)[name = tensor<string, []>("output_495_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_495 = add(x = x_479, y = output_495_cast_fp16)[name = tensor<string, []>("x_495")];
            tensor<fp16, []> var_22_promoted_124_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_124_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_4688_cast_fp16 = pow(x = x_495, y = var_22_promoted_124_to_fp16)[name = tensor<string, []>("op_4688_cast_fp16")];
            tensor<int32, [1]> var_4690_axes_0 = const()[name = tensor<string, []>("op_4690_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4690_keep_dims_0 = const()[name = tensor<string, []>("op_4690_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_4690_cast_fp16 = reduce_mean(axes = var_4690_axes_0, keep_dims = var_4690_keep_dims_0, x = var_4688_cast_fp16)[name = tensor<string, []>("op_4690_cast_fp16")];
            tensor<fp16, []> var_4691_to_fp16 = const()[name = tensor<string, []>("op_4691_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_4692_cast_fp16 = add(x = var_4690_cast_fp16, y = var_4691_to_fp16)[name = tensor<string, []>("op_4692_cast_fp16")];
            tensor<fp32, []> var_4693_epsilon_0 = const()[name = tensor<string, []>("op_4693_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_4693_cast_fp16 = rsqrt(epsilon = var_4693_epsilon_0, x = var_4692_cast_fp16)[name = tensor<string, []>("op_4693_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_497_cast_fp16 = mul(x = x_495, y = var_4693_cast_fp16)[name = tensor<string, []>("output_497_cast_fp16")];
            tensor<fp16, [2560]> var_4697_to_fp16 = const()[name = tensor<string, []>("op_4697_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940639552)))];
            tensor<fp16, [1, 256, 2560]> output_499_cast_fp16 = mul(x = output_497_cast_fp16, y = var_4697_to_fp16)[name = tensor<string, []>("output_499_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_249 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_20_mlp_gate_proj_weight_palettized, x = output_499_cast_fp16)[name = tensor<string, []>("linear_144")];
            tensor<string, []> var_4705_mode_0 = const()[name = tensor<string, []>("op_4705_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_4705 = gelu(mode = var_4705_mode_0, x = input_249)[name = tensor<string, []>("op_4705")];
            tensor<fp16, [1, 256, 10240]> var_4707 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_20_mlp_up_proj_weight_palettized, x = output_499_cast_fp16)[name = tensor<string, []>("linear_145")];
            tensor<fp16, [1, 256, 10240]> input_251 = mul(x = var_4705, y = var_4707)[name = tensor<string, []>("input_251")];
            tensor<fp16, [1, 256, 2560]> x_499 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_20_mlp_down_proj_weight_palettized, x = input_251)[name = tensor<string, []>("linear_146")];
            tensor<fp16, []> var_22_promoted_125_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_125_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_4713_cast_fp16 = pow(x = x_499, y = var_22_promoted_125_to_fp16)[name = tensor<string, []>("op_4713_cast_fp16")];
            tensor<int32, [1]> var_4715_axes_0 = const()[name = tensor<string, []>("op_4715_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4715_keep_dims_0 = const()[name = tensor<string, []>("op_4715_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_4715_cast_fp16 = reduce_mean(axes = var_4715_axes_0, keep_dims = var_4715_keep_dims_0, x = var_4713_cast_fp16)[name = tensor<string, []>("op_4715_cast_fp16")];
            tensor<fp16, []> var_4716_to_fp16 = const()[name = tensor<string, []>("op_4716_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_4717_cast_fp16 = add(x = var_4715_cast_fp16, y = var_4716_to_fp16)[name = tensor<string, []>("op_4717_cast_fp16")];
            tensor<fp32, []> var_4718_epsilon_0 = const()[name = tensor<string, []>("op_4718_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_4718_cast_fp16 = rsqrt(epsilon = var_4718_epsilon_0, x = var_4717_cast_fp16)[name = tensor<string, []>("op_4718_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_501_cast_fp16 = mul(x = x_499, y = var_4718_cast_fp16)[name = tensor<string, []>("output_501_cast_fp16")];
            tensor<fp16, [2560]> var_4722_to_fp16 = const()[name = tensor<string, []>("op_4722_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940644736)))];
            tensor<fp16, [1, 256, 2560]> output_503_cast_fp16 = mul(x = output_501_cast_fp16, y = var_4722_to_fp16)[name = tensor<string, []>("output_503_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_503 = add(x = x_495, y = output_503_cast_fp16)[name = tensor<string, []>("x_503")];
            tensor<fp16, []> var_22_promoted_126_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_126_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_4734_cast_fp16 = pow(x = x_503, y = var_22_promoted_126_to_fp16)[name = tensor<string, []>("op_4734_cast_fp16")];
            tensor<int32, [1]> var_4736_axes_0 = const()[name = tensor<string, []>("op_4736_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4736_keep_dims_0 = const()[name = tensor<string, []>("op_4736_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_4736_cast_fp16 = reduce_mean(axes = var_4736_axes_0, keep_dims = var_4736_keep_dims_0, x = var_4734_cast_fp16)[name = tensor<string, []>("op_4736_cast_fp16")];
            tensor<fp16, []> var_4737_to_fp16 = const()[name = tensor<string, []>("op_4737_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_4738_cast_fp16 = add(x = var_4736_cast_fp16, y = var_4737_to_fp16)[name = tensor<string, []>("op_4738_cast_fp16")];
            tensor<fp32, []> var_4739_epsilon_0 = const()[name = tensor<string, []>("op_4739_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_4739_cast_fp16 = rsqrt(epsilon = var_4739_epsilon_0, x = var_4738_cast_fp16)[name = tensor<string, []>("op_4739_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_505_cast_fp16 = mul(x = x_503, y = var_4739_cast_fp16)[name = tensor<string, []>("output_505_cast_fp16")];
            tensor<fp16, [2560]> var_4743_to_fp16 = const()[name = tensor<string, []>("op_4743_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940649920)))];
            tensor<fp16, [1, 256, 2560]> output_507_cast_fp16 = mul(x = output_505_cast_fp16, y = var_4743_to_fp16)[name = tensor<string, []>("output_507_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_4755 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_21_self_attn_q_proj_weight_palettized, x = output_507_cast_fp16)[name = tensor<string, []>("linear_147")];
            tensor<int32, [4]> var_4756 = const()[name = tensor<string, []>("op_4756"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_4757 = reshape(shape = var_4756, x = var_4755)[name = tensor<string, []>("op_4757")];
            tensor<int32, [4]> x_507_perm_0 = const()[name = tensor<string, []>("x_507_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_4760 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_21_self_attn_k_proj_weight_palettized, x = output_507_cast_fp16)[name = tensor<string, []>("linear_148")];
            tensor<int32, [4]> var_4761 = const()[name = tensor<string, []>("op_4761"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_4762 = reshape(shape = var_4761, x = var_4760)[name = tensor<string, []>("op_4762")];
            tensor<int32, [4]> x_511_perm_0 = const()[name = tensor<string, []>("x_511_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_4765 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_21_self_attn_v_proj_weight_palettized, x = output_507_cast_fp16)[name = tensor<string, []>("linear_149")];
            tensor<int32, [4]> var_4766 = const()[name = tensor<string, []>("op_4766"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_4767 = reshape(shape = var_4766, x = var_4765)[name = tensor<string, []>("op_4767")];
            tensor<int32, [4]> hidden_states_301_perm_0 = const()[name = tensor<string, []>("hidden_states_301_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_127_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_127_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_507 = transpose(perm = x_507_perm_0, x = var_4757)[name = tensor<string, []>("transpose_51")];
            tensor<fp16, [1, 8, 256, 256]> var_4771_cast_fp16 = pow(x = x_507, y = var_22_promoted_127_to_fp16)[name = tensor<string, []>("op_4771_cast_fp16")];
            tensor<int32, [1]> var_4773_axes_0 = const()[name = tensor<string, []>("op_4773_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4773_keep_dims_0 = const()[name = tensor<string, []>("op_4773_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_4773_cast_fp16 = reduce_mean(axes = var_4773_axes_0, keep_dims = var_4773_keep_dims_0, x = var_4771_cast_fp16)[name = tensor<string, []>("op_4773_cast_fp16")];
            tensor<fp16, []> var_4774_to_fp16 = const()[name = tensor<string, []>("op_4774_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_4775_cast_fp16 = add(x = var_4773_cast_fp16, y = var_4774_to_fp16)[name = tensor<string, []>("op_4775_cast_fp16")];
            tensor<fp32, []> var_4776_epsilon_0 = const()[name = tensor<string, []>("op_4776_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_4776_cast_fp16 = rsqrt(epsilon = var_4776_epsilon_0, x = var_4775_cast_fp16)[name = tensor<string, []>("op_4776_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_509_cast_fp16 = mul(x = x_507, y = var_4776_cast_fp16)[name = tensor<string, []>("output_509_cast_fp16")];
            tensor<fp16, [256]> var_4780_to_fp16 = const()[name = tensor<string, []>("op_4780_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940655104)))];
            tensor<fp16, [1, 8, 256, 256]> output_511_cast_fp16 = mul(x = output_509_cast_fp16, y = var_4780_to_fp16)[name = tensor<string, []>("output_511_cast_fp16")];
            tensor<fp16, []> var_22_promoted_128_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_128_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_511 = transpose(perm = x_511_perm_0, x = var_4762)[name = tensor<string, []>("transpose_50")];
            tensor<fp16, [1, 4, 256, 256]> var_4785_cast_fp16 = pow(x = x_511, y = var_22_promoted_128_to_fp16)[name = tensor<string, []>("op_4785_cast_fp16")];
            tensor<int32, [1]> var_4787_axes_0 = const()[name = tensor<string, []>("op_4787_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4787_keep_dims_0 = const()[name = tensor<string, []>("op_4787_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_4787_cast_fp16 = reduce_mean(axes = var_4787_axes_0, keep_dims = var_4787_keep_dims_0, x = var_4785_cast_fp16)[name = tensor<string, []>("op_4787_cast_fp16")];
            tensor<fp16, []> var_4788_to_fp16 = const()[name = tensor<string, []>("op_4788_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_4789_cast_fp16 = add(x = var_4787_cast_fp16, y = var_4788_to_fp16)[name = tensor<string, []>("op_4789_cast_fp16")];
            tensor<fp32, []> var_4790_epsilon_0 = const()[name = tensor<string, []>("op_4790_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_4790_cast_fp16 = rsqrt(epsilon = var_4790_epsilon_0, x = var_4789_cast_fp16)[name = tensor<string, []>("op_4790_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_513_cast_fp16 = mul(x = x_511, y = var_4790_cast_fp16)[name = tensor<string, []>("output_513_cast_fp16")];
            tensor<fp16, [256]> var_4794_to_fp16 = const()[name = tensor<string, []>("op_4794_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940655680)))];
            tensor<fp16, [1, 4, 256, 256]> output_515_cast_fp16 = mul(x = output_513_cast_fp16, y = var_4794_to_fp16)[name = tensor<string, []>("output_515_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_4799 = mul(x = output_511_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_4799")];
            tensor<int32, [4]> x1_85_begin_0 = const()[name = tensor<string, []>("x1_85_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_85_end_0 = const()[name = tensor<string, []>("x1_85_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_85_end_mask_0 = const()[name = tensor<string, []>("x1_85_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_85 = slice_by_index(begin = x1_85_begin_0, end = x1_85_end_0, end_mask = x1_85_end_mask_0, x = output_511_cast_fp16)[name = tensor<string, []>("x1_85")];
            tensor<int32, [4]> x2_85_begin_0 = const()[name = tensor<string, []>("x2_85_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_85_end_0 = const()[name = tensor<string, []>("x2_85_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_85_end_mask_0 = const()[name = tensor<string, []>("x2_85_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_85 = slice_by_index(begin = x2_85_begin_0, end = x2_85_end_0, end_mask = x2_85_end_mask_0, x = output_511_cast_fp16)[name = tensor<string, []>("x2_85")];
            tensor<fp16, []> const_518_promoted = const()[name = tensor<string, []>("const_518_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_4810 = mul(x = x2_85, y = const_518_promoted)[name = tensor<string, []>("op_4810")];
            tensor<bool, []> var_4812_interleave_0 = const()[name = tensor<string, []>("op_4812_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_4812 = concat(axis = var_24, interleave = var_4812_interleave_0, values = (var_4810, x1_85))[name = tensor<string, []>("op_4812")];
            tensor<fp16, [1, 8, 256, 256]> var_4813 = mul(x = var_4812, y = sin_7_palettized)[name = tensor<string, []>("op_4813")];
            tensor<fp16, [1, 8, 256, 256]> query_43 = add(x = var_4799, y = var_4813)[name = tensor<string, []>("query_43")];
            tensor<fp16, [1, 4, 256, 256]> var_4815 = mul(x = output_515_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_4815")];
            tensor<int32, [4]> x1_87_begin_0 = const()[name = tensor<string, []>("x1_87_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_87_end_0 = const()[name = tensor<string, []>("x1_87_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_87_end_mask_0 = const()[name = tensor<string, []>("x1_87_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_87 = slice_by_index(begin = x1_87_begin_0, end = x1_87_end_0, end_mask = x1_87_end_mask_0, x = output_515_cast_fp16)[name = tensor<string, []>("x1_87")];
            tensor<int32, [4]> x2_87_begin_0 = const()[name = tensor<string, []>("x2_87_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_87_end_0 = const()[name = tensor<string, []>("x2_87_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_87_end_mask_0 = const()[name = tensor<string, []>("x2_87_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_87 = slice_by_index(begin = x2_87_begin_0, end = x2_87_end_0, end_mask = x2_87_end_mask_0, x = output_515_cast_fp16)[name = tensor<string, []>("x2_87")];
            tensor<fp16, []> const_521_promoted = const()[name = tensor<string, []>("const_521_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_4826 = mul(x = x2_87, y = const_521_promoted)[name = tensor<string, []>("op_4826")];
            tensor<bool, []> var_4828_interleave_0 = const()[name = tensor<string, []>("op_4828_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_4828 = concat(axis = var_24, interleave = var_4828_interleave_0, values = (var_4826, x1_87))[name = tensor<string, []>("op_4828")];
            tensor<fp16, [1, 4, 256, 256]> var_4829 = mul(x = var_4828, y = sin_7_palettized)[name = tensor<string, []>("op_4829")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_297 = add(x = var_4815, y = var_4829)[name = tensor<string, []>("hidden_states_297")];
            tensor<int32, [1]> var_4838_axes_0 = const()[name = tensor<string, []>("op_4838_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_4838 = expand_dims(axes = var_4838_axes_0, x = hidden_states_297)[name = tensor<string, []>("op_4838")];
            tensor<int32, [5]> hidden_states_299_reps_0 = const()[name = tensor<string, []>("hidden_states_299_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_299 = tile(reps = hidden_states_299_reps_0, x = var_4838)[name = tensor<string, []>("hidden_states_299")];
            tensor<int32, [4]> var_4846 = const()[name = tensor<string, []>("op_4846"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_43 = reshape(shape = var_4846, x = hidden_states_299)[name = tensor<string, []>("key_states_43")];
            tensor<int32, [1]> var_4855_axes_0 = const()[name = tensor<string, []>("op_4855_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_301 = transpose(perm = hidden_states_301_perm_0, x = var_4767)[name = tensor<string, []>("transpose_49")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_4855 = expand_dims(axes = var_4855_axes_0, x = hidden_states_301)[name = tensor<string, []>("op_4855")];
            tensor<int32, [5]> hidden_states_303_reps_0 = const()[name = tensor<string, []>("hidden_states_303_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_303 = tile(reps = hidden_states_303_reps_0, x = var_4855)[name = tensor<string, []>("hidden_states_303")];
            tensor<int32, [4]> var_4863 = const()[name = tensor<string, []>("op_4863"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_43 = reshape(shape = var_4863, x = hidden_states_303)[name = tensor<string, []>("value_states_43")];
            tensor<bool, []> var_4866_transpose_x_1 = const()[name = tensor<string, []>("op_4866_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_4866_transpose_y_1 = const()[name = tensor<string, []>("op_4866_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_4866 = matmul(transpose_x = var_4866_transpose_x_1, transpose_y = var_4866_transpose_y_1, x = query_43, y = key_states_43)[name = tensor<string, []>("op_4866")];
            tensor<fp16, []> var_4867_to_fp16 = const()[name = tensor<string, []>("op_4867_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_85_cast_fp16 = mul(x = var_4866, y = var_4867_to_fp16)[name = tensor<string, []>("attn_weights_85_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_253_cast_fp16 = add(x = attn_weights_85_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_253_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_4875_cast_fp16 = softmax(axis = var_24, x = input_253_cast_fp16)[name = tensor<string, []>("op_4875_cast_fp16")];
            tensor<bool, []> attn_output_85_transpose_x_0 = const()[name = tensor<string, []>("attn_output_85_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_85_transpose_y_0 = const()[name = tensor<string, []>("attn_output_85_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_85 = matmul(transpose_x = attn_output_85_transpose_x_0, transpose_y = attn_output_85_transpose_y_0, x = var_4875_cast_fp16, y = value_states_43)[name = tensor<string, []>("attn_output_85")];
            tensor<int32, [4]> var_4879_perm_0 = const()[name = tensor<string, []>("op_4879_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_4881 = const()[name = tensor<string, []>("op_4881"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_4879 = transpose(perm = var_4879_perm_0, x = attn_output_85)[name = tensor<string, []>("transpose_48")];
            tensor<fp16, [1, 256, 2048]> var_4882 = reshape(shape = var_4881, x = var_4879)[name = tensor<string, []>("op_4882")];
            tensor<fp16, [1, 256, 2560]> x_515 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_21_self_attn_o_proj_weight_palettized, x = var_4882)[name = tensor<string, []>("linear_150")];
            tensor<fp16, []> var_22_promoted_129_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_129_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_4888_cast_fp16 = pow(x = x_515, y = var_22_promoted_129_to_fp16)[name = tensor<string, []>("op_4888_cast_fp16")];
            tensor<int32, [1]> var_4890_axes_0 = const()[name = tensor<string, []>("op_4890_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4890_keep_dims_0 = const()[name = tensor<string, []>("op_4890_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_4890_cast_fp16 = reduce_mean(axes = var_4890_axes_0, keep_dims = var_4890_keep_dims_0, x = var_4888_cast_fp16)[name = tensor<string, []>("op_4890_cast_fp16")];
            tensor<fp16, []> var_4891_to_fp16 = const()[name = tensor<string, []>("op_4891_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_4892_cast_fp16 = add(x = var_4890_cast_fp16, y = var_4891_to_fp16)[name = tensor<string, []>("op_4892_cast_fp16")];
            tensor<fp32, []> var_4893_epsilon_0 = const()[name = tensor<string, []>("op_4893_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_4893_cast_fp16 = rsqrt(epsilon = var_4893_epsilon_0, x = var_4892_cast_fp16)[name = tensor<string, []>("op_4893_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_517_cast_fp16 = mul(x = x_515, y = var_4893_cast_fp16)[name = tensor<string, []>("output_517_cast_fp16")];
            tensor<fp16, [2560]> var_4897_to_fp16 = const()[name = tensor<string, []>("op_4897_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940656256)))];
            tensor<fp16, [1, 256, 2560]> output_519_cast_fp16 = mul(x = output_517_cast_fp16, y = var_4897_to_fp16)[name = tensor<string, []>("output_519_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_519 = add(x = x_503, y = output_519_cast_fp16)[name = tensor<string, []>("x_519")];
            tensor<fp16, []> var_22_promoted_130_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_130_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_4903_cast_fp16 = pow(x = x_519, y = var_22_promoted_130_to_fp16)[name = tensor<string, []>("op_4903_cast_fp16")];
            tensor<int32, [1]> var_4905_axes_0 = const()[name = tensor<string, []>("op_4905_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4905_keep_dims_0 = const()[name = tensor<string, []>("op_4905_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_4905_cast_fp16 = reduce_mean(axes = var_4905_axes_0, keep_dims = var_4905_keep_dims_0, x = var_4903_cast_fp16)[name = tensor<string, []>("op_4905_cast_fp16")];
            tensor<fp16, []> var_4906_to_fp16 = const()[name = tensor<string, []>("op_4906_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_4907_cast_fp16 = add(x = var_4905_cast_fp16, y = var_4906_to_fp16)[name = tensor<string, []>("op_4907_cast_fp16")];
            tensor<fp32, []> var_4908_epsilon_0 = const()[name = tensor<string, []>("op_4908_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_4908_cast_fp16 = rsqrt(epsilon = var_4908_epsilon_0, x = var_4907_cast_fp16)[name = tensor<string, []>("op_4908_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_521_cast_fp16 = mul(x = x_519, y = var_4908_cast_fp16)[name = tensor<string, []>("output_521_cast_fp16")];
            tensor<fp16, [2560]> var_4912_to_fp16 = const()[name = tensor<string, []>("op_4912_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940661440)))];
            tensor<fp16, [1, 256, 2560]> output_523_cast_fp16 = mul(x = output_521_cast_fp16, y = var_4912_to_fp16)[name = tensor<string, []>("output_523_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_261 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_21_mlp_gate_proj_weight_palettized, x = output_523_cast_fp16)[name = tensor<string, []>("linear_151")];
            tensor<string, []> var_4920_mode_0 = const()[name = tensor<string, []>("op_4920_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_4920 = gelu(mode = var_4920_mode_0, x = input_261)[name = tensor<string, []>("op_4920")];
            tensor<fp16, [1, 256, 10240]> var_4922 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_21_mlp_up_proj_weight_palettized, x = output_523_cast_fp16)[name = tensor<string, []>("linear_152")];
            tensor<fp16, [1, 256, 10240]> input_263 = mul(x = var_4920, y = var_4922)[name = tensor<string, []>("input_263")];
            tensor<fp16, [1, 256, 2560]> x_523 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_21_mlp_down_proj_weight_palettized, x = input_263)[name = tensor<string, []>("linear_153")];
            tensor<fp16, []> var_22_promoted_131_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_131_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_4928_cast_fp16 = pow(x = x_523, y = var_22_promoted_131_to_fp16)[name = tensor<string, []>("op_4928_cast_fp16")];
            tensor<int32, [1]> var_4930_axes_0 = const()[name = tensor<string, []>("op_4930_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4930_keep_dims_0 = const()[name = tensor<string, []>("op_4930_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_4930_cast_fp16 = reduce_mean(axes = var_4930_axes_0, keep_dims = var_4930_keep_dims_0, x = var_4928_cast_fp16)[name = tensor<string, []>("op_4930_cast_fp16")];
            tensor<fp16, []> var_4931_to_fp16 = const()[name = tensor<string, []>("op_4931_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_4932_cast_fp16 = add(x = var_4930_cast_fp16, y = var_4931_to_fp16)[name = tensor<string, []>("op_4932_cast_fp16")];
            tensor<fp32, []> var_4933_epsilon_0 = const()[name = tensor<string, []>("op_4933_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_4933_cast_fp16 = rsqrt(epsilon = var_4933_epsilon_0, x = var_4932_cast_fp16)[name = tensor<string, []>("op_4933_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_525_cast_fp16 = mul(x = x_523, y = var_4933_cast_fp16)[name = tensor<string, []>("output_525_cast_fp16")];
            tensor<fp16, [2560]> var_4937_to_fp16 = const()[name = tensor<string, []>("op_4937_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940666624)))];
            tensor<fp16, [1, 256, 2560]> output_527_cast_fp16 = mul(x = output_525_cast_fp16, y = var_4937_to_fp16)[name = tensor<string, []>("output_527_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_527 = add(x = x_519, y = output_527_cast_fp16)[name = tensor<string, []>("x_527")];
            tensor<fp16, []> var_22_promoted_132_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_132_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_4949_cast_fp16 = pow(x = x_527, y = var_22_promoted_132_to_fp16)[name = tensor<string, []>("op_4949_cast_fp16")];
            tensor<int32, [1]> var_4951_axes_0 = const()[name = tensor<string, []>("op_4951_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4951_keep_dims_0 = const()[name = tensor<string, []>("op_4951_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_4951_cast_fp16 = reduce_mean(axes = var_4951_axes_0, keep_dims = var_4951_keep_dims_0, x = var_4949_cast_fp16)[name = tensor<string, []>("op_4951_cast_fp16")];
            tensor<fp16, []> var_4952_to_fp16 = const()[name = tensor<string, []>("op_4952_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_4953_cast_fp16 = add(x = var_4951_cast_fp16, y = var_4952_to_fp16)[name = tensor<string, []>("op_4953_cast_fp16")];
            tensor<fp32, []> var_4954_epsilon_0 = const()[name = tensor<string, []>("op_4954_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_4954_cast_fp16 = rsqrt(epsilon = var_4954_epsilon_0, x = var_4953_cast_fp16)[name = tensor<string, []>("op_4954_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_529_cast_fp16 = mul(x = x_527, y = var_4954_cast_fp16)[name = tensor<string, []>("output_529_cast_fp16")];
            tensor<fp16, [2560]> var_4958_to_fp16 = const()[name = tensor<string, []>("op_4958_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940671808)))];
            tensor<fp16, [1, 256, 2560]> output_531_cast_fp16 = mul(x = output_529_cast_fp16, y = var_4958_to_fp16)[name = tensor<string, []>("output_531_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_4970 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_22_self_attn_q_proj_weight_palettized, x = output_531_cast_fp16)[name = tensor<string, []>("linear_154")];
            tensor<int32, [4]> var_4971 = const()[name = tensor<string, []>("op_4971"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_4972 = reshape(shape = var_4971, x = var_4970)[name = tensor<string, []>("op_4972")];
            tensor<int32, [4]> x_531_perm_0 = const()[name = tensor<string, []>("x_531_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_4975 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_22_self_attn_k_proj_weight_palettized, x = output_531_cast_fp16)[name = tensor<string, []>("linear_155")];
            tensor<int32, [4]> var_4976 = const()[name = tensor<string, []>("op_4976"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_4977 = reshape(shape = var_4976, x = var_4975)[name = tensor<string, []>("op_4977")];
            tensor<int32, [4]> x_535_perm_0 = const()[name = tensor<string, []>("x_535_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_4980 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_22_self_attn_v_proj_weight_palettized, x = output_531_cast_fp16)[name = tensor<string, []>("linear_156")];
            tensor<int32, [4]> var_4981 = const()[name = tensor<string, []>("op_4981"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_4982 = reshape(shape = var_4981, x = var_4980)[name = tensor<string, []>("op_4982")];
            tensor<int32, [4]> hidden_states_315_perm_0 = const()[name = tensor<string, []>("hidden_states_315_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_133_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_133_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_531 = transpose(perm = x_531_perm_0, x = var_4972)[name = tensor<string, []>("transpose_47")];
            tensor<fp16, [1, 8, 256, 256]> var_4986_cast_fp16 = pow(x = x_531, y = var_22_promoted_133_to_fp16)[name = tensor<string, []>("op_4986_cast_fp16")];
            tensor<int32, [1]> var_4988_axes_0 = const()[name = tensor<string, []>("op_4988_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_4988_keep_dims_0 = const()[name = tensor<string, []>("op_4988_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_4988_cast_fp16 = reduce_mean(axes = var_4988_axes_0, keep_dims = var_4988_keep_dims_0, x = var_4986_cast_fp16)[name = tensor<string, []>("op_4988_cast_fp16")];
            tensor<fp16, []> var_4989_to_fp16 = const()[name = tensor<string, []>("op_4989_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_4990_cast_fp16 = add(x = var_4988_cast_fp16, y = var_4989_to_fp16)[name = tensor<string, []>("op_4990_cast_fp16")];
            tensor<fp32, []> var_4991_epsilon_0 = const()[name = tensor<string, []>("op_4991_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_4991_cast_fp16 = rsqrt(epsilon = var_4991_epsilon_0, x = var_4990_cast_fp16)[name = tensor<string, []>("op_4991_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_533_cast_fp16 = mul(x = x_531, y = var_4991_cast_fp16)[name = tensor<string, []>("output_533_cast_fp16")];
            tensor<fp16, [256]> var_4995_to_fp16 = const()[name = tensor<string, []>("op_4995_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940676992)))];
            tensor<fp16, [1, 8, 256, 256]> output_535_cast_fp16 = mul(x = output_533_cast_fp16, y = var_4995_to_fp16)[name = tensor<string, []>("output_535_cast_fp16")];
            tensor<fp16, []> var_22_promoted_134_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_134_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_535 = transpose(perm = x_535_perm_0, x = var_4977)[name = tensor<string, []>("transpose_46")];
            tensor<fp16, [1, 4, 256, 256]> var_5000_cast_fp16 = pow(x = x_535, y = var_22_promoted_134_to_fp16)[name = tensor<string, []>("op_5000_cast_fp16")];
            tensor<int32, [1]> var_5002_axes_0 = const()[name = tensor<string, []>("op_5002_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5002_keep_dims_0 = const()[name = tensor<string, []>("op_5002_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_5002_cast_fp16 = reduce_mean(axes = var_5002_axes_0, keep_dims = var_5002_keep_dims_0, x = var_5000_cast_fp16)[name = tensor<string, []>("op_5002_cast_fp16")];
            tensor<fp16, []> var_5003_to_fp16 = const()[name = tensor<string, []>("op_5003_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_5004_cast_fp16 = add(x = var_5002_cast_fp16, y = var_5003_to_fp16)[name = tensor<string, []>("op_5004_cast_fp16")];
            tensor<fp32, []> var_5005_epsilon_0 = const()[name = tensor<string, []>("op_5005_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_5005_cast_fp16 = rsqrt(epsilon = var_5005_epsilon_0, x = var_5004_cast_fp16)[name = tensor<string, []>("op_5005_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_537_cast_fp16 = mul(x = x_535, y = var_5005_cast_fp16)[name = tensor<string, []>("output_537_cast_fp16")];
            tensor<fp16, [256]> var_5009_to_fp16 = const()[name = tensor<string, []>("op_5009_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940677568)))];
            tensor<fp16, [1, 4, 256, 256]> output_539_cast_fp16 = mul(x = output_537_cast_fp16, y = var_5009_to_fp16)[name = tensor<string, []>("output_539_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_5014 = mul(x = output_535_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_5014")];
            tensor<int32, [4]> x1_89_begin_0 = const()[name = tensor<string, []>("x1_89_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_89_end_0 = const()[name = tensor<string, []>("x1_89_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_89_end_mask_0 = const()[name = tensor<string, []>("x1_89_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_89 = slice_by_index(begin = x1_89_begin_0, end = x1_89_end_0, end_mask = x1_89_end_mask_0, x = output_535_cast_fp16)[name = tensor<string, []>("x1_89")];
            tensor<int32, [4]> x2_89_begin_0 = const()[name = tensor<string, []>("x2_89_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_89_end_0 = const()[name = tensor<string, []>("x2_89_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_89_end_mask_0 = const()[name = tensor<string, []>("x2_89_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_89 = slice_by_index(begin = x2_89_begin_0, end = x2_89_end_0, end_mask = x2_89_end_mask_0, x = output_535_cast_fp16)[name = tensor<string, []>("x2_89")];
            tensor<fp16, []> const_541_promoted = const()[name = tensor<string, []>("const_541_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_5025 = mul(x = x2_89, y = const_541_promoted)[name = tensor<string, []>("op_5025")];
            tensor<bool, []> var_5027_interleave_0 = const()[name = tensor<string, []>("op_5027_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_5027 = concat(axis = var_24, interleave = var_5027_interleave_0, values = (var_5025, x1_89))[name = tensor<string, []>("op_5027")];
            tensor<fp16, [1, 8, 256, 256]> var_5028 = mul(x = var_5027, y = sin_7_palettized)[name = tensor<string, []>("op_5028")];
            tensor<fp16, [1, 8, 256, 256]> query_45 = add(x = var_5014, y = var_5028)[name = tensor<string, []>("query_45")];
            tensor<fp16, [1, 4, 256, 256]> var_5030 = mul(x = output_539_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_5030")];
            tensor<int32, [4]> x1_91_begin_0 = const()[name = tensor<string, []>("x1_91_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_91_end_0 = const()[name = tensor<string, []>("x1_91_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_91_end_mask_0 = const()[name = tensor<string, []>("x1_91_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_91 = slice_by_index(begin = x1_91_begin_0, end = x1_91_end_0, end_mask = x1_91_end_mask_0, x = output_539_cast_fp16)[name = tensor<string, []>("x1_91")];
            tensor<int32, [4]> x2_91_begin_0 = const()[name = tensor<string, []>("x2_91_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_91_end_0 = const()[name = tensor<string, []>("x2_91_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_91_end_mask_0 = const()[name = tensor<string, []>("x2_91_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_91 = slice_by_index(begin = x2_91_begin_0, end = x2_91_end_0, end_mask = x2_91_end_mask_0, x = output_539_cast_fp16)[name = tensor<string, []>("x2_91")];
            tensor<fp16, []> const_544_promoted = const()[name = tensor<string, []>("const_544_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_5041 = mul(x = x2_91, y = const_544_promoted)[name = tensor<string, []>("op_5041")];
            tensor<bool, []> var_5043_interleave_0 = const()[name = tensor<string, []>("op_5043_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_5043 = concat(axis = var_24, interleave = var_5043_interleave_0, values = (var_5041, x1_91))[name = tensor<string, []>("op_5043")];
            tensor<fp16, [1, 4, 256, 256]> var_5044 = mul(x = var_5043, y = sin_7_palettized)[name = tensor<string, []>("op_5044")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_311 = add(x = var_5030, y = var_5044)[name = tensor<string, []>("hidden_states_311")];
            tensor<int32, [1]> var_5053_axes_0 = const()[name = tensor<string, []>("op_5053_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_5053 = expand_dims(axes = var_5053_axes_0, x = hidden_states_311)[name = tensor<string, []>("op_5053")];
            tensor<int32, [5]> hidden_states_313_reps_0 = const()[name = tensor<string, []>("hidden_states_313_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_313 = tile(reps = hidden_states_313_reps_0, x = var_5053)[name = tensor<string, []>("hidden_states_313")];
            tensor<int32, [4]> var_5061 = const()[name = tensor<string, []>("op_5061"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_45 = reshape(shape = var_5061, x = hidden_states_313)[name = tensor<string, []>("key_states_45")];
            tensor<int32, [1]> var_5070_axes_0 = const()[name = tensor<string, []>("op_5070_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_315 = transpose(perm = hidden_states_315_perm_0, x = var_4982)[name = tensor<string, []>("transpose_45")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_5070 = expand_dims(axes = var_5070_axes_0, x = hidden_states_315)[name = tensor<string, []>("op_5070")];
            tensor<int32, [5]> hidden_states_317_reps_0 = const()[name = tensor<string, []>("hidden_states_317_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_317 = tile(reps = hidden_states_317_reps_0, x = var_5070)[name = tensor<string, []>("hidden_states_317")];
            tensor<int32, [4]> var_5078 = const()[name = tensor<string, []>("op_5078"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_45 = reshape(shape = var_5078, x = hidden_states_317)[name = tensor<string, []>("value_states_45")];
            tensor<bool, []> var_5081_transpose_x_1 = const()[name = tensor<string, []>("op_5081_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_5081_transpose_y_1 = const()[name = tensor<string, []>("op_5081_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_5081 = matmul(transpose_x = var_5081_transpose_x_1, transpose_y = var_5081_transpose_y_1, x = query_45, y = key_states_45)[name = tensor<string, []>("op_5081")];
            tensor<fp16, []> var_5082_to_fp16 = const()[name = tensor<string, []>("op_5082_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_89_cast_fp16 = mul(x = var_5081, y = var_5082_to_fp16)[name = tensor<string, []>("attn_weights_89_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_265_cast_fp16 = add(x = attn_weights_89_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_265_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_5090_cast_fp16 = softmax(axis = var_24, x = input_265_cast_fp16)[name = tensor<string, []>("op_5090_cast_fp16")];
            tensor<bool, []> attn_output_89_transpose_x_0 = const()[name = tensor<string, []>("attn_output_89_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_89_transpose_y_0 = const()[name = tensor<string, []>("attn_output_89_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_89 = matmul(transpose_x = attn_output_89_transpose_x_0, transpose_y = attn_output_89_transpose_y_0, x = var_5090_cast_fp16, y = value_states_45)[name = tensor<string, []>("attn_output_89")];
            tensor<int32, [4]> var_5094_perm_0 = const()[name = tensor<string, []>("op_5094_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_5096 = const()[name = tensor<string, []>("op_5096"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_5094 = transpose(perm = var_5094_perm_0, x = attn_output_89)[name = tensor<string, []>("transpose_44")];
            tensor<fp16, [1, 256, 2048]> var_5097 = reshape(shape = var_5096, x = var_5094)[name = tensor<string, []>("op_5097")];
            tensor<fp16, [1, 256, 2560]> x_539 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_22_self_attn_o_proj_weight_palettized, x = var_5097)[name = tensor<string, []>("linear_157")];
            tensor<fp16, []> var_22_promoted_135_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_135_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_5103_cast_fp16 = pow(x = x_539, y = var_22_promoted_135_to_fp16)[name = tensor<string, []>("op_5103_cast_fp16")];
            tensor<int32, [1]> var_5105_axes_0 = const()[name = tensor<string, []>("op_5105_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5105_keep_dims_0 = const()[name = tensor<string, []>("op_5105_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_5105_cast_fp16 = reduce_mean(axes = var_5105_axes_0, keep_dims = var_5105_keep_dims_0, x = var_5103_cast_fp16)[name = tensor<string, []>("op_5105_cast_fp16")];
            tensor<fp16, []> var_5106_to_fp16 = const()[name = tensor<string, []>("op_5106_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_5107_cast_fp16 = add(x = var_5105_cast_fp16, y = var_5106_to_fp16)[name = tensor<string, []>("op_5107_cast_fp16")];
            tensor<fp32, []> var_5108_epsilon_0 = const()[name = tensor<string, []>("op_5108_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_5108_cast_fp16 = rsqrt(epsilon = var_5108_epsilon_0, x = var_5107_cast_fp16)[name = tensor<string, []>("op_5108_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_541_cast_fp16 = mul(x = x_539, y = var_5108_cast_fp16)[name = tensor<string, []>("output_541_cast_fp16")];
            tensor<fp16, [2560]> var_5112_to_fp16 = const()[name = tensor<string, []>("op_5112_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940678144)))];
            tensor<fp16, [1, 256, 2560]> output_543_cast_fp16 = mul(x = output_541_cast_fp16, y = var_5112_to_fp16)[name = tensor<string, []>("output_543_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_543 = add(x = x_527, y = output_543_cast_fp16)[name = tensor<string, []>("x_543")];
            tensor<fp16, []> var_22_promoted_136_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_136_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_5118_cast_fp16 = pow(x = x_543, y = var_22_promoted_136_to_fp16)[name = tensor<string, []>("op_5118_cast_fp16")];
            tensor<int32, [1]> var_5120_axes_0 = const()[name = tensor<string, []>("op_5120_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5120_keep_dims_0 = const()[name = tensor<string, []>("op_5120_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_5120_cast_fp16 = reduce_mean(axes = var_5120_axes_0, keep_dims = var_5120_keep_dims_0, x = var_5118_cast_fp16)[name = tensor<string, []>("op_5120_cast_fp16")];
            tensor<fp16, []> var_5121_to_fp16 = const()[name = tensor<string, []>("op_5121_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_5122_cast_fp16 = add(x = var_5120_cast_fp16, y = var_5121_to_fp16)[name = tensor<string, []>("op_5122_cast_fp16")];
            tensor<fp32, []> var_5123_epsilon_0 = const()[name = tensor<string, []>("op_5123_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_5123_cast_fp16 = rsqrt(epsilon = var_5123_epsilon_0, x = var_5122_cast_fp16)[name = tensor<string, []>("op_5123_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_545_cast_fp16 = mul(x = x_543, y = var_5123_cast_fp16)[name = tensor<string, []>("output_545_cast_fp16")];
            tensor<fp16, [2560]> var_5127_to_fp16 = const()[name = tensor<string, []>("op_5127_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940683328)))];
            tensor<fp16, [1, 256, 2560]> output_547_cast_fp16 = mul(x = output_545_cast_fp16, y = var_5127_to_fp16)[name = tensor<string, []>("output_547_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_273 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_22_mlp_gate_proj_weight_palettized, x = output_547_cast_fp16)[name = tensor<string, []>("linear_158")];
            tensor<string, []> var_5135_mode_0 = const()[name = tensor<string, []>("op_5135_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_5135 = gelu(mode = var_5135_mode_0, x = input_273)[name = tensor<string, []>("op_5135")];
            tensor<fp16, [1, 256, 10240]> var_5137 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_22_mlp_up_proj_weight_palettized, x = output_547_cast_fp16)[name = tensor<string, []>("linear_159")];
            tensor<fp16, [1, 256, 10240]> input_275 = mul(x = var_5135, y = var_5137)[name = tensor<string, []>("input_275")];
            tensor<fp16, [1, 256, 2560]> x_547 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_22_mlp_down_proj_weight_palettized, x = input_275)[name = tensor<string, []>("linear_160")];
            tensor<fp16, []> var_22_promoted_137_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_137_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_5143_cast_fp16 = pow(x = x_547, y = var_22_promoted_137_to_fp16)[name = tensor<string, []>("op_5143_cast_fp16")];
            tensor<int32, [1]> var_5145_axes_0 = const()[name = tensor<string, []>("op_5145_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5145_keep_dims_0 = const()[name = tensor<string, []>("op_5145_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_5145_cast_fp16 = reduce_mean(axes = var_5145_axes_0, keep_dims = var_5145_keep_dims_0, x = var_5143_cast_fp16)[name = tensor<string, []>("op_5145_cast_fp16")];
            tensor<fp16, []> var_5146_to_fp16 = const()[name = tensor<string, []>("op_5146_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_5147_cast_fp16 = add(x = var_5145_cast_fp16, y = var_5146_to_fp16)[name = tensor<string, []>("op_5147_cast_fp16")];
            tensor<fp32, []> var_5148_epsilon_0 = const()[name = tensor<string, []>("op_5148_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_5148_cast_fp16 = rsqrt(epsilon = var_5148_epsilon_0, x = var_5147_cast_fp16)[name = tensor<string, []>("op_5148_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_549_cast_fp16 = mul(x = x_547, y = var_5148_cast_fp16)[name = tensor<string, []>("output_549_cast_fp16")];
            tensor<fp16, [2560]> var_5152_to_fp16 = const()[name = tensor<string, []>("op_5152_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940688512)))];
            tensor<fp16, [1, 256, 2560]> output_551_cast_fp16 = mul(x = output_549_cast_fp16, y = var_5152_to_fp16)[name = tensor<string, []>("output_551_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_551 = add(x = x_543, y = output_551_cast_fp16)[name = tensor<string, []>("x_551")];
            tensor<fp16, []> var_22_promoted_138_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_138_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_5164_cast_fp16 = pow(x = x_551, y = var_22_promoted_138_to_fp16)[name = tensor<string, []>("op_5164_cast_fp16")];
            tensor<int32, [1]> var_5166_axes_0 = const()[name = tensor<string, []>("op_5166_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5166_keep_dims_0 = const()[name = tensor<string, []>("op_5166_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_5166_cast_fp16 = reduce_mean(axes = var_5166_axes_0, keep_dims = var_5166_keep_dims_0, x = var_5164_cast_fp16)[name = tensor<string, []>("op_5166_cast_fp16")];
            tensor<fp16, []> var_5167_to_fp16 = const()[name = tensor<string, []>("op_5167_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_5168_cast_fp16 = add(x = var_5166_cast_fp16, y = var_5167_to_fp16)[name = tensor<string, []>("op_5168_cast_fp16")];
            tensor<fp32, []> var_5169_epsilon_0 = const()[name = tensor<string, []>("op_5169_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_5169_cast_fp16 = rsqrt(epsilon = var_5169_epsilon_0, x = var_5168_cast_fp16)[name = tensor<string, []>("op_5169_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_553_cast_fp16 = mul(x = x_551, y = var_5169_cast_fp16)[name = tensor<string, []>("output_553_cast_fp16")];
            tensor<fp16, [2560]> var_5173_to_fp16 = const()[name = tensor<string, []>("op_5173_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940693696)))];
            tensor<fp16, [1, 256, 2560]> output_555_cast_fp16 = mul(x = output_553_cast_fp16, y = var_5173_to_fp16)[name = tensor<string, []>("output_555_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_5185 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_23_self_attn_q_proj_weight_palettized, x = output_555_cast_fp16)[name = tensor<string, []>("linear_161")];
            tensor<int32, [4]> var_5186 = const()[name = tensor<string, []>("op_5186"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_5187 = reshape(shape = var_5186, x = var_5185)[name = tensor<string, []>("op_5187")];
            tensor<int32, [4]> x_555_perm_0 = const()[name = tensor<string, []>("x_555_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_5190 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_23_self_attn_k_proj_weight_palettized, x = output_555_cast_fp16)[name = tensor<string, []>("linear_162")];
            tensor<int32, [4]> var_5191 = const()[name = tensor<string, []>("op_5191"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_5192 = reshape(shape = var_5191, x = var_5190)[name = tensor<string, []>("op_5192")];
            tensor<int32, [4]> x_559_perm_0 = const()[name = tensor<string, []>("x_559_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_5195 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_23_self_attn_v_proj_weight_palettized, x = output_555_cast_fp16)[name = tensor<string, []>("linear_163")];
            tensor<int32, [4]> var_5196 = const()[name = tensor<string, []>("op_5196"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_5197 = reshape(shape = var_5196, x = var_5195)[name = tensor<string, []>("op_5197")];
            tensor<int32, [4]> hidden_states_329_perm_0 = const()[name = tensor<string, []>("hidden_states_329_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_139_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_139_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_555 = transpose(perm = x_555_perm_0, x = var_5187)[name = tensor<string, []>("transpose_43")];
            tensor<fp16, [1, 8, 256, 256]> var_5201_cast_fp16 = pow(x = x_555, y = var_22_promoted_139_to_fp16)[name = tensor<string, []>("op_5201_cast_fp16")];
            tensor<int32, [1]> var_5203_axes_0 = const()[name = tensor<string, []>("op_5203_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5203_keep_dims_0 = const()[name = tensor<string, []>("op_5203_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_5203_cast_fp16 = reduce_mean(axes = var_5203_axes_0, keep_dims = var_5203_keep_dims_0, x = var_5201_cast_fp16)[name = tensor<string, []>("op_5203_cast_fp16")];
            tensor<fp16, []> var_5204_to_fp16 = const()[name = tensor<string, []>("op_5204_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_5205_cast_fp16 = add(x = var_5203_cast_fp16, y = var_5204_to_fp16)[name = tensor<string, []>("op_5205_cast_fp16")];
            tensor<fp32, []> var_5206_epsilon_0 = const()[name = tensor<string, []>("op_5206_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_5206_cast_fp16 = rsqrt(epsilon = var_5206_epsilon_0, x = var_5205_cast_fp16)[name = tensor<string, []>("op_5206_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_557_cast_fp16 = mul(x = x_555, y = var_5206_cast_fp16)[name = tensor<string, []>("output_557_cast_fp16")];
            tensor<fp16, [256]> var_5210_to_fp16 = const()[name = tensor<string, []>("op_5210_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940698880)))];
            tensor<fp16, [1, 8, 256, 256]> output_559_cast_fp16 = mul(x = output_557_cast_fp16, y = var_5210_to_fp16)[name = tensor<string, []>("output_559_cast_fp16")];
            tensor<fp16, []> var_22_promoted_140_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_140_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_559 = transpose(perm = x_559_perm_0, x = var_5192)[name = tensor<string, []>("transpose_42")];
            tensor<fp16, [1, 4, 256, 256]> var_5215_cast_fp16 = pow(x = x_559, y = var_22_promoted_140_to_fp16)[name = tensor<string, []>("op_5215_cast_fp16")];
            tensor<int32, [1]> var_5217_axes_0 = const()[name = tensor<string, []>("op_5217_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5217_keep_dims_0 = const()[name = tensor<string, []>("op_5217_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_5217_cast_fp16 = reduce_mean(axes = var_5217_axes_0, keep_dims = var_5217_keep_dims_0, x = var_5215_cast_fp16)[name = tensor<string, []>("op_5217_cast_fp16")];
            tensor<fp16, []> var_5218_to_fp16 = const()[name = tensor<string, []>("op_5218_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_5219_cast_fp16 = add(x = var_5217_cast_fp16, y = var_5218_to_fp16)[name = tensor<string, []>("op_5219_cast_fp16")];
            tensor<fp32, []> var_5220_epsilon_0 = const()[name = tensor<string, []>("op_5220_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_5220_cast_fp16 = rsqrt(epsilon = var_5220_epsilon_0, x = var_5219_cast_fp16)[name = tensor<string, []>("op_5220_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_561_cast_fp16 = mul(x = x_559, y = var_5220_cast_fp16)[name = tensor<string, []>("output_561_cast_fp16")];
            tensor<fp16, [256]> var_5224_to_fp16 = const()[name = tensor<string, []>("op_5224_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940699456)))];
            tensor<fp16, [1, 4, 256, 256]> output_563_cast_fp16 = mul(x = output_561_cast_fp16, y = var_5224_to_fp16)[name = tensor<string, []>("output_563_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_5229 = mul(x = output_559_cast_fp16, y = cos_19_palettized)[name = tensor<string, []>("op_5229")];
            tensor<int32, [4]> x1_93_begin_0 = const()[name = tensor<string, []>("x1_93_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_93_end_0 = const()[name = tensor<string, []>("x1_93_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_93_end_mask_0 = const()[name = tensor<string, []>("x1_93_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_93 = slice_by_index(begin = x1_93_begin_0, end = x1_93_end_0, end_mask = x1_93_end_mask_0, x = output_559_cast_fp16)[name = tensor<string, []>("x1_93")];
            tensor<int32, [4]> x2_93_begin_0 = const()[name = tensor<string, []>("x2_93_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_93_end_0 = const()[name = tensor<string, []>("x2_93_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_93_end_mask_0 = const()[name = tensor<string, []>("x2_93_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_93 = slice_by_index(begin = x2_93_begin_0, end = x2_93_end_0, end_mask = x2_93_end_mask_0, x = output_559_cast_fp16)[name = tensor<string, []>("x2_93")];
            tensor<fp16, []> const_564_promoted = const()[name = tensor<string, []>("const_564_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_5240 = mul(x = x2_93, y = const_564_promoted)[name = tensor<string, []>("op_5240")];
            tensor<bool, []> var_5242_interleave_0 = const()[name = tensor<string, []>("op_5242_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_5242 = concat(axis = var_24, interleave = var_5242_interleave_0, values = (var_5240, x1_93))[name = tensor<string, []>("op_5242")];
            tensor<fp16, [1, 8, 256, 256]> var_5243 = mul(x = var_5242, y = sin_19_palettized)[name = tensor<string, []>("op_5243")];
            tensor<fp16, [1, 8, 256, 256]> query_47 = add(x = var_5229, y = var_5243)[name = tensor<string, []>("query_47")];
            tensor<fp16, [1, 4, 256, 256]> var_5245 = mul(x = output_563_cast_fp16, y = cos_19_palettized)[name = tensor<string, []>("op_5245")];
            tensor<int32, [4]> x1_95_begin_0 = const()[name = tensor<string, []>("x1_95_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_95_end_0 = const()[name = tensor<string, []>("x1_95_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_95_end_mask_0 = const()[name = tensor<string, []>("x1_95_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_95 = slice_by_index(begin = x1_95_begin_0, end = x1_95_end_0, end_mask = x1_95_end_mask_0, x = output_563_cast_fp16)[name = tensor<string, []>("x1_95")];
            tensor<int32, [4]> x2_95_begin_0 = const()[name = tensor<string, []>("x2_95_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_95_end_0 = const()[name = tensor<string, []>("x2_95_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_95_end_mask_0 = const()[name = tensor<string, []>("x2_95_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_95 = slice_by_index(begin = x2_95_begin_0, end = x2_95_end_0, end_mask = x2_95_end_mask_0, x = output_563_cast_fp16)[name = tensor<string, []>("x2_95")];
            tensor<fp16, []> const_567_promoted = const()[name = tensor<string, []>("const_567_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_5256 = mul(x = x2_95, y = const_567_promoted)[name = tensor<string, []>("op_5256")];
            tensor<bool, []> var_5258_interleave_0 = const()[name = tensor<string, []>("op_5258_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_5258 = concat(axis = var_24, interleave = var_5258_interleave_0, values = (var_5256, x1_95))[name = tensor<string, []>("op_5258")];
            tensor<fp16, [1, 4, 256, 256]> var_5259 = mul(x = var_5258, y = sin_19_palettized)[name = tensor<string, []>("op_5259")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_325 = add(x = var_5245, y = var_5259)[name = tensor<string, []>("hidden_states_325")];
            tensor<int32, [1]> var_5268_axes_0 = const()[name = tensor<string, []>("op_5268_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_5268 = expand_dims(axes = var_5268_axes_0, x = hidden_states_325)[name = tensor<string, []>("op_5268")];
            tensor<int32, [5]> hidden_states_327_reps_0 = const()[name = tensor<string, []>("hidden_states_327_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_327 = tile(reps = hidden_states_327_reps_0, x = var_5268)[name = tensor<string, []>("hidden_states_327")];
            tensor<int32, [4]> var_5276 = const()[name = tensor<string, []>("op_5276"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_47 = reshape(shape = var_5276, x = hidden_states_327)[name = tensor<string, []>("key_states_47")];
            tensor<int32, [1]> var_5285_axes_0 = const()[name = tensor<string, []>("op_5285_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_329 = transpose(perm = hidden_states_329_perm_0, x = var_5197)[name = tensor<string, []>("transpose_41")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_5285 = expand_dims(axes = var_5285_axes_0, x = hidden_states_329)[name = tensor<string, []>("op_5285")];
            tensor<int32, [5]> hidden_states_331_reps_0 = const()[name = tensor<string, []>("hidden_states_331_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_331 = tile(reps = hidden_states_331_reps_0, x = var_5285)[name = tensor<string, []>("hidden_states_331")];
            tensor<int32, [4]> var_5293 = const()[name = tensor<string, []>("op_5293"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_47 = reshape(shape = var_5293, x = hidden_states_331)[name = tensor<string, []>("value_states_47")];
            tensor<bool, []> var_5296_transpose_x_1 = const()[name = tensor<string, []>("op_5296_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_5296_transpose_y_1 = const()[name = tensor<string, []>("op_5296_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_5296 = matmul(transpose_x = var_5296_transpose_x_1, transpose_y = var_5296_transpose_y_1, x = query_47, y = key_states_47)[name = tensor<string, []>("op_5296")];
            tensor<fp16, []> var_5297_to_fp16 = const()[name = tensor<string, []>("op_5297_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_93_cast_fp16 = mul(x = var_5296, y = var_5297_to_fp16)[name = tensor<string, []>("attn_weights_93_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_277_cast_fp16 = add(x = attn_weights_93_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_277_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_5305_cast_fp16 = softmax(axis = var_24, x = input_277_cast_fp16)[name = tensor<string, []>("op_5305_cast_fp16")];
            tensor<bool, []> attn_output_93_transpose_x_0 = const()[name = tensor<string, []>("attn_output_93_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_93_transpose_y_0 = const()[name = tensor<string, []>("attn_output_93_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_93 = matmul(transpose_x = attn_output_93_transpose_x_0, transpose_y = attn_output_93_transpose_y_0, x = var_5305_cast_fp16, y = value_states_47)[name = tensor<string, []>("attn_output_93")];
            tensor<int32, [4]> var_5309_perm_0 = const()[name = tensor<string, []>("op_5309_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_5311 = const()[name = tensor<string, []>("op_5311"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_5309 = transpose(perm = var_5309_perm_0, x = attn_output_93)[name = tensor<string, []>("transpose_40")];
            tensor<fp16, [1, 256, 2048]> var_5312 = reshape(shape = var_5311, x = var_5309)[name = tensor<string, []>("op_5312")];
            tensor<fp16, [1, 256, 2560]> x_563 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_23_self_attn_o_proj_weight_palettized, x = var_5312)[name = tensor<string, []>("linear_164")];
            tensor<fp16, []> var_22_promoted_141_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_141_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_5318_cast_fp16 = pow(x = x_563, y = var_22_promoted_141_to_fp16)[name = tensor<string, []>("op_5318_cast_fp16")];
            tensor<int32, [1]> var_5320_axes_0 = const()[name = tensor<string, []>("op_5320_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5320_keep_dims_0 = const()[name = tensor<string, []>("op_5320_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_5320_cast_fp16 = reduce_mean(axes = var_5320_axes_0, keep_dims = var_5320_keep_dims_0, x = var_5318_cast_fp16)[name = tensor<string, []>("op_5320_cast_fp16")];
            tensor<fp16, []> var_5321_to_fp16 = const()[name = tensor<string, []>("op_5321_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_5322_cast_fp16 = add(x = var_5320_cast_fp16, y = var_5321_to_fp16)[name = tensor<string, []>("op_5322_cast_fp16")];
            tensor<fp32, []> var_5323_epsilon_0 = const()[name = tensor<string, []>("op_5323_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_5323_cast_fp16 = rsqrt(epsilon = var_5323_epsilon_0, x = var_5322_cast_fp16)[name = tensor<string, []>("op_5323_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_565_cast_fp16 = mul(x = x_563, y = var_5323_cast_fp16)[name = tensor<string, []>("output_565_cast_fp16")];
            tensor<fp16, [2560]> var_5327_to_fp16 = const()[name = tensor<string, []>("op_5327_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940700032)))];
            tensor<fp16, [1, 256, 2560]> output_567_cast_fp16 = mul(x = output_565_cast_fp16, y = var_5327_to_fp16)[name = tensor<string, []>("output_567_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_567 = add(x = x_551, y = output_567_cast_fp16)[name = tensor<string, []>("x_567")];
            tensor<fp16, []> var_22_promoted_142_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_142_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_5333_cast_fp16 = pow(x = x_567, y = var_22_promoted_142_to_fp16)[name = tensor<string, []>("op_5333_cast_fp16")];
            tensor<int32, [1]> var_5335_axes_0 = const()[name = tensor<string, []>("op_5335_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5335_keep_dims_0 = const()[name = tensor<string, []>("op_5335_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_5335_cast_fp16 = reduce_mean(axes = var_5335_axes_0, keep_dims = var_5335_keep_dims_0, x = var_5333_cast_fp16)[name = tensor<string, []>("op_5335_cast_fp16")];
            tensor<fp16, []> var_5336_to_fp16 = const()[name = tensor<string, []>("op_5336_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_5337_cast_fp16 = add(x = var_5335_cast_fp16, y = var_5336_to_fp16)[name = tensor<string, []>("op_5337_cast_fp16")];
            tensor<fp32, []> var_5338_epsilon_0 = const()[name = tensor<string, []>("op_5338_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_5338_cast_fp16 = rsqrt(epsilon = var_5338_epsilon_0, x = var_5337_cast_fp16)[name = tensor<string, []>("op_5338_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_569_cast_fp16 = mul(x = x_567, y = var_5338_cast_fp16)[name = tensor<string, []>("output_569_cast_fp16")];
            tensor<fp16, [2560]> var_5342_to_fp16 = const()[name = tensor<string, []>("op_5342_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940705216)))];
            tensor<fp16, [1, 256, 2560]> output_571_cast_fp16 = mul(x = output_569_cast_fp16, y = var_5342_to_fp16)[name = tensor<string, []>("output_571_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_285 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_23_mlp_gate_proj_weight_palettized, x = output_571_cast_fp16)[name = tensor<string, []>("linear_165")];
            tensor<string, []> var_5350_mode_0 = const()[name = tensor<string, []>("op_5350_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_5350 = gelu(mode = var_5350_mode_0, x = input_285)[name = tensor<string, []>("op_5350")];
            tensor<fp16, [1, 256, 10240]> var_5352 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_23_mlp_up_proj_weight_palettized, x = output_571_cast_fp16)[name = tensor<string, []>("linear_166")];
            tensor<fp16, [1, 256, 10240]> input_287 = mul(x = var_5350, y = var_5352)[name = tensor<string, []>("input_287")];
            tensor<fp16, [1, 256, 2560]> x_571 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_23_mlp_down_proj_weight_palettized, x = input_287)[name = tensor<string, []>("linear_167")];
            tensor<fp16, []> var_22_promoted_143_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_143_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_5358_cast_fp16 = pow(x = x_571, y = var_22_promoted_143_to_fp16)[name = tensor<string, []>("op_5358_cast_fp16")];
            tensor<int32, [1]> var_5360_axes_0 = const()[name = tensor<string, []>("op_5360_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5360_keep_dims_0 = const()[name = tensor<string, []>("op_5360_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_5360_cast_fp16 = reduce_mean(axes = var_5360_axes_0, keep_dims = var_5360_keep_dims_0, x = var_5358_cast_fp16)[name = tensor<string, []>("op_5360_cast_fp16")];
            tensor<fp16, []> var_5361_to_fp16 = const()[name = tensor<string, []>("op_5361_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_5362_cast_fp16 = add(x = var_5360_cast_fp16, y = var_5361_to_fp16)[name = tensor<string, []>("op_5362_cast_fp16")];
            tensor<fp32, []> var_5363_epsilon_0 = const()[name = tensor<string, []>("op_5363_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_5363_cast_fp16 = rsqrt(epsilon = var_5363_epsilon_0, x = var_5362_cast_fp16)[name = tensor<string, []>("op_5363_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_573_cast_fp16 = mul(x = x_571, y = var_5363_cast_fp16)[name = tensor<string, []>("output_573_cast_fp16")];
            tensor<fp16, [2560]> var_5367_to_fp16 = const()[name = tensor<string, []>("op_5367_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940710400)))];
            tensor<fp16, [1, 256, 2560]> output_575_cast_fp16 = mul(x = output_573_cast_fp16, y = var_5367_to_fp16)[name = tensor<string, []>("output_575_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_575 = add(x = x_567, y = output_575_cast_fp16)[name = tensor<string, []>("x_575")];
            tensor<fp16, []> var_22_promoted_144_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_144_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_5379_cast_fp16 = pow(x = x_575, y = var_22_promoted_144_to_fp16)[name = tensor<string, []>("op_5379_cast_fp16")];
            tensor<int32, [1]> var_5381_axes_0 = const()[name = tensor<string, []>("op_5381_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5381_keep_dims_0 = const()[name = tensor<string, []>("op_5381_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_5381_cast_fp16 = reduce_mean(axes = var_5381_axes_0, keep_dims = var_5381_keep_dims_0, x = var_5379_cast_fp16)[name = tensor<string, []>("op_5381_cast_fp16")];
            tensor<fp16, []> var_5382_to_fp16 = const()[name = tensor<string, []>("op_5382_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_5383_cast_fp16 = add(x = var_5381_cast_fp16, y = var_5382_to_fp16)[name = tensor<string, []>("op_5383_cast_fp16")];
            tensor<fp32, []> var_5384_epsilon_0 = const()[name = tensor<string, []>("op_5384_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_5384_cast_fp16 = rsqrt(epsilon = var_5384_epsilon_0, x = var_5383_cast_fp16)[name = tensor<string, []>("op_5384_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_577_cast_fp16 = mul(x = x_575, y = var_5384_cast_fp16)[name = tensor<string, []>("output_577_cast_fp16")];
            tensor<fp16, [2560]> var_5388_to_fp16 = const()[name = tensor<string, []>("op_5388_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940715584)))];
            tensor<fp16, [1, 256, 2560]> output_579_cast_fp16 = mul(x = output_577_cast_fp16, y = var_5388_to_fp16)[name = tensor<string, []>("output_579_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_5400 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_24_self_attn_q_proj_weight_palettized, x = output_579_cast_fp16)[name = tensor<string, []>("linear_168")];
            tensor<int32, [4]> var_5401 = const()[name = tensor<string, []>("op_5401"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_5402 = reshape(shape = var_5401, x = var_5400)[name = tensor<string, []>("op_5402")];
            tensor<int32, [4]> x_579_perm_0 = const()[name = tensor<string, []>("x_579_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_5405 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_24_self_attn_k_proj_weight_palettized, x = output_579_cast_fp16)[name = tensor<string, []>("linear_169")];
            tensor<int32, [4]> var_5406 = const()[name = tensor<string, []>("op_5406"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_5407 = reshape(shape = var_5406, x = var_5405)[name = tensor<string, []>("op_5407")];
            tensor<int32, [4]> x_583_perm_0 = const()[name = tensor<string, []>("x_583_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_5410 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_24_self_attn_v_proj_weight_palettized, x = output_579_cast_fp16)[name = tensor<string, []>("linear_170")];
            tensor<int32, [4]> var_5411 = const()[name = tensor<string, []>("op_5411"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_5412 = reshape(shape = var_5411, x = var_5410)[name = tensor<string, []>("op_5412")];
            tensor<int32, [4]> hidden_states_343_perm_0 = const()[name = tensor<string, []>("hidden_states_343_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_145_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_145_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_579 = transpose(perm = x_579_perm_0, x = var_5402)[name = tensor<string, []>("transpose_39")];
            tensor<fp16, [1, 8, 256, 256]> var_5416_cast_fp16 = pow(x = x_579, y = var_22_promoted_145_to_fp16)[name = tensor<string, []>("op_5416_cast_fp16")];
            tensor<int32, [1]> var_5418_axes_0 = const()[name = tensor<string, []>("op_5418_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5418_keep_dims_0 = const()[name = tensor<string, []>("op_5418_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_5418_cast_fp16 = reduce_mean(axes = var_5418_axes_0, keep_dims = var_5418_keep_dims_0, x = var_5416_cast_fp16)[name = tensor<string, []>("op_5418_cast_fp16")];
            tensor<fp16, []> var_5419_to_fp16 = const()[name = tensor<string, []>("op_5419_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_5420_cast_fp16 = add(x = var_5418_cast_fp16, y = var_5419_to_fp16)[name = tensor<string, []>("op_5420_cast_fp16")];
            tensor<fp32, []> var_5421_epsilon_0 = const()[name = tensor<string, []>("op_5421_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_5421_cast_fp16 = rsqrt(epsilon = var_5421_epsilon_0, x = var_5420_cast_fp16)[name = tensor<string, []>("op_5421_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_581_cast_fp16 = mul(x = x_579, y = var_5421_cast_fp16)[name = tensor<string, []>("output_581_cast_fp16")];
            tensor<fp16, [256]> var_5425_to_fp16 = const()[name = tensor<string, []>("op_5425_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940720768)))];
            tensor<fp16, [1, 8, 256, 256]> output_583_cast_fp16 = mul(x = output_581_cast_fp16, y = var_5425_to_fp16)[name = tensor<string, []>("output_583_cast_fp16")];
            tensor<fp16, []> var_22_promoted_146_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_146_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_583 = transpose(perm = x_583_perm_0, x = var_5407)[name = tensor<string, []>("transpose_38")];
            tensor<fp16, [1, 4, 256, 256]> var_5430_cast_fp16 = pow(x = x_583, y = var_22_promoted_146_to_fp16)[name = tensor<string, []>("op_5430_cast_fp16")];
            tensor<int32, [1]> var_5432_axes_0 = const()[name = tensor<string, []>("op_5432_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5432_keep_dims_0 = const()[name = tensor<string, []>("op_5432_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_5432_cast_fp16 = reduce_mean(axes = var_5432_axes_0, keep_dims = var_5432_keep_dims_0, x = var_5430_cast_fp16)[name = tensor<string, []>("op_5432_cast_fp16")];
            tensor<fp16, []> var_5433_to_fp16 = const()[name = tensor<string, []>("op_5433_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_5434_cast_fp16 = add(x = var_5432_cast_fp16, y = var_5433_to_fp16)[name = tensor<string, []>("op_5434_cast_fp16")];
            tensor<fp32, []> var_5435_epsilon_0 = const()[name = tensor<string, []>("op_5435_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_5435_cast_fp16 = rsqrt(epsilon = var_5435_epsilon_0, x = var_5434_cast_fp16)[name = tensor<string, []>("op_5435_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_585_cast_fp16 = mul(x = x_583, y = var_5435_cast_fp16)[name = tensor<string, []>("output_585_cast_fp16")];
            tensor<fp16, [256]> var_5439_to_fp16 = const()[name = tensor<string, []>("op_5439_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940721344)))];
            tensor<fp16, [1, 4, 256, 256]> output_587_cast_fp16 = mul(x = output_585_cast_fp16, y = var_5439_to_fp16)[name = tensor<string, []>("output_587_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_5444 = mul(x = output_583_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_5444")];
            tensor<int32, [4]> x1_97_begin_0 = const()[name = tensor<string, []>("x1_97_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_97_end_0 = const()[name = tensor<string, []>("x1_97_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_97_end_mask_0 = const()[name = tensor<string, []>("x1_97_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_97 = slice_by_index(begin = x1_97_begin_0, end = x1_97_end_0, end_mask = x1_97_end_mask_0, x = output_583_cast_fp16)[name = tensor<string, []>("x1_97")];
            tensor<int32, [4]> x2_97_begin_0 = const()[name = tensor<string, []>("x2_97_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_97_end_0 = const()[name = tensor<string, []>("x2_97_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_97_end_mask_0 = const()[name = tensor<string, []>("x2_97_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_97 = slice_by_index(begin = x2_97_begin_0, end = x2_97_end_0, end_mask = x2_97_end_mask_0, x = output_583_cast_fp16)[name = tensor<string, []>("x2_97")];
            tensor<fp16, []> const_587_promoted = const()[name = tensor<string, []>("const_587_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_5455 = mul(x = x2_97, y = const_587_promoted)[name = tensor<string, []>("op_5455")];
            tensor<bool, []> var_5457_interleave_0 = const()[name = tensor<string, []>("op_5457_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_5457 = concat(axis = var_24, interleave = var_5457_interleave_0, values = (var_5455, x1_97))[name = tensor<string, []>("op_5457")];
            tensor<fp16, [1, 8, 256, 256]> var_5458 = mul(x = var_5457, y = sin_7_palettized)[name = tensor<string, []>("op_5458")];
            tensor<fp16, [1, 8, 256, 256]> query_49 = add(x = var_5444, y = var_5458)[name = tensor<string, []>("query_49")];
            tensor<fp16, [1, 4, 256, 256]> var_5460 = mul(x = output_587_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_5460")];
            tensor<int32, [4]> x1_99_begin_0 = const()[name = tensor<string, []>("x1_99_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_99_end_0 = const()[name = tensor<string, []>("x1_99_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_99_end_mask_0 = const()[name = tensor<string, []>("x1_99_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_99 = slice_by_index(begin = x1_99_begin_0, end = x1_99_end_0, end_mask = x1_99_end_mask_0, x = output_587_cast_fp16)[name = tensor<string, []>("x1_99")];
            tensor<int32, [4]> x2_99_begin_0 = const()[name = tensor<string, []>("x2_99_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_99_end_0 = const()[name = tensor<string, []>("x2_99_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_99_end_mask_0 = const()[name = tensor<string, []>("x2_99_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_99 = slice_by_index(begin = x2_99_begin_0, end = x2_99_end_0, end_mask = x2_99_end_mask_0, x = output_587_cast_fp16)[name = tensor<string, []>("x2_99")];
            tensor<fp16, []> const_590_promoted = const()[name = tensor<string, []>("const_590_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_5471 = mul(x = x2_99, y = const_590_promoted)[name = tensor<string, []>("op_5471")];
            tensor<bool, []> var_5473_interleave_0 = const()[name = tensor<string, []>("op_5473_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_5473 = concat(axis = var_24, interleave = var_5473_interleave_0, values = (var_5471, x1_99))[name = tensor<string, []>("op_5473")];
            tensor<fp16, [1, 4, 256, 256]> var_5474 = mul(x = var_5473, y = sin_7_palettized)[name = tensor<string, []>("op_5474")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_339 = add(x = var_5460, y = var_5474)[name = tensor<string, []>("hidden_states_339")];
            tensor<int32, [1]> var_5483_axes_0 = const()[name = tensor<string, []>("op_5483_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_5483 = expand_dims(axes = var_5483_axes_0, x = hidden_states_339)[name = tensor<string, []>("op_5483")];
            tensor<int32, [5]> hidden_states_341_reps_0 = const()[name = tensor<string, []>("hidden_states_341_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_341 = tile(reps = hidden_states_341_reps_0, x = var_5483)[name = tensor<string, []>("hidden_states_341")];
            tensor<int32, [4]> var_5491 = const()[name = tensor<string, []>("op_5491"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_49 = reshape(shape = var_5491, x = hidden_states_341)[name = tensor<string, []>("key_states_49")];
            tensor<int32, [1]> var_5500_axes_0 = const()[name = tensor<string, []>("op_5500_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_343 = transpose(perm = hidden_states_343_perm_0, x = var_5412)[name = tensor<string, []>("transpose_37")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_5500 = expand_dims(axes = var_5500_axes_0, x = hidden_states_343)[name = tensor<string, []>("op_5500")];
            tensor<int32, [5]> hidden_states_345_reps_0 = const()[name = tensor<string, []>("hidden_states_345_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_345 = tile(reps = hidden_states_345_reps_0, x = var_5500)[name = tensor<string, []>("hidden_states_345")];
            tensor<int32, [4]> var_5508 = const()[name = tensor<string, []>("op_5508"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_49 = reshape(shape = var_5508, x = hidden_states_345)[name = tensor<string, []>("value_states_49")];
            tensor<bool, []> var_5511_transpose_x_1 = const()[name = tensor<string, []>("op_5511_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_5511_transpose_y_1 = const()[name = tensor<string, []>("op_5511_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_5511 = matmul(transpose_x = var_5511_transpose_x_1, transpose_y = var_5511_transpose_y_1, x = query_49, y = key_states_49)[name = tensor<string, []>("op_5511")];
            tensor<fp16, []> var_5512_to_fp16 = const()[name = tensor<string, []>("op_5512_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_97_cast_fp16 = mul(x = var_5511, y = var_5512_to_fp16)[name = tensor<string, []>("attn_weights_97_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_289_cast_fp16 = add(x = attn_weights_97_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_289_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_5520_cast_fp16 = softmax(axis = var_24, x = input_289_cast_fp16)[name = tensor<string, []>("op_5520_cast_fp16")];
            tensor<bool, []> attn_output_97_transpose_x_0 = const()[name = tensor<string, []>("attn_output_97_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_97_transpose_y_0 = const()[name = tensor<string, []>("attn_output_97_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_97 = matmul(transpose_x = attn_output_97_transpose_x_0, transpose_y = attn_output_97_transpose_y_0, x = var_5520_cast_fp16, y = value_states_49)[name = tensor<string, []>("attn_output_97")];
            tensor<int32, [4]> var_5524_perm_0 = const()[name = tensor<string, []>("op_5524_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_5526 = const()[name = tensor<string, []>("op_5526"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_5524 = transpose(perm = var_5524_perm_0, x = attn_output_97)[name = tensor<string, []>("transpose_36")];
            tensor<fp16, [1, 256, 2048]> var_5527 = reshape(shape = var_5526, x = var_5524)[name = tensor<string, []>("op_5527")];
            tensor<fp16, [1, 256, 2560]> x_587 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_24_self_attn_o_proj_weight_palettized, x = var_5527)[name = tensor<string, []>("linear_171")];
            tensor<fp16, []> var_22_promoted_147_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_147_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_5533_cast_fp16 = pow(x = x_587, y = var_22_promoted_147_to_fp16)[name = tensor<string, []>("op_5533_cast_fp16")];
            tensor<int32, [1]> var_5535_axes_0 = const()[name = tensor<string, []>("op_5535_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5535_keep_dims_0 = const()[name = tensor<string, []>("op_5535_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_5535_cast_fp16 = reduce_mean(axes = var_5535_axes_0, keep_dims = var_5535_keep_dims_0, x = var_5533_cast_fp16)[name = tensor<string, []>("op_5535_cast_fp16")];
            tensor<fp16, []> var_5536_to_fp16 = const()[name = tensor<string, []>("op_5536_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_5537_cast_fp16 = add(x = var_5535_cast_fp16, y = var_5536_to_fp16)[name = tensor<string, []>("op_5537_cast_fp16")];
            tensor<fp32, []> var_5538_epsilon_0 = const()[name = tensor<string, []>("op_5538_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_5538_cast_fp16 = rsqrt(epsilon = var_5538_epsilon_0, x = var_5537_cast_fp16)[name = tensor<string, []>("op_5538_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_589_cast_fp16 = mul(x = x_587, y = var_5538_cast_fp16)[name = tensor<string, []>("output_589_cast_fp16")];
            tensor<fp16, [2560]> var_5542_to_fp16 = const()[name = tensor<string, []>("op_5542_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940721920)))];
            tensor<fp16, [1, 256, 2560]> output_591_cast_fp16 = mul(x = output_589_cast_fp16, y = var_5542_to_fp16)[name = tensor<string, []>("output_591_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_591 = add(x = x_575, y = output_591_cast_fp16)[name = tensor<string, []>("x_591")];
            tensor<fp16, []> var_22_promoted_148_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_148_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_5548_cast_fp16 = pow(x = x_591, y = var_22_promoted_148_to_fp16)[name = tensor<string, []>("op_5548_cast_fp16")];
            tensor<int32, [1]> var_5550_axes_0 = const()[name = tensor<string, []>("op_5550_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5550_keep_dims_0 = const()[name = tensor<string, []>("op_5550_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_5550_cast_fp16 = reduce_mean(axes = var_5550_axes_0, keep_dims = var_5550_keep_dims_0, x = var_5548_cast_fp16)[name = tensor<string, []>("op_5550_cast_fp16")];
            tensor<fp16, []> var_5551_to_fp16 = const()[name = tensor<string, []>("op_5551_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_5552_cast_fp16 = add(x = var_5550_cast_fp16, y = var_5551_to_fp16)[name = tensor<string, []>("op_5552_cast_fp16")];
            tensor<fp32, []> var_5553_epsilon_0 = const()[name = tensor<string, []>("op_5553_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_5553_cast_fp16 = rsqrt(epsilon = var_5553_epsilon_0, x = var_5552_cast_fp16)[name = tensor<string, []>("op_5553_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_593_cast_fp16 = mul(x = x_591, y = var_5553_cast_fp16)[name = tensor<string, []>("output_593_cast_fp16")];
            tensor<fp16, [2560]> var_5557_to_fp16 = const()[name = tensor<string, []>("op_5557_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940727104)))];
            tensor<fp16, [1, 256, 2560]> output_595_cast_fp16 = mul(x = output_593_cast_fp16, y = var_5557_to_fp16)[name = tensor<string, []>("output_595_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_297 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_24_mlp_gate_proj_weight_palettized, x = output_595_cast_fp16)[name = tensor<string, []>("linear_172")];
            tensor<string, []> var_5565_mode_0 = const()[name = tensor<string, []>("op_5565_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_5565 = gelu(mode = var_5565_mode_0, x = input_297)[name = tensor<string, []>("op_5565")];
            tensor<fp16, [1, 256, 10240]> var_5567 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_24_mlp_up_proj_weight_palettized, x = output_595_cast_fp16)[name = tensor<string, []>("linear_173")];
            tensor<fp16, [1, 256, 10240]> input_299 = mul(x = var_5565, y = var_5567)[name = tensor<string, []>("input_299")];
            tensor<fp16, [1, 256, 2560]> x_595 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_24_mlp_down_proj_weight_palettized, x = input_299)[name = tensor<string, []>("linear_174")];
            tensor<fp16, []> var_22_promoted_149_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_149_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_5573_cast_fp16 = pow(x = x_595, y = var_22_promoted_149_to_fp16)[name = tensor<string, []>("op_5573_cast_fp16")];
            tensor<int32, [1]> var_5575_axes_0 = const()[name = tensor<string, []>("op_5575_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5575_keep_dims_0 = const()[name = tensor<string, []>("op_5575_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_5575_cast_fp16 = reduce_mean(axes = var_5575_axes_0, keep_dims = var_5575_keep_dims_0, x = var_5573_cast_fp16)[name = tensor<string, []>("op_5575_cast_fp16")];
            tensor<fp16, []> var_5576_to_fp16 = const()[name = tensor<string, []>("op_5576_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_5577_cast_fp16 = add(x = var_5575_cast_fp16, y = var_5576_to_fp16)[name = tensor<string, []>("op_5577_cast_fp16")];
            tensor<fp32, []> var_5578_epsilon_0 = const()[name = tensor<string, []>("op_5578_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_5578_cast_fp16 = rsqrt(epsilon = var_5578_epsilon_0, x = var_5577_cast_fp16)[name = tensor<string, []>("op_5578_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_597_cast_fp16 = mul(x = x_595, y = var_5578_cast_fp16)[name = tensor<string, []>("output_597_cast_fp16")];
            tensor<fp16, [2560]> var_5582_to_fp16 = const()[name = tensor<string, []>("op_5582_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940732288)))];
            tensor<fp16, [1, 256, 2560]> output_599_cast_fp16 = mul(x = output_597_cast_fp16, y = var_5582_to_fp16)[name = tensor<string, []>("output_599_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_599 = add(x = x_591, y = output_599_cast_fp16)[name = tensor<string, []>("x_599")];
            tensor<fp16, []> var_22_promoted_150_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_150_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_5594_cast_fp16 = pow(x = x_599, y = var_22_promoted_150_to_fp16)[name = tensor<string, []>("op_5594_cast_fp16")];
            tensor<int32, [1]> var_5596_axes_0 = const()[name = tensor<string, []>("op_5596_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5596_keep_dims_0 = const()[name = tensor<string, []>("op_5596_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_5596_cast_fp16 = reduce_mean(axes = var_5596_axes_0, keep_dims = var_5596_keep_dims_0, x = var_5594_cast_fp16)[name = tensor<string, []>("op_5596_cast_fp16")];
            tensor<fp16, []> var_5597_to_fp16 = const()[name = tensor<string, []>("op_5597_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_5598_cast_fp16 = add(x = var_5596_cast_fp16, y = var_5597_to_fp16)[name = tensor<string, []>("op_5598_cast_fp16")];
            tensor<fp32, []> var_5599_epsilon_0 = const()[name = tensor<string, []>("op_5599_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_5599_cast_fp16 = rsqrt(epsilon = var_5599_epsilon_0, x = var_5598_cast_fp16)[name = tensor<string, []>("op_5599_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_601_cast_fp16 = mul(x = x_599, y = var_5599_cast_fp16)[name = tensor<string, []>("output_601_cast_fp16")];
            tensor<fp16, [2560]> var_5603_to_fp16 = const()[name = tensor<string, []>("op_5603_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940737472)))];
            tensor<fp16, [1, 256, 2560]> output_603_cast_fp16 = mul(x = output_601_cast_fp16, y = var_5603_to_fp16)[name = tensor<string, []>("output_603_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_5615 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_25_self_attn_q_proj_weight_palettized, x = output_603_cast_fp16)[name = tensor<string, []>("linear_175")];
            tensor<int32, [4]> var_5616 = const()[name = tensor<string, []>("op_5616"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_5617 = reshape(shape = var_5616, x = var_5615)[name = tensor<string, []>("op_5617")];
            tensor<int32, [4]> x_603_perm_0 = const()[name = tensor<string, []>("x_603_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_5620 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_25_self_attn_k_proj_weight_palettized, x = output_603_cast_fp16)[name = tensor<string, []>("linear_176")];
            tensor<int32, [4]> var_5621 = const()[name = tensor<string, []>("op_5621"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_5622 = reshape(shape = var_5621, x = var_5620)[name = tensor<string, []>("op_5622")];
            tensor<int32, [4]> x_607_perm_0 = const()[name = tensor<string, []>("x_607_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_5625 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_25_self_attn_v_proj_weight_palettized, x = output_603_cast_fp16)[name = tensor<string, []>("linear_177")];
            tensor<int32, [4]> var_5626 = const()[name = tensor<string, []>("op_5626"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_5627 = reshape(shape = var_5626, x = var_5625)[name = tensor<string, []>("op_5627")];
            tensor<int32, [4]> hidden_states_357_perm_0 = const()[name = tensor<string, []>("hidden_states_357_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_151_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_151_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_603 = transpose(perm = x_603_perm_0, x = var_5617)[name = tensor<string, []>("transpose_35")];
            tensor<fp16, [1, 8, 256, 256]> var_5631_cast_fp16 = pow(x = x_603, y = var_22_promoted_151_to_fp16)[name = tensor<string, []>("op_5631_cast_fp16")];
            tensor<int32, [1]> var_5633_axes_0 = const()[name = tensor<string, []>("op_5633_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5633_keep_dims_0 = const()[name = tensor<string, []>("op_5633_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_5633_cast_fp16 = reduce_mean(axes = var_5633_axes_0, keep_dims = var_5633_keep_dims_0, x = var_5631_cast_fp16)[name = tensor<string, []>("op_5633_cast_fp16")];
            tensor<fp16, []> var_5634_to_fp16 = const()[name = tensor<string, []>("op_5634_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_5635_cast_fp16 = add(x = var_5633_cast_fp16, y = var_5634_to_fp16)[name = tensor<string, []>("op_5635_cast_fp16")];
            tensor<fp32, []> var_5636_epsilon_0 = const()[name = tensor<string, []>("op_5636_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_5636_cast_fp16 = rsqrt(epsilon = var_5636_epsilon_0, x = var_5635_cast_fp16)[name = tensor<string, []>("op_5636_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_605_cast_fp16 = mul(x = x_603, y = var_5636_cast_fp16)[name = tensor<string, []>("output_605_cast_fp16")];
            tensor<fp16, [256]> var_5640_to_fp16 = const()[name = tensor<string, []>("op_5640_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940742656)))];
            tensor<fp16, [1, 8, 256, 256]> output_607_cast_fp16 = mul(x = output_605_cast_fp16, y = var_5640_to_fp16)[name = tensor<string, []>("output_607_cast_fp16")];
            tensor<fp16, []> var_22_promoted_152_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_152_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_607 = transpose(perm = x_607_perm_0, x = var_5622)[name = tensor<string, []>("transpose_34")];
            tensor<fp16, [1, 4, 256, 256]> var_5645_cast_fp16 = pow(x = x_607, y = var_22_promoted_152_to_fp16)[name = tensor<string, []>("op_5645_cast_fp16")];
            tensor<int32, [1]> var_5647_axes_0 = const()[name = tensor<string, []>("op_5647_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5647_keep_dims_0 = const()[name = tensor<string, []>("op_5647_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_5647_cast_fp16 = reduce_mean(axes = var_5647_axes_0, keep_dims = var_5647_keep_dims_0, x = var_5645_cast_fp16)[name = tensor<string, []>("op_5647_cast_fp16")];
            tensor<fp16, []> var_5648_to_fp16 = const()[name = tensor<string, []>("op_5648_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_5649_cast_fp16 = add(x = var_5647_cast_fp16, y = var_5648_to_fp16)[name = tensor<string, []>("op_5649_cast_fp16")];
            tensor<fp32, []> var_5650_epsilon_0 = const()[name = tensor<string, []>("op_5650_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_5650_cast_fp16 = rsqrt(epsilon = var_5650_epsilon_0, x = var_5649_cast_fp16)[name = tensor<string, []>("op_5650_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_609_cast_fp16 = mul(x = x_607, y = var_5650_cast_fp16)[name = tensor<string, []>("output_609_cast_fp16")];
            tensor<fp16, [256]> var_5654_to_fp16 = const()[name = tensor<string, []>("op_5654_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940743232)))];
            tensor<fp16, [1, 4, 256, 256]> output_611_cast_fp16 = mul(x = output_609_cast_fp16, y = var_5654_to_fp16)[name = tensor<string, []>("output_611_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_5659 = mul(x = output_607_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_5659")];
            tensor<int32, [4]> x1_101_begin_0 = const()[name = tensor<string, []>("x1_101_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_101_end_0 = const()[name = tensor<string, []>("x1_101_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_101_end_mask_0 = const()[name = tensor<string, []>("x1_101_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_101 = slice_by_index(begin = x1_101_begin_0, end = x1_101_end_0, end_mask = x1_101_end_mask_0, x = output_607_cast_fp16)[name = tensor<string, []>("x1_101")];
            tensor<int32, [4]> x2_101_begin_0 = const()[name = tensor<string, []>("x2_101_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_101_end_0 = const()[name = tensor<string, []>("x2_101_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_101_end_mask_0 = const()[name = tensor<string, []>("x2_101_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_101 = slice_by_index(begin = x2_101_begin_0, end = x2_101_end_0, end_mask = x2_101_end_mask_0, x = output_607_cast_fp16)[name = tensor<string, []>("x2_101")];
            tensor<fp16, []> const_610_promoted = const()[name = tensor<string, []>("const_610_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_5670 = mul(x = x2_101, y = const_610_promoted)[name = tensor<string, []>("op_5670")];
            tensor<bool, []> var_5672_interleave_0 = const()[name = tensor<string, []>("op_5672_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_5672 = concat(axis = var_24, interleave = var_5672_interleave_0, values = (var_5670, x1_101))[name = tensor<string, []>("op_5672")];
            tensor<fp16, [1, 8, 256, 256]> var_5673 = mul(x = var_5672, y = sin_7_palettized)[name = tensor<string, []>("op_5673")];
            tensor<fp16, [1, 8, 256, 256]> query_51 = add(x = var_5659, y = var_5673)[name = tensor<string, []>("query_51")];
            tensor<fp16, [1, 4, 256, 256]> var_5675 = mul(x = output_611_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_5675")];
            tensor<int32, [4]> x1_103_begin_0 = const()[name = tensor<string, []>("x1_103_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_103_end_0 = const()[name = tensor<string, []>("x1_103_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_103_end_mask_0 = const()[name = tensor<string, []>("x1_103_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_103 = slice_by_index(begin = x1_103_begin_0, end = x1_103_end_0, end_mask = x1_103_end_mask_0, x = output_611_cast_fp16)[name = tensor<string, []>("x1_103")];
            tensor<int32, [4]> x2_103_begin_0 = const()[name = tensor<string, []>("x2_103_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_103_end_0 = const()[name = tensor<string, []>("x2_103_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_103_end_mask_0 = const()[name = tensor<string, []>("x2_103_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_103 = slice_by_index(begin = x2_103_begin_0, end = x2_103_end_0, end_mask = x2_103_end_mask_0, x = output_611_cast_fp16)[name = tensor<string, []>("x2_103")];
            tensor<fp16, []> const_613_promoted = const()[name = tensor<string, []>("const_613_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_5686 = mul(x = x2_103, y = const_613_promoted)[name = tensor<string, []>("op_5686")];
            tensor<bool, []> var_5688_interleave_0 = const()[name = tensor<string, []>("op_5688_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_5688 = concat(axis = var_24, interleave = var_5688_interleave_0, values = (var_5686, x1_103))[name = tensor<string, []>("op_5688")];
            tensor<fp16, [1, 4, 256, 256]> var_5689 = mul(x = var_5688, y = sin_7_palettized)[name = tensor<string, []>("op_5689")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_353 = add(x = var_5675, y = var_5689)[name = tensor<string, []>("hidden_states_353")];
            tensor<int32, [1]> var_5698_axes_0 = const()[name = tensor<string, []>("op_5698_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_5698 = expand_dims(axes = var_5698_axes_0, x = hidden_states_353)[name = tensor<string, []>("op_5698")];
            tensor<int32, [5]> hidden_states_355_reps_0 = const()[name = tensor<string, []>("hidden_states_355_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_355 = tile(reps = hidden_states_355_reps_0, x = var_5698)[name = tensor<string, []>("hidden_states_355")];
            tensor<int32, [4]> var_5706 = const()[name = tensor<string, []>("op_5706"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_51 = reshape(shape = var_5706, x = hidden_states_355)[name = tensor<string, []>("key_states_51")];
            tensor<int32, [1]> var_5715_axes_0 = const()[name = tensor<string, []>("op_5715_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_357 = transpose(perm = hidden_states_357_perm_0, x = var_5627)[name = tensor<string, []>("transpose_33")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_5715 = expand_dims(axes = var_5715_axes_0, x = hidden_states_357)[name = tensor<string, []>("op_5715")];
            tensor<int32, [5]> hidden_states_359_reps_0 = const()[name = tensor<string, []>("hidden_states_359_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_359 = tile(reps = hidden_states_359_reps_0, x = var_5715)[name = tensor<string, []>("hidden_states_359")];
            tensor<int32, [4]> var_5723 = const()[name = tensor<string, []>("op_5723"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_51 = reshape(shape = var_5723, x = hidden_states_359)[name = tensor<string, []>("value_states_51")];
            tensor<bool, []> var_5726_transpose_x_1 = const()[name = tensor<string, []>("op_5726_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_5726_transpose_y_1 = const()[name = tensor<string, []>("op_5726_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_5726 = matmul(transpose_x = var_5726_transpose_x_1, transpose_y = var_5726_transpose_y_1, x = query_51, y = key_states_51)[name = tensor<string, []>("op_5726")];
            tensor<fp16, []> var_5727_to_fp16 = const()[name = tensor<string, []>("op_5727_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_101_cast_fp16 = mul(x = var_5726, y = var_5727_to_fp16)[name = tensor<string, []>("attn_weights_101_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_301_cast_fp16 = add(x = attn_weights_101_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_301_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_5735_cast_fp16 = softmax(axis = var_24, x = input_301_cast_fp16)[name = tensor<string, []>("op_5735_cast_fp16")];
            tensor<bool, []> attn_output_101_transpose_x_0 = const()[name = tensor<string, []>("attn_output_101_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_101_transpose_y_0 = const()[name = tensor<string, []>("attn_output_101_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_101 = matmul(transpose_x = attn_output_101_transpose_x_0, transpose_y = attn_output_101_transpose_y_0, x = var_5735_cast_fp16, y = value_states_51)[name = tensor<string, []>("attn_output_101")];
            tensor<int32, [4]> var_5739_perm_0 = const()[name = tensor<string, []>("op_5739_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_5741 = const()[name = tensor<string, []>("op_5741"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_5739 = transpose(perm = var_5739_perm_0, x = attn_output_101)[name = tensor<string, []>("transpose_32")];
            tensor<fp16, [1, 256, 2048]> var_5742 = reshape(shape = var_5741, x = var_5739)[name = tensor<string, []>("op_5742")];
            tensor<fp16, [1, 256, 2560]> x_611 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_25_self_attn_o_proj_weight_palettized, x = var_5742)[name = tensor<string, []>("linear_178")];
            tensor<fp16, []> var_22_promoted_153_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_153_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_5748_cast_fp16 = pow(x = x_611, y = var_22_promoted_153_to_fp16)[name = tensor<string, []>("op_5748_cast_fp16")];
            tensor<int32, [1]> var_5750_axes_0 = const()[name = tensor<string, []>("op_5750_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5750_keep_dims_0 = const()[name = tensor<string, []>("op_5750_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_5750_cast_fp16 = reduce_mean(axes = var_5750_axes_0, keep_dims = var_5750_keep_dims_0, x = var_5748_cast_fp16)[name = tensor<string, []>("op_5750_cast_fp16")];
            tensor<fp16, []> var_5751_to_fp16 = const()[name = tensor<string, []>("op_5751_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_5752_cast_fp16 = add(x = var_5750_cast_fp16, y = var_5751_to_fp16)[name = tensor<string, []>("op_5752_cast_fp16")];
            tensor<fp32, []> var_5753_epsilon_0 = const()[name = tensor<string, []>("op_5753_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_5753_cast_fp16 = rsqrt(epsilon = var_5753_epsilon_0, x = var_5752_cast_fp16)[name = tensor<string, []>("op_5753_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_613_cast_fp16 = mul(x = x_611, y = var_5753_cast_fp16)[name = tensor<string, []>("output_613_cast_fp16")];
            tensor<fp16, [2560]> var_5757_to_fp16 = const()[name = tensor<string, []>("op_5757_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940743808)))];
            tensor<fp16, [1, 256, 2560]> output_615_cast_fp16 = mul(x = output_613_cast_fp16, y = var_5757_to_fp16)[name = tensor<string, []>("output_615_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_615 = add(x = x_599, y = output_615_cast_fp16)[name = tensor<string, []>("x_615")];
            tensor<fp16, []> var_22_promoted_154_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_154_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_5763_cast_fp16 = pow(x = x_615, y = var_22_promoted_154_to_fp16)[name = tensor<string, []>("op_5763_cast_fp16")];
            tensor<int32, [1]> var_5765_axes_0 = const()[name = tensor<string, []>("op_5765_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5765_keep_dims_0 = const()[name = tensor<string, []>("op_5765_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_5765_cast_fp16 = reduce_mean(axes = var_5765_axes_0, keep_dims = var_5765_keep_dims_0, x = var_5763_cast_fp16)[name = tensor<string, []>("op_5765_cast_fp16")];
            tensor<fp16, []> var_5766_to_fp16 = const()[name = tensor<string, []>("op_5766_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_5767_cast_fp16 = add(x = var_5765_cast_fp16, y = var_5766_to_fp16)[name = tensor<string, []>("op_5767_cast_fp16")];
            tensor<fp32, []> var_5768_epsilon_0 = const()[name = tensor<string, []>("op_5768_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_5768_cast_fp16 = rsqrt(epsilon = var_5768_epsilon_0, x = var_5767_cast_fp16)[name = tensor<string, []>("op_5768_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_617_cast_fp16 = mul(x = x_615, y = var_5768_cast_fp16)[name = tensor<string, []>("output_617_cast_fp16")];
            tensor<fp16, [2560]> var_5772_to_fp16 = const()[name = tensor<string, []>("op_5772_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940748992)))];
            tensor<fp16, [1, 256, 2560]> output_619_cast_fp16 = mul(x = output_617_cast_fp16, y = var_5772_to_fp16)[name = tensor<string, []>("output_619_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_309 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_25_mlp_gate_proj_weight_palettized, x = output_619_cast_fp16)[name = tensor<string, []>("linear_179")];
            tensor<string, []> var_5780_mode_0 = const()[name = tensor<string, []>("op_5780_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_5780 = gelu(mode = var_5780_mode_0, x = input_309)[name = tensor<string, []>("op_5780")];
            tensor<fp16, [1, 256, 10240]> var_5782 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_25_mlp_up_proj_weight_palettized, x = output_619_cast_fp16)[name = tensor<string, []>("linear_180")];
            tensor<fp16, [1, 256, 10240]> input_311 = mul(x = var_5780, y = var_5782)[name = tensor<string, []>("input_311")];
            tensor<fp16, [1, 256, 2560]> x_619 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_25_mlp_down_proj_weight_palettized, x = input_311)[name = tensor<string, []>("linear_181")];
            tensor<fp16, []> var_22_promoted_155_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_155_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_5788_cast_fp16 = pow(x = x_619, y = var_22_promoted_155_to_fp16)[name = tensor<string, []>("op_5788_cast_fp16")];
            tensor<int32, [1]> var_5790_axes_0 = const()[name = tensor<string, []>("op_5790_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5790_keep_dims_0 = const()[name = tensor<string, []>("op_5790_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_5790_cast_fp16 = reduce_mean(axes = var_5790_axes_0, keep_dims = var_5790_keep_dims_0, x = var_5788_cast_fp16)[name = tensor<string, []>("op_5790_cast_fp16")];
            tensor<fp16, []> var_5791_to_fp16 = const()[name = tensor<string, []>("op_5791_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_5792_cast_fp16 = add(x = var_5790_cast_fp16, y = var_5791_to_fp16)[name = tensor<string, []>("op_5792_cast_fp16")];
            tensor<fp32, []> var_5793_epsilon_0 = const()[name = tensor<string, []>("op_5793_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_5793_cast_fp16 = rsqrt(epsilon = var_5793_epsilon_0, x = var_5792_cast_fp16)[name = tensor<string, []>("op_5793_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_621_cast_fp16 = mul(x = x_619, y = var_5793_cast_fp16)[name = tensor<string, []>("output_621_cast_fp16")];
            tensor<fp16, [2560]> var_5797_to_fp16 = const()[name = tensor<string, []>("op_5797_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940754176)))];
            tensor<fp16, [1, 256, 2560]> output_623_cast_fp16 = mul(x = output_621_cast_fp16, y = var_5797_to_fp16)[name = tensor<string, []>("output_623_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_623 = add(x = x_615, y = output_623_cast_fp16)[name = tensor<string, []>("x_623")];
            tensor<fp16, []> var_22_promoted_156_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_156_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_5809_cast_fp16 = pow(x = x_623, y = var_22_promoted_156_to_fp16)[name = tensor<string, []>("op_5809_cast_fp16")];
            tensor<int32, [1]> var_5811_axes_0 = const()[name = tensor<string, []>("op_5811_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5811_keep_dims_0 = const()[name = tensor<string, []>("op_5811_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_5811_cast_fp16 = reduce_mean(axes = var_5811_axes_0, keep_dims = var_5811_keep_dims_0, x = var_5809_cast_fp16)[name = tensor<string, []>("op_5811_cast_fp16")];
            tensor<fp16, []> var_5812_to_fp16 = const()[name = tensor<string, []>("op_5812_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_5813_cast_fp16 = add(x = var_5811_cast_fp16, y = var_5812_to_fp16)[name = tensor<string, []>("op_5813_cast_fp16")];
            tensor<fp32, []> var_5814_epsilon_0 = const()[name = tensor<string, []>("op_5814_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_5814_cast_fp16 = rsqrt(epsilon = var_5814_epsilon_0, x = var_5813_cast_fp16)[name = tensor<string, []>("op_5814_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_625_cast_fp16 = mul(x = x_623, y = var_5814_cast_fp16)[name = tensor<string, []>("output_625_cast_fp16")];
            tensor<fp16, [2560]> var_5818_to_fp16 = const()[name = tensor<string, []>("op_5818_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940759360)))];
            tensor<fp16, [1, 256, 2560]> output_627_cast_fp16 = mul(x = output_625_cast_fp16, y = var_5818_to_fp16)[name = tensor<string, []>("output_627_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_5830 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_26_self_attn_q_proj_weight_palettized, x = output_627_cast_fp16)[name = tensor<string, []>("linear_182")];
            tensor<int32, [4]> var_5831 = const()[name = tensor<string, []>("op_5831"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_5832 = reshape(shape = var_5831, x = var_5830)[name = tensor<string, []>("op_5832")];
            tensor<int32, [4]> x_627_perm_0 = const()[name = tensor<string, []>("x_627_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_5835 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_26_self_attn_k_proj_weight_palettized, x = output_627_cast_fp16)[name = tensor<string, []>("linear_183")];
            tensor<int32, [4]> var_5836 = const()[name = tensor<string, []>("op_5836"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_5837 = reshape(shape = var_5836, x = var_5835)[name = tensor<string, []>("op_5837")];
            tensor<int32, [4]> x_631_perm_0 = const()[name = tensor<string, []>("x_631_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_5840 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_26_self_attn_v_proj_weight_palettized, x = output_627_cast_fp16)[name = tensor<string, []>("linear_184")];
            tensor<int32, [4]> var_5841 = const()[name = tensor<string, []>("op_5841"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_5842 = reshape(shape = var_5841, x = var_5840)[name = tensor<string, []>("op_5842")];
            tensor<int32, [4]> hidden_states_371_perm_0 = const()[name = tensor<string, []>("hidden_states_371_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_157_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_157_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_627 = transpose(perm = x_627_perm_0, x = var_5832)[name = tensor<string, []>("transpose_31")];
            tensor<fp16, [1, 8, 256, 256]> var_5846_cast_fp16 = pow(x = x_627, y = var_22_promoted_157_to_fp16)[name = tensor<string, []>("op_5846_cast_fp16")];
            tensor<int32, [1]> var_5848_axes_0 = const()[name = tensor<string, []>("op_5848_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5848_keep_dims_0 = const()[name = tensor<string, []>("op_5848_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_5848_cast_fp16 = reduce_mean(axes = var_5848_axes_0, keep_dims = var_5848_keep_dims_0, x = var_5846_cast_fp16)[name = tensor<string, []>("op_5848_cast_fp16")];
            tensor<fp16, []> var_5849_to_fp16 = const()[name = tensor<string, []>("op_5849_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_5850_cast_fp16 = add(x = var_5848_cast_fp16, y = var_5849_to_fp16)[name = tensor<string, []>("op_5850_cast_fp16")];
            tensor<fp32, []> var_5851_epsilon_0 = const()[name = tensor<string, []>("op_5851_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_5851_cast_fp16 = rsqrt(epsilon = var_5851_epsilon_0, x = var_5850_cast_fp16)[name = tensor<string, []>("op_5851_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_629_cast_fp16 = mul(x = x_627, y = var_5851_cast_fp16)[name = tensor<string, []>("output_629_cast_fp16")];
            tensor<fp16, [256]> var_5855_to_fp16 = const()[name = tensor<string, []>("op_5855_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940764544)))];
            tensor<fp16, [1, 8, 256, 256]> output_631_cast_fp16 = mul(x = output_629_cast_fp16, y = var_5855_to_fp16)[name = tensor<string, []>("output_631_cast_fp16")];
            tensor<fp16, []> var_22_promoted_158_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_158_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_631 = transpose(perm = x_631_perm_0, x = var_5837)[name = tensor<string, []>("transpose_30")];
            tensor<fp16, [1, 4, 256, 256]> var_5860_cast_fp16 = pow(x = x_631, y = var_22_promoted_158_to_fp16)[name = tensor<string, []>("op_5860_cast_fp16")];
            tensor<int32, [1]> var_5862_axes_0 = const()[name = tensor<string, []>("op_5862_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5862_keep_dims_0 = const()[name = tensor<string, []>("op_5862_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_5862_cast_fp16 = reduce_mean(axes = var_5862_axes_0, keep_dims = var_5862_keep_dims_0, x = var_5860_cast_fp16)[name = tensor<string, []>("op_5862_cast_fp16")];
            tensor<fp16, []> var_5863_to_fp16 = const()[name = tensor<string, []>("op_5863_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_5864_cast_fp16 = add(x = var_5862_cast_fp16, y = var_5863_to_fp16)[name = tensor<string, []>("op_5864_cast_fp16")];
            tensor<fp32, []> var_5865_epsilon_0 = const()[name = tensor<string, []>("op_5865_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_5865_cast_fp16 = rsqrt(epsilon = var_5865_epsilon_0, x = var_5864_cast_fp16)[name = tensor<string, []>("op_5865_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_633_cast_fp16 = mul(x = x_631, y = var_5865_cast_fp16)[name = tensor<string, []>("output_633_cast_fp16")];
            tensor<fp16, [256]> var_5869_to_fp16 = const()[name = tensor<string, []>("op_5869_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940765120)))];
            tensor<fp16, [1, 4, 256, 256]> output_635_cast_fp16 = mul(x = output_633_cast_fp16, y = var_5869_to_fp16)[name = tensor<string, []>("output_635_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_5874 = mul(x = output_631_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_5874")];
            tensor<int32, [4]> x1_105_begin_0 = const()[name = tensor<string, []>("x1_105_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_105_end_0 = const()[name = tensor<string, []>("x1_105_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_105_end_mask_0 = const()[name = tensor<string, []>("x1_105_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_105 = slice_by_index(begin = x1_105_begin_0, end = x1_105_end_0, end_mask = x1_105_end_mask_0, x = output_631_cast_fp16)[name = tensor<string, []>("x1_105")];
            tensor<int32, [4]> x2_105_begin_0 = const()[name = tensor<string, []>("x2_105_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_105_end_0 = const()[name = tensor<string, []>("x2_105_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_105_end_mask_0 = const()[name = tensor<string, []>("x2_105_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_105 = slice_by_index(begin = x2_105_begin_0, end = x2_105_end_0, end_mask = x2_105_end_mask_0, x = output_631_cast_fp16)[name = tensor<string, []>("x2_105")];
            tensor<fp16, []> const_633_promoted = const()[name = tensor<string, []>("const_633_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_5885 = mul(x = x2_105, y = const_633_promoted)[name = tensor<string, []>("op_5885")];
            tensor<bool, []> var_5887_interleave_0 = const()[name = tensor<string, []>("op_5887_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_5887 = concat(axis = var_24, interleave = var_5887_interleave_0, values = (var_5885, x1_105))[name = tensor<string, []>("op_5887")];
            tensor<fp16, [1, 8, 256, 256]> var_5888 = mul(x = var_5887, y = sin_7_palettized)[name = tensor<string, []>("op_5888")];
            tensor<fp16, [1, 8, 256, 256]> query_53 = add(x = var_5874, y = var_5888)[name = tensor<string, []>("query_53")];
            tensor<fp16, [1, 4, 256, 256]> var_5890 = mul(x = output_635_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_5890")];
            tensor<int32, [4]> x1_107_begin_0 = const()[name = tensor<string, []>("x1_107_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_107_end_0 = const()[name = tensor<string, []>("x1_107_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_107_end_mask_0 = const()[name = tensor<string, []>("x1_107_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_107 = slice_by_index(begin = x1_107_begin_0, end = x1_107_end_0, end_mask = x1_107_end_mask_0, x = output_635_cast_fp16)[name = tensor<string, []>("x1_107")];
            tensor<int32, [4]> x2_107_begin_0 = const()[name = tensor<string, []>("x2_107_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_107_end_0 = const()[name = tensor<string, []>("x2_107_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_107_end_mask_0 = const()[name = tensor<string, []>("x2_107_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_107 = slice_by_index(begin = x2_107_begin_0, end = x2_107_end_0, end_mask = x2_107_end_mask_0, x = output_635_cast_fp16)[name = tensor<string, []>("x2_107")];
            tensor<fp16, []> const_636_promoted = const()[name = tensor<string, []>("const_636_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_5901 = mul(x = x2_107, y = const_636_promoted)[name = tensor<string, []>("op_5901")];
            tensor<bool, []> var_5903_interleave_0 = const()[name = tensor<string, []>("op_5903_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_5903 = concat(axis = var_24, interleave = var_5903_interleave_0, values = (var_5901, x1_107))[name = tensor<string, []>("op_5903")];
            tensor<fp16, [1, 4, 256, 256]> var_5904 = mul(x = var_5903, y = sin_7_palettized)[name = tensor<string, []>("op_5904")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_367 = add(x = var_5890, y = var_5904)[name = tensor<string, []>("hidden_states_367")];
            tensor<int32, [1]> var_5913_axes_0 = const()[name = tensor<string, []>("op_5913_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_5913 = expand_dims(axes = var_5913_axes_0, x = hidden_states_367)[name = tensor<string, []>("op_5913")];
            tensor<int32, [5]> hidden_states_369_reps_0 = const()[name = tensor<string, []>("hidden_states_369_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_369 = tile(reps = hidden_states_369_reps_0, x = var_5913)[name = tensor<string, []>("hidden_states_369")];
            tensor<int32, [4]> var_5921 = const()[name = tensor<string, []>("op_5921"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_53 = reshape(shape = var_5921, x = hidden_states_369)[name = tensor<string, []>("key_states_53")];
            tensor<int32, [1]> var_5930_axes_0 = const()[name = tensor<string, []>("op_5930_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_371 = transpose(perm = hidden_states_371_perm_0, x = var_5842)[name = tensor<string, []>("transpose_29")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_5930 = expand_dims(axes = var_5930_axes_0, x = hidden_states_371)[name = tensor<string, []>("op_5930")];
            tensor<int32, [5]> hidden_states_373_reps_0 = const()[name = tensor<string, []>("hidden_states_373_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_373 = tile(reps = hidden_states_373_reps_0, x = var_5930)[name = tensor<string, []>("hidden_states_373")];
            tensor<int32, [4]> var_5938 = const()[name = tensor<string, []>("op_5938"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_53 = reshape(shape = var_5938, x = hidden_states_373)[name = tensor<string, []>("value_states_53")];
            tensor<bool, []> var_5941_transpose_x_1 = const()[name = tensor<string, []>("op_5941_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_5941_transpose_y_1 = const()[name = tensor<string, []>("op_5941_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_5941 = matmul(transpose_x = var_5941_transpose_x_1, transpose_y = var_5941_transpose_y_1, x = query_53, y = key_states_53)[name = tensor<string, []>("op_5941")];
            tensor<fp16, []> var_5942_to_fp16 = const()[name = tensor<string, []>("op_5942_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_105_cast_fp16 = mul(x = var_5941, y = var_5942_to_fp16)[name = tensor<string, []>("attn_weights_105_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_313_cast_fp16 = add(x = attn_weights_105_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_313_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_5950_cast_fp16 = softmax(axis = var_24, x = input_313_cast_fp16)[name = tensor<string, []>("op_5950_cast_fp16")];
            tensor<bool, []> attn_output_105_transpose_x_0 = const()[name = tensor<string, []>("attn_output_105_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_105_transpose_y_0 = const()[name = tensor<string, []>("attn_output_105_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_105 = matmul(transpose_x = attn_output_105_transpose_x_0, transpose_y = attn_output_105_transpose_y_0, x = var_5950_cast_fp16, y = value_states_53)[name = tensor<string, []>("attn_output_105")];
            tensor<int32, [4]> var_5954_perm_0 = const()[name = tensor<string, []>("op_5954_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_5956 = const()[name = tensor<string, []>("op_5956"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_5954 = transpose(perm = var_5954_perm_0, x = attn_output_105)[name = tensor<string, []>("transpose_28")];
            tensor<fp16, [1, 256, 2048]> var_5957 = reshape(shape = var_5956, x = var_5954)[name = tensor<string, []>("op_5957")];
            tensor<fp16, [1, 256, 2560]> x_635 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_26_self_attn_o_proj_weight_palettized, x = var_5957)[name = tensor<string, []>("linear_185")];
            tensor<fp16, []> var_22_promoted_159_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_159_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_5963_cast_fp16 = pow(x = x_635, y = var_22_promoted_159_to_fp16)[name = tensor<string, []>("op_5963_cast_fp16")];
            tensor<int32, [1]> var_5965_axes_0 = const()[name = tensor<string, []>("op_5965_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5965_keep_dims_0 = const()[name = tensor<string, []>("op_5965_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_5965_cast_fp16 = reduce_mean(axes = var_5965_axes_0, keep_dims = var_5965_keep_dims_0, x = var_5963_cast_fp16)[name = tensor<string, []>("op_5965_cast_fp16")];
            tensor<fp16, []> var_5966_to_fp16 = const()[name = tensor<string, []>("op_5966_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_5967_cast_fp16 = add(x = var_5965_cast_fp16, y = var_5966_to_fp16)[name = tensor<string, []>("op_5967_cast_fp16")];
            tensor<fp32, []> var_5968_epsilon_0 = const()[name = tensor<string, []>("op_5968_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_5968_cast_fp16 = rsqrt(epsilon = var_5968_epsilon_0, x = var_5967_cast_fp16)[name = tensor<string, []>("op_5968_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_637_cast_fp16 = mul(x = x_635, y = var_5968_cast_fp16)[name = tensor<string, []>("output_637_cast_fp16")];
            tensor<fp16, [2560]> var_5972_to_fp16 = const()[name = tensor<string, []>("op_5972_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940765696)))];
            tensor<fp16, [1, 256, 2560]> output_639_cast_fp16 = mul(x = output_637_cast_fp16, y = var_5972_to_fp16)[name = tensor<string, []>("output_639_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_639 = add(x = x_623, y = output_639_cast_fp16)[name = tensor<string, []>("x_639")];
            tensor<fp16, []> var_22_promoted_160_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_160_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_5978_cast_fp16 = pow(x = x_639, y = var_22_promoted_160_to_fp16)[name = tensor<string, []>("op_5978_cast_fp16")];
            tensor<int32, [1]> var_5980_axes_0 = const()[name = tensor<string, []>("op_5980_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_5980_keep_dims_0 = const()[name = tensor<string, []>("op_5980_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_5980_cast_fp16 = reduce_mean(axes = var_5980_axes_0, keep_dims = var_5980_keep_dims_0, x = var_5978_cast_fp16)[name = tensor<string, []>("op_5980_cast_fp16")];
            tensor<fp16, []> var_5981_to_fp16 = const()[name = tensor<string, []>("op_5981_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_5982_cast_fp16 = add(x = var_5980_cast_fp16, y = var_5981_to_fp16)[name = tensor<string, []>("op_5982_cast_fp16")];
            tensor<fp32, []> var_5983_epsilon_0 = const()[name = tensor<string, []>("op_5983_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_5983_cast_fp16 = rsqrt(epsilon = var_5983_epsilon_0, x = var_5982_cast_fp16)[name = tensor<string, []>("op_5983_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_641_cast_fp16 = mul(x = x_639, y = var_5983_cast_fp16)[name = tensor<string, []>("output_641_cast_fp16")];
            tensor<fp16, [2560]> var_5987_to_fp16 = const()[name = tensor<string, []>("op_5987_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940770880)))];
            tensor<fp16, [1, 256, 2560]> output_643_cast_fp16 = mul(x = output_641_cast_fp16, y = var_5987_to_fp16)[name = tensor<string, []>("output_643_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_321 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_26_mlp_gate_proj_weight_palettized, x = output_643_cast_fp16)[name = tensor<string, []>("linear_186")];
            tensor<string, []> var_5995_mode_0 = const()[name = tensor<string, []>("op_5995_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_5995 = gelu(mode = var_5995_mode_0, x = input_321)[name = tensor<string, []>("op_5995")];
            tensor<fp16, [1, 256, 10240]> var_5997 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_26_mlp_up_proj_weight_palettized, x = output_643_cast_fp16)[name = tensor<string, []>("linear_187")];
            tensor<fp16, [1, 256, 10240]> input_323 = mul(x = var_5995, y = var_5997)[name = tensor<string, []>("input_323")];
            tensor<fp16, [1, 256, 2560]> x_643 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_26_mlp_down_proj_weight_palettized, x = input_323)[name = tensor<string, []>("linear_188")];
            tensor<fp16, []> var_22_promoted_161_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_161_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_6003_cast_fp16 = pow(x = x_643, y = var_22_promoted_161_to_fp16)[name = tensor<string, []>("op_6003_cast_fp16")];
            tensor<int32, [1]> var_6005_axes_0 = const()[name = tensor<string, []>("op_6005_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6005_keep_dims_0 = const()[name = tensor<string, []>("op_6005_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_6005_cast_fp16 = reduce_mean(axes = var_6005_axes_0, keep_dims = var_6005_keep_dims_0, x = var_6003_cast_fp16)[name = tensor<string, []>("op_6005_cast_fp16")];
            tensor<fp16, []> var_6006_to_fp16 = const()[name = tensor<string, []>("op_6006_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_6007_cast_fp16 = add(x = var_6005_cast_fp16, y = var_6006_to_fp16)[name = tensor<string, []>("op_6007_cast_fp16")];
            tensor<fp32, []> var_6008_epsilon_0 = const()[name = tensor<string, []>("op_6008_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_6008_cast_fp16 = rsqrt(epsilon = var_6008_epsilon_0, x = var_6007_cast_fp16)[name = tensor<string, []>("op_6008_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_645_cast_fp16 = mul(x = x_643, y = var_6008_cast_fp16)[name = tensor<string, []>("output_645_cast_fp16")];
            tensor<fp16, [2560]> var_6012_to_fp16 = const()[name = tensor<string, []>("op_6012_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940776064)))];
            tensor<fp16, [1, 256, 2560]> output_647_cast_fp16 = mul(x = output_645_cast_fp16, y = var_6012_to_fp16)[name = tensor<string, []>("output_647_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_647 = add(x = x_639, y = output_647_cast_fp16)[name = tensor<string, []>("x_647")];
            tensor<fp16, []> var_22_promoted_162_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_162_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_6024_cast_fp16 = pow(x = x_647, y = var_22_promoted_162_to_fp16)[name = tensor<string, []>("op_6024_cast_fp16")];
            tensor<int32, [1]> var_6026_axes_0 = const()[name = tensor<string, []>("op_6026_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6026_keep_dims_0 = const()[name = tensor<string, []>("op_6026_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_6026_cast_fp16 = reduce_mean(axes = var_6026_axes_0, keep_dims = var_6026_keep_dims_0, x = var_6024_cast_fp16)[name = tensor<string, []>("op_6026_cast_fp16")];
            tensor<fp16, []> var_6027_to_fp16 = const()[name = tensor<string, []>("op_6027_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_6028_cast_fp16 = add(x = var_6026_cast_fp16, y = var_6027_to_fp16)[name = tensor<string, []>("op_6028_cast_fp16")];
            tensor<fp32, []> var_6029_epsilon_0 = const()[name = tensor<string, []>("op_6029_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_6029_cast_fp16 = rsqrt(epsilon = var_6029_epsilon_0, x = var_6028_cast_fp16)[name = tensor<string, []>("op_6029_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_649_cast_fp16 = mul(x = x_647, y = var_6029_cast_fp16)[name = tensor<string, []>("output_649_cast_fp16")];
            tensor<fp16, [2560]> var_6033_to_fp16 = const()[name = tensor<string, []>("op_6033_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940781248)))];
            tensor<fp16, [1, 256, 2560]> output_651_cast_fp16 = mul(x = output_649_cast_fp16, y = var_6033_to_fp16)[name = tensor<string, []>("output_651_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_6045 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_27_self_attn_q_proj_weight_palettized, x = output_651_cast_fp16)[name = tensor<string, []>("linear_189")];
            tensor<int32, [4]> var_6046 = const()[name = tensor<string, []>("op_6046"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_6047 = reshape(shape = var_6046, x = var_6045)[name = tensor<string, []>("op_6047")];
            tensor<int32, [4]> x_651_perm_0 = const()[name = tensor<string, []>("x_651_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_6050 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_27_self_attn_k_proj_weight_palettized, x = output_651_cast_fp16)[name = tensor<string, []>("linear_190")];
            tensor<int32, [4]> var_6051 = const()[name = tensor<string, []>("op_6051"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_6052 = reshape(shape = var_6051, x = var_6050)[name = tensor<string, []>("op_6052")];
            tensor<int32, [4]> x_655_perm_0 = const()[name = tensor<string, []>("x_655_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_6055 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_27_self_attn_v_proj_weight_palettized, x = output_651_cast_fp16)[name = tensor<string, []>("linear_191")];
            tensor<int32, [4]> var_6056 = const()[name = tensor<string, []>("op_6056"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_6057 = reshape(shape = var_6056, x = var_6055)[name = tensor<string, []>("op_6057")];
            tensor<int32, [4]> hidden_states_385_perm_0 = const()[name = tensor<string, []>("hidden_states_385_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_163_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_163_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_651 = transpose(perm = x_651_perm_0, x = var_6047)[name = tensor<string, []>("transpose_27")];
            tensor<fp16, [1, 8, 256, 256]> var_6061_cast_fp16 = pow(x = x_651, y = var_22_promoted_163_to_fp16)[name = tensor<string, []>("op_6061_cast_fp16")];
            tensor<int32, [1]> var_6063_axes_0 = const()[name = tensor<string, []>("op_6063_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6063_keep_dims_0 = const()[name = tensor<string, []>("op_6063_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_6063_cast_fp16 = reduce_mean(axes = var_6063_axes_0, keep_dims = var_6063_keep_dims_0, x = var_6061_cast_fp16)[name = tensor<string, []>("op_6063_cast_fp16")];
            tensor<fp16, []> var_6064_to_fp16 = const()[name = tensor<string, []>("op_6064_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_6065_cast_fp16 = add(x = var_6063_cast_fp16, y = var_6064_to_fp16)[name = tensor<string, []>("op_6065_cast_fp16")];
            tensor<fp32, []> var_6066_epsilon_0 = const()[name = tensor<string, []>("op_6066_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_6066_cast_fp16 = rsqrt(epsilon = var_6066_epsilon_0, x = var_6065_cast_fp16)[name = tensor<string, []>("op_6066_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_653_cast_fp16 = mul(x = x_651, y = var_6066_cast_fp16)[name = tensor<string, []>("output_653_cast_fp16")];
            tensor<fp16, [256]> var_6070_to_fp16 = const()[name = tensor<string, []>("op_6070_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940786432)))];
            tensor<fp16, [1, 8, 256, 256]> output_655_cast_fp16 = mul(x = output_653_cast_fp16, y = var_6070_to_fp16)[name = tensor<string, []>("output_655_cast_fp16")];
            tensor<fp16, []> var_22_promoted_164_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_164_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_655 = transpose(perm = x_655_perm_0, x = var_6052)[name = tensor<string, []>("transpose_26")];
            tensor<fp16, [1, 4, 256, 256]> var_6075_cast_fp16 = pow(x = x_655, y = var_22_promoted_164_to_fp16)[name = tensor<string, []>("op_6075_cast_fp16")];
            tensor<int32, [1]> var_6077_axes_0 = const()[name = tensor<string, []>("op_6077_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6077_keep_dims_0 = const()[name = tensor<string, []>("op_6077_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_6077_cast_fp16 = reduce_mean(axes = var_6077_axes_0, keep_dims = var_6077_keep_dims_0, x = var_6075_cast_fp16)[name = tensor<string, []>("op_6077_cast_fp16")];
            tensor<fp16, []> var_6078_to_fp16 = const()[name = tensor<string, []>("op_6078_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_6079_cast_fp16 = add(x = var_6077_cast_fp16, y = var_6078_to_fp16)[name = tensor<string, []>("op_6079_cast_fp16")];
            tensor<fp32, []> var_6080_epsilon_0 = const()[name = tensor<string, []>("op_6080_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_6080_cast_fp16 = rsqrt(epsilon = var_6080_epsilon_0, x = var_6079_cast_fp16)[name = tensor<string, []>("op_6080_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_657_cast_fp16 = mul(x = x_655, y = var_6080_cast_fp16)[name = tensor<string, []>("output_657_cast_fp16")];
            tensor<fp16, [256]> var_6084_to_fp16 = const()[name = tensor<string, []>("op_6084_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940787008)))];
            tensor<fp16, [1, 4, 256, 256]> output_659_cast_fp16 = mul(x = output_657_cast_fp16, y = var_6084_to_fp16)[name = tensor<string, []>("output_659_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_6089 = mul(x = output_655_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_6089")];
            tensor<int32, [4]> x1_109_begin_0 = const()[name = tensor<string, []>("x1_109_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_109_end_0 = const()[name = tensor<string, []>("x1_109_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_109_end_mask_0 = const()[name = tensor<string, []>("x1_109_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_109 = slice_by_index(begin = x1_109_begin_0, end = x1_109_end_0, end_mask = x1_109_end_mask_0, x = output_655_cast_fp16)[name = tensor<string, []>("x1_109")];
            tensor<int32, [4]> x2_109_begin_0 = const()[name = tensor<string, []>("x2_109_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_109_end_0 = const()[name = tensor<string, []>("x2_109_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_109_end_mask_0 = const()[name = tensor<string, []>("x2_109_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_109 = slice_by_index(begin = x2_109_begin_0, end = x2_109_end_0, end_mask = x2_109_end_mask_0, x = output_655_cast_fp16)[name = tensor<string, []>("x2_109")];
            tensor<fp16, []> const_656_promoted = const()[name = tensor<string, []>("const_656_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_6100 = mul(x = x2_109, y = const_656_promoted)[name = tensor<string, []>("op_6100")];
            tensor<bool, []> var_6102_interleave_0 = const()[name = tensor<string, []>("op_6102_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_6102 = concat(axis = var_24, interleave = var_6102_interleave_0, values = (var_6100, x1_109))[name = tensor<string, []>("op_6102")];
            tensor<fp16, [1, 8, 256, 256]> var_6103 = mul(x = var_6102, y = sin_7_palettized)[name = tensor<string, []>("op_6103")];
            tensor<fp16, [1, 8, 256, 256]> query_55 = add(x = var_6089, y = var_6103)[name = tensor<string, []>("query_55")];
            tensor<fp16, [1, 4, 256, 256]> var_6105 = mul(x = output_659_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_6105")];
            tensor<int32, [4]> x1_111_begin_0 = const()[name = tensor<string, []>("x1_111_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_111_end_0 = const()[name = tensor<string, []>("x1_111_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_111_end_mask_0 = const()[name = tensor<string, []>("x1_111_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_111 = slice_by_index(begin = x1_111_begin_0, end = x1_111_end_0, end_mask = x1_111_end_mask_0, x = output_659_cast_fp16)[name = tensor<string, []>("x1_111")];
            tensor<int32, [4]> x2_111_begin_0 = const()[name = tensor<string, []>("x2_111_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_111_end_0 = const()[name = tensor<string, []>("x2_111_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_111_end_mask_0 = const()[name = tensor<string, []>("x2_111_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_111 = slice_by_index(begin = x2_111_begin_0, end = x2_111_end_0, end_mask = x2_111_end_mask_0, x = output_659_cast_fp16)[name = tensor<string, []>("x2_111")];
            tensor<fp16, []> const_659_promoted = const()[name = tensor<string, []>("const_659_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_6116 = mul(x = x2_111, y = const_659_promoted)[name = tensor<string, []>("op_6116")];
            tensor<bool, []> var_6118_interleave_0 = const()[name = tensor<string, []>("op_6118_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_6118 = concat(axis = var_24, interleave = var_6118_interleave_0, values = (var_6116, x1_111))[name = tensor<string, []>("op_6118")];
            tensor<fp16, [1, 4, 256, 256]> var_6119 = mul(x = var_6118, y = sin_7_palettized)[name = tensor<string, []>("op_6119")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_381 = add(x = var_6105, y = var_6119)[name = tensor<string, []>("hidden_states_381")];
            tensor<int32, [1]> var_6128_axes_0 = const()[name = tensor<string, []>("op_6128_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_6128 = expand_dims(axes = var_6128_axes_0, x = hidden_states_381)[name = tensor<string, []>("op_6128")];
            tensor<int32, [5]> hidden_states_383_reps_0 = const()[name = tensor<string, []>("hidden_states_383_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_383 = tile(reps = hidden_states_383_reps_0, x = var_6128)[name = tensor<string, []>("hidden_states_383")];
            tensor<int32, [4]> var_6136 = const()[name = tensor<string, []>("op_6136"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_55 = reshape(shape = var_6136, x = hidden_states_383)[name = tensor<string, []>("key_states_55")];
            tensor<int32, [1]> var_6145_axes_0 = const()[name = tensor<string, []>("op_6145_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_385 = transpose(perm = hidden_states_385_perm_0, x = var_6057)[name = tensor<string, []>("transpose_25")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_6145 = expand_dims(axes = var_6145_axes_0, x = hidden_states_385)[name = tensor<string, []>("op_6145")];
            tensor<int32, [5]> hidden_states_387_reps_0 = const()[name = tensor<string, []>("hidden_states_387_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_387 = tile(reps = hidden_states_387_reps_0, x = var_6145)[name = tensor<string, []>("hidden_states_387")];
            tensor<int32, [4]> var_6153 = const()[name = tensor<string, []>("op_6153"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_55 = reshape(shape = var_6153, x = hidden_states_387)[name = tensor<string, []>("value_states_55")];
            tensor<bool, []> var_6156_transpose_x_1 = const()[name = tensor<string, []>("op_6156_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_6156_transpose_y_1 = const()[name = tensor<string, []>("op_6156_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_6156 = matmul(transpose_x = var_6156_transpose_x_1, transpose_y = var_6156_transpose_y_1, x = query_55, y = key_states_55)[name = tensor<string, []>("op_6156")];
            tensor<fp16, []> var_6157_to_fp16 = const()[name = tensor<string, []>("op_6157_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_109_cast_fp16 = mul(x = var_6156, y = var_6157_to_fp16)[name = tensor<string, []>("attn_weights_109_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_325_cast_fp16 = add(x = attn_weights_109_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_325_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_6165_cast_fp16 = softmax(axis = var_24, x = input_325_cast_fp16)[name = tensor<string, []>("op_6165_cast_fp16")];
            tensor<bool, []> attn_output_109_transpose_x_0 = const()[name = tensor<string, []>("attn_output_109_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_109_transpose_y_0 = const()[name = tensor<string, []>("attn_output_109_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_109 = matmul(transpose_x = attn_output_109_transpose_x_0, transpose_y = attn_output_109_transpose_y_0, x = var_6165_cast_fp16, y = value_states_55)[name = tensor<string, []>("attn_output_109")];
            tensor<int32, [4]> var_6169_perm_0 = const()[name = tensor<string, []>("op_6169_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_6171 = const()[name = tensor<string, []>("op_6171"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_6169 = transpose(perm = var_6169_perm_0, x = attn_output_109)[name = tensor<string, []>("transpose_24")];
            tensor<fp16, [1, 256, 2048]> var_6172 = reshape(shape = var_6171, x = var_6169)[name = tensor<string, []>("op_6172")];
            tensor<fp16, [1, 256, 2560]> x_659 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_27_self_attn_o_proj_weight_palettized, x = var_6172)[name = tensor<string, []>("linear_192")];
            tensor<fp16, []> var_22_promoted_165_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_165_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_6178_cast_fp16 = pow(x = x_659, y = var_22_promoted_165_to_fp16)[name = tensor<string, []>("op_6178_cast_fp16")];
            tensor<int32, [1]> var_6180_axes_0 = const()[name = tensor<string, []>("op_6180_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6180_keep_dims_0 = const()[name = tensor<string, []>("op_6180_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_6180_cast_fp16 = reduce_mean(axes = var_6180_axes_0, keep_dims = var_6180_keep_dims_0, x = var_6178_cast_fp16)[name = tensor<string, []>("op_6180_cast_fp16")];
            tensor<fp16, []> var_6181_to_fp16 = const()[name = tensor<string, []>("op_6181_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_6182_cast_fp16 = add(x = var_6180_cast_fp16, y = var_6181_to_fp16)[name = tensor<string, []>("op_6182_cast_fp16")];
            tensor<fp32, []> var_6183_epsilon_0 = const()[name = tensor<string, []>("op_6183_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_6183_cast_fp16 = rsqrt(epsilon = var_6183_epsilon_0, x = var_6182_cast_fp16)[name = tensor<string, []>("op_6183_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_661_cast_fp16 = mul(x = x_659, y = var_6183_cast_fp16)[name = tensor<string, []>("output_661_cast_fp16")];
            tensor<fp16, [2560]> var_6187_to_fp16 = const()[name = tensor<string, []>("op_6187_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940787584)))];
            tensor<fp16, [1, 256, 2560]> output_663_cast_fp16 = mul(x = output_661_cast_fp16, y = var_6187_to_fp16)[name = tensor<string, []>("output_663_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_663 = add(x = x_647, y = output_663_cast_fp16)[name = tensor<string, []>("x_663")];
            tensor<fp16, []> var_22_promoted_166_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_166_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_6193_cast_fp16 = pow(x = x_663, y = var_22_promoted_166_to_fp16)[name = tensor<string, []>("op_6193_cast_fp16")];
            tensor<int32, [1]> var_6195_axes_0 = const()[name = tensor<string, []>("op_6195_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6195_keep_dims_0 = const()[name = tensor<string, []>("op_6195_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_6195_cast_fp16 = reduce_mean(axes = var_6195_axes_0, keep_dims = var_6195_keep_dims_0, x = var_6193_cast_fp16)[name = tensor<string, []>("op_6195_cast_fp16")];
            tensor<fp16, []> var_6196_to_fp16 = const()[name = tensor<string, []>("op_6196_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_6197_cast_fp16 = add(x = var_6195_cast_fp16, y = var_6196_to_fp16)[name = tensor<string, []>("op_6197_cast_fp16")];
            tensor<fp32, []> var_6198_epsilon_0 = const()[name = tensor<string, []>("op_6198_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_6198_cast_fp16 = rsqrt(epsilon = var_6198_epsilon_0, x = var_6197_cast_fp16)[name = tensor<string, []>("op_6198_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_665_cast_fp16 = mul(x = x_663, y = var_6198_cast_fp16)[name = tensor<string, []>("output_665_cast_fp16")];
            tensor<fp16, [2560]> var_6202_to_fp16 = const()[name = tensor<string, []>("op_6202_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940792768)))];
            tensor<fp16, [1, 256, 2560]> output_667_cast_fp16 = mul(x = output_665_cast_fp16, y = var_6202_to_fp16)[name = tensor<string, []>("output_667_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_333 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_27_mlp_gate_proj_weight_palettized, x = output_667_cast_fp16)[name = tensor<string, []>("linear_193")];
            tensor<string, []> var_6210_mode_0 = const()[name = tensor<string, []>("op_6210_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_6210 = gelu(mode = var_6210_mode_0, x = input_333)[name = tensor<string, []>("op_6210")];
            tensor<fp16, [1, 256, 10240]> var_6212 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_27_mlp_up_proj_weight_palettized, x = output_667_cast_fp16)[name = tensor<string, []>("linear_194")];
            tensor<fp16, [1, 256, 10240]> input_335 = mul(x = var_6210, y = var_6212)[name = tensor<string, []>("input_335")];
            tensor<fp16, [1, 256, 2560]> x_667 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_27_mlp_down_proj_weight_palettized, x = input_335)[name = tensor<string, []>("linear_195")];
            tensor<fp16, []> var_22_promoted_167_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_167_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_6218_cast_fp16 = pow(x = x_667, y = var_22_promoted_167_to_fp16)[name = tensor<string, []>("op_6218_cast_fp16")];
            tensor<int32, [1]> var_6220_axes_0 = const()[name = tensor<string, []>("op_6220_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6220_keep_dims_0 = const()[name = tensor<string, []>("op_6220_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_6220_cast_fp16 = reduce_mean(axes = var_6220_axes_0, keep_dims = var_6220_keep_dims_0, x = var_6218_cast_fp16)[name = tensor<string, []>("op_6220_cast_fp16")];
            tensor<fp16, []> var_6221_to_fp16 = const()[name = tensor<string, []>("op_6221_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_6222_cast_fp16 = add(x = var_6220_cast_fp16, y = var_6221_to_fp16)[name = tensor<string, []>("op_6222_cast_fp16")];
            tensor<fp32, []> var_6223_epsilon_0 = const()[name = tensor<string, []>("op_6223_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_6223_cast_fp16 = rsqrt(epsilon = var_6223_epsilon_0, x = var_6222_cast_fp16)[name = tensor<string, []>("op_6223_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_669_cast_fp16 = mul(x = x_667, y = var_6223_cast_fp16)[name = tensor<string, []>("output_669_cast_fp16")];
            tensor<fp16, [2560]> var_6227_to_fp16 = const()[name = tensor<string, []>("op_6227_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940797952)))];
            tensor<fp16, [1, 256, 2560]> output_671_cast_fp16 = mul(x = output_669_cast_fp16, y = var_6227_to_fp16)[name = tensor<string, []>("output_671_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_671 = add(x = x_663, y = output_671_cast_fp16)[name = tensor<string, []>("x_671")];
            tensor<fp16, []> var_22_promoted_168_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_168_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_6239_cast_fp16 = pow(x = x_671, y = var_22_promoted_168_to_fp16)[name = tensor<string, []>("op_6239_cast_fp16")];
            tensor<int32, [1]> var_6241_axes_0 = const()[name = tensor<string, []>("op_6241_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6241_keep_dims_0 = const()[name = tensor<string, []>("op_6241_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_6241_cast_fp16 = reduce_mean(axes = var_6241_axes_0, keep_dims = var_6241_keep_dims_0, x = var_6239_cast_fp16)[name = tensor<string, []>("op_6241_cast_fp16")];
            tensor<fp16, []> var_6242_to_fp16 = const()[name = tensor<string, []>("op_6242_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_6243_cast_fp16 = add(x = var_6241_cast_fp16, y = var_6242_to_fp16)[name = tensor<string, []>("op_6243_cast_fp16")];
            tensor<fp32, []> var_6244_epsilon_0 = const()[name = tensor<string, []>("op_6244_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_6244_cast_fp16 = rsqrt(epsilon = var_6244_epsilon_0, x = var_6243_cast_fp16)[name = tensor<string, []>("op_6244_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_673_cast_fp16 = mul(x = x_671, y = var_6244_cast_fp16)[name = tensor<string, []>("output_673_cast_fp16")];
            tensor<fp16, [2560]> var_6248_to_fp16 = const()[name = tensor<string, []>("op_6248_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940803136)))];
            tensor<fp16, [1, 256, 2560]> output_675_cast_fp16 = mul(x = output_673_cast_fp16, y = var_6248_to_fp16)[name = tensor<string, []>("output_675_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_6260 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_28_self_attn_q_proj_weight_palettized, x = output_675_cast_fp16)[name = tensor<string, []>("linear_196")];
            tensor<int32, [4]> var_6261 = const()[name = tensor<string, []>("op_6261"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_6262 = reshape(shape = var_6261, x = var_6260)[name = tensor<string, []>("op_6262")];
            tensor<int32, [4]> x_675_perm_0 = const()[name = tensor<string, []>("x_675_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_6265 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_28_self_attn_k_proj_weight_palettized, x = output_675_cast_fp16)[name = tensor<string, []>("linear_197")];
            tensor<int32, [4]> var_6266 = const()[name = tensor<string, []>("op_6266"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_6267 = reshape(shape = var_6266, x = var_6265)[name = tensor<string, []>("op_6267")];
            tensor<int32, [4]> x_679_perm_0 = const()[name = tensor<string, []>("x_679_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_6270 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_28_self_attn_v_proj_weight_palettized, x = output_675_cast_fp16)[name = tensor<string, []>("linear_198")];
            tensor<int32, [4]> var_6271 = const()[name = tensor<string, []>("op_6271"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_6272 = reshape(shape = var_6271, x = var_6270)[name = tensor<string, []>("op_6272")];
            tensor<int32, [4]> hidden_states_399_perm_0 = const()[name = tensor<string, []>("hidden_states_399_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_169_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_169_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_675 = transpose(perm = x_675_perm_0, x = var_6262)[name = tensor<string, []>("transpose_23")];
            tensor<fp16, [1, 8, 256, 256]> var_6276_cast_fp16 = pow(x = x_675, y = var_22_promoted_169_to_fp16)[name = tensor<string, []>("op_6276_cast_fp16")];
            tensor<int32, [1]> var_6278_axes_0 = const()[name = tensor<string, []>("op_6278_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6278_keep_dims_0 = const()[name = tensor<string, []>("op_6278_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_6278_cast_fp16 = reduce_mean(axes = var_6278_axes_0, keep_dims = var_6278_keep_dims_0, x = var_6276_cast_fp16)[name = tensor<string, []>("op_6278_cast_fp16")];
            tensor<fp16, []> var_6279_to_fp16 = const()[name = tensor<string, []>("op_6279_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_6280_cast_fp16 = add(x = var_6278_cast_fp16, y = var_6279_to_fp16)[name = tensor<string, []>("op_6280_cast_fp16")];
            tensor<fp32, []> var_6281_epsilon_0 = const()[name = tensor<string, []>("op_6281_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_6281_cast_fp16 = rsqrt(epsilon = var_6281_epsilon_0, x = var_6280_cast_fp16)[name = tensor<string, []>("op_6281_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_677_cast_fp16 = mul(x = x_675, y = var_6281_cast_fp16)[name = tensor<string, []>("output_677_cast_fp16")];
            tensor<fp16, [256]> var_6285_to_fp16 = const()[name = tensor<string, []>("op_6285_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940808320)))];
            tensor<fp16, [1, 8, 256, 256]> output_679_cast_fp16 = mul(x = output_677_cast_fp16, y = var_6285_to_fp16)[name = tensor<string, []>("output_679_cast_fp16")];
            tensor<fp16, []> var_22_promoted_170_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_170_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_679 = transpose(perm = x_679_perm_0, x = var_6267)[name = tensor<string, []>("transpose_22")];
            tensor<fp16, [1, 4, 256, 256]> var_6290_cast_fp16 = pow(x = x_679, y = var_22_promoted_170_to_fp16)[name = tensor<string, []>("op_6290_cast_fp16")];
            tensor<int32, [1]> var_6292_axes_0 = const()[name = tensor<string, []>("op_6292_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6292_keep_dims_0 = const()[name = tensor<string, []>("op_6292_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_6292_cast_fp16 = reduce_mean(axes = var_6292_axes_0, keep_dims = var_6292_keep_dims_0, x = var_6290_cast_fp16)[name = tensor<string, []>("op_6292_cast_fp16")];
            tensor<fp16, []> var_6293_to_fp16 = const()[name = tensor<string, []>("op_6293_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_6294_cast_fp16 = add(x = var_6292_cast_fp16, y = var_6293_to_fp16)[name = tensor<string, []>("op_6294_cast_fp16")];
            tensor<fp32, []> var_6295_epsilon_0 = const()[name = tensor<string, []>("op_6295_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_6295_cast_fp16 = rsqrt(epsilon = var_6295_epsilon_0, x = var_6294_cast_fp16)[name = tensor<string, []>("op_6295_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_681_cast_fp16 = mul(x = x_679, y = var_6295_cast_fp16)[name = tensor<string, []>("output_681_cast_fp16")];
            tensor<fp16, [256]> var_6299_to_fp16 = const()[name = tensor<string, []>("op_6299_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940808896)))];
            tensor<fp16, [1, 4, 256, 256]> output_683_cast_fp16 = mul(x = output_681_cast_fp16, y = var_6299_to_fp16)[name = tensor<string, []>("output_683_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_6304 = mul(x = output_679_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_6304")];
            tensor<int32, [4]> x1_113_begin_0 = const()[name = tensor<string, []>("x1_113_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_113_end_0 = const()[name = tensor<string, []>("x1_113_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_113_end_mask_0 = const()[name = tensor<string, []>("x1_113_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_113 = slice_by_index(begin = x1_113_begin_0, end = x1_113_end_0, end_mask = x1_113_end_mask_0, x = output_679_cast_fp16)[name = tensor<string, []>("x1_113")];
            tensor<int32, [4]> x2_113_begin_0 = const()[name = tensor<string, []>("x2_113_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_113_end_0 = const()[name = tensor<string, []>("x2_113_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_113_end_mask_0 = const()[name = tensor<string, []>("x2_113_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_113 = slice_by_index(begin = x2_113_begin_0, end = x2_113_end_0, end_mask = x2_113_end_mask_0, x = output_679_cast_fp16)[name = tensor<string, []>("x2_113")];
            tensor<fp16, []> const_679_promoted = const()[name = tensor<string, []>("const_679_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_6315 = mul(x = x2_113, y = const_679_promoted)[name = tensor<string, []>("op_6315")];
            tensor<bool, []> var_6317_interleave_0 = const()[name = tensor<string, []>("op_6317_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_6317 = concat(axis = var_24, interleave = var_6317_interleave_0, values = (var_6315, x1_113))[name = tensor<string, []>("op_6317")];
            tensor<fp16, [1, 8, 256, 256]> var_6318 = mul(x = var_6317, y = sin_7_palettized)[name = tensor<string, []>("op_6318")];
            tensor<fp16, [1, 8, 256, 256]> query_57 = add(x = var_6304, y = var_6318)[name = tensor<string, []>("query_57")];
            tensor<fp16, [1, 4, 256, 256]> var_6320 = mul(x = output_683_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_6320")];
            tensor<int32, [4]> x1_115_begin_0 = const()[name = tensor<string, []>("x1_115_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_115_end_0 = const()[name = tensor<string, []>("x1_115_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_115_end_mask_0 = const()[name = tensor<string, []>("x1_115_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_115 = slice_by_index(begin = x1_115_begin_0, end = x1_115_end_0, end_mask = x1_115_end_mask_0, x = output_683_cast_fp16)[name = tensor<string, []>("x1_115")];
            tensor<int32, [4]> x2_115_begin_0 = const()[name = tensor<string, []>("x2_115_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_115_end_0 = const()[name = tensor<string, []>("x2_115_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_115_end_mask_0 = const()[name = tensor<string, []>("x2_115_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_115 = slice_by_index(begin = x2_115_begin_0, end = x2_115_end_0, end_mask = x2_115_end_mask_0, x = output_683_cast_fp16)[name = tensor<string, []>("x2_115")];
            tensor<fp16, []> const_682_promoted = const()[name = tensor<string, []>("const_682_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_6331 = mul(x = x2_115, y = const_682_promoted)[name = tensor<string, []>("op_6331")];
            tensor<bool, []> var_6333_interleave_0 = const()[name = tensor<string, []>("op_6333_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_6333 = concat(axis = var_24, interleave = var_6333_interleave_0, values = (var_6331, x1_115))[name = tensor<string, []>("op_6333")];
            tensor<fp16, [1, 4, 256, 256]> var_6334 = mul(x = var_6333, y = sin_7_palettized)[name = tensor<string, []>("op_6334")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_395 = add(x = var_6320, y = var_6334)[name = tensor<string, []>("hidden_states_395")];
            tensor<int32, [1]> var_6343_axes_0 = const()[name = tensor<string, []>("op_6343_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_6343 = expand_dims(axes = var_6343_axes_0, x = hidden_states_395)[name = tensor<string, []>("op_6343")];
            tensor<int32, [5]> hidden_states_397_reps_0 = const()[name = tensor<string, []>("hidden_states_397_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_397 = tile(reps = hidden_states_397_reps_0, x = var_6343)[name = tensor<string, []>("hidden_states_397")];
            tensor<int32, [4]> var_6351 = const()[name = tensor<string, []>("op_6351"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_57 = reshape(shape = var_6351, x = hidden_states_397)[name = tensor<string, []>("key_states_57")];
            tensor<int32, [1]> var_6360_axes_0 = const()[name = tensor<string, []>("op_6360_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_399 = transpose(perm = hidden_states_399_perm_0, x = var_6272)[name = tensor<string, []>("transpose_21")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_6360 = expand_dims(axes = var_6360_axes_0, x = hidden_states_399)[name = tensor<string, []>("op_6360")];
            tensor<int32, [5]> hidden_states_401_reps_0 = const()[name = tensor<string, []>("hidden_states_401_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_401 = tile(reps = hidden_states_401_reps_0, x = var_6360)[name = tensor<string, []>("hidden_states_401")];
            tensor<int32, [4]> var_6368 = const()[name = tensor<string, []>("op_6368"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_57 = reshape(shape = var_6368, x = hidden_states_401)[name = tensor<string, []>("value_states_57")];
            tensor<bool, []> var_6371_transpose_x_1 = const()[name = tensor<string, []>("op_6371_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_6371_transpose_y_1 = const()[name = tensor<string, []>("op_6371_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_6371 = matmul(transpose_x = var_6371_transpose_x_1, transpose_y = var_6371_transpose_y_1, x = query_57, y = key_states_57)[name = tensor<string, []>("op_6371")];
            tensor<fp16, []> var_6372_to_fp16 = const()[name = tensor<string, []>("op_6372_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_113_cast_fp16 = mul(x = var_6371, y = var_6372_to_fp16)[name = tensor<string, []>("attn_weights_113_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_337_cast_fp16 = add(x = attn_weights_113_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_337_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_6380_cast_fp16 = softmax(axis = var_24, x = input_337_cast_fp16)[name = tensor<string, []>("op_6380_cast_fp16")];
            tensor<bool, []> attn_output_113_transpose_x_0 = const()[name = tensor<string, []>("attn_output_113_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_113_transpose_y_0 = const()[name = tensor<string, []>("attn_output_113_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_113 = matmul(transpose_x = attn_output_113_transpose_x_0, transpose_y = attn_output_113_transpose_y_0, x = var_6380_cast_fp16, y = value_states_57)[name = tensor<string, []>("attn_output_113")];
            tensor<int32, [4]> var_6384_perm_0 = const()[name = tensor<string, []>("op_6384_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_6386 = const()[name = tensor<string, []>("op_6386"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_6384 = transpose(perm = var_6384_perm_0, x = attn_output_113)[name = tensor<string, []>("transpose_20")];
            tensor<fp16, [1, 256, 2048]> var_6387 = reshape(shape = var_6386, x = var_6384)[name = tensor<string, []>("op_6387")];
            tensor<fp16, [1, 256, 2560]> x_683 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_28_self_attn_o_proj_weight_palettized, x = var_6387)[name = tensor<string, []>("linear_199")];
            tensor<fp16, []> var_22_promoted_171_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_171_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_6393_cast_fp16 = pow(x = x_683, y = var_22_promoted_171_to_fp16)[name = tensor<string, []>("op_6393_cast_fp16")];
            tensor<int32, [1]> var_6395_axes_0 = const()[name = tensor<string, []>("op_6395_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6395_keep_dims_0 = const()[name = tensor<string, []>("op_6395_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_6395_cast_fp16 = reduce_mean(axes = var_6395_axes_0, keep_dims = var_6395_keep_dims_0, x = var_6393_cast_fp16)[name = tensor<string, []>("op_6395_cast_fp16")];
            tensor<fp16, []> var_6396_to_fp16 = const()[name = tensor<string, []>("op_6396_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_6397_cast_fp16 = add(x = var_6395_cast_fp16, y = var_6396_to_fp16)[name = tensor<string, []>("op_6397_cast_fp16")];
            tensor<fp32, []> var_6398_epsilon_0 = const()[name = tensor<string, []>("op_6398_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_6398_cast_fp16 = rsqrt(epsilon = var_6398_epsilon_0, x = var_6397_cast_fp16)[name = tensor<string, []>("op_6398_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_685_cast_fp16 = mul(x = x_683, y = var_6398_cast_fp16)[name = tensor<string, []>("output_685_cast_fp16")];
            tensor<fp16, [2560]> var_6402_to_fp16 = const()[name = tensor<string, []>("op_6402_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940809472)))];
            tensor<fp16, [1, 256, 2560]> output_687_cast_fp16 = mul(x = output_685_cast_fp16, y = var_6402_to_fp16)[name = tensor<string, []>("output_687_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_687 = add(x = x_671, y = output_687_cast_fp16)[name = tensor<string, []>("x_687")];
            tensor<fp16, []> var_22_promoted_172_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_172_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_6408_cast_fp16 = pow(x = x_687, y = var_22_promoted_172_to_fp16)[name = tensor<string, []>("op_6408_cast_fp16")];
            tensor<int32, [1]> var_6410_axes_0 = const()[name = tensor<string, []>("op_6410_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6410_keep_dims_0 = const()[name = tensor<string, []>("op_6410_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_6410_cast_fp16 = reduce_mean(axes = var_6410_axes_0, keep_dims = var_6410_keep_dims_0, x = var_6408_cast_fp16)[name = tensor<string, []>("op_6410_cast_fp16")];
            tensor<fp16, []> var_6411_to_fp16 = const()[name = tensor<string, []>("op_6411_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_6412_cast_fp16 = add(x = var_6410_cast_fp16, y = var_6411_to_fp16)[name = tensor<string, []>("op_6412_cast_fp16")];
            tensor<fp32, []> var_6413_epsilon_0 = const()[name = tensor<string, []>("op_6413_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_6413_cast_fp16 = rsqrt(epsilon = var_6413_epsilon_0, x = var_6412_cast_fp16)[name = tensor<string, []>("op_6413_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_689_cast_fp16 = mul(x = x_687, y = var_6413_cast_fp16)[name = tensor<string, []>("output_689_cast_fp16")];
            tensor<fp16, [2560]> var_6417_to_fp16 = const()[name = tensor<string, []>("op_6417_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940814656)))];
            tensor<fp16, [1, 256, 2560]> output_691_cast_fp16 = mul(x = output_689_cast_fp16, y = var_6417_to_fp16)[name = tensor<string, []>("output_691_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_345 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_28_mlp_gate_proj_weight_palettized, x = output_691_cast_fp16)[name = tensor<string, []>("linear_200")];
            tensor<string, []> var_6425_mode_0 = const()[name = tensor<string, []>("op_6425_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_6425 = gelu(mode = var_6425_mode_0, x = input_345)[name = tensor<string, []>("op_6425")];
            tensor<fp16, [1, 256, 10240]> var_6427 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_28_mlp_up_proj_weight_palettized, x = output_691_cast_fp16)[name = tensor<string, []>("linear_201")];
            tensor<fp16, [1, 256, 10240]> input_347 = mul(x = var_6425, y = var_6427)[name = tensor<string, []>("input_347")];
            tensor<fp16, [1, 256, 2560]> x_691 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_28_mlp_down_proj_weight_palettized, x = input_347)[name = tensor<string, []>("linear_202")];
            tensor<fp16, []> var_22_promoted_173_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_173_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_6433_cast_fp16 = pow(x = x_691, y = var_22_promoted_173_to_fp16)[name = tensor<string, []>("op_6433_cast_fp16")];
            tensor<int32, [1]> var_6435_axes_0 = const()[name = tensor<string, []>("op_6435_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6435_keep_dims_0 = const()[name = tensor<string, []>("op_6435_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_6435_cast_fp16 = reduce_mean(axes = var_6435_axes_0, keep_dims = var_6435_keep_dims_0, x = var_6433_cast_fp16)[name = tensor<string, []>("op_6435_cast_fp16")];
            tensor<fp16, []> var_6436_to_fp16 = const()[name = tensor<string, []>("op_6436_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_6437_cast_fp16 = add(x = var_6435_cast_fp16, y = var_6436_to_fp16)[name = tensor<string, []>("op_6437_cast_fp16")];
            tensor<fp32, []> var_6438_epsilon_0 = const()[name = tensor<string, []>("op_6438_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_6438_cast_fp16 = rsqrt(epsilon = var_6438_epsilon_0, x = var_6437_cast_fp16)[name = tensor<string, []>("op_6438_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_693_cast_fp16 = mul(x = x_691, y = var_6438_cast_fp16)[name = tensor<string, []>("output_693_cast_fp16")];
            tensor<fp16, [2560]> var_6442_to_fp16 = const()[name = tensor<string, []>("op_6442_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940819840)))];
            tensor<fp16, [1, 256, 2560]> output_695_cast_fp16 = mul(x = output_693_cast_fp16, y = var_6442_to_fp16)[name = tensor<string, []>("output_695_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_695 = add(x = x_687, y = output_695_cast_fp16)[name = tensor<string, []>("x_695")];
            tensor<fp16, []> var_22_promoted_174_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_174_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_6454_cast_fp16 = pow(x = x_695, y = var_22_promoted_174_to_fp16)[name = tensor<string, []>("op_6454_cast_fp16")];
            tensor<int32, [1]> var_6456_axes_0 = const()[name = tensor<string, []>("op_6456_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6456_keep_dims_0 = const()[name = tensor<string, []>("op_6456_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_6456_cast_fp16 = reduce_mean(axes = var_6456_axes_0, keep_dims = var_6456_keep_dims_0, x = var_6454_cast_fp16)[name = tensor<string, []>("op_6456_cast_fp16")];
            tensor<fp16, []> var_6457_to_fp16 = const()[name = tensor<string, []>("op_6457_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_6458_cast_fp16 = add(x = var_6456_cast_fp16, y = var_6457_to_fp16)[name = tensor<string, []>("op_6458_cast_fp16")];
            tensor<fp32, []> var_6459_epsilon_0 = const()[name = tensor<string, []>("op_6459_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_6459_cast_fp16 = rsqrt(epsilon = var_6459_epsilon_0, x = var_6458_cast_fp16)[name = tensor<string, []>("op_6459_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_697_cast_fp16 = mul(x = x_695, y = var_6459_cast_fp16)[name = tensor<string, []>("output_697_cast_fp16")];
            tensor<fp16, [2560]> var_6463_to_fp16 = const()[name = tensor<string, []>("op_6463_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940825024)))];
            tensor<fp16, [1, 256, 2560]> output_699_cast_fp16 = mul(x = output_697_cast_fp16, y = var_6463_to_fp16)[name = tensor<string, []>("output_699_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_6475 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_29_self_attn_q_proj_weight_palettized, x = output_699_cast_fp16)[name = tensor<string, []>("linear_203")];
            tensor<int32, [4]> var_6476 = const()[name = tensor<string, []>("op_6476"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_6477 = reshape(shape = var_6476, x = var_6475)[name = tensor<string, []>("op_6477")];
            tensor<int32, [4]> x_699_perm_0 = const()[name = tensor<string, []>("x_699_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_6480 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_29_self_attn_k_proj_weight_palettized, x = output_699_cast_fp16)[name = tensor<string, []>("linear_204")];
            tensor<int32, [4]> var_6481 = const()[name = tensor<string, []>("op_6481"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_6482 = reshape(shape = var_6481, x = var_6480)[name = tensor<string, []>("op_6482")];
            tensor<int32, [4]> x_703_perm_0 = const()[name = tensor<string, []>("x_703_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_6485 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_29_self_attn_v_proj_weight_palettized, x = output_699_cast_fp16)[name = tensor<string, []>("linear_205")];
            tensor<int32, [4]> var_6486 = const()[name = tensor<string, []>("op_6486"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_6487 = reshape(shape = var_6486, x = var_6485)[name = tensor<string, []>("op_6487")];
            tensor<int32, [4]> hidden_states_413_perm_0 = const()[name = tensor<string, []>("hidden_states_413_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_175_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_175_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_699 = transpose(perm = x_699_perm_0, x = var_6477)[name = tensor<string, []>("transpose_19")];
            tensor<fp16, [1, 8, 256, 256]> var_6491_cast_fp16 = pow(x = x_699, y = var_22_promoted_175_to_fp16)[name = tensor<string, []>("op_6491_cast_fp16")];
            tensor<int32, [1]> var_6493_axes_0 = const()[name = tensor<string, []>("op_6493_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6493_keep_dims_0 = const()[name = tensor<string, []>("op_6493_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_6493_cast_fp16 = reduce_mean(axes = var_6493_axes_0, keep_dims = var_6493_keep_dims_0, x = var_6491_cast_fp16)[name = tensor<string, []>("op_6493_cast_fp16")];
            tensor<fp16, []> var_6494_to_fp16 = const()[name = tensor<string, []>("op_6494_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_6495_cast_fp16 = add(x = var_6493_cast_fp16, y = var_6494_to_fp16)[name = tensor<string, []>("op_6495_cast_fp16")];
            tensor<fp32, []> var_6496_epsilon_0 = const()[name = tensor<string, []>("op_6496_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_6496_cast_fp16 = rsqrt(epsilon = var_6496_epsilon_0, x = var_6495_cast_fp16)[name = tensor<string, []>("op_6496_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_701_cast_fp16 = mul(x = x_699, y = var_6496_cast_fp16)[name = tensor<string, []>("output_701_cast_fp16")];
            tensor<fp16, [256]> var_6500_to_fp16 = const()[name = tensor<string, []>("op_6500_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940830208)))];
            tensor<fp16, [1, 8, 256, 256]> output_703_cast_fp16 = mul(x = output_701_cast_fp16, y = var_6500_to_fp16)[name = tensor<string, []>("output_703_cast_fp16")];
            tensor<fp16, []> var_22_promoted_176_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_176_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_703 = transpose(perm = x_703_perm_0, x = var_6482)[name = tensor<string, []>("transpose_18")];
            tensor<fp16, [1, 4, 256, 256]> var_6505_cast_fp16 = pow(x = x_703, y = var_22_promoted_176_to_fp16)[name = tensor<string, []>("op_6505_cast_fp16")];
            tensor<int32, [1]> var_6507_axes_0 = const()[name = tensor<string, []>("op_6507_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6507_keep_dims_0 = const()[name = tensor<string, []>("op_6507_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_6507_cast_fp16 = reduce_mean(axes = var_6507_axes_0, keep_dims = var_6507_keep_dims_0, x = var_6505_cast_fp16)[name = tensor<string, []>("op_6507_cast_fp16")];
            tensor<fp16, []> var_6508_to_fp16 = const()[name = tensor<string, []>("op_6508_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_6509_cast_fp16 = add(x = var_6507_cast_fp16, y = var_6508_to_fp16)[name = tensor<string, []>("op_6509_cast_fp16")];
            tensor<fp32, []> var_6510_epsilon_0 = const()[name = tensor<string, []>("op_6510_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_6510_cast_fp16 = rsqrt(epsilon = var_6510_epsilon_0, x = var_6509_cast_fp16)[name = tensor<string, []>("op_6510_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_705_cast_fp16 = mul(x = x_703, y = var_6510_cast_fp16)[name = tensor<string, []>("output_705_cast_fp16")];
            tensor<fp16, [256]> var_6514_to_fp16 = const()[name = tensor<string, []>("op_6514_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940830784)))];
            tensor<fp16, [1, 4, 256, 256]> output_707_cast_fp16 = mul(x = output_705_cast_fp16, y = var_6514_to_fp16)[name = tensor<string, []>("output_707_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_6519 = mul(x = output_703_cast_fp16, y = cos_19_palettized)[name = tensor<string, []>("op_6519")];
            tensor<int32, [4]> x1_117_begin_0 = const()[name = tensor<string, []>("x1_117_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_117_end_0 = const()[name = tensor<string, []>("x1_117_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_117_end_mask_0 = const()[name = tensor<string, []>("x1_117_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_117 = slice_by_index(begin = x1_117_begin_0, end = x1_117_end_0, end_mask = x1_117_end_mask_0, x = output_703_cast_fp16)[name = tensor<string, []>("x1_117")];
            tensor<int32, [4]> x2_117_begin_0 = const()[name = tensor<string, []>("x2_117_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_117_end_0 = const()[name = tensor<string, []>("x2_117_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_117_end_mask_0 = const()[name = tensor<string, []>("x2_117_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_117 = slice_by_index(begin = x2_117_begin_0, end = x2_117_end_0, end_mask = x2_117_end_mask_0, x = output_703_cast_fp16)[name = tensor<string, []>("x2_117")];
            tensor<fp16, []> const_702_promoted = const()[name = tensor<string, []>("const_702_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_6530 = mul(x = x2_117, y = const_702_promoted)[name = tensor<string, []>("op_6530")];
            tensor<bool, []> var_6532_interleave_0 = const()[name = tensor<string, []>("op_6532_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_6532 = concat(axis = var_24, interleave = var_6532_interleave_0, values = (var_6530, x1_117))[name = tensor<string, []>("op_6532")];
            tensor<fp16, [1, 8, 256, 256]> var_6533 = mul(x = var_6532, y = sin_19_palettized)[name = tensor<string, []>("op_6533")];
            tensor<fp16, [1, 8, 256, 256]> query_59 = add(x = var_6519, y = var_6533)[name = tensor<string, []>("query_59")];
            tensor<fp16, [1, 4, 256, 256]> var_6535 = mul(x = output_707_cast_fp16, y = cos_19_palettized)[name = tensor<string, []>("op_6535")];
            tensor<int32, [4]> x1_119_begin_0 = const()[name = tensor<string, []>("x1_119_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_119_end_0 = const()[name = tensor<string, []>("x1_119_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_119_end_mask_0 = const()[name = tensor<string, []>("x1_119_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_119 = slice_by_index(begin = x1_119_begin_0, end = x1_119_end_0, end_mask = x1_119_end_mask_0, x = output_707_cast_fp16)[name = tensor<string, []>("x1_119")];
            tensor<int32, [4]> x2_119_begin_0 = const()[name = tensor<string, []>("x2_119_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_119_end_0 = const()[name = tensor<string, []>("x2_119_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_119_end_mask_0 = const()[name = tensor<string, []>("x2_119_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_119 = slice_by_index(begin = x2_119_begin_0, end = x2_119_end_0, end_mask = x2_119_end_mask_0, x = output_707_cast_fp16)[name = tensor<string, []>("x2_119")];
            tensor<fp16, []> const_705_promoted = const()[name = tensor<string, []>("const_705_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_6546 = mul(x = x2_119, y = const_705_promoted)[name = tensor<string, []>("op_6546")];
            tensor<bool, []> var_6548_interleave_0 = const()[name = tensor<string, []>("op_6548_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_6548 = concat(axis = var_24, interleave = var_6548_interleave_0, values = (var_6546, x1_119))[name = tensor<string, []>("op_6548")];
            tensor<fp16, [1, 4, 256, 256]> var_6549 = mul(x = var_6548, y = sin_19_palettized)[name = tensor<string, []>("op_6549")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_409 = add(x = var_6535, y = var_6549)[name = tensor<string, []>("hidden_states_409")];
            tensor<int32, [1]> var_6558_axes_0 = const()[name = tensor<string, []>("op_6558_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_6558 = expand_dims(axes = var_6558_axes_0, x = hidden_states_409)[name = tensor<string, []>("op_6558")];
            tensor<int32, [5]> hidden_states_411_reps_0 = const()[name = tensor<string, []>("hidden_states_411_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_411 = tile(reps = hidden_states_411_reps_0, x = var_6558)[name = tensor<string, []>("hidden_states_411")];
            tensor<int32, [4]> var_6566 = const()[name = tensor<string, []>("op_6566"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_59 = reshape(shape = var_6566, x = hidden_states_411)[name = tensor<string, []>("key_states_59")];
            tensor<int32, [1]> var_6575_axes_0 = const()[name = tensor<string, []>("op_6575_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_413 = transpose(perm = hidden_states_413_perm_0, x = var_6487)[name = tensor<string, []>("transpose_17")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_6575 = expand_dims(axes = var_6575_axes_0, x = hidden_states_413)[name = tensor<string, []>("op_6575")];
            tensor<int32, [5]> hidden_states_415_reps_0 = const()[name = tensor<string, []>("hidden_states_415_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_415 = tile(reps = hidden_states_415_reps_0, x = var_6575)[name = tensor<string, []>("hidden_states_415")];
            tensor<int32, [4]> var_6583 = const()[name = tensor<string, []>("op_6583"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_59 = reshape(shape = var_6583, x = hidden_states_415)[name = tensor<string, []>("value_states_59")];
            tensor<bool, []> var_6586_transpose_x_1 = const()[name = tensor<string, []>("op_6586_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_6586_transpose_y_1 = const()[name = tensor<string, []>("op_6586_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_6586 = matmul(transpose_x = var_6586_transpose_x_1, transpose_y = var_6586_transpose_y_1, x = query_59, y = key_states_59)[name = tensor<string, []>("op_6586")];
            tensor<fp16, []> var_6587_to_fp16 = const()[name = tensor<string, []>("op_6587_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_117_cast_fp16 = mul(x = var_6586, y = var_6587_to_fp16)[name = tensor<string, []>("attn_weights_117_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_349_cast_fp16 = add(x = attn_weights_117_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_349_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_6595_cast_fp16 = softmax(axis = var_24, x = input_349_cast_fp16)[name = tensor<string, []>("op_6595_cast_fp16")];
            tensor<bool, []> attn_output_117_transpose_x_0 = const()[name = tensor<string, []>("attn_output_117_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_117_transpose_y_0 = const()[name = tensor<string, []>("attn_output_117_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_117 = matmul(transpose_x = attn_output_117_transpose_x_0, transpose_y = attn_output_117_transpose_y_0, x = var_6595_cast_fp16, y = value_states_59)[name = tensor<string, []>("attn_output_117")];
            tensor<int32, [4]> var_6599_perm_0 = const()[name = tensor<string, []>("op_6599_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_6601 = const()[name = tensor<string, []>("op_6601"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_6599 = transpose(perm = var_6599_perm_0, x = attn_output_117)[name = tensor<string, []>("transpose_16")];
            tensor<fp16, [1, 256, 2048]> var_6602 = reshape(shape = var_6601, x = var_6599)[name = tensor<string, []>("op_6602")];
            tensor<fp16, [1, 256, 2560]> x_707 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_29_self_attn_o_proj_weight_palettized, x = var_6602)[name = tensor<string, []>("linear_206")];
            tensor<fp16, []> var_22_promoted_177_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_177_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_6608_cast_fp16 = pow(x = x_707, y = var_22_promoted_177_to_fp16)[name = tensor<string, []>("op_6608_cast_fp16")];
            tensor<int32, [1]> var_6610_axes_0 = const()[name = tensor<string, []>("op_6610_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6610_keep_dims_0 = const()[name = tensor<string, []>("op_6610_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_6610_cast_fp16 = reduce_mean(axes = var_6610_axes_0, keep_dims = var_6610_keep_dims_0, x = var_6608_cast_fp16)[name = tensor<string, []>("op_6610_cast_fp16")];
            tensor<fp16, []> var_6611_to_fp16 = const()[name = tensor<string, []>("op_6611_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_6612_cast_fp16 = add(x = var_6610_cast_fp16, y = var_6611_to_fp16)[name = tensor<string, []>("op_6612_cast_fp16")];
            tensor<fp32, []> var_6613_epsilon_0 = const()[name = tensor<string, []>("op_6613_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_6613_cast_fp16 = rsqrt(epsilon = var_6613_epsilon_0, x = var_6612_cast_fp16)[name = tensor<string, []>("op_6613_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_709_cast_fp16 = mul(x = x_707, y = var_6613_cast_fp16)[name = tensor<string, []>("output_709_cast_fp16")];
            tensor<fp16, [2560]> var_6617_to_fp16 = const()[name = tensor<string, []>("op_6617_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940831360)))];
            tensor<fp16, [1, 256, 2560]> output_711_cast_fp16 = mul(x = output_709_cast_fp16, y = var_6617_to_fp16)[name = tensor<string, []>("output_711_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_711 = add(x = x_695, y = output_711_cast_fp16)[name = tensor<string, []>("x_711")];
            tensor<fp16, []> var_22_promoted_178_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_178_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_6623_cast_fp16 = pow(x = x_711, y = var_22_promoted_178_to_fp16)[name = tensor<string, []>("op_6623_cast_fp16")];
            tensor<int32, [1]> var_6625_axes_0 = const()[name = tensor<string, []>("op_6625_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6625_keep_dims_0 = const()[name = tensor<string, []>("op_6625_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_6625_cast_fp16 = reduce_mean(axes = var_6625_axes_0, keep_dims = var_6625_keep_dims_0, x = var_6623_cast_fp16)[name = tensor<string, []>("op_6625_cast_fp16")];
            tensor<fp16, []> var_6626_to_fp16 = const()[name = tensor<string, []>("op_6626_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_6627_cast_fp16 = add(x = var_6625_cast_fp16, y = var_6626_to_fp16)[name = tensor<string, []>("op_6627_cast_fp16")];
            tensor<fp32, []> var_6628_epsilon_0 = const()[name = tensor<string, []>("op_6628_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_6628_cast_fp16 = rsqrt(epsilon = var_6628_epsilon_0, x = var_6627_cast_fp16)[name = tensor<string, []>("op_6628_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_713_cast_fp16 = mul(x = x_711, y = var_6628_cast_fp16)[name = tensor<string, []>("output_713_cast_fp16")];
            tensor<fp16, [2560]> var_6632_to_fp16 = const()[name = tensor<string, []>("op_6632_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940836544)))];
            tensor<fp16, [1, 256, 2560]> output_715_cast_fp16 = mul(x = output_713_cast_fp16, y = var_6632_to_fp16)[name = tensor<string, []>("output_715_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_357 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_29_mlp_gate_proj_weight_palettized, x = output_715_cast_fp16)[name = tensor<string, []>("linear_207")];
            tensor<string, []> var_6640_mode_0 = const()[name = tensor<string, []>("op_6640_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_6640 = gelu(mode = var_6640_mode_0, x = input_357)[name = tensor<string, []>("op_6640")];
            tensor<fp16, [1, 256, 10240]> var_6642 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_29_mlp_up_proj_weight_palettized, x = output_715_cast_fp16)[name = tensor<string, []>("linear_208")];
            tensor<fp16, [1, 256, 10240]> input_359 = mul(x = var_6640, y = var_6642)[name = tensor<string, []>("input_359")];
            tensor<fp16, [1, 256, 2560]> x_715 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_29_mlp_down_proj_weight_palettized, x = input_359)[name = tensor<string, []>("linear_209")];
            tensor<fp16, []> var_22_promoted_179_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_179_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_6648_cast_fp16 = pow(x = x_715, y = var_22_promoted_179_to_fp16)[name = tensor<string, []>("op_6648_cast_fp16")];
            tensor<int32, [1]> var_6650_axes_0 = const()[name = tensor<string, []>("op_6650_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6650_keep_dims_0 = const()[name = tensor<string, []>("op_6650_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_6650_cast_fp16 = reduce_mean(axes = var_6650_axes_0, keep_dims = var_6650_keep_dims_0, x = var_6648_cast_fp16)[name = tensor<string, []>("op_6650_cast_fp16")];
            tensor<fp16, []> var_6651_to_fp16 = const()[name = tensor<string, []>("op_6651_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_6652_cast_fp16 = add(x = var_6650_cast_fp16, y = var_6651_to_fp16)[name = tensor<string, []>("op_6652_cast_fp16")];
            tensor<fp32, []> var_6653_epsilon_0 = const()[name = tensor<string, []>("op_6653_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_6653_cast_fp16 = rsqrt(epsilon = var_6653_epsilon_0, x = var_6652_cast_fp16)[name = tensor<string, []>("op_6653_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_717_cast_fp16 = mul(x = x_715, y = var_6653_cast_fp16)[name = tensor<string, []>("output_717_cast_fp16")];
            tensor<fp16, [2560]> var_6657_to_fp16 = const()[name = tensor<string, []>("op_6657_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940841728)))];
            tensor<fp16, [1, 256, 2560]> output_719_cast_fp16 = mul(x = output_717_cast_fp16, y = var_6657_to_fp16)[name = tensor<string, []>("output_719_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_719 = add(x = x_711, y = output_719_cast_fp16)[name = tensor<string, []>("x_719")];
            tensor<fp16, []> var_22_promoted_180_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_180_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_6669_cast_fp16 = pow(x = x_719, y = var_22_promoted_180_to_fp16)[name = tensor<string, []>("op_6669_cast_fp16")];
            tensor<int32, [1]> var_6671_axes_0 = const()[name = tensor<string, []>("op_6671_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6671_keep_dims_0 = const()[name = tensor<string, []>("op_6671_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_6671_cast_fp16 = reduce_mean(axes = var_6671_axes_0, keep_dims = var_6671_keep_dims_0, x = var_6669_cast_fp16)[name = tensor<string, []>("op_6671_cast_fp16")];
            tensor<fp16, []> var_6672_to_fp16 = const()[name = tensor<string, []>("op_6672_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_6673_cast_fp16 = add(x = var_6671_cast_fp16, y = var_6672_to_fp16)[name = tensor<string, []>("op_6673_cast_fp16")];
            tensor<fp32, []> var_6674_epsilon_0 = const()[name = tensor<string, []>("op_6674_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_6674_cast_fp16 = rsqrt(epsilon = var_6674_epsilon_0, x = var_6673_cast_fp16)[name = tensor<string, []>("op_6674_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_721_cast_fp16 = mul(x = x_719, y = var_6674_cast_fp16)[name = tensor<string, []>("output_721_cast_fp16")];
            tensor<fp16, [2560]> var_6678_to_fp16 = const()[name = tensor<string, []>("op_6678_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940846912)))];
            tensor<fp16, [1, 256, 2560]> output_723_cast_fp16 = mul(x = output_721_cast_fp16, y = var_6678_to_fp16)[name = tensor<string, []>("output_723_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_6690 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_30_self_attn_q_proj_weight_palettized, x = output_723_cast_fp16)[name = tensor<string, []>("linear_210")];
            tensor<int32, [4]> var_6691 = const()[name = tensor<string, []>("op_6691"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_6692 = reshape(shape = var_6691, x = var_6690)[name = tensor<string, []>("op_6692")];
            tensor<int32, [4]> x_723_perm_0 = const()[name = tensor<string, []>("x_723_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_6695 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_30_self_attn_k_proj_weight_palettized, x = output_723_cast_fp16)[name = tensor<string, []>("linear_211")];
            tensor<int32, [4]> var_6696 = const()[name = tensor<string, []>("op_6696"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_6697 = reshape(shape = var_6696, x = var_6695)[name = tensor<string, []>("op_6697")];
            tensor<int32, [4]> x_727_perm_0 = const()[name = tensor<string, []>("x_727_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_6700 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_30_self_attn_v_proj_weight_palettized, x = output_723_cast_fp16)[name = tensor<string, []>("linear_212")];
            tensor<int32, [4]> var_6701 = const()[name = tensor<string, []>("op_6701"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_6702 = reshape(shape = var_6701, x = var_6700)[name = tensor<string, []>("op_6702")];
            tensor<int32, [4]> hidden_states_427_perm_0 = const()[name = tensor<string, []>("hidden_states_427_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_181_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_181_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_723 = transpose(perm = x_723_perm_0, x = var_6692)[name = tensor<string, []>("transpose_15")];
            tensor<fp16, [1, 8, 256, 256]> var_6706_cast_fp16 = pow(x = x_723, y = var_22_promoted_181_to_fp16)[name = tensor<string, []>("op_6706_cast_fp16")];
            tensor<int32, [1]> var_6708_axes_0 = const()[name = tensor<string, []>("op_6708_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6708_keep_dims_0 = const()[name = tensor<string, []>("op_6708_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_6708_cast_fp16 = reduce_mean(axes = var_6708_axes_0, keep_dims = var_6708_keep_dims_0, x = var_6706_cast_fp16)[name = tensor<string, []>("op_6708_cast_fp16")];
            tensor<fp16, []> var_6709_to_fp16 = const()[name = tensor<string, []>("op_6709_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_6710_cast_fp16 = add(x = var_6708_cast_fp16, y = var_6709_to_fp16)[name = tensor<string, []>("op_6710_cast_fp16")];
            tensor<fp32, []> var_6711_epsilon_0 = const()[name = tensor<string, []>("op_6711_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_6711_cast_fp16 = rsqrt(epsilon = var_6711_epsilon_0, x = var_6710_cast_fp16)[name = tensor<string, []>("op_6711_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_725_cast_fp16 = mul(x = x_723, y = var_6711_cast_fp16)[name = tensor<string, []>("output_725_cast_fp16")];
            tensor<fp16, [256]> var_6715_to_fp16 = const()[name = tensor<string, []>("op_6715_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940852096)))];
            tensor<fp16, [1, 8, 256, 256]> output_727_cast_fp16 = mul(x = output_725_cast_fp16, y = var_6715_to_fp16)[name = tensor<string, []>("output_727_cast_fp16")];
            tensor<fp16, []> var_22_promoted_182_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_182_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_727 = transpose(perm = x_727_perm_0, x = var_6697)[name = tensor<string, []>("transpose_14")];
            tensor<fp16, [1, 4, 256, 256]> var_6720_cast_fp16 = pow(x = x_727, y = var_22_promoted_182_to_fp16)[name = tensor<string, []>("op_6720_cast_fp16")];
            tensor<int32, [1]> var_6722_axes_0 = const()[name = tensor<string, []>("op_6722_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6722_keep_dims_0 = const()[name = tensor<string, []>("op_6722_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_6722_cast_fp16 = reduce_mean(axes = var_6722_axes_0, keep_dims = var_6722_keep_dims_0, x = var_6720_cast_fp16)[name = tensor<string, []>("op_6722_cast_fp16")];
            tensor<fp16, []> var_6723_to_fp16 = const()[name = tensor<string, []>("op_6723_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_6724_cast_fp16 = add(x = var_6722_cast_fp16, y = var_6723_to_fp16)[name = tensor<string, []>("op_6724_cast_fp16")];
            tensor<fp32, []> var_6725_epsilon_0 = const()[name = tensor<string, []>("op_6725_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_6725_cast_fp16 = rsqrt(epsilon = var_6725_epsilon_0, x = var_6724_cast_fp16)[name = tensor<string, []>("op_6725_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_729_cast_fp16 = mul(x = x_727, y = var_6725_cast_fp16)[name = tensor<string, []>("output_729_cast_fp16")];
            tensor<fp16, [256]> var_6729_to_fp16 = const()[name = tensor<string, []>("op_6729_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940852672)))];
            tensor<fp16, [1, 4, 256, 256]> output_731_cast_fp16 = mul(x = output_729_cast_fp16, y = var_6729_to_fp16)[name = tensor<string, []>("output_731_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_6734 = mul(x = output_727_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_6734")];
            tensor<int32, [4]> x1_121_begin_0 = const()[name = tensor<string, []>("x1_121_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_121_end_0 = const()[name = tensor<string, []>("x1_121_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_121_end_mask_0 = const()[name = tensor<string, []>("x1_121_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_121 = slice_by_index(begin = x1_121_begin_0, end = x1_121_end_0, end_mask = x1_121_end_mask_0, x = output_727_cast_fp16)[name = tensor<string, []>("x1_121")];
            tensor<int32, [4]> x2_121_begin_0 = const()[name = tensor<string, []>("x2_121_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_121_end_0 = const()[name = tensor<string, []>("x2_121_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_121_end_mask_0 = const()[name = tensor<string, []>("x2_121_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_121 = slice_by_index(begin = x2_121_begin_0, end = x2_121_end_0, end_mask = x2_121_end_mask_0, x = output_727_cast_fp16)[name = tensor<string, []>("x2_121")];
            tensor<fp16, []> const_725_promoted = const()[name = tensor<string, []>("const_725_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_6745 = mul(x = x2_121, y = const_725_promoted)[name = tensor<string, []>("op_6745")];
            tensor<bool, []> var_6747_interleave_0 = const()[name = tensor<string, []>("op_6747_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_6747 = concat(axis = var_24, interleave = var_6747_interleave_0, values = (var_6745, x1_121))[name = tensor<string, []>("op_6747")];
            tensor<fp16, [1, 8, 256, 256]> var_6748 = mul(x = var_6747, y = sin_7_palettized)[name = tensor<string, []>("op_6748")];
            tensor<fp16, [1, 8, 256, 256]> query_61 = add(x = var_6734, y = var_6748)[name = tensor<string, []>("query_61")];
            tensor<fp16, [1, 4, 256, 256]> var_6750 = mul(x = output_731_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_6750")];
            tensor<int32, [4]> x1_123_begin_0 = const()[name = tensor<string, []>("x1_123_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_123_end_0 = const()[name = tensor<string, []>("x1_123_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_123_end_mask_0 = const()[name = tensor<string, []>("x1_123_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_123 = slice_by_index(begin = x1_123_begin_0, end = x1_123_end_0, end_mask = x1_123_end_mask_0, x = output_731_cast_fp16)[name = tensor<string, []>("x1_123")];
            tensor<int32, [4]> x2_123_begin_0 = const()[name = tensor<string, []>("x2_123_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_123_end_0 = const()[name = tensor<string, []>("x2_123_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_123_end_mask_0 = const()[name = tensor<string, []>("x2_123_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_123 = slice_by_index(begin = x2_123_begin_0, end = x2_123_end_0, end_mask = x2_123_end_mask_0, x = output_731_cast_fp16)[name = tensor<string, []>("x2_123")];
            tensor<fp16, []> const_728_promoted = const()[name = tensor<string, []>("const_728_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_6761 = mul(x = x2_123, y = const_728_promoted)[name = tensor<string, []>("op_6761")];
            tensor<bool, []> var_6763_interleave_0 = const()[name = tensor<string, []>("op_6763_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_6763 = concat(axis = var_24, interleave = var_6763_interleave_0, values = (var_6761, x1_123))[name = tensor<string, []>("op_6763")];
            tensor<fp16, [1, 4, 256, 256]> var_6764 = mul(x = var_6763, y = sin_7_palettized)[name = tensor<string, []>("op_6764")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_423 = add(x = var_6750, y = var_6764)[name = tensor<string, []>("hidden_states_423")];
            tensor<int32, [1]> var_6773_axes_0 = const()[name = tensor<string, []>("op_6773_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_6773 = expand_dims(axes = var_6773_axes_0, x = hidden_states_423)[name = tensor<string, []>("op_6773")];
            tensor<int32, [5]> hidden_states_425_reps_0 = const()[name = tensor<string, []>("hidden_states_425_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_425 = tile(reps = hidden_states_425_reps_0, x = var_6773)[name = tensor<string, []>("hidden_states_425")];
            tensor<int32, [4]> var_6781 = const()[name = tensor<string, []>("op_6781"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_61 = reshape(shape = var_6781, x = hidden_states_425)[name = tensor<string, []>("key_states_61")];
            tensor<int32, [1]> var_6790_axes_0 = const()[name = tensor<string, []>("op_6790_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_427 = transpose(perm = hidden_states_427_perm_0, x = var_6702)[name = tensor<string, []>("transpose_13")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_6790 = expand_dims(axes = var_6790_axes_0, x = hidden_states_427)[name = tensor<string, []>("op_6790")];
            tensor<int32, [5]> hidden_states_429_reps_0 = const()[name = tensor<string, []>("hidden_states_429_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_429 = tile(reps = hidden_states_429_reps_0, x = var_6790)[name = tensor<string, []>("hidden_states_429")];
            tensor<int32, [4]> var_6798 = const()[name = tensor<string, []>("op_6798"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_61 = reshape(shape = var_6798, x = hidden_states_429)[name = tensor<string, []>("value_states_61")];
            tensor<bool, []> var_6801_transpose_x_1 = const()[name = tensor<string, []>("op_6801_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_6801_transpose_y_1 = const()[name = tensor<string, []>("op_6801_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_6801 = matmul(transpose_x = var_6801_transpose_x_1, transpose_y = var_6801_transpose_y_1, x = query_61, y = key_states_61)[name = tensor<string, []>("op_6801")];
            tensor<fp16, []> var_6802_to_fp16 = const()[name = tensor<string, []>("op_6802_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_121_cast_fp16 = mul(x = var_6801, y = var_6802_to_fp16)[name = tensor<string, []>("attn_weights_121_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_361_cast_fp16 = add(x = attn_weights_121_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_361_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_6810_cast_fp16 = softmax(axis = var_24, x = input_361_cast_fp16)[name = tensor<string, []>("op_6810_cast_fp16")];
            tensor<bool, []> attn_output_121_transpose_x_0 = const()[name = tensor<string, []>("attn_output_121_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_121_transpose_y_0 = const()[name = tensor<string, []>("attn_output_121_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_121 = matmul(transpose_x = attn_output_121_transpose_x_0, transpose_y = attn_output_121_transpose_y_0, x = var_6810_cast_fp16, y = value_states_61)[name = tensor<string, []>("attn_output_121")];
            tensor<int32, [4]> var_6814_perm_0 = const()[name = tensor<string, []>("op_6814_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_6816 = const()[name = tensor<string, []>("op_6816"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_6814 = transpose(perm = var_6814_perm_0, x = attn_output_121)[name = tensor<string, []>("transpose_12")];
            tensor<fp16, [1, 256, 2048]> var_6817 = reshape(shape = var_6816, x = var_6814)[name = tensor<string, []>("op_6817")];
            tensor<fp16, [1, 256, 2560]> x_731 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_30_self_attn_o_proj_weight_palettized, x = var_6817)[name = tensor<string, []>("linear_213")];
            tensor<fp16, []> var_22_promoted_183_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_183_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_6823_cast_fp16 = pow(x = x_731, y = var_22_promoted_183_to_fp16)[name = tensor<string, []>("op_6823_cast_fp16")];
            tensor<int32, [1]> var_6825_axes_0 = const()[name = tensor<string, []>("op_6825_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6825_keep_dims_0 = const()[name = tensor<string, []>("op_6825_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_6825_cast_fp16 = reduce_mean(axes = var_6825_axes_0, keep_dims = var_6825_keep_dims_0, x = var_6823_cast_fp16)[name = tensor<string, []>("op_6825_cast_fp16")];
            tensor<fp16, []> var_6826_to_fp16 = const()[name = tensor<string, []>("op_6826_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_6827_cast_fp16 = add(x = var_6825_cast_fp16, y = var_6826_to_fp16)[name = tensor<string, []>("op_6827_cast_fp16")];
            tensor<fp32, []> var_6828_epsilon_0 = const()[name = tensor<string, []>("op_6828_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_6828_cast_fp16 = rsqrt(epsilon = var_6828_epsilon_0, x = var_6827_cast_fp16)[name = tensor<string, []>("op_6828_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_733_cast_fp16 = mul(x = x_731, y = var_6828_cast_fp16)[name = tensor<string, []>("output_733_cast_fp16")];
            tensor<fp16, [2560]> var_6832_to_fp16 = const()[name = tensor<string, []>("op_6832_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940853248)))];
            tensor<fp16, [1, 256, 2560]> output_735_cast_fp16 = mul(x = output_733_cast_fp16, y = var_6832_to_fp16)[name = tensor<string, []>("output_735_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_735 = add(x = x_719, y = output_735_cast_fp16)[name = tensor<string, []>("x_735")];
            tensor<fp16, []> var_22_promoted_184_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_184_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_6838_cast_fp16 = pow(x = x_735, y = var_22_promoted_184_to_fp16)[name = tensor<string, []>("op_6838_cast_fp16")];
            tensor<int32, [1]> var_6840_axes_0 = const()[name = tensor<string, []>("op_6840_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6840_keep_dims_0 = const()[name = tensor<string, []>("op_6840_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_6840_cast_fp16 = reduce_mean(axes = var_6840_axes_0, keep_dims = var_6840_keep_dims_0, x = var_6838_cast_fp16)[name = tensor<string, []>("op_6840_cast_fp16")];
            tensor<fp16, []> var_6841_to_fp16 = const()[name = tensor<string, []>("op_6841_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_6842_cast_fp16 = add(x = var_6840_cast_fp16, y = var_6841_to_fp16)[name = tensor<string, []>("op_6842_cast_fp16")];
            tensor<fp32, []> var_6843_epsilon_0 = const()[name = tensor<string, []>("op_6843_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_6843_cast_fp16 = rsqrt(epsilon = var_6843_epsilon_0, x = var_6842_cast_fp16)[name = tensor<string, []>("op_6843_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_737_cast_fp16 = mul(x = x_735, y = var_6843_cast_fp16)[name = tensor<string, []>("output_737_cast_fp16")];
            tensor<fp16, [2560]> var_6847_to_fp16 = const()[name = tensor<string, []>("op_6847_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940858432)))];
            tensor<fp16, [1, 256, 2560]> output_739_cast_fp16 = mul(x = output_737_cast_fp16, y = var_6847_to_fp16)[name = tensor<string, []>("output_739_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_369 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_30_mlp_gate_proj_weight_palettized, x = output_739_cast_fp16)[name = tensor<string, []>("linear_214")];
            tensor<string, []> var_6855_mode_0 = const()[name = tensor<string, []>("op_6855_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_6855 = gelu(mode = var_6855_mode_0, x = input_369)[name = tensor<string, []>("op_6855")];
            tensor<fp16, [1, 256, 10240]> var_6857 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_30_mlp_up_proj_weight_palettized, x = output_739_cast_fp16)[name = tensor<string, []>("linear_215")];
            tensor<fp16, [1, 256, 10240]> input_371 = mul(x = var_6855, y = var_6857)[name = tensor<string, []>("input_371")];
            tensor<fp16, [1, 256, 2560]> x_739 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_30_mlp_down_proj_weight_palettized, x = input_371)[name = tensor<string, []>("linear_216")];
            tensor<fp16, []> var_22_promoted_185_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_185_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_6863_cast_fp16 = pow(x = x_739, y = var_22_promoted_185_to_fp16)[name = tensor<string, []>("op_6863_cast_fp16")];
            tensor<int32, [1]> var_6865_axes_0 = const()[name = tensor<string, []>("op_6865_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6865_keep_dims_0 = const()[name = tensor<string, []>("op_6865_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_6865_cast_fp16 = reduce_mean(axes = var_6865_axes_0, keep_dims = var_6865_keep_dims_0, x = var_6863_cast_fp16)[name = tensor<string, []>("op_6865_cast_fp16")];
            tensor<fp16, []> var_6866_to_fp16 = const()[name = tensor<string, []>("op_6866_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_6867_cast_fp16 = add(x = var_6865_cast_fp16, y = var_6866_to_fp16)[name = tensor<string, []>("op_6867_cast_fp16")];
            tensor<fp32, []> var_6868_epsilon_0 = const()[name = tensor<string, []>("op_6868_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_6868_cast_fp16 = rsqrt(epsilon = var_6868_epsilon_0, x = var_6867_cast_fp16)[name = tensor<string, []>("op_6868_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_741_cast_fp16 = mul(x = x_739, y = var_6868_cast_fp16)[name = tensor<string, []>("output_741_cast_fp16")];
            tensor<fp16, [2560]> var_6872_to_fp16 = const()[name = tensor<string, []>("op_6872_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940863616)))];
            tensor<fp16, [1, 256, 2560]> output_743_cast_fp16 = mul(x = output_741_cast_fp16, y = var_6872_to_fp16)[name = tensor<string, []>("output_743_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_743 = add(x = x_735, y = output_743_cast_fp16)[name = tensor<string, []>("x_743")];
            tensor<fp16, []> var_22_promoted_186_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_186_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_6884_cast_fp16 = pow(x = x_743, y = var_22_promoted_186_to_fp16)[name = tensor<string, []>("op_6884_cast_fp16")];
            tensor<int32, [1]> var_6886_axes_0 = const()[name = tensor<string, []>("op_6886_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6886_keep_dims_0 = const()[name = tensor<string, []>("op_6886_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_6886_cast_fp16 = reduce_mean(axes = var_6886_axes_0, keep_dims = var_6886_keep_dims_0, x = var_6884_cast_fp16)[name = tensor<string, []>("op_6886_cast_fp16")];
            tensor<fp16, []> var_6887_to_fp16 = const()[name = tensor<string, []>("op_6887_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_6888_cast_fp16 = add(x = var_6886_cast_fp16, y = var_6887_to_fp16)[name = tensor<string, []>("op_6888_cast_fp16")];
            tensor<fp32, []> var_6889_epsilon_0 = const()[name = tensor<string, []>("op_6889_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_6889_cast_fp16 = rsqrt(epsilon = var_6889_epsilon_0, x = var_6888_cast_fp16)[name = tensor<string, []>("op_6889_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_745_cast_fp16 = mul(x = x_743, y = var_6889_cast_fp16)[name = tensor<string, []>("output_745_cast_fp16")];
            tensor<fp16, [2560]> var_6893_to_fp16 = const()[name = tensor<string, []>("op_6893_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940868800)))];
            tensor<fp16, [1, 256, 2560]> output_747_cast_fp16 = mul(x = output_745_cast_fp16, y = var_6893_to_fp16)[name = tensor<string, []>("output_747_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_6905 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_31_self_attn_q_proj_weight_palettized, x = output_747_cast_fp16)[name = tensor<string, []>("linear_217")];
            tensor<int32, [4]> var_6906 = const()[name = tensor<string, []>("op_6906"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_6907 = reshape(shape = var_6906, x = var_6905)[name = tensor<string, []>("op_6907")];
            tensor<int32, [4]> x_747_perm_0 = const()[name = tensor<string, []>("x_747_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_6910 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_31_self_attn_k_proj_weight_palettized, x = output_747_cast_fp16)[name = tensor<string, []>("linear_218")];
            tensor<int32, [4]> var_6911 = const()[name = tensor<string, []>("op_6911"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_6912 = reshape(shape = var_6911, x = var_6910)[name = tensor<string, []>("op_6912")];
            tensor<int32, [4]> x_751_perm_0 = const()[name = tensor<string, []>("x_751_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_6915 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_31_self_attn_v_proj_weight_palettized, x = output_747_cast_fp16)[name = tensor<string, []>("linear_219")];
            tensor<int32, [4]> var_6916 = const()[name = tensor<string, []>("op_6916"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_6917 = reshape(shape = var_6916, x = var_6915)[name = tensor<string, []>("op_6917")];
            tensor<int32, [4]> hidden_states_441_perm_0 = const()[name = tensor<string, []>("hidden_states_441_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_187_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_187_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_747 = transpose(perm = x_747_perm_0, x = var_6907)[name = tensor<string, []>("transpose_11")];
            tensor<fp16, [1, 8, 256, 256]> var_6921_cast_fp16 = pow(x = x_747, y = var_22_promoted_187_to_fp16)[name = tensor<string, []>("op_6921_cast_fp16")];
            tensor<int32, [1]> var_6923_axes_0 = const()[name = tensor<string, []>("op_6923_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6923_keep_dims_0 = const()[name = tensor<string, []>("op_6923_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_6923_cast_fp16 = reduce_mean(axes = var_6923_axes_0, keep_dims = var_6923_keep_dims_0, x = var_6921_cast_fp16)[name = tensor<string, []>("op_6923_cast_fp16")];
            tensor<fp16, []> var_6924_to_fp16 = const()[name = tensor<string, []>("op_6924_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_6925_cast_fp16 = add(x = var_6923_cast_fp16, y = var_6924_to_fp16)[name = tensor<string, []>("op_6925_cast_fp16")];
            tensor<fp32, []> var_6926_epsilon_0 = const()[name = tensor<string, []>("op_6926_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_6926_cast_fp16 = rsqrt(epsilon = var_6926_epsilon_0, x = var_6925_cast_fp16)[name = tensor<string, []>("op_6926_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_749_cast_fp16 = mul(x = x_747, y = var_6926_cast_fp16)[name = tensor<string, []>("output_749_cast_fp16")];
            tensor<fp16, [256]> var_6930_to_fp16 = const()[name = tensor<string, []>("op_6930_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940873984)))];
            tensor<fp16, [1, 8, 256, 256]> output_751_cast_fp16 = mul(x = output_749_cast_fp16, y = var_6930_to_fp16)[name = tensor<string, []>("output_751_cast_fp16")];
            tensor<fp16, []> var_22_promoted_188_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_188_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_751 = transpose(perm = x_751_perm_0, x = var_6912)[name = tensor<string, []>("transpose_10")];
            tensor<fp16, [1, 4, 256, 256]> var_6935_cast_fp16 = pow(x = x_751, y = var_22_promoted_188_to_fp16)[name = tensor<string, []>("op_6935_cast_fp16")];
            tensor<int32, [1]> var_6937_axes_0 = const()[name = tensor<string, []>("op_6937_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_6937_keep_dims_0 = const()[name = tensor<string, []>("op_6937_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_6937_cast_fp16 = reduce_mean(axes = var_6937_axes_0, keep_dims = var_6937_keep_dims_0, x = var_6935_cast_fp16)[name = tensor<string, []>("op_6937_cast_fp16")];
            tensor<fp16, []> var_6938_to_fp16 = const()[name = tensor<string, []>("op_6938_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_6939_cast_fp16 = add(x = var_6937_cast_fp16, y = var_6938_to_fp16)[name = tensor<string, []>("op_6939_cast_fp16")];
            tensor<fp32, []> var_6940_epsilon_0 = const()[name = tensor<string, []>("op_6940_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_6940_cast_fp16 = rsqrt(epsilon = var_6940_epsilon_0, x = var_6939_cast_fp16)[name = tensor<string, []>("op_6940_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_753_cast_fp16 = mul(x = x_751, y = var_6940_cast_fp16)[name = tensor<string, []>("output_753_cast_fp16")];
            tensor<fp16, [256]> var_6944_to_fp16 = const()[name = tensor<string, []>("op_6944_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940874560)))];
            tensor<fp16, [1, 4, 256, 256]> output_755_cast_fp16 = mul(x = output_753_cast_fp16, y = var_6944_to_fp16)[name = tensor<string, []>("output_755_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_6949 = mul(x = output_751_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_6949")];
            tensor<int32, [4]> x1_125_begin_0 = const()[name = tensor<string, []>("x1_125_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_125_end_0 = const()[name = tensor<string, []>("x1_125_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_125_end_mask_0 = const()[name = tensor<string, []>("x1_125_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_125 = slice_by_index(begin = x1_125_begin_0, end = x1_125_end_0, end_mask = x1_125_end_mask_0, x = output_751_cast_fp16)[name = tensor<string, []>("x1_125")];
            tensor<int32, [4]> x2_125_begin_0 = const()[name = tensor<string, []>("x2_125_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_125_end_0 = const()[name = tensor<string, []>("x2_125_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_125_end_mask_0 = const()[name = tensor<string, []>("x2_125_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_125 = slice_by_index(begin = x2_125_begin_0, end = x2_125_end_0, end_mask = x2_125_end_mask_0, x = output_751_cast_fp16)[name = tensor<string, []>("x2_125")];
            tensor<fp16, []> const_748_promoted = const()[name = tensor<string, []>("const_748_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_6960 = mul(x = x2_125, y = const_748_promoted)[name = tensor<string, []>("op_6960")];
            tensor<bool, []> var_6962_interleave_0 = const()[name = tensor<string, []>("op_6962_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_6962 = concat(axis = var_24, interleave = var_6962_interleave_0, values = (var_6960, x1_125))[name = tensor<string, []>("op_6962")];
            tensor<fp16, [1, 8, 256, 256]> var_6963 = mul(x = var_6962, y = sin_7_palettized)[name = tensor<string, []>("op_6963")];
            tensor<fp16, [1, 8, 256, 256]> query_63 = add(x = var_6949, y = var_6963)[name = tensor<string, []>("query_63")];
            tensor<fp16, [1, 4, 256, 256]> var_6965 = mul(x = output_755_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_6965")];
            tensor<int32, [4]> x1_127_begin_0 = const()[name = tensor<string, []>("x1_127_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_127_end_0 = const()[name = tensor<string, []>("x1_127_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_127_end_mask_0 = const()[name = tensor<string, []>("x1_127_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_127 = slice_by_index(begin = x1_127_begin_0, end = x1_127_end_0, end_mask = x1_127_end_mask_0, x = output_755_cast_fp16)[name = tensor<string, []>("x1_127")];
            tensor<int32, [4]> x2_127_begin_0 = const()[name = tensor<string, []>("x2_127_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_127_end_0 = const()[name = tensor<string, []>("x2_127_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_127_end_mask_0 = const()[name = tensor<string, []>("x2_127_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_127 = slice_by_index(begin = x2_127_begin_0, end = x2_127_end_0, end_mask = x2_127_end_mask_0, x = output_755_cast_fp16)[name = tensor<string, []>("x2_127")];
            tensor<fp16, []> const_751_promoted = const()[name = tensor<string, []>("const_751_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_6976 = mul(x = x2_127, y = const_751_promoted)[name = tensor<string, []>("op_6976")];
            tensor<bool, []> var_6978_interleave_0 = const()[name = tensor<string, []>("op_6978_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_6978 = concat(axis = var_24, interleave = var_6978_interleave_0, values = (var_6976, x1_127))[name = tensor<string, []>("op_6978")];
            tensor<fp16, [1, 4, 256, 256]> var_6979 = mul(x = var_6978, y = sin_7_palettized)[name = tensor<string, []>("op_6979")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_437 = add(x = var_6965, y = var_6979)[name = tensor<string, []>("hidden_states_437")];
            tensor<int32, [1]> var_6988_axes_0 = const()[name = tensor<string, []>("op_6988_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_6988 = expand_dims(axes = var_6988_axes_0, x = hidden_states_437)[name = tensor<string, []>("op_6988")];
            tensor<int32, [5]> hidden_states_439_reps_0 = const()[name = tensor<string, []>("hidden_states_439_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_439 = tile(reps = hidden_states_439_reps_0, x = var_6988)[name = tensor<string, []>("hidden_states_439")];
            tensor<int32, [4]> var_6996 = const()[name = tensor<string, []>("op_6996"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_63 = reshape(shape = var_6996, x = hidden_states_439)[name = tensor<string, []>("key_states_63")];
            tensor<int32, [1]> var_7005_axes_0 = const()[name = tensor<string, []>("op_7005_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_441 = transpose(perm = hidden_states_441_perm_0, x = var_6917)[name = tensor<string, []>("transpose_9")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_7005 = expand_dims(axes = var_7005_axes_0, x = hidden_states_441)[name = tensor<string, []>("op_7005")];
            tensor<int32, [5]> hidden_states_443_reps_0 = const()[name = tensor<string, []>("hidden_states_443_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_443 = tile(reps = hidden_states_443_reps_0, x = var_7005)[name = tensor<string, []>("hidden_states_443")];
            tensor<int32, [4]> var_7013 = const()[name = tensor<string, []>("op_7013"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_63 = reshape(shape = var_7013, x = hidden_states_443)[name = tensor<string, []>("value_states_63")];
            tensor<bool, []> var_7016_transpose_x_1 = const()[name = tensor<string, []>("op_7016_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_7016_transpose_y_1 = const()[name = tensor<string, []>("op_7016_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_7016 = matmul(transpose_x = var_7016_transpose_x_1, transpose_y = var_7016_transpose_y_1, x = query_63, y = key_states_63)[name = tensor<string, []>("op_7016")];
            tensor<fp16, []> var_7017_to_fp16 = const()[name = tensor<string, []>("op_7017_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_125_cast_fp16 = mul(x = var_7016, y = var_7017_to_fp16)[name = tensor<string, []>("attn_weights_125_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_373_cast_fp16 = add(x = attn_weights_125_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_373_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_7025_cast_fp16 = softmax(axis = var_24, x = input_373_cast_fp16)[name = tensor<string, []>("op_7025_cast_fp16")];
            tensor<bool, []> attn_output_125_transpose_x_0 = const()[name = tensor<string, []>("attn_output_125_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_125_transpose_y_0 = const()[name = tensor<string, []>("attn_output_125_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_125 = matmul(transpose_x = attn_output_125_transpose_x_0, transpose_y = attn_output_125_transpose_y_0, x = var_7025_cast_fp16, y = value_states_63)[name = tensor<string, []>("attn_output_125")];
            tensor<int32, [4]> var_7029_perm_0 = const()[name = tensor<string, []>("op_7029_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_7031 = const()[name = tensor<string, []>("op_7031"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_7029 = transpose(perm = var_7029_perm_0, x = attn_output_125)[name = tensor<string, []>("transpose_8")];
            tensor<fp16, [1, 256, 2048]> var_7032 = reshape(shape = var_7031, x = var_7029)[name = tensor<string, []>("op_7032")];
            tensor<fp16, [1, 256, 2560]> x_755 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_31_self_attn_o_proj_weight_palettized, x = var_7032)[name = tensor<string, []>("linear_220")];
            tensor<fp16, []> var_22_promoted_189_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_189_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_7038_cast_fp16 = pow(x = x_755, y = var_22_promoted_189_to_fp16)[name = tensor<string, []>("op_7038_cast_fp16")];
            tensor<int32, [1]> var_7040_axes_0 = const()[name = tensor<string, []>("op_7040_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_7040_keep_dims_0 = const()[name = tensor<string, []>("op_7040_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_7040_cast_fp16 = reduce_mean(axes = var_7040_axes_0, keep_dims = var_7040_keep_dims_0, x = var_7038_cast_fp16)[name = tensor<string, []>("op_7040_cast_fp16")];
            tensor<fp16, []> var_7041_to_fp16 = const()[name = tensor<string, []>("op_7041_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_7042_cast_fp16 = add(x = var_7040_cast_fp16, y = var_7041_to_fp16)[name = tensor<string, []>("op_7042_cast_fp16")];
            tensor<fp32, []> var_7043_epsilon_0 = const()[name = tensor<string, []>("op_7043_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_7043_cast_fp16 = rsqrt(epsilon = var_7043_epsilon_0, x = var_7042_cast_fp16)[name = tensor<string, []>("op_7043_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_757_cast_fp16 = mul(x = x_755, y = var_7043_cast_fp16)[name = tensor<string, []>("output_757_cast_fp16")];
            tensor<fp16, [2560]> var_7047_to_fp16 = const()[name = tensor<string, []>("op_7047_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940875136)))];
            tensor<fp16, [1, 256, 2560]> output_759_cast_fp16 = mul(x = output_757_cast_fp16, y = var_7047_to_fp16)[name = tensor<string, []>("output_759_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_759 = add(x = x_743, y = output_759_cast_fp16)[name = tensor<string, []>("x_759")];
            tensor<fp16, []> var_22_promoted_190_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_190_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_7053_cast_fp16 = pow(x = x_759, y = var_22_promoted_190_to_fp16)[name = tensor<string, []>("op_7053_cast_fp16")];
            tensor<int32, [1]> var_7055_axes_0 = const()[name = tensor<string, []>("op_7055_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_7055_keep_dims_0 = const()[name = tensor<string, []>("op_7055_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_7055_cast_fp16 = reduce_mean(axes = var_7055_axes_0, keep_dims = var_7055_keep_dims_0, x = var_7053_cast_fp16)[name = tensor<string, []>("op_7055_cast_fp16")];
            tensor<fp16, []> var_7056_to_fp16 = const()[name = tensor<string, []>("op_7056_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_7057_cast_fp16 = add(x = var_7055_cast_fp16, y = var_7056_to_fp16)[name = tensor<string, []>("op_7057_cast_fp16")];
            tensor<fp32, []> var_7058_epsilon_0 = const()[name = tensor<string, []>("op_7058_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_7058_cast_fp16 = rsqrt(epsilon = var_7058_epsilon_0, x = var_7057_cast_fp16)[name = tensor<string, []>("op_7058_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_761_cast_fp16 = mul(x = x_759, y = var_7058_cast_fp16)[name = tensor<string, []>("output_761_cast_fp16")];
            tensor<fp16, [2560]> var_7062_to_fp16 = const()[name = tensor<string, []>("op_7062_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940880320)))];
            tensor<fp16, [1, 256, 2560]> output_763_cast_fp16 = mul(x = output_761_cast_fp16, y = var_7062_to_fp16)[name = tensor<string, []>("output_763_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_381 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_31_mlp_gate_proj_weight_palettized, x = output_763_cast_fp16)[name = tensor<string, []>("linear_221")];
            tensor<string, []> var_7070_mode_0 = const()[name = tensor<string, []>("op_7070_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_7070 = gelu(mode = var_7070_mode_0, x = input_381)[name = tensor<string, []>("op_7070")];
            tensor<fp16, [1, 256, 10240]> var_7072 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_31_mlp_up_proj_weight_palettized, x = output_763_cast_fp16)[name = tensor<string, []>("linear_222")];
            tensor<fp16, [1, 256, 10240]> input_383 = mul(x = var_7070, y = var_7072)[name = tensor<string, []>("input_383")];
            tensor<fp16, [1, 256, 2560]> x_763 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_31_mlp_down_proj_weight_palettized, x = input_383)[name = tensor<string, []>("linear_223")];
            tensor<fp16, []> var_22_promoted_191_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_191_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_7078_cast_fp16 = pow(x = x_763, y = var_22_promoted_191_to_fp16)[name = tensor<string, []>("op_7078_cast_fp16")];
            tensor<int32, [1]> var_7080_axes_0 = const()[name = tensor<string, []>("op_7080_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_7080_keep_dims_0 = const()[name = tensor<string, []>("op_7080_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_7080_cast_fp16 = reduce_mean(axes = var_7080_axes_0, keep_dims = var_7080_keep_dims_0, x = var_7078_cast_fp16)[name = tensor<string, []>("op_7080_cast_fp16")];
            tensor<fp16, []> var_7081_to_fp16 = const()[name = tensor<string, []>("op_7081_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_7082_cast_fp16 = add(x = var_7080_cast_fp16, y = var_7081_to_fp16)[name = tensor<string, []>("op_7082_cast_fp16")];
            tensor<fp32, []> var_7083_epsilon_0 = const()[name = tensor<string, []>("op_7083_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_7083_cast_fp16 = rsqrt(epsilon = var_7083_epsilon_0, x = var_7082_cast_fp16)[name = tensor<string, []>("op_7083_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_765_cast_fp16 = mul(x = x_763, y = var_7083_cast_fp16)[name = tensor<string, []>("output_765_cast_fp16")];
            tensor<fp16, [2560]> var_7087_to_fp16 = const()[name = tensor<string, []>("op_7087_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940885504)))];
            tensor<fp16, [1, 256, 2560]> output_767_cast_fp16 = mul(x = output_765_cast_fp16, y = var_7087_to_fp16)[name = tensor<string, []>("output_767_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_767 = add(x = x_759, y = output_767_cast_fp16)[name = tensor<string, []>("x_767")];
            tensor<fp16, []> var_22_promoted_192_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_192_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_7099_cast_fp16 = pow(x = x_767, y = var_22_promoted_192_to_fp16)[name = tensor<string, []>("op_7099_cast_fp16")];
            tensor<int32, [1]> var_7101_axes_0 = const()[name = tensor<string, []>("op_7101_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_7101_keep_dims_0 = const()[name = tensor<string, []>("op_7101_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_7101_cast_fp16 = reduce_mean(axes = var_7101_axes_0, keep_dims = var_7101_keep_dims_0, x = var_7099_cast_fp16)[name = tensor<string, []>("op_7101_cast_fp16")];
            tensor<fp16, []> var_7102_to_fp16 = const()[name = tensor<string, []>("op_7102_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_7103_cast_fp16 = add(x = var_7101_cast_fp16, y = var_7102_to_fp16)[name = tensor<string, []>("op_7103_cast_fp16")];
            tensor<fp32, []> var_7104_epsilon_0 = const()[name = tensor<string, []>("op_7104_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_7104_cast_fp16 = rsqrt(epsilon = var_7104_epsilon_0, x = var_7103_cast_fp16)[name = tensor<string, []>("op_7104_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_769_cast_fp16 = mul(x = x_767, y = var_7104_cast_fp16)[name = tensor<string, []>("output_769_cast_fp16")];
            tensor<fp16, [2560]> var_7108_to_fp16 = const()[name = tensor<string, []>("op_7108_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940890688)))];
            tensor<fp16, [1, 256, 2560]> output_771_cast_fp16 = mul(x = output_769_cast_fp16, y = var_7108_to_fp16)[name = tensor<string, []>("output_771_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_7120 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_32_self_attn_q_proj_weight_palettized, x = output_771_cast_fp16)[name = tensor<string, []>("linear_224")];
            tensor<int32, [4]> var_7121 = const()[name = tensor<string, []>("op_7121"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_7122 = reshape(shape = var_7121, x = var_7120)[name = tensor<string, []>("op_7122")];
            tensor<int32, [4]> x_771_perm_0 = const()[name = tensor<string, []>("x_771_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_7125 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_32_self_attn_k_proj_weight_palettized, x = output_771_cast_fp16)[name = tensor<string, []>("linear_225")];
            tensor<int32, [4]> var_7126 = const()[name = tensor<string, []>("op_7126"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_7127 = reshape(shape = var_7126, x = var_7125)[name = tensor<string, []>("op_7127")];
            tensor<int32, [4]> x_775_perm_0 = const()[name = tensor<string, []>("x_775_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_7130 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_32_self_attn_v_proj_weight_palettized, x = output_771_cast_fp16)[name = tensor<string, []>("linear_226")];
            tensor<int32, [4]> var_7131 = const()[name = tensor<string, []>("op_7131"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_7132 = reshape(shape = var_7131, x = var_7130)[name = tensor<string, []>("op_7132")];
            tensor<int32, [4]> hidden_states_455_perm_0 = const()[name = tensor<string, []>("hidden_states_455_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_193_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_193_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_771 = transpose(perm = x_771_perm_0, x = var_7122)[name = tensor<string, []>("transpose_7")];
            tensor<fp16, [1, 8, 256, 256]> var_7136_cast_fp16 = pow(x = x_771, y = var_22_promoted_193_to_fp16)[name = tensor<string, []>("op_7136_cast_fp16")];
            tensor<int32, [1]> var_7138_axes_0 = const()[name = tensor<string, []>("op_7138_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_7138_keep_dims_0 = const()[name = tensor<string, []>("op_7138_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_7138_cast_fp16 = reduce_mean(axes = var_7138_axes_0, keep_dims = var_7138_keep_dims_0, x = var_7136_cast_fp16)[name = tensor<string, []>("op_7138_cast_fp16")];
            tensor<fp16, []> var_7139_to_fp16 = const()[name = tensor<string, []>("op_7139_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_7140_cast_fp16 = add(x = var_7138_cast_fp16, y = var_7139_to_fp16)[name = tensor<string, []>("op_7140_cast_fp16")];
            tensor<fp32, []> var_7141_epsilon_0 = const()[name = tensor<string, []>("op_7141_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_7141_cast_fp16 = rsqrt(epsilon = var_7141_epsilon_0, x = var_7140_cast_fp16)[name = tensor<string, []>("op_7141_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_773_cast_fp16 = mul(x = x_771, y = var_7141_cast_fp16)[name = tensor<string, []>("output_773_cast_fp16")];
            tensor<fp16, [256]> var_7145_to_fp16 = const()[name = tensor<string, []>("op_7145_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940895872)))];
            tensor<fp16, [1, 8, 256, 256]> output_775_cast_fp16 = mul(x = output_773_cast_fp16, y = var_7145_to_fp16)[name = tensor<string, []>("output_775_cast_fp16")];
            tensor<fp16, []> var_22_promoted_194_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_194_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_775 = transpose(perm = x_775_perm_0, x = var_7127)[name = tensor<string, []>("transpose_6")];
            tensor<fp16, [1, 4, 256, 256]> var_7150_cast_fp16 = pow(x = x_775, y = var_22_promoted_194_to_fp16)[name = tensor<string, []>("op_7150_cast_fp16")];
            tensor<int32, [1]> var_7152_axes_0 = const()[name = tensor<string, []>("op_7152_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_7152_keep_dims_0 = const()[name = tensor<string, []>("op_7152_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_7152_cast_fp16 = reduce_mean(axes = var_7152_axes_0, keep_dims = var_7152_keep_dims_0, x = var_7150_cast_fp16)[name = tensor<string, []>("op_7152_cast_fp16")];
            tensor<fp16, []> var_7153_to_fp16 = const()[name = tensor<string, []>("op_7153_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_7154_cast_fp16 = add(x = var_7152_cast_fp16, y = var_7153_to_fp16)[name = tensor<string, []>("op_7154_cast_fp16")];
            tensor<fp32, []> var_7155_epsilon_0 = const()[name = tensor<string, []>("op_7155_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_7155_cast_fp16 = rsqrt(epsilon = var_7155_epsilon_0, x = var_7154_cast_fp16)[name = tensor<string, []>("op_7155_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_777_cast_fp16 = mul(x = x_775, y = var_7155_cast_fp16)[name = tensor<string, []>("output_777_cast_fp16")];
            tensor<fp16, [256]> var_7159_to_fp16 = const()[name = tensor<string, []>("op_7159_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940896448)))];
            tensor<fp16, [1, 4, 256, 256]> output_779_cast_fp16 = mul(x = output_777_cast_fp16, y = var_7159_to_fp16)[name = tensor<string, []>("output_779_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_7164 = mul(x = output_775_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_7164")];
            tensor<int32, [4]> x1_129_begin_0 = const()[name = tensor<string, []>("x1_129_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_129_end_0 = const()[name = tensor<string, []>("x1_129_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_129_end_mask_0 = const()[name = tensor<string, []>("x1_129_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_129 = slice_by_index(begin = x1_129_begin_0, end = x1_129_end_0, end_mask = x1_129_end_mask_0, x = output_775_cast_fp16)[name = tensor<string, []>("x1_129")];
            tensor<int32, [4]> x2_129_begin_0 = const()[name = tensor<string, []>("x2_129_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_129_end_0 = const()[name = tensor<string, []>("x2_129_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_129_end_mask_0 = const()[name = tensor<string, []>("x2_129_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_129 = slice_by_index(begin = x2_129_begin_0, end = x2_129_end_0, end_mask = x2_129_end_mask_0, x = output_775_cast_fp16)[name = tensor<string, []>("x2_129")];
            tensor<fp16, []> const_771_promoted = const()[name = tensor<string, []>("const_771_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_7175 = mul(x = x2_129, y = const_771_promoted)[name = tensor<string, []>("op_7175")];
            tensor<bool, []> var_7177_interleave_0 = const()[name = tensor<string, []>("op_7177_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_7177 = concat(axis = var_24, interleave = var_7177_interleave_0, values = (var_7175, x1_129))[name = tensor<string, []>("op_7177")];
            tensor<fp16, [1, 8, 256, 256]> var_7178 = mul(x = var_7177, y = sin_7_palettized)[name = tensor<string, []>("op_7178")];
            tensor<fp16, [1, 8, 256, 256]> query_65 = add(x = var_7164, y = var_7178)[name = tensor<string, []>("query_65")];
            tensor<fp16, [1, 4, 256, 256]> var_7180 = mul(x = output_779_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_7180")];
            tensor<int32, [4]> x1_131_begin_0 = const()[name = tensor<string, []>("x1_131_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_131_end_0 = const()[name = tensor<string, []>("x1_131_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_131_end_mask_0 = const()[name = tensor<string, []>("x1_131_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1_131 = slice_by_index(begin = x1_131_begin_0, end = x1_131_end_0, end_mask = x1_131_end_mask_0, x = output_779_cast_fp16)[name = tensor<string, []>("x1_131")];
            tensor<int32, [4]> x2_131_begin_0 = const()[name = tensor<string, []>("x2_131_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_131_end_0 = const()[name = tensor<string, []>("x2_131_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_131_end_mask_0 = const()[name = tensor<string, []>("x2_131_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2_131 = slice_by_index(begin = x2_131_begin_0, end = x2_131_end_0, end_mask = x2_131_end_mask_0, x = output_779_cast_fp16)[name = tensor<string, []>("x2_131")];
            tensor<fp16, []> const_774_promoted = const()[name = tensor<string, []>("const_774_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_7191 = mul(x = x2_131, y = const_774_promoted)[name = tensor<string, []>("op_7191")];
            tensor<bool, []> var_7193_interleave_0 = const()[name = tensor<string, []>("op_7193_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_7193 = concat(axis = var_24, interleave = var_7193_interleave_0, values = (var_7191, x1_131))[name = tensor<string, []>("op_7193")];
            tensor<fp16, [1, 4, 256, 256]> var_7194 = mul(x = var_7193, y = sin_7_palettized)[name = tensor<string, []>("op_7194")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_451 = add(x = var_7180, y = var_7194)[name = tensor<string, []>("hidden_states_451")];
            tensor<int32, [1]> var_7203_axes_0 = const()[name = tensor<string, []>("op_7203_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_7203 = expand_dims(axes = var_7203_axes_0, x = hidden_states_451)[name = tensor<string, []>("op_7203")];
            tensor<int32, [5]> hidden_states_453_reps_0 = const()[name = tensor<string, []>("hidden_states_453_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_453 = tile(reps = hidden_states_453_reps_0, x = var_7203)[name = tensor<string, []>("hidden_states_453")];
            tensor<int32, [4]> var_7211 = const()[name = tensor<string, []>("op_7211"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states_65 = reshape(shape = var_7211, x = hidden_states_453)[name = tensor<string, []>("key_states_65")];
            tensor<int32, [1]> var_7220_axes_0 = const()[name = tensor<string, []>("op_7220_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_455 = transpose(perm = hidden_states_455_perm_0, x = var_7132)[name = tensor<string, []>("transpose_5")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_7220 = expand_dims(axes = var_7220_axes_0, x = hidden_states_455)[name = tensor<string, []>("op_7220")];
            tensor<int32, [5]> hidden_states_457_reps_0 = const()[name = tensor<string, []>("hidden_states_457_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_457 = tile(reps = hidden_states_457_reps_0, x = var_7220)[name = tensor<string, []>("hidden_states_457")];
            tensor<int32, [4]> var_7228 = const()[name = tensor<string, []>("op_7228"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states_65 = reshape(shape = var_7228, x = hidden_states_457)[name = tensor<string, []>("value_states_65")];
            tensor<bool, []> var_7231_transpose_x_1 = const()[name = tensor<string, []>("op_7231_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_7231_transpose_y_1 = const()[name = tensor<string, []>("op_7231_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_7231 = matmul(transpose_x = var_7231_transpose_x_1, transpose_y = var_7231_transpose_y_1, x = query_65, y = key_states_65)[name = tensor<string, []>("op_7231")];
            tensor<fp16, []> var_7232_to_fp16 = const()[name = tensor<string, []>("op_7232_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_129_cast_fp16 = mul(x = var_7231, y = var_7232_to_fp16)[name = tensor<string, []>("attn_weights_129_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_385_cast_fp16 = add(x = attn_weights_129_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_385_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_7240_cast_fp16 = softmax(axis = var_24, x = input_385_cast_fp16)[name = tensor<string, []>("op_7240_cast_fp16")];
            tensor<bool, []> attn_output_129_transpose_x_0 = const()[name = tensor<string, []>("attn_output_129_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_129_transpose_y_0 = const()[name = tensor<string, []>("attn_output_129_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_129 = matmul(transpose_x = attn_output_129_transpose_x_0, transpose_y = attn_output_129_transpose_y_0, x = var_7240_cast_fp16, y = value_states_65)[name = tensor<string, []>("attn_output_129")];
            tensor<int32, [4]> var_7244_perm_0 = const()[name = tensor<string, []>("op_7244_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_7246 = const()[name = tensor<string, []>("op_7246"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_7244 = transpose(perm = var_7244_perm_0, x = attn_output_129)[name = tensor<string, []>("transpose_4")];
            tensor<fp16, [1, 256, 2048]> var_7247 = reshape(shape = var_7246, x = var_7244)[name = tensor<string, []>("op_7247")];
            tensor<fp16, [1, 256, 2560]> x_779 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_32_self_attn_o_proj_weight_palettized, x = var_7247)[name = tensor<string, []>("linear_227")];
            tensor<fp16, []> var_22_promoted_195_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_195_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_7253_cast_fp16 = pow(x = x_779, y = var_22_promoted_195_to_fp16)[name = tensor<string, []>("op_7253_cast_fp16")];
            tensor<int32, [1]> var_7255_axes_0 = const()[name = tensor<string, []>("op_7255_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_7255_keep_dims_0 = const()[name = tensor<string, []>("op_7255_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_7255_cast_fp16 = reduce_mean(axes = var_7255_axes_0, keep_dims = var_7255_keep_dims_0, x = var_7253_cast_fp16)[name = tensor<string, []>("op_7255_cast_fp16")];
            tensor<fp16, []> var_7256_to_fp16 = const()[name = tensor<string, []>("op_7256_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_7257_cast_fp16 = add(x = var_7255_cast_fp16, y = var_7256_to_fp16)[name = tensor<string, []>("op_7257_cast_fp16")];
            tensor<fp32, []> var_7258_epsilon_0 = const()[name = tensor<string, []>("op_7258_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_7258_cast_fp16 = rsqrt(epsilon = var_7258_epsilon_0, x = var_7257_cast_fp16)[name = tensor<string, []>("op_7258_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_781_cast_fp16 = mul(x = x_779, y = var_7258_cast_fp16)[name = tensor<string, []>("output_781_cast_fp16")];
            tensor<fp16, [2560]> var_7262_to_fp16 = const()[name = tensor<string, []>("op_7262_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940897024)))];
            tensor<fp16, [1, 256, 2560]> output_783_cast_fp16 = mul(x = output_781_cast_fp16, y = var_7262_to_fp16)[name = tensor<string, []>("output_783_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_783 = add(x = x_767, y = output_783_cast_fp16)[name = tensor<string, []>("x_783")];
            tensor<fp16, []> var_22_promoted_196_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_196_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_7268_cast_fp16 = pow(x = x_783, y = var_22_promoted_196_to_fp16)[name = tensor<string, []>("op_7268_cast_fp16")];
            tensor<int32, [1]> var_7270_axes_0 = const()[name = tensor<string, []>("op_7270_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_7270_keep_dims_0 = const()[name = tensor<string, []>("op_7270_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_7270_cast_fp16 = reduce_mean(axes = var_7270_axes_0, keep_dims = var_7270_keep_dims_0, x = var_7268_cast_fp16)[name = tensor<string, []>("op_7270_cast_fp16")];
            tensor<fp16, []> var_7271_to_fp16 = const()[name = tensor<string, []>("op_7271_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_7272_cast_fp16 = add(x = var_7270_cast_fp16, y = var_7271_to_fp16)[name = tensor<string, []>("op_7272_cast_fp16")];
            tensor<fp32, []> var_7273_epsilon_0 = const()[name = tensor<string, []>("op_7273_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_7273_cast_fp16 = rsqrt(epsilon = var_7273_epsilon_0, x = var_7272_cast_fp16)[name = tensor<string, []>("op_7273_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_785_cast_fp16 = mul(x = x_783, y = var_7273_cast_fp16)[name = tensor<string, []>("output_785_cast_fp16")];
            tensor<fp16, [2560]> var_7277_to_fp16 = const()[name = tensor<string, []>("op_7277_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940902208)))];
            tensor<fp16, [1, 256, 2560]> output_787_cast_fp16 = mul(x = output_785_cast_fp16, y = var_7277_to_fp16)[name = tensor<string, []>("output_787_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_393 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_32_mlp_gate_proj_weight_palettized, x = output_787_cast_fp16)[name = tensor<string, []>("linear_228")];
            tensor<string, []> var_7285_mode_0 = const()[name = tensor<string, []>("op_7285_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_7285 = gelu(mode = var_7285_mode_0, x = input_393)[name = tensor<string, []>("op_7285")];
            tensor<fp16, [1, 256, 10240]> var_7287 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_32_mlp_up_proj_weight_palettized, x = output_787_cast_fp16)[name = tensor<string, []>("linear_229")];
            tensor<fp16, [1, 256, 10240]> input_395 = mul(x = var_7285, y = var_7287)[name = tensor<string, []>("input_395")];
            tensor<fp16, [1, 256, 2560]> x_787 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_32_mlp_down_proj_weight_palettized, x = input_395)[name = tensor<string, []>("linear_230")];
            tensor<fp16, []> var_22_promoted_197_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_197_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_7293_cast_fp16 = pow(x = x_787, y = var_22_promoted_197_to_fp16)[name = tensor<string, []>("op_7293_cast_fp16")];
            tensor<int32, [1]> var_7295_axes_0 = const()[name = tensor<string, []>("op_7295_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_7295_keep_dims_0 = const()[name = tensor<string, []>("op_7295_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_7295_cast_fp16 = reduce_mean(axes = var_7295_axes_0, keep_dims = var_7295_keep_dims_0, x = var_7293_cast_fp16)[name = tensor<string, []>("op_7295_cast_fp16")];
            tensor<fp16, []> var_7296_to_fp16 = const()[name = tensor<string, []>("op_7296_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_7297_cast_fp16 = add(x = var_7295_cast_fp16, y = var_7296_to_fp16)[name = tensor<string, []>("op_7297_cast_fp16")];
            tensor<fp32, []> var_7298_epsilon_0 = const()[name = tensor<string, []>("op_7298_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_7298_cast_fp16 = rsqrt(epsilon = var_7298_epsilon_0, x = var_7297_cast_fp16)[name = tensor<string, []>("op_7298_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_789_cast_fp16 = mul(x = x_787, y = var_7298_cast_fp16)[name = tensor<string, []>("output_789_cast_fp16")];
            tensor<fp16, [2560]> var_7302_to_fp16 = const()[name = tensor<string, []>("op_7302_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940907392)))];
            tensor<fp16, [1, 256, 2560]> output_791_cast_fp16 = mul(x = output_789_cast_fp16, y = var_7302_to_fp16)[name = tensor<string, []>("output_791_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_791 = add(x = x_783, y = output_791_cast_fp16)[name = tensor<string, []>("x_791")];
            tensor<fp16, []> var_22_promoted_198_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_198_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_7314_cast_fp16 = pow(x = x_791, y = var_22_promoted_198_to_fp16)[name = tensor<string, []>("op_7314_cast_fp16")];
            tensor<int32, [1]> var_7316_axes_0 = const()[name = tensor<string, []>("op_7316_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_7316_keep_dims_0 = const()[name = tensor<string, []>("op_7316_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_7316_cast_fp16 = reduce_mean(axes = var_7316_axes_0, keep_dims = var_7316_keep_dims_0, x = var_7314_cast_fp16)[name = tensor<string, []>("op_7316_cast_fp16")];
            tensor<fp16, []> var_7317_to_fp16 = const()[name = tensor<string, []>("op_7317_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_7318_cast_fp16 = add(x = var_7316_cast_fp16, y = var_7317_to_fp16)[name = tensor<string, []>("op_7318_cast_fp16")];
            tensor<fp32, []> var_7319_epsilon_0 = const()[name = tensor<string, []>("op_7319_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_7319_cast_fp16 = rsqrt(epsilon = var_7319_epsilon_0, x = var_7318_cast_fp16)[name = tensor<string, []>("op_7319_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_793_cast_fp16 = mul(x = x_791, y = var_7319_cast_fp16)[name = tensor<string, []>("output_793_cast_fp16")];
            tensor<fp16, [2560]> var_7323_to_fp16 = const()[name = tensor<string, []>("op_7323_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940912576)))];
            tensor<fp16, [1, 256, 2560]> output_795_cast_fp16 = mul(x = output_793_cast_fp16, y = var_7323_to_fp16)[name = tensor<string, []>("output_795_cast_fp16")];
            tensor<fp16, [1, 256, 2048]> var_7335 = linear(bias = linear_0_bias_0, weight = wrapped_model_language_model_layers_33_self_attn_q_proj_weight_palettized, x = output_795_cast_fp16)[name = tensor<string, []>("linear_231")];
            tensor<int32, [4]> var_7336 = const()[name = tensor<string, []>("op_7336"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 8, 256]> var_7337 = reshape(shape = var_7336, x = var_7335)[name = tensor<string, []>("op_7337")];
            tensor<int32, [4]> x_795_perm_0 = const()[name = tensor<string, []>("x_795_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_7340 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_33_self_attn_k_proj_weight_palettized, x = output_795_cast_fp16)[name = tensor<string, []>("linear_232")];
            tensor<int32, [4]> var_7341 = const()[name = tensor<string, []>("op_7341"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_7342 = reshape(shape = var_7341, x = var_7340)[name = tensor<string, []>("op_7342")];
            tensor<int32, [4]> x_799_perm_0 = const()[name = tensor<string, []>("x_799_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 256, 1024]> var_7345 = linear(bias = linear_1_bias_0, weight = wrapped_model_language_model_layers_33_self_attn_v_proj_weight_palettized, x = output_795_cast_fp16)[name = tensor<string, []>("linear_233")];
            tensor<int32, [4]> var_7346 = const()[name = tensor<string, []>("op_7346"), val = tensor<int32, [4]>([1, 256, -1, 256])];
            tensor<fp16, [1, 256, 4, 256]> var_7347 = reshape(shape = var_7346, x = var_7345)[name = tensor<string, []>("op_7347")];
            tensor<int32, [4]> hidden_states_469_perm_0 = const()[name = tensor<string, []>("hidden_states_469_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> var_22_promoted_199_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_199_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 8, 256, 256]> x_795 = transpose(perm = x_795_perm_0, x = var_7337)[name = tensor<string, []>("transpose_3")];
            tensor<fp16, [1, 8, 256, 256]> var_7351_cast_fp16 = pow(x = x_795, y = var_22_promoted_199_to_fp16)[name = tensor<string, []>("op_7351_cast_fp16")];
            tensor<int32, [1]> var_7353_axes_0 = const()[name = tensor<string, []>("op_7353_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_7353_keep_dims_0 = const()[name = tensor<string, []>("op_7353_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 1]> var_7353_cast_fp16 = reduce_mean(axes = var_7353_axes_0, keep_dims = var_7353_keep_dims_0, x = var_7351_cast_fp16)[name = tensor<string, []>("op_7353_cast_fp16")];
            tensor<fp16, []> var_7354_to_fp16 = const()[name = tensor<string, []>("op_7354_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 8, 256, 1]> var_7355_cast_fp16 = add(x = var_7353_cast_fp16, y = var_7354_to_fp16)[name = tensor<string, []>("op_7355_cast_fp16")];
            tensor<fp32, []> var_7356_epsilon_0 = const()[name = tensor<string, []>("op_7356_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 8, 256, 1]> var_7356_cast_fp16 = rsqrt(epsilon = var_7356_epsilon_0, x = var_7355_cast_fp16)[name = tensor<string, []>("op_7356_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> output_797_cast_fp16 = mul(x = x_795, y = var_7356_cast_fp16)[name = tensor<string, []>("output_797_cast_fp16")];
            tensor<fp16, [256]> var_7360_to_fp16 = const()[name = tensor<string, []>("op_7360_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940917760)))];
            tensor<fp16, [1, 8, 256, 256]> output_799_cast_fp16 = mul(x = output_797_cast_fp16, y = var_7360_to_fp16)[name = tensor<string, []>("output_799_cast_fp16")];
            tensor<fp16, []> var_22_promoted_200_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_200_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 4, 256, 256]> x_799 = transpose(perm = x_799_perm_0, x = var_7342)[name = tensor<string, []>("transpose_2")];
            tensor<fp16, [1, 4, 256, 256]> var_7365_cast_fp16 = pow(x = x_799, y = var_22_promoted_200_to_fp16)[name = tensor<string, []>("op_7365_cast_fp16")];
            tensor<int32, [1]> var_7367_axes_0 = const()[name = tensor<string, []>("op_7367_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_7367_keep_dims_0 = const()[name = tensor<string, []>("op_7367_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 4, 256, 1]> var_7367_cast_fp16 = reduce_mean(axes = var_7367_axes_0, keep_dims = var_7367_keep_dims_0, x = var_7365_cast_fp16)[name = tensor<string, []>("op_7367_cast_fp16")];
            tensor<fp16, []> var_7368_to_fp16 = const()[name = tensor<string, []>("op_7368_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 4, 256, 1]> var_7369_cast_fp16 = add(x = var_7367_cast_fp16, y = var_7368_to_fp16)[name = tensor<string, []>("op_7369_cast_fp16")];
            tensor<fp32, []> var_7370_epsilon_0 = const()[name = tensor<string, []>("op_7370_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 4, 256, 1]> var_7370_cast_fp16 = rsqrt(epsilon = var_7370_epsilon_0, x = var_7369_cast_fp16)[name = tensor<string, []>("op_7370_cast_fp16")];
            tensor<fp16, [1, 4, 256, 256]> output_801_cast_fp16 = mul(x = x_799, y = var_7370_cast_fp16)[name = tensor<string, []>("output_801_cast_fp16")];
            tensor<fp16, [256]> var_7374_to_fp16 = const()[name = tensor<string, []>("op_7374_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940918336)))];
            tensor<fp16, [1, 4, 256, 256]> output_803_cast_fp16 = mul(x = output_801_cast_fp16, y = var_7374_to_fp16)[name = tensor<string, []>("output_803_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_7379 = mul(x = output_799_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_7379")];
            tensor<int32, [4]> x1_133_begin_0 = const()[name = tensor<string, []>("x1_133_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_133_end_0 = const()[name = tensor<string, []>("x1_133_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
            tensor<bool, [4]> x1_133_end_mask_0 = const()[name = tensor<string, []>("x1_133_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 256, 128]> x1_133 = slice_by_index(begin = x1_133_begin_0, end = x1_133_end_0, end_mask = x1_133_end_mask_0, x = output_799_cast_fp16)[name = tensor<string, []>("x1_133")];
            tensor<int32, [4]> x2_133_begin_0 = const()[name = tensor<string, []>("x2_133_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_133_end_0 = const()[name = tensor<string, []>("x2_133_end_0"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<bool, [4]> x2_133_end_mask_0 = const()[name = tensor<string, []>("x2_133_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 256, 128]> x2_133 = slice_by_index(begin = x2_133_begin_0, end = x2_133_end_0, end_mask = x2_133_end_mask_0, x = output_799_cast_fp16)[name = tensor<string, []>("x2_133")];
            tensor<fp16, []> const_794_promoted = const()[name = tensor<string, []>("const_794_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 8, 256, 128]> var_7390 = mul(x = x2_133, y = const_794_promoted)[name = tensor<string, []>("op_7390")];
            tensor<bool, []> var_7392_interleave_0 = const()[name = tensor<string, []>("op_7392_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> var_7392 = concat(axis = var_24, interleave = var_7392_interleave_0, values = (var_7390, x1_133))[name = tensor<string, []>("op_7392")];
            tensor<fp16, [1, 8, 256, 256]> var_7393 = mul(x = var_7392, y = sin_7_palettized)[name = tensor<string, []>("op_7393")];
            tensor<fp16, [1, 8, 256, 256]> query = add(x = var_7379, y = var_7393)[name = tensor<string, []>("query")];
            tensor<fp16, [1, 4, 256, 256]> var_7395 = mul(x = output_803_cast_fp16, y = cos_7_palettized)[name = tensor<string, []>("op_7395")];
            tensor<int32, [4]> x1_begin_0 = const()[name = tensor<string, []>("x1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_end_0 = const()[name = tensor<string, []>("x1_end_0"), val = tensor<int32, [4]>([1, 4, 256, 128])];
            tensor<bool, [4]> x1_end_mask_0 = const()[name = tensor<string, []>("x1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 256, 128]> x1 = slice_by_index(begin = x1_begin_0, end = x1_end_0, end_mask = x1_end_mask_0, x = output_803_cast_fp16)[name = tensor<string, []>("x1")];
            tensor<int32, [4]> x2_begin_0 = const()[name = tensor<string, []>("x2_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> x2_end_0 = const()[name = tensor<string, []>("x2_end_0"), val = tensor<int32, [4]>([1, 4, 256, 256])];
            tensor<bool, [4]> x2_end_mask_0 = const()[name = tensor<string, []>("x2_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 256, 128]> x2 = slice_by_index(begin = x2_begin_0, end = x2_end_0, end_mask = x2_end_mask_0, x = output_803_cast_fp16)[name = tensor<string, []>("x2")];
            tensor<fp16, []> const_797_promoted = const()[name = tensor<string, []>("const_797_promoted"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 256, 128]> var_7406 = mul(x = x2, y = const_797_promoted)[name = tensor<string, []>("op_7406")];
            tensor<bool, []> var_7408_interleave_0 = const()[name = tensor<string, []>("op_7408_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 256, 256]> var_7408 = concat(axis = var_24, interleave = var_7408_interleave_0, values = (var_7406, x1))[name = tensor<string, []>("op_7408")];
            tensor<fp16, [1, 4, 256, 256]> var_7409 = mul(x = var_7408, y = sin_7_palettized)[name = tensor<string, []>("op_7409")];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_465 = add(x = var_7395, y = var_7409)[name = tensor<string, []>("hidden_states_465")];
            tensor<int32, [1]> var_7418_axes_0 = const()[name = tensor<string, []>("op_7418_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 256, 256]> var_7418 = expand_dims(axes = var_7418_axes_0, x = hidden_states_465)[name = tensor<string, []>("op_7418")];
            tensor<int32, [5]> hidden_states_467_reps_0 = const()[name = tensor<string, []>("hidden_states_467_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_467 = tile(reps = hidden_states_467_reps_0, x = var_7418)[name = tensor<string, []>("hidden_states_467")];
            tensor<int32, [4]> var_7426 = const()[name = tensor<string, []>("op_7426"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> key_states = reshape(shape = var_7426, x = hidden_states_467)[name = tensor<string, []>("key_states")];
            tensor<int32, [1]> var_7435_axes_0 = const()[name = tensor<string, []>("op_7435_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 256, 256]> hidden_states_469 = transpose(perm = hidden_states_469_perm_0, x = var_7347)[name = tensor<string, []>("transpose_1")];
            tensor<fp16, [1, 4, 1, 256, 256]> var_7435 = expand_dims(axes = var_7435_axes_0, x = hidden_states_469)[name = tensor<string, []>("op_7435")];
            tensor<int32, [5]> hidden_states_471_reps_0 = const()[name = tensor<string, []>("hidden_states_471_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 256, 256]> hidden_states_471 = tile(reps = hidden_states_471_reps_0, x = var_7435)[name = tensor<string, []>("hidden_states_471")];
            tensor<int32, [4]> var_7443 = const()[name = tensor<string, []>("op_7443"), val = tensor<int32, [4]>([1, 8, 256, 256])];
            tensor<fp16, [1, 8, 256, 256]> value_states = reshape(shape = var_7443, x = hidden_states_471)[name = tensor<string, []>("value_states")];
            tensor<bool, []> var_7446_transpose_x_1 = const()[name = tensor<string, []>("op_7446_transpose_x_1"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_7446_transpose_y_1 = const()[name = tensor<string, []>("op_7446_transpose_y_1"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 8, 256, 256]> var_7446 = matmul(transpose_x = var_7446_transpose_x_1, transpose_y = var_7446_transpose_y_1, x = query, y = key_states)[name = tensor<string, []>("op_7446")];
            tensor<fp16, []> var_7447_to_fp16 = const()[name = tensor<string, []>("op_7447_to_fp16"), val = tensor<fp16, []>(0x1p-4)];
            tensor<fp16, [1, 8, 256, 256]> attn_weights_133_cast_fp16 = mul(x = var_7446, y = var_7447_to_fp16)[name = tensor<string, []>("attn_weights_133_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> input_397_cast_fp16 = add(x = attn_weights_133_cast_fp16, y = attention_mask_to_fp16_palettized)[name = tensor<string, []>("input_397_cast_fp16")];
            tensor<fp16, [1, 8, 256, 256]> var_7455_cast_fp16 = softmax(axis = var_24, x = input_397_cast_fp16)[name = tensor<string, []>("op_7455_cast_fp16")];
            tensor<bool, []> attn_output_133_transpose_x_0 = const()[name = tensor<string, []>("attn_output_133_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_133_transpose_y_0 = const()[name = tensor<string, []>("attn_output_133_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 256, 256]> attn_output_133 = matmul(transpose_x = attn_output_133_transpose_x_0, transpose_y = attn_output_133_transpose_y_0, x = var_7455_cast_fp16, y = value_states)[name = tensor<string, []>("attn_output_133")];
            tensor<int32, [4]> var_7459_perm_0 = const()[name = tensor<string, []>("op_7459_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_7461 = const()[name = tensor<string, []>("op_7461"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 8, 256]> var_7459 = transpose(perm = var_7459_perm_0, x = attn_output_133)[name = tensor<string, []>("transpose_0")];
            tensor<fp16, [1, 256, 2048]> var_7462 = reshape(shape = var_7461, x = var_7459)[name = tensor<string, []>("op_7462")];
            tensor<fp16, [1, 256, 2560]> x_803 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_33_self_attn_o_proj_weight_palettized, x = var_7462)[name = tensor<string, []>("linear_234")];
            tensor<fp16, []> var_22_promoted_201_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_201_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_7468_cast_fp16 = pow(x = x_803, y = var_22_promoted_201_to_fp16)[name = tensor<string, []>("op_7468_cast_fp16")];
            tensor<int32, [1]> var_7470_axes_0 = const()[name = tensor<string, []>("op_7470_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_7470_keep_dims_0 = const()[name = tensor<string, []>("op_7470_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_7470_cast_fp16 = reduce_mean(axes = var_7470_axes_0, keep_dims = var_7470_keep_dims_0, x = var_7468_cast_fp16)[name = tensor<string, []>("op_7470_cast_fp16")];
            tensor<fp16, []> var_7471_to_fp16 = const()[name = tensor<string, []>("op_7471_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_7472_cast_fp16 = add(x = var_7470_cast_fp16, y = var_7471_to_fp16)[name = tensor<string, []>("op_7472_cast_fp16")];
            tensor<fp32, []> var_7473_epsilon_0 = const()[name = tensor<string, []>("op_7473_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_7473_cast_fp16 = rsqrt(epsilon = var_7473_epsilon_0, x = var_7472_cast_fp16)[name = tensor<string, []>("op_7473_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_805_cast_fp16 = mul(x = x_803, y = var_7473_cast_fp16)[name = tensor<string, []>("output_805_cast_fp16")];
            tensor<fp16, [2560]> var_7477_to_fp16 = const()[name = tensor<string, []>("op_7477_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940918912)))];
            tensor<fp16, [1, 256, 2560]> output_807_cast_fp16 = mul(x = output_805_cast_fp16, y = var_7477_to_fp16)[name = tensor<string, []>("output_807_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_807 = add(x = x_791, y = output_807_cast_fp16)[name = tensor<string, []>("x_807")];
            tensor<fp16, []> var_22_promoted_202_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_202_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_7483_cast_fp16 = pow(x = x_807, y = var_22_promoted_202_to_fp16)[name = tensor<string, []>("op_7483_cast_fp16")];
            tensor<int32, [1]> var_7485_axes_0 = const()[name = tensor<string, []>("op_7485_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_7485_keep_dims_0 = const()[name = tensor<string, []>("op_7485_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_7485_cast_fp16 = reduce_mean(axes = var_7485_axes_0, keep_dims = var_7485_keep_dims_0, x = var_7483_cast_fp16)[name = tensor<string, []>("op_7485_cast_fp16")];
            tensor<fp16, []> var_7486_to_fp16 = const()[name = tensor<string, []>("op_7486_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_7487_cast_fp16 = add(x = var_7485_cast_fp16, y = var_7486_to_fp16)[name = tensor<string, []>("op_7487_cast_fp16")];
            tensor<fp32, []> var_7488_epsilon_0 = const()[name = tensor<string, []>("op_7488_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_7488_cast_fp16 = rsqrt(epsilon = var_7488_epsilon_0, x = var_7487_cast_fp16)[name = tensor<string, []>("op_7488_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_809_cast_fp16 = mul(x = x_807, y = var_7488_cast_fp16)[name = tensor<string, []>("output_809_cast_fp16")];
            tensor<fp16, [2560]> var_7492_to_fp16 = const()[name = tensor<string, []>("op_7492_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940924096)))];
            tensor<fp16, [1, 256, 2560]> output_811_cast_fp16 = mul(x = output_809_cast_fp16, y = var_7492_to_fp16)[name = tensor<string, []>("output_811_cast_fp16")];
            tensor<fp16, [1, 256, 10240]> input_405 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_33_mlp_gate_proj_weight_palettized, x = output_811_cast_fp16)[name = tensor<string, []>("linear_235")];
            tensor<string, []> var_7500_mode_0 = const()[name = tensor<string, []>("op_7500_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 256, 10240]> var_7500 = gelu(mode = var_7500_mode_0, x = input_405)[name = tensor<string, []>("op_7500")];
            tensor<fp16, [1, 256, 10240]> var_7502 = linear(bias = linear_4_bias_0, weight = wrapped_model_language_model_layers_33_mlp_up_proj_weight_palettized, x = output_811_cast_fp16)[name = tensor<string, []>("linear_236")];
            tensor<fp16, [1, 256, 10240]> input_407 = mul(x = var_7500, y = var_7502)[name = tensor<string, []>("input_407")];
            tensor<fp16, [1, 256, 2560]> x_811 = linear(bias = linear_3_bias_0, weight = wrapped_model_language_model_layers_33_mlp_down_proj_weight_palettized, x = input_407)[name = tensor<string, []>("linear_237")];
            tensor<fp16, []> var_22_promoted_203_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_203_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_7508_cast_fp16 = pow(x = x_811, y = var_22_promoted_203_to_fp16)[name = tensor<string, []>("op_7508_cast_fp16")];
            tensor<int32, [1]> var_7510_axes_0 = const()[name = tensor<string, []>("op_7510_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_7510_keep_dims_0 = const()[name = tensor<string, []>("op_7510_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_7510_cast_fp16 = reduce_mean(axes = var_7510_axes_0, keep_dims = var_7510_keep_dims_0, x = var_7508_cast_fp16)[name = tensor<string, []>("op_7510_cast_fp16")];
            tensor<fp16, []> var_7511_to_fp16 = const()[name = tensor<string, []>("op_7511_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_7512_cast_fp16 = add(x = var_7510_cast_fp16, y = var_7511_to_fp16)[name = tensor<string, []>("op_7512_cast_fp16")];
            tensor<fp32, []> var_7513_epsilon_0 = const()[name = tensor<string, []>("op_7513_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_7513_cast_fp16 = rsqrt(epsilon = var_7513_epsilon_0, x = var_7512_cast_fp16)[name = tensor<string, []>("op_7513_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_813_cast_fp16 = mul(x = x_811, y = var_7513_cast_fp16)[name = tensor<string, []>("output_813_cast_fp16")];
            tensor<fp16, [2560]> var_7517_to_fp16 = const()[name = tensor<string, []>("op_7517_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940929280)))];
            tensor<fp16, [1, 256, 2560]> output_815_cast_fp16 = mul(x = output_813_cast_fp16, y = var_7517_to_fp16)[name = tensor<string, []>("output_815_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> x_815 = add(x = x_807, y = output_815_cast_fp16)[name = tensor<string, []>("x_815")];
            tensor<fp16, []> var_22_promoted_204_to_fp16 = const()[name = tensor<string, []>("op_22_promoted_204_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 256, 2560]> var_7523_cast_fp16 = pow(x = x_815, y = var_22_promoted_204_to_fp16)[name = tensor<string, []>("op_7523_cast_fp16")];
            tensor<int32, [1]> var_7525_axes_0 = const()[name = tensor<string, []>("op_7525_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_7525_keep_dims_0 = const()[name = tensor<string, []>("op_7525_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 256, 1]> var_7525_cast_fp16 = reduce_mean(axes = var_7525_axes_0, keep_dims = var_7525_keep_dims_0, x = var_7523_cast_fp16)[name = tensor<string, []>("op_7525_cast_fp16")];
            tensor<fp16, []> var_7526_to_fp16 = const()[name = tensor<string, []>("op_7526_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 256, 1]> var_7527_cast_fp16 = add(x = var_7525_cast_fp16, y = var_7526_to_fp16)[name = tensor<string, []>("op_7527_cast_fp16")];
            tensor<fp32, []> var_7528_epsilon_0 = const()[name = tensor<string, []>("op_7528_epsilon_0"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp16, [1, 256, 1]> var_7528_cast_fp16 = rsqrt(epsilon = var_7528_epsilon_0, x = var_7527_cast_fp16)[name = tensor<string, []>("op_7528_cast_fp16")];
            tensor<fp16, [1, 256, 2560]> output_817_cast_fp16 = mul(x = x_815, y = var_7528_cast_fp16)[name = tensor<string, []>("output_817_cast_fp16")];
            tensor<fp16, [2560]> var_7532_to_fp16 = const()[name = tensor<string, []>("op_7532_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940934464)))];
            tensor<fp16, [1, 256, 2560]> output_cast_fp16 = mul(x = output_817_cast_fp16, y = var_7532_to_fp16)[name = tensor<string, []>("output_cast_fp16")];
            tensor<fp16, [262208]> linear_238_bias_0 = const()[name = tensor<string, []>("linear_238_bias_0"), val = tensor<fp16, [262208]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1940939648)))];
            tensor<fp16, [1, 256, 262208]> logits = linear(bias = linear_238_bias_0, weight = wrapped_model_language_model_embed_tokens_weight_palettized, x = output_cast_fp16)[name = tensor<string, []>("linear_238")];
        } -> (logits);
}