Hugging Face's logo

FluidInference
/
pocket-tts-coreml

Text-to-Speech
Core ML
English
tts
apple
on-device
phonon
Model card Files
xet
 Community
4
alexwengg commited on
Commit
94fe1e4
·
verified ·
1 Parent(s): 26a814a

feat(voice-cloning): per-language mimi_encoderv3 for italian_24l (#793)

Browse files
Files changed (4) hide show
  1. v2.1/italian_24l/mimi_encoderv3.mlmodelc/analytics/coremldata.bin +3 -0
  2. v2.1/italian_24l/mimi_encoderv3.mlmodelc/coremldata.bin +3 -0
  3. v2.1/italian_24l/mimi_encoderv3.mlmodelc/model.mil +376 -0
  4. v2.1/italian_24l/mimi_encoderv3.mlmodelc/weights/weight.bin +3 -0
v2.1/italian_24l/mimi_encoderv3.mlmodelc/analytics/coremldata.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1e34526bdf29f919365c3dee8e5bc0924bd290539b16836248d9c4e8b74d71ed
3
+ size 243
v2.1/italian_24l/mimi_encoderv3.mlmodelc/coremldata.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:118b73b9ca6f5a37444aecd26f22fd8179552109c82d489310145a696bf1fe0f
3
+ size 379
v2.1/italian_24l/mimi_encoderv3.mlmodelc/model.mil ADDED
@@ -0,0 +1,376 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ program(1.0)
2
+ [buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3520.4.1"}, {"coremlc-version", "3520.5.1.2"}, {"coremltools-component-torch", "2.7.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})]
3
+ {
4
+ func main<ios17>(tensor<fp32, [1, 1, 240000]> audio) {
5
+ tensor<fp32, []> var_8 = const()[name = tensor<string, []>("op_8"), val = tensor<fp32, []>(0x1p+0)];
6
+ tensor<int32, [6]> input_1_pad_0 = const()[name = tensor<string, []>("input_1_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 6, 0])];
7
+ tensor<string, []> input_1_mode_0 = const()[name = tensor<string, []>("input_1_mode_0"), val = tensor<string, []>("constant")];
8
+ tensor<string, []> audio_to_fp16_dtype_0 = const()[name = tensor<string, []>("audio_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
9
+ tensor<fp16, []> const_0_to_fp16 = const()[name = tensor<string, []>("const_0_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
10
+ tensor<fp16, [1, 1, 240000]> audio_to_fp16 = cast(dtype = audio_to_fp16_dtype_0, x = audio)[name = tensor<string, []>("cast_27")];
11
+ tensor<fp16, [1, 1, 240006]> input_1_cast_fp16 = pad(constant_val = const_0_to_fp16, mode = input_1_mode_0, pad = input_1_pad_0, x = audio_to_fp16)[name = tensor<string, []>("input_1_cast_fp16")];
12
+ tensor<string, []> input_3_pad_type_0 = const()[name = tensor<string, []>("input_3_pad_type_0"), val = tensor<string, []>("valid")];
13
+ tensor<int32, [1]> input_3_strides_0 = const()[name = tensor<string, []>("input_3_strides_0"), val = tensor<int32, [1]>([1])];
14
+ tensor<int32, [2]> input_3_pad_0 = const()[name = tensor<string, []>("input_3_pad_0"), val = tensor<int32, [2]>([0, 0])];
15
+ tensor<int32, [1]> input_3_dilations_0 = const()[name = tensor<string, []>("input_3_dilations_0"), val = tensor<int32, [1]>([1])];
16
+ tensor<int32, []> input_3_groups_0 = const()[name = tensor<string, []>("input_3_groups_0"), val = tensor<int32, []>(1)];
17
+ tensor<fp16, [64, 1, 7]> encoder_model_0_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_model_0_conv_weight_to_fp16"), val = tensor<fp16, [64, 1, 7]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
18
+ tensor<fp16, [64]> encoder_model_0_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_model_0_conv_bias_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1024)))];
19
+ tensor<fp16, [1, 64, 240000]> input_3_cast_fp16 = conv(bias = encoder_model_0_conv_bias_to_fp16, dilations = input_3_dilations_0, groups = input_3_groups_0, pad = input_3_pad_0, pad_type = input_3_pad_type_0, strides = input_3_strides_0, weight = encoder_model_0_conv_weight_to_fp16, x = input_1_cast_fp16)[name = tensor<string, []>("input_3_cast_fp16")];
20
+ tensor<fp16, [1, 64, 240000]> input_5_cast_fp16 = elu(alpha = var_8, x = input_3_cast_fp16)[name = tensor<string, []>("input_5_cast_fp16")];
21
+ tensor<int32, [6]> input_7_pad_0 = const()[name = tensor<string, []>("input_7_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
22
+ tensor<string, []> input_7_mode_0 = const()[name = tensor<string, []>("input_7_mode_0"), val = tensor<string, []>("constant")];
23
+ tensor<fp16, []> const_1_to_fp16 = const()[name = tensor<string, []>("const_1_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
24
+ tensor<fp16, [1, 64, 240002]> input_7_cast_fp16 = pad(constant_val = const_1_to_fp16, mode = input_7_mode_0, pad = input_7_pad_0, x = input_5_cast_fp16)[name = tensor<string, []>("input_7_cast_fp16")];
25
+ tensor<string, []> input_9_pad_type_0 = const()[name = tensor<string, []>("input_9_pad_type_0"), val = tensor<string, []>("valid")];
26
+ tensor<int32, [1]> input_9_strides_0 = const()[name = tensor<string, []>("input_9_strides_0"), val = tensor<int32, [1]>([1])];
27
+ tensor<int32, [2]> input_9_pad_0 = const()[name = tensor<string, []>("input_9_pad_0"), val = tensor<int32, [2]>([0, 0])];
28
+ tensor<int32, [1]> input_9_dilations_0 = const()[name = tensor<string, []>("input_9_dilations_0"), val = tensor<int32, [1]>([1])];
29
+ tensor<int32, []> input_9_groups_0 = const()[name = tensor<string, []>("input_9_groups_0"), val = tensor<int32, []>(1)];
30
+ tensor<fp16, [32, 64, 3]> encoder_model_1_block_1_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_model_1_block_1_conv_weight_to_fp16"), val = tensor<fp16, [32, 64, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1216)))];
31
+ tensor<fp16, [32]> encoder_model_1_block_1_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_model_1_block_1_conv_bias_to_fp16"), val = tensor<fp16, [32]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13568)))];
32
+ tensor<fp16, [1, 32, 240000]> input_9_cast_fp16 = conv(bias = encoder_model_1_block_1_conv_bias_to_fp16, dilations = input_9_dilations_0, groups = input_9_groups_0, pad = input_9_pad_0, pad_type = input_9_pad_type_0, strides = input_9_strides_0, weight = encoder_model_1_block_1_conv_weight_to_fp16, x = input_7_cast_fp16)[name = tensor<string, []>("input_9_cast_fp16")];
33
+ tensor<fp16, [1, 32, 240000]> input_11_cast_fp16 = elu(alpha = var_8, x = input_9_cast_fp16)[name = tensor<string, []>("input_11_cast_fp16")];
34
+ tensor<string, []> v_1_pad_type_0 = const()[name = tensor<string, []>("v_1_pad_type_0"), val = tensor<string, []>("valid")];
35
+ tensor<int32, [1]> v_1_strides_0 = const()[name = tensor<string, []>("v_1_strides_0"), val = tensor<int32, [1]>([1])];
36
+ tensor<int32, [2]> v_1_pad_0 = const()[name = tensor<string, []>("v_1_pad_0"), val = tensor<int32, [2]>([0, 0])];
37
+ tensor<int32, [1]> v_1_dilations_0 = const()[name = tensor<string, []>("v_1_dilations_0"), val = tensor<int32, [1]>([1])];
38
+ tensor<int32, []> v_1_groups_0 = const()[name = tensor<string, []>("v_1_groups_0"), val = tensor<int32, []>(1)];
39
+ tensor<fp16, [64, 32, 1]> encoder_model_1_block_3_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_model_1_block_3_conv_weight_to_fp16"), val = tensor<fp16, [64, 32, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13696)))];
40
+ tensor<fp16, [64]> encoder_model_1_block_3_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_model_1_block_3_conv_bias_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17856)))];
41
+ tensor<fp16, [1, 64, 240000]> v_1_cast_fp16 = conv(bias = encoder_model_1_block_3_conv_bias_to_fp16, dilations = v_1_dilations_0, groups = v_1_groups_0, pad = v_1_pad_0, pad_type = v_1_pad_type_0, strides = v_1_strides_0, weight = encoder_model_1_block_3_conv_weight_to_fp16, x = input_11_cast_fp16)[name = tensor<string, []>("v_1_cast_fp16")];
42
+ tensor<fp16, [1, 64, 240000]> input_13_cast_fp16 = add(x = input_3_cast_fp16, y = v_1_cast_fp16)[name = tensor<string, []>("input_13_cast_fp16")];
43
+ tensor<fp16, [1, 64, 240000]> input_15_cast_fp16 = elu(alpha = var_8, x = input_13_cast_fp16)[name = tensor<string, []>("input_15_cast_fp16")];
44
+ tensor<int32, [6]> input_17_pad_0 = const()[name = tensor<string, []>("input_17_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 0])];
45
+ tensor<string, []> input_17_mode_0 = const()[name = tensor<string, []>("input_17_mode_0"), val = tensor<string, []>("constant")];
46
+ tensor<fp16, []> const_2_to_fp16 = const()[name = tensor<string, []>("const_2_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
47
+ tensor<fp16, [1, 64, 240004]> input_17_cast_fp16 = pad(constant_val = const_2_to_fp16, mode = input_17_mode_0, pad = input_17_pad_0, x = input_15_cast_fp16)[name = tensor<string, []>("input_17_cast_fp16")];
48
+ tensor<string, []> input_19_pad_type_0 = const()[name = tensor<string, []>("input_19_pad_type_0"), val = tensor<string, []>("valid")];
49
+ tensor<int32, [1]> input_19_strides_0 = const()[name = tensor<string, []>("input_19_strides_0"), val = tensor<int32, [1]>([4])];
50
+ tensor<int32, [2]> input_19_pad_0 = const()[name = tensor<string, []>("input_19_pad_0"), val = tensor<int32, [2]>([0, 0])];
51
+ tensor<int32, [1]> input_19_dilations_0 = const()[name = tensor<string, []>("input_19_dilations_0"), val = tensor<int32, [1]>([1])];
52
+ tensor<int32, []> input_19_groups_0 = const()[name = tensor<string, []>("input_19_groups_0"), val = tensor<int32, []>(1)];
53
+ tensor<fp16, [128, 64, 8]> encoder_model_3_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_model_3_conv_weight_to_fp16"), val = tensor<fp16, [128, 64, 8]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18048)))];
54
+ tensor<fp16, [128]> encoder_model_3_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_model_3_conv_bias_to_fp16"), val = tensor<fp16, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149184)))];
55
+ tensor<fp16, [1, 128, 60000]> input_19_cast_fp16 = conv(bias = encoder_model_3_conv_bias_to_fp16, dilations = input_19_dilations_0, groups = input_19_groups_0, pad = input_19_pad_0, pad_type = input_19_pad_type_0, strides = input_19_strides_0, weight = encoder_model_3_conv_weight_to_fp16, x = input_17_cast_fp16)[name = tensor<string, []>("input_19_cast_fp16")];
56
+ tensor<fp16, [1, 128, 60000]> input_21_cast_fp16 = elu(alpha = var_8, x = input_19_cast_fp16)[name = tensor<string, []>("input_21_cast_fp16")];
57
+ tensor<int32, [6]> input_23_pad_0 = const()[name = tensor<string, []>("input_23_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
58
+ tensor<string, []> input_23_mode_0 = const()[name = tensor<string, []>("input_23_mode_0"), val = tensor<string, []>("constant")];
59
+ tensor<fp16, []> const_3_to_fp16 = const()[name = tensor<string, []>("const_3_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
60
+ tensor<fp16, [1, 128, 60002]> input_23_cast_fp16 = pad(constant_val = const_3_to_fp16, mode = input_23_mode_0, pad = input_23_pad_0, x = input_21_cast_fp16)[name = tensor<string, []>("input_23_cast_fp16")];
61
+ tensor<string, []> input_25_pad_type_0 = const()[name = tensor<string, []>("input_25_pad_type_0"), val = tensor<string, []>("valid")];
62
+ tensor<int32, [1]> input_25_strides_0 = const()[name = tensor<string, []>("input_25_strides_0"), val = tensor<int32, [1]>([1])];
63
+ tensor<int32, [2]> input_25_pad_0 = const()[name = tensor<string, []>("input_25_pad_0"), val = tensor<int32, [2]>([0, 0])];
64
+ tensor<int32, [1]> input_25_dilations_0 = const()[name = tensor<string, []>("input_25_dilations_0"), val = tensor<int32, [1]>([1])];
65
+ tensor<int32, []> input_25_groups_0 = const()[name = tensor<string, []>("input_25_groups_0"), val = tensor<int32, []>(1)];
66
+ tensor<fp16, [64, 128, 3]> encoder_model_4_block_1_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_model_4_block_1_conv_weight_to_fp16"), val = tensor<fp16, [64, 128, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149504)))];
67
+ tensor<fp16, [64]> encoder_model_4_block_1_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_model_4_block_1_conv_bias_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198720)))];
68
+ tensor<fp16, [1, 64, 60000]> input_25_cast_fp16 = conv(bias = encoder_model_4_block_1_conv_bias_to_fp16, dilations = input_25_dilations_0, groups = input_25_groups_0, pad = input_25_pad_0, pad_type = input_25_pad_type_0, strides = input_25_strides_0, weight = encoder_model_4_block_1_conv_weight_to_fp16, x = input_23_cast_fp16)[name = tensor<string, []>("input_25_cast_fp16")];
69
+ tensor<fp16, [1, 64, 60000]> input_27_cast_fp16 = elu(alpha = var_8, x = input_25_cast_fp16)[name = tensor<string, []>("input_27_cast_fp16")];
70
+ tensor<string, []> v_3_pad_type_0 = const()[name = tensor<string, []>("v_3_pad_type_0"), val = tensor<string, []>("valid")];
71
+ tensor<int32, [1]> v_3_strides_0 = const()[name = tensor<string, []>("v_3_strides_0"), val = tensor<int32, [1]>([1])];
72
+ tensor<int32, [2]> v_3_pad_0 = const()[name = tensor<string, []>("v_3_pad_0"), val = tensor<int32, [2]>([0, 0])];
73
+ tensor<int32, [1]> v_3_dilations_0 = const()[name = tensor<string, []>("v_3_dilations_0"), val = tensor<int32, [1]>([1])];
74
+ tensor<int32, []> v_3_groups_0 = const()[name = tensor<string, []>("v_3_groups_0"), val = tensor<int32, []>(1)];
75
+ tensor<fp16, [128, 64, 1]> encoder_model_4_block_3_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_model_4_block_3_conv_weight_to_fp16"), val = tensor<fp16, [128, 64, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198912)))];
76
+ tensor<fp16, [128]> encoder_model_4_block_3_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_model_4_block_3_conv_bias_to_fp16"), val = tensor<fp16, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215360)))];
77
+ tensor<fp16, [1, 128, 60000]> v_3_cast_fp16 = conv(bias = encoder_model_4_block_3_conv_bias_to_fp16, dilations = v_3_dilations_0, groups = v_3_groups_0, pad = v_3_pad_0, pad_type = v_3_pad_type_0, strides = v_3_strides_0, weight = encoder_model_4_block_3_conv_weight_to_fp16, x = input_27_cast_fp16)[name = tensor<string, []>("v_3_cast_fp16")];
78
+ tensor<fp16, [1, 128, 60000]> input_29_cast_fp16 = add(x = input_19_cast_fp16, y = v_3_cast_fp16)[name = tensor<string, []>("input_29_cast_fp16")];
79
+ tensor<fp16, [1, 128, 60000]> input_31_cast_fp16 = elu(alpha = var_8, x = input_29_cast_fp16)[name = tensor<string, []>("input_31_cast_fp16")];
80
+ tensor<int32, [6]> input_33_pad_0 = const()[name = tensor<string, []>("input_33_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 5, 0])];
81
+ tensor<string, []> input_33_mode_0 = const()[name = tensor<string, []>("input_33_mode_0"), val = tensor<string, []>("constant")];
82
+ tensor<fp16, []> const_4_to_fp16 = const()[name = tensor<string, []>("const_4_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
83
+ tensor<fp16, [1, 128, 60005]> input_33_cast_fp16 = pad(constant_val = const_4_to_fp16, mode = input_33_mode_0, pad = input_33_pad_0, x = input_31_cast_fp16)[name = tensor<string, []>("input_33_cast_fp16")];
84
+ tensor<string, []> input_35_pad_type_0 = const()[name = tensor<string, []>("input_35_pad_type_0"), val = tensor<string, []>("valid")];
85
+ tensor<int32, [1]> input_35_strides_0 = const()[name = tensor<string, []>("input_35_strides_0"), val = tensor<int32, [1]>([5])];
86
+ tensor<int32, [2]> input_35_pad_0 = const()[name = tensor<string, []>("input_35_pad_0"), val = tensor<int32, [2]>([0, 0])];
87
+ tensor<int32, [1]> input_35_dilations_0 = const()[name = tensor<string, []>("input_35_dilations_0"), val = tensor<int32, [1]>([1])];
88
+ tensor<int32, []> input_35_groups_0 = const()[name = tensor<string, []>("input_35_groups_0"), val = tensor<int32, []>(1)];
89
+ tensor<fp16, [256, 128, 10]> encoder_model_6_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_model_6_conv_weight_to_fp16"), val = tensor<fp16, [256, 128, 10]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215680)))];
90
+ tensor<fp16, [256]> encoder_model_6_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_model_6_conv_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(871104)))];
91
+ tensor<fp16, [1, 256, 12000]> input_35_cast_fp16 = conv(bias = encoder_model_6_conv_bias_to_fp16, dilations = input_35_dilations_0, groups = input_35_groups_0, pad = input_35_pad_0, pad_type = input_35_pad_type_0, strides = input_35_strides_0, weight = encoder_model_6_conv_weight_to_fp16, x = input_33_cast_fp16)[name = tensor<string, []>("input_35_cast_fp16")];
92
+ tensor<fp16, [1, 256, 12000]> input_37_cast_fp16 = elu(alpha = var_8, x = input_35_cast_fp16)[name = tensor<string, []>("input_37_cast_fp16")];
93
+ tensor<int32, [6]> input_39_pad_0 = const()[name = tensor<string, []>("input_39_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
94
+ tensor<string, []> input_39_mode_0 = const()[name = tensor<string, []>("input_39_mode_0"), val = tensor<string, []>("constant")];
95
+ tensor<fp16, []> const_5_to_fp16 = const()[name = tensor<string, []>("const_5_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
96
+ tensor<fp16, [1, 256, 12002]> input_39_cast_fp16 = pad(constant_val = const_5_to_fp16, mode = input_39_mode_0, pad = input_39_pad_0, x = input_37_cast_fp16)[name = tensor<string, []>("input_39_cast_fp16")];
97
+ tensor<string, []> input_41_pad_type_0 = const()[name = tensor<string, []>("input_41_pad_type_0"), val = tensor<string, []>("valid")];
98
+ tensor<int32, [1]> input_41_strides_0 = const()[name = tensor<string, []>("input_41_strides_0"), val = tensor<int32, [1]>([1])];
99
+ tensor<int32, [2]> input_41_pad_0 = const()[name = tensor<string, []>("input_41_pad_0"), val = tensor<int32, [2]>([0, 0])];
100
+ tensor<int32, [1]> input_41_dilations_0 = const()[name = tensor<string, []>("input_41_dilations_0"), val = tensor<int32, [1]>([1])];
101
+ tensor<int32, []> input_41_groups_0 = const()[name = tensor<string, []>("input_41_groups_0"), val = tensor<int32, []>(1)];
102
+ tensor<fp16, [128, 256, 3]> encoder_model_7_block_1_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_model_7_block_1_conv_weight_to_fp16"), val = tensor<fp16, [128, 256, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(871680)))];
103
+ tensor<fp16, [128]> encoder_model_7_block_1_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_model_7_block_1_conv_bias_to_fp16"), val = tensor<fp16, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1068352)))];
104
+ tensor<fp16, [1, 128, 12000]> input_41_cast_fp16 = conv(bias = encoder_model_7_block_1_conv_bias_to_fp16, dilations = input_41_dilations_0, groups = input_41_groups_0, pad = input_41_pad_0, pad_type = input_41_pad_type_0, strides = input_41_strides_0, weight = encoder_model_7_block_1_conv_weight_to_fp16, x = input_39_cast_fp16)[name = tensor<string, []>("input_41_cast_fp16")];
105
+ tensor<fp16, [1, 128, 12000]> input_43_cast_fp16 = elu(alpha = var_8, x = input_41_cast_fp16)[name = tensor<string, []>("input_43_cast_fp16")];
106
+ tensor<string, []> v_5_pad_type_0 = const()[name = tensor<string, []>("v_5_pad_type_0"), val = tensor<string, []>("valid")];
107
+ tensor<int32, [1]> v_5_strides_0 = const()[name = tensor<string, []>("v_5_strides_0"), val = tensor<int32, [1]>([1])];
108
+ tensor<int32, [2]> v_5_pad_0 = const()[name = tensor<string, []>("v_5_pad_0"), val = tensor<int32, [2]>([0, 0])];
109
+ tensor<int32, [1]> v_5_dilations_0 = const()[name = tensor<string, []>("v_5_dilations_0"), val = tensor<int32, [1]>([1])];
110
+ tensor<int32, []> v_5_groups_0 = const()[name = tensor<string, []>("v_5_groups_0"), val = tensor<int32, []>(1)];
111
+ tensor<fp16, [256, 128, 1]> encoder_model_7_block_3_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_model_7_block_3_conv_weight_to_fp16"), val = tensor<fp16, [256, 128, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1068672)))];
112
+ tensor<fp16, [256]> encoder_model_7_block_3_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_model_7_block_3_conv_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1134272)))];
113
+ tensor<fp16, [1, 256, 12000]> v_5_cast_fp16 = conv(bias = encoder_model_7_block_3_conv_bias_to_fp16, dilations = v_5_dilations_0, groups = v_5_groups_0, pad = v_5_pad_0, pad_type = v_5_pad_type_0, strides = v_5_strides_0, weight = encoder_model_7_block_3_conv_weight_to_fp16, x = input_43_cast_fp16)[name = tensor<string, []>("v_5_cast_fp16")];
114
+ tensor<fp16, [1, 256, 12000]> input_45_cast_fp16 = add(x = input_35_cast_fp16, y = v_5_cast_fp16)[name = tensor<string, []>("input_45_cast_fp16")];
115
+ tensor<fp16, [1, 256, 12000]> input_47_cast_fp16 = elu(alpha = var_8, x = input_45_cast_fp16)[name = tensor<string, []>("input_47_cast_fp16")];
116
+ tensor<int32, [6]> input_49_pad_0 = const()[name = tensor<string, []>("input_49_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 6, 0])];
117
+ tensor<string, []> input_49_mode_0 = const()[name = tensor<string, []>("input_49_mode_0"), val = tensor<string, []>("constant")];
118
+ tensor<fp16, []> const_6_to_fp16 = const()[name = tensor<string, []>("const_6_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
119
+ tensor<fp16, [1, 256, 12006]> input_49_cast_fp16 = pad(constant_val = const_6_to_fp16, mode = input_49_mode_0, pad = input_49_pad_0, x = input_47_cast_fp16)[name = tensor<string, []>("input_49_cast_fp16")];
120
+ tensor<string, []> input_51_pad_type_0 = const()[name = tensor<string, []>("input_51_pad_type_0"), val = tensor<string, []>("valid")];
121
+ tensor<int32, [1]> input_51_strides_0 = const()[name = tensor<string, []>("input_51_strides_0"), val = tensor<int32, [1]>([6])];
122
+ tensor<int32, [2]> input_51_pad_0 = const()[name = tensor<string, []>("input_51_pad_0"), val = tensor<int32, [2]>([0, 0])];
123
+ tensor<int32, [1]> input_51_dilations_0 = const()[name = tensor<string, []>("input_51_dilations_0"), val = tensor<int32, [1]>([1])];
124
+ tensor<int32, []> input_51_groups_0 = const()[name = tensor<string, []>("input_51_groups_0"), val = tensor<int32, []>(1)];
125
+ tensor<fp16, [512, 256, 12]> encoder_model_9_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_model_9_conv_weight_to_fp16"), val = tensor<fp16, [512, 256, 12]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1134848)))];
126
+ tensor<fp16, [512]> encoder_model_9_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_model_9_conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4280640)))];
127
+ tensor<fp16, [1, 512, 2000]> input_51_cast_fp16 = conv(bias = encoder_model_9_conv_bias_to_fp16, dilations = input_51_dilations_0, groups = input_51_groups_0, pad = input_51_pad_0, pad_type = input_51_pad_type_0, strides = input_51_strides_0, weight = encoder_model_9_conv_weight_to_fp16, x = input_49_cast_fp16)[name = tensor<string, []>("input_51_cast_fp16")];
128
+ tensor<fp16, [1, 512, 2000]> input_53_cast_fp16 = elu(alpha = var_8, x = input_51_cast_fp16)[name = tensor<string, []>("input_53_cast_fp16")];
129
+ tensor<int32, [6]> input_55_pad_0 = const()[name = tensor<string, []>("input_55_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
130
+ tensor<string, []> input_55_mode_0 = const()[name = tensor<string, []>("input_55_mode_0"), val = tensor<string, []>("constant")];
131
+ tensor<fp16, []> const_7_to_fp16 = const()[name = tensor<string, []>("const_7_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
132
+ tensor<fp16, [1, 512, 2002]> input_55_cast_fp16 = pad(constant_val = const_7_to_fp16, mode = input_55_mode_0, pad = input_55_pad_0, x = input_53_cast_fp16)[name = tensor<string, []>("input_55_cast_fp16")];
133
+ tensor<string, []> x_1_pad_type_0 = const()[name = tensor<string, []>("x_1_pad_type_0"), val = tensor<string, []>("valid")];
134
+ tensor<int32, [1]> x_1_strides_0 = const()[name = tensor<string, []>("x_1_strides_0"), val = tensor<int32, [1]>([1])];
135
+ tensor<int32, [2]> x_1_pad_0 = const()[name = tensor<string, []>("x_1_pad_0"), val = tensor<int32, [2]>([0, 0])];
136
+ tensor<int32, [1]> x_1_dilations_0 = const()[name = tensor<string, []>("x_1_dilations_0"), val = tensor<int32, [1]>([1])];
137
+ tensor<int32, []> x_1_groups_0 = const()[name = tensor<string, []>("x_1_groups_0"), val = tensor<int32, []>(1)];
138
+ tensor<fp16, [512, 512, 3]> encoder_model_11_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_model_11_conv_weight_to_fp16"), val = tensor<fp16, [512, 512, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4281728)))];
139
+ tensor<fp16, [512]> encoder_model_11_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_model_11_conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5854656)))];
140
+ tensor<fp16, [1, 512, 2000]> x_1_cast_fp16 = conv(bias = encoder_model_11_conv_bias_to_fp16, dilations = x_1_dilations_0, groups = x_1_groups_0, pad = x_1_pad_0, pad_type = x_1_pad_type_0, strides = x_1_strides_0, weight = encoder_model_11_conv_weight_to_fp16, x = input_55_cast_fp16)[name = tensor<string, []>("x_1_cast_fp16")];
141
+ tensor<int32, [3]> input_57_perm_0 = const()[name = tensor<string, []>("input_57_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
142
+ tensor<int32, [1]> query_1_axes_0 = const()[name = tensor<string, []>("query_1_axes_0"), val = tensor<int32, [1]>([-1])];
143
+ tensor<fp16, [512]> encoder_transformer_transformer_layers_0_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_transformer_transformer_layers_0_norm1_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5855744)))];
144
+ tensor<fp16, [512]> encoder_transformer_transformer_layers_0_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_transformer_transformer_layers_0_norm1_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5856832)))];
145
+ tensor<fp16, []> var_178_to_fp16 = const()[name = tensor<string, []>("op_178_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
146
+ tensor<fp16, [1, 2000, 512]> input_57_cast_fp16 = transpose(perm = input_57_perm_0, x = x_1_cast_fp16)[name = tensor<string, []>("transpose_22")];
147
+ tensor<fp16, [1, 2000, 512]> query_1_cast_fp16 = layer_norm(axes = query_1_axes_0, beta = encoder_transformer_transformer_layers_0_norm1_bias_to_fp16, epsilon = var_178_to_fp16, gamma = encoder_transformer_transformer_layers_0_norm1_weight_to_fp16, x = input_57_cast_fp16)[name = tensor<string, []>("query_1_cast_fp16")];
148
+ tensor<fp16, [1536, 512]> encoder_transformer_transformer_layers_0_self_attn_in_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_transformer_transformer_layers_0_self_attn_in_proj_weight_to_fp16"), val = tensor<fp16, [1536, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5857920)))];
149
+ tensor<fp16, [1536]> linear_0_bias_0_to_fp16 = const()[name = tensor<string, []>("linear_0_bias_0_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7430848)))];
150
+ tensor<fp16, [1, 2000, 1536]> linear_0_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = encoder_transformer_transformer_layers_0_self_attn_in_proj_weight_to_fp16, x = query_1_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")];
151
+ tensor<int32, [5]> var_203 = const()[name = tensor<string, []>("op_203"), val = tensor<int32, [5]>([1, 2000, 3, 8, 64])];
152
+ tensor<fp16, [1, 2000, 3, 8, 64]> packed_1_cast_fp16 = reshape(shape = var_203, x = linear_0_cast_fp16)[name = tensor<string, []>("packed_1_cast_fp16")];
153
+ tensor<int32, [3]> var_205_split_sizes_0 = const()[name = tensor<string, []>("op_205_split_sizes_0"), val = tensor<int32, [3]>([1, 1, 1])];
154
+ tensor<int32, []> var_205_axis_0 = const()[name = tensor<string, []>("op_205_axis_0"), val = tensor<int32, []>(2)];
155
+ tensor<fp16, [1, 2000, 1, 8, 64]> var_205_cast_fp16_0, tensor<fp16, [1, 2000, 1, 8, 64]> var_205_cast_fp16_1, tensor<fp16, [1, 2000, 1, 8, 64]> var_205_cast_fp16_2 = split(axis = var_205_axis_0, split_sizes = var_205_split_sizes_0, x = packed_1_cast_fp16)[name = tensor<string, []>("op_205_cast_fp16")];
156
+ tensor<int32, [1]> squeeze_0_axes_0 = const()[name = tensor<string, []>("squeeze_0_axes_0"), val = tensor<int32, [1]>([2])];
157
+ tensor<fp16, [1, 2000, 8, 64]> squeeze_0_cast_fp16 = squeeze(axes = squeeze_0_axes_0, x = var_205_cast_fp16_0)[name = tensor<string, []>("squeeze_0_cast_fp16")];
158
+ tensor<int32, [1]> squeeze_1_axes_0 = const()[name = tensor<string, []>("squeeze_1_axes_0"), val = tensor<int32, [1]>([2])];
159
+ tensor<fp16, [1, 2000, 8, 64]> squeeze_1_cast_fp16 = squeeze(axes = squeeze_1_axes_0, x = var_205_cast_fp16_1)[name = tensor<string, []>("squeeze_1_cast_fp16")];
160
+ tensor<int32, [1]> squeeze_2_axes_0 = const()[name = tensor<string, []>("squeeze_2_axes_0"), val = tensor<int32, [1]>([2])];
161
+ tensor<fp16, [1, 2000, 8, 64]> squeeze_2_cast_fp16 = squeeze(axes = squeeze_2_axes_0, x = var_205_cast_fp16_2)[name = tensor<string, []>("squeeze_2_cast_fp16")];
162
+ tensor<int32, [5]> var_218 = const()[name = tensor<string, []>("op_218"), val = tensor<int32, [5]>([1, 2000, 8, 32, 2])];
163
+ tensor<fp16, [1, 2000, 8, 32, 2]> q_3_cast_fp16 = reshape(shape = var_218, x = squeeze_0_cast_fp16)[name = tensor<string, []>("q_3_cast_fp16")];
164
+ tensor<int32, [5]> var_220 = const()[name = tensor<string, []>("op_220"), val = tensor<int32, [5]>([1, 2000, 8, 32, 2])];
165
+ tensor<fp16, [1, 2000, 8, 32, 2]> k_3_cast_fp16 = reshape(shape = var_220, x = squeeze_1_cast_fp16)[name = tensor<string, []>("k_3_cast_fp16")];
166
+ tensor<int32, [5]> var_222_begin_0 = const()[name = tensor<string, []>("op_222_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
167
+ tensor<int32, [5]> var_222_end_0 = const()[name = tensor<string, []>("op_222_end_0"), val = tensor<int32, [5]>([1, 2000, 8, 32, 1])];
168
+ tensor<bool, [5]> var_222_end_mask_0 = const()[name = tensor<string, []>("op_222_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
169
+ tensor<bool, [5]> var_222_squeeze_mask_0 = const()[name = tensor<string, []>("op_222_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
170
+ tensor<fp16, [1, 2000, 8, 32]> var_222_cast_fp16 = slice_by_index(begin = var_222_begin_0, end = var_222_end_0, end_mask = var_222_end_mask_0, squeeze_mask = var_222_squeeze_mask_0, x = q_3_cast_fp16)[name = tensor<string, []>("op_222_cast_fp16")];
171
+ tensor<int32, [5]> var_224_begin_0 = const()[name = tensor<string, []>("op_224_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 1])];
172
+ tensor<int32, [5]> var_224_end_0 = const()[name = tensor<string, []>("op_224_end_0"), val = tensor<int32, [5]>([1, 2000, 8, 32, 2])];
173
+ tensor<bool, [5]> var_224_end_mask_0 = const()[name = tensor<string, []>("op_224_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
174
+ tensor<bool, [5]> var_224_squeeze_mask_0 = const()[name = tensor<string, []>("op_224_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
175
+ tensor<fp16, [1, 2000, 8, 32]> var_224_cast_fp16 = slice_by_index(begin = var_224_begin_0, end = var_224_end_0, end_mask = var_224_end_mask_0, squeeze_mask = var_224_squeeze_mask_0, x = q_3_cast_fp16)[name = tensor<string, []>("op_224_cast_fp16")];
176
+ tensor<int32, [5]> var_226_begin_0 = const()[name = tensor<string, []>("op_226_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
177
+ tensor<int32, [5]> var_226_end_0 = const()[name = tensor<string, []>("op_226_end_0"), val = tensor<int32, [5]>([1, 2000, 8, 32, 1])];
178
+ tensor<bool, [5]> var_226_end_mask_0 = const()[name = tensor<string, []>("op_226_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
179
+ tensor<bool, [5]> var_226_squeeze_mask_0 = const()[name = tensor<string, []>("op_226_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
180
+ tensor<fp16, [1, 2000, 8, 32]> var_226_cast_fp16 = slice_by_index(begin = var_226_begin_0, end = var_226_end_0, end_mask = var_226_end_mask_0, squeeze_mask = var_226_squeeze_mask_0, x = k_3_cast_fp16)[name = tensor<string, []>("op_226_cast_fp16")];
181
+ tensor<int32, [5]> var_228_begin_0 = const()[name = tensor<string, []>("op_228_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 1])];
182
+ tensor<int32, [5]> var_228_end_0 = const()[name = tensor<string, []>("op_228_end_0"), val = tensor<int32, [5]>([1, 2000, 8, 32, 2])];
183
+ tensor<bool, [5]> var_228_end_mask_0 = const()[name = tensor<string, []>("op_228_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
184
+ tensor<bool, [5]> var_228_squeeze_mask_0 = const()[name = tensor<string, []>("op_228_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
185
+ tensor<fp16, [1, 2000, 8, 32]> var_228_cast_fp16 = slice_by_index(begin = var_228_begin_0, end = var_228_end_0, end_mask = var_228_end_mask_0, squeeze_mask = var_228_squeeze_mask_0, x = k_3_cast_fp16)[name = tensor<string, []>("op_228_cast_fp16")];
186
+ tensor<fp16, [2000, 1, 32]> rotr_1_to_fp16 = const()[name = tensor<string, []>("rotr_1_to_fp16"), val = tensor<fp16, [2000, 1, 32]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7433984)))];
187
+ tensor<fp16, [1, 2000, 8, 32]> var_234_cast_fp16 = mul(x = var_222_cast_fp16, y = rotr_1_to_fp16)[name = tensor<string, []>("op_234_cast_fp16")];
188
+ tensor<fp16, [2000, 1, 32]> roti_1_to_fp16 = const()[name = tensor<string, []>("roti_1_to_fp16"), val = tensor<fp16, [2000, 1, 32]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7562048)))];
189
+ tensor<fp16, [1, 2000, 8, 32]> var_235_cast_fp16 = mul(x = var_224_cast_fp16, y = roti_1_to_fp16)[name = tensor<string, []>("op_235_cast_fp16")];
190
+ tensor<fp16, [1, 2000, 8, 32]> qor_1_cast_fp16 = sub(x = var_234_cast_fp16, y = var_235_cast_fp16)[name = tensor<string, []>("qor_1_cast_fp16")];
191
+ tensor<fp16, [1, 2000, 8, 32]> var_237_cast_fp16 = mul(x = var_222_cast_fp16, y = roti_1_to_fp16)[name = tensor<string, []>("op_237_cast_fp16")];
192
+ tensor<fp16, [1, 2000, 8, 32]> var_238_cast_fp16 = mul(x = var_224_cast_fp16, y = rotr_1_to_fp16)[name = tensor<string, []>("op_238_cast_fp16")];
193
+ tensor<fp16, [1, 2000, 8, 32]> qoi_1_cast_fp16 = add(x = var_237_cast_fp16, y = var_238_cast_fp16)[name = tensor<string, []>("qoi_1_cast_fp16")];
194
+ tensor<fp16, [1, 2000, 8, 32]> var_240_cast_fp16 = mul(x = var_226_cast_fp16, y = rotr_1_to_fp16)[name = tensor<string, []>("op_240_cast_fp16")];
195
+ tensor<fp16, [1, 2000, 8, 32]> var_241_cast_fp16 = mul(x = var_228_cast_fp16, y = roti_1_to_fp16)[name = tensor<string, []>("op_241_cast_fp16")];
196
+ tensor<fp16, [1, 2000, 8, 32]> kor_1_cast_fp16 = sub(x = var_240_cast_fp16, y = var_241_cast_fp16)[name = tensor<string, []>("kor_1_cast_fp16")];
197
+ tensor<fp16, [1, 2000, 8, 32]> var_243_cast_fp16 = mul(x = var_226_cast_fp16, y = roti_1_to_fp16)[name = tensor<string, []>("op_243_cast_fp16")];
198
+ tensor<fp16, [1, 2000, 8, 32]> var_244_cast_fp16 = mul(x = var_228_cast_fp16, y = rotr_1_to_fp16)[name = tensor<string, []>("op_244_cast_fp16")];
199
+ tensor<fp16, [1, 2000, 8, 32]> koi_1_cast_fp16 = add(x = var_243_cast_fp16, y = var_244_cast_fp16)[name = tensor<string, []>("koi_1_cast_fp16")];
200
+ tensor<int32, []> qo_1_axis_0 = const()[name = tensor<string, []>("qo_1_axis_0"), val = tensor<int32, []>(-1)];
201
+ tensor<fp16, [1, 2000, 8, 32, 2]> qo_1_cast_fp16 = stack(axis = qo_1_axis_0, values = (qor_1_cast_fp16, qoi_1_cast_fp16))[name = tensor<string, []>("qo_1_cast_fp16")];
202
+ tensor<int32, []> ko_1_axis_0 = const()[name = tensor<string, []>("ko_1_axis_0"), val = tensor<int32, []>(-1)];
203
+ tensor<fp16, [1, 2000, 8, 32, 2]> ko_1_cast_fp16 = stack(axis = ko_1_axis_0, values = (kor_1_cast_fp16, koi_1_cast_fp16))[name = tensor<string, []>("ko_1_cast_fp16")];
204
+ tensor<int32, [4]> var_254 = const()[name = tensor<string, []>("op_254"), val = tensor<int32, [4]>([1, 2000, 8, 64])];
205
+ tensor<fp16, [1, 2000, 8, 64]> q_5_cast_fp16 = reshape(shape = var_254, x = qo_1_cast_fp16)[name = tensor<string, []>("q_5_cast_fp16")];
206
+ tensor<int32, [4]> var_256 = const()[name = tensor<string, []>("op_256"), val = tensor<int32, [4]>([1, 2000, 8, 64])];
207
+ tensor<fp16, [1, 2000, 8, 64]> k_5_cast_fp16 = reshape(shape = var_256, x = ko_1_cast_fp16)[name = tensor<string, []>("k_5_cast_fp16")];
208
+ tensor<int32, [4]> v_9_perm_0 = const()[name = tensor<string, []>("v_9_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
209
+ tensor<fp16, []> mul_1_y_0_to_fp16 = const()[name = tensor<string, []>("mul_1_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
210
+ tensor<fp16, [1, 2000, 8, 64]> mul_1_cast_fp16 = mul(x = q_5_cast_fp16, y = mul_1_y_0_to_fp16)[name = tensor<string, []>("mul_1_cast_fp16")];
211
+ tensor<bool, []> matmul_0_transpose_y_0 = const()[name = tensor<string, []>("matmul_0_transpose_y_0"), val = tensor<bool, []>(true)];
212
+ tensor<bool, []> matmul_0_transpose_x_0 = const()[name = tensor<string, []>("matmul_0_transpose_x_0"), val = tensor<bool, []>(false)];
213
+ tensor<int32, [4]> transpose_8_perm_0 = const()[name = tensor<string, []>("transpose_8_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
214
+ tensor<int32, [4]> transpose_9_perm_0 = const()[name = tensor<string, []>("transpose_9_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
215
+ tensor<fp16, [1, 8, 2000, 64]> transpose_9 = transpose(perm = transpose_9_perm_0, x = k_5_cast_fp16)[name = tensor<string, []>("transpose_19")];
216
+ tensor<fp16, [1, 8, 2000, 64]> transpose_8 = transpose(perm = transpose_8_perm_0, x = mul_1_cast_fp16)[name = tensor<string, []>("transpose_20")];
217
+ tensor<fp16, [1, 8, 2000, 2000]> matmul_0_cast_fp16 = matmul(transpose_x = matmul_0_transpose_x_0, transpose_y = matmul_0_transpose_y_0, x = transpose_8, y = transpose_9)[name = tensor<string, []>("matmul_0_cast_fp16")];
218
+ tensor<fp16, [1, 1, 2000, 2000]> mul_0_to_fp16 = const()[name = tensor<string, []>("mul_0_to_fp16"), val = tensor<fp16, [1, 1, 2000, 2000]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7690112)))];
219
+ tensor<fp16, [1, 8, 2000, 2000]> add_0_cast_fp16 = add(x = matmul_0_cast_fp16, y = mul_0_to_fp16)[name = tensor<string, []>("add_0_cast_fp16")];
220
+ tensor<int32, []> softmax_0_axis_0 = const()[name = tensor<string, []>("softmax_0_axis_0"), val = tensor<int32, []>(-1)];
221
+ tensor<fp16, [1, 8, 2000, 2000]> softmax_0_cast_fp16 = softmax(axis = softmax_0_axis_0, x = add_0_cast_fp16)[name = tensor<string, []>("softmax_0_cast_fp16")];
222
+ tensor<bool, []> x_3_transpose_x_0 = const()[name = tensor<string, []>("x_3_transpose_x_0"), val = tensor<bool, []>(false)];
223
+ tensor<bool, []> x_3_transpose_y_0 = const()[name = tensor<string, []>("x_3_transpose_y_0"), val = tensor<bool, []>(false)];
224
+ tensor<fp16, [1, 8, 2000, 64]> v_9_cast_fp16 = transpose(perm = v_9_perm_0, x = squeeze_2_cast_fp16)[name = tensor<string, []>("transpose_21")];
225
+ tensor<fp16, [1, 8, 2000, 64]> x_3_cast_fp16 = matmul(transpose_x = x_3_transpose_x_0, transpose_y = x_3_transpose_y_0, x = softmax_0_cast_fp16, y = v_9_cast_fp16)[name = tensor<string, []>("x_3_cast_fp16")];
226
+ tensor<int32, [4]> var_276_perm_0 = const()[name = tensor<string, []>("op_276_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
227
+ tensor<int32, [3]> var_277 = const()[name = tensor<string, []>("op_277"), val = tensor<int32, [3]>([1, 2000, 512])];
228
+ tensor<fp16, [1, 2000, 8, 64]> var_276_cast_fp16 = transpose(perm = var_276_perm_0, x = x_3_cast_fp16)[name = tensor<string, []>("transpose_18")];
229
+ tensor<fp16, [1, 2000, 512]> input_59_cast_fp16 = reshape(shape = var_277, x = var_276_cast_fp16)[name = tensor<string, []>("input_59_cast_fp16")];
230
+ tensor<fp16, [512, 512]> encoder_transformer_transformer_layers_0_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_transformer_transformer_layers_0_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15690176)))];
231
+ tensor<fp16, [512]> linear_1_bias_0_to_fp16 = const()[name = tensor<string, []>("linear_1_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16214528)))];
232
+ tensor<fp16, [1, 2000, 512]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_transformer_transformer_layers_0_self_attn_out_proj_weight_to_fp16, x = input_59_cast_fp16)[name = tensor<string, []>("linear_1_cast_fp16")];
233
+ tensor<fp16, [512]> encoder_transformer_transformer_layers_0_layer_scale_1_scale_to_fp16 = const()[name = tensor<string, []>("encoder_transformer_transformer_layers_0_layer_scale_1_scale_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16215616)))];
234
+ tensor<fp16, [1, 2000, 512]> var_283_cast_fp16 = mul(x = encoder_transformer_transformer_layers_0_layer_scale_1_scale_to_fp16, y = linear_1_cast_fp16)[name = tensor<string, []>("op_283_cast_fp16")];
235
+ tensor<fp16, [1, 2000, 512]> input_61_cast_fp16 = add(x = input_57_cast_fp16, y = var_283_cast_fp16)[name = tensor<string, []>("input_61_cast_fp16")];
236
+ tensor<int32, [1]> input_63_axes_0 = const()[name = tensor<string, []>("input_63_axes_0"), val = tensor<int32, [1]>([-1])];
237
+ tensor<fp16, [512]> encoder_transformer_transformer_layers_0_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_transformer_transformer_layers_0_norm2_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16216704)))];
238
+ tensor<fp16, [512]> encoder_transformer_transformer_layers_0_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_transformer_transformer_layers_0_norm2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16217792)))];
239
+ tensor<fp16, [1, 2000, 512]> input_63_cast_fp16 = layer_norm(axes = input_63_axes_0, beta = encoder_transformer_transformer_layers_0_norm2_bias_to_fp16, epsilon = var_178_to_fp16, gamma = encoder_transformer_transformer_layers_0_norm2_weight_to_fp16, x = input_61_cast_fp16)[name = tensor<string, []>("input_63_cast_fp16")];
240
+ tensor<fp16, [2048, 512]> encoder_transformer_transformer_layers_0_linear1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_transformer_transformer_layers_0_linear1_weight_to_fp16"), val = tensor<fp16, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16218880)))];
241
+ tensor<fp16, [2048]> linear_2_bias_0_to_fp16 = const()[name = tensor<string, []>("linear_2_bias_0_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18316096)))];
242
+ tensor<fp16, [1, 2000, 2048]> linear_2_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = encoder_transformer_transformer_layers_0_linear1_weight_to_fp16, x = input_63_cast_fp16)[name = tensor<string, []>("linear_2_cast_fp16")];
243
+ tensor<string, []> input_65_mode_0 = const()[name = tensor<string, []>("input_65_mode_0"), val = tensor<string, []>("EXACT")];
244
+ tensor<fp16, [1, 2000, 2048]> input_65_cast_fp16 = gelu(mode = input_65_mode_0, x = linear_2_cast_fp16)[name = tensor<string, []>("input_65_cast_fp16")];
245
+ tensor<fp16, [512, 2048]> encoder_transformer_transformer_layers_0_linear2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_transformer_transformer_layers_0_linear2_weight_to_fp16"), val = tensor<fp16, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18320256)))];
246
+ tensor<fp16, [1, 2000, 512]> linear_3_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_transformer_transformer_layers_0_linear2_weight_to_fp16, x = input_65_cast_fp16)[name = tensor<string, []>("linear_3_cast_fp16")];
247
+ tensor<fp16, [512]> encoder_transformer_transformer_layers_0_layer_scale_2_scale_to_fp16 = const()[name = tensor<string, []>("encoder_transformer_transformer_layers_0_layer_scale_2_scale_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20417472)))];
248
+ tensor<fp16, [1, 2000, 512]> var_296_cast_fp16 = mul(x = encoder_transformer_transformer_layers_0_layer_scale_2_scale_to_fp16, y = linear_3_cast_fp16)[name = tensor<string, []>("op_296_cast_fp16")];
249
+ tensor<fp16, [1, 2000, 512]> input_67_cast_fp16 = add(x = input_61_cast_fp16, y = var_296_cast_fp16)[name = tensor<string, []>("input_67_cast_fp16")];
250
+ tensor<int32, [1]> query_axes_0 = const()[name = tensor<string, []>("query_axes_0"), val = tensor<int32, [1]>([-1])];
251
+ tensor<fp16, [512]> encoder_transformer_transformer_layers_1_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_transformer_transformer_layers_1_norm1_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20418560)))];
252
+ tensor<fp16, [512]> encoder_transformer_transformer_layers_1_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_transformer_transformer_layers_1_norm1_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20419648)))];
253
+ tensor<fp16, [1, 2000, 512]> query_cast_fp16 = layer_norm(axes = query_axes_0, beta = encoder_transformer_transformer_layers_1_norm1_bias_to_fp16, epsilon = var_178_to_fp16, gamma = encoder_transformer_transformer_layers_1_norm1_weight_to_fp16, x = input_67_cast_fp16)[name = tensor<string, []>("query_cast_fp16")];
254
+ tensor<fp16, [1536, 512]> encoder_transformer_transformer_layers_1_self_attn_in_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_transformer_transformer_layers_1_self_attn_in_proj_weight_to_fp16"), val = tensor<fp16, [1536, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20420736)))];
255
+ tensor<fp16, [1, 2000, 1536]> linear_4_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = encoder_transformer_transformer_layers_1_self_attn_in_proj_weight_to_fp16, x = query_cast_fp16)[name = tensor<string, []>("linear_4_cast_fp16")];
256
+ tensor<int32, [5]> var_313 = const()[name = tensor<string, []>("op_313"), val = tensor<int32, [5]>([1, 2000, 3, 8, 64])];
257
+ tensor<fp16, [1, 2000, 3, 8, 64]> packed_cast_fp16 = reshape(shape = var_313, x = linear_4_cast_fp16)[name = tensor<string, []>("packed_cast_fp16")];
258
+ tensor<int32, [3]> var_315_split_sizes_0 = const()[name = tensor<string, []>("op_315_split_sizes_0"), val = tensor<int32, [3]>([1, 1, 1])];
259
+ tensor<int32, []> var_315_axis_0 = const()[name = tensor<string, []>("op_315_axis_0"), val = tensor<int32, []>(2)];
260
+ tensor<fp16, [1, 2000, 1, 8, 64]> var_315_cast_fp16_0, tensor<fp16, [1, 2000, 1, 8, 64]> var_315_cast_fp16_1, tensor<fp16, [1, 2000, 1, 8, 64]> var_315_cast_fp16_2 = split(axis = var_315_axis_0, split_sizes = var_315_split_sizes_0, x = packed_cast_fp16)[name = tensor<string, []>("op_315_cast_fp16")];
261
+ tensor<int32, [1]> squeeze_3_axes_0 = const()[name = tensor<string, []>("squeeze_3_axes_0"), val = tensor<int32, [1]>([2])];
262
+ tensor<fp16, [1, 2000, 8, 64]> squeeze_3_cast_fp16 = squeeze(axes = squeeze_3_axes_0, x = var_315_cast_fp16_0)[name = tensor<string, []>("squeeze_3_cast_fp16")];
263
+ tensor<int32, [1]> squeeze_4_axes_0 = const()[name = tensor<string, []>("squeeze_4_axes_0"), val = tensor<int32, [1]>([2])];
264
+ tensor<fp16, [1, 2000, 8, 64]> squeeze_4_cast_fp16 = squeeze(axes = squeeze_4_axes_0, x = var_315_cast_fp16_1)[name = tensor<string, []>("squeeze_4_cast_fp16")];
265
+ tensor<int32, [1]> squeeze_5_axes_0 = const()[name = tensor<string, []>("squeeze_5_axes_0"), val = tensor<int32, [1]>([2])];
266
+ tensor<fp16, [1, 2000, 8, 64]> squeeze_5_cast_fp16 = squeeze(axes = squeeze_5_axes_0, x = var_315_cast_fp16_2)[name = tensor<string, []>("squeeze_5_cast_fp16")];
267
+ tensor<int32, [5]> var_328 = const()[name = tensor<string, []>("op_328"), val = tensor<int32, [5]>([1, 2000, 8, 32, 2])];
268
+ tensor<fp16, [1, 2000, 8, 32, 2]> q_11_cast_fp16 = reshape(shape = var_328, x = squeeze_3_cast_fp16)[name = tensor<string, []>("q_11_cast_fp16")];
269
+ tensor<int32, [5]> var_330 = const()[name = tensor<string, []>("op_330"), val = tensor<int32, [5]>([1, 2000, 8, 32, 2])];
270
+ tensor<fp16, [1, 2000, 8, 32, 2]> k_11_cast_fp16 = reshape(shape = var_330, x = squeeze_4_cast_fp16)[name = tensor<string, []>("k_11_cast_fp16")];
271
+ tensor<int32, [5]> var_332_begin_0 = const()[name = tensor<string, []>("op_332_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
272
+ tensor<int32, [5]> var_332_end_0 = const()[name = tensor<string, []>("op_332_end_0"), val = tensor<int32, [5]>([1, 2000, 8, 32, 1])];
273
+ tensor<bool, [5]> var_332_end_mask_0 = const()[name = tensor<string, []>("op_332_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
274
+ tensor<bool, [5]> var_332_squeeze_mask_0 = const()[name = tensor<string, []>("op_332_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
275
+ tensor<fp16, [1, 2000, 8, 32]> var_332_cast_fp16 = slice_by_index(begin = var_332_begin_0, end = var_332_end_0, end_mask = var_332_end_mask_0, squeeze_mask = var_332_squeeze_mask_0, x = q_11_cast_fp16)[name = tensor<string, []>("op_332_cast_fp16")];
276
+ tensor<int32, [5]> var_334_begin_0 = const()[name = tensor<string, []>("op_334_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 1])];
277
+ tensor<int32, [5]> var_334_end_0 = const()[name = tensor<string, []>("op_334_end_0"), val = tensor<int32, [5]>([1, 2000, 8, 32, 2])];
278
+ tensor<bool, [5]> var_334_end_mask_0 = const()[name = tensor<string, []>("op_334_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
279
+ tensor<bool, [5]> var_334_squeeze_mask_0 = const()[name = tensor<string, []>("op_334_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
280
+ tensor<fp16, [1, 2000, 8, 32]> var_334_cast_fp16 = slice_by_index(begin = var_334_begin_0, end = var_334_end_0, end_mask = var_334_end_mask_0, squeeze_mask = var_334_squeeze_mask_0, x = q_11_cast_fp16)[name = tensor<string, []>("op_334_cast_fp16")];
281
+ tensor<int32, [5]> var_336_begin_0 = const()[name = tensor<string, []>("op_336_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
282
+ tensor<int32, [5]> var_336_end_0 = const()[name = tensor<string, []>("op_336_end_0"), val = tensor<int32, [5]>([1, 2000, 8, 32, 1])];
283
+ tensor<bool, [5]> var_336_end_mask_0 = const()[name = tensor<string, []>("op_336_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
284
+ tensor<bool, [5]> var_336_squeeze_mask_0 = const()[name = tensor<string, []>("op_336_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
285
+ tensor<fp16, [1, 2000, 8, 32]> var_336_cast_fp16 = slice_by_index(begin = var_336_begin_0, end = var_336_end_0, end_mask = var_336_end_mask_0, squeeze_mask = var_336_squeeze_mask_0, x = k_11_cast_fp16)[name = tensor<string, []>("op_336_cast_fp16")];
286
+ tensor<int32, [5]> var_338_begin_0 = const()[name = tensor<string, []>("op_338_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 1])];
287
+ tensor<int32, [5]> var_338_end_0 = const()[name = tensor<string, []>("op_338_end_0"), val = tensor<int32, [5]>([1, 2000, 8, 32, 2])];
288
+ tensor<bool, [5]> var_338_end_mask_0 = const()[name = tensor<string, []>("op_338_end_mask_0"), val = tensor<bool, [5]>([true, true, true, true, false])];
289
+ tensor<bool, [5]> var_338_squeeze_mask_0 = const()[name = tensor<string, []>("op_338_squeeze_mask_0"), val = tensor<bool, [5]>([false, false, false, false, true])];
290
+ tensor<fp16, [1, 2000, 8, 32]> var_338_cast_fp16 = slice_by_index(begin = var_338_begin_0, end = var_338_end_0, end_mask = var_338_end_mask_0, squeeze_mask = var_338_squeeze_mask_0, x = k_11_cast_fp16)[name = tensor<string, []>("op_338_cast_fp16")];
291
+ tensor<fp16, [1, 2000, 8, 32]> var_344_cast_fp16 = mul(x = var_332_cast_fp16, y = rotr_1_to_fp16)[name = tensor<string, []>("op_344_cast_fp16")];
292
+ tensor<fp16, [1, 2000, 8, 32]> var_345_cast_fp16 = mul(x = var_334_cast_fp16, y = roti_1_to_fp16)[name = tensor<string, []>("op_345_cast_fp16")];
293
+ tensor<fp16, [1, 2000, 8, 32]> qor_5_cast_fp16 = sub(x = var_344_cast_fp16, y = var_345_cast_fp16)[name = tensor<string, []>("qor_5_cast_fp16")];
294
+ tensor<fp16, [1, 2000, 8, 32]> var_347_cast_fp16 = mul(x = var_332_cast_fp16, y = roti_1_to_fp16)[name = tensor<string, []>("op_347_cast_fp16")];
295
+ tensor<fp16, [1, 2000, 8, 32]> var_348_cast_fp16 = mul(x = var_334_cast_fp16, y = rotr_1_to_fp16)[name = tensor<string, []>("op_348_cast_fp16")];
296
+ tensor<fp16, [1, 2000, 8, 32]> qoi_5_cast_fp16 = add(x = var_347_cast_fp16, y = var_348_cast_fp16)[name = tensor<string, []>("qoi_5_cast_fp16")];
297
+ tensor<fp16, [1, 2000, 8, 32]> var_350_cast_fp16 = mul(x = var_336_cast_fp16, y = rotr_1_to_fp16)[name = tensor<string, []>("op_350_cast_fp16")];
298
+ tensor<fp16, [1, 2000, 8, 32]> var_351_cast_fp16 = mul(x = var_338_cast_fp16, y = roti_1_to_fp16)[name = tensor<string, []>("op_351_cast_fp16")];
299
+ tensor<fp16, [1, 2000, 8, 32]> kor_5_cast_fp16 = sub(x = var_350_cast_fp16, y = var_351_cast_fp16)[name = tensor<string, []>("kor_5_cast_fp16")];
300
+ tensor<fp16, [1, 2000, 8, 32]> var_353_cast_fp16 = mul(x = var_336_cast_fp16, y = roti_1_to_fp16)[name = tensor<string, []>("op_353_cast_fp16")];
301
+ tensor<fp16, [1, 2000, 8, 32]> var_354_cast_fp16 = mul(x = var_338_cast_fp16, y = rotr_1_to_fp16)[name = tensor<string, []>("op_354_cast_fp16")];
302
+ tensor<fp16, [1, 2000, 8, 32]> koi_5_cast_fp16 = add(x = var_353_cast_fp16, y = var_354_cast_fp16)[name = tensor<string, []>("koi_5_cast_fp16")];
303
+ tensor<int32, []> qo_axis_0 = const()[name = tensor<string, []>("qo_axis_0"), val = tensor<int32, []>(-1)];
304
+ tensor<fp16, [1, 2000, 8, 32, 2]> qo_cast_fp16 = stack(axis = qo_axis_0, values = (qor_5_cast_fp16, qoi_5_cast_fp16))[name = tensor<string, []>("qo_cast_fp16")];
305
+ tensor<int32, []> ko_axis_0 = const()[name = tensor<string, []>("ko_axis_0"), val = tensor<int32, []>(-1)];
306
+ tensor<fp16, [1, 2000, 8, 32, 2]> ko_cast_fp16 = stack(axis = ko_axis_0, values = (kor_5_cast_fp16, koi_5_cast_fp16))[name = tensor<string, []>("ko_cast_fp16")];
307
+ tensor<int32, [4]> var_364 = const()[name = tensor<string, []>("op_364"), val = tensor<int32, [4]>([1, 2000, 8, 64])];
308
+ tensor<fp16, [1, 2000, 8, 64]> q_13_cast_fp16 = reshape(shape = var_364, x = qo_cast_fp16)[name = tensor<string, []>("q_13_cast_fp16")];
309
+ tensor<int32, [4]> var_366 = const()[name = tensor<string, []>("op_366"), val = tensor<int32, [4]>([1, 2000, 8, 64])];
310
+ tensor<fp16, [1, 2000, 8, 64]> k_13_cast_fp16 = reshape(shape = var_366, x = ko_cast_fp16)[name = tensor<string, []>("k_13_cast_fp16")];
311
+ tensor<int32, [4]> v_perm_0 = const()[name = tensor<string, []>("v_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
312
+ tensor<fp16, []> mul_3_y_0_to_fp16 = const()[name = tensor<string, []>("mul_3_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
313
+ tensor<fp16, [1, 2000, 8, 64]> mul_3_cast_fp16 = mul(x = q_13_cast_fp16, y = mul_3_y_0_to_fp16)[name = tensor<string, []>("mul_3_cast_fp16")];
314
+ tensor<bool, []> matmul_1_transpose_y_0 = const()[name = tensor<string, []>("matmul_1_transpose_y_0"), val = tensor<bool, []>(true)];
315
+ tensor<bool, []> matmul_1_transpose_x_0 = const()[name = tensor<string, []>("matmul_1_transpose_x_0"), val = tensor<bool, []>(false)];
316
+ tensor<int32, [4]> transpose_10_perm_0 = const()[name = tensor<string, []>("transpose_10_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
317
+ tensor<int32, [4]> transpose_11_perm_0 = const()[name = tensor<string, []>("transpose_11_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
318
+ tensor<fp16, [1, 8, 2000, 64]> transpose_11 = transpose(perm = transpose_11_perm_0, x = k_13_cast_fp16)[name = tensor<string, []>("transpose_15")];
319
+ tensor<fp16, [1, 8, 2000, 64]> transpose_10 = transpose(perm = transpose_10_perm_0, x = mul_3_cast_fp16)[name = tensor<string, []>("transpose_16")];
320
+ tensor<fp16, [1, 8, 2000, 2000]> matmul_1_cast_fp16 = matmul(transpose_x = matmul_1_transpose_x_0, transpose_y = matmul_1_transpose_y_0, x = transpose_10, y = transpose_11)[name = tensor<string, []>("matmul_1_cast_fp16")];
321
+ tensor<fp16, [1, 8, 2000, 2000]> add_1_cast_fp16 = add(x = matmul_1_cast_fp16, y = mul_0_to_fp16)[name = tensor<string, []>("add_1_cast_fp16")];
322
+ tensor<int32, []> softmax_1_axis_0 = const()[name = tensor<string, []>("softmax_1_axis_0"), val = tensor<int32, []>(-1)];
323
+ tensor<fp16, [1, 8, 2000, 2000]> softmax_1_cast_fp16 = softmax(axis = softmax_1_axis_0, x = add_1_cast_fp16)[name = tensor<string, []>("softmax_1_cast_fp16")];
324
+ tensor<bool, []> x_9_transpose_x_0 = const()[name = tensor<string, []>("x_9_transpose_x_0"), val = tensor<bool, []>(false)];
325
+ tensor<bool, []> x_9_transpose_y_0 = const()[name = tensor<string, []>("x_9_transpose_y_0"), val = tensor<bool, []>(false)];
326
+ tensor<fp16, [1, 8, 2000, 64]> v_cast_fp16 = transpose(perm = v_perm_0, x = squeeze_5_cast_fp16)[name = tensor<string, []>("transpose_17")];
327
+ tensor<fp16, [1, 8, 2000, 64]> x_9_cast_fp16 = matmul(transpose_x = x_9_transpose_x_0, transpose_y = x_9_transpose_y_0, x = softmax_1_cast_fp16, y = v_cast_fp16)[name = tensor<string, []>("x_9_cast_fp16")];
328
+ tensor<int32, [4]> var_386_perm_0 = const()[name = tensor<string, []>("op_386_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
329
+ tensor<int32, [3]> var_387 = const()[name = tensor<string, []>("op_387"), val = tensor<int32, [3]>([1, 2000, 512])];
330
+ tensor<fp16, [1, 2000, 8, 64]> var_386_cast_fp16 = transpose(perm = var_386_perm_0, x = x_9_cast_fp16)[name = tensor<string, []>("transpose_14")];
331
+ tensor<fp16, [1, 2000, 512]> input_69_cast_fp16 = reshape(shape = var_387, x = var_386_cast_fp16)[name = tensor<string, []>("input_69_cast_fp16")];
332
+ tensor<fp16, [512, 512]> encoder_transformer_transformer_layers_1_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_transformer_transformer_layers_1_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21993664)))];
333
+ tensor<fp16, [1, 2000, 512]> linear_5_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_transformer_transformer_layers_1_self_attn_out_proj_weight_to_fp16, x = input_69_cast_fp16)[name = tensor<string, []>("linear_5_cast_fp16")];
334
+ tensor<fp16, [512]> encoder_transformer_transformer_layers_1_layer_scale_1_scale_to_fp16 = const()[name = tensor<string, []>("encoder_transformer_transformer_layers_1_layer_scale_1_scale_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22518016)))];
335
+ tensor<fp16, [1, 2000, 512]> var_393_cast_fp16 = mul(x = encoder_transformer_transformer_layers_1_layer_scale_1_scale_to_fp16, y = linear_5_cast_fp16)[name = tensor<string, []>("op_393_cast_fp16")];
336
+ tensor<fp16, [1, 2000, 512]> input_71_cast_fp16 = add(x = input_67_cast_fp16, y = var_393_cast_fp16)[name = tensor<string, []>("input_71_cast_fp16")];
337
+ tensor<int32, [1]> input_73_axes_0 = const()[name = tensor<string, []>("input_73_axes_0"), val = tensor<int32, [1]>([-1])];
338
+ tensor<fp16, [512]> encoder_transformer_transformer_layers_1_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_transformer_transformer_layers_1_norm2_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22519104)))];
339
+ tensor<fp16, [512]> encoder_transformer_transformer_layers_1_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_transformer_transformer_layers_1_norm2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22520192)))];
340
+ tensor<fp16, [1, 2000, 512]> input_73_cast_fp16 = layer_norm(axes = input_73_axes_0, beta = encoder_transformer_transformer_layers_1_norm2_bias_to_fp16, epsilon = var_178_to_fp16, gamma = encoder_transformer_transformer_layers_1_norm2_weight_to_fp16, x = input_71_cast_fp16)[name = tensor<string, []>("input_73_cast_fp16")];
341
+ tensor<fp16, [2048, 512]> encoder_transformer_transformer_layers_1_linear1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_transformer_transformer_layers_1_linear1_weight_to_fp16"), val = tensor<fp16, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22521280)))];
342
+ tensor<fp16, [1, 2000, 2048]> linear_6_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = encoder_transformer_transformer_layers_1_linear1_weight_to_fp16, x = input_73_cast_fp16)[name = tensor<string, []>("linear_6_cast_fp16")];
343
+ tensor<string, []> input_75_mode_0 = const()[name = tensor<string, []>("input_75_mode_0"), val = tensor<string, []>("EXACT")];
344
+ tensor<fp16, [1, 2000, 2048]> input_75_cast_fp16 = gelu(mode = input_75_mode_0, x = linear_6_cast_fp16)[name = tensor<string, []>("input_75_cast_fp16")];
345
+ tensor<fp16, [512, 2048]> encoder_transformer_transformer_layers_1_linear2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_transformer_transformer_layers_1_linear2_weight_to_fp16"), val = tensor<fp16, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24618496)))];
346
+ tensor<fp16, [1, 2000, 512]> linear_7_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_transformer_transformer_layers_1_linear2_weight_to_fp16, x = input_75_cast_fp16)[name = tensor<string, []>("linear_7_cast_fp16")];
347
+ tensor<fp16, [512]> encoder_transformer_transformer_layers_1_layer_scale_2_scale_to_fp16 = const()[name = tensor<string, []>("encoder_transformer_transformer_layers_1_layer_scale_2_scale_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26715712)))];
348
+ tensor<fp16, [1, 2000, 512]> var_406_cast_fp16 = mul(x = encoder_transformer_transformer_layers_1_layer_scale_2_scale_to_fp16, y = linear_7_cast_fp16)[name = tensor<string, []>("op_406_cast_fp16")];
349
+ tensor<fp16, [1, 2000, 512]> z_cast_fp16 = add(x = input_71_cast_fp16, y = var_406_cast_fp16)[name = tensor<string, []>("z_cast_fp16")];
350
+ tensor<int32, [3]> x_perm_0 = const()[name = tensor<string, []>("x_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
351
+ tensor<int32, []> var_412 = const()[name = tensor<string, []>("op_412"), val = tensor<int32, []>(-1)];
352
+ tensor<int32, [3]> var_419_begin_0 = const()[name = tensor<string, []>("op_419_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
353
+ tensor<int32, [3]> var_419_end_0 = const()[name = tensor<string, []>("op_419_end_0"), val = tensor<int32, [3]>([1, 512, 1])];
354
+ tensor<bool, [3]> var_419_end_mask_0 = const()[name = tensor<string, []>("op_419_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
355
+ tensor<fp16, [1, 512, 2000]> x_cast_fp16 = transpose(perm = x_perm_0, x = z_cast_fp16)[name = tensor<string, []>("transpose_13")];
356
+ tensor<fp16, [1, 512, 1]> var_419_cast_fp16 = slice_by_index(begin = var_419_begin_0, end = var_419_end_0, end_mask = var_419_end_mask_0, x = x_cast_fp16)[name = tensor<string, []>("op_419_cast_fp16")];
357
+ tensor<int32, [3]> init_reps_0 = const()[name = tensor<string, []>("init_reps_0"), val = tensor<int32, [3]>([1, 1, 16])];
358
+ tensor<fp16, [1, 512, 16]> init_cast_fp16 = tile(reps = init_reps_0, x = var_419_cast_fp16)[name = tensor<string, []>("init_cast_fp16")];
359
+ tensor<bool, []> input_interleave_0 = const()[name = tensor<string, []>("input_interleave_0"), val = tensor<bool, []>(false)];
360
+ tensor<fp16, [1, 512, 2016]> input_cast_fp16 = concat(axis = var_412, interleave = input_interleave_0, values = (init_cast_fp16, x_cast_fp16))[name = tensor<string, []>("input_cast_fp16")];
361
+ tensor<string, []> emb_pad_type_0 = const()[name = tensor<string, []>("emb_pad_type_0"), val = tensor<string, []>("valid")];
362
+ tensor<int32, [1]> emb_strides_0 = const()[name = tensor<string, []>("emb_strides_0"), val = tensor<int32, [1]>([16])];
363
+ tensor<int32, [2]> emb_pad_0 = const()[name = tensor<string, []>("emb_pad_0"), val = tensor<int32, [2]>([0, 0])];
364
+ tensor<int32, [1]> emb_dilations_0 = const()[name = tensor<string, []>("emb_dilations_0"), val = tensor<int32, [1]>([1])];
365
+ tensor<int32, []> emb_groups_0 = const()[name = tensor<string, []>("emb_groups_0"), val = tensor<int32, []>(1)];
366
+ tensor<fp16, [32, 512, 32]> downsample_conv_conv_weight_to_fp16 = const()[name = tensor<string, []>("downsample_conv_conv_weight_to_fp16"), val = tensor<fp16, [32, 512, 32]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26716800)))];
367
+ tensor<fp16, [1, 32, 125]> emb_cast_fp16 = conv(dilations = emb_dilations_0, groups = emb_groups_0, pad = emb_pad_0, pad_type = emb_pad_type_0, strides = emb_strides_0, weight = downsample_conv_conv_weight_to_fp16, x = input_cast_fp16)[name = tensor<string, []>("emb_cast_fp16")];
368
+ tensor<int32, [3]> var_432_perm_0 = const()[name = tensor<string, []>("op_432_perm_0"), val = tensor<int32, [3]>([0, -1, -2])];
369
+ tensor<fp16, [1024, 32]> speaker_proj_weight_to_fp16 = const()[name = tensor<string, []>("speaker_proj_weight_to_fp16"), val = tensor<fp16, [1024, 32]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27765440)))];
370
+ tensor<fp16, [1024]> linear_8_bias_0_to_fp16 = const()[name = tensor<string, []>("linear_8_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27831040)))];
371
+ tensor<fp16, [1, 125, 32]> var_432_cast_fp16 = transpose(perm = var_432_perm_0, x = emb_cast_fp16)[name = tensor<string, []>("transpose_12")];
372
+ tensor<fp16, [1, 125, 1024]> linear_8_cast_fp16 = linear(bias = linear_8_bias_0_to_fp16, weight = speaker_proj_weight_to_fp16, x = var_432_cast_fp16)[name = tensor<string, []>("linear_8_cast_fp16")];
373
+ tensor<string, []> linear_8_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("linear_8_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
374
+ tensor<fp32, [1, 125, 1024]> conditioning = cast(dtype = linear_8_cast_fp16_to_fp32_dtype_0, x = linear_8_cast_fp16)[name = tensor<string, []>("cast_26")];
375
+ } -> (conditioning);
376
+ }
v2.1/italian_24l/mimi_encoderv3.mlmodelc/weights/weight.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1d6856ad52c13647e26777220d45043edf735e69ea8bad2f8f67f54ef2268782
3
+ size 27833152