Hugging Face's logo

lithium0003
/
ggml-coreml-whisper

Model card Files
xet
 Community
lithium0003 commited on
Commit
8a23912
·
1 Parent(s): 1f717f0

20251213

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. ggml-base-encoder.mlmodelc/analytics/coremldata.bin +1 -1
  2. ggml-base-encoder.mlmodelc/coremldata.bin +2 -2
  3. ggml-base-encoder.mlmodelc/metadata.json +22 -20
  4. ggml-base-encoder.mlmodelc/model.mil +0 -0
  5. ggml-base-encoder.mlmodelc/weights/weight.bin +1 -1
  6. ggml-large-v3-turbo-encoder.mlmodelc/model0/analytics/coremldata.bin → ggml-base-q8_0.bin +2 -2
  7. ggml-large-v2-encoder.mlmodelc/metadata.json +0 -66
  8. ggml-large-v2-encoder.mlmodelc/model.mil +0 -0
  9. ggml-large-v2-q8_0.bin +0 -3
  10. ggml-large-v3-encoder.mlmodelc/analytics/coremldata.bin +1 -1
  11. ggml-large-v3-encoder.mlmodelc/coremldata.bin +2 -2
  12. ggml-large-v3-encoder.mlmodelc/metadata.json +22 -20
  13. ggml-large-v3-encoder.mlmodelc/model.mil +0 -0
  14. ggml-large-v3-encoder.mlmodelc/weights/weight.bin +2 -2
  15. ggml-large-v3-turbo-encoder.mlmodelc/analytics/coremldata.bin +2 -2
  16. ggml-large-v3-turbo-encoder.mlmodelc/coremldata.bin +2 -2
  17. ggml-large-v3-turbo-encoder.mlmodelc/metadata.json +10 -16
  18. ggml-large-v3-turbo-encoder.mlmodelc/model.mil +0 -0
  19. ggml-large-v3-turbo-encoder.mlmodelc/model0/coremldata.bin +0 -3
  20. ggml-large-v3-turbo-encoder.mlmodelc/model0/model.mil +0 -0
  21. ggml-large-v3-turbo-encoder.mlmodelc/model0/weights/0-weight.bin +0 -3
  22. ggml-large-v3-turbo-encoder.mlmodelc/model1/analytics/coremldata.bin +0 -3
  23. ggml-large-v3-turbo-encoder.mlmodelc/model1/coremldata.bin +0 -3
  24. ggml-large-v3-turbo-encoder.mlmodelc/model1/model.mil +0 -0
  25. ggml-large-v3-turbo-encoder.mlmodelc/model1/weights/1-weight.bin +0 -3
  26. {ggml-large-v2-encoder.mlmodelc → ggml-large-v3-turbo-encoder.mlmodelc}/weights/weight.bin +2 -2
  27. ggml-medium-encoder.mlmodelc/analytics/coremldata.bin +1 -1
  28. ggml-medium-encoder.mlmodelc/coremldata.bin +2 -2
  29. ggml-medium-encoder.mlmodelc/metadata.json +22 -20
  30. ggml-medium-encoder.mlmodelc/model.mil +0 -0
  31. ggml-medium-encoder.mlmodelc/weights/weight.bin +1 -1
  32. ggml-base.bin → ggml-medium-q8_0.bin +2 -2
  33. ggml-medium.bin +0 -3
  34. ggml-small-encoder.mlmodelc/analytics/coremldata.bin +1 -1
  35. ggml-small-encoder.mlmodelc/coremldata.bin +2 -2
  36. ggml-small-encoder.mlmodelc/metadata.json +22 -20
  37. ggml-small-encoder.mlmodelc/model.mil +0 -0
  38. ggml-small-encoder.mlmodelc/weights/weight.bin +1 -1
  39. ggml-large-v2-encoder.mlmodelc/coremldata.bin → ggml-small-q8_0.bin +2 -2
  40. ggml-small.bin +0 -3
  41. ggml-tiny-encoder.mlmodelc/analytics/coremldata.bin +1 -1
  42. ggml-tiny-encoder.mlmodelc/coremldata.bin +2 -2
  43. ggml-tiny-encoder.mlmodelc/metadata.json +22 -20
  44. ggml-tiny-encoder.mlmodelc/model.mil +221 -265
  45. ggml-tiny-encoder.mlmodelc/weights/weight.bin +1 -1
  46. ggml-large-v2-encoder.mlmodelc/analytics/coremldata.bin → ggml-tiny-q8_0.bin +2 -2
  47. ggml-tiny.bin +0 -3
  48. index/base +1 -1
  49. index/large-v2 +0 -6
  50. index/large-v3-turbo +2 -8
ggml-base-encoder.mlmodelc/analytics/coremldata.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:31fb009a1caa38a49165cb418f454ef1e5d3cd8e7a1bc37a721575d800a8c712
3
  size 243
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6ce0ee7deaec2f6b2c5927b6a124f866047ab9bfd04e92c30aadae7b9b45e0d7
3
  size 243
ggml-base-encoder.mlmodelc/coremldata.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b353632cc6cb8774fdd42eab32407ce491f6e786a1e15c30fc9c58c7e39cd437
3
- size 318
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:98b532b5a0af2ecdf289928a9b20e6a5aa780da7fd5833fde5c4f90dfa8747ef
3
+ size 379
ggml-base-encoder.mlmodelc/metadata.json CHANGED
@@ -17,36 +17,38 @@
17
  "modelParameters" : [
18
 
19
  ],
20
- "specificationVersion" : 8,
21
  "mlProgramOperationTypeHistogram" : {
22
- "Ios17.layerNorm" : 13,
23
- "Ios17.reshape" : 24,
24
- "Ios17.conv" : 2,
25
- "Ios17.linear" : 36,
26
- "Ios17.add" : 13,
27
- "Ios17.matmul" : 12,
28
- "Ios16.gelu" : 8,
29
- "Ios16.softmax" : 6,
30
- "Ios17.mul" : 12,
31
- "Ios17.transpose" : 25
32
  },
33
  "computePrecision" : "Mixed (Float16, Int32)",
34
  "isUpdatable" : "0",
 
 
 
35
  "availability" : {
36
- "macOS" : "14.0",
37
- "tvOS" : "17.0",
38
- "visionOS" : "1.0",
39
- "watchOS" : "10.0",
40
- "iOS" : "17.0",
41
- "macCatalyst" : "17.0"
42
  },
43
  "modelType" : {
44
  "name" : "MLModelType_mlProgram"
45
  },
46
  "userDefinedMetadata" : {
47
- "com.github.apple.coremltools.source_dialect" : "TorchScript",
48
- "com.github.apple.coremltools.source" : "torch==2.2.2",
49
- "com.github.apple.coremltools.version" : "7.2"
 
50
  },
51
  "inputSchema" : [
52
  {
 
17
  "modelParameters" : [
18
 
19
  ],
20
+ "specificationVersion" : 9,
21
  "mlProgramOperationTypeHistogram" : {
22
+ "Ios18.scaledDotProductAttention" : 6,
23
+ "Ios18.linear" : 36,
24
+ "Ios18.gelu" : 8,
25
+ "Ios18.layerNorm" : 13,
26
+ "Ios18.transpose" : 25,
27
+ "Ios18.conv" : 2,
28
+ "Ios18.add" : 13,
29
+ "Ios18.reshape" : 24
 
 
30
  },
31
  "computePrecision" : "Mixed (Float16, Int32)",
32
  "isUpdatable" : "0",
33
+ "stateSchema" : [
34
+
35
+ ],
36
  "availability" : {
37
+ "macOS" : "15.0",
38
+ "tvOS" : "18.0",
39
+ "visionOS" : "2.0",
40
+ "watchOS" : "11.0",
41
+ "iOS" : "18.0",
42
+ "macCatalyst" : "18.0"
43
  },
44
  "modelType" : {
45
  "name" : "MLModelType_mlProgram"
46
  },
47
  "userDefinedMetadata" : {
48
+ "com.github.apple.coremltools.conversion_date" : "2025-12-13",
49
+ "com.github.apple.coremltools.source" : "torch==2.9.1",
50
+ "com.github.apple.coremltools.version" : "9.0",
51
+ "com.github.apple.coremltools.source_dialect" : "TorchScript"
52
  },
53
  "inputSchema" : [
54
  {
ggml-base-encoder.mlmodelc/model.mil CHANGED
The diff for this file is too large to render. See raw diff
 
ggml-base-encoder.mlmodelc/weights/weight.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b072fd4aebbd60ad4e35df05d3c2ea47267c00f25ac82f4dbcf49fb38e19faec
3
  size 41188544
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c45bee989219532c4cec616d439c51f280ac9d7b04f7847c4b7d7daba1d47523
3
  size 41188544
ggml-large-v3-turbo-encoder.mlmodelc/model0/analytics/coremldata.bin → ggml-base-q8_0.bin RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:5a8281049b2a65a3be541cfd9f949e84b8fe1c5251ce90e46da1626fed54e58a
3
- size 108
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f1a911a31e5812ebffef4c517c19186627a0203b7e95b5dbb6db884a9dc446ef
3
+ size 56663708
ggml-large-v2-encoder.mlmodelc/metadata.json DELETED
@@ -1,66 +0,0 @@
1
- [
2
- {
3
- "metadataOutputVersion" : "3.0",
4
- "storagePrecision" : "Float16",
5
- "outputSchema" : [
6
- {
7
- "hasShapeFlexibility" : "0",
8
- "isOptional" : "0",
9
- "dataType" : "Float16",
10
- "formattedType" : "MultiArray (Float16 1 × 1500 × 1280)",
11
- "shortDescription" : "",
12
- "shape" : "[1, 1500, 1280]",
13
- "name" : "output",
14
- "type" : "MultiArray"
15
- }
16
- ],
17
- "modelParameters" : [
18
-
19
- ],
20
- "specificationVersion" : 8,
21
- "mlProgramOperationTypeHistogram" : {
22
- "Ios17.layerNorm" : 65,
23
- "Ios17.reshape" : 128,
24
- "Ios17.conv" : 2,
25
- "Ios17.linear" : 192,
26
- "Ios17.add" : 65,
27
- "Ios17.matmul" : 64,
28
- "Ios16.gelu" : 34,
29
- "Ios16.softmax" : 32,
30
- "Ios17.mul" : 64,
31
- "Ios17.transpose" : 129
32
- },
33
- "computePrecision" : "Mixed (Float16, Int32)",
34
- "isUpdatable" : "0",
35
- "availability" : {
36
- "macOS" : "14.0",
37
- "tvOS" : "17.0",
38
- "visionOS" : "1.0",
39
- "watchOS" : "10.0",
40
- "iOS" : "17.0",
41
- "macCatalyst" : "17.0"
42
- },
43
- "modelType" : {
44
- "name" : "MLModelType_mlProgram"
45
- },
46
- "userDefinedMetadata" : {
47
- "com.github.apple.coremltools.source_dialect" : "TorchScript",
48
- "com.github.apple.coremltools.source" : "torch==2.3.0",
49
- "com.github.apple.coremltools.version" : "7.2"
50
- },
51
- "inputSchema" : [
52
- {
53
- "hasShapeFlexibility" : "0",
54
- "isOptional" : "0",
55
- "dataType" : "Float16",
56
- "formattedType" : "MultiArray (Float16 1 × 80 × 3000)",
57
- "shortDescription" : "",
58
- "shape" : "[1, 80, 3000]",
59
- "name" : "logmel_data",
60
- "type" : "MultiArray"
61
- }
62
- ],
63
- "generatedClassName" : "ggml_large_v2_encoder",
64
- "method" : "predict"
65
- }
66
- ]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ggml-large-v2-encoder.mlmodelc/model.mil DELETED
The diff for this file is too large to render. See raw diff
 
ggml-large-v2-q8_0.bin DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:3a0b8cd189b25520643b628d08dd100a8f0c884c048c25950302bd01a259ae58
3
- size 967527492
 
 
 
 
ggml-large-v3-encoder.mlmodelc/analytics/coremldata.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2f3c468ec7667cd29ab246467a25b2b93f67b72690248d50d5477ea32e2af932
3
  size 243
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2c012cd81c638ceccf50d4eb96a26f7a440a3a0c677a35120b927a047d1bccc5
3
  size 243
ggml-large-v3-encoder.mlmodelc/coremldata.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:f8cf0ad791b6d5613788de024fd72ea2af5e4c08b696fa2528f5ed25a25b5b4b
3
- size 319
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:435c32326d0bd2151a3fefa83c80af6c7bab4b3f797aeb6e5f18ab131dd2459b
3
+ size 380
ggml-large-v3-encoder.mlmodelc/metadata.json CHANGED
@@ -17,36 +17,38 @@
17
  "modelParameters" : [
18
 
19
  ],
20
- "specificationVersion" : 8,
21
  "mlProgramOperationTypeHistogram" : {
22
- "Ios17.layerNorm" : 65,
23
- "Ios17.reshape" : 128,
24
- "Ios17.conv" : 2,
25
- "Ios17.linear" : 192,
26
- "Ios17.add" : 65,
27
- "Ios17.matmul" : 64,
28
- "Ios16.gelu" : 34,
29
- "Ios16.softmax" : 32,
30
- "Ios17.mul" : 64,
31
- "Ios17.transpose" : 129
32
  },
33
  "computePrecision" : "Mixed (Float16, Int32)",
34
  "isUpdatable" : "0",
 
 
 
35
  "availability" : {
36
- "macOS" : "14.0",
37
- "tvOS" : "17.0",
38
- "visionOS" : "1.0",
39
- "watchOS" : "10.0",
40
- "iOS" : "17.0",
41
- "macCatalyst" : "17.0"
42
  },
43
  "modelType" : {
44
  "name" : "MLModelType_mlProgram"
45
  },
46
  "userDefinedMetadata" : {
47
- "com.github.apple.coremltools.source_dialect" : "TorchScript",
48
- "com.github.apple.coremltools.source" : "torch==2.3.0",
49
- "com.github.apple.coremltools.version" : "7.2"
 
50
  },
51
  "inputSchema" : [
52
  {
 
17
  "modelParameters" : [
18
 
19
  ],
20
+ "specificationVersion" : 9,
21
  "mlProgramOperationTypeHistogram" : {
22
+ "Ios18.scaledDotProductAttention" : 32,
23
+ "Ios18.linear" : 192,
24
+ "Ios18.gelu" : 34,
25
+ "Ios18.layerNorm" : 65,
26
+ "Ios18.transpose" : 129,
27
+ "Ios18.conv" : 2,
28
+ "Ios18.add" : 65,
29
+ "Ios18.reshape" : 128
 
 
30
  },
31
  "computePrecision" : "Mixed (Float16, Int32)",
32
  "isUpdatable" : "0",
33
+ "stateSchema" : [
34
+
35
+ ],
36
  "availability" : {
37
+ "macOS" : "15.0",
38
+ "tvOS" : "18.0",
39
+ "visionOS" : "2.0",
40
+ "watchOS" : "11.0",
41
+ "iOS" : "18.0",
42
+ "macCatalyst" : "18.0"
43
  },
44
  "modelType" : {
45
  "name" : "MLModelType_mlProgram"
46
  },
47
  "userDefinedMetadata" : {
48
+ "com.github.apple.coremltools.conversion_date" : "2025-12-13",
49
+ "com.github.apple.coremltools.source" : "torch==2.9.1",
50
+ "com.github.apple.coremltools.version" : "9.0",
51
+ "com.github.apple.coremltools.source_dialect" : "TorchScript"
52
  },
53
  "inputSchema" : [
54
  {
ggml-large-v3-encoder.mlmodelc/model.mil CHANGED
The diff for this file is too large to render. See raw diff
 
ggml-large-v3-encoder.mlmodelc/weights/weight.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:086e3063e54313291f2d4199d1303e64e466d638384c9469ac3f6ad95ce0c895
3
- size 1064174656
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7cdd15ff2f714bdca758058f70490c89617f764ddc1b25602a6c60eaa408856a
3
+ size 1273971776
ggml-large-v3-turbo-encoder.mlmodelc/analytics/coremldata.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7be9acf764f6e452c949d3a3b27bf0afe3609629db9ba9dd8069127ad8e43c20
3
- size 202
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:521d0f794f87e8fea91108f66d373485c89fce52d861e3cca3e4a5a01b6734fd
3
+ size 243
ggml-large-v3-turbo-encoder.mlmodelc/coremldata.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:46dc321dd0ff6005125dc0365c3e0ecb2413f838328888df48578af4d2869749
3
- size 197
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d188b1f3d2d1c8e15381f6c75bb2a2caa03a5094c30136fe897b6a1d91ce907c
3
+ size 380
ggml-large-v3-turbo-encoder.mlmodelc/metadata.json CHANGED
@@ -19,17 +19,16 @@
19
  ],
20
  "specificationVersion" : 9,
21
  "mlProgramOperationTypeHistogram" : {
22
- "Ios18.reshape" : 128,
23
  "Ios18.linear" : 192,
24
  "Ios18.gelu" : 34,
25
  "Ios18.layerNorm" : 65,
26
  "Ios18.transpose" : 129,
27
  "Ios18.conv" : 2,
28
- "Ios18.cast" : 4,
29
- "Ios18.scaledDotProductAttention" : 32,
30
- "Ios18.add" : 65
31
  },
32
- "computePrecision" : "Mixed (Float16, Float32, Int32)",
33
  "isUpdatable" : "0",
34
  "stateSchema" : [
35
 
@@ -43,18 +42,13 @@
43
  "macCatalyst" : "18.0"
44
  },
45
  "modelType" : {
46
- "name" : "MLModelType_pipeline",
47
- "structure" : [
48
- {
49
- "name" : "MLModelType_mlProgram"
50
- },
51
- {
52
- "name" : "MLModelType_mlProgram"
53
- }
54
- ]
55
  },
56
  "userDefinedMetadata" : {
57
-
 
 
 
58
  },
59
  "inputSchema" : [
60
  {
@@ -68,7 +62,7 @@
68
  "type" : "MultiArray"
69
  }
70
  ],
71
- "generatedClassName" : "chunked_pipeline",
72
  "method" : "predict"
73
  }
74
  ]
 
19
  ],
20
  "specificationVersion" : 9,
21
  "mlProgramOperationTypeHistogram" : {
22
+ "Ios18.scaledDotProductAttention" : 32,
23
  "Ios18.linear" : 192,
24
  "Ios18.gelu" : 34,
25
  "Ios18.layerNorm" : 65,
26
  "Ios18.transpose" : 129,
27
  "Ios18.conv" : 2,
28
+ "Ios18.add" : 65,
29
+ "Ios18.reshape" : 128
 
30
  },
31
+ "computePrecision" : "Mixed (Float16, Int32)",
32
  "isUpdatable" : "0",
33
  "stateSchema" : [
34
 
 
42
  "macCatalyst" : "18.0"
43
  },
44
  "modelType" : {
45
+ "name" : "MLModelType_mlProgram"
 
 
 
 
 
 
 
 
46
  },
47
  "userDefinedMetadata" : {
48
+ "com.github.apple.coremltools.conversion_date" : "2025-12-13",
49
+ "com.github.apple.coremltools.source" : "torch==2.9.1",
50
+ "com.github.apple.coremltools.version" : "9.0",
51
+ "com.github.apple.coremltools.source_dialect" : "TorchScript"
52
  },
53
  "inputSchema" : [
54
  {
 
62
  "type" : "MultiArray"
63
  }
64
  ],
65
+ "generatedClassName" : "ggml_large_v3_turbo_encoder",
66
  "method" : "predict"
67
  }
68
  ]
ggml-large-v3-turbo-encoder.mlmodelc/model.mil ADDED
The diff for this file is too large to render. See raw diff
 
ggml-large-v3-turbo-encoder.mlmodelc/model0/coremldata.bin DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:a2b0461e225831cc34e0017a300f867929784559e2ee471f01ddfd3452381076
3
- size 201
 
 
 
 
ggml-large-v3-turbo-encoder.mlmodelc/model0/model.mil DELETED
The diff for this file is too large to render. See raw diff
 
ggml-large-v3-turbo-encoder.mlmodelc/model0/weights/0-weight.bin DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:6d26bc07916a863ace6c1bb750aad546bdc7b16c12ebe198d3b511134b09eb68
3
- size 644314048
 
 
 
 
ggml-large-v3-turbo-encoder.mlmodelc/model1/analytics/coremldata.bin DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:5a8281049b2a65a3be541cfd9f949e84b8fe1c5251ce90e46da1626fed54e58a
3
- size 108
 
 
 
 
ggml-large-v3-turbo-encoder.mlmodelc/model1/coremldata.bin DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:ee5eef16c4adf0aac1e84f5676c8942aa2741fe919f65d72b1b23bd9dd417ab2
3
- size 196
 
 
 
 
ggml-large-v3-turbo-encoder.mlmodelc/model1/model.mil DELETED
The diff for this file is too large to render. See raw diff
 
ggml-large-v3-turbo-encoder.mlmodelc/model1/weights/1-weight.bin DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:28e9965c3b4ed7db29d226652d29ca6f09a35ee68d2cd61100e7b0f4d0a7095c
3
- size 629660416
 
 
 
 
{ggml-large-v2-encoder.mlmodelc → ggml-large-v3-turbo-encoder.mlmodelc}/weights/weight.bin RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:38680c45f8b18d9586187777968cf81e3e7b7af6baa88a52fc1a06bdb8798548
3
- size 1063806016
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e77bb855473f07c5e56a2343dcf1f3af9c80b6e61ce4247268d67f929e47a1c8
3
+ size 1273971776
ggml-medium-encoder.mlmodelc/analytics/coremldata.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:975604071b5e16602d99864f0c9bd082c0114ced9fea1cce6367ee7a33e99591
3
  size 243
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f3328e031445e499158cd4fc091b4ca7da84476e629db98372b1c52ed49e8f49
3
  size 243
ggml-medium-encoder.mlmodelc/coremldata.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e781c7fcb346060d55fa9f2066f27fdc0e58b6018c50400488144d06e6dd7109
3
- size 318
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1085b1ad3f7ed6de1553948e97404683f50b555a59ee26d06b8c425e8f8f30d1
3
+ size 379
ggml-medium-encoder.mlmodelc/metadata.json CHANGED
@@ -17,36 +17,38 @@
17
  "modelParameters" : [
18
 
19
  ],
20
- "specificationVersion" : 8,
21
  "mlProgramOperationTypeHistogram" : {
22
- "Ios17.layerNorm" : 49,
23
- "Ios17.reshape" : 96,
24
- "Ios17.conv" : 2,
25
- "Ios17.linear" : 144,
26
- "Ios17.add" : 49,
27
- "Ios17.matmul" : 48,
28
- "Ios16.gelu" : 26,
29
- "Ios16.softmax" : 24,
30
- "Ios17.mul" : 48,
31
- "Ios17.transpose" : 97
32
  },
33
  "computePrecision" : "Mixed (Float16, Int32)",
34
  "isUpdatable" : "0",
 
 
 
35
  "availability" : {
36
- "macOS" : "14.0",
37
- "tvOS" : "17.0",
38
- "visionOS" : "1.0",
39
- "watchOS" : "10.0",
40
- "iOS" : "17.0",
41
- "macCatalyst" : "17.0"
42
  },
43
  "modelType" : {
44
  "name" : "MLModelType_mlProgram"
45
  },
46
  "userDefinedMetadata" : {
47
- "com.github.apple.coremltools.source_dialect" : "TorchScript",
48
- "com.github.apple.coremltools.source" : "torch==2.2.2",
49
- "com.github.apple.coremltools.version" : "7.2"
 
50
  },
51
  "inputSchema" : [
52
  {
 
17
  "modelParameters" : [
18
 
19
  ],
20
+ "specificationVersion" : 9,
21
  "mlProgramOperationTypeHistogram" : {
22
+ "Ios18.scaledDotProductAttention" : 24,
23
+ "Ios18.linear" : 144,
24
+ "Ios18.gelu" : 26,
25
+ "Ios18.layerNorm" : 49,
26
+ "Ios18.transpose" : 97,
27
+ "Ios18.conv" : 2,
28
+ "Ios18.add" : 49,
29
+ "Ios18.reshape" : 96
 
 
30
  },
31
  "computePrecision" : "Mixed (Float16, Int32)",
32
  "isUpdatable" : "0",
33
+ "stateSchema" : [
34
+
35
+ ],
36
  "availability" : {
37
+ "macOS" : "15.0",
38
+ "tvOS" : "18.0",
39
+ "visionOS" : "2.0",
40
+ "watchOS" : "11.0",
41
+ "iOS" : "18.0",
42
+ "macCatalyst" : "18.0"
43
  },
44
  "modelType" : {
45
  "name" : "MLModelType_mlProgram"
46
  },
47
  "userDefinedMetadata" : {
48
+ "com.github.apple.coremltools.conversion_date" : "2025-12-13",
49
+ "com.github.apple.coremltools.source" : "torch==2.9.1",
50
+ "com.github.apple.coremltools.version" : "9.0",
51
+ "com.github.apple.coremltools.source_dialect" : "TorchScript"
52
  },
53
  "inputSchema" : [
54
  {
ggml-medium-encoder.mlmodelc/model.mil CHANGED
The diff for this file is too large to render. See raw diff
 
ggml-medium-encoder.mlmodelc/weights/weight.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:cbbdb8a10cee8abc0d0c0d86af33ea4b1d8453380df103fd49272a28143a892b
3
  size 614458432
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2cd1baef4c7d8260ea817ea56705b3700155c01e8d3ea4bc8e364a8674a88d15
3
  size 614458432
ggml-base.bin → ggml-medium-q8_0.bin RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:84ceff52bae6817d2b3c3f5c96e6cfad66352047dd390b86b8aaf0e6f30dff9c
3
- size 105151868
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1562f96f3dd57b9517f286ea4b5843ed70759bce397035ffe84f0bb2d407d62b
3
+ size 488364756
ggml-medium.bin DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:076f80d119389b32fb3ff7ee78330418de19970090183807dfa3da56a4f69fda
3
- size 915642516
 
 
 
 
ggml-small-encoder.mlmodelc/analytics/coremldata.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:0e33d81d8cf4a206337049c6f0379ed5370da2b0fc3d822a6065b2c021664a46
3
  size 243
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ad9997a136336812138a5f02e4977a2a83c5112a5c2928beb7b43bafc55853df
3
  size 243
ggml-small-encoder.mlmodelc/coremldata.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:f6643c80fd2477e972234f11c3d21cc415963809f6bc560980f83bda4c62f31b
3
- size 318
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c6253da76ccc8aefef97c4bc85a4f2a5687e4ed4db03a41c34efff4a7ec9de3e
3
+ size 379
ggml-small-encoder.mlmodelc/metadata.json CHANGED
@@ -17,36 +17,38 @@
17
  "modelParameters" : [
18
 
19
  ],
20
- "specificationVersion" : 8,
21
  "mlProgramOperationTypeHistogram" : {
22
- "Ios17.layerNorm" : 25,
23
- "Ios17.reshape" : 48,
24
- "Ios17.conv" : 2,
25
- "Ios17.linear" : 72,
26
- "Ios17.add" : 25,
27
- "Ios17.matmul" : 24,
28
- "Ios16.gelu" : 14,
29
- "Ios16.softmax" : 12,
30
- "Ios17.mul" : 24,
31
- "Ios17.transpose" : 49
32
  },
33
  "computePrecision" : "Mixed (Float16, Int32)",
34
  "isUpdatable" : "0",
 
 
 
35
  "availability" : {
36
- "macOS" : "14.0",
37
- "tvOS" : "17.0",
38
- "visionOS" : "1.0",
39
- "watchOS" : "10.0",
40
- "iOS" : "17.0",
41
- "macCatalyst" : "17.0"
42
  },
43
  "modelType" : {
44
  "name" : "MLModelType_mlProgram"
45
  },
46
  "userDefinedMetadata" : {
47
- "com.github.apple.coremltools.source_dialect" : "TorchScript",
48
- "com.github.apple.coremltools.source" : "torch==2.2.2",
49
- "com.github.apple.coremltools.version" : "7.2"
 
50
  },
51
  "inputSchema" : [
52
  {
 
17
  "modelParameters" : [
18
 
19
  ],
20
+ "specificationVersion" : 9,
21
  "mlProgramOperationTypeHistogram" : {
22
+ "Ios18.scaledDotProductAttention" : 12,
23
+ "Ios18.linear" : 72,
24
+ "Ios18.gelu" : 14,
25
+ "Ios18.layerNorm" : 25,
26
+ "Ios18.transpose" : 49,
27
+ "Ios18.conv" : 2,
28
+ "Ios18.add" : 25,
29
+ "Ios18.reshape" : 48
 
 
30
  },
31
  "computePrecision" : "Mixed (Float16, Int32)",
32
  "isUpdatable" : "0",
33
+ "stateSchema" : [
34
+
35
+ ],
36
  "availability" : {
37
+ "macOS" : "15.0",
38
+ "tvOS" : "18.0",
39
+ "visionOS" : "2.0",
40
+ "watchOS" : "11.0",
41
+ "iOS" : "18.0",
42
+ "macCatalyst" : "18.0"
43
  },
44
  "modelType" : {
45
  "name" : "MLModelType_mlProgram"
46
  },
47
  "userDefinedMetadata" : {
48
+ "com.github.apple.coremltools.conversion_date" : "2025-12-13",
49
+ "com.github.apple.coremltools.source" : "torch==2.9.1",
50
+ "com.github.apple.coremltools.version" : "9.0",
51
+ "com.github.apple.coremltools.source_dialect" : "TorchScript"
52
  },
53
  "inputSchema" : [
54
  {
ggml-small-encoder.mlmodelc/model.mil CHANGED
The diff for this file is too large to render. See raw diff
 
ggml-small-encoder.mlmodelc/weights/weight.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:c2362f377002638f1f3ec3a413d80c9d82410b537d776a1a75d6571c8e40ace6
3
  size 176321856
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4d3ab676977d57b06993ee7ebc638fc8568a99ddb11cb7a445328ce50fbd8b36
3
  size 176321856
ggml-large-v2-encoder.mlmodelc/coremldata.bin → ggml-small-q8_0.bin RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ae8e3d2e9d47c7254f44d42290d3bfc325144bd7671d7631b3eb222484a1006e
3
- size 318
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2f9f98684c133761578bb32314e4b978c203bc6d4d1b6f5553c356104b9a7571
3
+ size 165242356
ggml-small.bin DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:42ea1b1ba6ffa8cde9da77664db1ac5e24378f9d3d0363ca0104deebedac7732
3
- size 308753476
 
 
 
 
ggml-tiny-encoder.mlmodelc/analytics/coremldata.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:3e6edb6be522938b84fd688c4aef4180f652783aaa157e274bd1ba221c2f9322
3
  size 243
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e4caac0b38c9cca17df0f212e16a01d7046e2f454767565a9161988640492be8
3
  size 243
ggml-tiny-encoder.mlmodelc/coremldata.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:4a2e684990429a39584b0adf645ea565057cb6bdcfd404f2dba45a1031e4faaa
3
- size 318
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:de0977d50f2b62990f3fe458c5f94e6e05147e197dfce5ce9c6fadc370ea0e7f
3
+ size 379
ggml-tiny-encoder.mlmodelc/metadata.json CHANGED
@@ -17,36 +17,38 @@
17
  "modelParameters" : [
18
 
19
  ],
20
- "specificationVersion" : 8,
21
  "mlProgramOperationTypeHistogram" : {
22
- "Ios17.layerNorm" : 9,
23
- "Ios17.reshape" : 16,
24
- "Ios17.conv" : 2,
25
- "Ios17.linear" : 24,
26
- "Ios17.add" : 9,
27
- "Ios17.matmul" : 8,
28
- "Ios16.gelu" : 6,
29
- "Ios16.softmax" : 4,
30
- "Ios17.mul" : 8,
31
- "Ios17.transpose" : 17
32
  },
33
  "computePrecision" : "Mixed (Float16, Int32)",
34
  "isUpdatable" : "0",
 
 
 
35
  "availability" : {
36
- "macOS" : "14.0",
37
- "tvOS" : "17.0",
38
- "visionOS" : "1.0",
39
- "watchOS" : "10.0",
40
- "iOS" : "17.0",
41
- "macCatalyst" : "17.0"
42
  },
43
  "modelType" : {
44
  "name" : "MLModelType_mlProgram"
45
  },
46
  "userDefinedMetadata" : {
47
- "com.github.apple.coremltools.source_dialect" : "TorchScript",
48
- "com.github.apple.coremltools.source" : "torch==2.2.2",
49
- "com.github.apple.coremltools.version" : "7.2"
 
50
  },
51
  "inputSchema" : [
52
  {
 
17
  "modelParameters" : [
18
 
19
  ],
20
+ "specificationVersion" : 9,
21
  "mlProgramOperationTypeHistogram" : {
22
+ "Ios18.scaledDotProductAttention" : 4,
23
+ "Ios18.linear" : 24,
24
+ "Ios18.gelu" : 6,
25
+ "Ios18.layerNorm" : 9,
26
+ "Ios18.transpose" : 17,
27
+ "Ios18.conv" : 2,
28
+ "Ios18.add" : 9,
29
+ "Ios18.reshape" : 16
 
 
30
  },
31
  "computePrecision" : "Mixed (Float16, Int32)",
32
  "isUpdatable" : "0",
33
+ "stateSchema" : [
34
+
35
+ ],
36
  "availability" : {
37
+ "macOS" : "15.0",
38
+ "tvOS" : "18.0",
39
+ "visionOS" : "2.0",
40
+ "watchOS" : "11.0",
41
+ "iOS" : "18.0",
42
+ "macCatalyst" : "18.0"
43
  },
44
  "modelType" : {
45
  "name" : "MLModelType_mlProgram"
46
  },
47
  "userDefinedMetadata" : {
48
+ "com.github.apple.coremltools.conversion_date" : "2025-12-13",
49
+ "com.github.apple.coremltools.source" : "torch==2.9.1",
50
+ "com.github.apple.coremltools.version" : "9.0",
51
+ "com.github.apple.coremltools.source_dialect" : "TorchScript"
52
  },
53
  "inputSchema" : [
54
  {
ggml-tiny-encoder.mlmodelc/model.mil CHANGED
@@ -1,268 +1,224 @@
1
- program(1.0)
2
- [buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3304.5.2"}, {"coremlc-version", "3304.6.2"}, {"coremltools-component-torch", "2.2.2"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "7.2"}})]
3
  {
4
- func main<ios17>(tensor<fp16, [1, 80, 3000]> logmel_data) {
5
- tensor<int32, []> var_16 = const()[name = tensor<string, []>("op_16"), val = tensor<int32, []>(1)];
6
- tensor<int32, [1]> var_24 = const()[name = tensor<string, []>("op_24"), val = tensor<int32, [1]>([1])];
7
- tensor<int32, [1]> var_26 = const()[name = tensor<string, []>("op_26"), val = tensor<int32, [1]>([1])];
8
- tensor<string, []> var_28_pad_type_0 = const()[name = tensor<string, []>("op_28_pad_type_0"), val = tensor<string, []>("custom")];
9
- tensor<int32, [2]> var_28_pad_0 = const()[name = tensor<string, []>("op_28_pad_0"), val = tensor<int32, [2]>([1, 1])];
10
- tensor<fp16, [384, 80, 3]> weight_3_to_fp16 = const()[name = tensor<string, []>("weight_3_to_fp16"), val = tensor<fp16, [384, 80, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
11
- tensor<fp16, [384]> bias_3_to_fp16 = const()[name = tensor<string, []>("bias_3_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(184448)))];
12
- tensor<fp16, [1, 384, 3000]> var_28_cast_fp16 = conv(bias = bias_3_to_fp16, dilations = var_26, groups = var_16, pad = var_28_pad_0, pad_type = var_28_pad_type_0, strides = var_24, weight = weight_3_to_fp16, x = logmel_data)[name = tensor<string, []>("op_28_cast_fp16")];
13
- tensor<string, []> input_1_mode_0 = const()[name = tensor<string, []>("input_1_mode_0"), val = tensor<string, []>("EXACT")];
14
- tensor<fp16, [1, 384, 3000]> input_1_cast_fp16 = gelu(mode = input_1_mode_0, x = var_28_cast_fp16)[name = tensor<string, []>("input_1_cast_fp16")];
15
- tensor<int32, []> var_33 = const()[name = tensor<string, []>("op_33"), val = tensor<int32, []>(1)];
16
- tensor<int32, [1]> var_42 = const()[name = tensor<string, []>("op_42"), val = tensor<int32, [1]>([2])];
17
- tensor<int32, [1]> var_44 = const()[name = tensor<string, []>("op_44"), val = tensor<int32, [1]>([1])];
18
- tensor<string, []> var_46_pad_type_0 = const()[name = tensor<string, []>("op_46_pad_type_0"), val = tensor<string, []>("custom")];
19
- tensor<int32, [2]> var_46_pad_0 = const()[name = tensor<string, []>("op_46_pad_0"), val = tensor<int32, [2]>([1, 1])];
20
- tensor<fp16, [384, 384, 3]> weight_7_to_fp16 = const()[name = tensor<string, []>("weight_7_to_fp16"), val = tensor<fp16, [384, 384, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(185280)))];
21
- tensor<fp16, [384]> bias_7_to_fp16 = const()[name = tensor<string, []>("bias_7_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1070080)))];
22
- tensor<fp16, [1, 384, 1500]> var_46_cast_fp16 = conv(bias = bias_7_to_fp16, dilations = var_44, groups = var_33, pad = var_46_pad_0, pad_type = var_46_pad_type_0, strides = var_42, weight = weight_7_to_fp16, x = input_1_cast_fp16)[name = tensor<string, []>("op_46_cast_fp16")];
23
- tensor<string, []> x_3_mode_0 = const()[name = tensor<string, []>("x_3_mode_0"), val = tensor<string, []>("EXACT")];
24
- tensor<fp16, [1, 384, 1500]> x_3_cast_fp16 = gelu(mode = x_3_mode_0, x = var_46_cast_fp16)[name = tensor<string, []>("x_3_cast_fp16")];
25
- tensor<int32, [3]> var_52 = const()[name = tensor<string, []>("op_52"), val = tensor<int32, [3]>([0, 2, 1])];
26
- tensor<fp16, [1500, 384]> positional_embedding_to_fp16 = const()[name = tensor<string, []>("positional_embedding_to_fp16"), val = tensor<fp16, [1500, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1070912)))];
27
- tensor<fp16, [1, 1500, 384]> transpose_40 = transpose(perm = var_52, x = x_3_cast_fp16)[name = tensor<string, []>("transpose_40")];
28
- tensor<fp16, [1, 1500, 384]> var_55_cast_fp16 = add(x = transpose_40, y = positional_embedding_to_fp16)[name = tensor<string, []>("op_55_cast_fp16")];
29
- tensor<int32, []> var_67 = const()[name = tensor<string, []>("op_67"), val = tensor<int32, []>(-1)];
30
- tensor<int32, [1]> var_83_axes_0 = const()[name = tensor<string, []>("op_83_axes_0"), val = tensor<int32, [1]>([-1])];
31
- tensor<fp16, [384]> blocks_0_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_0_attn_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2222976)))];
32
- tensor<fp16, [384]> blocks_0_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_0_attn_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2223808)))];
33
- tensor<fp16, []> var_73_to_fp16 = const()[name = tensor<string, []>("op_73_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
34
- tensor<fp16, [1, 1500, 384]> var_83_cast_fp16 = layer_norm(axes = var_83_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_73_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_55_cast_fp16)[name = tensor<string, []>("op_83_cast_fp16")];
35
- tensor<fp16, [384, 384]> var_94_to_fp16 = const()[name = tensor<string, []>("op_94_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2224640)))];
36
- tensor<fp16, [384]> var_95_to_fp16 = const()[name = tensor<string, []>("op_95_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2519616)))];
37
- tensor<fp16, [1, 1500, 384]> linear_0_cast_fp16 = linear(bias = var_95_to_fp16, weight = var_94_to_fp16, x = var_83_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")];
38
- tensor<fp16, [384, 384]> var_98_to_fp16 = const()[name = tensor<string, []>("op_98_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2520448)))];
39
- tensor<fp16, [384]> linear_1_bias_0_to_fp16 = const()[name = tensor<string, []>("linear_1_bias_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2815424)))];
40
- tensor<fp16, [1, 1500, 384]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_98_to_fp16, x = var_83_cast_fp16)[name = tensor<string, []>("linear_1_cast_fp16")];
41
- tensor<fp16, [384, 384]> var_102_to_fp16 = const()[name = tensor<string, []>("op_102_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2816256)))];
42
- tensor<fp16, [384]> var_103_to_fp16 = const()[name = tensor<string, []>("op_103_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3111232)))];
43
- tensor<fp16, [1, 1500, 384]> linear_2_cast_fp16 = linear(bias = var_103_to_fp16, weight = var_102_to_fp16, x = var_83_cast_fp16)[name = tensor<string, []>("linear_2_cast_fp16")];
44
- tensor<int32, [4]> var_111 = const()[name = tensor<string, []>("op_111"), val = tensor<int32, [4]>([1, 1500, 6, -1])];
45
- tensor<fp16, [1, 1500, 6, 64]> var_112_cast_fp16 = reshape(shape = var_111, x = linear_0_cast_fp16)[name = tensor<string, []>("op_112_cast_fp16")];
46
- tensor<fp16, [1, 1, 1, 1]> const_28_to_fp16 = const()[name = tensor<string, []>("const_28_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
47
- tensor<fp16, [1, 1500, 6, 64]> q_3_cast_fp16 = mul(x = var_112_cast_fp16, y = const_28_to_fp16)[name = tensor<string, []>("q_3_cast_fp16")];
48
- tensor<int32, [4]> var_118 = const()[name = tensor<string, []>("op_118"), val = tensor<int32, [4]>([1, 1500, 6, -1])];
49
- tensor<fp16, [1, 1500, 6, 64]> var_119_cast_fp16 = reshape(shape = var_118, x = linear_1_cast_fp16)[name = tensor<string, []>("op_119_cast_fp16")];
50
- tensor<fp16, [1, 1, 1, 1]> const_29_to_fp16 = const()[name = tensor<string, []>("const_29_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
51
- tensor<fp16, [1, 1500, 6, 64]> k_3_cast_fp16 = mul(x = var_119_cast_fp16, y = const_29_to_fp16)[name = tensor<string, []>("k_3_cast_fp16")];
52
- tensor<int32, [4]> var_125 = const()[name = tensor<string, []>("op_125"), val = tensor<int32, [4]>([1, 1500, 6, -1])];
53
- tensor<fp16, [1, 1500, 6, 64]> var_126_cast_fp16 = reshape(shape = var_125, x = linear_2_cast_fp16)[name = tensor<string, []>("op_126_cast_fp16")];
54
- tensor<int32, [4]> var_127 = const()[name = tensor<string, []>("op_127"), val = tensor<int32, [4]>([0, 2, 1, 3])];
55
- tensor<bool, []> qk_1_transpose_x_0 = const()[name = tensor<string, []>("qk_1_transpose_x_0"), val = tensor<bool, []>(false)];
56
- tensor<bool, []> qk_1_transpose_y_0 = const()[name = tensor<string, []>("qk_1_transpose_y_0"), val = tensor<bool, []>(false)];
57
- tensor<int32, [4]> transpose_16_perm_0 = const()[name = tensor<string, []>("transpose_16_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
58
- tensor<int32, [4]> transpose_17_perm_0 = const()[name = tensor<string, []>("transpose_17_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
59
- tensor<fp16, [1, 6, 64, 1500]> transpose_37 = transpose(perm = transpose_17_perm_0, x = k_3_cast_fp16)[name = tensor<string, []>("transpose_37")];
60
- tensor<fp16, [1, 6, 1500, 64]> transpose_38 = transpose(perm = transpose_16_perm_0, x = q_3_cast_fp16)[name = tensor<string, []>("transpose_38")];
61
- tensor<fp16, [1, 6, 1500, 1500]> qk_1_cast_fp16 = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = transpose_38, y = transpose_37)[name = tensor<string, []>("qk_1_cast_fp16")];
62
- tensor<fp16, [1, 6, 1500, 1500]> var_131_cast_fp16 = softmax(axis = var_67, x = qk_1_cast_fp16)[name = tensor<string, []>("op_131_cast_fp16")];
63
- tensor<bool, []> var_133_transpose_x_0 = const()[name = tensor<string, []>("op_133_transpose_x_0"), val = tensor<bool, []>(false)];
64
- tensor<bool, []> var_133_transpose_y_0 = const()[name = tensor<string, []>("op_133_transpose_y_0"), val = tensor<bool, []>(false)];
65
- tensor<fp16, [1, 6, 1500, 64]> transpose_39 = transpose(perm = var_127, x = var_126_cast_fp16)[name = tensor<string, []>("transpose_39")];
66
- tensor<fp16, [1, 6, 1500, 64]> var_133_cast_fp16 = matmul(transpose_x = var_133_transpose_x_0, transpose_y = var_133_transpose_y_0, x = var_131_cast_fp16, y = transpose_39)[name = tensor<string, []>("op_133_cast_fp16")];
67
- tensor<int32, [4]> var_134 = const()[name = tensor<string, []>("op_134"), val = tensor<int32, [4]>([0, 2, 1, 3])];
68
- tensor<int32, [3]> concat_0 = const()[name = tensor<string, []>("concat_0"), val = tensor<int32, [3]>([1, 1500, 384])];
69
- tensor<fp16, [1, 1500, 6, 64]> transpose_36 = transpose(perm = var_134, x = var_133_cast_fp16)[name = tensor<string, []>("transpose_36")];
70
- tensor<fp16, [1, 1500, 384]> x_11_cast_fp16 = reshape(shape = concat_0, x = transpose_36)[name = tensor<string, []>("x_11_cast_fp16")];
71
- tensor<fp16, [384, 384]> var_139_to_fp16 = const()[name = tensor<string, []>("op_139_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3112064)))];
72
- tensor<fp16, [384]> var_140_to_fp16 = const()[name = tensor<string, []>("op_140_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3407040)))];
73
- tensor<fp16, [1, 1500, 384]> linear_3_cast_fp16 = linear(bias = var_140_to_fp16, weight = var_139_to_fp16, x = x_11_cast_fp16)[name = tensor<string, []>("linear_3_cast_fp16")];
74
- tensor<fp16, [1, 1500, 384]> x_13_cast_fp16 = add(x = var_55_cast_fp16, y = linear_3_cast_fp16)[name = tensor<string, []>("x_13_cast_fp16")];
75
- tensor<int32, [1]> var_147_axes_0 = const()[name = tensor<string, []>("op_147_axes_0"), val = tensor<int32, [1]>([-1])];
76
- tensor<fp16, [384]> blocks_0_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_0_mlp_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3407872)))];
77
- tensor<fp16, [384]> blocks_0_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_0_mlp_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3408704)))];
78
- tensor<fp16, [1, 1500, 384]> var_147_cast_fp16 = layer_norm(axes = var_147_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_73_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast_fp16)[name = tensor<string, []>("op_147_cast_fp16")];
79
- tensor<fp16, [1536, 384]> var_156_to_fp16 = const()[name = tensor<string, []>("op_156_to_fp16"), val = tensor<fp16, [1536, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3409536)))];
80
- tensor<fp16, [1536]> var_157_to_fp16 = const()[name = tensor<string, []>("op_157_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4589248)))];
81
- tensor<fp16, [1, 1500, 1536]> linear_4_cast_fp16 = linear(bias = var_157_to_fp16, weight = var_156_to_fp16, x = var_147_cast_fp16)[name = tensor<string, []>("linear_4_cast_fp16")];
82
- tensor<string, []> x_17_mode_0 = const()[name = tensor<string, []>("x_17_mode_0"), val = tensor<string, []>("EXACT")];
83
- tensor<fp16, [1, 1500, 1536]> x_17_cast_fp16 = gelu(mode = x_17_mode_0, x = linear_4_cast_fp16)[name = tensor<string, []>("x_17_cast_fp16")];
84
- tensor<fp16, [384, 1536]> var_162_to_fp16 = const()[name = tensor<string, []>("op_162_to_fp16"), val = tensor<fp16, [384, 1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4592384)))];
85
- tensor<fp16, [384]> var_163_to_fp16 = const()[name = tensor<string, []>("op_163_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5772096)))];
86
- tensor<fp16, [1, 1500, 384]> linear_5_cast_fp16 = linear(bias = var_163_to_fp16, weight = var_162_to_fp16, x = x_17_cast_fp16)[name = tensor<string, []>("linear_5_cast_fp16")];
87
- tensor<fp16, [1, 1500, 384]> x_19_cast_fp16 = add(x = x_13_cast_fp16, y = linear_5_cast_fp16)[name = tensor<string, []>("x_19_cast_fp16")];
88
- tensor<int32, []> var_172 = const()[name = tensor<string, []>("op_172"), val = tensor<int32, []>(-1)];
89
- tensor<int32, [1]> var_188_axes_0 = const()[name = tensor<string, []>("op_188_axes_0"), val = tensor<int32, [1]>([-1])];
90
- tensor<fp16, [384]> blocks_1_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_1_attn_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5772928)))];
91
- tensor<fp16, [384]> blocks_1_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_1_attn_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5773760)))];
92
- tensor<fp16, []> var_178_to_fp16 = const()[name = tensor<string, []>("op_178_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
93
- tensor<fp16, [1, 1500, 384]> var_188_cast_fp16 = layer_norm(axes = var_188_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_178_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast_fp16)[name = tensor<string, []>("op_188_cast_fp16")];
94
- tensor<fp16, [384, 384]> var_199_to_fp16 = const()[name = tensor<string, []>("op_199_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5774592)))];
95
- tensor<fp16, [384]> var_200_to_fp16 = const()[name = tensor<string, []>("op_200_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6069568)))];
96
- tensor<fp16, [1, 1500, 384]> linear_6_cast_fp16 = linear(bias = var_200_to_fp16, weight = var_199_to_fp16, x = var_188_cast_fp16)[name = tensor<string, []>("linear_6_cast_fp16")];
97
- tensor<fp16, [384, 384]> var_203_to_fp16 = const()[name = tensor<string, []>("op_203_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6070400)))];
98
- tensor<fp16, [1, 1500, 384]> linear_7_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_203_to_fp16, x = var_188_cast_fp16)[name = tensor<string, []>("linear_7_cast_fp16")];
99
- tensor<fp16, [384, 384]> var_207_to_fp16 = const()[name = tensor<string, []>("op_207_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6365376)))];
100
- tensor<fp16, [384]> var_208_to_fp16 = const()[name = tensor<string, []>("op_208_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6660352)))];
101
- tensor<fp16, [1, 1500, 384]> linear_8_cast_fp16 = linear(bias = var_208_to_fp16, weight = var_207_to_fp16, x = var_188_cast_fp16)[name = tensor<string, []>("linear_8_cast_fp16")];
102
- tensor<int32, [4]> var_216 = const()[name = tensor<string, []>("op_216"), val = tensor<int32, [4]>([1, 1500, 6, -1])];
103
- tensor<fp16, [1, 1500, 6, 64]> var_217_cast_fp16 = reshape(shape = var_216, x = linear_6_cast_fp16)[name = tensor<string, []>("op_217_cast_fp16")];
104
- tensor<fp16, [1, 1, 1, 1]> const_30_to_fp16 = const()[name = tensor<string, []>("const_30_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
105
- tensor<fp16, [1, 1500, 6, 64]> q_7_cast_fp16 = mul(x = var_217_cast_fp16, y = const_30_to_fp16)[name = tensor<string, []>("q_7_cast_fp16")];
106
- tensor<int32, [4]> var_223 = const()[name = tensor<string, []>("op_223"), val = tensor<int32, [4]>([1, 1500, 6, -1])];
107
- tensor<fp16, [1, 1500, 6, 64]> var_224_cast_fp16 = reshape(shape = var_223, x = linear_7_cast_fp16)[name = tensor<string, []>("op_224_cast_fp16")];
108
- tensor<fp16, [1, 1, 1, 1]> const_31_to_fp16 = const()[name = tensor<string, []>("const_31_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
109
- tensor<fp16, [1, 1500, 6, 64]> k_7_cast_fp16 = mul(x = var_224_cast_fp16, y = const_31_to_fp16)[name = tensor<string, []>("k_7_cast_fp16")];
110
- tensor<int32, [4]> var_230 = const()[name = tensor<string, []>("op_230"), val = tensor<int32, [4]>([1, 1500, 6, -1])];
111
- tensor<fp16, [1, 1500, 6, 64]> var_231_cast_fp16 = reshape(shape = var_230, x = linear_8_cast_fp16)[name = tensor<string, []>("op_231_cast_fp16")];
112
- tensor<int32, [4]> var_232 = const()[name = tensor<string, []>("op_232"), val = tensor<int32, [4]>([0, 2, 1, 3])];
113
- tensor<bool, []> qk_3_transpose_x_0 = const()[name = tensor<string, []>("qk_3_transpose_x_0"), val = tensor<bool, []>(false)];
114
- tensor<bool, []> qk_3_transpose_y_0 = const()[name = tensor<string, []>("qk_3_transpose_y_0"), val = tensor<bool, []>(false)];
115
- tensor<int32, [4]> transpose_18_perm_0 = const()[name = tensor<string, []>("transpose_18_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
116
- tensor<int32, [4]> transpose_19_perm_0 = const()[name = tensor<string, []>("transpose_19_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
117
- tensor<fp16, [1, 6, 64, 1500]> transpose_33 = transpose(perm = transpose_19_perm_0, x = k_7_cast_fp16)[name = tensor<string, []>("transpose_33")];
118
- tensor<fp16, [1, 6, 1500, 64]> transpose_34 = transpose(perm = transpose_18_perm_0, x = q_7_cast_fp16)[name = tensor<string, []>("transpose_34")];
119
- tensor<fp16, [1, 6, 1500, 1500]> qk_3_cast_fp16 = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = transpose_34, y = transpose_33)[name = tensor<string, []>("qk_3_cast_fp16")];
120
- tensor<fp16, [1, 6, 1500, 1500]> var_236_cast_fp16 = softmax(axis = var_172, x = qk_3_cast_fp16)[name = tensor<string, []>("op_236_cast_fp16")];
121
- tensor<bool, []> var_238_transpose_x_0 = const()[name = tensor<string, []>("op_238_transpose_x_0"), val = tensor<bool, []>(false)];
122
- tensor<bool, []> var_238_transpose_y_0 = const()[name = tensor<string, []>("op_238_transpose_y_0"), val = tensor<bool, []>(false)];
123
- tensor<fp16, [1, 6, 1500, 64]> transpose_35 = transpose(perm = var_232, x = var_231_cast_fp16)[name = tensor<string, []>("transpose_35")];
124
- tensor<fp16, [1, 6, 1500, 64]> var_238_cast_fp16 = matmul(transpose_x = var_238_transpose_x_0, transpose_y = var_238_transpose_y_0, x = var_236_cast_fp16, y = transpose_35)[name = tensor<string, []>("op_238_cast_fp16")];
125
- tensor<int32, [4]> var_239 = const()[name = tensor<string, []>("op_239"), val = tensor<int32, [4]>([0, 2, 1, 3])];
126
- tensor<int32, [3]> concat_1 = const()[name = tensor<string, []>("concat_1"), val = tensor<int32, [3]>([1, 1500, 384])];
127
- tensor<fp16, [1, 1500, 6, 64]> transpose_32 = transpose(perm = var_239, x = var_238_cast_fp16)[name = tensor<string, []>("transpose_32")];
128
- tensor<fp16, [1, 1500, 384]> x_23_cast_fp16 = reshape(shape = concat_1, x = transpose_32)[name = tensor<string, []>("x_23_cast_fp16")];
129
- tensor<fp16, [384, 384]> var_244_to_fp16 = const()[name = tensor<string, []>("op_244_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6661184)))];
130
- tensor<fp16, [384]> var_245_to_fp16 = const()[name = tensor<string, []>("op_245_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6956160)))];
131
- tensor<fp16, [1, 1500, 384]> linear_9_cast_fp16 = linear(bias = var_245_to_fp16, weight = var_244_to_fp16, x = x_23_cast_fp16)[name = tensor<string, []>("linear_9_cast_fp16")];
132
- tensor<fp16, [1, 1500, 384]> x_25_cast_fp16 = add(x = x_19_cast_fp16, y = linear_9_cast_fp16)[name = tensor<string, []>("x_25_cast_fp16")];
133
- tensor<int32, [1]> var_252_axes_0 = const()[name = tensor<string, []>("op_252_axes_0"), val = tensor<int32, [1]>([-1])];
134
- tensor<fp16, [384]> blocks_1_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_1_mlp_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6956992)))];
135
- tensor<fp16, [384]> blocks_1_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_1_mlp_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6957824)))];
136
- tensor<fp16, [1, 1500, 384]> var_252_cast_fp16 = layer_norm(axes = var_252_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_178_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast_fp16)[name = tensor<string, []>("op_252_cast_fp16")];
137
- tensor<fp16, [1536, 384]> var_261_to_fp16 = const()[name = tensor<string, []>("op_261_to_fp16"), val = tensor<fp16, [1536, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6958656)))];
138
- tensor<fp16, [1536]> var_262_to_fp16 = const()[name = tensor<string, []>("op_262_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8138368)))];
139
- tensor<fp16, [1, 1500, 1536]> linear_10_cast_fp16 = linear(bias = var_262_to_fp16, weight = var_261_to_fp16, x = var_252_cast_fp16)[name = tensor<string, []>("linear_10_cast_fp16")];
140
- tensor<string, []> x_29_mode_0 = const()[name = tensor<string, []>("x_29_mode_0"), val = tensor<string, []>("EXACT")];
141
- tensor<fp16, [1, 1500, 1536]> x_29_cast_fp16 = gelu(mode = x_29_mode_0, x = linear_10_cast_fp16)[name = tensor<string, []>("x_29_cast_fp16")];
142
- tensor<fp16, [384, 1536]> var_267_to_fp16 = const()[name = tensor<string, []>("op_267_to_fp16"), val = tensor<fp16, [384, 1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8141504)))];
143
- tensor<fp16, [384]> var_268_to_fp16 = const()[name = tensor<string, []>("op_268_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9321216)))];
144
- tensor<fp16, [1, 1500, 384]> linear_11_cast_fp16 = linear(bias = var_268_to_fp16, weight = var_267_to_fp16, x = x_29_cast_fp16)[name = tensor<string, []>("linear_11_cast_fp16")];
145
- tensor<fp16, [1, 1500, 384]> x_31_cast_fp16 = add(x = x_25_cast_fp16, y = linear_11_cast_fp16)[name = tensor<string, []>("x_31_cast_fp16")];
146
- tensor<int32, []> var_277 = const()[name = tensor<string, []>("op_277"), val = tensor<int32, []>(-1)];
147
- tensor<int32, [1]> var_293_axes_0 = const()[name = tensor<string, []>("op_293_axes_0"), val = tensor<int32, [1]>([-1])];
148
- tensor<fp16, [384]> blocks_2_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_2_attn_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9322048)))];
149
- tensor<fp16, [384]> blocks_2_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_2_attn_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9322880)))];
150
- tensor<fp16, []> var_283_to_fp16 = const()[name = tensor<string, []>("op_283_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
151
- tensor<fp16, [1, 1500, 384]> var_293_cast_fp16 = layer_norm(axes = var_293_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_283_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast_fp16)[name = tensor<string, []>("op_293_cast_fp16")];
152
- tensor<fp16, [384, 384]> var_304_to_fp16 = const()[name = tensor<string, []>("op_304_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9323712)))];
153
- tensor<fp16, [384]> var_305_to_fp16 = const()[name = tensor<string, []>("op_305_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9618688)))];
154
- tensor<fp16, [1, 1500, 384]> linear_12_cast_fp16 = linear(bias = var_305_to_fp16, weight = var_304_to_fp16, x = var_293_cast_fp16)[name = tensor<string, []>("linear_12_cast_fp16")];
155
- tensor<fp16, [384, 384]> var_308_to_fp16 = const()[name = tensor<string, []>("op_308_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9619520)))];
156
- tensor<fp16, [1, 1500, 384]> linear_13_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_308_to_fp16, x = var_293_cast_fp16)[name = tensor<string, []>("linear_13_cast_fp16")];
157
- tensor<fp16, [384, 384]> var_312_to_fp16 = const()[name = tensor<string, []>("op_312_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9914496)))];
158
- tensor<fp16, [384]> var_313_to_fp16 = const()[name = tensor<string, []>("op_313_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10209472)))];
159
- tensor<fp16, [1, 1500, 384]> linear_14_cast_fp16 = linear(bias = var_313_to_fp16, weight = var_312_to_fp16, x = var_293_cast_fp16)[name = tensor<string, []>("linear_14_cast_fp16")];
160
- tensor<int32, [4]> var_321 = const()[name = tensor<string, []>("op_321"), val = tensor<int32, [4]>([1, 1500, 6, -1])];
161
- tensor<fp16, [1, 1500, 6, 64]> var_322_cast_fp16 = reshape(shape = var_321, x = linear_12_cast_fp16)[name = tensor<string, []>("op_322_cast_fp16")];
162
- tensor<fp16, [1, 1, 1, 1]> const_32_to_fp16 = const()[name = tensor<string, []>("const_32_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
163
- tensor<fp16, [1, 1500, 6, 64]> q_11_cast_fp16 = mul(x = var_322_cast_fp16, y = const_32_to_fp16)[name = tensor<string, []>("q_11_cast_fp16")];
164
- tensor<int32, [4]> var_328 = const()[name = tensor<string, []>("op_328"), val = tensor<int32, [4]>([1, 1500, 6, -1])];
165
- tensor<fp16, [1, 1500, 6, 64]> var_329_cast_fp16 = reshape(shape = var_328, x = linear_13_cast_fp16)[name = tensor<string, []>("op_329_cast_fp16")];
166
- tensor<fp16, [1, 1, 1, 1]> const_33_to_fp16 = const()[name = tensor<string, []>("const_33_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
167
- tensor<fp16, [1, 1500, 6, 64]> k_11_cast_fp16 = mul(x = var_329_cast_fp16, y = const_33_to_fp16)[name = tensor<string, []>("k_11_cast_fp16")];
168
- tensor<int32, [4]> var_335 = const()[name = tensor<string, []>("op_335"), val = tensor<int32, [4]>([1, 1500, 6, -1])];
169
- tensor<fp16, [1, 1500, 6, 64]> var_336_cast_fp16 = reshape(shape = var_335, x = linear_14_cast_fp16)[name = tensor<string, []>("op_336_cast_fp16")];
170
- tensor<int32, [4]> var_337 = const()[name = tensor<string, []>("op_337"), val = tensor<int32, [4]>([0, 2, 1, 3])];
171
- tensor<bool, []> qk_5_transpose_x_0 = const()[name = tensor<string, []>("qk_5_transpose_x_0"), val = tensor<bool, []>(false)];
172
- tensor<bool, []> qk_5_transpose_y_0 = const()[name = tensor<string, []>("qk_5_transpose_y_0"), val = tensor<bool, []>(false)];
173
- tensor<int32, [4]> transpose_20_perm_0 = const()[name = tensor<string, []>("transpose_20_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
174
- tensor<int32, [4]> transpose_21_perm_0 = const()[name = tensor<string, []>("transpose_21_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
175
- tensor<fp16, [1, 6, 64, 1500]> transpose_29 = transpose(perm = transpose_21_perm_0, x = k_11_cast_fp16)[name = tensor<string, []>("transpose_29")];
176
- tensor<fp16, [1, 6, 1500, 64]> transpose_30 = transpose(perm = transpose_20_perm_0, x = q_11_cast_fp16)[name = tensor<string, []>("transpose_30")];
177
- tensor<fp16, [1, 6, 1500, 1500]> qk_5_cast_fp16 = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = transpose_30, y = transpose_29)[name = tensor<string, []>("qk_5_cast_fp16")];
178
- tensor<fp16, [1, 6, 1500, 1500]> var_341_cast_fp16 = softmax(axis = var_277, x = qk_5_cast_fp16)[name = tensor<string, []>("op_341_cast_fp16")];
179
- tensor<bool, []> var_343_transpose_x_0 = const()[name = tensor<string, []>("op_343_transpose_x_0"), val = tensor<bool, []>(false)];
180
- tensor<bool, []> var_343_transpose_y_0 = const()[name = tensor<string, []>("op_343_transpose_y_0"), val = tensor<bool, []>(false)];
181
- tensor<fp16, [1, 6, 1500, 64]> transpose_31 = transpose(perm = var_337, x = var_336_cast_fp16)[name = tensor<string, []>("transpose_31")];
182
- tensor<fp16, [1, 6, 1500, 64]> var_343_cast_fp16 = matmul(transpose_x = var_343_transpose_x_0, transpose_y = var_343_transpose_y_0, x = var_341_cast_fp16, y = transpose_31)[name = tensor<string, []>("op_343_cast_fp16")];
183
- tensor<int32, [4]> var_344 = const()[name = tensor<string, []>("op_344"), val = tensor<int32, [4]>([0, 2, 1, 3])];
184
- tensor<int32, [3]> concat_2 = const()[name = tensor<string, []>("concat_2"), val = tensor<int32, [3]>([1, 1500, 384])];
185
- tensor<fp16, [1, 1500, 6, 64]> transpose_28 = transpose(perm = var_344, x = var_343_cast_fp16)[name = tensor<string, []>("transpose_28")];
186
- tensor<fp16, [1, 1500, 384]> x_35_cast_fp16 = reshape(shape = concat_2, x = transpose_28)[name = tensor<string, []>("x_35_cast_fp16")];
187
- tensor<fp16, [384, 384]> var_349_to_fp16 = const()[name = tensor<string, []>("op_349_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10210304)))];
188
- tensor<fp16, [384]> var_350_to_fp16 = const()[name = tensor<string, []>("op_350_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10505280)))];
189
- tensor<fp16, [1, 1500, 384]> linear_15_cast_fp16 = linear(bias = var_350_to_fp16, weight = var_349_to_fp16, x = x_35_cast_fp16)[name = tensor<string, []>("linear_15_cast_fp16")];
190
- tensor<fp16, [1, 1500, 384]> x_37_cast_fp16 = add(x = x_31_cast_fp16, y = linear_15_cast_fp16)[name = tensor<string, []>("x_37_cast_fp16")];
191
- tensor<int32, [1]> var_357_axes_0 = const()[name = tensor<string, []>("op_357_axes_0"), val = tensor<int32, [1]>([-1])];
192
- tensor<fp16, [384]> blocks_2_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_2_mlp_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10506112)))];
193
- tensor<fp16, [384]> blocks_2_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_2_mlp_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10506944)))];
194
- tensor<fp16, [1, 1500, 384]> var_357_cast_fp16 = layer_norm(axes = var_357_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_283_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast_fp16)[name = tensor<string, []>("op_357_cast_fp16")];
195
- tensor<fp16, [1536, 384]> var_366_to_fp16 = const()[name = tensor<string, []>("op_366_to_fp16"), val = tensor<fp16, [1536, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10507776)))];
196
- tensor<fp16, [1536]> var_367_to_fp16 = const()[name = tensor<string, []>("op_367_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11687488)))];
197
- tensor<fp16, [1, 1500, 1536]> linear_16_cast_fp16 = linear(bias = var_367_to_fp16, weight = var_366_to_fp16, x = var_357_cast_fp16)[name = tensor<string, []>("linear_16_cast_fp16")];
198
- tensor<string, []> x_41_mode_0 = const()[name = tensor<string, []>("x_41_mode_0"), val = tensor<string, []>("EXACT")];
199
- tensor<fp16, [1, 1500, 1536]> x_41_cast_fp16 = gelu(mode = x_41_mode_0, x = linear_16_cast_fp16)[name = tensor<string, []>("x_41_cast_fp16")];
200
- tensor<fp16, [384, 1536]> var_372_to_fp16 = const()[name = tensor<string, []>("op_372_to_fp16"), val = tensor<fp16, [384, 1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11690624)))];
201
- tensor<fp16, [384]> var_373_to_fp16 = const()[name = tensor<string, []>("op_373_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12870336)))];
202
- tensor<fp16, [1, 1500, 384]> linear_17_cast_fp16 = linear(bias = var_373_to_fp16, weight = var_372_to_fp16, x = x_41_cast_fp16)[name = tensor<string, []>("linear_17_cast_fp16")];
203
- tensor<fp16, [1, 1500, 384]> x_43_cast_fp16 = add(x = x_37_cast_fp16, y = linear_17_cast_fp16)[name = tensor<string, []>("x_43_cast_fp16")];
204
- tensor<int32, []> var_382 = const()[name = tensor<string, []>("op_382"), val = tensor<int32, []>(-1)];
205
- tensor<int32, [1]> var_398_axes_0 = const()[name = tensor<string, []>("op_398_axes_0"), val = tensor<int32, [1]>([-1])];
206
- tensor<fp16, [384]> blocks_3_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_3_attn_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12871168)))];
207
- tensor<fp16, [384]> blocks_3_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_3_attn_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12872000)))];
208
- tensor<fp16, []> var_388_to_fp16 = const()[name = tensor<string, []>("op_388_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
209
- tensor<fp16, [1, 1500, 384]> var_398_cast_fp16 = layer_norm(axes = var_398_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_388_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast_fp16)[name = tensor<string, []>("op_398_cast_fp16")];
210
- tensor<fp16, [384, 384]> var_409_to_fp16 = const()[name = tensor<string, []>("op_409_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12872832)))];
211
- tensor<fp16, [384]> var_410_to_fp16 = const()[name = tensor<string, []>("op_410_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13167808)))];
212
- tensor<fp16, [1, 1500, 384]> linear_18_cast_fp16 = linear(bias = var_410_to_fp16, weight = var_409_to_fp16, x = var_398_cast_fp16)[name = tensor<string, []>("linear_18_cast_fp16")];
213
- tensor<fp16, [384, 384]> var_413_to_fp16 = const()[name = tensor<string, []>("op_413_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13168640)))];
214
- tensor<fp16, [1, 1500, 384]> linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_413_to_fp16, x = var_398_cast_fp16)[name = tensor<string, []>("linear_19_cast_fp16")];
215
- tensor<fp16, [384, 384]> var_417_to_fp16 = const()[name = tensor<string, []>("op_417_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13463616)))];
216
- tensor<fp16, [384]> var_418_to_fp16 = const()[name = tensor<string, []>("op_418_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13758592)))];
217
- tensor<fp16, [1, 1500, 384]> linear_20_cast_fp16 = linear(bias = var_418_to_fp16, weight = var_417_to_fp16, x = var_398_cast_fp16)[name = tensor<string, []>("linear_20_cast_fp16")];
218
- tensor<int32, [4]> var_426 = const()[name = tensor<string, []>("op_426"), val = tensor<int32, [4]>([1, 1500, 6, -1])];
219
- tensor<fp16, [1, 1500, 6, 64]> var_427_cast_fp16 = reshape(shape = var_426, x = linear_18_cast_fp16)[name = tensor<string, []>("op_427_cast_fp16")];
220
- tensor<fp16, [1, 1, 1, 1]> const_34_to_fp16 = const()[name = tensor<string, []>("const_34_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
221
- tensor<fp16, [1, 1500, 6, 64]> q_cast_fp16 = mul(x = var_427_cast_fp16, y = const_34_to_fp16)[name = tensor<string, []>("q_cast_fp16")];
222
- tensor<int32, [4]> var_433 = const()[name = tensor<string, []>("op_433"), val = tensor<int32, [4]>([1, 1500, 6, -1])];
223
- tensor<fp16, [1, 1500, 6, 64]> var_434_cast_fp16 = reshape(shape = var_433, x = linear_19_cast_fp16)[name = tensor<string, []>("op_434_cast_fp16")];
224
- tensor<fp16, [1, 1, 1, 1]> const_35_to_fp16 = const()[name = tensor<string, []>("const_35_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
225
- tensor<fp16, [1, 1500, 6, 64]> k_cast_fp16 = mul(x = var_434_cast_fp16, y = const_35_to_fp16)[name = tensor<string, []>("k_cast_fp16")];
226
- tensor<int32, [4]> var_440 = const()[name = tensor<string, []>("op_440"), val = tensor<int32, [4]>([1, 1500, 6, -1])];
227
- tensor<fp16, [1, 1500, 6, 64]> var_441_cast_fp16 = reshape(shape = var_440, x = linear_20_cast_fp16)[name = tensor<string, []>("op_441_cast_fp16")];
228
- tensor<int32, [4]> var_442 = const()[name = tensor<string, []>("op_442"), val = tensor<int32, [4]>([0, 2, 1, 3])];
229
- tensor<bool, []> qk_transpose_x_0 = const()[name = tensor<string, []>("qk_transpose_x_0"), val = tensor<bool, []>(false)];
230
- tensor<bool, []> qk_transpose_y_0 = const()[name = tensor<string, []>("qk_transpose_y_0"), val = tensor<bool, []>(false)];
231
- tensor<int32, [4]> transpose_22_perm_0 = const()[name = tensor<string, []>("transpose_22_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
232
- tensor<int32, [4]> transpose_23_perm_0 = const()[name = tensor<string, []>("transpose_23_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
233
- tensor<fp16, [1, 6, 64, 1500]> transpose_25 = transpose(perm = transpose_23_perm_0, x = k_cast_fp16)[name = tensor<string, []>("transpose_25")];
234
- tensor<fp16, [1, 6, 1500, 64]> transpose_26 = transpose(perm = transpose_22_perm_0, x = q_cast_fp16)[name = tensor<string, []>("transpose_26")];
235
- tensor<fp16, [1, 6, 1500, 1500]> qk_cast_fp16 = matmul(transpose_x = qk_transpose_x_0, transpose_y = qk_transpose_y_0, x = transpose_26, y = transpose_25)[name = tensor<string, []>("qk_cast_fp16")];
236
- tensor<fp16, [1, 6, 1500, 1500]> var_446_cast_fp16 = softmax(axis = var_382, x = qk_cast_fp16)[name = tensor<string, []>("op_446_cast_fp16")];
237
- tensor<bool, []> var_448_transpose_x_0 = const()[name = tensor<string, []>("op_448_transpose_x_0"), val = tensor<bool, []>(false)];
238
- tensor<bool, []> var_448_transpose_y_0 = const()[name = tensor<string, []>("op_448_transpose_y_0"), val = tensor<bool, []>(false)];
239
- tensor<fp16, [1, 6, 1500, 64]> transpose_27 = transpose(perm = var_442, x = var_441_cast_fp16)[name = tensor<string, []>("transpose_27")];
240
- tensor<fp16, [1, 6, 1500, 64]> var_448_cast_fp16 = matmul(transpose_x = var_448_transpose_x_0, transpose_y = var_448_transpose_y_0, x = var_446_cast_fp16, y = transpose_27)[name = tensor<string, []>("op_448_cast_fp16")];
241
- tensor<int32, [4]> var_449 = const()[name = tensor<string, []>("op_449"), val = tensor<int32, [4]>([0, 2, 1, 3])];
242
- tensor<int32, [3]> concat_3 = const()[name = tensor<string, []>("concat_3"), val = tensor<int32, [3]>([1, 1500, 384])];
243
- tensor<fp16, [1, 1500, 6, 64]> transpose_24 = transpose(perm = var_449, x = var_448_cast_fp16)[name = tensor<string, []>("transpose_24")];
244
- tensor<fp16, [1, 1500, 384]> x_47_cast_fp16 = reshape(shape = concat_3, x = transpose_24)[name = tensor<string, []>("x_47_cast_fp16")];
245
- tensor<fp16, [384, 384]> var_454_to_fp16 = const()[name = tensor<string, []>("op_454_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13759424)))];
246
- tensor<fp16, [384]> var_455_to_fp16 = const()[name = tensor<string, []>("op_455_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14054400)))];
247
- tensor<fp16, [1, 1500, 384]> linear_21_cast_fp16 = linear(bias = var_455_to_fp16, weight = var_454_to_fp16, x = x_47_cast_fp16)[name = tensor<string, []>("linear_21_cast_fp16")];
248
- tensor<fp16, [1, 1500, 384]> x_49_cast_fp16 = add(x = x_43_cast_fp16, y = linear_21_cast_fp16)[name = tensor<string, []>("x_49_cast_fp16")];
249
- tensor<int32, [1]> var_462_axes_0 = const()[name = tensor<string, []>("op_462_axes_0"), val = tensor<int32, [1]>([-1])];
250
- tensor<fp16, [384]> blocks_3_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_3_mlp_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14055232)))];
251
- tensor<fp16, [384]> blocks_3_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_3_mlp_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14056064)))];
252
- tensor<fp16, [1, 1500, 384]> var_462_cast_fp16 = layer_norm(axes = var_462_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_388_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast_fp16)[name = tensor<string, []>("op_462_cast_fp16")];
253
- tensor<fp16, [1536, 384]> var_471_to_fp16 = const()[name = tensor<string, []>("op_471_to_fp16"), val = tensor<fp16, [1536, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14056896)))];
254
- tensor<fp16, [1536]> var_472_to_fp16 = const()[name = tensor<string, []>("op_472_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15236608)))];
255
- tensor<fp16, [1, 1500, 1536]> linear_22_cast_fp16 = linear(bias = var_472_to_fp16, weight = var_471_to_fp16, x = var_462_cast_fp16)[name = tensor<string, []>("linear_22_cast_fp16")];
256
- tensor<string, []> x_53_mode_0 = const()[name = tensor<string, []>("x_53_mode_0"), val = tensor<string, []>("EXACT")];
257
- tensor<fp16, [1, 1500, 1536]> x_53_cast_fp16 = gelu(mode = x_53_mode_0, x = linear_22_cast_fp16)[name = tensor<string, []>("x_53_cast_fp16")];
258
- tensor<fp16, [384, 1536]> var_477_to_fp16 = const()[name = tensor<string, []>("op_477_to_fp16"), val = tensor<fp16, [384, 1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15239744)))];
259
- tensor<fp16, [384]> var_478_to_fp16 = const()[name = tensor<string, []>("op_478_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16419456)))];
260
- tensor<fp16, [1, 1500, 384]> linear_23_cast_fp16 = linear(bias = var_478_to_fp16, weight = var_477_to_fp16, x = x_53_cast_fp16)[name = tensor<string, []>("linear_23_cast_fp16")];
261
- tensor<fp16, [1, 1500, 384]> x_cast_fp16 = add(x = x_49_cast_fp16, y = linear_23_cast_fp16)[name = tensor<string, []>("x_cast_fp16")];
262
- tensor<int32, [1]> var_491_axes_0 = const()[name = tensor<string, []>("op_491_axes_0"), val = tensor<int32, [1]>([-1])];
263
- tensor<fp16, [384]> ln_post_weight_to_fp16 = const()[name = tensor<string, []>("ln_post_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16420288)))];
264
- tensor<fp16, [384]> ln_post_bias_to_fp16 = const()[name = tensor<string, []>("ln_post_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16421120)))];
265
- tensor<fp16, []> var_482_to_fp16 = const()[name = tensor<string, []>("op_482_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
266
- tensor<fp16, [1, 1500, 384]> output = layer_norm(axes = var_491_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_482_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast_fp16)[name = tensor<string, []>("op_491_cast_fp16")];
267
  } -> (output);
268
  }
 
1
+ program(1.3)
2
+ [buildInfo = dict<string, 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"}})]
3
  {
4
+ func main<ios18>(tensor<fp16, [1, 80, 3000]> logmel_data) {
5
+ string var_28_pad_type_0 = const()[name = string("op_28_pad_type_0"), val = string("custom")];
6
+ tensor<int32, [2]> var_28_pad_0 = const()[name = string("op_28_pad_0"), val = tensor<int32, [2]>([1, 1])];
7
+ tensor<int32, [1]> var_28_strides_0 = const()[name = string("op_28_strides_0"), val = tensor<int32, [1]>([1])];
8
+ tensor<int32, [1]> var_28_dilations_0 = const()[name = string("op_28_dilations_0"), val = tensor<int32, [1]>([1])];
9
+ int32 var_28_groups_0 = const()[name = string("op_28_groups_0"), val = int32(1)];
10
+ tensor<fp16, [384, 80, 3]> const_0_to_fp16 = const()[name = string("const_0_to_fp16"), val = tensor<fp16, [384, 80, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))];
11
+ tensor<fp16, [384]> const_1_to_fp16 = const()[name = string("const_1_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184448)))];
12
+ tensor<fp16, [1, 384, 3000]> var_28_cast_fp16 = conv(bias = const_1_to_fp16, dilations = var_28_dilations_0, groups = var_28_groups_0, pad = var_28_pad_0, pad_type = var_28_pad_type_0, strides = var_28_strides_0, weight = const_0_to_fp16, x = logmel_data)[name = string("op_28_cast_fp16")];
13
+ string input_1_mode_0 = const()[name = string("input_1_mode_0"), val = string("EXACT")];
14
+ tensor<fp16, [1, 384, 3000]> input_1_cast_fp16 = gelu(mode = input_1_mode_0, x = var_28_cast_fp16)[name = string("input_1_cast_fp16")];
15
+ string var_46_pad_type_0 = const()[name = string("op_46_pad_type_0"), val = string("custom")];
16
+ tensor<int32, [2]> var_46_pad_0 = const()[name = string("op_46_pad_0"), val = tensor<int32, [2]>([1, 1])];
17
+ tensor<int32, [1]> var_46_strides_0 = const()[name = string("op_46_strides_0"), val = tensor<int32, [1]>([2])];
18
+ tensor<int32, [1]> var_46_dilations_0 = const()[name = string("op_46_dilations_0"), val = tensor<int32, [1]>([1])];
19
+ int32 var_46_groups_0 = const()[name = string("op_46_groups_0"), val = int32(1)];
20
+ tensor<fp16, [384, 384, 3]> const_2_to_fp16 = const()[name = string("const_2_to_fp16"), val = tensor<fp16, [384, 384, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185280)))];
21
+ tensor<fp16, [384]> const_3_to_fp16 = const()[name = string("const_3_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1070080)))];
22
+ tensor<fp16, [1, 384, 1500]> var_46_cast_fp16 = conv(bias = const_3_to_fp16, dilations = var_46_dilations_0, groups = var_46_groups_0, pad = var_46_pad_0, pad_type = var_46_pad_type_0, strides = var_46_strides_0, weight = const_2_to_fp16, x = input_1_cast_fp16)[name = string("op_46_cast_fp16")];
23
+ string x_3_mode_0 = const()[name = string("x_3_mode_0"), val = string("EXACT")];
24
+ tensor<fp16, [1, 384, 1500]> x_3_cast_fp16 = gelu(mode = x_3_mode_0, x = var_46_cast_fp16)[name = string("x_3_cast_fp16")];
25
+ tensor<int32, [3]> var_52 = const()[name = string("op_52"), val = tensor<int32, [3]>([0, 2, 1])];
26
+ tensor<fp16, [1500, 384]> positional_embedding_to_fp16 = const()[name = string("positional_embedding_to_fp16"), val = tensor<fp16, [1500, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1070912)))];
27
+ tensor<fp16, [1, 1500, 384]> x_5_cast_fp16 = transpose(perm = var_52, x = x_3_cast_fp16)[name = string("transpose_52")];
28
+ tensor<fp16, [1, 1500, 384]> var_55_cast_fp16 = add(x = x_5_cast_fp16, y = positional_embedding_to_fp16)[name = string("op_55_cast_fp16")];
29
+ tensor<int32, [1]> var_82_axes_0 = const()[name = string("op_82_axes_0"), val = tensor<int32, [1]>([-1])];
30
+ tensor<fp16, [384]> blocks_0_attn_ln_weight_to_fp16 = const()[name = string("blocks_0_attn_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2222976)))];
31
+ tensor<fp16, [384]> blocks_0_attn_ln_bias_to_fp16 = const()[name = string("blocks_0_attn_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2223808)))];
32
+ fp16 var_72_to_fp16 = const()[name = string("op_72_to_fp16"), val = fp16(0x1.5p-17)];
33
+ tensor<fp16, [1, 1500, 384]> var_82_cast_fp16 = layer_norm(axes = var_82_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_72_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_55_cast_fp16)[name = string("op_82_cast_fp16")];
34
+ tensor<fp16, [384, 384]> const_4_to_fp16 = const()[name = string("const_4_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2224640)))];
35
+ tensor<fp16, [384]> const_5_to_fp16 = const()[name = string("const_5_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2519616)))];
36
+ tensor<fp16, [1, 1500, 384]> linear_0_cast_fp16 = linear(bias = const_5_to_fp16, weight = const_4_to_fp16, x = var_82_cast_fp16)[name = string("linear_0_cast_fp16")];
37
+ tensor<fp16, [384, 384]> const_6_to_fp16 = const()[name = string("const_6_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2520448)))];
38
+ tensor<fp16, [384]> linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2815424)))];
39
+ tensor<fp16, [1, 1500, 384]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_6_to_fp16, x = var_82_cast_fp16)[name = string("linear_1_cast_fp16")];
40
+ tensor<fp16, [384, 384]> const_7_to_fp16 = const()[name = string("const_7_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2816256)))];
41
+ tensor<fp16, [384]> const_8_to_fp16 = const()[name = string("const_8_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3111232)))];
42
+ tensor<fp16, [1, 1500, 384]> linear_2_cast_fp16 = linear(bias = const_8_to_fp16, weight = const_7_to_fp16, x = var_82_cast_fp16)[name = string("linear_2_cast_fp16")];
43
+ tensor<int32, [4]> var_106 = const()[name = string("op_106"), val = tensor<int32, [4]>([1, 1500, 6, -1])];
44
+ tensor<fp16, [1, 1500, 6, 64]> var_107_cast_fp16 = reshape(shape = var_106, x = linear_0_cast_fp16)[name = string("op_107_cast_fp16")];
45
+ tensor<int32, [4]> var_112 = const()[name = string("op_112"), val = tensor<int32, [4]>([1, 1500, 6, -1])];
46
+ tensor<fp16, [1, 1500, 6, 64]> var_113_cast_fp16 = reshape(shape = var_112, x = linear_1_cast_fp16)[name = string("op_113_cast_fp16")];
47
+ tensor<int32, [4]> var_118 = const()[name = string("op_118"), val = tensor<int32, [4]>([1, 1500, 6, -1])];
48
+ tensor<fp16, [1, 1500, 6, 64]> var_119_cast_fp16 = reshape(shape = var_118, x = linear_2_cast_fp16)[name = string("op_119_cast_fp16")];
49
+ tensor<int32, [4]> transpose_24_perm_0 = const()[name = string("transpose_24_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
50
+ tensor<int32, [4]> transpose_25_perm_0 = const()[name = string("transpose_25_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
51
+ tensor<int32, [4]> transpose_26_perm_0 = const()[name = string("transpose_26_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
52
+ tensor<fp16, [1, 6, 1500, 64]> transpose_26 = transpose(perm = transpose_26_perm_0, x = var_119_cast_fp16)[name = string("transpose_49")];
53
+ tensor<fp16, [1, 6, 1500, 64]> transpose_25 = transpose(perm = transpose_25_perm_0, x = var_113_cast_fp16)[name = string("transpose_50")];
54
+ tensor<fp16, [1, 6, 1500, 64]> transpose_24 = transpose(perm = transpose_24_perm_0, x = var_107_cast_fp16)[name = string("transpose_51")];
55
+ tensor<fp16, [1, 6, 1500, 64]> a_1_cast_fp16 = scaled_dot_product_attention(key = transpose_25, query = transpose_24, value = transpose_26)[name = string("a_1_cast_fp16")];
56
+ tensor<int32, [4]> var_123 = const()[name = string("op_123"), val = tensor<int32, [4]>([0, 2, 1, 3])];
57
+ tensor<int32, [3]> concat_0 = const()[name = string("concat_0"), val = tensor<int32, [3]>([1, 1500, 384])];
58
+ tensor<fp16, [1, 1500, 6, 64]> var_124_cast_fp16 = transpose(perm = var_123, x = a_1_cast_fp16)[name = string("transpose_48")];
59
+ tensor<fp16, [1, 1500, 384]> x_11_cast_fp16 = reshape(shape = concat_0, x = var_124_cast_fp16)[name = string("x_11_cast_fp16")];
60
+ tensor<fp16, [384, 384]> const_15_to_fp16 = const()[name = string("const_15_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3112064)))];
61
+ tensor<fp16, [384]> const_16_to_fp16 = const()[name = string("const_16_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3407040)))];
62
+ tensor<fp16, [1, 1500, 384]> linear_3_cast_fp16 = linear(bias = const_16_to_fp16, weight = const_15_to_fp16, x = x_11_cast_fp16)[name = string("linear_3_cast_fp16")];
63
+ tensor<fp16, [1, 1500, 384]> x_13_cast_fp16 = add(x = var_55_cast_fp16, y = linear_3_cast_fp16)[name = string("x_13_cast_fp16")];
64
+ tensor<int32, [1]> var_136_axes_0 = const()[name = string("op_136_axes_0"), val = tensor<int32, [1]>([-1])];
65
+ tensor<fp16, [384]> blocks_0_mlp_ln_weight_to_fp16 = const()[name = string("blocks_0_mlp_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3407872)))];
66
+ tensor<fp16, [384]> blocks_0_mlp_ln_bias_to_fp16 = const()[name = string("blocks_0_mlp_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3408704)))];
67
+ tensor<fp16, [1, 1500, 384]> var_136_cast_fp16 = layer_norm(axes = var_136_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_72_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast_fp16)[name = string("op_136_cast_fp16")];
68
+ tensor<fp16, [1536, 384]> const_17_to_fp16 = const()[name = string("const_17_to_fp16"), val = tensor<fp16, [1536, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3409536)))];
69
+ tensor<fp16, [1536]> const_18_to_fp16 = const()[name = string("const_18_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4589248)))];
70
+ tensor<fp16, [1, 1500, 1536]> linear_4_cast_fp16 = linear(bias = const_18_to_fp16, weight = const_17_to_fp16, x = var_136_cast_fp16)[name = string("linear_4_cast_fp16")];
71
+ string x_17_mode_0 = const()[name = string("x_17_mode_0"), val = string("EXACT")];
72
+ tensor<fp16, [1, 1500, 1536]> x_17_cast_fp16 = gelu(mode = x_17_mode_0, x = linear_4_cast_fp16)[name = string("x_17_cast_fp16")];
73
+ tensor<fp16, [384, 1536]> const_19_to_fp16 = const()[name = string("const_19_to_fp16"), val = tensor<fp16, [384, 1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4592384)))];
74
+ tensor<fp16, [384]> const_20_to_fp16 = const()[name = string("const_20_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5772096)))];
75
+ tensor<fp16, [1, 1500, 384]> linear_5_cast_fp16 = linear(bias = const_20_to_fp16, weight = const_19_to_fp16, x = x_17_cast_fp16)[name = string("linear_5_cast_fp16")];
76
+ tensor<fp16, [1, 1500, 384]> x_19_cast_fp16 = add(x = x_13_cast_fp16, y = linear_5_cast_fp16)[name = string("x_19_cast_fp16")];
77
+ tensor<int32, [1]> var_176_axes_0 = const()[name = string("op_176_axes_0"), val = tensor<int32, [1]>([-1])];
78
+ tensor<fp16, [384]> blocks_1_attn_ln_weight_to_fp16 = const()[name = string("blocks_1_attn_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5772928)))];
79
+ tensor<fp16, [384]> blocks_1_attn_ln_bias_to_fp16 = const()[name = string("blocks_1_attn_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5773760)))];
80
+ fp16 var_166_to_fp16 = const()[name = string("op_166_to_fp16"), val = fp16(0x1.5p-17)];
81
+ tensor<fp16, [1, 1500, 384]> var_176_cast_fp16 = layer_norm(axes = var_176_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_166_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast_fp16)[name = string("op_176_cast_fp16")];
82
+ tensor<fp16, [384, 384]> const_21_to_fp16 = const()[name = string("const_21_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5774592)))];
83
+ tensor<fp16, [384]> const_22_to_fp16 = const()[name = string("const_22_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6069568)))];
84
+ tensor<fp16, [1, 1500, 384]> linear_6_cast_fp16 = linear(bias = const_22_to_fp16, weight = const_21_to_fp16, x = var_176_cast_fp16)[name = string("linear_6_cast_fp16")];
85
+ tensor<fp16, [384, 384]> const_23_to_fp16 = const()[name = string("const_23_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6070400)))];
86
+ tensor<fp16, [1, 1500, 384]> linear_7_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_23_to_fp16, x = var_176_cast_fp16)[name = string("linear_7_cast_fp16")];
87
+ tensor<fp16, [384, 384]> const_24_to_fp16 = const()[name = string("const_24_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6365376)))];
88
+ tensor<fp16, [384]> const_25_to_fp16 = const()[name = string("const_25_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6660352)))];
89
+ tensor<fp16, [1, 1500, 384]> linear_8_cast_fp16 = linear(bias = const_25_to_fp16, weight = const_24_to_fp16, x = var_176_cast_fp16)[name = string("linear_8_cast_fp16")];
90
+ tensor<int32, [4]> var_200 = const()[name = string("op_200"), val = tensor<int32, [4]>([1, 1500, 6, -1])];
91
+ tensor<fp16, [1, 1500, 6, 64]> var_201_cast_fp16 = reshape(shape = var_200, x = linear_6_cast_fp16)[name = string("op_201_cast_fp16")];
92
+ tensor<int32, [4]> var_206 = const()[name = string("op_206"), val = tensor<int32, [4]>([1, 1500, 6, -1])];
93
+ tensor<fp16, [1, 1500, 6, 64]> var_207_cast_fp16 = reshape(shape = var_206, x = linear_7_cast_fp16)[name = string("op_207_cast_fp16")];
94
+ tensor<int32, [4]> var_212 = const()[name = string("op_212"), val = tensor<int32, [4]>([1, 1500, 6, -1])];
95
+ tensor<fp16, [1, 1500, 6, 64]> var_213_cast_fp16 = reshape(shape = var_212, x = linear_8_cast_fp16)[name = string("op_213_cast_fp16")];
96
+ tensor<int32, [4]> transpose_27_perm_0 = const()[name = string("transpose_27_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
97
+ tensor<int32, [4]> transpose_28_perm_0 = const()[name = string("transpose_28_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
98
+ tensor<int32, [4]> transpose_29_perm_0 = const()[name = string("transpose_29_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
99
+ tensor<fp16, [1, 6, 1500, 64]> transpose_29 = transpose(perm = transpose_29_perm_0, x = var_213_cast_fp16)[name = string("transpose_45")];
100
+ tensor<fp16, [1, 6, 1500, 64]> transpose_28 = transpose(perm = transpose_28_perm_0, x = var_207_cast_fp16)[name = string("transpose_46")];
101
+ tensor<fp16, [1, 6, 1500, 64]> transpose_27 = transpose(perm = transpose_27_perm_0, x = var_201_cast_fp16)[name = string("transpose_47")];
102
+ tensor<fp16, [1, 6, 1500, 64]> a_3_cast_fp16 = scaled_dot_product_attention(key = transpose_28, query = transpose_27, value = transpose_29)[name = string("a_3_cast_fp16")];
103
+ tensor<int32, [4]> var_217 = const()[name = string("op_217"), val = tensor<int32, [4]>([0, 2, 1, 3])];
104
+ tensor<int32, [3]> concat_1 = const()[name = string("concat_1"), val = tensor<int32, [3]>([1, 1500, 384])];
105
+ tensor<fp16, [1, 1500, 6, 64]> var_218_cast_fp16 = transpose(perm = var_217, x = a_3_cast_fp16)[name = string("transpose_44")];
106
+ tensor<fp16, [1, 1500, 384]> x_23_cast_fp16 = reshape(shape = concat_1, x = var_218_cast_fp16)[name = string("x_23_cast_fp16")];
107
+ tensor<fp16, [384, 384]> const_32_to_fp16 = const()[name = string("const_32_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6661184)))];
108
+ tensor<fp16, [384]> const_33_to_fp16 = const()[name = string("const_33_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6956160)))];
109
+ tensor<fp16, [1, 1500, 384]> linear_9_cast_fp16 = linear(bias = const_33_to_fp16, weight = const_32_to_fp16, x = x_23_cast_fp16)[name = string("linear_9_cast_fp16")];
110
+ tensor<fp16, [1, 1500, 384]> x_25_cast_fp16 = add(x = x_19_cast_fp16, y = linear_9_cast_fp16)[name = string("x_25_cast_fp16")];
111
+ tensor<int32, [1]> var_230_axes_0 = const()[name = string("op_230_axes_0"), val = tensor<int32, [1]>([-1])];
112
+ tensor<fp16, [384]> blocks_1_mlp_ln_weight_to_fp16 = const()[name = string("blocks_1_mlp_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6956992)))];
113
+ tensor<fp16, [384]> blocks_1_mlp_ln_bias_to_fp16 = const()[name = string("blocks_1_mlp_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6957824)))];
114
+ tensor<fp16, [1, 1500, 384]> var_230_cast_fp16 = layer_norm(axes = var_230_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_166_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast_fp16)[name = string("op_230_cast_fp16")];
115
+ tensor<fp16, [1536, 384]> const_34_to_fp16 = const()[name = string("const_34_to_fp16"), val = tensor<fp16, [1536, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6958656)))];
116
+ tensor<fp16, [1536]> const_35_to_fp16 = const()[name = string("const_35_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8138368)))];
117
+ tensor<fp16, [1, 1500, 1536]> linear_10_cast_fp16 = linear(bias = const_35_to_fp16, weight = const_34_to_fp16, x = var_230_cast_fp16)[name = string("linear_10_cast_fp16")];
118
+ string x_29_mode_0 = const()[name = string("x_29_mode_0"), val = string("EXACT")];
119
+ tensor<fp16, [1, 1500, 1536]> x_29_cast_fp16 = gelu(mode = x_29_mode_0, x = linear_10_cast_fp16)[name = string("x_29_cast_fp16")];
120
+ tensor<fp16, [384, 1536]> const_36_to_fp16 = const()[name = string("const_36_to_fp16"), val = tensor<fp16, [384, 1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8141504)))];
121
+ tensor<fp16, [384]> const_37_to_fp16 = const()[name = string("const_37_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9321216)))];
122
+ tensor<fp16, [1, 1500, 384]> linear_11_cast_fp16 = linear(bias = const_37_to_fp16, weight = const_36_to_fp16, x = x_29_cast_fp16)[name = string("linear_11_cast_fp16")];
123
+ tensor<fp16, [1, 1500, 384]> x_31_cast_fp16 = add(x = x_25_cast_fp16, y = linear_11_cast_fp16)[name = string("x_31_cast_fp16")];
124
+ tensor<int32, [1]> var_270_axes_0 = const()[name = string("op_270_axes_0"), val = tensor<int32, [1]>([-1])];
125
+ tensor<fp16, [384]> blocks_2_attn_ln_weight_to_fp16 = const()[name = string("blocks_2_attn_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9322048)))];
126
+ tensor<fp16, [384]> blocks_2_attn_ln_bias_to_fp16 = const()[name = string("blocks_2_attn_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9322880)))];
127
+ fp16 var_260_to_fp16 = const()[name = string("op_260_to_fp16"), val = fp16(0x1.5p-17)];
128
+ tensor<fp16, [1, 1500, 384]> var_270_cast_fp16 = layer_norm(axes = var_270_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_260_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast_fp16)[name = string("op_270_cast_fp16")];
129
+ tensor<fp16, [384, 384]> const_38_to_fp16 = const()[name = string("const_38_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9323712)))];
130
+ tensor<fp16, [384]> const_39_to_fp16 = const()[name = string("const_39_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9618688)))];
131
+ tensor<fp16, [1, 1500, 384]> linear_12_cast_fp16 = linear(bias = const_39_to_fp16, weight = const_38_to_fp16, x = var_270_cast_fp16)[name = string("linear_12_cast_fp16")];
132
+ tensor<fp16, [384, 384]> const_40_to_fp16 = const()[name = string("const_40_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9619520)))];
133
+ tensor<fp16, [1, 1500, 384]> linear_13_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_40_to_fp16, x = var_270_cast_fp16)[name = string("linear_13_cast_fp16")];
134
+ tensor<fp16, [384, 384]> const_41_to_fp16 = const()[name = string("const_41_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9914496)))];
135
+ tensor<fp16, [384]> const_42_to_fp16 = const()[name = string("const_42_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10209472)))];
136
+ tensor<fp16, [1, 1500, 384]> linear_14_cast_fp16 = linear(bias = const_42_to_fp16, weight = const_41_to_fp16, x = var_270_cast_fp16)[name = string("linear_14_cast_fp16")];
137
+ tensor<int32, [4]> var_294 = const()[name = string("op_294"), val = tensor<int32, [4]>([1, 1500, 6, -1])];
138
+ tensor<fp16, [1, 1500, 6, 64]> var_295_cast_fp16 = reshape(shape = var_294, x = linear_12_cast_fp16)[name = string("op_295_cast_fp16")];
139
+ tensor<int32, [4]> var_300 = const()[name = string("op_300"), val = tensor<int32, [4]>([1, 1500, 6, -1])];
140
+ tensor<fp16, [1, 1500, 6, 64]> var_301_cast_fp16 = reshape(shape = var_300, x = linear_13_cast_fp16)[name = string("op_301_cast_fp16")];
141
+ tensor<int32, [4]> var_306 = const()[name = string("op_306"), val = tensor<int32, [4]>([1, 1500, 6, -1])];
142
+ tensor<fp16, [1, 1500, 6, 64]> var_307_cast_fp16 = reshape(shape = var_306, x = linear_14_cast_fp16)[name = string("op_307_cast_fp16")];
143
+ tensor<int32, [4]> transpose_30_perm_0 = const()[name = string("transpose_30_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
144
+ tensor<int32, [4]> transpose_31_perm_0 = const()[name = string("transpose_31_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
145
+ tensor<int32, [4]> transpose_32_perm_0 = const()[name = string("transpose_32_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
146
+ tensor<fp16, [1, 6, 1500, 64]> transpose_32 = transpose(perm = transpose_32_perm_0, x = var_307_cast_fp16)[name = string("transpose_41")];
147
+ tensor<fp16, [1, 6, 1500, 64]> transpose_31 = transpose(perm = transpose_31_perm_0, x = var_301_cast_fp16)[name = string("transpose_42")];
148
+ tensor<fp16, [1, 6, 1500, 64]> transpose_30 = transpose(perm = transpose_30_perm_0, x = var_295_cast_fp16)[name = string("transpose_43")];
149
+ tensor<fp16, [1, 6, 1500, 64]> a_5_cast_fp16 = scaled_dot_product_attention(key = transpose_31, query = transpose_30, value = transpose_32)[name = string("a_5_cast_fp16")];
150
+ tensor<int32, [4]> var_311 = const()[name = string("op_311"), val = tensor<int32, [4]>([0, 2, 1, 3])];
151
+ tensor<int32, [3]> concat_2 = const()[name = string("concat_2"), val = tensor<int32, [3]>([1, 1500, 384])];
152
+ tensor<fp16, [1, 1500, 6, 64]> var_312_cast_fp16 = transpose(perm = var_311, x = a_5_cast_fp16)[name = string("transpose_40")];
153
+ tensor<fp16, [1, 1500, 384]> x_35_cast_fp16 = reshape(shape = concat_2, x = var_312_cast_fp16)[name = string("x_35_cast_fp16")];
154
+ tensor<fp16, [384, 384]> const_49_to_fp16 = const()[name = string("const_49_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10210304)))];
155
+ tensor<fp16, [384]> const_50_to_fp16 = const()[name = string("const_50_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10505280)))];
156
+ tensor<fp16, [1, 1500, 384]> linear_15_cast_fp16 = linear(bias = const_50_to_fp16, weight = const_49_to_fp16, x = x_35_cast_fp16)[name = string("linear_15_cast_fp16")];
157
+ tensor<fp16, [1, 1500, 384]> x_37_cast_fp16 = add(x = x_31_cast_fp16, y = linear_15_cast_fp16)[name = string("x_37_cast_fp16")];
158
+ tensor<int32, [1]> var_324_axes_0 = const()[name = string("op_324_axes_0"), val = tensor<int32, [1]>([-1])];
159
+ tensor<fp16, [384]> blocks_2_mlp_ln_weight_to_fp16 = const()[name = string("blocks_2_mlp_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10506112)))];
160
+ tensor<fp16, [384]> blocks_2_mlp_ln_bias_to_fp16 = const()[name = string("blocks_2_mlp_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10506944)))];
161
+ tensor<fp16, [1, 1500, 384]> var_324_cast_fp16 = layer_norm(axes = var_324_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_260_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast_fp16)[name = string("op_324_cast_fp16")];
162
+ tensor<fp16, [1536, 384]> const_51_to_fp16 = const()[name = string("const_51_to_fp16"), val = tensor<fp16, [1536, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10507776)))];
163
+ tensor<fp16, [1536]> const_52_to_fp16 = const()[name = string("const_52_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11687488)))];
164
+ tensor<fp16, [1, 1500, 1536]> linear_16_cast_fp16 = linear(bias = const_52_to_fp16, weight = const_51_to_fp16, x = var_324_cast_fp16)[name = string("linear_16_cast_fp16")];
165
+ string x_41_mode_0 = const()[name = string("x_41_mode_0"), val = string("EXACT")];
166
+ tensor<fp16, [1, 1500, 1536]> x_41_cast_fp16 = gelu(mode = x_41_mode_0, x = linear_16_cast_fp16)[name = string("x_41_cast_fp16")];
167
+ tensor<fp16, [384, 1536]> const_53_to_fp16 = const()[name = string("const_53_to_fp16"), val = tensor<fp16, [384, 1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11690624)))];
168
+ tensor<fp16, [384]> const_54_to_fp16 = const()[name = string("const_54_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12870336)))];
169
+ tensor<fp16, [1, 1500, 384]> linear_17_cast_fp16 = linear(bias = const_54_to_fp16, weight = const_53_to_fp16, x = x_41_cast_fp16)[name = string("linear_17_cast_fp16")];
170
+ tensor<fp16, [1, 1500, 384]> x_43_cast_fp16 = add(x = x_37_cast_fp16, y = linear_17_cast_fp16)[name = string("x_43_cast_fp16")];
171
+ tensor<int32, [1]> var_364_axes_0 = const()[name = string("op_364_axes_0"), val = tensor<int32, [1]>([-1])];
172
+ tensor<fp16, [384]> blocks_3_attn_ln_weight_to_fp16 = const()[name = string("blocks_3_attn_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12871168)))];
173
+ tensor<fp16, [384]> blocks_3_attn_ln_bias_to_fp16 = const()[name = string("blocks_3_attn_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12872000)))];
174
+ fp16 var_354_to_fp16 = const()[name = string("op_354_to_fp16"), val = fp16(0x1.5p-17)];
175
+ tensor<fp16, [1, 1500, 384]> var_364_cast_fp16 = layer_norm(axes = var_364_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_354_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast_fp16)[name = string("op_364_cast_fp16")];
176
+ tensor<fp16, [384, 384]> const_55_to_fp16 = const()[name = string("const_55_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12872832)))];
177
+ tensor<fp16, [384]> const_56_to_fp16 = const()[name = string("const_56_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13167808)))];
178
+ tensor<fp16, [1, 1500, 384]> linear_18_cast_fp16 = linear(bias = const_56_to_fp16, weight = const_55_to_fp16, x = var_364_cast_fp16)[name = string("linear_18_cast_fp16")];
179
+ tensor<fp16, [384, 384]> const_57_to_fp16 = const()[name = string("const_57_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13168640)))];
180
+ tensor<fp16, [1, 1500, 384]> linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_57_to_fp16, x = var_364_cast_fp16)[name = string("linear_19_cast_fp16")];
181
+ tensor<fp16, [384, 384]> const_58_to_fp16 = const()[name = string("const_58_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13463616)))];
182
+ tensor<fp16, [384]> const_59_to_fp16 = const()[name = string("const_59_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13758592)))];
183
+ tensor<fp16, [1, 1500, 384]> linear_20_cast_fp16 = linear(bias = const_59_to_fp16, weight = const_58_to_fp16, x = var_364_cast_fp16)[name = string("linear_20_cast_fp16")];
184
+ tensor<int32, [4]> var_388 = const()[name = string("op_388"), val = tensor<int32, [4]>([1, 1500, 6, -1])];
185
+ tensor<fp16, [1, 1500, 6, 64]> var_389_cast_fp16 = reshape(shape = var_388, x = linear_18_cast_fp16)[name = string("op_389_cast_fp16")];
186
+ tensor<int32, [4]> var_394 = const()[name = string("op_394"), val = tensor<int32, [4]>([1, 1500, 6, -1])];
187
+ tensor<fp16, [1, 1500, 6, 64]> var_395_cast_fp16 = reshape(shape = var_394, x = linear_19_cast_fp16)[name = string("op_395_cast_fp16")];
188
+ tensor<int32, [4]> var_400 = const()[name = string("op_400"), val = tensor<int32, [4]>([1, 1500, 6, -1])];
189
+ tensor<fp16, [1, 1500, 6, 64]> var_401_cast_fp16 = reshape(shape = var_400, x = linear_20_cast_fp16)[name = string("op_401_cast_fp16")];
190
+ tensor<int32, [4]> transpose_33_perm_0 = const()[name = string("transpose_33_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
191
+ tensor<int32, [4]> transpose_34_perm_0 = const()[name = string("transpose_34_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
192
+ tensor<int32, [4]> transpose_35_perm_0 = const()[name = string("transpose_35_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
193
+ tensor<fp16, [1, 6, 1500, 64]> transpose_35 = transpose(perm = transpose_35_perm_0, x = var_401_cast_fp16)[name = string("transpose_37")];
194
+ tensor<fp16, [1, 6, 1500, 64]> transpose_34 = transpose(perm = transpose_34_perm_0, x = var_395_cast_fp16)[name = string("transpose_38")];
195
+ tensor<fp16, [1, 6, 1500, 64]> transpose_33 = transpose(perm = transpose_33_perm_0, x = var_389_cast_fp16)[name = string("transpose_39")];
196
+ tensor<fp16, [1, 6, 1500, 64]> a_cast_fp16 = scaled_dot_product_attention(key = transpose_34, query = transpose_33, value = transpose_35)[name = string("a_cast_fp16")];
197
+ tensor<int32, [4]> var_405 = const()[name = string("op_405"), val = tensor<int32, [4]>([0, 2, 1, 3])];
198
+ tensor<int32, [3]> concat_3 = const()[name = string("concat_3"), val = tensor<int32, [3]>([1, 1500, 384])];
199
+ tensor<fp16, [1, 1500, 6, 64]> var_406_cast_fp16 = transpose(perm = var_405, x = a_cast_fp16)[name = string("transpose_36")];
200
+ tensor<fp16, [1, 1500, 384]> x_47_cast_fp16 = reshape(shape = concat_3, x = var_406_cast_fp16)[name = string("x_47_cast_fp16")];
201
+ tensor<fp16, [384, 384]> const_66_to_fp16 = const()[name = string("const_66_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13759424)))];
202
+ tensor<fp16, [384]> const_67_to_fp16 = const()[name = string("const_67_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14054400)))];
203
+ tensor<fp16, [1, 1500, 384]> linear_21_cast_fp16 = linear(bias = const_67_to_fp16, weight = const_66_to_fp16, x = x_47_cast_fp16)[name = string("linear_21_cast_fp16")];
204
+ tensor<fp16, [1, 1500, 384]> x_49_cast_fp16 = add(x = x_43_cast_fp16, y = linear_21_cast_fp16)[name = string("x_49_cast_fp16")];
205
+ tensor<int32, [1]> var_418_axes_0 = const()[name = string("op_418_axes_0"), val = tensor<int32, [1]>([-1])];
206
+ tensor<fp16, [384]> blocks_3_mlp_ln_weight_to_fp16 = const()[name = string("blocks_3_mlp_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14055232)))];
207
+ tensor<fp16, [384]> blocks_3_mlp_ln_bias_to_fp16 = const()[name = string("blocks_3_mlp_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14056064)))];
208
+ tensor<fp16, [1, 1500, 384]> var_418_cast_fp16 = layer_norm(axes = var_418_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_354_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast_fp16)[name = string("op_418_cast_fp16")];
209
+ tensor<fp16, [1536, 384]> const_68_to_fp16 = const()[name = string("const_68_to_fp16"), val = tensor<fp16, [1536, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14056896)))];
210
+ tensor<fp16, [1536]> const_69_to_fp16 = const()[name = string("const_69_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15236608)))];
211
+ tensor<fp16, [1, 1500, 1536]> linear_22_cast_fp16 = linear(bias = const_69_to_fp16, weight = const_68_to_fp16, x = var_418_cast_fp16)[name = string("linear_22_cast_fp16")];
212
+ string x_53_mode_0 = const()[name = string("x_53_mode_0"), val = string("EXACT")];
213
+ tensor<fp16, [1, 1500, 1536]> x_53_cast_fp16 = gelu(mode = x_53_mode_0, x = linear_22_cast_fp16)[name = string("x_53_cast_fp16")];
214
+ tensor<fp16, [384, 1536]> const_70_to_fp16 = const()[name = string("const_70_to_fp16"), val = tensor<fp16, [384, 1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15239744)))];
215
+ tensor<fp16, [384]> const_71_to_fp16 = const()[name = string("const_71_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16419456)))];
216
+ tensor<fp16, [1, 1500, 384]> linear_23_cast_fp16 = linear(bias = const_71_to_fp16, weight = const_70_to_fp16, x = x_53_cast_fp16)[name = string("linear_23_cast_fp16")];
217
+ tensor<fp16, [1, 1500, 384]> x_cast_fp16 = add(x = x_49_cast_fp16, y = linear_23_cast_fp16)[name = string("x_cast_fp16")];
218
+ tensor<int32, [1]> var_447_axes_0 = const()[name = string("op_447_axes_0"), val = tensor<int32, [1]>([-1])];
219
+ tensor<fp16, [384]> ln_post_weight_to_fp16 = const()[name = string("ln_post_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16420288)))];
220
+ tensor<fp16, [384]> ln_post_bias_to_fp16 = const()[name = string("ln_post_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16421120)))];
221
+ fp16 var_438_to_fp16 = const()[name = string("op_438_to_fp16"), val = fp16(0x1.5p-17)];
222
+ tensor<fp16, [1, 1500, 384]> output = layer_norm(axes = var_447_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_438_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast_fp16)[name = string("op_447_cast_fp16")];
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
223
  } -> (output);
224
  }
ggml-tiny-encoder.mlmodelc/weights/weight.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:cb82c7078e49e07a517e598cce7e3b6dded7397efc495be1992ef822570284a4
3
  size 16421952
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4efa9bb81afaf12ac6d7cf7a3a4ba1e6b92f05f96ae77fd55cf725e2ecd3a5fd
3
  size 16421952
ggml-large-v2-encoder.mlmodelc/analytics/coremldata.bin → ggml-tiny-q8_0.bin RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ad0b061176b791284f765ca1e72c9c7fed728ef6cf0af875a51c93fd1a8e525f
3
- size 243
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:98bfdeb46504a115c4c0dcbaf14713b003152b2ce23f14f639d7f6d39dee043a
3
+ size 32561100
ggml-tiny.bin DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:c7e4b084cefeebeed66fb9d096a29b836125edbb8456fea5a9c77b4efc085323
3
- size 60079860
 
 
 
 
index/base CHANGED
@@ -3,4 +3,4 @@ ggml-base-encoder.mlmodelc/metadata.json
3
  ggml-base-encoder.mlmodelc/model.mil
4
  ggml-base-encoder.mlmodelc/coremldata.bin
5
  ggml-base-encoder.mlmodelc/analytics/coremldata.bin
6
- ggml-base.bin
 
3
  ggml-base-encoder.mlmodelc/model.mil
4
  ggml-base-encoder.mlmodelc/coremldata.bin
5
  ggml-base-encoder.mlmodelc/analytics/coremldata.bin
6
+ ggml-base-q8_0.bin
index/large-v2 DELETED
@@ -1,6 +0,0 @@
1
- ggml-large-v2-encoder.mlmodelc/weights/weight.bin
2
- ggml-large-v2-encoder.mlmodelc/metadata.json
3
- ggml-large-v2-encoder.mlmodelc/model.mil
4
- ggml-large-v2-encoder.mlmodelc/coremldata.bin
5
- ggml-large-v2-encoder.mlmodelc/analytics/coremldata.bin
6
- ggml-large-v2-q8_0.bin
 
 
 
 
 
 
 
index/large-v3-turbo CHANGED
@@ -1,12 +1,6 @@
 
1
  ggml-large-v3-turbo-encoder.mlmodelc/metadata.json
2
- ggml-large-v3-turbo-encoder.mlmodelc/model0/weights/0-weight.bin
3
- ggml-large-v3-turbo-encoder.mlmodelc/model0/model.mil
4
- ggml-large-v3-turbo-encoder.mlmodelc/model0/coremldata.bin
5
- ggml-large-v3-turbo-encoder.mlmodelc/model0/analytics/coremldata.bin
6
- ggml-large-v3-turbo-encoder.mlmodelc/model1/weights/1-weight.bin
7
- ggml-large-v3-turbo-encoder.mlmodelc/model1/model.mil
8
- ggml-large-v3-turbo-encoder.mlmodelc/model1/coremldata.bin
9
- ggml-large-v3-turbo-encoder.mlmodelc/model1/analytics/coremldata.bin
10
  ggml-large-v3-turbo-encoder.mlmodelc/coremldata.bin
11
  ggml-large-v3-turbo-encoder.mlmodelc/analytics/coremldata.bin
12
  ggml-large-v3-turbo-q8_0.bin
 
1
+ ggml-large-v3-turbo-encoder.mlmodelc/weights/weight.bin
2
  ggml-large-v3-turbo-encoder.mlmodelc/metadata.json
3
+ ggml-large-v3-turbo-encoder.mlmodelc/model.mil
 
 
 
 
 
 
 
4
  ggml-large-v3-turbo-encoder.mlmodelc/coremldata.bin
5
  ggml-large-v3-turbo-encoder.mlmodelc/analytics/coremldata.bin
6
  ggml-large-v3-turbo-q8_0.bin