alexwengg's picture
Upload 5 files
5d8407a verified
program(1.0)
[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3510.2.1"}, {"coremlc-version", "3500.32.1"}, {"coremltools-component-torch", "2.9.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})]
{
func main<ios17>(tensor<fp32, [2, 1, 512, 16, 64]> cache0, tensor<fp32, [2, 1, 512, 16, 64]> cache1, tensor<fp32, [2, 1, 512, 16, 64]> cache2, tensor<fp32, [2, 1, 512, 16, 64]> cache3, tensor<fp32, [2, 1, 512, 16, 64]> cache4, tensor<fp32, [2, 1, 512, 16, 64]> cache5, tensor<fp32, [1, 175, 1024]> conditioning, tensor<fp32, [1]> position0, tensor<fp32, [1]> position1, tensor<fp32, [1]> position2, tensor<fp32, [1]> position3, tensor<fp32, [1]> position4, tensor<fp32, [1]> position5) {
tensor<fp32, [1024]> norm0_1_bias = const()[name = tensor<string, []>("norm0_1_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
tensor<fp32, [1024]> norm0_1_weight = const()[name = tensor<string, []>("norm0_1_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4224)))];
tensor<fp32, [3072, 1024]> attn0_in_proj_weight = const()[name = tensor<string, []>("attn0_in_proj_weight"), val = tensor<fp32, [3072, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8384)))];
tensor<fp32, [1024, 1024]> attn0_out_proj_weight = const()[name = tensor<string, []>("attn0_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12591360)))];
tensor<fp32, [1024]> norm0_2_bias = const()[name = tensor<string, []>("norm0_2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16785728)))];
tensor<fp32, [1024]> norm0_2_weight = const()[name = tensor<string, []>("norm0_2_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16789888)))];
tensor<fp32, [4096, 1024]> linear0_1_weight = const()[name = tensor<string, []>("linear0_1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16794048)))];
tensor<fp32, [1024, 4096]> linear0_2_weight = const()[name = tensor<string, []>("linear0_2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33571328)))];
tensor<fp32, [1024]> norm1_1_bias = const()[name = tensor<string, []>("norm1_1_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(50348608)))];
tensor<fp32, [1024]> norm1_1_weight = const()[name = tensor<string, []>("norm1_1_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(50352768)))];
tensor<fp32, [3072, 1024]> attn1_in_proj_weight = const()[name = tensor<string, []>("attn1_in_proj_weight"), val = tensor<fp32, [3072, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(50356928)))];
tensor<fp32, [1024, 1024]> attn1_out_proj_weight = const()[name = tensor<string, []>("attn1_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62939904)))];
tensor<fp32, [1024]> norm1_2_bias = const()[name = tensor<string, []>("norm1_2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67134272)))];
tensor<fp32, [1024]> norm1_2_weight = const()[name = tensor<string, []>("norm1_2_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67138432)))];
tensor<fp32, [4096, 1024]> linear1_1_weight = const()[name = tensor<string, []>("linear1_1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67142592)))];
tensor<fp32, [1024, 4096]> linear1_2_weight = const()[name = tensor<string, []>("linear1_2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83919872)))];
tensor<fp32, [1024]> norm2_1_bias = const()[name = tensor<string, []>("norm2_1_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100697152)))];
tensor<fp32, [1024]> norm2_1_weight = const()[name = tensor<string, []>("norm2_1_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100701312)))];
tensor<fp32, [3072, 1024]> attn2_in_proj_weight = const()[name = tensor<string, []>("attn2_in_proj_weight"), val = tensor<fp32, [3072, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100705472)))];
tensor<fp32, [1024, 1024]> attn2_out_proj_weight = const()[name = tensor<string, []>("attn2_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113288448)))];
tensor<fp32, [1024]> norm2_2_bias = const()[name = tensor<string, []>("norm2_2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117482816)))];
tensor<fp32, [1024]> norm2_2_weight = const()[name = tensor<string, []>("norm2_2_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117486976)))];
tensor<fp32, [4096, 1024]> linear2_1_weight = const()[name = tensor<string, []>("linear2_1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117491136)))];
tensor<fp32, [1024, 4096]> linear2_2_weight = const()[name = tensor<string, []>("linear2_2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134268416)))];
tensor<fp32, [1024]> norm3_1_bias = const()[name = tensor<string, []>("norm3_1_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151045696)))];
tensor<fp32, [1024]> norm3_1_weight = const()[name = tensor<string, []>("norm3_1_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151049856)))];
tensor<fp32, [3072, 1024]> attn3_in_proj_weight = const()[name = tensor<string, []>("attn3_in_proj_weight"), val = tensor<fp32, [3072, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151054016)))];
tensor<fp32, [1024, 1024]> attn3_out_proj_weight = const()[name = tensor<string, []>("attn3_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163636992)))];
tensor<fp32, [1024]> norm3_2_bias = const()[name = tensor<string, []>("norm3_2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167831360)))];
tensor<fp32, [1024]> norm3_2_weight = const()[name = tensor<string, []>("norm3_2_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167835520)))];
tensor<fp32, [4096, 1024]> linear3_1_weight = const()[name = tensor<string, []>("linear3_1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167839680)))];
tensor<fp32, [1024, 4096]> linear3_2_weight = const()[name = tensor<string, []>("linear3_2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(184616960)))];
tensor<fp32, [1024]> norm4_1_bias = const()[name = tensor<string, []>("norm4_1_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(201394240)))];
tensor<fp32, [1024]> norm4_1_weight = const()[name = tensor<string, []>("norm4_1_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(201398400)))];
tensor<fp32, [3072, 1024]> attn4_in_proj_weight = const()[name = tensor<string, []>("attn4_in_proj_weight"), val = tensor<fp32, [3072, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(201402560)))];
tensor<fp32, [1024, 1024]> attn4_out_proj_weight = const()[name = tensor<string, []>("attn4_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213985536)))];
tensor<fp32, [1024]> norm4_2_bias = const()[name = tensor<string, []>("norm4_2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(218179904)))];
tensor<fp32, [1024]> norm4_2_weight = const()[name = tensor<string, []>("norm4_2_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(218184064)))];
tensor<fp32, [4096, 1024]> linear4_1_weight = const()[name = tensor<string, []>("linear4_1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(218188224)))];
tensor<fp32, [1024, 4096]> linear4_2_weight = const()[name = tensor<string, []>("linear4_2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(234965504)))];
tensor<fp32, [1024]> norm5_1_bias = const()[name = tensor<string, []>("norm5_1_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(251742784)))];
tensor<fp32, [1024]> norm5_1_weight = const()[name = tensor<string, []>("norm5_1_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(251746944)))];
tensor<fp32, [3072, 1024]> attn5_in_proj_weight = const()[name = tensor<string, []>("attn5_in_proj_weight"), val = tensor<fp32, [3072, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(251751104)))];
tensor<fp32, []> var_47 = const()[name = tensor<string, []>("op_47"), val = tensor<fp32, []>(0x1.4f8b58p-17)];
tensor<int32, [1]> x_1_axes_0 = const()[name = tensor<string, []>("x_1_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [1, 175, 1024]> x_1 = layer_norm(axes = x_1_axes_0, beta = norm0_1_bias, epsilon = var_47, gamma = norm0_1_weight, x = conditioning)[name = tensor<string, []>("x_1")];
tensor<fp32, [3072]> linear_0_bias_0 = const()[name = tensor<string, []>("linear_0_bias_0"), val = tensor<fp32, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(264334080)))];
tensor<fp32, [1, 175, 3072]> var_79 = linear(bias = linear_0_bias_0, weight = attn0_in_proj_weight, x = x_1)[name = tensor<string, []>("linear_0")];
tensor<int32, [5]> var_83 = const()[name = tensor<string, []>("op_83"), val = tensor<int32, [5]>([1, 175, 3, 16, 64])];
tensor<fp32, [1, 175, 3, 16, 64]> qkv_1 = reshape(shape = var_83, x = var_79)[name = tensor<string, []>("qkv_1")];
tensor<int32, [5]> q_1_begin_0 = const()[name = tensor<string, []>("q_1_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> q_1_end_0 = const()[name = tensor<string, []>("q_1_end_0"), val = tensor<int32, [5]>([1, 175, 1, 16, 64])];
tensor<bool, [5]> q_1_end_mask_0 = const()[name = tensor<string, []>("q_1_end_mask_0"), val = tensor<bool, [5]>([true, true, false, true, true])];
tensor<bool, [5]> q_1_squeeze_mask_0 = const()[name = tensor<string, []>("q_1_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, true, false, false])];
tensor<fp32, [1, 175, 16, 64]> q_1 = slice_by_index(begin = q_1_begin_0, end = q_1_end_0, end_mask = q_1_end_mask_0, squeeze_mask = q_1_squeeze_mask_0, x = qkv_1)[name = tensor<string, []>("q_1")];
tensor<int32, [5]> k_1_begin_0 = const()[name = tensor<string, []>("k_1_begin_0"), val = tensor<int32, [5]>([0, 0, 1, 0, 0])];
tensor<int32, [5]> k_1_end_0 = const()[name = tensor<string, []>("k_1_end_0"), val = tensor<int32, [5]>([1, 175, 2, 16, 64])];
tensor<bool, [5]> k_1_end_mask_0 = const()[name = tensor<string, []>("k_1_end_mask_0"), val = tensor<bool, [5]>([true, true, false, true, true])];
tensor<bool, [5]> k_1_squeeze_mask_0 = const()[name = tensor<string, []>("k_1_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, true, false, false])];
tensor<fp32, [1, 175, 16, 64]> k_1 = slice_by_index(begin = k_1_begin_0, end = k_1_end_0, end_mask = k_1_end_mask_0, squeeze_mask = k_1_squeeze_mask_0, x = qkv_1)[name = tensor<string, []>("k_1")];
tensor<int32, [5]> v_1_begin_0 = const()[name = tensor<string, []>("v_1_begin_0"), val = tensor<int32, [5]>([0, 0, 2, 0, 0])];
tensor<int32, [5]> v_1_end_0 = const()[name = tensor<string, []>("v_1_end_0"), val = tensor<int32, [5]>([1, 175, 3, 16, 64])];
tensor<bool, [5]> v_1_end_mask_0 = const()[name = tensor<string, []>("v_1_end_mask_0"), val = tensor<bool, [5]>([true, true, false, true, true])];
tensor<bool, [5]> v_1_squeeze_mask_0 = const()[name = tensor<string, []>("v_1_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, true, false, false])];
tensor<fp32, [1, 175, 16, 64]> v_1 = slice_by_index(begin = v_1_begin_0, end = v_1_end_0, end_mask = v_1_end_mask_0, squeeze_mask = v_1_squeeze_mask_0, x = qkv_1)[name = tensor<string, []>("v_1")];
tensor<fp32, [32]> freqs_1 = const()[name = tensor<string, []>("freqs_1"), val = tensor<fp32, [32]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(264346432)))];
tensor<int32, [2]> var_181 = const()[name = tensor<string, []>("op_181"), val = tensor<int32, [2]>([1, 1])];
tensor<fp32, [1, 1]> var_182 = reshape(shape = var_181, x = position0)[name = tensor<string, []>("op_182")];
tensor<fp32, [175]> ts_1_promoted = const()[name = tensor<string, []>("ts_1_promoted"), val = tensor<fp32, [175]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(264346624)))];
tensor<fp32, [1, 175]> ts_3 = add(x = ts_1_promoted, y = var_182)[name = tensor<string, []>("ts_3")];
tensor<int32, [4]> var_187 = const()[name = tensor<string, []>("op_187"), val = tensor<int32, [4]>([1, 175, 1, 1])];
tensor<fp32, [1, 175, 1, 1]> ts_5 = reshape(shape = var_187, x = ts_3)[name = tensor<string, []>("ts_5")];
tensor<int32, [5]> var_191 = const()[name = tensor<string, []>("op_191"), val = tensor<int32, [5]>([1, 175, 16, 32, 2])];
tensor<fp32, [1, 175, 16, 32, 2]> q_complex_1 = reshape(shape = var_191, x = q_1)[name = tensor<string, []>("q_complex_1")];
tensor<int32, [5]> var_195 = const()[name = tensor<string, []>("op_195"), val = tensor<int32, [5]>([1, 175, 16, 32, 2])];
tensor<fp32, [1, 175, 16, 32, 2]> k_complex_1 = reshape(shape = var_195, x = k_1)[name = tensor<string, []>("k_complex_1")];
tensor<int32, [5]> var_199_begin_0 = const()[name = tensor<string, []>("op_199_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> var_199_end_0 = const()[name = tensor<string, []>("op_199_end_0"), val = tensor<int32, [5]>([1, 175, 16, 32, 1])];
tensor<bool, [5]> var_199_end_mask_0 = const()[name = tensor<string, []>("op_199_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
tensor<bool, [5]> var_199_squeeze_mask_0 = const()[name = tensor<string, []>("op_199_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
tensor<fp32, [1, 175, 16, 32]> var_199 = slice_by_index(begin = var_199_begin_0, end = var_199_end_0, end_mask = var_199_end_mask_0, squeeze_mask = var_199_squeeze_mask_0, x = q_complex_1)[name = tensor<string, []>("op_199")];
tensor<int32, [5]> var_207_begin_0 = const()[name = tensor<string, []>("op_207_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 1])];
tensor<int32, [5]> var_207_end_0 = const()[name = tensor<string, []>("op_207_end_0"), val = tensor<int32, [5]>([1, 175, 16, 32, 2])];
tensor<bool, [5]> var_207_end_mask_0 = const()[name = tensor<string, []>("op_207_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
tensor<bool, [5]> var_207_squeeze_mask_0 = const()[name = tensor<string, []>("op_207_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
tensor<fp32, [1, 175, 16, 32]> var_207 = slice_by_index(begin = var_207_begin_0, end = var_207_end_0, end_mask = var_207_end_mask_0, squeeze_mask = var_207_squeeze_mask_0, x = q_complex_1)[name = tensor<string, []>("op_207")];
tensor<int32, [5]> var_215_begin_0 = const()[name = tensor<string, []>("op_215_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> var_215_end_0 = const()[name = tensor<string, []>("op_215_end_0"), val = tensor<int32, [5]>([1, 175, 16, 32, 1])];
tensor<bool, [5]> var_215_end_mask_0 = const()[name = tensor<string, []>("op_215_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
tensor<bool, [5]> var_215_squeeze_mask_0 = const()[name = tensor<string, []>("op_215_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
tensor<fp32, [1, 175, 16, 32]> var_215 = slice_by_index(begin = var_215_begin_0, end = var_215_end_0, end_mask = var_215_end_mask_0, squeeze_mask = var_215_squeeze_mask_0, x = k_complex_1)[name = tensor<string, []>("op_215")];
tensor<int32, [5]> var_223_begin_0 = const()[name = tensor<string, []>("op_223_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 1])];
tensor<int32, [5]> var_223_end_0 = const()[name = tensor<string, []>("op_223_end_0"), val = tensor<int32, [5]>([1, 175, 16, 32, 2])];
tensor<bool, [5]> var_223_end_mask_0 = const()[name = tensor<string, []>("op_223_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
tensor<bool, [5]> var_223_squeeze_mask_0 = const()[name = tensor<string, []>("op_223_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
tensor<fp32, [1, 175, 16, 32]> var_223 = slice_by_index(begin = var_223_begin_0, end = var_223_end_0, end_mask = var_223_end_mask_0, squeeze_mask = var_223_squeeze_mask_0, x = k_complex_1)[name = tensor<string, []>("op_223")];
tensor<fp32, [1, 175, 1, 32]> var_229 = mul(x = freqs_1, y = ts_5)[name = tensor<string, []>("op_229")];
tensor<fp32, [1, 175, 1, 32]> rotr_1 = cos(x = var_229)[name = tensor<string, []>("rotr_1")];
tensor<fp32, [1, 175, 1, 32]> roti_1 = sin(x = var_229)[name = tensor<string, []>("roti_1")];
tensor<fp32, [1, 175, 16, 32]> var_233 = mul(x = var_199, y = rotr_1)[name = tensor<string, []>("op_233")];
tensor<fp32, [1, 175, 16, 32]> var_234 = mul(x = var_207, y = roti_1)[name = tensor<string, []>("op_234")];
tensor<fp32, [1, 175, 16, 32]> qor_1 = sub(x = var_233, y = var_234)[name = tensor<string, []>("qor_1")];
tensor<fp32, [1, 175, 16, 32]> var_237 = mul(x = var_199, y = roti_1)[name = tensor<string, []>("op_237")];
tensor<fp32, [1, 175, 16, 32]> var_238 = mul(x = var_207, y = rotr_1)[name = tensor<string, []>("op_238")];
tensor<fp32, [1, 175, 16, 32]> qoi_1 = add(x = var_237, y = var_238)[name = tensor<string, []>("qoi_1")];
tensor<fp32, [1, 175, 16, 32]> var_241 = mul(x = var_215, y = rotr_1)[name = tensor<string, []>("op_241")];
tensor<fp32, [1, 175, 16, 32]> var_242 = mul(x = var_223, y = roti_1)[name = tensor<string, []>("op_242")];
tensor<fp32, [1, 175, 16, 32]> kor_1 = sub(x = var_241, y = var_242)[name = tensor<string, []>("kor_1")];
tensor<fp32, [1, 175, 16, 32]> var_245 = mul(x = var_215, y = roti_1)[name = tensor<string, []>("op_245")];
tensor<fp32, [1, 175, 16, 32]> var_246 = mul(x = var_223, y = rotr_1)[name = tensor<string, []>("op_246")];
tensor<fp32, [1, 175, 16, 32]> koi_1 = add(x = var_245, y = var_246)[name = tensor<string, []>("koi_1")];
tensor<int32, []> qo_1_axis_0 = const()[name = tensor<string, []>("qo_1_axis_0"), val = tensor<int32, []>(-1)];
tensor<fp32, [1, 175, 16, 32, 2]> qo_1 = stack(axis = qo_1_axis_0, values = (qor_1, qoi_1))[name = tensor<string, []>("qo_1")];
tensor<int32, []> ko_1_axis_0 = const()[name = tensor<string, []>("ko_1_axis_0"), val = tensor<int32, []>(-1)];
tensor<fp32, [1, 175, 16, 32, 2]> ko_1 = stack(axis = ko_1_axis_0, values = (kor_1, koi_1))[name = tensor<string, []>("ko_1")];
tensor<int32, [4]> var_275 = const()[name = tensor<string, []>("op_275"), val = tensor<int32, [4]>([1, 175, 16, 64])];
tensor<fp32, [1, 175, 16, 64]> q_3 = reshape(shape = var_275, x = qo_1)[name = tensor<string, []>("q_3")];
tensor<int32, [4]> var_277 = const()[name = tensor<string, []>("op_277"), val = tensor<int32, [4]>([1, 175, 16, 64])];
tensor<fp32, [1, 175, 16, 64]> k_3 = reshape(shape = var_277, x = ko_1)[name = tensor<string, []>("k_3")];
tensor<int32, [4]> var_284 = const()[name = tensor<string, []>("op_284"), val = tensor<int32, [4]>([1, 1, 1, 1])];
tensor<fp32, [1, 1, 1, 1]> write_base_float_1 = reshape(shape = var_284, x = position0)[name = tensor<string, []>("write_base_float_1")];
tensor<fp32, [1, 175, 1, 1]> write_offsets_float_1_promoted = const()[name = tensor<string, []>("write_offsets_float_1_promoted"), val = tensor<fp32, [1, 175, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(264347392)))];
tensor<fp32, [1, 175, 1, 1]> write_indices_float_1 = add(x = write_base_float_1, y = write_offsets_float_1_promoted)[name = tensor<string, []>("write_indices_float_1")];
tensor<fp32, []> _inversed_299_y_0 = const()[name = tensor<string, []>("_inversed_299_y_0"), val = tensor<fp32, []>(0x1p-9)];
tensor<fp32, [1, 175, 1, 1]> _inversed_299 = mul(x = write_indices_float_1, y = _inversed_299_y_0)[name = tensor<string, []>("_inversed_299")];
tensor<fp32, [1, 175, 1, 1]> var_300 = floor(x = _inversed_299)[name = tensor<string, []>("op_300")];
tensor<fp32, []> var_301 = const()[name = tensor<string, []>("op_301"), val = tensor<fp32, []>(0x1p+9)];
tensor<fp32, [1, 175, 1, 1]> var_302 = mul(x = var_300, y = var_301)[name = tensor<string, []>("op_302")];
tensor<fp32, [1, 175, 1, 1]> write_indices_float_3 = sub(x = write_indices_float_1, y = var_302)[name = tensor<string, []>("write_indices_float_3")];
tensor<string, []> var_309_dtype_0 = const()[name = tensor<string, []>("op_309_dtype_0"), val = tensor<string, []>("int32")];
tensor<int32, [4]> write_indices_1_reps_0 = const()[name = tensor<string, []>("write_indices_1_reps_0"), val = tensor<int32, [4]>([1, 1, 16, 64])];
tensor<int32, [1, 175, 1, 1]> var_309 = cast(dtype = var_309_dtype_0, x = write_indices_float_3)[name = tensor<string, []>("cast_104")];
tensor<int32, [1, 175, 16, 64]> write_indices_1 = tile(reps = write_indices_1_reps_0, x = var_309)[name = tensor<string, []>("write_indices_1")];
tensor<int32, [5]> var_317_begin_0 = const()[name = tensor<string, []>("op_317_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> var_317_end_0 = const()[name = tensor<string, []>("op_317_end_0"), val = tensor<int32, [5]>([1, 1, 512, 16, 64])];
tensor<bool, [5]> var_317_end_mask_0 = const()[name = tensor<string, []>("op_317_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> var_317_squeeze_mask_0 = const()[name = tensor<string, []>("op_317_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<fp32, [1, 512, 16, 64]> var_317 = slice_by_index(begin = var_317_begin_0, end = var_317_end_0, end_mask = var_317_end_mask_0, squeeze_mask = var_317_squeeze_mask_0, x = cache0)[name = tensor<string, []>("op_317")];
tensor<int32, []> var_319_axis_0 = const()[name = tensor<string, []>("op_319_axis_0"), val = tensor<int32, []>(1)];
tensor<string, []> var_319_mode_0 = const()[name = tensor<string, []>("op_319_mode_0"), val = tensor<string, []>("update")];
tensor<bool, []> var_319_validate_indices_0 = const()[name = tensor<string, []>("op_319_validate_indices_0"), val = tensor<bool, []>(false)];
tensor<fp32, [1, 512, 16, 64]> var_319 = scatter_along_axis(axis = var_319_axis_0, data = var_317, indices = write_indices_1, mode = var_319_mode_0, updates = k_3, validate_indices = var_319_validate_indices_0)[name = tensor<string, []>("op_319")];
tensor<int32, [5]> concat_1 = const()[name = tensor<string, []>("concat_1"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> concat_2 = const()[name = tensor<string, []>("concat_2"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> new_cache_1_internal_tensor_assign_1_stride_0 = const()[name = tensor<string, []>("new_cache_1_internal_tensor_assign_1_stride_0"), val = tensor<int32, [5]>([1, 1, 1, 1, 1])];
tensor<bool, [5]> new_cache_1_internal_tensor_assign_1_begin_mask_0 = const()[name = tensor<string, []>("new_cache_1_internal_tensor_assign_1_begin_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> new_cache_1_internal_tensor_assign_1_end_mask_0 = const()[name = tensor<string, []>("new_cache_1_internal_tensor_assign_1_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> new_cache_1_internal_tensor_assign_1_squeeze_mask_0 = const()[name = tensor<string, []>("new_cache_1_internal_tensor_assign_1_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<int32, [5]> shape_10 = const()[name = tensor<string, []>("shape_10"), val = tensor<int32, [5]>([2, 1, 512, 16, 64])];
tensor<int32, []> reduce_prod_0 = const()[name = tensor<string, []>("reduce_prod_0"), val = tensor<int32, []>(1048576)];
tensor<int32, []> range_1d_0_start_0 = const()[name = tensor<string, []>("range_1d_0_start_0"), val = tensor<int32, []>(0)];
tensor<int32, []> range_1d_0_step_0 = const()[name = tensor<string, []>("range_1d_0_step_0"), val = tensor<int32, []>(1)];
tensor<int32, [1048576]> range_1d_0 = range_1d(end = reduce_prod_0, start = range_1d_0_start_0, step = range_1d_0_step_0)[name = tensor<string, []>("range_1d_0")];
tensor<int32, [2, 1, 512, 16, 64]> reshape_0 = reshape(shape = shape_10, x = range_1d_0)[name = tensor<string, []>("reshape_0")];
tensor<int32, [1, 512, 16, 64]> slice_by_index_0 = slice_by_index(begin = concat_1, begin_mask = new_cache_1_internal_tensor_assign_1_begin_mask_0, end = concat_2, end_mask = new_cache_1_internal_tensor_assign_1_end_mask_0, squeeze_mask = new_cache_1_internal_tensor_assign_1_squeeze_mask_0, stride = new_cache_1_internal_tensor_assign_1_stride_0, x = reshape_0)[name = tensor<string, []>("slice_by_index_0")];
tensor<int32, [1]> reshape_1_shape_0 = const()[name = tensor<string, []>("reshape_1_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<int32, [524288]> reshape_1 = reshape(shape = reshape_1_shape_0, x = slice_by_index_0)[name = tensor<string, []>("reshape_1")];
tensor<int32, [1]> reshape_2_shape_0 = const()[name = tensor<string, []>("reshape_2_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [524288]> reshape_2 = reshape(shape = reshape_2_shape_0, x = var_319)[name = tensor<string, []>("reshape_2")];
tensor<int32, [1]> reshape_3_shape_0 = const()[name = tensor<string, []>("reshape_3_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [1048576]> reshape_3 = reshape(shape = reshape_3_shape_0, x = cache0)[name = tensor<string, []>("reshape_3")];
tensor<string, []> scatter_0_mode_0 = const()[name = tensor<string, []>("scatter_0_mode_0"), val = tensor<string, []>("update")];
tensor<int32, []> scatter_0_axis_0 = const()[name = tensor<string, []>("scatter_0_axis_0"), val = tensor<int32, []>(0)];
tensor<bool, []> scatter_0_validate_indices_0 = const()[name = tensor<string, []>("scatter_0_validate_indices_0"), val = tensor<bool, []>(false)];
tensor<fp32, [1048576]> scatter_0 = scatter(axis = scatter_0_axis_0, data = reshape_3, indices = reshape_1, mode = scatter_0_mode_0, updates = reshape_2, validate_indices = scatter_0_validate_indices_0)[name = tensor<string, []>("scatter_0")];
tensor<fp32, [2, 1, 512, 16, 64]> reshape_4 = reshape(shape = shape_10, x = scatter_0)[name = tensor<string, []>("reshape_4")];
tensor<int32, [5]> var_327_begin_0 = const()[name = tensor<string, []>("op_327_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
tensor<int32, [5]> var_327_end_0 = const()[name = tensor<string, []>("op_327_end_0"), val = tensor<int32, [5]>([2, 1, 512, 16, 64])];
tensor<bool, [5]> var_327_end_mask_0 = const()[name = tensor<string, []>("op_327_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> var_327_squeeze_mask_0 = const()[name = tensor<string, []>("op_327_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<fp32, [1, 512, 16, 64]> var_327 = slice_by_index(begin = var_327_begin_0, end = var_327_end_0, end_mask = var_327_end_mask_0, squeeze_mask = var_327_squeeze_mask_0, x = reshape_4)[name = tensor<string, []>("op_327")];
tensor<int32, []> var_329_axis_0 = const()[name = tensor<string, []>("op_329_axis_0"), val = tensor<int32, []>(1)];
tensor<string, []> var_329_mode_0 = const()[name = tensor<string, []>("op_329_mode_0"), val = tensor<string, []>("update")];
tensor<bool, []> var_329_validate_indices_0 = const()[name = tensor<string, []>("op_329_validate_indices_0"), val = tensor<bool, []>(false)];
tensor<fp32, [1, 512, 16, 64]> var_329 = scatter_along_axis(axis = var_329_axis_0, data = var_327, indices = write_indices_1, mode = var_329_mode_0, updates = v_1, validate_indices = var_329_validate_indices_0)[name = tensor<string, []>("op_329")];
tensor<int32, [5]> concat_3 = const()[name = tensor<string, []>("concat_3"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
tensor<int32, [5]> concat_4 = const()[name = tensor<string, []>("concat_4"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> new_cache_1_internal_tensor_assign_2_stride_0 = const()[name = tensor<string, []>("new_cache_1_internal_tensor_assign_2_stride_0"), val = tensor<int32, [5]>([1, 1, 1, 1, 1])];
tensor<bool, [5]> new_cache_1_internal_tensor_assign_2_begin_mask_0 = const()[name = tensor<string, []>("new_cache_1_internal_tensor_assign_2_begin_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> new_cache_1_internal_tensor_assign_2_end_mask_0 = const()[name = tensor<string, []>("new_cache_1_internal_tensor_assign_2_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> new_cache_1_internal_tensor_assign_2_squeeze_mask_0 = const()[name = tensor<string, []>("new_cache_1_internal_tensor_assign_2_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<int32, [5]> shape_11 = const()[name = tensor<string, []>("shape_11"), val = tensor<int32, [5]>([2, 1, 512, 16, 64])];
tensor<int32, []> reduce_prod_1 = const()[name = tensor<string, []>("reduce_prod_1"), val = tensor<int32, []>(1048576)];
tensor<int32, []> range_1d_1_start_0 = const()[name = tensor<string, []>("range_1d_1_start_0"), val = tensor<int32, []>(0)];
tensor<int32, []> range_1d_1_step_0 = const()[name = tensor<string, []>("range_1d_1_step_0"), val = tensor<int32, []>(1)];
tensor<int32, [1048576]> range_1d_1 = range_1d(end = reduce_prod_1, start = range_1d_1_start_0, step = range_1d_1_step_0)[name = tensor<string, []>("range_1d_1")];
tensor<int32, [2, 1, 512, 16, 64]> reshape_5 = reshape(shape = shape_11, x = range_1d_1)[name = tensor<string, []>("reshape_5")];
tensor<int32, [1, 512, 16, 64]> slice_by_index_1 = slice_by_index(begin = concat_3, begin_mask = new_cache_1_internal_tensor_assign_2_begin_mask_0, end = concat_4, end_mask = new_cache_1_internal_tensor_assign_2_end_mask_0, squeeze_mask = new_cache_1_internal_tensor_assign_2_squeeze_mask_0, stride = new_cache_1_internal_tensor_assign_2_stride_0, x = reshape_5)[name = tensor<string, []>("slice_by_index_1")];
tensor<int32, [1]> reshape_6_shape_0 = const()[name = tensor<string, []>("reshape_6_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<int32, [524288]> reshape_6 = reshape(shape = reshape_6_shape_0, x = slice_by_index_1)[name = tensor<string, []>("reshape_6")];
tensor<int32, [1]> reshape_7_shape_0 = const()[name = tensor<string, []>("reshape_7_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [524288]> reshape_7 = reshape(shape = reshape_7_shape_0, x = var_329)[name = tensor<string, []>("reshape_7")];
tensor<int32, [1]> reshape_8_shape_0 = const()[name = tensor<string, []>("reshape_8_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [1048576]> reshape_8 = reshape(shape = reshape_8_shape_0, x = reshape_4)[name = tensor<string, []>("reshape_8")];
tensor<string, []> scatter_1_mode_0 = const()[name = tensor<string, []>("scatter_1_mode_0"), val = tensor<string, []>("update")];
tensor<int32, []> scatter_1_axis_0 = const()[name = tensor<string, []>("scatter_1_axis_0"), val = tensor<int32, []>(0)];
tensor<bool, []> scatter_1_validate_indices_0 = const()[name = tensor<string, []>("scatter_1_validate_indices_0"), val = tensor<bool, []>(false)];
tensor<fp32, [1048576]> scatter_1 = scatter(axis = scatter_1_axis_0, data = reshape_8, indices = reshape_6, mode = scatter_1_mode_0, updates = reshape_7, validate_indices = scatter_1_validate_indices_0)[name = tensor<string, []>("scatter_1")];
tensor<fp32, [2, 1, 512, 16, 64]> new_cache_1_internal_tensor_assign_2 = reshape(shape = shape_11, x = scatter_1)[name = tensor<string, []>("reshape_9")];
tensor<int32, [5]> keys_1_begin_0 = const()[name = tensor<string, []>("keys_1_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> keys_1_end_0 = const()[name = tensor<string, []>("keys_1_end_0"), val = tensor<int32, [5]>([1, 1, 512, 16, 64])];
tensor<bool, [5]> keys_1_end_mask_0 = const()[name = tensor<string, []>("keys_1_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> keys_1_squeeze_mask_0 = const()[name = tensor<string, []>("keys_1_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<fp32, [1, 512, 16, 64]> keys_1 = slice_by_index(begin = keys_1_begin_0, end = keys_1_end_0, end_mask = keys_1_end_mask_0, squeeze_mask = keys_1_squeeze_mask_0, x = new_cache_1_internal_tensor_assign_2)[name = tensor<string, []>("keys_1")];
tensor<int32, [5]> values_1_begin_0 = const()[name = tensor<string, []>("values_1_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
tensor<int32, [5]> values_1_end_0 = const()[name = tensor<string, []>("values_1_end_0"), val = tensor<int32, [5]>([2, 1, 512, 16, 64])];
tensor<bool, [5]> values_1_end_mask_0 = const()[name = tensor<string, []>("values_1_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> values_1_squeeze_mask_0 = const()[name = tensor<string, []>("values_1_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<fp32, [1, 512, 16, 64]> values_1 = slice_by_index(begin = values_1_begin_0, end = values_1_end_0, end_mask = values_1_end_mask_0, squeeze_mask = values_1_squeeze_mask_0, x = new_cache_1_internal_tensor_assign_2)[name = tensor<string, []>("values_1")];
tensor<bool, [1, 512, 16, 64]> var_341 = not_equal(x = keys_1, y = keys_1)[name = tensor<string, []>("op_341")];
tensor<fp32, [1, 512, 16, 64]> var_347 = const()[name = tensor<string, []>("op_347"), val = tensor<fp32, [1, 512, 16, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(264348160)))];
tensor<fp32, [1, 512, 16, 64]> keys_3 = select(a = var_347, b = keys_1, cond = var_341)[name = tensor<string, []>("keys_3")];
tensor<bool, [1, 512, 16, 64]> var_349 = not_equal(x = values_1, y = values_1)[name = tensor<string, []>("op_349")];
tensor<fp32, [1, 512, 16, 64]> values_3 = select(a = var_347, b = values_1, cond = var_349)[name = tensor<string, []>("values_3")];
tensor<int32, [4]> var_373 = const()[name = tensor<string, []>("op_373"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_386 = const()[name = tensor<string, []>("op_386"), val = tensor<int32, [3]>([1, 1, 1])];
tensor<fp32, [1, 1, 1]> var_387 = reshape(shape = var_386, x = position0)[name = tensor<string, []>("op_387")];
tensor<fp32, [1, 175, 1]> q_offsets_1_promoted = const()[name = tensor<string, []>("q_offsets_1_promoted"), val = tensor<fp32, [1, 175, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(266445376)))];
tensor<fp32, [1, 175, 1]> q_positions_1 = add(x = var_387, y = q_offsets_1_promoted)[name = tensor<string, []>("q_positions_1")];
tensor<fp32, []> var_404 = const()[name = tensor<string, []>("op_404"), val = tensor<fp32, []>(0x1.5ep+7)];
tensor<fp32, [1, 1, 1]> valid_len_1 = add(x = var_387, y = var_404)[name = tensor<string, []>("valid_len_1")];
tensor<fp32, [1, 1, 512]> k_positions_1_promoted = const()[name = tensor<string, []>("k_positions_1_promoted"), val = tensor<fp32, [1, 1, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(266446144)))];
tensor<bool, [1, 1, 512]> valid_mask_1 = less(x = k_positions_1_promoted, y = valid_len_1)[name = tensor<string, []>("valid_mask_1")];
tensor<bool, [1, 175, 512]> causal_mask_1 = less_equal(x = k_positions_1_promoted, y = q_positions_1)[name = tensor<string, []>("causal_mask_1")];
tensor<bool, [1, 175, 512]> attn_mask_1 = logical_and(x = valid_mask_1, y = causal_mask_1)[name = tensor<string, []>("attn_mask_1")];
tensor<int32, [1]> attn_mask_3_axes_0 = const()[name = tensor<string, []>("attn_mask_3_axes_0"), val = tensor<int32, [1]>([1])];
tensor<bool, [1, 1, 175, 512]> attn_mask_3 = expand_dims(axes = attn_mask_3_axes_0, x = attn_mask_1)[name = tensor<string, []>("attn_mask_3")];
tensor<fp32, [1]> var_416 = const()[name = tensor<string, []>("op_416"), val = tensor<fp32, [1]>([0x1.fffe5cp-4])];
tensor<bool, []> var_422_transpose_x_0 = const()[name = tensor<string, []>("op_422_transpose_x_0"), val = tensor<bool, []>(false)];
tensor<bool, []> var_422_transpose_y_0 = const()[name = tensor<string, []>("op_422_transpose_y_0"), val = tensor<bool, []>(false)];
tensor<int32, [4]> transpose_15_perm_0 = const()[name = tensor<string, []>("transpose_15_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
tensor<int32, [4]> transpose_16_perm_0 = const()[name = tensor<string, []>("transpose_16_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
tensor<fp32, [1, 16, 64, 512]> transpose_16 = transpose(perm = transpose_16_perm_0, x = keys_3)[name = tensor<string, []>("transpose_42")];
tensor<fp32, [1, 16, 175, 64]> transpose_15 = transpose(perm = transpose_15_perm_0, x = q_3)[name = tensor<string, []>("transpose_43")];
tensor<fp32, [1, 16, 175, 512]> var_422 = matmul(transpose_x = var_422_transpose_x_0, transpose_y = var_422_transpose_y_0, x = transpose_15, y = transpose_16)[name = tensor<string, []>("op_422")];
tensor<fp32, [1, 16, 175, 512]> attn_weights_1 = mul(x = var_422, y = var_416)[name = tensor<string, []>("attn_weights_1")];
tensor<bool, [1, 1, 175, 512]> var_424 = logical_not(x = attn_mask_3)[name = tensor<string, []>("op_424")];
tensor<fp32, []> var_425 = const()[name = tensor<string, []>("op_425"), val = tensor<fp32, []>(-0x1.ff933cp+127)];
tensor<fp32, [1, 16, 175, 512]> attn_weights_3 = select(a = var_425, b = attn_weights_1, cond = var_424)[name = tensor<string, []>("attn_weights_3")];
tensor<int32, []> var_427 = const()[name = tensor<string, []>("op_427"), val = tensor<int32, []>(-1)];
tensor<fp32, [1, 16, 175, 512]> attn_weights_5 = softmax(axis = var_427, x = attn_weights_3)[name = tensor<string, []>("attn_weights_5")];
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<fp32, [1, 16, 512, 64]> values_5 = transpose(perm = var_373, x = values_3)[name = tensor<string, []>("transpose_44")];
tensor<fp32, [1, 16, 175, 64]> attn_output_1 = matmul(transpose_x = attn_output_1_transpose_x_0, transpose_y = attn_output_1_transpose_y_0, x = attn_weights_5, y = values_5)[name = tensor<string, []>("attn_output_1")];
tensor<int32, [4]> var_435 = const()[name = tensor<string, []>("op_435"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_438 = const()[name = tensor<string, []>("op_438"), val = tensor<int32, [3]>([1, 175, 1024])];
tensor<fp32, [1, 175, 16, 64]> var_436 = transpose(perm = var_435, x = attn_output_1)[name = tensor<string, []>("transpose_41")];
tensor<fp32, [1, 175, 1024]> input_1 = reshape(shape = var_438, x = var_436)[name = tensor<string, []>("input_1")];
tensor<fp32, [1024]> linear_1_bias_0 = const()[name = tensor<string, []>("linear_1_bias_0"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(266448256)))];
tensor<fp32, [1, 175, 1024]> attn_out_1 = linear(bias = linear_1_bias_0, weight = attn0_out_proj_weight, x = input_1)[name = tensor<string, []>("linear_1")];
tensor<fp32, []> var_444 = const()[name = tensor<string, []>("op_444"), val = tensor<fp32, []>(0x1.5ep+7)];
tensor<fp32, [1]> var_445 = add(x = position0, y = var_444)[name = tensor<string, []>("op_445")];
tensor<fp32, [1, 175, 1024]> input_3 = add(x = conditioning, y = attn_out_1)[name = tensor<string, []>("input_3")];
tensor<fp32, []> var_449 = const()[name = tensor<string, []>("op_449"), val = tensor<fp32, []>(0x1.4f8b58p-17)];
tensor<int32, [1]> input_5_axes_0 = const()[name = tensor<string, []>("input_5_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [1, 175, 1024]> input_5 = layer_norm(axes = input_5_axes_0, beta = norm0_2_bias, epsilon = var_449, gamma = norm0_2_weight, x = input_3)[name = tensor<string, []>("input_5")];
tensor<fp32, [4096]> linear_2_bias_0 = const()[name = tensor<string, []>("linear_2_bias_0"), val = tensor<fp32, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(266452416)))];
tensor<fp32, [1, 175, 4096]> var_457 = linear(bias = linear_2_bias_0, weight = linear0_1_weight, x = input_5)[name = tensor<string, []>("linear_2")];
tensor<string, []> input_7_mode_0 = const()[name = tensor<string, []>("input_7_mode_0"), val = tensor<string, []>("EXACT")];
tensor<fp32, [1, 175, 4096]> input_7 = gelu(mode = input_7_mode_0, x = var_457)[name = tensor<string, []>("input_7")];
tensor<fp32, [1, 175, 1024]> ffn_out_1 = linear(bias = linear_1_bias_0, weight = linear0_2_weight, x = input_7)[name = tensor<string, []>("linear_3")];
tensor<fp32, [1, 175, 1024]> input_9 = add(x = input_3, y = ffn_out_1)[name = tensor<string, []>("input_9")];
tensor<fp32, []> var_466 = const()[name = tensor<string, []>("op_466"), val = tensor<fp32, []>(0x1.4f8b58p-17)];
tensor<int32, [1]> x_3_axes_0 = const()[name = tensor<string, []>("x_3_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [1, 175, 1024]> x_3 = layer_norm(axes = x_3_axes_0, beta = norm1_1_bias, epsilon = var_466, gamma = norm1_1_weight, x = input_9)[name = tensor<string, []>("x_3")];
tensor<fp32, [1, 175, 3072]> var_498 = linear(bias = linear_0_bias_0, weight = attn1_in_proj_weight, x = x_3)[name = tensor<string, []>("linear_4")];
tensor<int32, [5]> var_502 = const()[name = tensor<string, []>("op_502"), val = tensor<int32, [5]>([1, 175, 3, 16, 64])];
tensor<fp32, [1, 175, 3, 16, 64]> qkv_3 = reshape(shape = var_502, x = var_498)[name = tensor<string, []>("qkv_3")];
tensor<int32, [5]> q_7_begin_0 = const()[name = tensor<string, []>("q_7_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> q_7_end_0 = const()[name = tensor<string, []>("q_7_end_0"), val = tensor<int32, [5]>([1, 175, 1, 16, 64])];
tensor<bool, [5]> q_7_end_mask_0 = const()[name = tensor<string, []>("q_7_end_mask_0"), val = tensor<bool, [5]>([true, true, false, true, true])];
tensor<bool, [5]> q_7_squeeze_mask_0 = const()[name = tensor<string, []>("q_7_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, true, false, false])];
tensor<fp32, [1, 175, 16, 64]> q_7 = slice_by_index(begin = q_7_begin_0, end = q_7_end_0, end_mask = q_7_end_mask_0, squeeze_mask = q_7_squeeze_mask_0, x = qkv_3)[name = tensor<string, []>("q_7")];
tensor<int32, [5]> k_5_begin_0 = const()[name = tensor<string, []>("k_5_begin_0"), val = tensor<int32, [5]>([0, 0, 1, 0, 0])];
tensor<int32, [5]> k_5_end_0 = const()[name = tensor<string, []>("k_5_end_0"), val = tensor<int32, [5]>([1, 175, 2, 16, 64])];
tensor<bool, [5]> k_5_end_mask_0 = const()[name = tensor<string, []>("k_5_end_mask_0"), val = tensor<bool, [5]>([true, true, false, true, true])];
tensor<bool, [5]> k_5_squeeze_mask_0 = const()[name = tensor<string, []>("k_5_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, true, false, false])];
tensor<fp32, [1, 175, 16, 64]> k_5 = slice_by_index(begin = k_5_begin_0, end = k_5_end_0, end_mask = k_5_end_mask_0, squeeze_mask = k_5_squeeze_mask_0, x = qkv_3)[name = tensor<string, []>("k_5")];
tensor<int32, [5]> v_3_begin_0 = const()[name = tensor<string, []>("v_3_begin_0"), val = tensor<int32, [5]>([0, 0, 2, 0, 0])];
tensor<int32, [5]> v_3_end_0 = const()[name = tensor<string, []>("v_3_end_0"), val = tensor<int32, [5]>([1, 175, 3, 16, 64])];
tensor<bool, [5]> v_3_end_mask_0 = const()[name = tensor<string, []>("v_3_end_mask_0"), val = tensor<bool, [5]>([true, true, false, true, true])];
tensor<bool, [5]> v_3_squeeze_mask_0 = const()[name = tensor<string, []>("v_3_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, true, false, false])];
tensor<fp32, [1, 175, 16, 64]> v_3 = slice_by_index(begin = v_3_begin_0, end = v_3_end_0, end_mask = v_3_end_mask_0, squeeze_mask = v_3_squeeze_mask_0, x = qkv_3)[name = tensor<string, []>("v_3")];
tensor<fp32, [32]> freqs_3 = const()[name = tensor<string, []>("freqs_3"), val = tensor<fp32, [32]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(266468864)))];
tensor<int32, [2]> var_600 = const()[name = tensor<string, []>("op_600"), val = tensor<int32, [2]>([1, 1])];
tensor<fp32, [1, 1]> var_601 = reshape(shape = var_600, x = position1)[name = tensor<string, []>("op_601")];
tensor<fp32, [1, 175]> ts_9 = add(x = ts_1_promoted, y = var_601)[name = tensor<string, []>("ts_9")];
tensor<int32, [4]> var_606 = const()[name = tensor<string, []>("op_606"), val = tensor<int32, [4]>([1, 175, 1, 1])];
tensor<fp32, [1, 175, 1, 1]> ts_11 = reshape(shape = var_606, x = ts_9)[name = tensor<string, []>("ts_11")];
tensor<int32, [5]> var_610 = const()[name = tensor<string, []>("op_610"), val = tensor<int32, [5]>([1, 175, 16, 32, 2])];
tensor<fp32, [1, 175, 16, 32, 2]> q_complex_3 = reshape(shape = var_610, x = q_7)[name = tensor<string, []>("q_complex_3")];
tensor<int32, [5]> var_614 = const()[name = tensor<string, []>("op_614"), val = tensor<int32, [5]>([1, 175, 16, 32, 2])];
tensor<fp32, [1, 175, 16, 32, 2]> k_complex_3 = reshape(shape = var_614, x = k_5)[name = tensor<string, []>("k_complex_3")];
tensor<int32, [5]> var_618_begin_0 = const()[name = tensor<string, []>("op_618_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> var_618_end_0 = const()[name = tensor<string, []>("op_618_end_0"), val = tensor<int32, [5]>([1, 175, 16, 32, 1])];
tensor<bool, [5]> var_618_end_mask_0 = const()[name = tensor<string, []>("op_618_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
tensor<bool, [5]> var_618_squeeze_mask_0 = const()[name = tensor<string, []>("op_618_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
tensor<fp32, [1, 175, 16, 32]> var_618 = slice_by_index(begin = var_618_begin_0, end = var_618_end_0, end_mask = var_618_end_mask_0, squeeze_mask = var_618_squeeze_mask_0, x = q_complex_3)[name = tensor<string, []>("op_618")];
tensor<int32, [5]> var_626_begin_0 = const()[name = tensor<string, []>("op_626_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 1])];
tensor<int32, [5]> var_626_end_0 = const()[name = tensor<string, []>("op_626_end_0"), val = tensor<int32, [5]>([1, 175, 16, 32, 2])];
tensor<bool, [5]> var_626_end_mask_0 = const()[name = tensor<string, []>("op_626_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
tensor<bool, [5]> var_626_squeeze_mask_0 = const()[name = tensor<string, []>("op_626_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
tensor<fp32, [1, 175, 16, 32]> var_626 = slice_by_index(begin = var_626_begin_0, end = var_626_end_0, end_mask = var_626_end_mask_0, squeeze_mask = var_626_squeeze_mask_0, x = q_complex_3)[name = tensor<string, []>("op_626")];
tensor<int32, [5]> var_634_begin_0 = const()[name = tensor<string, []>("op_634_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> var_634_end_0 = const()[name = tensor<string, []>("op_634_end_0"), val = tensor<int32, [5]>([1, 175, 16, 32, 1])];
tensor<bool, [5]> var_634_end_mask_0 = const()[name = tensor<string, []>("op_634_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
tensor<bool, [5]> var_634_squeeze_mask_0 = const()[name = tensor<string, []>("op_634_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
tensor<fp32, [1, 175, 16, 32]> var_634 = slice_by_index(begin = var_634_begin_0, end = var_634_end_0, end_mask = var_634_end_mask_0, squeeze_mask = var_634_squeeze_mask_0, x = k_complex_3)[name = tensor<string, []>("op_634")];
tensor<int32, [5]> var_642_begin_0 = const()[name = tensor<string, []>("op_642_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 1])];
tensor<int32, [5]> var_642_end_0 = const()[name = tensor<string, []>("op_642_end_0"), val = tensor<int32, [5]>([1, 175, 16, 32, 2])];
tensor<bool, [5]> var_642_end_mask_0 = const()[name = tensor<string, []>("op_642_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
tensor<bool, [5]> var_642_squeeze_mask_0 = const()[name = tensor<string, []>("op_642_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
tensor<fp32, [1, 175, 16, 32]> var_642 = slice_by_index(begin = var_642_begin_0, end = var_642_end_0, end_mask = var_642_end_mask_0, squeeze_mask = var_642_squeeze_mask_0, x = k_complex_3)[name = tensor<string, []>("op_642")];
tensor<fp32, [1, 175, 1, 32]> var_648 = mul(x = freqs_3, y = ts_11)[name = tensor<string, []>("op_648")];
tensor<fp32, [1, 175, 1, 32]> rotr_3 = cos(x = var_648)[name = tensor<string, []>("rotr_3")];
tensor<fp32, [1, 175, 1, 32]> roti_3 = sin(x = var_648)[name = tensor<string, []>("roti_3")];
tensor<fp32, [1, 175, 16, 32]> var_652 = mul(x = var_618, y = rotr_3)[name = tensor<string, []>("op_652")];
tensor<fp32, [1, 175, 16, 32]> var_653 = mul(x = var_626, y = roti_3)[name = tensor<string, []>("op_653")];
tensor<fp32, [1, 175, 16, 32]> qor_5 = sub(x = var_652, y = var_653)[name = tensor<string, []>("qor_5")];
tensor<fp32, [1, 175, 16, 32]> var_656 = mul(x = var_618, y = roti_3)[name = tensor<string, []>("op_656")];
tensor<fp32, [1, 175, 16, 32]> var_657 = mul(x = var_626, y = rotr_3)[name = tensor<string, []>("op_657")];
tensor<fp32, [1, 175, 16, 32]> qoi_5 = add(x = var_656, y = var_657)[name = tensor<string, []>("qoi_5")];
tensor<fp32, [1, 175, 16, 32]> var_660 = mul(x = var_634, y = rotr_3)[name = tensor<string, []>("op_660")];
tensor<fp32, [1, 175, 16, 32]> var_661 = mul(x = var_642, y = roti_3)[name = tensor<string, []>("op_661")];
tensor<fp32, [1, 175, 16, 32]> kor_5 = sub(x = var_660, y = var_661)[name = tensor<string, []>("kor_5")];
tensor<fp32, [1, 175, 16, 32]> var_664 = mul(x = var_634, y = roti_3)[name = tensor<string, []>("op_664")];
tensor<fp32, [1, 175, 16, 32]> var_665 = mul(x = var_642, y = rotr_3)[name = tensor<string, []>("op_665")];
tensor<fp32, [1, 175, 16, 32]> koi_5 = add(x = var_664, y = var_665)[name = tensor<string, []>("koi_5")];
tensor<int32, []> qo_3_axis_0 = const()[name = tensor<string, []>("qo_3_axis_0"), val = tensor<int32, []>(-1)];
tensor<fp32, [1, 175, 16, 32, 2]> qo_3 = stack(axis = qo_3_axis_0, values = (qor_5, qoi_5))[name = tensor<string, []>("qo_3")];
tensor<int32, []> ko_3_axis_0 = const()[name = tensor<string, []>("ko_3_axis_0"), val = tensor<int32, []>(-1)];
tensor<fp32, [1, 175, 16, 32, 2]> ko_3 = stack(axis = ko_3_axis_0, values = (kor_5, koi_5))[name = tensor<string, []>("ko_3")];
tensor<int32, [4]> var_694 = const()[name = tensor<string, []>("op_694"), val = tensor<int32, [4]>([1, 175, 16, 64])];
tensor<fp32, [1, 175, 16, 64]> q_9 = reshape(shape = var_694, x = qo_3)[name = tensor<string, []>("q_9")];
tensor<int32, [4]> var_696 = const()[name = tensor<string, []>("op_696"), val = tensor<int32, [4]>([1, 175, 16, 64])];
tensor<fp32, [1, 175, 16, 64]> k_7 = reshape(shape = var_696, x = ko_3)[name = tensor<string, []>("k_7")];
tensor<int32, [4]> var_703 = const()[name = tensor<string, []>("op_703"), val = tensor<int32, [4]>([1, 1, 1, 1])];
tensor<fp32, [1, 1, 1, 1]> write_base_float_3 = reshape(shape = var_703, x = position1)[name = tensor<string, []>("write_base_float_3")];
tensor<fp32, [1, 175, 1, 1]> write_indices_float_5 = add(x = write_base_float_3, y = write_offsets_float_1_promoted)[name = tensor<string, []>("write_indices_float_5")];
tensor<fp32, []> _inversed_718_y_0 = const()[name = tensor<string, []>("_inversed_718_y_0"), val = tensor<fp32, []>(0x1p-9)];
tensor<fp32, [1, 175, 1, 1]> _inversed_718 = mul(x = write_indices_float_5, y = _inversed_718_y_0)[name = tensor<string, []>("_inversed_718")];
tensor<fp32, [1, 175, 1, 1]> var_719 = floor(x = _inversed_718)[name = tensor<string, []>("op_719")];
tensor<fp32, []> var_720 = const()[name = tensor<string, []>("op_720"), val = tensor<fp32, []>(0x1p+9)];
tensor<fp32, [1, 175, 1, 1]> var_721 = mul(x = var_719, y = var_720)[name = tensor<string, []>("op_721")];
tensor<fp32, [1, 175, 1, 1]> write_indices_float_7 = sub(x = write_indices_float_5, y = var_721)[name = tensor<string, []>("write_indices_float_7")];
tensor<string, []> var_728_dtype_0 = const()[name = tensor<string, []>("op_728_dtype_0"), val = tensor<string, []>("int32")];
tensor<int32, [4]> write_indices_3_reps_0 = const()[name = tensor<string, []>("write_indices_3_reps_0"), val = tensor<int32, [4]>([1, 1, 16, 64])];
tensor<int32, [1, 175, 1, 1]> var_728 = cast(dtype = var_728_dtype_0, x = write_indices_float_7)[name = tensor<string, []>("cast_103")];
tensor<int32, [1, 175, 16, 64]> write_indices_3 = tile(reps = write_indices_3_reps_0, x = var_728)[name = tensor<string, []>("write_indices_3")];
tensor<int32, [5]> var_736_begin_0 = const()[name = tensor<string, []>("op_736_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> var_736_end_0 = const()[name = tensor<string, []>("op_736_end_0"), val = tensor<int32, [5]>([1, 1, 512, 16, 64])];
tensor<bool, [5]> var_736_end_mask_0 = const()[name = tensor<string, []>("op_736_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> var_736_squeeze_mask_0 = const()[name = tensor<string, []>("op_736_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<fp32, [1, 512, 16, 64]> var_736 = slice_by_index(begin = var_736_begin_0, end = var_736_end_0, end_mask = var_736_end_mask_0, squeeze_mask = var_736_squeeze_mask_0, x = cache1)[name = tensor<string, []>("op_736")];
tensor<int32, []> var_738_axis_0 = const()[name = tensor<string, []>("op_738_axis_0"), val = tensor<int32, []>(1)];
tensor<string, []> var_738_mode_0 = const()[name = tensor<string, []>("op_738_mode_0"), val = tensor<string, []>("update")];
tensor<bool, []> var_738_validate_indices_0 = const()[name = tensor<string, []>("op_738_validate_indices_0"), val = tensor<bool, []>(false)];
tensor<fp32, [1, 512, 16, 64]> var_738 = scatter_along_axis(axis = var_738_axis_0, data = var_736, indices = write_indices_3, mode = var_738_mode_0, updates = k_7, validate_indices = var_738_validate_indices_0)[name = tensor<string, []>("op_738")];
tensor<int32, [5]> concat_8 = const()[name = tensor<string, []>("concat_8"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> concat_9 = const()[name = tensor<string, []>("concat_9"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> new_cache_3_internal_tensor_assign_1_stride_0 = const()[name = tensor<string, []>("new_cache_3_internal_tensor_assign_1_stride_0"), val = tensor<int32, [5]>([1, 1, 1, 1, 1])];
tensor<bool, [5]> new_cache_3_internal_tensor_assign_1_begin_mask_0 = const()[name = tensor<string, []>("new_cache_3_internal_tensor_assign_1_begin_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> new_cache_3_internal_tensor_assign_1_end_mask_0 = const()[name = tensor<string, []>("new_cache_3_internal_tensor_assign_1_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> new_cache_3_internal_tensor_assign_1_squeeze_mask_0 = const()[name = tensor<string, []>("new_cache_3_internal_tensor_assign_1_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<int32, [5]> shape_12 = const()[name = tensor<string, []>("shape_12"), val = tensor<int32, [5]>([2, 1, 512, 16, 64])];
tensor<int32, []> reduce_prod_2 = const()[name = tensor<string, []>("reduce_prod_2"), val = tensor<int32, []>(1048576)];
tensor<int32, []> range_1d_2_start_0 = const()[name = tensor<string, []>("range_1d_2_start_0"), val = tensor<int32, []>(0)];
tensor<int32, []> range_1d_2_step_0 = const()[name = tensor<string, []>("range_1d_2_step_0"), val = tensor<int32, []>(1)];
tensor<int32, [1048576]> range_1d_2 = range_1d(end = reduce_prod_2, start = range_1d_2_start_0, step = range_1d_2_step_0)[name = tensor<string, []>("range_1d_2")];
tensor<int32, [2, 1, 512, 16, 64]> reshape_10 = reshape(shape = shape_12, x = range_1d_2)[name = tensor<string, []>("reshape_10")];
tensor<int32, [1, 512, 16, 64]> slice_by_index_2 = slice_by_index(begin = concat_8, begin_mask = new_cache_3_internal_tensor_assign_1_begin_mask_0, end = concat_9, end_mask = new_cache_3_internal_tensor_assign_1_end_mask_0, squeeze_mask = new_cache_3_internal_tensor_assign_1_squeeze_mask_0, stride = new_cache_3_internal_tensor_assign_1_stride_0, x = reshape_10)[name = tensor<string, []>("slice_by_index_2")];
tensor<int32, [1]> reshape_11_shape_0 = const()[name = tensor<string, []>("reshape_11_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<int32, [524288]> reshape_11 = reshape(shape = reshape_11_shape_0, x = slice_by_index_2)[name = tensor<string, []>("reshape_11")];
tensor<int32, [1]> reshape_12_shape_0 = const()[name = tensor<string, []>("reshape_12_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [524288]> reshape_12 = reshape(shape = reshape_12_shape_0, x = var_738)[name = tensor<string, []>("reshape_12")];
tensor<int32, [1]> reshape_13_shape_0 = const()[name = tensor<string, []>("reshape_13_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [1048576]> reshape_13 = reshape(shape = reshape_13_shape_0, x = cache1)[name = tensor<string, []>("reshape_13")];
tensor<string, []> scatter_2_mode_0 = const()[name = tensor<string, []>("scatter_2_mode_0"), val = tensor<string, []>("update")];
tensor<int32, []> scatter_2_axis_0 = const()[name = tensor<string, []>("scatter_2_axis_0"), val = tensor<int32, []>(0)];
tensor<bool, []> scatter_2_validate_indices_0 = const()[name = tensor<string, []>("scatter_2_validate_indices_0"), val = tensor<bool, []>(false)];
tensor<fp32, [1048576]> scatter_2 = scatter(axis = scatter_2_axis_0, data = reshape_13, indices = reshape_11, mode = scatter_2_mode_0, updates = reshape_12, validate_indices = scatter_2_validate_indices_0)[name = tensor<string, []>("scatter_2")];
tensor<fp32, [2, 1, 512, 16, 64]> reshape_14 = reshape(shape = shape_12, x = scatter_2)[name = tensor<string, []>("reshape_14")];
tensor<int32, [5]> var_746_begin_0 = const()[name = tensor<string, []>("op_746_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
tensor<int32, [5]> var_746_end_0 = const()[name = tensor<string, []>("op_746_end_0"), val = tensor<int32, [5]>([2, 1, 512, 16, 64])];
tensor<bool, [5]> var_746_end_mask_0 = const()[name = tensor<string, []>("op_746_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> var_746_squeeze_mask_0 = const()[name = tensor<string, []>("op_746_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<fp32, [1, 512, 16, 64]> var_746 = slice_by_index(begin = var_746_begin_0, end = var_746_end_0, end_mask = var_746_end_mask_0, squeeze_mask = var_746_squeeze_mask_0, x = reshape_14)[name = tensor<string, []>("op_746")];
tensor<int32, []> var_748_axis_0 = const()[name = tensor<string, []>("op_748_axis_0"), val = tensor<int32, []>(1)];
tensor<string, []> var_748_mode_0 = const()[name = tensor<string, []>("op_748_mode_0"), val = tensor<string, []>("update")];
tensor<bool, []> var_748_validate_indices_0 = const()[name = tensor<string, []>("op_748_validate_indices_0"), val = tensor<bool, []>(false)];
tensor<fp32, [1, 512, 16, 64]> var_748 = scatter_along_axis(axis = var_748_axis_0, data = var_746, indices = write_indices_3, mode = var_748_mode_0, updates = v_3, validate_indices = var_748_validate_indices_0)[name = tensor<string, []>("op_748")];
tensor<int32, [5]> concat_10 = const()[name = tensor<string, []>("concat_10"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
tensor<int32, [5]> concat_11 = const()[name = tensor<string, []>("concat_11"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> new_cache_3_internal_tensor_assign_2_stride_0 = const()[name = tensor<string, []>("new_cache_3_internal_tensor_assign_2_stride_0"), val = tensor<int32, [5]>([1, 1, 1, 1, 1])];
tensor<bool, [5]> new_cache_3_internal_tensor_assign_2_begin_mask_0 = const()[name = tensor<string, []>("new_cache_3_internal_tensor_assign_2_begin_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> new_cache_3_internal_tensor_assign_2_end_mask_0 = const()[name = tensor<string, []>("new_cache_3_internal_tensor_assign_2_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> new_cache_3_internal_tensor_assign_2_squeeze_mask_0 = const()[name = tensor<string, []>("new_cache_3_internal_tensor_assign_2_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<int32, [5]> shape_13 = const()[name = tensor<string, []>("shape_13"), val = tensor<int32, [5]>([2, 1, 512, 16, 64])];
tensor<int32, []> reduce_prod_3 = const()[name = tensor<string, []>("reduce_prod_3"), val = tensor<int32, []>(1048576)];
tensor<int32, []> range_1d_3_start_0 = const()[name = tensor<string, []>("range_1d_3_start_0"), val = tensor<int32, []>(0)];
tensor<int32, []> range_1d_3_step_0 = const()[name = tensor<string, []>("range_1d_3_step_0"), val = tensor<int32, []>(1)];
tensor<int32, [1048576]> range_1d_3 = range_1d(end = reduce_prod_3, start = range_1d_3_start_0, step = range_1d_3_step_0)[name = tensor<string, []>("range_1d_3")];
tensor<int32, [2, 1, 512, 16, 64]> reshape_15 = reshape(shape = shape_13, x = range_1d_3)[name = tensor<string, []>("reshape_15")];
tensor<int32, [1, 512, 16, 64]> slice_by_index_3 = slice_by_index(begin = concat_10, begin_mask = new_cache_3_internal_tensor_assign_2_begin_mask_0, end = concat_11, end_mask = new_cache_3_internal_tensor_assign_2_end_mask_0, squeeze_mask = new_cache_3_internal_tensor_assign_2_squeeze_mask_0, stride = new_cache_3_internal_tensor_assign_2_stride_0, x = reshape_15)[name = tensor<string, []>("slice_by_index_3")];
tensor<int32, [1]> reshape_16_shape_0 = const()[name = tensor<string, []>("reshape_16_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<int32, [524288]> reshape_16 = reshape(shape = reshape_16_shape_0, x = slice_by_index_3)[name = tensor<string, []>("reshape_16")];
tensor<int32, [1]> reshape_17_shape_0 = const()[name = tensor<string, []>("reshape_17_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [524288]> reshape_17 = reshape(shape = reshape_17_shape_0, x = var_748)[name = tensor<string, []>("reshape_17")];
tensor<int32, [1]> reshape_18_shape_0 = const()[name = tensor<string, []>("reshape_18_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [1048576]> reshape_18 = reshape(shape = reshape_18_shape_0, x = reshape_14)[name = tensor<string, []>("reshape_18")];
tensor<string, []> scatter_3_mode_0 = const()[name = tensor<string, []>("scatter_3_mode_0"), val = tensor<string, []>("update")];
tensor<int32, []> scatter_3_axis_0 = const()[name = tensor<string, []>("scatter_3_axis_0"), val = tensor<int32, []>(0)];
tensor<bool, []> scatter_3_validate_indices_0 = const()[name = tensor<string, []>("scatter_3_validate_indices_0"), val = tensor<bool, []>(false)];
tensor<fp32, [1048576]> scatter_3 = scatter(axis = scatter_3_axis_0, data = reshape_18, indices = reshape_16, mode = scatter_3_mode_0, updates = reshape_17, validate_indices = scatter_3_validate_indices_0)[name = tensor<string, []>("scatter_3")];
tensor<fp32, [2, 1, 512, 16, 64]> new_cache_3_internal_tensor_assign_2 = reshape(shape = shape_13, x = scatter_3)[name = tensor<string, []>("reshape_19")];
tensor<int32, [5]> keys_7_begin_0 = const()[name = tensor<string, []>("keys_7_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> keys_7_end_0 = const()[name = tensor<string, []>("keys_7_end_0"), val = tensor<int32, [5]>([1, 1, 512, 16, 64])];
tensor<bool, [5]> keys_7_end_mask_0 = const()[name = tensor<string, []>("keys_7_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> keys_7_squeeze_mask_0 = const()[name = tensor<string, []>("keys_7_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<fp32, [1, 512, 16, 64]> keys_7 = slice_by_index(begin = keys_7_begin_0, end = keys_7_end_0, end_mask = keys_7_end_mask_0, squeeze_mask = keys_7_squeeze_mask_0, x = new_cache_3_internal_tensor_assign_2)[name = tensor<string, []>("keys_7")];
tensor<int32, [5]> values_7_begin_0 = const()[name = tensor<string, []>("values_7_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
tensor<int32, [5]> values_7_end_0 = const()[name = tensor<string, []>("values_7_end_0"), val = tensor<int32, [5]>([2, 1, 512, 16, 64])];
tensor<bool, [5]> values_7_end_mask_0 = const()[name = tensor<string, []>("values_7_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> values_7_squeeze_mask_0 = const()[name = tensor<string, []>("values_7_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<fp32, [1, 512, 16, 64]> values_7 = slice_by_index(begin = values_7_begin_0, end = values_7_end_0, end_mask = values_7_end_mask_0, squeeze_mask = values_7_squeeze_mask_0, x = new_cache_3_internal_tensor_assign_2)[name = tensor<string, []>("values_7")];
tensor<bool, [1, 512, 16, 64]> var_760 = not_equal(x = keys_7, y = keys_7)[name = tensor<string, []>("op_760")];
tensor<fp32, [1, 512, 16, 64]> keys_9 = select(a = var_347, b = keys_7, cond = var_760)[name = tensor<string, []>("keys_9")];
tensor<bool, [1, 512, 16, 64]> var_768 = not_equal(x = values_7, y = values_7)[name = tensor<string, []>("op_768")];
tensor<fp32, [1, 512, 16, 64]> values_9 = select(a = var_347, b = values_7, cond = var_768)[name = tensor<string, []>("values_9")];
tensor<int32, [4]> var_792 = const()[name = tensor<string, []>("op_792"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_805 = const()[name = tensor<string, []>("op_805"), val = tensor<int32, [3]>([1, 1, 1])];
tensor<fp32, [1, 1, 1]> var_806 = reshape(shape = var_805, x = position1)[name = tensor<string, []>("op_806")];
tensor<fp32, [1, 175, 1]> q_positions_3 = add(x = var_806, y = q_offsets_1_promoted)[name = tensor<string, []>("q_positions_3")];
tensor<fp32, []> var_823 = const()[name = tensor<string, []>("op_823"), val = tensor<fp32, []>(0x1.5ep+7)];
tensor<fp32, [1, 1, 1]> valid_len_3 = add(x = var_806, y = var_823)[name = tensor<string, []>("valid_len_3")];
tensor<bool, [1, 1, 512]> valid_mask_3 = less(x = k_positions_1_promoted, y = valid_len_3)[name = tensor<string, []>("valid_mask_3")];
tensor<bool, [1, 175, 512]> causal_mask_3 = less_equal(x = k_positions_1_promoted, y = q_positions_3)[name = tensor<string, []>("causal_mask_3")];
tensor<bool, [1, 175, 512]> attn_mask_5 = logical_and(x = valid_mask_3, y = causal_mask_3)[name = tensor<string, []>("attn_mask_5")];
tensor<int32, [1]> attn_mask_7_axes_0 = const()[name = tensor<string, []>("attn_mask_7_axes_0"), val = tensor<int32, [1]>([1])];
tensor<bool, [1, 1, 175, 512]> attn_mask_7 = expand_dims(axes = attn_mask_7_axes_0, x = attn_mask_5)[name = tensor<string, []>("attn_mask_7")];
tensor<fp32, [1]> var_835 = const()[name = tensor<string, []>("op_835"), val = tensor<fp32, [1]>([0x1.fffe5cp-4])];
tensor<bool, []> var_841_transpose_x_0 = const()[name = tensor<string, []>("op_841_transpose_x_0"), val = tensor<bool, []>(false)];
tensor<bool, []> var_841_transpose_y_0 = const()[name = tensor<string, []>("op_841_transpose_y_0"), val = tensor<bool, []>(false)];
tensor<int32, [4]> transpose_17_perm_0 = const()[name = tensor<string, []>("transpose_17_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
tensor<int32, [4]> transpose_18_perm_0 = const()[name = tensor<string, []>("transpose_18_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
tensor<fp32, [1, 16, 64, 512]> transpose_18 = transpose(perm = transpose_18_perm_0, x = keys_9)[name = tensor<string, []>("transpose_38")];
tensor<fp32, [1, 16, 175, 64]> transpose_17 = transpose(perm = transpose_17_perm_0, x = q_9)[name = tensor<string, []>("transpose_39")];
tensor<fp32, [1, 16, 175, 512]> var_841 = matmul(transpose_x = var_841_transpose_x_0, transpose_y = var_841_transpose_y_0, x = transpose_17, y = transpose_18)[name = tensor<string, []>("op_841")];
tensor<fp32, [1, 16, 175, 512]> attn_weights_7 = mul(x = var_841, y = var_835)[name = tensor<string, []>("attn_weights_7")];
tensor<bool, [1, 1, 175, 512]> var_843 = logical_not(x = attn_mask_7)[name = tensor<string, []>("op_843")];
tensor<fp32, []> var_844 = const()[name = tensor<string, []>("op_844"), val = tensor<fp32, []>(-0x1.ff933cp+127)];
tensor<fp32, [1, 16, 175, 512]> attn_weights_9 = select(a = var_844, b = attn_weights_7, cond = var_843)[name = tensor<string, []>("attn_weights_9")];
tensor<int32, []> var_846 = const()[name = tensor<string, []>("op_846"), val = tensor<int32, []>(-1)];
tensor<fp32, [1, 16, 175, 512]> attn_weights_11 = softmax(axis = var_846, x = attn_weights_9)[name = tensor<string, []>("attn_weights_11")];
tensor<bool, []> attn_output_3_transpose_x_0 = const()[name = tensor<string, []>("attn_output_3_transpose_x_0"), val = tensor<bool, []>(false)];
tensor<bool, []> attn_output_3_transpose_y_0 = const()[name = tensor<string, []>("attn_output_3_transpose_y_0"), val = tensor<bool, []>(false)];
tensor<fp32, [1, 16, 512, 64]> values_11 = transpose(perm = var_792, x = values_9)[name = tensor<string, []>("transpose_40")];
tensor<fp32, [1, 16, 175, 64]> attn_output_3 = matmul(transpose_x = attn_output_3_transpose_x_0, transpose_y = attn_output_3_transpose_y_0, x = attn_weights_11, y = values_11)[name = tensor<string, []>("attn_output_3")];
tensor<int32, [4]> var_854 = const()[name = tensor<string, []>("op_854"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_857 = const()[name = tensor<string, []>("op_857"), val = tensor<int32, [3]>([1, 175, 1024])];
tensor<fp32, [1, 175, 16, 64]> var_855 = transpose(perm = var_854, x = attn_output_3)[name = tensor<string, []>("transpose_37")];
tensor<fp32, [1, 175, 1024]> input_11 = reshape(shape = var_857, x = var_855)[name = tensor<string, []>("input_11")];
tensor<fp32, [1, 175, 1024]> attn_out_3 = linear(bias = linear_1_bias_0, weight = attn1_out_proj_weight, x = input_11)[name = tensor<string, []>("linear_5")];
tensor<fp32, []> var_863 = const()[name = tensor<string, []>("op_863"), val = tensor<fp32, []>(0x1.5ep+7)];
tensor<fp32, [1]> var_864 = add(x = position1, y = var_863)[name = tensor<string, []>("op_864")];
tensor<fp32, [1, 175, 1024]> input_13 = add(x = input_9, y = attn_out_3)[name = tensor<string, []>("input_13")];
tensor<fp32, []> var_868 = const()[name = tensor<string, []>("op_868"), val = tensor<fp32, []>(0x1.4f8b58p-17)];
tensor<int32, [1]> input_15_axes_0 = const()[name = tensor<string, []>("input_15_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [1, 175, 1024]> input_15 = layer_norm(axes = input_15_axes_0, beta = norm1_2_bias, epsilon = var_868, gamma = norm1_2_weight, x = input_13)[name = tensor<string, []>("input_15")];
tensor<fp32, [1, 175, 4096]> var_876 = linear(bias = linear_2_bias_0, weight = linear1_1_weight, x = input_15)[name = tensor<string, []>("linear_6")];
tensor<string, []> input_17_mode_0 = const()[name = tensor<string, []>("input_17_mode_0"), val = tensor<string, []>("EXACT")];
tensor<fp32, [1, 175, 4096]> input_17 = gelu(mode = input_17_mode_0, x = var_876)[name = tensor<string, []>("input_17")];
tensor<fp32, [1, 175, 1024]> ffn_out_3 = linear(bias = linear_1_bias_0, weight = linear1_2_weight, x = input_17)[name = tensor<string, []>("linear_7")];
tensor<fp32, [1, 175, 1024]> input_19 = add(x = input_13, y = ffn_out_3)[name = tensor<string, []>("input_19")];
tensor<fp32, []> var_885 = const()[name = tensor<string, []>("op_885"), val = tensor<fp32, []>(0x1.4f8b58p-17)];
tensor<int32, [1]> x_5_axes_0 = const()[name = tensor<string, []>("x_5_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [1, 175, 1024]> x_5 = layer_norm(axes = x_5_axes_0, beta = norm2_1_bias, epsilon = var_885, gamma = norm2_1_weight, x = input_19)[name = tensor<string, []>("x_5")];
tensor<fp32, [1, 175, 3072]> var_917 = linear(bias = linear_0_bias_0, weight = attn2_in_proj_weight, x = x_5)[name = tensor<string, []>("linear_8")];
tensor<int32, [5]> var_921 = const()[name = tensor<string, []>("op_921"), val = tensor<int32, [5]>([1, 175, 3, 16, 64])];
tensor<fp32, [1, 175, 3, 16, 64]> qkv_5 = reshape(shape = var_921, x = var_917)[name = tensor<string, []>("qkv_5")];
tensor<int32, [5]> q_13_begin_0 = const()[name = tensor<string, []>("q_13_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> q_13_end_0 = const()[name = tensor<string, []>("q_13_end_0"), val = tensor<int32, [5]>([1, 175, 1, 16, 64])];
tensor<bool, [5]> q_13_end_mask_0 = const()[name = tensor<string, []>("q_13_end_mask_0"), val = tensor<bool, [5]>([true, true, false, true, true])];
tensor<bool, [5]> q_13_squeeze_mask_0 = const()[name = tensor<string, []>("q_13_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, true, false, false])];
tensor<fp32, [1, 175, 16, 64]> q_13 = slice_by_index(begin = q_13_begin_0, end = q_13_end_0, end_mask = q_13_end_mask_0, squeeze_mask = q_13_squeeze_mask_0, x = qkv_5)[name = tensor<string, []>("q_13")];
tensor<int32, [5]> k_9_begin_0 = const()[name = tensor<string, []>("k_9_begin_0"), val = tensor<int32, [5]>([0, 0, 1, 0, 0])];
tensor<int32, [5]> k_9_end_0 = const()[name = tensor<string, []>("k_9_end_0"), val = tensor<int32, [5]>([1, 175, 2, 16, 64])];
tensor<bool, [5]> k_9_end_mask_0 = const()[name = tensor<string, []>("k_9_end_mask_0"), val = tensor<bool, [5]>([true, true, false, true, true])];
tensor<bool, [5]> k_9_squeeze_mask_0 = const()[name = tensor<string, []>("k_9_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, true, false, false])];
tensor<fp32, [1, 175, 16, 64]> k_9 = slice_by_index(begin = k_9_begin_0, end = k_9_end_0, end_mask = k_9_end_mask_0, squeeze_mask = k_9_squeeze_mask_0, x = qkv_5)[name = tensor<string, []>("k_9")];
tensor<int32, [5]> v_5_begin_0 = const()[name = tensor<string, []>("v_5_begin_0"), val = tensor<int32, [5]>([0, 0, 2, 0, 0])];
tensor<int32, [5]> v_5_end_0 = const()[name = tensor<string, []>("v_5_end_0"), val = tensor<int32, [5]>([1, 175, 3, 16, 64])];
tensor<bool, [5]> v_5_end_mask_0 = const()[name = tensor<string, []>("v_5_end_mask_0"), val = tensor<bool, [5]>([true, true, false, true, true])];
tensor<bool, [5]> v_5_squeeze_mask_0 = const()[name = tensor<string, []>("v_5_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, true, false, false])];
tensor<fp32, [1, 175, 16, 64]> v_5 = slice_by_index(begin = v_5_begin_0, end = v_5_end_0, end_mask = v_5_end_mask_0, squeeze_mask = v_5_squeeze_mask_0, x = qkv_5)[name = tensor<string, []>("v_5")];
tensor<fp32, [32]> freqs_5 = const()[name = tensor<string, []>("freqs_5"), val = tensor<fp32, [32]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(266469056)))];
tensor<int32, [2]> var_1019 = const()[name = tensor<string, []>("op_1019"), val = tensor<int32, [2]>([1, 1])];
tensor<fp32, [1, 1]> var_1020 = reshape(shape = var_1019, x = position2)[name = tensor<string, []>("op_1020")];
tensor<fp32, [1, 175]> ts_15 = add(x = ts_1_promoted, y = var_1020)[name = tensor<string, []>("ts_15")];
tensor<int32, [4]> var_1025 = const()[name = tensor<string, []>("op_1025"), val = tensor<int32, [4]>([1, 175, 1, 1])];
tensor<fp32, [1, 175, 1, 1]> ts_17 = reshape(shape = var_1025, x = ts_15)[name = tensor<string, []>("ts_17")];
tensor<int32, [5]> var_1029 = const()[name = tensor<string, []>("op_1029"), val = tensor<int32, [5]>([1, 175, 16, 32, 2])];
tensor<fp32, [1, 175, 16, 32, 2]> q_complex_5 = reshape(shape = var_1029, x = q_13)[name = tensor<string, []>("q_complex_5")];
tensor<int32, [5]> var_1033 = const()[name = tensor<string, []>("op_1033"), val = tensor<int32, [5]>([1, 175, 16, 32, 2])];
tensor<fp32, [1, 175, 16, 32, 2]> k_complex_5 = reshape(shape = var_1033, x = k_9)[name = tensor<string, []>("k_complex_5")];
tensor<int32, [5]> var_1037_begin_0 = const()[name = tensor<string, []>("op_1037_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> var_1037_end_0 = const()[name = tensor<string, []>("op_1037_end_0"), val = tensor<int32, [5]>([1, 175, 16, 32, 1])];
tensor<bool, [5]> var_1037_end_mask_0 = const()[name = tensor<string, []>("op_1037_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
tensor<bool, [5]> var_1037_squeeze_mask_0 = const()[name = tensor<string, []>("op_1037_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
tensor<fp32, [1, 175, 16, 32]> var_1037 = slice_by_index(begin = var_1037_begin_0, end = var_1037_end_0, end_mask = var_1037_end_mask_0, squeeze_mask = var_1037_squeeze_mask_0, x = q_complex_5)[name = tensor<string, []>("op_1037")];
tensor<int32, [5]> var_1045_begin_0 = const()[name = tensor<string, []>("op_1045_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 1])];
tensor<int32, [5]> var_1045_end_0 = const()[name = tensor<string, []>("op_1045_end_0"), val = tensor<int32, [5]>([1, 175, 16, 32, 2])];
tensor<bool, [5]> var_1045_end_mask_0 = const()[name = tensor<string, []>("op_1045_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
tensor<bool, [5]> var_1045_squeeze_mask_0 = const()[name = tensor<string, []>("op_1045_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
tensor<fp32, [1, 175, 16, 32]> var_1045 = slice_by_index(begin = var_1045_begin_0, end = var_1045_end_0, end_mask = var_1045_end_mask_0, squeeze_mask = var_1045_squeeze_mask_0, x = q_complex_5)[name = tensor<string, []>("op_1045")];
tensor<int32, [5]> var_1053_begin_0 = const()[name = tensor<string, []>("op_1053_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> var_1053_end_0 = const()[name = tensor<string, []>("op_1053_end_0"), val = tensor<int32, [5]>([1, 175, 16, 32, 1])];
tensor<bool, [5]> var_1053_end_mask_0 = const()[name = tensor<string, []>("op_1053_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
tensor<bool, [5]> var_1053_squeeze_mask_0 = const()[name = tensor<string, []>("op_1053_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
tensor<fp32, [1, 175, 16, 32]> var_1053 = slice_by_index(begin = var_1053_begin_0, end = var_1053_end_0, end_mask = var_1053_end_mask_0, squeeze_mask = var_1053_squeeze_mask_0, x = k_complex_5)[name = tensor<string, []>("op_1053")];
tensor<int32, [5]> var_1061_begin_0 = const()[name = tensor<string, []>("op_1061_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 1])];
tensor<int32, [5]> var_1061_end_0 = const()[name = tensor<string, []>("op_1061_end_0"), val = tensor<int32, [5]>([1, 175, 16, 32, 2])];
tensor<bool, [5]> var_1061_end_mask_0 = const()[name = tensor<string, []>("op_1061_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
tensor<bool, [5]> var_1061_squeeze_mask_0 = const()[name = tensor<string, []>("op_1061_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
tensor<fp32, [1, 175, 16, 32]> var_1061 = slice_by_index(begin = var_1061_begin_0, end = var_1061_end_0, end_mask = var_1061_end_mask_0, squeeze_mask = var_1061_squeeze_mask_0, x = k_complex_5)[name = tensor<string, []>("op_1061")];
tensor<fp32, [1, 175, 1, 32]> var_1067 = mul(x = freqs_5, y = ts_17)[name = tensor<string, []>("op_1067")];
tensor<fp32, [1, 175, 1, 32]> rotr_5 = cos(x = var_1067)[name = tensor<string, []>("rotr_5")];
tensor<fp32, [1, 175, 1, 32]> roti_5 = sin(x = var_1067)[name = tensor<string, []>("roti_5")];
tensor<fp32, [1, 175, 16, 32]> var_1071 = mul(x = var_1037, y = rotr_5)[name = tensor<string, []>("op_1071")];
tensor<fp32, [1, 175, 16, 32]> var_1072 = mul(x = var_1045, y = roti_5)[name = tensor<string, []>("op_1072")];
tensor<fp32, [1, 175, 16, 32]> qor_9 = sub(x = var_1071, y = var_1072)[name = tensor<string, []>("qor_9")];
tensor<fp32, [1, 175, 16, 32]> var_1075 = mul(x = var_1037, y = roti_5)[name = tensor<string, []>("op_1075")];
tensor<fp32, [1, 175, 16, 32]> var_1076 = mul(x = var_1045, y = rotr_5)[name = tensor<string, []>("op_1076")];
tensor<fp32, [1, 175, 16, 32]> qoi_9 = add(x = var_1075, y = var_1076)[name = tensor<string, []>("qoi_9")];
tensor<fp32, [1, 175, 16, 32]> var_1079 = mul(x = var_1053, y = rotr_5)[name = tensor<string, []>("op_1079")];
tensor<fp32, [1, 175, 16, 32]> var_1080 = mul(x = var_1061, y = roti_5)[name = tensor<string, []>("op_1080")];
tensor<fp32, [1, 175, 16, 32]> kor_9 = sub(x = var_1079, y = var_1080)[name = tensor<string, []>("kor_9")];
tensor<fp32, [1, 175, 16, 32]> var_1083 = mul(x = var_1053, y = roti_5)[name = tensor<string, []>("op_1083")];
tensor<fp32, [1, 175, 16, 32]> var_1084 = mul(x = var_1061, y = rotr_5)[name = tensor<string, []>("op_1084")];
tensor<fp32, [1, 175, 16, 32]> koi_9 = add(x = var_1083, y = var_1084)[name = tensor<string, []>("koi_9")];
tensor<int32, []> qo_5_axis_0 = const()[name = tensor<string, []>("qo_5_axis_0"), val = tensor<int32, []>(-1)];
tensor<fp32, [1, 175, 16, 32, 2]> qo_5 = stack(axis = qo_5_axis_0, values = (qor_9, qoi_9))[name = tensor<string, []>("qo_5")];
tensor<int32, []> ko_5_axis_0 = const()[name = tensor<string, []>("ko_5_axis_0"), val = tensor<int32, []>(-1)];
tensor<fp32, [1, 175, 16, 32, 2]> ko_5 = stack(axis = ko_5_axis_0, values = (kor_9, koi_9))[name = tensor<string, []>("ko_5")];
tensor<int32, [4]> var_1113 = const()[name = tensor<string, []>("op_1113"), val = tensor<int32, [4]>([1, 175, 16, 64])];
tensor<fp32, [1, 175, 16, 64]> q_15 = reshape(shape = var_1113, x = qo_5)[name = tensor<string, []>("q_15")];
tensor<int32, [4]> var_1115 = const()[name = tensor<string, []>("op_1115"), val = tensor<int32, [4]>([1, 175, 16, 64])];
tensor<fp32, [1, 175, 16, 64]> k_11 = reshape(shape = var_1115, x = ko_5)[name = tensor<string, []>("k_11")];
tensor<int32, [4]> var_1122 = const()[name = tensor<string, []>("op_1122"), val = tensor<int32, [4]>([1, 1, 1, 1])];
tensor<fp32, [1, 1, 1, 1]> write_base_float_5 = reshape(shape = var_1122, x = position2)[name = tensor<string, []>("write_base_float_5")];
tensor<fp32, [1, 175, 1, 1]> write_indices_float_9 = add(x = write_base_float_5, y = write_offsets_float_1_promoted)[name = tensor<string, []>("write_indices_float_9")];
tensor<fp32, []> _inversed_1137_y_0 = const()[name = tensor<string, []>("_inversed_1137_y_0"), val = tensor<fp32, []>(0x1p-9)];
tensor<fp32, [1, 175, 1, 1]> _inversed_1137 = mul(x = write_indices_float_9, y = _inversed_1137_y_0)[name = tensor<string, []>("_inversed_1137")];
tensor<fp32, [1, 175, 1, 1]> var_1138 = floor(x = _inversed_1137)[name = tensor<string, []>("op_1138")];
tensor<fp32, []> var_1139 = const()[name = tensor<string, []>("op_1139"), val = tensor<fp32, []>(0x1p+9)];
tensor<fp32, [1, 175, 1, 1]> var_1140 = mul(x = var_1138, y = var_1139)[name = tensor<string, []>("op_1140")];
tensor<fp32, [1, 175, 1, 1]> write_indices_float_11 = sub(x = write_indices_float_9, y = var_1140)[name = tensor<string, []>("write_indices_float_11")];
tensor<string, []> var_1147_dtype_0 = const()[name = tensor<string, []>("op_1147_dtype_0"), val = tensor<string, []>("int32")];
tensor<int32, [4]> write_indices_5_reps_0 = const()[name = tensor<string, []>("write_indices_5_reps_0"), val = tensor<int32, [4]>([1, 1, 16, 64])];
tensor<int32, [1, 175, 1, 1]> var_1147 = cast(dtype = var_1147_dtype_0, x = write_indices_float_11)[name = tensor<string, []>("cast_102")];
tensor<int32, [1, 175, 16, 64]> write_indices_5 = tile(reps = write_indices_5_reps_0, x = var_1147)[name = tensor<string, []>("write_indices_5")];
tensor<int32, [5]> var_1155_begin_0 = const()[name = tensor<string, []>("op_1155_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> var_1155_end_0 = const()[name = tensor<string, []>("op_1155_end_0"), val = tensor<int32, [5]>([1, 1, 512, 16, 64])];
tensor<bool, [5]> var_1155_end_mask_0 = const()[name = tensor<string, []>("op_1155_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> var_1155_squeeze_mask_0 = const()[name = tensor<string, []>("op_1155_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<fp32, [1, 512, 16, 64]> var_1155 = slice_by_index(begin = var_1155_begin_0, end = var_1155_end_0, end_mask = var_1155_end_mask_0, squeeze_mask = var_1155_squeeze_mask_0, x = cache2)[name = tensor<string, []>("op_1155")];
tensor<int32, []> var_1157_axis_0 = const()[name = tensor<string, []>("op_1157_axis_0"), val = tensor<int32, []>(1)];
tensor<string, []> var_1157_mode_0 = const()[name = tensor<string, []>("op_1157_mode_0"), val = tensor<string, []>("update")];
tensor<bool, []> var_1157_validate_indices_0 = const()[name = tensor<string, []>("op_1157_validate_indices_0"), val = tensor<bool, []>(false)];
tensor<fp32, [1, 512, 16, 64]> var_1157 = scatter_along_axis(axis = var_1157_axis_0, data = var_1155, indices = write_indices_5, mode = var_1157_mode_0, updates = k_11, validate_indices = var_1157_validate_indices_0)[name = tensor<string, []>("op_1157")];
tensor<int32, [5]> concat_15 = const()[name = tensor<string, []>("concat_15"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> concat_16 = const()[name = tensor<string, []>("concat_16"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> new_cache_5_internal_tensor_assign_1_stride_0 = const()[name = tensor<string, []>("new_cache_5_internal_tensor_assign_1_stride_0"), val = tensor<int32, [5]>([1, 1, 1, 1, 1])];
tensor<bool, [5]> new_cache_5_internal_tensor_assign_1_begin_mask_0 = const()[name = tensor<string, []>("new_cache_5_internal_tensor_assign_1_begin_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> new_cache_5_internal_tensor_assign_1_end_mask_0 = const()[name = tensor<string, []>("new_cache_5_internal_tensor_assign_1_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> new_cache_5_internal_tensor_assign_1_squeeze_mask_0 = const()[name = tensor<string, []>("new_cache_5_internal_tensor_assign_1_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<int32, [5]> shape_14 = const()[name = tensor<string, []>("shape_14"), val = tensor<int32, [5]>([2, 1, 512, 16, 64])];
tensor<int32, []> reduce_prod_4 = const()[name = tensor<string, []>("reduce_prod_4"), val = tensor<int32, []>(1048576)];
tensor<int32, []> range_1d_4_start_0 = const()[name = tensor<string, []>("range_1d_4_start_0"), val = tensor<int32, []>(0)];
tensor<int32, []> range_1d_4_step_0 = const()[name = tensor<string, []>("range_1d_4_step_0"), val = tensor<int32, []>(1)];
tensor<int32, [1048576]> range_1d_4 = range_1d(end = reduce_prod_4, start = range_1d_4_start_0, step = range_1d_4_step_0)[name = tensor<string, []>("range_1d_4")];
tensor<int32, [2, 1, 512, 16, 64]> reshape_20 = reshape(shape = shape_14, x = range_1d_4)[name = tensor<string, []>("reshape_20")];
tensor<int32, [1, 512, 16, 64]> slice_by_index_4 = slice_by_index(begin = concat_15, begin_mask = new_cache_5_internal_tensor_assign_1_begin_mask_0, end = concat_16, end_mask = new_cache_5_internal_tensor_assign_1_end_mask_0, squeeze_mask = new_cache_5_internal_tensor_assign_1_squeeze_mask_0, stride = new_cache_5_internal_tensor_assign_1_stride_0, x = reshape_20)[name = tensor<string, []>("slice_by_index_4")];
tensor<int32, [1]> reshape_21_shape_0 = const()[name = tensor<string, []>("reshape_21_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<int32, [524288]> reshape_21 = reshape(shape = reshape_21_shape_0, x = slice_by_index_4)[name = tensor<string, []>("reshape_21")];
tensor<int32, [1]> reshape_22_shape_0 = const()[name = tensor<string, []>("reshape_22_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [524288]> reshape_22 = reshape(shape = reshape_22_shape_0, x = var_1157)[name = tensor<string, []>("reshape_22")];
tensor<int32, [1]> reshape_23_shape_0 = const()[name = tensor<string, []>("reshape_23_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [1048576]> reshape_23 = reshape(shape = reshape_23_shape_0, x = cache2)[name = tensor<string, []>("reshape_23")];
tensor<string, []> scatter_4_mode_0 = const()[name = tensor<string, []>("scatter_4_mode_0"), val = tensor<string, []>("update")];
tensor<int32, []> scatter_4_axis_0 = const()[name = tensor<string, []>("scatter_4_axis_0"), val = tensor<int32, []>(0)];
tensor<bool, []> scatter_4_validate_indices_0 = const()[name = tensor<string, []>("scatter_4_validate_indices_0"), val = tensor<bool, []>(false)];
tensor<fp32, [1048576]> scatter_4 = scatter(axis = scatter_4_axis_0, data = reshape_23, indices = reshape_21, mode = scatter_4_mode_0, updates = reshape_22, validate_indices = scatter_4_validate_indices_0)[name = tensor<string, []>("scatter_4")];
tensor<fp32, [2, 1, 512, 16, 64]> reshape_24 = reshape(shape = shape_14, x = scatter_4)[name = tensor<string, []>("reshape_24")];
tensor<int32, [5]> var_1165_begin_0 = const()[name = tensor<string, []>("op_1165_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
tensor<int32, [5]> var_1165_end_0 = const()[name = tensor<string, []>("op_1165_end_0"), val = tensor<int32, [5]>([2, 1, 512, 16, 64])];
tensor<bool, [5]> var_1165_end_mask_0 = const()[name = tensor<string, []>("op_1165_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> var_1165_squeeze_mask_0 = const()[name = tensor<string, []>("op_1165_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<fp32, [1, 512, 16, 64]> var_1165 = slice_by_index(begin = var_1165_begin_0, end = var_1165_end_0, end_mask = var_1165_end_mask_0, squeeze_mask = var_1165_squeeze_mask_0, x = reshape_24)[name = tensor<string, []>("op_1165")];
tensor<int32, []> var_1167_axis_0 = const()[name = tensor<string, []>("op_1167_axis_0"), val = tensor<int32, []>(1)];
tensor<string, []> var_1167_mode_0 = const()[name = tensor<string, []>("op_1167_mode_0"), val = tensor<string, []>("update")];
tensor<bool, []> var_1167_validate_indices_0 = const()[name = tensor<string, []>("op_1167_validate_indices_0"), val = tensor<bool, []>(false)];
tensor<fp32, [1, 512, 16, 64]> var_1167 = scatter_along_axis(axis = var_1167_axis_0, data = var_1165, indices = write_indices_5, mode = var_1167_mode_0, updates = v_5, validate_indices = var_1167_validate_indices_0)[name = tensor<string, []>("op_1167")];
tensor<int32, [5]> concat_17 = const()[name = tensor<string, []>("concat_17"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
tensor<int32, [5]> concat_18 = const()[name = tensor<string, []>("concat_18"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> new_cache_5_internal_tensor_assign_2_stride_0 = const()[name = tensor<string, []>("new_cache_5_internal_tensor_assign_2_stride_0"), val = tensor<int32, [5]>([1, 1, 1, 1, 1])];
tensor<bool, [5]> new_cache_5_internal_tensor_assign_2_begin_mask_0 = const()[name = tensor<string, []>("new_cache_5_internal_tensor_assign_2_begin_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> new_cache_5_internal_tensor_assign_2_end_mask_0 = const()[name = tensor<string, []>("new_cache_5_internal_tensor_assign_2_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> new_cache_5_internal_tensor_assign_2_squeeze_mask_0 = const()[name = tensor<string, []>("new_cache_5_internal_tensor_assign_2_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<int32, [5]> shape_15 = const()[name = tensor<string, []>("shape_15"), val = tensor<int32, [5]>([2, 1, 512, 16, 64])];
tensor<int32, []> reduce_prod_5 = const()[name = tensor<string, []>("reduce_prod_5"), val = tensor<int32, []>(1048576)];
tensor<int32, []> range_1d_5_start_0 = const()[name = tensor<string, []>("range_1d_5_start_0"), val = tensor<int32, []>(0)];
tensor<int32, []> range_1d_5_step_0 = const()[name = tensor<string, []>("range_1d_5_step_0"), val = tensor<int32, []>(1)];
tensor<int32, [1048576]> range_1d_5 = range_1d(end = reduce_prod_5, start = range_1d_5_start_0, step = range_1d_5_step_0)[name = tensor<string, []>("range_1d_5")];
tensor<int32, [2, 1, 512, 16, 64]> reshape_25 = reshape(shape = shape_15, x = range_1d_5)[name = tensor<string, []>("reshape_25")];
tensor<int32, [1, 512, 16, 64]> slice_by_index_5 = slice_by_index(begin = concat_17, begin_mask = new_cache_5_internal_tensor_assign_2_begin_mask_0, end = concat_18, end_mask = new_cache_5_internal_tensor_assign_2_end_mask_0, squeeze_mask = new_cache_5_internal_tensor_assign_2_squeeze_mask_0, stride = new_cache_5_internal_tensor_assign_2_stride_0, x = reshape_25)[name = tensor<string, []>("slice_by_index_5")];
tensor<int32, [1]> reshape_26_shape_0 = const()[name = tensor<string, []>("reshape_26_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<int32, [524288]> reshape_26 = reshape(shape = reshape_26_shape_0, x = slice_by_index_5)[name = tensor<string, []>("reshape_26")];
tensor<int32, [1]> reshape_27_shape_0 = const()[name = tensor<string, []>("reshape_27_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [524288]> reshape_27 = reshape(shape = reshape_27_shape_0, x = var_1167)[name = tensor<string, []>("reshape_27")];
tensor<int32, [1]> reshape_28_shape_0 = const()[name = tensor<string, []>("reshape_28_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [1048576]> reshape_28 = reshape(shape = reshape_28_shape_0, x = reshape_24)[name = tensor<string, []>("reshape_28")];
tensor<string, []> scatter_5_mode_0 = const()[name = tensor<string, []>("scatter_5_mode_0"), val = tensor<string, []>("update")];
tensor<int32, []> scatter_5_axis_0 = const()[name = tensor<string, []>("scatter_5_axis_0"), val = tensor<int32, []>(0)];
tensor<bool, []> scatter_5_validate_indices_0 = const()[name = tensor<string, []>("scatter_5_validate_indices_0"), val = tensor<bool, []>(false)];
tensor<fp32, [1048576]> scatter_5 = scatter(axis = scatter_5_axis_0, data = reshape_28, indices = reshape_26, mode = scatter_5_mode_0, updates = reshape_27, validate_indices = scatter_5_validate_indices_0)[name = tensor<string, []>("scatter_5")];
tensor<fp32, [2, 1, 512, 16, 64]> new_cache_5_internal_tensor_assign_2 = reshape(shape = shape_15, x = scatter_5)[name = tensor<string, []>("reshape_29")];
tensor<int32, [5]> keys_13_begin_0 = const()[name = tensor<string, []>("keys_13_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> keys_13_end_0 = const()[name = tensor<string, []>("keys_13_end_0"), val = tensor<int32, [5]>([1, 1, 512, 16, 64])];
tensor<bool, [5]> keys_13_end_mask_0 = const()[name = tensor<string, []>("keys_13_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> keys_13_squeeze_mask_0 = const()[name = tensor<string, []>("keys_13_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<fp32, [1, 512, 16, 64]> keys_13 = slice_by_index(begin = keys_13_begin_0, end = keys_13_end_0, end_mask = keys_13_end_mask_0, squeeze_mask = keys_13_squeeze_mask_0, x = new_cache_5_internal_tensor_assign_2)[name = tensor<string, []>("keys_13")];
tensor<int32, [5]> values_13_begin_0 = const()[name = tensor<string, []>("values_13_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
tensor<int32, [5]> values_13_end_0 = const()[name = tensor<string, []>("values_13_end_0"), val = tensor<int32, [5]>([2, 1, 512, 16, 64])];
tensor<bool, [5]> values_13_end_mask_0 = const()[name = tensor<string, []>("values_13_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> values_13_squeeze_mask_0 = const()[name = tensor<string, []>("values_13_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<fp32, [1, 512, 16, 64]> values_13 = slice_by_index(begin = values_13_begin_0, end = values_13_end_0, end_mask = values_13_end_mask_0, squeeze_mask = values_13_squeeze_mask_0, x = new_cache_5_internal_tensor_assign_2)[name = tensor<string, []>("values_13")];
tensor<bool, [1, 512, 16, 64]> var_1179 = not_equal(x = keys_13, y = keys_13)[name = tensor<string, []>("op_1179")];
tensor<fp32, [1, 512, 16, 64]> keys_15 = select(a = var_347, b = keys_13, cond = var_1179)[name = tensor<string, []>("keys_15")];
tensor<bool, [1, 512, 16, 64]> var_1187 = not_equal(x = values_13, y = values_13)[name = tensor<string, []>("op_1187")];
tensor<fp32, [1, 512, 16, 64]> values_15 = select(a = var_347, b = values_13, cond = var_1187)[name = tensor<string, []>("values_15")];
tensor<int32, [4]> var_1211 = const()[name = tensor<string, []>("op_1211"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_1224 = const()[name = tensor<string, []>("op_1224"), val = tensor<int32, [3]>([1, 1, 1])];
tensor<fp32, [1, 1, 1]> var_1225 = reshape(shape = var_1224, x = position2)[name = tensor<string, []>("op_1225")];
tensor<fp32, [1, 175, 1]> q_positions_5 = add(x = var_1225, y = q_offsets_1_promoted)[name = tensor<string, []>("q_positions_5")];
tensor<fp32, []> var_1242 = const()[name = tensor<string, []>("op_1242"), val = tensor<fp32, []>(0x1.5ep+7)];
tensor<fp32, [1, 1, 1]> valid_len_5 = add(x = var_1225, y = var_1242)[name = tensor<string, []>("valid_len_5")];
tensor<bool, [1, 1, 512]> valid_mask_5 = less(x = k_positions_1_promoted, y = valid_len_5)[name = tensor<string, []>("valid_mask_5")];
tensor<bool, [1, 175, 512]> causal_mask_5 = less_equal(x = k_positions_1_promoted, y = q_positions_5)[name = tensor<string, []>("causal_mask_5")];
tensor<bool, [1, 175, 512]> attn_mask_9 = logical_and(x = valid_mask_5, y = causal_mask_5)[name = tensor<string, []>("attn_mask_9")];
tensor<int32, [1]> attn_mask_11_axes_0 = const()[name = tensor<string, []>("attn_mask_11_axes_0"), val = tensor<int32, [1]>([1])];
tensor<bool, [1, 1, 175, 512]> attn_mask_11 = expand_dims(axes = attn_mask_11_axes_0, x = attn_mask_9)[name = tensor<string, []>("attn_mask_11")];
tensor<fp32, [1]> var_1254 = const()[name = tensor<string, []>("op_1254"), val = tensor<fp32, [1]>([0x1.fffe5cp-4])];
tensor<bool, []> var_1260_transpose_x_0 = const()[name = tensor<string, []>("op_1260_transpose_x_0"), val = tensor<bool, []>(false)];
tensor<bool, []> var_1260_transpose_y_0 = const()[name = tensor<string, []>("op_1260_transpose_y_0"), val = tensor<bool, []>(false)];
tensor<int32, [4]> transpose_19_perm_0 = const()[name = tensor<string, []>("transpose_19_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
tensor<int32, [4]> transpose_20_perm_0 = const()[name = tensor<string, []>("transpose_20_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
tensor<fp32, [1, 16, 64, 512]> transpose_20 = transpose(perm = transpose_20_perm_0, x = keys_15)[name = tensor<string, []>("transpose_34")];
tensor<fp32, [1, 16, 175, 64]> transpose_19 = transpose(perm = transpose_19_perm_0, x = q_15)[name = tensor<string, []>("transpose_35")];
tensor<fp32, [1, 16, 175, 512]> var_1260 = matmul(transpose_x = var_1260_transpose_x_0, transpose_y = var_1260_transpose_y_0, x = transpose_19, y = transpose_20)[name = tensor<string, []>("op_1260")];
tensor<fp32, [1, 16, 175, 512]> attn_weights_13 = mul(x = var_1260, y = var_1254)[name = tensor<string, []>("attn_weights_13")];
tensor<bool, [1, 1, 175, 512]> var_1262 = logical_not(x = attn_mask_11)[name = tensor<string, []>("op_1262")];
tensor<fp32, []> var_1263 = const()[name = tensor<string, []>("op_1263"), val = tensor<fp32, []>(-0x1.ff933cp+127)];
tensor<fp32, [1, 16, 175, 512]> attn_weights_15 = select(a = var_1263, b = attn_weights_13, cond = var_1262)[name = tensor<string, []>("attn_weights_15")];
tensor<int32, []> var_1265 = const()[name = tensor<string, []>("op_1265"), val = tensor<int32, []>(-1)];
tensor<fp32, [1, 16, 175, 512]> attn_weights_17 = softmax(axis = var_1265, x = attn_weights_15)[name = tensor<string, []>("attn_weights_17")];
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<fp32, [1, 16, 512, 64]> values_17 = transpose(perm = var_1211, x = values_15)[name = tensor<string, []>("transpose_36")];
tensor<fp32, [1, 16, 175, 64]> attn_output_5 = matmul(transpose_x = attn_output_5_transpose_x_0, transpose_y = attn_output_5_transpose_y_0, x = attn_weights_17, y = values_17)[name = tensor<string, []>("attn_output_5")];
tensor<int32, [4]> var_1273 = const()[name = tensor<string, []>("op_1273"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_1276 = const()[name = tensor<string, []>("op_1276"), val = tensor<int32, [3]>([1, 175, 1024])];
tensor<fp32, [1, 175, 16, 64]> var_1274 = transpose(perm = var_1273, x = attn_output_5)[name = tensor<string, []>("transpose_33")];
tensor<fp32, [1, 175, 1024]> input_21 = reshape(shape = var_1276, x = var_1274)[name = tensor<string, []>("input_21")];
tensor<fp32, [1, 175, 1024]> attn_out_5 = linear(bias = linear_1_bias_0, weight = attn2_out_proj_weight, x = input_21)[name = tensor<string, []>("linear_9")];
tensor<fp32, []> var_1282 = const()[name = tensor<string, []>("op_1282"), val = tensor<fp32, []>(0x1.5ep+7)];
tensor<fp32, [1]> var_1283 = add(x = position2, y = var_1282)[name = tensor<string, []>("op_1283")];
tensor<fp32, [1, 175, 1024]> input_23 = add(x = input_19, y = attn_out_5)[name = tensor<string, []>("input_23")];
tensor<fp32, []> var_1287 = const()[name = tensor<string, []>("op_1287"), val = tensor<fp32, []>(0x1.4f8b58p-17)];
tensor<int32, [1]> input_25_axes_0 = const()[name = tensor<string, []>("input_25_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [1, 175, 1024]> input_25 = layer_norm(axes = input_25_axes_0, beta = norm2_2_bias, epsilon = var_1287, gamma = norm2_2_weight, x = input_23)[name = tensor<string, []>("input_25")];
tensor<fp32, [1, 175, 4096]> var_1295 = linear(bias = linear_2_bias_0, weight = linear2_1_weight, x = input_25)[name = tensor<string, []>("linear_10")];
tensor<string, []> input_27_mode_0 = const()[name = tensor<string, []>("input_27_mode_0"), val = tensor<string, []>("EXACT")];
tensor<fp32, [1, 175, 4096]> input_27 = gelu(mode = input_27_mode_0, x = var_1295)[name = tensor<string, []>("input_27")];
tensor<fp32, [1, 175, 1024]> ffn_out_5 = linear(bias = linear_1_bias_0, weight = linear2_2_weight, x = input_27)[name = tensor<string, []>("linear_11")];
tensor<fp32, [1, 175, 1024]> input_29 = add(x = input_23, y = ffn_out_5)[name = tensor<string, []>("input_29")];
tensor<fp32, []> var_1304 = const()[name = tensor<string, []>("op_1304"), val = tensor<fp32, []>(0x1.4f8b58p-17)];
tensor<int32, [1]> x_7_axes_0 = const()[name = tensor<string, []>("x_7_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [1, 175, 1024]> x_7 = layer_norm(axes = x_7_axes_0, beta = norm3_1_bias, epsilon = var_1304, gamma = norm3_1_weight, x = input_29)[name = tensor<string, []>("x_7")];
tensor<fp32, [1, 175, 3072]> var_1336 = linear(bias = linear_0_bias_0, weight = attn3_in_proj_weight, x = x_7)[name = tensor<string, []>("linear_12")];
tensor<int32, [5]> var_1340 = const()[name = tensor<string, []>("op_1340"), val = tensor<int32, [5]>([1, 175, 3, 16, 64])];
tensor<fp32, [1, 175, 3, 16, 64]> qkv_7 = reshape(shape = var_1340, x = var_1336)[name = tensor<string, []>("qkv_7")];
tensor<int32, [5]> q_19_begin_0 = const()[name = tensor<string, []>("q_19_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> q_19_end_0 = const()[name = tensor<string, []>("q_19_end_0"), val = tensor<int32, [5]>([1, 175, 1, 16, 64])];
tensor<bool, [5]> q_19_end_mask_0 = const()[name = tensor<string, []>("q_19_end_mask_0"), val = tensor<bool, [5]>([true, true, false, true, true])];
tensor<bool, [5]> q_19_squeeze_mask_0 = const()[name = tensor<string, []>("q_19_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, true, false, false])];
tensor<fp32, [1, 175, 16, 64]> q_19 = slice_by_index(begin = q_19_begin_0, end = q_19_end_0, end_mask = q_19_end_mask_0, squeeze_mask = q_19_squeeze_mask_0, x = qkv_7)[name = tensor<string, []>("q_19")];
tensor<int32, [5]> k_13_begin_0 = const()[name = tensor<string, []>("k_13_begin_0"), val = tensor<int32, [5]>([0, 0, 1, 0, 0])];
tensor<int32, [5]> k_13_end_0 = const()[name = tensor<string, []>("k_13_end_0"), val = tensor<int32, [5]>([1, 175, 2, 16, 64])];
tensor<bool, [5]> k_13_end_mask_0 = const()[name = tensor<string, []>("k_13_end_mask_0"), val = tensor<bool, [5]>([true, true, false, true, true])];
tensor<bool, [5]> k_13_squeeze_mask_0 = const()[name = tensor<string, []>("k_13_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, true, false, false])];
tensor<fp32, [1, 175, 16, 64]> k_13 = slice_by_index(begin = k_13_begin_0, end = k_13_end_0, end_mask = k_13_end_mask_0, squeeze_mask = k_13_squeeze_mask_0, x = qkv_7)[name = tensor<string, []>("k_13")];
tensor<int32, [5]> v_7_begin_0 = const()[name = tensor<string, []>("v_7_begin_0"), val = tensor<int32, [5]>([0, 0, 2, 0, 0])];
tensor<int32, [5]> v_7_end_0 = const()[name = tensor<string, []>("v_7_end_0"), val = tensor<int32, [5]>([1, 175, 3, 16, 64])];
tensor<bool, [5]> v_7_end_mask_0 = const()[name = tensor<string, []>("v_7_end_mask_0"), val = tensor<bool, [5]>([true, true, false, true, true])];
tensor<bool, [5]> v_7_squeeze_mask_0 = const()[name = tensor<string, []>("v_7_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, true, false, false])];
tensor<fp32, [1, 175, 16, 64]> v_7 = slice_by_index(begin = v_7_begin_0, end = v_7_end_0, end_mask = v_7_end_mask_0, squeeze_mask = v_7_squeeze_mask_0, x = qkv_7)[name = tensor<string, []>("v_7")];
tensor<fp32, [32]> freqs_7 = const()[name = tensor<string, []>("freqs_7"), val = tensor<fp32, [32]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(266469248)))];
tensor<int32, [2]> var_1438 = const()[name = tensor<string, []>("op_1438"), val = tensor<int32, [2]>([1, 1])];
tensor<fp32, [1, 1]> var_1439 = reshape(shape = var_1438, x = position3)[name = tensor<string, []>("op_1439")];
tensor<fp32, [1, 175]> ts_21 = add(x = ts_1_promoted, y = var_1439)[name = tensor<string, []>("ts_21")];
tensor<int32, [4]> var_1444 = const()[name = tensor<string, []>("op_1444"), val = tensor<int32, [4]>([1, 175, 1, 1])];
tensor<fp32, [1, 175, 1, 1]> ts_23 = reshape(shape = var_1444, x = ts_21)[name = tensor<string, []>("ts_23")];
tensor<int32, [5]> var_1448 = const()[name = tensor<string, []>("op_1448"), val = tensor<int32, [5]>([1, 175, 16, 32, 2])];
tensor<fp32, [1, 175, 16, 32, 2]> q_complex_7 = reshape(shape = var_1448, x = q_19)[name = tensor<string, []>("q_complex_7")];
tensor<int32, [5]> var_1452 = const()[name = tensor<string, []>("op_1452"), val = tensor<int32, [5]>([1, 175, 16, 32, 2])];
tensor<fp32, [1, 175, 16, 32, 2]> k_complex_7 = reshape(shape = var_1452, x = k_13)[name = tensor<string, []>("k_complex_7")];
tensor<int32, [5]> var_1456_begin_0 = const()[name = tensor<string, []>("op_1456_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> var_1456_end_0 = const()[name = tensor<string, []>("op_1456_end_0"), val = tensor<int32, [5]>([1, 175, 16, 32, 1])];
tensor<bool, [5]> var_1456_end_mask_0 = const()[name = tensor<string, []>("op_1456_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
tensor<bool, [5]> var_1456_squeeze_mask_0 = const()[name = tensor<string, []>("op_1456_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
tensor<fp32, [1, 175, 16, 32]> var_1456 = slice_by_index(begin = var_1456_begin_0, end = var_1456_end_0, end_mask = var_1456_end_mask_0, squeeze_mask = var_1456_squeeze_mask_0, x = q_complex_7)[name = tensor<string, []>("op_1456")];
tensor<int32, [5]> var_1464_begin_0 = const()[name = tensor<string, []>("op_1464_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 1])];
tensor<int32, [5]> var_1464_end_0 = const()[name = tensor<string, []>("op_1464_end_0"), val = tensor<int32, [5]>([1, 175, 16, 32, 2])];
tensor<bool, [5]> var_1464_end_mask_0 = const()[name = tensor<string, []>("op_1464_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
tensor<bool, [5]> var_1464_squeeze_mask_0 = const()[name = tensor<string, []>("op_1464_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
tensor<fp32, [1, 175, 16, 32]> var_1464 = slice_by_index(begin = var_1464_begin_0, end = var_1464_end_0, end_mask = var_1464_end_mask_0, squeeze_mask = var_1464_squeeze_mask_0, x = q_complex_7)[name = tensor<string, []>("op_1464")];
tensor<int32, [5]> var_1472_begin_0 = const()[name = tensor<string, []>("op_1472_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> var_1472_end_0 = const()[name = tensor<string, []>("op_1472_end_0"), val = tensor<int32, [5]>([1, 175, 16, 32, 1])];
tensor<bool, [5]> var_1472_end_mask_0 = const()[name = tensor<string, []>("op_1472_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
tensor<bool, [5]> var_1472_squeeze_mask_0 = const()[name = tensor<string, []>("op_1472_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
tensor<fp32, [1, 175, 16, 32]> var_1472 = slice_by_index(begin = var_1472_begin_0, end = var_1472_end_0, end_mask = var_1472_end_mask_0, squeeze_mask = var_1472_squeeze_mask_0, x = k_complex_7)[name = tensor<string, []>("op_1472")];
tensor<int32, [5]> var_1480_begin_0 = const()[name = tensor<string, []>("op_1480_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 1])];
tensor<int32, [5]> var_1480_end_0 = const()[name = tensor<string, []>("op_1480_end_0"), val = tensor<int32, [5]>([1, 175, 16, 32, 2])];
tensor<bool, [5]> var_1480_end_mask_0 = const()[name = tensor<string, []>("op_1480_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
tensor<bool, [5]> var_1480_squeeze_mask_0 = const()[name = tensor<string, []>("op_1480_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
tensor<fp32, [1, 175, 16, 32]> var_1480 = slice_by_index(begin = var_1480_begin_0, end = var_1480_end_0, end_mask = var_1480_end_mask_0, squeeze_mask = var_1480_squeeze_mask_0, x = k_complex_7)[name = tensor<string, []>("op_1480")];
tensor<fp32, [1, 175, 1, 32]> var_1486 = mul(x = freqs_7, y = ts_23)[name = tensor<string, []>("op_1486")];
tensor<fp32, [1, 175, 1, 32]> rotr_7 = cos(x = var_1486)[name = tensor<string, []>("rotr_7")];
tensor<fp32, [1, 175, 1, 32]> roti_7 = sin(x = var_1486)[name = tensor<string, []>("roti_7")];
tensor<fp32, [1, 175, 16, 32]> var_1490 = mul(x = var_1456, y = rotr_7)[name = tensor<string, []>("op_1490")];
tensor<fp32, [1, 175, 16, 32]> var_1491 = mul(x = var_1464, y = roti_7)[name = tensor<string, []>("op_1491")];
tensor<fp32, [1, 175, 16, 32]> qor_13 = sub(x = var_1490, y = var_1491)[name = tensor<string, []>("qor_13")];
tensor<fp32, [1, 175, 16, 32]> var_1494 = mul(x = var_1456, y = roti_7)[name = tensor<string, []>("op_1494")];
tensor<fp32, [1, 175, 16, 32]> var_1495 = mul(x = var_1464, y = rotr_7)[name = tensor<string, []>("op_1495")];
tensor<fp32, [1, 175, 16, 32]> qoi_13 = add(x = var_1494, y = var_1495)[name = tensor<string, []>("qoi_13")];
tensor<fp32, [1, 175, 16, 32]> var_1498 = mul(x = var_1472, y = rotr_7)[name = tensor<string, []>("op_1498")];
tensor<fp32, [1, 175, 16, 32]> var_1499 = mul(x = var_1480, y = roti_7)[name = tensor<string, []>("op_1499")];
tensor<fp32, [1, 175, 16, 32]> kor_13 = sub(x = var_1498, y = var_1499)[name = tensor<string, []>("kor_13")];
tensor<fp32, [1, 175, 16, 32]> var_1502 = mul(x = var_1472, y = roti_7)[name = tensor<string, []>("op_1502")];
tensor<fp32, [1, 175, 16, 32]> var_1503 = mul(x = var_1480, y = rotr_7)[name = tensor<string, []>("op_1503")];
tensor<fp32, [1, 175, 16, 32]> koi_13 = add(x = var_1502, y = var_1503)[name = tensor<string, []>("koi_13")];
tensor<int32, []> qo_7_axis_0 = const()[name = tensor<string, []>("qo_7_axis_0"), val = tensor<int32, []>(-1)];
tensor<fp32, [1, 175, 16, 32, 2]> qo_7 = stack(axis = qo_7_axis_0, values = (qor_13, qoi_13))[name = tensor<string, []>("qo_7")];
tensor<int32, []> ko_7_axis_0 = const()[name = tensor<string, []>("ko_7_axis_0"), val = tensor<int32, []>(-1)];
tensor<fp32, [1, 175, 16, 32, 2]> ko_7 = stack(axis = ko_7_axis_0, values = (kor_13, koi_13))[name = tensor<string, []>("ko_7")];
tensor<int32, [4]> var_1532 = const()[name = tensor<string, []>("op_1532"), val = tensor<int32, [4]>([1, 175, 16, 64])];
tensor<fp32, [1, 175, 16, 64]> q_21 = reshape(shape = var_1532, x = qo_7)[name = tensor<string, []>("q_21")];
tensor<int32, [4]> var_1534 = const()[name = tensor<string, []>("op_1534"), val = tensor<int32, [4]>([1, 175, 16, 64])];
tensor<fp32, [1, 175, 16, 64]> k_15 = reshape(shape = var_1534, x = ko_7)[name = tensor<string, []>("k_15")];
tensor<int32, [4]> var_1541 = const()[name = tensor<string, []>("op_1541"), val = tensor<int32, [4]>([1, 1, 1, 1])];
tensor<fp32, [1, 1, 1, 1]> write_base_float_7 = reshape(shape = var_1541, x = position3)[name = tensor<string, []>("write_base_float_7")];
tensor<fp32, [1, 175, 1, 1]> write_indices_float_13 = add(x = write_base_float_7, y = write_offsets_float_1_promoted)[name = tensor<string, []>("write_indices_float_13")];
tensor<fp32, []> _inversed_1556_y_0 = const()[name = tensor<string, []>("_inversed_1556_y_0"), val = tensor<fp32, []>(0x1p-9)];
tensor<fp32, [1, 175, 1, 1]> _inversed_1556 = mul(x = write_indices_float_13, y = _inversed_1556_y_0)[name = tensor<string, []>("_inversed_1556")];
tensor<fp32, [1, 175, 1, 1]> var_1557 = floor(x = _inversed_1556)[name = tensor<string, []>("op_1557")];
tensor<fp32, []> var_1558 = const()[name = tensor<string, []>("op_1558"), val = tensor<fp32, []>(0x1p+9)];
tensor<fp32, [1, 175, 1, 1]> var_1559 = mul(x = var_1557, y = var_1558)[name = tensor<string, []>("op_1559")];
tensor<fp32, [1, 175, 1, 1]> write_indices_float_15 = sub(x = write_indices_float_13, y = var_1559)[name = tensor<string, []>("write_indices_float_15")];
tensor<string, []> var_1566_dtype_0 = const()[name = tensor<string, []>("op_1566_dtype_0"), val = tensor<string, []>("int32")];
tensor<int32, [4]> write_indices_7_reps_0 = const()[name = tensor<string, []>("write_indices_7_reps_0"), val = tensor<int32, [4]>([1, 1, 16, 64])];
tensor<int32, [1, 175, 1, 1]> var_1566 = cast(dtype = var_1566_dtype_0, x = write_indices_float_15)[name = tensor<string, []>("cast_101")];
tensor<int32, [1, 175, 16, 64]> write_indices_7 = tile(reps = write_indices_7_reps_0, x = var_1566)[name = tensor<string, []>("write_indices_7")];
tensor<int32, [5]> var_1574_begin_0 = const()[name = tensor<string, []>("op_1574_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> var_1574_end_0 = const()[name = tensor<string, []>("op_1574_end_0"), val = tensor<int32, [5]>([1, 1, 512, 16, 64])];
tensor<bool, [5]> var_1574_end_mask_0 = const()[name = tensor<string, []>("op_1574_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> var_1574_squeeze_mask_0 = const()[name = tensor<string, []>("op_1574_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<fp32, [1, 512, 16, 64]> var_1574 = slice_by_index(begin = var_1574_begin_0, end = var_1574_end_0, end_mask = var_1574_end_mask_0, squeeze_mask = var_1574_squeeze_mask_0, x = cache3)[name = tensor<string, []>("op_1574")];
tensor<int32, []> var_1576_axis_0 = const()[name = tensor<string, []>("op_1576_axis_0"), val = tensor<int32, []>(1)];
tensor<string, []> var_1576_mode_0 = const()[name = tensor<string, []>("op_1576_mode_0"), val = tensor<string, []>("update")];
tensor<bool, []> var_1576_validate_indices_0 = const()[name = tensor<string, []>("op_1576_validate_indices_0"), val = tensor<bool, []>(false)];
tensor<fp32, [1, 512, 16, 64]> var_1576 = scatter_along_axis(axis = var_1576_axis_0, data = var_1574, indices = write_indices_7, mode = var_1576_mode_0, updates = k_15, validate_indices = var_1576_validate_indices_0)[name = tensor<string, []>("op_1576")];
tensor<int32, [5]> concat_22 = const()[name = tensor<string, []>("concat_22"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> concat_23 = const()[name = tensor<string, []>("concat_23"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> new_cache_7_internal_tensor_assign_1_stride_0 = const()[name = tensor<string, []>("new_cache_7_internal_tensor_assign_1_stride_0"), val = tensor<int32, [5]>([1, 1, 1, 1, 1])];
tensor<bool, [5]> new_cache_7_internal_tensor_assign_1_begin_mask_0 = const()[name = tensor<string, []>("new_cache_7_internal_tensor_assign_1_begin_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> new_cache_7_internal_tensor_assign_1_end_mask_0 = const()[name = tensor<string, []>("new_cache_7_internal_tensor_assign_1_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> new_cache_7_internal_tensor_assign_1_squeeze_mask_0 = const()[name = tensor<string, []>("new_cache_7_internal_tensor_assign_1_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<int32, [5]> shape_16 = const()[name = tensor<string, []>("shape_16"), val = tensor<int32, [5]>([2, 1, 512, 16, 64])];
tensor<int32, []> reduce_prod_6 = const()[name = tensor<string, []>("reduce_prod_6"), val = tensor<int32, []>(1048576)];
tensor<int32, []> range_1d_6_start_0 = const()[name = tensor<string, []>("range_1d_6_start_0"), val = tensor<int32, []>(0)];
tensor<int32, []> range_1d_6_step_0 = const()[name = tensor<string, []>("range_1d_6_step_0"), val = tensor<int32, []>(1)];
tensor<int32, [1048576]> range_1d_6 = range_1d(end = reduce_prod_6, start = range_1d_6_start_0, step = range_1d_6_step_0)[name = tensor<string, []>("range_1d_6")];
tensor<int32, [2, 1, 512, 16, 64]> reshape_30 = reshape(shape = shape_16, x = range_1d_6)[name = tensor<string, []>("reshape_30")];
tensor<int32, [1, 512, 16, 64]> slice_by_index_6 = slice_by_index(begin = concat_22, begin_mask = new_cache_7_internal_tensor_assign_1_begin_mask_0, end = concat_23, end_mask = new_cache_7_internal_tensor_assign_1_end_mask_0, squeeze_mask = new_cache_7_internal_tensor_assign_1_squeeze_mask_0, stride = new_cache_7_internal_tensor_assign_1_stride_0, x = reshape_30)[name = tensor<string, []>("slice_by_index_6")];
tensor<int32, [1]> reshape_31_shape_0 = const()[name = tensor<string, []>("reshape_31_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<int32, [524288]> reshape_31 = reshape(shape = reshape_31_shape_0, x = slice_by_index_6)[name = tensor<string, []>("reshape_31")];
tensor<int32, [1]> reshape_32_shape_0 = const()[name = tensor<string, []>("reshape_32_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [524288]> reshape_32 = reshape(shape = reshape_32_shape_0, x = var_1576)[name = tensor<string, []>("reshape_32")];
tensor<int32, [1]> reshape_33_shape_0 = const()[name = tensor<string, []>("reshape_33_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [1048576]> reshape_33 = reshape(shape = reshape_33_shape_0, x = cache3)[name = tensor<string, []>("reshape_33")];
tensor<string, []> scatter_6_mode_0 = const()[name = tensor<string, []>("scatter_6_mode_0"), val = tensor<string, []>("update")];
tensor<int32, []> scatter_6_axis_0 = const()[name = tensor<string, []>("scatter_6_axis_0"), val = tensor<int32, []>(0)];
tensor<bool, []> scatter_6_validate_indices_0 = const()[name = tensor<string, []>("scatter_6_validate_indices_0"), val = tensor<bool, []>(false)];
tensor<fp32, [1048576]> scatter_6 = scatter(axis = scatter_6_axis_0, data = reshape_33, indices = reshape_31, mode = scatter_6_mode_0, updates = reshape_32, validate_indices = scatter_6_validate_indices_0)[name = tensor<string, []>("scatter_6")];
tensor<fp32, [2, 1, 512, 16, 64]> reshape_34 = reshape(shape = shape_16, x = scatter_6)[name = tensor<string, []>("reshape_34")];
tensor<int32, [5]> var_1584_begin_0 = const()[name = tensor<string, []>("op_1584_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
tensor<int32, [5]> var_1584_end_0 = const()[name = tensor<string, []>("op_1584_end_0"), val = tensor<int32, [5]>([2, 1, 512, 16, 64])];
tensor<bool, [5]> var_1584_end_mask_0 = const()[name = tensor<string, []>("op_1584_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> var_1584_squeeze_mask_0 = const()[name = tensor<string, []>("op_1584_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<fp32, [1, 512, 16, 64]> var_1584 = slice_by_index(begin = var_1584_begin_0, end = var_1584_end_0, end_mask = var_1584_end_mask_0, squeeze_mask = var_1584_squeeze_mask_0, x = reshape_34)[name = tensor<string, []>("op_1584")];
tensor<int32, []> var_1586_axis_0 = const()[name = tensor<string, []>("op_1586_axis_0"), val = tensor<int32, []>(1)];
tensor<string, []> var_1586_mode_0 = const()[name = tensor<string, []>("op_1586_mode_0"), val = tensor<string, []>("update")];
tensor<bool, []> var_1586_validate_indices_0 = const()[name = tensor<string, []>("op_1586_validate_indices_0"), val = tensor<bool, []>(false)];
tensor<fp32, [1, 512, 16, 64]> var_1586 = scatter_along_axis(axis = var_1586_axis_0, data = var_1584, indices = write_indices_7, mode = var_1586_mode_0, updates = v_7, validate_indices = var_1586_validate_indices_0)[name = tensor<string, []>("op_1586")];
tensor<int32, [5]> concat_24 = const()[name = tensor<string, []>("concat_24"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
tensor<int32, [5]> concat_25 = const()[name = tensor<string, []>("concat_25"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> new_cache_7_internal_tensor_assign_2_stride_0 = const()[name = tensor<string, []>("new_cache_7_internal_tensor_assign_2_stride_0"), val = tensor<int32, [5]>([1, 1, 1, 1, 1])];
tensor<bool, [5]> new_cache_7_internal_tensor_assign_2_begin_mask_0 = const()[name = tensor<string, []>("new_cache_7_internal_tensor_assign_2_begin_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> new_cache_7_internal_tensor_assign_2_end_mask_0 = const()[name = tensor<string, []>("new_cache_7_internal_tensor_assign_2_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> new_cache_7_internal_tensor_assign_2_squeeze_mask_0 = const()[name = tensor<string, []>("new_cache_7_internal_tensor_assign_2_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<int32, [5]> shape_17 = const()[name = tensor<string, []>("shape_17"), val = tensor<int32, [5]>([2, 1, 512, 16, 64])];
tensor<int32, []> reduce_prod_7 = const()[name = tensor<string, []>("reduce_prod_7"), val = tensor<int32, []>(1048576)];
tensor<int32, []> range_1d_7_start_0 = const()[name = tensor<string, []>("range_1d_7_start_0"), val = tensor<int32, []>(0)];
tensor<int32, []> range_1d_7_step_0 = const()[name = tensor<string, []>("range_1d_7_step_0"), val = tensor<int32, []>(1)];
tensor<int32, [1048576]> range_1d_7 = range_1d(end = reduce_prod_7, start = range_1d_7_start_0, step = range_1d_7_step_0)[name = tensor<string, []>("range_1d_7")];
tensor<int32, [2, 1, 512, 16, 64]> reshape_35 = reshape(shape = shape_17, x = range_1d_7)[name = tensor<string, []>("reshape_35")];
tensor<int32, [1, 512, 16, 64]> slice_by_index_7 = slice_by_index(begin = concat_24, begin_mask = new_cache_7_internal_tensor_assign_2_begin_mask_0, end = concat_25, end_mask = new_cache_7_internal_tensor_assign_2_end_mask_0, squeeze_mask = new_cache_7_internal_tensor_assign_2_squeeze_mask_0, stride = new_cache_7_internal_tensor_assign_2_stride_0, x = reshape_35)[name = tensor<string, []>("slice_by_index_7")];
tensor<int32, [1]> reshape_36_shape_0 = const()[name = tensor<string, []>("reshape_36_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<int32, [524288]> reshape_36 = reshape(shape = reshape_36_shape_0, x = slice_by_index_7)[name = tensor<string, []>("reshape_36")];
tensor<int32, [1]> reshape_37_shape_0 = const()[name = tensor<string, []>("reshape_37_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [524288]> reshape_37 = reshape(shape = reshape_37_shape_0, x = var_1586)[name = tensor<string, []>("reshape_37")];
tensor<int32, [1]> reshape_38_shape_0 = const()[name = tensor<string, []>("reshape_38_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [1048576]> reshape_38 = reshape(shape = reshape_38_shape_0, x = reshape_34)[name = tensor<string, []>("reshape_38")];
tensor<string, []> scatter_7_mode_0 = const()[name = tensor<string, []>("scatter_7_mode_0"), val = tensor<string, []>("update")];
tensor<int32, []> scatter_7_axis_0 = const()[name = tensor<string, []>("scatter_7_axis_0"), val = tensor<int32, []>(0)];
tensor<bool, []> scatter_7_validate_indices_0 = const()[name = tensor<string, []>("scatter_7_validate_indices_0"), val = tensor<bool, []>(false)];
tensor<fp32, [1048576]> scatter_7 = scatter(axis = scatter_7_axis_0, data = reshape_38, indices = reshape_36, mode = scatter_7_mode_0, updates = reshape_37, validate_indices = scatter_7_validate_indices_0)[name = tensor<string, []>("scatter_7")];
tensor<fp32, [2, 1, 512, 16, 64]> new_cache_7_internal_tensor_assign_2 = reshape(shape = shape_17, x = scatter_7)[name = tensor<string, []>("reshape_39")];
tensor<int32, [5]> keys_19_begin_0 = const()[name = tensor<string, []>("keys_19_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> keys_19_end_0 = const()[name = tensor<string, []>("keys_19_end_0"), val = tensor<int32, [5]>([1, 1, 512, 16, 64])];
tensor<bool, [5]> keys_19_end_mask_0 = const()[name = tensor<string, []>("keys_19_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> keys_19_squeeze_mask_0 = const()[name = tensor<string, []>("keys_19_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<fp32, [1, 512, 16, 64]> keys_19 = slice_by_index(begin = keys_19_begin_0, end = keys_19_end_0, end_mask = keys_19_end_mask_0, squeeze_mask = keys_19_squeeze_mask_0, x = new_cache_7_internal_tensor_assign_2)[name = tensor<string, []>("keys_19")];
tensor<int32, [5]> values_19_begin_0 = const()[name = tensor<string, []>("values_19_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
tensor<int32, [5]> values_19_end_0 = const()[name = tensor<string, []>("values_19_end_0"), val = tensor<int32, [5]>([2, 1, 512, 16, 64])];
tensor<bool, [5]> values_19_end_mask_0 = const()[name = tensor<string, []>("values_19_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> values_19_squeeze_mask_0 = const()[name = tensor<string, []>("values_19_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<fp32, [1, 512, 16, 64]> values_19 = slice_by_index(begin = values_19_begin_0, end = values_19_end_0, end_mask = values_19_end_mask_0, squeeze_mask = values_19_squeeze_mask_0, x = new_cache_7_internal_tensor_assign_2)[name = tensor<string, []>("values_19")];
tensor<bool, [1, 512, 16, 64]> var_1598 = not_equal(x = keys_19, y = keys_19)[name = tensor<string, []>("op_1598")];
tensor<fp32, [1, 512, 16, 64]> keys_21 = select(a = var_347, b = keys_19, cond = var_1598)[name = tensor<string, []>("keys_21")];
tensor<bool, [1, 512, 16, 64]> var_1606 = not_equal(x = values_19, y = values_19)[name = tensor<string, []>("op_1606")];
tensor<fp32, [1, 512, 16, 64]> values_21 = select(a = var_347, b = values_19, cond = var_1606)[name = tensor<string, []>("values_21")];
tensor<int32, [4]> var_1630 = const()[name = tensor<string, []>("op_1630"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_1643 = const()[name = tensor<string, []>("op_1643"), val = tensor<int32, [3]>([1, 1, 1])];
tensor<fp32, [1, 1, 1]> var_1644 = reshape(shape = var_1643, x = position3)[name = tensor<string, []>("op_1644")];
tensor<fp32, [1, 175, 1]> q_positions_7 = add(x = var_1644, y = q_offsets_1_promoted)[name = tensor<string, []>("q_positions_7")];
tensor<fp32, []> var_1661 = const()[name = tensor<string, []>("op_1661"), val = tensor<fp32, []>(0x1.5ep+7)];
tensor<fp32, [1, 1, 1]> valid_len_7 = add(x = var_1644, y = var_1661)[name = tensor<string, []>("valid_len_7")];
tensor<bool, [1, 1, 512]> valid_mask_7 = less(x = k_positions_1_promoted, y = valid_len_7)[name = tensor<string, []>("valid_mask_7")];
tensor<bool, [1, 175, 512]> causal_mask_7 = less_equal(x = k_positions_1_promoted, y = q_positions_7)[name = tensor<string, []>("causal_mask_7")];
tensor<bool, [1, 175, 512]> attn_mask_13 = logical_and(x = valid_mask_7, y = causal_mask_7)[name = tensor<string, []>("attn_mask_13")];
tensor<int32, [1]> attn_mask_15_axes_0 = const()[name = tensor<string, []>("attn_mask_15_axes_0"), val = tensor<int32, [1]>([1])];
tensor<bool, [1, 1, 175, 512]> attn_mask_15 = expand_dims(axes = attn_mask_15_axes_0, x = attn_mask_13)[name = tensor<string, []>("attn_mask_15")];
tensor<fp32, [1]> var_1673 = const()[name = tensor<string, []>("op_1673"), val = tensor<fp32, [1]>([0x1.fffe5cp-4])];
tensor<bool, []> var_1679_transpose_x_0 = const()[name = tensor<string, []>("op_1679_transpose_x_0"), val = tensor<bool, []>(false)];
tensor<bool, []> var_1679_transpose_y_0 = const()[name = tensor<string, []>("op_1679_transpose_y_0"), val = tensor<bool, []>(false)];
tensor<int32, [4]> transpose_21_perm_0 = const()[name = tensor<string, []>("transpose_21_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
tensor<int32, [4]> transpose_22_perm_0 = const()[name = tensor<string, []>("transpose_22_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
tensor<fp32, [1, 16, 64, 512]> transpose_22 = transpose(perm = transpose_22_perm_0, x = keys_21)[name = tensor<string, []>("transpose_30")];
tensor<fp32, [1, 16, 175, 64]> transpose_21 = transpose(perm = transpose_21_perm_0, x = q_21)[name = tensor<string, []>("transpose_31")];
tensor<fp32, [1, 16, 175, 512]> var_1679 = matmul(transpose_x = var_1679_transpose_x_0, transpose_y = var_1679_transpose_y_0, x = transpose_21, y = transpose_22)[name = tensor<string, []>("op_1679")];
tensor<fp32, [1, 16, 175, 512]> attn_weights_19 = mul(x = var_1679, y = var_1673)[name = tensor<string, []>("attn_weights_19")];
tensor<bool, [1, 1, 175, 512]> var_1681 = logical_not(x = attn_mask_15)[name = tensor<string, []>("op_1681")];
tensor<fp32, []> var_1682 = const()[name = tensor<string, []>("op_1682"), val = tensor<fp32, []>(-0x1.ff933cp+127)];
tensor<fp32, [1, 16, 175, 512]> attn_weights_21 = select(a = var_1682, b = attn_weights_19, cond = var_1681)[name = tensor<string, []>("attn_weights_21")];
tensor<int32, []> var_1684 = const()[name = tensor<string, []>("op_1684"), val = tensor<int32, []>(-1)];
tensor<fp32, [1, 16, 175, 512]> attn_weights_23 = softmax(axis = var_1684, x = attn_weights_21)[name = tensor<string, []>("attn_weights_23")];
tensor<bool, []> attn_output_7_transpose_x_0 = const()[name = tensor<string, []>("attn_output_7_transpose_x_0"), val = tensor<bool, []>(false)];
tensor<bool, []> attn_output_7_transpose_y_0 = const()[name = tensor<string, []>("attn_output_7_transpose_y_0"), val = tensor<bool, []>(false)];
tensor<fp32, [1, 16, 512, 64]> values_23 = transpose(perm = var_1630, x = values_21)[name = tensor<string, []>("transpose_32")];
tensor<fp32, [1, 16, 175, 64]> attn_output_7 = matmul(transpose_x = attn_output_7_transpose_x_0, transpose_y = attn_output_7_transpose_y_0, x = attn_weights_23, y = values_23)[name = tensor<string, []>("attn_output_7")];
tensor<int32, [4]> var_1692 = const()[name = tensor<string, []>("op_1692"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_1695 = const()[name = tensor<string, []>("op_1695"), val = tensor<int32, [3]>([1, 175, 1024])];
tensor<fp32, [1, 175, 16, 64]> var_1693 = transpose(perm = var_1692, x = attn_output_7)[name = tensor<string, []>("transpose_29")];
tensor<fp32, [1, 175, 1024]> input_31 = reshape(shape = var_1695, x = var_1693)[name = tensor<string, []>("input_31")];
tensor<fp32, [1, 175, 1024]> attn_out_7 = linear(bias = linear_1_bias_0, weight = attn3_out_proj_weight, x = input_31)[name = tensor<string, []>("linear_13")];
tensor<fp32, []> var_1701 = const()[name = tensor<string, []>("op_1701"), val = tensor<fp32, []>(0x1.5ep+7)];
tensor<fp32, [1]> var_1702 = add(x = position3, y = var_1701)[name = tensor<string, []>("op_1702")];
tensor<fp32, [1, 175, 1024]> input_33 = add(x = input_29, y = attn_out_7)[name = tensor<string, []>("input_33")];
tensor<fp32, []> var_1706 = const()[name = tensor<string, []>("op_1706"), val = tensor<fp32, []>(0x1.4f8b58p-17)];
tensor<int32, [1]> input_35_axes_0 = const()[name = tensor<string, []>("input_35_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [1, 175, 1024]> input_35 = layer_norm(axes = input_35_axes_0, beta = norm3_2_bias, epsilon = var_1706, gamma = norm3_2_weight, x = input_33)[name = tensor<string, []>("input_35")];
tensor<fp32, [1, 175, 4096]> var_1714 = linear(bias = linear_2_bias_0, weight = linear3_1_weight, x = input_35)[name = tensor<string, []>("linear_14")];
tensor<string, []> input_37_mode_0 = const()[name = tensor<string, []>("input_37_mode_0"), val = tensor<string, []>("EXACT")];
tensor<fp32, [1, 175, 4096]> input_37 = gelu(mode = input_37_mode_0, x = var_1714)[name = tensor<string, []>("input_37")];
tensor<fp32, [1, 175, 1024]> ffn_out_7 = linear(bias = linear_1_bias_0, weight = linear3_2_weight, x = input_37)[name = tensor<string, []>("linear_15")];
tensor<fp32, [1, 175, 1024]> input_39 = add(x = input_33, y = ffn_out_7)[name = tensor<string, []>("input_39")];
tensor<fp32, []> var_1723 = const()[name = tensor<string, []>("op_1723"), val = tensor<fp32, []>(0x1.4f8b58p-17)];
tensor<int32, [1]> x_9_axes_0 = const()[name = tensor<string, []>("x_9_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [1, 175, 1024]> x_9 = layer_norm(axes = x_9_axes_0, beta = norm4_1_bias, epsilon = var_1723, gamma = norm4_1_weight, x = input_39)[name = tensor<string, []>("x_9")];
tensor<fp32, [1, 175, 3072]> var_1755 = linear(bias = linear_0_bias_0, weight = attn4_in_proj_weight, x = x_9)[name = tensor<string, []>("linear_16")];
tensor<int32, [5]> var_1759 = const()[name = tensor<string, []>("op_1759"), val = tensor<int32, [5]>([1, 175, 3, 16, 64])];
tensor<fp32, [1, 175, 3, 16, 64]> qkv_9 = reshape(shape = var_1759, x = var_1755)[name = tensor<string, []>("qkv_9")];
tensor<int32, [5]> q_25_begin_0 = const()[name = tensor<string, []>("q_25_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> q_25_end_0 = const()[name = tensor<string, []>("q_25_end_0"), val = tensor<int32, [5]>([1, 175, 1, 16, 64])];
tensor<bool, [5]> q_25_end_mask_0 = const()[name = tensor<string, []>("q_25_end_mask_0"), val = tensor<bool, [5]>([true, true, false, true, true])];
tensor<bool, [5]> q_25_squeeze_mask_0 = const()[name = tensor<string, []>("q_25_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, true, false, false])];
tensor<fp32, [1, 175, 16, 64]> q_25 = slice_by_index(begin = q_25_begin_0, end = q_25_end_0, end_mask = q_25_end_mask_0, squeeze_mask = q_25_squeeze_mask_0, x = qkv_9)[name = tensor<string, []>("q_25")];
tensor<int32, [5]> k_17_begin_0 = const()[name = tensor<string, []>("k_17_begin_0"), val = tensor<int32, [5]>([0, 0, 1, 0, 0])];
tensor<int32, [5]> k_17_end_0 = const()[name = tensor<string, []>("k_17_end_0"), val = tensor<int32, [5]>([1, 175, 2, 16, 64])];
tensor<bool, [5]> k_17_end_mask_0 = const()[name = tensor<string, []>("k_17_end_mask_0"), val = tensor<bool, [5]>([true, true, false, true, true])];
tensor<bool, [5]> k_17_squeeze_mask_0 = const()[name = tensor<string, []>("k_17_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, true, false, false])];
tensor<fp32, [1, 175, 16, 64]> k_17 = slice_by_index(begin = k_17_begin_0, end = k_17_end_0, end_mask = k_17_end_mask_0, squeeze_mask = k_17_squeeze_mask_0, x = qkv_9)[name = tensor<string, []>("k_17")];
tensor<int32, [5]> v_9_begin_0 = const()[name = tensor<string, []>("v_9_begin_0"), val = tensor<int32, [5]>([0, 0, 2, 0, 0])];
tensor<int32, [5]> v_9_end_0 = const()[name = tensor<string, []>("v_9_end_0"), val = tensor<int32, [5]>([1, 175, 3, 16, 64])];
tensor<bool, [5]> v_9_end_mask_0 = const()[name = tensor<string, []>("v_9_end_mask_0"), val = tensor<bool, [5]>([true, true, false, true, true])];
tensor<bool, [5]> v_9_squeeze_mask_0 = const()[name = tensor<string, []>("v_9_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, true, false, false])];
tensor<fp32, [1, 175, 16, 64]> v_9 = slice_by_index(begin = v_9_begin_0, end = v_9_end_0, end_mask = v_9_end_mask_0, squeeze_mask = v_9_squeeze_mask_0, x = qkv_9)[name = tensor<string, []>("v_9")];
tensor<fp32, [32]> freqs_9 = const()[name = tensor<string, []>("freqs_9"), val = tensor<fp32, [32]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(266469440)))];
tensor<int32, [2]> var_1857 = const()[name = tensor<string, []>("op_1857"), val = tensor<int32, [2]>([1, 1])];
tensor<fp32, [1, 1]> var_1858 = reshape(shape = var_1857, x = position4)[name = tensor<string, []>("op_1858")];
tensor<fp32, [1, 175]> ts_27 = add(x = ts_1_promoted, y = var_1858)[name = tensor<string, []>("ts_27")];
tensor<int32, [4]> var_1863 = const()[name = tensor<string, []>("op_1863"), val = tensor<int32, [4]>([1, 175, 1, 1])];
tensor<fp32, [1, 175, 1, 1]> ts_29 = reshape(shape = var_1863, x = ts_27)[name = tensor<string, []>("ts_29")];
tensor<int32, [5]> var_1867 = const()[name = tensor<string, []>("op_1867"), val = tensor<int32, [5]>([1, 175, 16, 32, 2])];
tensor<fp32, [1, 175, 16, 32, 2]> q_complex_9 = reshape(shape = var_1867, x = q_25)[name = tensor<string, []>("q_complex_9")];
tensor<int32, [5]> var_1871 = const()[name = tensor<string, []>("op_1871"), val = tensor<int32, [5]>([1, 175, 16, 32, 2])];
tensor<fp32, [1, 175, 16, 32, 2]> k_complex_9 = reshape(shape = var_1871, x = k_17)[name = tensor<string, []>("k_complex_9")];
tensor<int32, [5]> var_1875_begin_0 = const()[name = tensor<string, []>("op_1875_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> var_1875_end_0 = const()[name = tensor<string, []>("op_1875_end_0"), val = tensor<int32, [5]>([1, 175, 16, 32, 1])];
tensor<bool, [5]> var_1875_end_mask_0 = const()[name = tensor<string, []>("op_1875_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
tensor<bool, [5]> var_1875_squeeze_mask_0 = const()[name = tensor<string, []>("op_1875_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
tensor<fp32, [1, 175, 16, 32]> var_1875 = slice_by_index(begin = var_1875_begin_0, end = var_1875_end_0, end_mask = var_1875_end_mask_0, squeeze_mask = var_1875_squeeze_mask_0, x = q_complex_9)[name = tensor<string, []>("op_1875")];
tensor<int32, [5]> var_1883_begin_0 = const()[name = tensor<string, []>("op_1883_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 1])];
tensor<int32, [5]> var_1883_end_0 = const()[name = tensor<string, []>("op_1883_end_0"), val = tensor<int32, [5]>([1, 175, 16, 32, 2])];
tensor<bool, [5]> var_1883_end_mask_0 = const()[name = tensor<string, []>("op_1883_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
tensor<bool, [5]> var_1883_squeeze_mask_0 = const()[name = tensor<string, []>("op_1883_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
tensor<fp32, [1, 175, 16, 32]> var_1883 = slice_by_index(begin = var_1883_begin_0, end = var_1883_end_0, end_mask = var_1883_end_mask_0, squeeze_mask = var_1883_squeeze_mask_0, x = q_complex_9)[name = tensor<string, []>("op_1883")];
tensor<int32, [5]> var_1891_begin_0 = const()[name = tensor<string, []>("op_1891_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> var_1891_end_0 = const()[name = tensor<string, []>("op_1891_end_0"), val = tensor<int32, [5]>([1, 175, 16, 32, 1])];
tensor<bool, [5]> var_1891_end_mask_0 = const()[name = tensor<string, []>("op_1891_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
tensor<bool, [5]> var_1891_squeeze_mask_0 = const()[name = tensor<string, []>("op_1891_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
tensor<fp32, [1, 175, 16, 32]> var_1891 = slice_by_index(begin = var_1891_begin_0, end = var_1891_end_0, end_mask = var_1891_end_mask_0, squeeze_mask = var_1891_squeeze_mask_0, x = k_complex_9)[name = tensor<string, []>("op_1891")];
tensor<int32, [5]> var_1899_begin_0 = const()[name = tensor<string, []>("op_1899_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 1])];
tensor<int32, [5]> var_1899_end_0 = const()[name = tensor<string, []>("op_1899_end_0"), val = tensor<int32, [5]>([1, 175, 16, 32, 2])];
tensor<bool, [5]> var_1899_end_mask_0 = const()[name = tensor<string, []>("op_1899_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
tensor<bool, [5]> var_1899_squeeze_mask_0 = const()[name = tensor<string, []>("op_1899_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
tensor<fp32, [1, 175, 16, 32]> var_1899 = slice_by_index(begin = var_1899_begin_0, end = var_1899_end_0, end_mask = var_1899_end_mask_0, squeeze_mask = var_1899_squeeze_mask_0, x = k_complex_9)[name = tensor<string, []>("op_1899")];
tensor<fp32, [1, 175, 1, 32]> var_1905 = mul(x = freqs_9, y = ts_29)[name = tensor<string, []>("op_1905")];
tensor<fp32, [1, 175, 1, 32]> rotr_9 = cos(x = var_1905)[name = tensor<string, []>("rotr_9")];
tensor<fp32, [1, 175, 1, 32]> roti_9 = sin(x = var_1905)[name = tensor<string, []>("roti_9")];
tensor<fp32, [1, 175, 16, 32]> var_1909 = mul(x = var_1875, y = rotr_9)[name = tensor<string, []>("op_1909")];
tensor<fp32, [1, 175, 16, 32]> var_1910 = mul(x = var_1883, y = roti_9)[name = tensor<string, []>("op_1910")];
tensor<fp32, [1, 175, 16, 32]> qor_17 = sub(x = var_1909, y = var_1910)[name = tensor<string, []>("qor_17")];
tensor<fp32, [1, 175, 16, 32]> var_1913 = mul(x = var_1875, y = roti_9)[name = tensor<string, []>("op_1913")];
tensor<fp32, [1, 175, 16, 32]> var_1914 = mul(x = var_1883, y = rotr_9)[name = tensor<string, []>("op_1914")];
tensor<fp32, [1, 175, 16, 32]> qoi_17 = add(x = var_1913, y = var_1914)[name = tensor<string, []>("qoi_17")];
tensor<fp32, [1, 175, 16, 32]> var_1917 = mul(x = var_1891, y = rotr_9)[name = tensor<string, []>("op_1917")];
tensor<fp32, [1, 175, 16, 32]> var_1918 = mul(x = var_1899, y = roti_9)[name = tensor<string, []>("op_1918")];
tensor<fp32, [1, 175, 16, 32]> kor_17 = sub(x = var_1917, y = var_1918)[name = tensor<string, []>("kor_17")];
tensor<fp32, [1, 175, 16, 32]> var_1921 = mul(x = var_1891, y = roti_9)[name = tensor<string, []>("op_1921")];
tensor<fp32, [1, 175, 16, 32]> var_1922 = mul(x = var_1899, y = rotr_9)[name = tensor<string, []>("op_1922")];
tensor<fp32, [1, 175, 16, 32]> koi_17 = add(x = var_1921, y = var_1922)[name = tensor<string, []>("koi_17")];
tensor<int32, []> qo_9_axis_0 = const()[name = tensor<string, []>("qo_9_axis_0"), val = tensor<int32, []>(-1)];
tensor<fp32, [1, 175, 16, 32, 2]> qo_9 = stack(axis = qo_9_axis_0, values = (qor_17, qoi_17))[name = tensor<string, []>("qo_9")];
tensor<int32, []> ko_9_axis_0 = const()[name = tensor<string, []>("ko_9_axis_0"), val = tensor<int32, []>(-1)];
tensor<fp32, [1, 175, 16, 32, 2]> ko_9 = stack(axis = ko_9_axis_0, values = (kor_17, koi_17))[name = tensor<string, []>("ko_9")];
tensor<int32, [4]> var_1951 = const()[name = tensor<string, []>("op_1951"), val = tensor<int32, [4]>([1, 175, 16, 64])];
tensor<fp32, [1, 175, 16, 64]> q_27 = reshape(shape = var_1951, x = qo_9)[name = tensor<string, []>("q_27")];
tensor<int32, [4]> var_1953 = const()[name = tensor<string, []>("op_1953"), val = tensor<int32, [4]>([1, 175, 16, 64])];
tensor<fp32, [1, 175, 16, 64]> k_19 = reshape(shape = var_1953, x = ko_9)[name = tensor<string, []>("k_19")];
tensor<int32, [4]> var_1960 = const()[name = tensor<string, []>("op_1960"), val = tensor<int32, [4]>([1, 1, 1, 1])];
tensor<fp32, [1, 1, 1, 1]> write_base_float_9 = reshape(shape = var_1960, x = position4)[name = tensor<string, []>("write_base_float_9")];
tensor<fp32, [1, 175, 1, 1]> write_indices_float_17 = add(x = write_base_float_9, y = write_offsets_float_1_promoted)[name = tensor<string, []>("write_indices_float_17")];
tensor<fp32, []> _inversed_1975_y_0 = const()[name = tensor<string, []>("_inversed_1975_y_0"), val = tensor<fp32, []>(0x1p-9)];
tensor<fp32, [1, 175, 1, 1]> _inversed_1975 = mul(x = write_indices_float_17, y = _inversed_1975_y_0)[name = tensor<string, []>("_inversed_1975")];
tensor<fp32, [1, 175, 1, 1]> var_1976 = floor(x = _inversed_1975)[name = tensor<string, []>("op_1976")];
tensor<fp32, []> var_1977 = const()[name = tensor<string, []>("op_1977"), val = tensor<fp32, []>(0x1p+9)];
tensor<fp32, [1, 175, 1, 1]> var_1978 = mul(x = var_1976, y = var_1977)[name = tensor<string, []>("op_1978")];
tensor<fp32, [1, 175, 1, 1]> write_indices_float_19 = sub(x = write_indices_float_17, y = var_1978)[name = tensor<string, []>("write_indices_float_19")];
tensor<string, []> var_1985_dtype_0 = const()[name = tensor<string, []>("op_1985_dtype_0"), val = tensor<string, []>("int32")];
tensor<int32, [4]> write_indices_9_reps_0 = const()[name = tensor<string, []>("write_indices_9_reps_0"), val = tensor<int32, [4]>([1, 1, 16, 64])];
tensor<int32, [1, 175, 1, 1]> var_1985 = cast(dtype = var_1985_dtype_0, x = write_indices_float_19)[name = tensor<string, []>("cast_100")];
tensor<int32, [1, 175, 16, 64]> write_indices_9 = tile(reps = write_indices_9_reps_0, x = var_1985)[name = tensor<string, []>("write_indices_9")];
tensor<int32, [5]> var_1993_begin_0 = const()[name = tensor<string, []>("op_1993_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> var_1993_end_0 = const()[name = tensor<string, []>("op_1993_end_0"), val = tensor<int32, [5]>([1, 1, 512, 16, 64])];
tensor<bool, [5]> var_1993_end_mask_0 = const()[name = tensor<string, []>("op_1993_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> var_1993_squeeze_mask_0 = const()[name = tensor<string, []>("op_1993_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<fp32, [1, 512, 16, 64]> var_1993 = slice_by_index(begin = var_1993_begin_0, end = var_1993_end_0, end_mask = var_1993_end_mask_0, squeeze_mask = var_1993_squeeze_mask_0, x = cache4)[name = tensor<string, []>("op_1993")];
tensor<int32, []> var_1995_axis_0 = const()[name = tensor<string, []>("op_1995_axis_0"), val = tensor<int32, []>(1)];
tensor<string, []> var_1995_mode_0 = const()[name = tensor<string, []>("op_1995_mode_0"), val = tensor<string, []>("update")];
tensor<bool, []> var_1995_validate_indices_0 = const()[name = tensor<string, []>("op_1995_validate_indices_0"), val = tensor<bool, []>(false)];
tensor<fp32, [1, 512, 16, 64]> var_1995 = scatter_along_axis(axis = var_1995_axis_0, data = var_1993, indices = write_indices_9, mode = var_1995_mode_0, updates = k_19, validate_indices = var_1995_validate_indices_0)[name = tensor<string, []>("op_1995")];
tensor<int32, [5]> concat_29 = const()[name = tensor<string, []>("concat_29"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> concat_30 = const()[name = tensor<string, []>("concat_30"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> new_cache_9_internal_tensor_assign_1_stride_0 = const()[name = tensor<string, []>("new_cache_9_internal_tensor_assign_1_stride_0"), val = tensor<int32, [5]>([1, 1, 1, 1, 1])];
tensor<bool, [5]> new_cache_9_internal_tensor_assign_1_begin_mask_0 = const()[name = tensor<string, []>("new_cache_9_internal_tensor_assign_1_begin_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> new_cache_9_internal_tensor_assign_1_end_mask_0 = const()[name = tensor<string, []>("new_cache_9_internal_tensor_assign_1_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> new_cache_9_internal_tensor_assign_1_squeeze_mask_0 = const()[name = tensor<string, []>("new_cache_9_internal_tensor_assign_1_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<int32, [5]> shape_18 = const()[name = tensor<string, []>("shape_18"), val = tensor<int32, [5]>([2, 1, 512, 16, 64])];
tensor<int32, []> reduce_prod_8 = const()[name = tensor<string, []>("reduce_prod_8"), val = tensor<int32, []>(1048576)];
tensor<int32, []> range_1d_8_start_0 = const()[name = tensor<string, []>("range_1d_8_start_0"), val = tensor<int32, []>(0)];
tensor<int32, []> range_1d_8_step_0 = const()[name = tensor<string, []>("range_1d_8_step_0"), val = tensor<int32, []>(1)];
tensor<int32, [1048576]> range_1d_8 = range_1d(end = reduce_prod_8, start = range_1d_8_start_0, step = range_1d_8_step_0)[name = tensor<string, []>("range_1d_8")];
tensor<int32, [2, 1, 512, 16, 64]> reshape_40 = reshape(shape = shape_18, x = range_1d_8)[name = tensor<string, []>("reshape_40")];
tensor<int32, [1, 512, 16, 64]> slice_by_index_8 = slice_by_index(begin = concat_29, begin_mask = new_cache_9_internal_tensor_assign_1_begin_mask_0, end = concat_30, end_mask = new_cache_9_internal_tensor_assign_1_end_mask_0, squeeze_mask = new_cache_9_internal_tensor_assign_1_squeeze_mask_0, stride = new_cache_9_internal_tensor_assign_1_stride_0, x = reshape_40)[name = tensor<string, []>("slice_by_index_8")];
tensor<int32, [1]> reshape_41_shape_0 = const()[name = tensor<string, []>("reshape_41_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<int32, [524288]> reshape_41 = reshape(shape = reshape_41_shape_0, x = slice_by_index_8)[name = tensor<string, []>("reshape_41")];
tensor<int32, [1]> reshape_42_shape_0 = const()[name = tensor<string, []>("reshape_42_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [524288]> reshape_42 = reshape(shape = reshape_42_shape_0, x = var_1995)[name = tensor<string, []>("reshape_42")];
tensor<int32, [1]> reshape_43_shape_0 = const()[name = tensor<string, []>("reshape_43_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [1048576]> reshape_43 = reshape(shape = reshape_43_shape_0, x = cache4)[name = tensor<string, []>("reshape_43")];
tensor<string, []> scatter_8_mode_0 = const()[name = tensor<string, []>("scatter_8_mode_0"), val = tensor<string, []>("update")];
tensor<int32, []> scatter_8_axis_0 = const()[name = tensor<string, []>("scatter_8_axis_0"), val = tensor<int32, []>(0)];
tensor<bool, []> scatter_8_validate_indices_0 = const()[name = tensor<string, []>("scatter_8_validate_indices_0"), val = tensor<bool, []>(false)];
tensor<fp32, [1048576]> scatter_8 = scatter(axis = scatter_8_axis_0, data = reshape_43, indices = reshape_41, mode = scatter_8_mode_0, updates = reshape_42, validate_indices = scatter_8_validate_indices_0)[name = tensor<string, []>("scatter_8")];
tensor<fp32, [2, 1, 512, 16, 64]> reshape_44 = reshape(shape = shape_18, x = scatter_8)[name = tensor<string, []>("reshape_44")];
tensor<int32, [5]> var_2003_begin_0 = const()[name = tensor<string, []>("op_2003_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
tensor<int32, [5]> var_2003_end_0 = const()[name = tensor<string, []>("op_2003_end_0"), val = tensor<int32, [5]>([2, 1, 512, 16, 64])];
tensor<bool, [5]> var_2003_end_mask_0 = const()[name = tensor<string, []>("op_2003_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> var_2003_squeeze_mask_0 = const()[name = tensor<string, []>("op_2003_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<fp32, [1, 512, 16, 64]> var_2003 = slice_by_index(begin = var_2003_begin_0, end = var_2003_end_0, end_mask = var_2003_end_mask_0, squeeze_mask = var_2003_squeeze_mask_0, x = reshape_44)[name = tensor<string, []>("op_2003")];
tensor<int32, []> var_2005_axis_0 = const()[name = tensor<string, []>("op_2005_axis_0"), val = tensor<int32, []>(1)];
tensor<string, []> var_2005_mode_0 = const()[name = tensor<string, []>("op_2005_mode_0"), val = tensor<string, []>("update")];
tensor<bool, []> var_2005_validate_indices_0 = const()[name = tensor<string, []>("op_2005_validate_indices_0"), val = tensor<bool, []>(false)];
tensor<fp32, [1, 512, 16, 64]> var_2005 = scatter_along_axis(axis = var_2005_axis_0, data = var_2003, indices = write_indices_9, mode = var_2005_mode_0, updates = v_9, validate_indices = var_2005_validate_indices_0)[name = tensor<string, []>("op_2005")];
tensor<int32, [5]> concat_31 = const()[name = tensor<string, []>("concat_31"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
tensor<int32, [5]> concat_32 = const()[name = tensor<string, []>("concat_32"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> new_cache_9_internal_tensor_assign_2_stride_0 = const()[name = tensor<string, []>("new_cache_9_internal_tensor_assign_2_stride_0"), val = tensor<int32, [5]>([1, 1, 1, 1, 1])];
tensor<bool, [5]> new_cache_9_internal_tensor_assign_2_begin_mask_0 = const()[name = tensor<string, []>("new_cache_9_internal_tensor_assign_2_begin_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> new_cache_9_internal_tensor_assign_2_end_mask_0 = const()[name = tensor<string, []>("new_cache_9_internal_tensor_assign_2_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> new_cache_9_internal_tensor_assign_2_squeeze_mask_0 = const()[name = tensor<string, []>("new_cache_9_internal_tensor_assign_2_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<int32, [5]> shape_19 = const()[name = tensor<string, []>("shape_19"), val = tensor<int32, [5]>([2, 1, 512, 16, 64])];
tensor<int32, []> reduce_prod_9 = const()[name = tensor<string, []>("reduce_prod_9"), val = tensor<int32, []>(1048576)];
tensor<int32, []> range_1d_9_start_0 = const()[name = tensor<string, []>("range_1d_9_start_0"), val = tensor<int32, []>(0)];
tensor<int32, []> range_1d_9_step_0 = const()[name = tensor<string, []>("range_1d_9_step_0"), val = tensor<int32, []>(1)];
tensor<int32, [1048576]> range_1d_9 = range_1d(end = reduce_prod_9, start = range_1d_9_start_0, step = range_1d_9_step_0)[name = tensor<string, []>("range_1d_9")];
tensor<int32, [2, 1, 512, 16, 64]> reshape_45 = reshape(shape = shape_19, x = range_1d_9)[name = tensor<string, []>("reshape_45")];
tensor<int32, [1, 512, 16, 64]> slice_by_index_9 = slice_by_index(begin = concat_31, begin_mask = new_cache_9_internal_tensor_assign_2_begin_mask_0, end = concat_32, end_mask = new_cache_9_internal_tensor_assign_2_end_mask_0, squeeze_mask = new_cache_9_internal_tensor_assign_2_squeeze_mask_0, stride = new_cache_9_internal_tensor_assign_2_stride_0, x = reshape_45)[name = tensor<string, []>("slice_by_index_9")];
tensor<int32, [1]> reshape_46_shape_0 = const()[name = tensor<string, []>("reshape_46_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<int32, [524288]> reshape_46 = reshape(shape = reshape_46_shape_0, x = slice_by_index_9)[name = tensor<string, []>("reshape_46")];
tensor<int32, [1]> reshape_47_shape_0 = const()[name = tensor<string, []>("reshape_47_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [524288]> reshape_47 = reshape(shape = reshape_47_shape_0, x = var_2005)[name = tensor<string, []>("reshape_47")];
tensor<int32, [1]> reshape_48_shape_0 = const()[name = tensor<string, []>("reshape_48_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [1048576]> reshape_48 = reshape(shape = reshape_48_shape_0, x = reshape_44)[name = tensor<string, []>("reshape_48")];
tensor<string, []> scatter_9_mode_0 = const()[name = tensor<string, []>("scatter_9_mode_0"), val = tensor<string, []>("update")];
tensor<int32, []> scatter_9_axis_0 = const()[name = tensor<string, []>("scatter_9_axis_0"), val = tensor<int32, []>(0)];
tensor<bool, []> scatter_9_validate_indices_0 = const()[name = tensor<string, []>("scatter_9_validate_indices_0"), val = tensor<bool, []>(false)];
tensor<fp32, [1048576]> scatter_9 = scatter(axis = scatter_9_axis_0, data = reshape_48, indices = reshape_46, mode = scatter_9_mode_0, updates = reshape_47, validate_indices = scatter_9_validate_indices_0)[name = tensor<string, []>("scatter_9")];
tensor<fp32, [2, 1, 512, 16, 64]> new_cache_9_internal_tensor_assign_2 = reshape(shape = shape_19, x = scatter_9)[name = tensor<string, []>("reshape_49")];
tensor<int32, [5]> keys_25_begin_0 = const()[name = tensor<string, []>("keys_25_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> keys_25_end_0 = const()[name = tensor<string, []>("keys_25_end_0"), val = tensor<int32, [5]>([1, 1, 512, 16, 64])];
tensor<bool, [5]> keys_25_end_mask_0 = const()[name = tensor<string, []>("keys_25_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> keys_25_squeeze_mask_0 = const()[name = tensor<string, []>("keys_25_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<fp32, [1, 512, 16, 64]> keys_25 = slice_by_index(begin = keys_25_begin_0, end = keys_25_end_0, end_mask = keys_25_end_mask_0, squeeze_mask = keys_25_squeeze_mask_0, x = new_cache_9_internal_tensor_assign_2)[name = tensor<string, []>("keys_25")];
tensor<int32, [5]> values_25_begin_0 = const()[name = tensor<string, []>("values_25_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
tensor<int32, [5]> values_25_end_0 = const()[name = tensor<string, []>("values_25_end_0"), val = tensor<int32, [5]>([2, 1, 512, 16, 64])];
tensor<bool, [5]> values_25_end_mask_0 = const()[name = tensor<string, []>("values_25_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> values_25_squeeze_mask_0 = const()[name = tensor<string, []>("values_25_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<fp32, [1, 512, 16, 64]> values_25 = slice_by_index(begin = values_25_begin_0, end = values_25_end_0, end_mask = values_25_end_mask_0, squeeze_mask = values_25_squeeze_mask_0, x = new_cache_9_internal_tensor_assign_2)[name = tensor<string, []>("values_25")];
tensor<bool, [1, 512, 16, 64]> var_2017 = not_equal(x = keys_25, y = keys_25)[name = tensor<string, []>("op_2017")];
tensor<fp32, [1, 512, 16, 64]> keys_27 = select(a = var_347, b = keys_25, cond = var_2017)[name = tensor<string, []>("keys_27")];
tensor<bool, [1, 512, 16, 64]> var_2025 = not_equal(x = values_25, y = values_25)[name = tensor<string, []>("op_2025")];
tensor<fp32, [1, 512, 16, 64]> values_27 = select(a = var_347, b = values_25, cond = var_2025)[name = tensor<string, []>("values_27")];
tensor<int32, [4]> var_2049 = const()[name = tensor<string, []>("op_2049"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_2062 = const()[name = tensor<string, []>("op_2062"), val = tensor<int32, [3]>([1, 1, 1])];
tensor<fp32, [1, 1, 1]> var_2063 = reshape(shape = var_2062, x = position4)[name = tensor<string, []>("op_2063")];
tensor<fp32, [1, 175, 1]> q_positions_9 = add(x = var_2063, y = q_offsets_1_promoted)[name = tensor<string, []>("q_positions_9")];
tensor<fp32, []> var_2080 = const()[name = tensor<string, []>("op_2080"), val = tensor<fp32, []>(0x1.5ep+7)];
tensor<fp32, [1, 1, 1]> valid_len_9 = add(x = var_2063, y = var_2080)[name = tensor<string, []>("valid_len_9")];
tensor<bool, [1, 1, 512]> valid_mask_9 = less(x = k_positions_1_promoted, y = valid_len_9)[name = tensor<string, []>("valid_mask_9")];
tensor<bool, [1, 175, 512]> causal_mask_9 = less_equal(x = k_positions_1_promoted, y = q_positions_9)[name = tensor<string, []>("causal_mask_9")];
tensor<bool, [1, 175, 512]> attn_mask_17 = logical_and(x = valid_mask_9, y = causal_mask_9)[name = tensor<string, []>("attn_mask_17")];
tensor<int32, [1]> attn_mask_19_axes_0 = const()[name = tensor<string, []>("attn_mask_19_axes_0"), val = tensor<int32, [1]>([1])];
tensor<bool, [1, 1, 175, 512]> attn_mask_19 = expand_dims(axes = attn_mask_19_axes_0, x = attn_mask_17)[name = tensor<string, []>("attn_mask_19")];
tensor<fp32, [1]> var_2092 = const()[name = tensor<string, []>("op_2092"), val = tensor<fp32, [1]>([0x1.fffe5cp-4])];
tensor<bool, []> var_2098_transpose_x_0 = const()[name = tensor<string, []>("op_2098_transpose_x_0"), val = tensor<bool, []>(false)];
tensor<bool, []> var_2098_transpose_y_0 = const()[name = tensor<string, []>("op_2098_transpose_y_0"), val = tensor<bool, []>(false)];
tensor<int32, [4]> transpose_23_perm_0 = const()[name = tensor<string, []>("transpose_23_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
tensor<int32, [4]> transpose_24_perm_0 = const()[name = tensor<string, []>("transpose_24_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
tensor<fp32, [1, 16, 64, 512]> transpose_24 = transpose(perm = transpose_24_perm_0, x = keys_27)[name = tensor<string, []>("transpose_26")];
tensor<fp32, [1, 16, 175, 64]> transpose_23 = transpose(perm = transpose_23_perm_0, x = q_27)[name = tensor<string, []>("transpose_27")];
tensor<fp32, [1, 16, 175, 512]> var_2098 = matmul(transpose_x = var_2098_transpose_x_0, transpose_y = var_2098_transpose_y_0, x = transpose_23, y = transpose_24)[name = tensor<string, []>("op_2098")];
tensor<fp32, [1, 16, 175, 512]> attn_weights_25 = mul(x = var_2098, y = var_2092)[name = tensor<string, []>("attn_weights_25")];
tensor<bool, [1, 1, 175, 512]> var_2100 = logical_not(x = attn_mask_19)[name = tensor<string, []>("op_2100")];
tensor<fp32, []> var_2101 = const()[name = tensor<string, []>("op_2101"), val = tensor<fp32, []>(-0x1.ff933cp+127)];
tensor<fp32, [1, 16, 175, 512]> attn_weights_27 = select(a = var_2101, b = attn_weights_25, cond = var_2100)[name = tensor<string, []>("attn_weights_27")];
tensor<int32, []> var_2103 = const()[name = tensor<string, []>("op_2103"), val = tensor<int32, []>(-1)];
tensor<fp32, [1, 16, 175, 512]> attn_weights_29 = softmax(axis = var_2103, x = attn_weights_27)[name = tensor<string, []>("attn_weights_29")];
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<fp32, [1, 16, 512, 64]> values_29 = transpose(perm = var_2049, x = values_27)[name = tensor<string, []>("transpose_28")];
tensor<fp32, [1, 16, 175, 64]> attn_output_9 = matmul(transpose_x = attn_output_9_transpose_x_0, transpose_y = attn_output_9_transpose_y_0, x = attn_weights_29, y = values_29)[name = tensor<string, []>("attn_output_9")];
tensor<int32, [4]> var_2111 = const()[name = tensor<string, []>("op_2111"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_2114 = const()[name = tensor<string, []>("op_2114"), val = tensor<int32, [3]>([1, 175, 1024])];
tensor<fp32, [1, 175, 16, 64]> var_2112 = transpose(perm = var_2111, x = attn_output_9)[name = tensor<string, []>("transpose_25")];
tensor<fp32, [1, 175, 1024]> input_41 = reshape(shape = var_2114, x = var_2112)[name = tensor<string, []>("input_41")];
tensor<fp32, [1, 175, 1024]> attn_out_9 = linear(bias = linear_1_bias_0, weight = attn4_out_proj_weight, x = input_41)[name = tensor<string, []>("linear_17")];
tensor<fp32, []> var_2120 = const()[name = tensor<string, []>("op_2120"), val = tensor<fp32, []>(0x1.5ep+7)];
tensor<fp32, [1]> var_2121 = add(x = position4, y = var_2120)[name = tensor<string, []>("op_2121")];
tensor<fp32, [1, 175, 1024]> input_43 = add(x = input_39, y = attn_out_9)[name = tensor<string, []>("input_43")];
tensor<fp32, []> var_2125 = const()[name = tensor<string, []>("op_2125"), val = tensor<fp32, []>(0x1.4f8b58p-17)];
tensor<int32, [1]> input_45_axes_0 = const()[name = tensor<string, []>("input_45_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [1, 175, 1024]> input_45 = layer_norm(axes = input_45_axes_0, beta = norm4_2_bias, epsilon = var_2125, gamma = norm4_2_weight, x = input_43)[name = tensor<string, []>("input_45")];
tensor<fp32, [1, 175, 4096]> var_2133 = linear(bias = linear_2_bias_0, weight = linear4_1_weight, x = input_45)[name = tensor<string, []>("linear_18")];
tensor<string, []> input_47_mode_0 = const()[name = tensor<string, []>("input_47_mode_0"), val = tensor<string, []>("EXACT")];
tensor<fp32, [1, 175, 4096]> input_47 = gelu(mode = input_47_mode_0, x = var_2133)[name = tensor<string, []>("input_47")];
tensor<fp32, [1, 175, 1024]> ffn_out_9 = linear(bias = linear_1_bias_0, weight = linear4_2_weight, x = input_47)[name = tensor<string, []>("linear_19")];
tensor<fp32, [1, 175, 1024]> input_49 = add(x = input_43, y = ffn_out_9)[name = tensor<string, []>("input_49")];
tensor<fp32, []> var_2142 = const()[name = tensor<string, []>("op_2142"), val = tensor<fp32, []>(0x1.4f8b58p-17)];
tensor<int32, [1]> x_axes_0 = const()[name = tensor<string, []>("x_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [1, 175, 1024]> x = layer_norm(axes = x_axes_0, beta = norm5_1_bias, epsilon = var_2142, gamma = norm5_1_weight, x = input_49)[name = tensor<string, []>("x")];
tensor<fp32, [1, 175, 3072]> var_2163 = linear(bias = linear_0_bias_0, weight = attn5_in_proj_weight, x = x)[name = tensor<string, []>("linear_20")];
tensor<int32, [5]> var_2167 = const()[name = tensor<string, []>("op_2167"), val = tensor<int32, [5]>([1, 175, 3, 16, 64])];
tensor<fp32, [1, 175, 3, 16, 64]> qkv = reshape(shape = var_2167, x = var_2163)[name = tensor<string, []>("qkv")];
tensor<int32, [5]> k_21_begin_0 = const()[name = tensor<string, []>("k_21_begin_0"), val = tensor<int32, [5]>([0, 0, 1, 0, 0])];
tensor<int32, [5]> k_21_end_0 = const()[name = tensor<string, []>("k_21_end_0"), val = tensor<int32, [5]>([1, 175, 2, 16, 64])];
tensor<bool, [5]> k_21_end_mask_0 = const()[name = tensor<string, []>("k_21_end_mask_0"), val = tensor<bool, [5]>([true, true, false, true, true])];
tensor<bool, [5]> k_21_squeeze_mask_0 = const()[name = tensor<string, []>("k_21_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, true, false, false])];
tensor<fp32, [1, 175, 16, 64]> k_21 = slice_by_index(begin = k_21_begin_0, end = k_21_end_0, end_mask = k_21_end_mask_0, squeeze_mask = k_21_squeeze_mask_0, x = qkv)[name = tensor<string, []>("k_21")];
tensor<int32, [5]> v_begin_0 = const()[name = tensor<string, []>("v_begin_0"), val = tensor<int32, [5]>([0, 0, 2, 0, 0])];
tensor<int32, [5]> v_end_0 = const()[name = tensor<string, []>("v_end_0"), val = tensor<int32, [5]>([1, 175, 3, 16, 64])];
tensor<bool, [5]> v_end_mask_0 = const()[name = tensor<string, []>("v_end_mask_0"), val = tensor<bool, [5]>([true, true, false, true, true])];
tensor<bool, [5]> v_squeeze_mask_0 = const()[name = tensor<string, []>("v_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, true, false, false])];
tensor<fp32, [1, 175, 16, 64]> v = slice_by_index(begin = v_begin_0, end = v_end_0, end_mask = v_end_mask_0, squeeze_mask = v_squeeze_mask_0, x = qkv)[name = tensor<string, []>("v")];
tensor<fp32, [32]> freqs = const()[name = tensor<string, []>("freqs"), val = tensor<fp32, [32]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(266469632)))];
tensor<int32, [2]> var_2252 = const()[name = tensor<string, []>("op_2252"), val = tensor<int32, [2]>([1, 1])];
tensor<fp32, [1, 1]> var_2253 = reshape(shape = var_2252, x = position5)[name = tensor<string, []>("op_2253")];
tensor<fp32, [1, 175]> ts_33 = add(x = ts_1_promoted, y = var_2253)[name = tensor<string, []>("ts_33")];
tensor<int32, [4]> var_2258 = const()[name = tensor<string, []>("op_2258"), val = tensor<int32, [4]>([1, 175, 1, 1])];
tensor<fp32, [1, 175, 1, 1]> ts = reshape(shape = var_2258, x = ts_33)[name = tensor<string, []>("ts")];
tensor<int32, [5]> var_2262 = const()[name = tensor<string, []>("op_2262"), val = tensor<int32, [5]>([1, 175, 16, 32, 2])];
tensor<fp32, [1, 175, 16, 32, 2]> k_complex = reshape(shape = var_2262, x = k_21)[name = tensor<string, []>("k_complex")];
tensor<int32, [5]> var_2266_begin_0 = const()[name = tensor<string, []>("op_2266_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> var_2266_end_0 = const()[name = tensor<string, []>("op_2266_end_0"), val = tensor<int32, [5]>([1, 175, 16, 32, 1])];
tensor<bool, [5]> var_2266_end_mask_0 = const()[name = tensor<string, []>("op_2266_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
tensor<bool, [5]> var_2266_squeeze_mask_0 = const()[name = tensor<string, []>("op_2266_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
tensor<fp32, [1, 175, 16, 32]> var_2266 = slice_by_index(begin = var_2266_begin_0, end = var_2266_end_0, end_mask = var_2266_end_mask_0, squeeze_mask = var_2266_squeeze_mask_0, x = k_complex)[name = tensor<string, []>("op_2266")];
tensor<int32, [5]> var_2274_begin_0 = const()[name = tensor<string, []>("op_2274_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 1])];
tensor<int32, [5]> var_2274_end_0 = const()[name = tensor<string, []>("op_2274_end_0"), val = tensor<int32, [5]>([1, 175, 16, 32, 2])];
tensor<bool, [5]> var_2274_end_mask_0 = const()[name = tensor<string, []>("op_2274_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
tensor<bool, [5]> var_2274_squeeze_mask_0 = const()[name = tensor<string, []>("op_2274_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
tensor<fp32, [1, 175, 16, 32]> var_2274 = slice_by_index(begin = var_2274_begin_0, end = var_2274_end_0, end_mask = var_2274_end_mask_0, squeeze_mask = var_2274_squeeze_mask_0, x = k_complex)[name = tensor<string, []>("op_2274")];
tensor<fp32, [1, 175, 1, 32]> var_2280 = mul(x = freqs, y = ts)[name = tensor<string, []>("op_2280")];
tensor<fp32, [1, 175, 1, 32]> rotr = cos(x = var_2280)[name = tensor<string, []>("rotr")];
tensor<fp32, [1, 175, 1, 32]> roti = sin(x = var_2280)[name = tensor<string, []>("roti")];
tensor<fp32, [1, 175, 16, 32]> var_2284 = mul(x = var_2266, y = rotr)[name = tensor<string, []>("op_2284")];
tensor<fp32, [1, 175, 16, 32]> var_2285 = mul(x = var_2274, y = roti)[name = tensor<string, []>("op_2285")];
tensor<fp32, [1, 175, 16, 32]> kor_21 = sub(x = var_2284, y = var_2285)[name = tensor<string, []>("kor_21")];
tensor<fp32, [1, 175, 16, 32]> var_2288 = mul(x = var_2266, y = roti)[name = tensor<string, []>("op_2288")];
tensor<fp32, [1, 175, 16, 32]> var_2289 = mul(x = var_2274, y = rotr)[name = tensor<string, []>("op_2289")];
tensor<fp32, [1, 175, 16, 32]> koi_21 = add(x = var_2288, y = var_2289)[name = tensor<string, []>("koi_21")];
tensor<int32, []> ko_axis_0 = const()[name = tensor<string, []>("ko_axis_0"), val = tensor<int32, []>(-1)];
tensor<fp32, [1, 175, 16, 32, 2]> ko = stack(axis = ko_axis_0, values = (kor_21, koi_21))[name = tensor<string, []>("ko")];
tensor<int32, [4]> var_2305 = const()[name = tensor<string, []>("op_2305"), val = tensor<int32, [4]>([1, 175, 16, 64])];
tensor<fp32, [1, 175, 16, 64]> k = reshape(shape = var_2305, x = ko)[name = tensor<string, []>("k")];
tensor<int32, [4]> var_2312 = const()[name = tensor<string, []>("op_2312"), val = tensor<int32, [4]>([1, 1, 1, 1])];
tensor<fp32, [1, 1, 1, 1]> write_base_float = reshape(shape = var_2312, x = position5)[name = tensor<string, []>("write_base_float")];
tensor<fp32, [1, 175, 1, 1]> write_indices_float_21 = add(x = write_base_float, y = write_offsets_float_1_promoted)[name = tensor<string, []>("write_indices_float_21")];
tensor<fp32, []> _inversed_2327_y_0 = const()[name = tensor<string, []>("_inversed_2327_y_0"), val = tensor<fp32, []>(0x1p-9)];
tensor<fp32, [1, 175, 1, 1]> _inversed_2327 = mul(x = write_indices_float_21, y = _inversed_2327_y_0)[name = tensor<string, []>("_inversed_2327")];
tensor<fp32, [1, 175, 1, 1]> var_2328 = floor(x = _inversed_2327)[name = tensor<string, []>("op_2328")];
tensor<fp32, []> var_2329 = const()[name = tensor<string, []>("op_2329"), val = tensor<fp32, []>(0x1p+9)];
tensor<fp32, [1, 175, 1, 1]> var_2330 = mul(x = var_2328, y = var_2329)[name = tensor<string, []>("op_2330")];
tensor<fp32, [1, 175, 1, 1]> write_indices_float = sub(x = write_indices_float_21, y = var_2330)[name = tensor<string, []>("write_indices_float")];
tensor<string, []> var_2337_dtype_0 = const()[name = tensor<string, []>("op_2337_dtype_0"), val = tensor<string, []>("int32")];
tensor<int32, [4]> write_indices_reps_0 = const()[name = tensor<string, []>("write_indices_reps_0"), val = tensor<int32, [4]>([1, 1, 16, 64])];
tensor<int32, [1, 175, 1, 1]> var_2337 = cast(dtype = var_2337_dtype_0, x = write_indices_float)[name = tensor<string, []>("cast_99")];
tensor<int32, [1, 175, 16, 64]> write_indices = tile(reps = write_indices_reps_0, x = var_2337)[name = tensor<string, []>("write_indices")];
tensor<int32, [5]> var_2345_begin_0 = const()[name = tensor<string, []>("op_2345_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> var_2345_end_0 = const()[name = tensor<string, []>("op_2345_end_0"), val = tensor<int32, [5]>([1, 1, 512, 16, 64])];
tensor<bool, [5]> var_2345_end_mask_0 = const()[name = tensor<string, []>("op_2345_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> var_2345_squeeze_mask_0 = const()[name = tensor<string, []>("op_2345_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<fp32, [1, 512, 16, 64]> var_2345 = slice_by_index(begin = var_2345_begin_0, end = var_2345_end_0, end_mask = var_2345_end_mask_0, squeeze_mask = var_2345_squeeze_mask_0, x = cache5)[name = tensor<string, []>("op_2345")];
tensor<int32, []> var_2347_axis_0 = const()[name = tensor<string, []>("op_2347_axis_0"), val = tensor<int32, []>(1)];
tensor<string, []> var_2347_mode_0 = const()[name = tensor<string, []>("op_2347_mode_0"), val = tensor<string, []>("update")];
tensor<bool, []> var_2347_validate_indices_0 = const()[name = tensor<string, []>("op_2347_validate_indices_0"), val = tensor<bool, []>(false)];
tensor<fp32, [1, 512, 16, 64]> var_2347 = scatter_along_axis(axis = var_2347_axis_0, data = var_2345, indices = write_indices, mode = var_2347_mode_0, updates = k, validate_indices = var_2347_validate_indices_0)[name = tensor<string, []>("op_2347")];
tensor<int32, [5]> concat_36 = const()[name = tensor<string, []>("concat_36"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> concat_37 = const()[name = tensor<string, []>("concat_37"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> new_cache_internal_tensor_assign_1_stride_0 = const()[name = tensor<string, []>("new_cache_internal_tensor_assign_1_stride_0"), val = tensor<int32, [5]>([1, 1, 1, 1, 1])];
tensor<bool, [5]> new_cache_internal_tensor_assign_1_begin_mask_0 = const()[name = tensor<string, []>("new_cache_internal_tensor_assign_1_begin_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> new_cache_internal_tensor_assign_1_end_mask_0 = const()[name = tensor<string, []>("new_cache_internal_tensor_assign_1_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> new_cache_internal_tensor_assign_1_squeeze_mask_0 = const()[name = tensor<string, []>("new_cache_internal_tensor_assign_1_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<int32, [5]> shape_20 = const()[name = tensor<string, []>("shape_20"), val = tensor<int32, [5]>([2, 1, 512, 16, 64])];
tensor<int32, []> reduce_prod_10 = const()[name = tensor<string, []>("reduce_prod_10"), val = tensor<int32, []>(1048576)];
tensor<int32, []> range_1d_10_start_0 = const()[name = tensor<string, []>("range_1d_10_start_0"), val = tensor<int32, []>(0)];
tensor<int32, []> range_1d_10_step_0 = const()[name = tensor<string, []>("range_1d_10_step_0"), val = tensor<int32, []>(1)];
tensor<int32, [1048576]> range_1d_10 = range_1d(end = reduce_prod_10, start = range_1d_10_start_0, step = range_1d_10_step_0)[name = tensor<string, []>("range_1d_10")];
tensor<int32, [2, 1, 512, 16, 64]> reshape_50 = reshape(shape = shape_20, x = range_1d_10)[name = tensor<string, []>("reshape_50")];
tensor<int32, [1, 512, 16, 64]> slice_by_index_10 = slice_by_index(begin = concat_36, begin_mask = new_cache_internal_tensor_assign_1_begin_mask_0, end = concat_37, end_mask = new_cache_internal_tensor_assign_1_end_mask_0, squeeze_mask = new_cache_internal_tensor_assign_1_squeeze_mask_0, stride = new_cache_internal_tensor_assign_1_stride_0, x = reshape_50)[name = tensor<string, []>("slice_by_index_10")];
tensor<int32, [1]> reshape_51_shape_0 = const()[name = tensor<string, []>("reshape_51_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<int32, [524288]> reshape_51 = reshape(shape = reshape_51_shape_0, x = slice_by_index_10)[name = tensor<string, []>("reshape_51")];
tensor<int32, [1]> reshape_52_shape_0 = const()[name = tensor<string, []>("reshape_52_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [524288]> reshape_52 = reshape(shape = reshape_52_shape_0, x = var_2347)[name = tensor<string, []>("reshape_52")];
tensor<int32, [1]> reshape_53_shape_0 = const()[name = tensor<string, []>("reshape_53_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [1048576]> reshape_53 = reshape(shape = reshape_53_shape_0, x = cache5)[name = tensor<string, []>("reshape_53")];
tensor<string, []> scatter_10_mode_0 = const()[name = tensor<string, []>("scatter_10_mode_0"), val = tensor<string, []>("update")];
tensor<int32, []> scatter_10_axis_0 = const()[name = tensor<string, []>("scatter_10_axis_0"), val = tensor<int32, []>(0)];
tensor<bool, []> scatter_10_validate_indices_0 = const()[name = tensor<string, []>("scatter_10_validate_indices_0"), val = tensor<bool, []>(false)];
tensor<fp32, [1048576]> scatter_10 = scatter(axis = scatter_10_axis_0, data = reshape_53, indices = reshape_51, mode = scatter_10_mode_0, updates = reshape_52, validate_indices = scatter_10_validate_indices_0)[name = tensor<string, []>("scatter_10")];
tensor<fp32, [2, 1, 512, 16, 64]> reshape_54 = reshape(shape = shape_20, x = scatter_10)[name = tensor<string, []>("reshape_54")];
tensor<int32, [5]> var_2355_begin_0 = const()[name = tensor<string, []>("op_2355_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
tensor<int32, [5]> var_2355_end_0 = const()[name = tensor<string, []>("op_2355_end_0"), val = tensor<int32, [5]>([2, 1, 512, 16, 64])];
tensor<bool, [5]> var_2355_end_mask_0 = const()[name = tensor<string, []>("op_2355_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> var_2355_squeeze_mask_0 = const()[name = tensor<string, []>("op_2355_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<fp32, [1, 512, 16, 64]> var_2355 = slice_by_index(begin = var_2355_begin_0, end = var_2355_end_0, end_mask = var_2355_end_mask_0, squeeze_mask = var_2355_squeeze_mask_0, x = reshape_54)[name = tensor<string, []>("op_2355")];
tensor<int32, []> var_2357_axis_0 = const()[name = tensor<string, []>("op_2357_axis_0"), val = tensor<int32, []>(1)];
tensor<string, []> var_2357_mode_0 = const()[name = tensor<string, []>("op_2357_mode_0"), val = tensor<string, []>("update")];
tensor<bool, []> var_2357_validate_indices_0 = const()[name = tensor<string, []>("op_2357_validate_indices_0"), val = tensor<bool, []>(false)];
tensor<fp32, [1, 512, 16, 64]> var_2357 = scatter_along_axis(axis = var_2357_axis_0, data = var_2355, indices = write_indices, mode = var_2357_mode_0, updates = v, validate_indices = var_2357_validate_indices_0)[name = tensor<string, []>("op_2357")];
tensor<int32, [5]> concat_38 = const()[name = tensor<string, []>("concat_38"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
tensor<int32, [5]> concat_39 = const()[name = tensor<string, []>("concat_39"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
tensor<int32, [5]> new_cache_internal_tensor_assign_2_stride_0 = const()[name = tensor<string, []>("new_cache_internal_tensor_assign_2_stride_0"), val = tensor<int32, [5]>([1, 1, 1, 1, 1])];
tensor<bool, [5]> new_cache_internal_tensor_assign_2_begin_mask_0 = const()[name = tensor<string, []>("new_cache_internal_tensor_assign_2_begin_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> new_cache_internal_tensor_assign_2_end_mask_0 = const()[name = tensor<string, []>("new_cache_internal_tensor_assign_2_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
tensor<bool, [5]> new_cache_internal_tensor_assign_2_squeeze_mask_0 = const()[name = tensor<string, []>("new_cache_internal_tensor_assign_2_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
tensor<int32, [5]> shape_21 = const()[name = tensor<string, []>("shape_21"), val = tensor<int32, [5]>([2, 1, 512, 16, 64])];
tensor<int32, []> reduce_prod_11 = const()[name = tensor<string, []>("reduce_prod_11"), val = tensor<int32, []>(1048576)];
tensor<int32, []> range_1d_11_start_0 = const()[name = tensor<string, []>("range_1d_11_start_0"), val = tensor<int32, []>(0)];
tensor<int32, []> range_1d_11_step_0 = const()[name = tensor<string, []>("range_1d_11_step_0"), val = tensor<int32, []>(1)];
tensor<int32, [1048576]> range_1d_11 = range_1d(end = reduce_prod_11, start = range_1d_11_start_0, step = range_1d_11_step_0)[name = tensor<string, []>("range_1d_11")];
tensor<int32, [2, 1, 512, 16, 64]> reshape_55 = reshape(shape = shape_21, x = range_1d_11)[name = tensor<string, []>("reshape_55")];
tensor<int32, [1, 512, 16, 64]> slice_by_index_11 = slice_by_index(begin = concat_38, begin_mask = new_cache_internal_tensor_assign_2_begin_mask_0, end = concat_39, end_mask = new_cache_internal_tensor_assign_2_end_mask_0, squeeze_mask = new_cache_internal_tensor_assign_2_squeeze_mask_0, stride = new_cache_internal_tensor_assign_2_stride_0, x = reshape_55)[name = tensor<string, []>("slice_by_index_11")];
tensor<int32, [1]> reshape_56_shape_0 = const()[name = tensor<string, []>("reshape_56_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<int32, [524288]> reshape_56 = reshape(shape = reshape_56_shape_0, x = slice_by_index_11)[name = tensor<string, []>("reshape_56")];
tensor<int32, [1]> reshape_57_shape_0 = const()[name = tensor<string, []>("reshape_57_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [524288]> reshape_57 = reshape(shape = reshape_57_shape_0, x = var_2357)[name = tensor<string, []>("reshape_57")];
tensor<int32, [1]> reshape_58_shape_0 = const()[name = tensor<string, []>("reshape_58_shape_0"), val = tensor<int32, [1]>([-1])];
tensor<fp32, [1048576]> reshape_58 = reshape(shape = reshape_58_shape_0, x = reshape_54)[name = tensor<string, []>("reshape_58")];
tensor<string, []> scatter_11_mode_0 = const()[name = tensor<string, []>("scatter_11_mode_0"), val = tensor<string, []>("update")];
tensor<int32, []> scatter_11_axis_0 = const()[name = tensor<string, []>("scatter_11_axis_0"), val = tensor<int32, []>(0)];
tensor<bool, []> scatter_11_validate_indices_0 = const()[name = tensor<string, []>("scatter_11_validate_indices_0"), val = tensor<bool, []>(false)];
tensor<fp32, [1048576]> scatter_11 = scatter(axis = scatter_11_axis_0, data = reshape_58, indices = reshape_56, mode = scatter_11_mode_0, updates = reshape_57, validate_indices = scatter_11_validate_indices_0)[name = tensor<string, []>("scatter_11")];
tensor<fp32, [2, 1, 512, 16, 64]> new_cache_internal_tensor_assign_2 = reshape(shape = shape_21, x = scatter_11)[name = tensor<string, []>("reshape_59")];
tensor<fp32, []> var_2364 = const()[name = tensor<string, []>("op_2364"), val = tensor<fp32, []>(0x1.5ep+7)];
tensor<fp32, [1]> var_2365 = add(x = position5, y = var_2364)[name = tensor<string, []>("op_2365")];
} -> (new_cache_1_internal_tensor_assign_2, var_445, new_cache_3_internal_tensor_assign_2, var_864, new_cache_5_internal_tensor_assign_2, var_1283, new_cache_7_internal_tensor_assign_2, var_1702, new_cache_9_internal_tensor_assign_2, var_2121, new_cache_internal_tensor_assign_2, var_2365);
}