chua commited on
Commit
8d8aaf9
·
verified ·
1 Parent(s): 64f7370

Add ASR model bundle metadata and small assets

Browse files
dist/ASRModels.bundle/asr_manifest.json ADDED
@@ -0,0 +1,79 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "mel": {
3
+ "n_fft": 400,
4
+ "hop": 160,
5
+ "n_mels": 128,
6
+ "sample_rate": 16000,
7
+ "frames_padded": 3000,
8
+ "max_samples": 480000
9
+ },
10
+ "mel_filters_shape": [
11
+ 201,
12
+ 128
13
+ ],
14
+ "projector": {
15
+ "in": 1024,
16
+ "out": 512
17
+ },
18
+ "embedding": {
19
+ "vocab": 151936,
20
+ "dim": 512,
21
+ "dtype": "float16"
22
+ },
23
+ "lm": {
24
+ "num_layers": 8,
25
+ "num_heads": 8,
26
+ "head_dim": 64,
27
+ "hidden": 512,
28
+ "max_total_len": 512
29
+ },
30
+ "tokens": {
31
+ "user_token": "<|user|>",
32
+ "assistant_token": "<|assistant|>",
33
+ "bos_audio_token": "<|begin_of_audio|>",
34
+ "audio_token": "<|audio|>",
35
+ "eos_audio_token": "<|end_of_audio|>",
36
+ "assistant_end_token": "<|im_end|>",
37
+ "audio_token_id": 151646,
38
+ "pad_token_id": 151643,
39
+ "eos_token_ids": [
40
+ 151645
41
+ ],
42
+ "asr_block_token_id_from": -1,
43
+ "extra_block_token_ids": [
44
+ 151647,
45
+ 151650,
46
+ 151648,
47
+ 151646,
48
+ 151649
49
+ ],
50
+ "vocab_size": 151652
51
+ },
52
+ "prompt": {
53
+ "template": "{user}{bos_audio}{audio}{eos_audio}Please transcribe this audio.{assistant}",
54
+ "merge_factor": 4
55
+ },
56
+ "generation": {
57
+ "max_new_tokens": 128,
58
+ "repetition_penalty": 1.0,
59
+ "greedy": true
60
+ },
61
+ "prompt_ids": {
62
+ "user": 151647,
63
+ "bos_audio": 151648,
64
+ "audio": 151646,
65
+ "eos_audio": 151649,
66
+ "text": [
67
+ 5501,
68
+ 1356,
69
+ 3114,
70
+ 419,
71
+ 7699,
72
+ 13
73
+ ],
74
+ "assistant": 151650
75
+ },
76
+ "vocab_table": {
77
+ "size": 151652
78
+ }
79
+ }
dist/ASRModels.bundle/hann_window.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2b66ad855bb20e73e657a4df6d8f4a4c840a18431c35cf323f29143f04796c7e
3
+ size 1600
dist/ASRModels.bundle/lm_decode.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ce33ce3f1e0e0992b71de0c567fd572a1be5f2c384749150114607938b813f1d
3
+ size 163575
dist/ASRModels.bundle/lm_prefill.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:79d2ff37c7dbbf8a466f4058aededa0452b899f85a7ce5840a95579d5b1005ab
3
+ size 162606
dist/ASRModels.bundle/manifest.json ADDED
@@ -0,0 +1,85 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "schemaVersion": 2,
3
+ "bundleVersion": "1.0.0",
4
+ "assets": {
5
+ "tower": {
6
+ "file": "tower.mlmodelc",
7
+ "sha256": "29d6ff275182f31eb51bfb0ac782c2d119c87c1277aa7ee59178979b67895127",
8
+ "sizeBytes": 221403139
9
+ },
10
+ "lm_prefill": {
11
+ "file": "lm_prefill.onnx",
12
+ "sha256": "79d2ff37c7dbbf8a466f4058aededa0452b899f85a7ce5840a95579d5b1005ab",
13
+ "sizeBytes": 162606
14
+ },
15
+ "lm_decode": {
16
+ "file": "lm_decode.onnx",
17
+ "sha256": "ce33ce3f1e0e0992b71de0c567fd572a1be5f2c384749150114607938b813f1d",
18
+ "sizeBytes": 163575
19
+ },
20
+ "lm_shared_data": {
21
+ "file": "lm_shared_int4.data",
22
+ "sha256": "f7ffe71cc716a7b2db49fd9a39cca9b714b5d30680d93760003c0b63109bf8e8",
23
+ "sizeBytes": 55464704
24
+ },
25
+ "mask_gen": {
26
+ "file": "mask_gen.onnx",
27
+ "sha256": "bae8051fa484580e5e5828a851ebed6f45877a0348b1ef225667eb9e99e26bd4",
28
+ "sizeBytes": 166605
29
+ },
30
+ "asr_manifest": {
31
+ "file": "asr_manifest.json",
32
+ "sha256": "bae102bbab352f3a550c02bfa7bb1356c23e31a5b20af1f94989f202489edcac",
33
+ "sizeBytes": 1476
34
+ },
35
+ "hann_window": {
36
+ "file": "hann_window.bin",
37
+ "sha256": "2b66ad855bb20e73e657a4df6d8f4a4c840a18431c35cf323f29143f04796c7e",
38
+ "sizeBytes": 1600
39
+ },
40
+ "mel_filters": {
41
+ "file": "mel_filters.bin",
42
+ "sha256": "a3bdc67fe03fb9464ae3dd00533bb422f7530fa6eaf5bdda4faf72c156aad17b",
43
+ "sizeBytes": 102912
44
+ },
45
+ "projector_norm_w": {
46
+ "file": "projector_norm_w.bin",
47
+ "sha256": "41580bc0d9f389f58b0fed9e287851ef97f7c089f2670655ac756482dfaa5932",
48
+ "sizeBytes": 4096
49
+ },
50
+ "projector_norm_b": {
51
+ "file": "projector_norm_b.bin",
52
+ "sha256": "a1cf72db77e96997f775303a9a284d73b379f815a2ea02942dd1bf6603f3fa3a",
53
+ "sizeBytes": 4096
54
+ },
55
+ "projector_lin_w": {
56
+ "file": "projector_lin_w.bin",
57
+ "sha256": "c94c5e7aac7ee5ef891fcfc8acae04f2fbbba9cd77cad2fd877d3611e453f16a",
58
+ "sizeBytes": 2097152
59
+ },
60
+ "projector_lin_b": {
61
+ "file": "projector_lin_b.bin",
62
+ "sha256": "d954ed1c7800f115ebcfcb21930f7db8c46ac4dea2bd935be4f55ba58478c9ec",
63
+ "sizeBytes": 2048
64
+ },
65
+ "token_embedding": {
66
+ "file": "token_embedding_fp16.bin",
67
+ "sha256": "15aef975b5283283de86ed009750ac73e0314b918c7aed4ed455bb24e7f6bbcb",
68
+ "sizeBytes": 155582464
69
+ },
70
+ "vocab_bytes": {
71
+ "file": "vocab_bytes.bin",
72
+ "sha256": "047830a7230b3609a4afce91f1e10cd0c488f7e908ca381e4435c743c10850c5",
73
+ "sizeBytes": 976078
74
+ },
75
+ "vocab_offsets": {
76
+ "file": "vocab_offsets.bin",
77
+ "sha256": "af2cc5f3e54f2eaa69d21636bb696cc99c55d10aeaf31280f3b5a1c622475a52",
78
+ "sizeBytes": 606612
79
+ }
80
+ },
81
+ "extra": {
82
+ "model": "Audio8-ASR-0.1B",
83
+ "builtBy": "make_asset_bundle.py"
84
+ }
85
+ }
dist/ASRModels.bundle/mask_gen.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bae8051fa484580e5e5828a851ebed6f45877a0348b1ef225667eb9e99e26bd4
3
+ size 166605
dist/ASRModels.bundle/mel_filters.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a3bdc67fe03fb9464ae3dd00533bb422f7530fa6eaf5bdda4faf72c156aad17b
3
+ size 102912
dist/ASRModels.bundle/projector_lin_b.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d954ed1c7800f115ebcfcb21930f7db8c46ac4dea2bd935be4f55ba58478c9ec
3
+ size 2048
dist/ASRModels.bundle/projector_lin_w.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c94c5e7aac7ee5ef891fcfc8acae04f2fbbba9cd77cad2fd877d3611e453f16a
3
+ size 2097152
dist/ASRModels.bundle/projector_norm_b.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a1cf72db77e96997f775303a9a284d73b379f815a2ea02942dd1bf6603f3fa3a
3
+ size 4096
dist/ASRModels.bundle/projector_norm_w.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:41580bc0d9f389f58b0fed9e287851ef97f7c089f2670655ac756482dfaa5932
3
+ size 4096
dist/ASRModels.bundle/tower.mlmodelc/analytics/coremldata.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a6b986d3cad99739964a83157016239050e93490db6675ce3fa3b5b1f70d438b
3
+ size 243
dist/ASRModels.bundle/tower.mlmodelc/coremldata.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:85b64cc1cba68b025f7019d4a455486f1f9b3178db8f6ff8c632cc19d4a37c32
3
+ size 418
dist/ASRModels.bundle/tower.mlmodelc/metadata.json ADDED
@@ -0,0 +1,262 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [
2
+ {
3
+ "metadataOutputVersion" : "3.0",
4
+ "storagePrecision" : "Mixed (Float16, Int8)",
5
+ "outputSchema" : [
6
+ {
7
+ "hasShapeFlexibility" : "0",
8
+ "isOptional" : "0",
9
+ "dataType" : "Float16",
10
+ "formattedType" : "MultiArray (Float16 390 × 1024)",
11
+ "shortDescription" : "",
12
+ "shape" : "[390, 1024]",
13
+ "name" : "hidden",
14
+ "type" : "MultiArray"
15
+ }
16
+ ],
17
+ "modelParameters" : [
18
+
19
+ ],
20
+ "specificationVersion" : 9,
21
+ "functions" : [
22
+ {
23
+ "inputSchema" : [
24
+ {
25
+ "hasShapeFlexibility" : "0",
26
+ "isOptional" : "0",
27
+ "dataType" : "Float32",
28
+ "formattedType" : "MultiArray (Float32 1 × 128 × 500)",
29
+ "shortDescription" : "",
30
+ "shape" : "[1, 128, 500]",
31
+ "name" : "audios",
32
+ "type" : "MultiArray"
33
+ },
34
+ {
35
+ "hasShapeFlexibility" : "0",
36
+ "isOptional" : "0",
37
+ "dataType" : "Float32",
38
+ "formattedType" : "MultiArray (Float32 1 × 1 × 65 × 65)",
39
+ "shortDescription" : "",
40
+ "shape" : "[1, 1, 65, 65]",
41
+ "name" : "attn_mask",
42
+ "type" : "MultiArray"
43
+ }
44
+ ],
45
+ "computePrecision" : "Mixed (Float16, Int32)",
46
+ "storagePrecision" : "Mixed (Float16, Int8)",
47
+ "stateSchema" : [
48
+
49
+ ],
50
+ "outputSchema" : [
51
+ {
52
+ "hasShapeFlexibility" : "0",
53
+ "isOptional" : "0",
54
+ "dataType" : "Float16",
55
+ "formattedType" : "MultiArray (Float16 65 × 1024)",
56
+ "shortDescription" : "",
57
+ "shape" : "[65, 1024]",
58
+ "name" : "hidden",
59
+ "type" : "MultiArray"
60
+ }
61
+ ],
62
+ "name" : "tower_5s",
63
+ "mlProgramOperationTypeHistogram" : {
64
+ "Ios18.linear" : 119,
65
+ "Ios18.conv" : 3,
66
+ "Ios18.expandDims" : 55,
67
+ "Ios18.matmul" : 36,
68
+ "Ios18.gelu" : 26,
69
+ "Ios16.constexprAffineDequantize" : 123,
70
+ "Ios18.add" : 59,
71
+ "Ios18.softmax" : 18,
72
+ "Ios18.sliceByIndex" : 1,
73
+ "Ios18.layerNorm" : 42,
74
+ "Ios18.transpose" : 75,
75
+ "Ios18.cast" : 2,
76
+ "Ios18.squeeze" : 18,
77
+ "Ios18.reshape" : 75,
78
+ "Ios18.mul" : 18
79
+ }
80
+ },
81
+ {
82
+ "inputSchema" : [
83
+ {
84
+ "hasShapeFlexibility" : "0",
85
+ "isOptional" : "0",
86
+ "dataType" : "Float32",
87
+ "formattedType" : "MultiArray (Float32 1 × 128 × 1000)",
88
+ "shortDescription" : "",
89
+ "shape" : "[1, 128, 1000]",
90
+ "name" : "audios",
91
+ "type" : "MultiArray"
92
+ },
93
+ {
94
+ "hasShapeFlexibility" : "0",
95
+ "isOptional" : "0",
96
+ "dataType" : "Float32",
97
+ "formattedType" : "MultiArray (Float32 1 × 1 × 130 × 130)",
98
+ "shortDescription" : "",
99
+ "shape" : "[1, 1, 130, 130]",
100
+ "name" : "attn_mask",
101
+ "type" : "MultiArray"
102
+ }
103
+ ],
104
+ "computePrecision" : "Mixed (Float16, Int32)",
105
+ "storagePrecision" : "Mixed (Float16, Int8)",
106
+ "stateSchema" : [
107
+
108
+ ],
109
+ "outputSchema" : [
110
+ {
111
+ "hasShapeFlexibility" : "0",
112
+ "isOptional" : "0",
113
+ "dataType" : "Float16",
114
+ "formattedType" : "MultiArray (Float16 130 × 1024)",
115
+ "shortDescription" : "",
116
+ "shape" : "[130, 1024]",
117
+ "name" : "hidden",
118
+ "type" : "MultiArray"
119
+ }
120
+ ],
121
+ "name" : "tower_10s",
122
+ "mlProgramOperationTypeHistogram" : {
123
+ "Ios18.linear" : 119,
124
+ "Ios18.conv" : 3,
125
+ "Ios18.expandDims" : 55,
126
+ "Ios18.matmul" : 36,
127
+ "Ios18.gelu" : 26,
128
+ "Ios16.constexprAffineDequantize" : 123,
129
+ "Ios18.add" : 59,
130
+ "Ios18.softmax" : 18,
131
+ "Ios18.sliceByIndex" : 1,
132
+ "Ios18.layerNorm" : 42,
133
+ "Ios18.transpose" : 75,
134
+ "Ios18.cast" : 2,
135
+ "Ios18.squeeze" : 18,
136
+ "Ios18.reshape" : 75,
137
+ "Ios18.mul" : 18
138
+ }
139
+ },
140
+ {
141
+ "inputSchema" : [
142
+ {
143
+ "hasShapeFlexibility" : "0",
144
+ "isOptional" : "0",
145
+ "dataType" : "Float32",
146
+ "formattedType" : "MultiArray (Float32 1 × 128 × 3000)",
147
+ "shortDescription" : "",
148
+ "shape" : "[1, 128, 3000]",
149
+ "name" : "audios",
150
+ "type" : "MultiArray"
151
+ },
152
+ {
153
+ "hasShapeFlexibility" : "0",
154
+ "isOptional" : "0",
155
+ "dataType" : "Float32",
156
+ "formattedType" : "MultiArray (Float32 1 × 1 × 390 × 390)",
157
+ "shortDescription" : "",
158
+ "shape" : "[1, 1, 390, 390]",
159
+ "name" : "attn_mask",
160
+ "type" : "MultiArray"
161
+ }
162
+ ],
163
+ "computePrecision" : "Mixed (Float16, Int32)",
164
+ "storagePrecision" : "Mixed (Float16, Int8)",
165
+ "stateSchema" : [
166
+
167
+ ],
168
+ "outputSchema" : [
169
+ {
170
+ "hasShapeFlexibility" : "0",
171
+ "isOptional" : "0",
172
+ "dataType" : "Float16",
173
+ "formattedType" : "MultiArray (Float16 390 × 1024)",
174
+ "shortDescription" : "",
175
+ "shape" : "[390, 1024]",
176
+ "name" : "hidden",
177
+ "type" : "MultiArray"
178
+ }
179
+ ],
180
+ "name" : "tower_30s",
181
+ "mlProgramOperationTypeHistogram" : {
182
+ "Ios18.linear" : 119,
183
+ "Ios18.conv" : 3,
184
+ "Ios18.expandDims" : 55,
185
+ "Ios18.matmul" : 36,
186
+ "Ios18.gelu" : 26,
187
+ "Ios16.constexprAffineDequantize" : 123,
188
+ "Ios18.add" : 59,
189
+ "Ios18.softmax" : 18,
190
+ "Ios18.sliceByIndex" : 1,
191
+ "Ios18.layerNorm" : 42,
192
+ "Ios18.transpose" : 75,
193
+ "Ios18.cast" : 2,
194
+ "Ios18.squeeze" : 18,
195
+ "Ios18.reshape" : 75,
196
+ "Ios18.mul" : 18
197
+ }
198
+ }
199
+ ],
200
+ "mlProgramOperationTypeHistogram" : {
201
+ "Ios18.linear" : 119,
202
+ "Ios18.conv" : 3,
203
+ "Ios18.expandDims" : 55,
204
+ "Ios18.matmul" : 36,
205
+ "Ios18.gelu" : 26,
206
+ "Ios16.constexprAffineDequantize" : 123,
207
+ "Ios18.add" : 59,
208
+ "Ios18.softmax" : 18,
209
+ "Ios18.sliceByIndex" : 1,
210
+ "Ios18.layerNorm" : 42,
211
+ "Ios18.transpose" : 75,
212
+ "Ios18.cast" : 2,
213
+ "Ios18.squeeze" : 18,
214
+ "Ios18.reshape" : 75,
215
+ "Ios18.mul" : 18
216
+ },
217
+ "isUpdatable" : "0",
218
+ "stateSchema" : [
219
+
220
+ ],
221
+ "availability" : {
222
+ "macOS" : "15.0",
223
+ "tvOS" : "18.0",
224
+ "visionOS" : "2.0",
225
+ "watchOS" : "11.0",
226
+ "iOS" : "18.0",
227
+ "macCatalyst" : "18.0"
228
+ },
229
+ "computePrecision" : "Mixed (Float16, Int32)",
230
+ "modelType" : {
231
+ "name" : "MLModelType_mlProgram"
232
+ },
233
+ "inputSchema" : [
234
+ {
235
+ "hasShapeFlexibility" : "0",
236
+ "isOptional" : "0",
237
+ "dataType" : "Float32",
238
+ "formattedType" : "MultiArray (Float32 1 × 128 × 3000)",
239
+ "shortDescription" : "",
240
+ "shape" : "[1, 128, 3000]",
241
+ "name" : "audios",
242
+ "type" : "MultiArray"
243
+ },
244
+ {
245
+ "hasShapeFlexibility" : "0",
246
+ "isOptional" : "0",
247
+ "dataType" : "Float32",
248
+ "formattedType" : "MultiArray (Float32 1 × 1 × 390 × 390)",
249
+ "shortDescription" : "",
250
+ "shape" : "[1, 1, 390, 390]",
251
+ "name" : "attn_mask",
252
+ "type" : "MultiArray"
253
+ }
254
+ ],
255
+ "defaultFunctionName" : "tower_30s",
256
+ "generatedClassName" : "audio_tower_multifunction_w8",
257
+ "userDefinedMetadata" : {
258
+
259
+ },
260
+ "method" : "predict"
261
+ }
262
+ ]
dist/ASRModels.bundle/tower.mlmodelc/model.mil ADDED
The diff for this file is too large to render. See raw diff
 
dist/ASRModels.bundle/vocab_bytes.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:047830a7230b3609a4afce91f1e10cd0c488f7e908ca381e4435c743c10850c5
3
+ size 976078
dist/ASRModels.bundle/vocab_offsets.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:af2cc5f3e54f2eaa69d21636bb696cc99c55d10aeaf31280f3b5a1c622475a52
3
+ size 606612