Hugging Face's logo

NexaAI
/
Granite-4-Micro-ANE

Text Generation
Model card Files
xet
Community
nexaml commited on
Commit
dc492e4
·
verified ·
1 Parent(s): de699ac

Upload 65 files

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. granite-3B-part0-ar1-32.mlmodelc/analytics/coremldata.bin +3 -0
  2. granite-3B-part0-ar1-32.mlmodelc/coremldata.bin +3 -0
  3. granite-3B-part0-ar1-32.mlmodelc/metadata.json +350 -0
  4. granite-3B-part0-ar1-32.mlmodelc/model.mil +0 -0
  5. granite-3B-part0-ar1-32.mlmodelc/weights/weight.bin +3 -0
  6. granite-3B-part1-ar1-32.mlmodelc/analytics/coremldata.bin +3 -0
  7. granite-3B-part1-ar1-32.mlmodelc/coremldata.bin +3 -0
  8. granite-3B-part1-ar1-32.mlmodelc/metadata.json +350 -0
  9. granite-3B-part1-ar1-32.mlmodelc/model.mil +0 -0
  10. granite-3B-part1-ar1-32.mlmodelc/weights/weight.bin +3 -0
  11. granite-3B-part2-ar1-32.mlmodelc/analytics/coremldata.bin +3 -0
  12. granite-3B-part2-ar1-32.mlmodelc/coremldata.bin +3 -0
  13. granite-3B-part2-ar1-32.mlmodelc/metadata.json +350 -0
  14. granite-3B-part2-ar1-32.mlmodelc/model.mil +0 -0
  15. granite-3B-part2-ar1-32.mlmodelc/weights/weight.bin +3 -0
  16. granite-3B-part3-ar1-32.mlmodelc/analytics/coremldata.bin +3 -0
  17. granite-3B-part3-ar1-32.mlmodelc/coremldata.bin +3 -0
  18. granite-3B-part3-ar1-32.mlmodelc/metadata.json +350 -0
  19. granite-3B-part3-ar1-32.mlmodelc/model.mil +0 -0
  20. granite-3B-part3-ar1-32.mlmodelc/weights/weight.bin +3 -0
  21. granite-3B-part4-ar1-32.mlmodelc/analytics/coremldata.bin +3 -0
  22. granite-3B-part4-ar1-32.mlmodelc/coremldata.bin +3 -0
  23. granite-3B-part4-ar1-32.mlmodelc/metadata.json +350 -0
  24. granite-3B-part4-ar1-32.mlmodelc/model.mil +0 -0
  25. granite-3B-part4-ar1-32.mlmodelc/weights/weight.bin +3 -0
  26. granite-3B-part5-ar1-32.mlmodelc/analytics/coremldata.bin +3 -0
  27. granite-3B-part5-ar1-32.mlmodelc/coremldata.bin +3 -0
  28. granite-3B-part5-ar1-32.mlmodelc/metadata.json +350 -0
  29. granite-3B-part5-ar1-32.mlmodelc/model.mil +0 -0
  30. granite-3B-part5-ar1-32.mlmodelc/weights/weight.bin +3 -0
  31. granite-3B-part6-ar1-32.mlmodelc/analytics/coremldata.bin +3 -0
  32. granite-3B-part6-ar1-32.mlmodelc/coremldata.bin +3 -0
  33. granite-3B-part6-ar1-32.mlmodelc/metadata.json +350 -0
  34. granite-3B-part6-ar1-32.mlmodelc/model.mil +0 -0
  35. granite-3B-part6-ar1-32.mlmodelc/weights/weight.bin +3 -0
  36. granite-3B-part7-ar1-32.mlmodelc/analytics/coremldata.bin +3 -0
  37. granite-3B-part7-ar1-32.mlmodelc/coremldata.bin +3 -0
  38. granite-3B-part7-ar1-32.mlmodelc/metadata.json +350 -0
  39. granite-3B-part7-ar1-32.mlmodelc/model.mil +0 -0
  40. granite-3B-part7-ar1-32.mlmodelc/weights/weight.bin +3 -0
  41. utils/.cache/huggingface/.gitignore +1 -0
  42. utils/.cache/huggingface/download/utils/chat_template.jinja.lock +0 -0
  43. utils/.cache/huggingface/download/utils/chat_template.jinja.metadata +3 -0
  44. utils/.cache/huggingface/download/utils/config.json.lock +0 -0
  45. utils/.cache/huggingface/download/utils/config.json.metadata +3 -0
  46. utils/.cache/huggingface/download/utils/embed.pt.lock +0 -0
  47. utils/.cache/huggingface/download/utils/embed.pt.metadata +3 -0
  48. utils/.cache/huggingface/download/utils/merges.txt.lock +0 -0
  49. utils/.cache/huggingface/download/utils/merges.txt.metadata +3 -0
  50. utils/.cache/huggingface/download/utils/special_tokens_map.json.lock +0 -0
granite-3B-part0-ar1-32.mlmodelc/analytics/coremldata.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c124ee22bb353127161251e0c4f467faf8714d5d4fee30eefa230413a20a96e4
3
+ size 243
granite-3B-part0-ar1-32.mlmodelc/coremldata.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:61c212ece12e17d28f26f8124a14e9b0d5292b3f86c46a60a3d93b3e371f5926
3
+ size 569
granite-3B-part0-ar1-32.mlmodelc/metadata.json ADDED
@@ -0,0 +1,350 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 × 1 × 2560)",
11
+ "shortDescription" : "",
12
+ "shape" : "[1, 1, 2560]",
13
+ "name" : "output",
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" : "Float16",
28
+ "formattedType" : "MultiArray (Float16 1 × 1 × 2560)",
29
+ "shortDescription" : "",
30
+ "shape" : "[1, 1, 2560]",
31
+ "name" : "input_embedding",
32
+ "type" : "MultiArray"
33
+ },
34
+ {
35
+ "hasShapeFlexibility" : "0",
36
+ "isOptional" : "0",
37
+ "dataType" : "Float16",
38
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
39
+ "shortDescription" : "",
40
+ "shape" : "[1, 1, 64]",
41
+ "name" : "cos",
42
+ "type" : "MultiArray"
43
+ },
44
+ {
45
+ "hasShapeFlexibility" : "0",
46
+ "isOptional" : "0",
47
+ "dataType" : "Float16",
48
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
49
+ "shortDescription" : "",
50
+ "shape" : "[1, 1, 64]",
51
+ "name" : "sin",
52
+ "type" : "MultiArray"
53
+ },
54
+ {
55
+ "hasShapeFlexibility" : "0",
56
+ "isOptional" : "0",
57
+ "dataType" : "Float16",
58
+ "formattedType" : "MultiArray (Float16 1 × 1 × 1 × 1024)",
59
+ "shortDescription" : "",
60
+ "shape" : "[1, 1, 1, 1024]",
61
+ "name" : "attention_mask",
62
+ "type" : "MultiArray"
63
+ },
64
+ {
65
+ "hasShapeFlexibility" : "0",
66
+ "isOptional" : "0",
67
+ "dataType" : "Int32",
68
+ "formattedType" : "MultiArray (Int32 1)",
69
+ "shortDescription" : "",
70
+ "shape" : "[1]",
71
+ "name" : "n_past",
72
+ "type" : "MultiArray"
73
+ }
74
+ ],
75
+ "computePrecision" : "Mixed (Float16, Int32)",
76
+ "storagePrecision" : "Float16",
77
+ "stateSchema" : [
78
+ {
79
+ "dataType" : "Float16",
80
+ "isOptional" : "0",
81
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
82
+ "shortDescription" : "",
83
+ "shape" : "[5, 8, 1024, 64]",
84
+ "name" : "key_cache",
85
+ "type" : "State"
86
+ },
87
+ {
88
+ "dataType" : "Float16",
89
+ "isOptional" : "0",
90
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
91
+ "shortDescription" : "",
92
+ "shape" : "[5, 8, 1024, 64]",
93
+ "name" : "value_cache",
94
+ "type" : "State"
95
+ }
96
+ ],
97
+ "outputSchema" : [
98
+ {
99
+ "hasShapeFlexibility" : "0",
100
+ "isOptional" : "0",
101
+ "dataType" : "Float16",
102
+ "formattedType" : "MultiArray (Float16 1 × 1 × 2560)",
103
+ "shortDescription" : "",
104
+ "shape" : "[1, 1, 2560]",
105
+ "name" : "output",
106
+ "type" : "MultiArray"
107
+ }
108
+ ],
109
+ "name" : "ar1",
110
+ "mlProgramOperationTypeHistogram" : {
111
+ "Ios18.readState" : 12,
112
+ "Ios18.conv" : 35,
113
+ "Ios18.expandDims" : 32,
114
+ "Ios18.matmul" : 10,
115
+ "Ios18.concat" : 26,
116
+ "Ios18.add" : 26,
117
+ "Ios18.silu" : 5,
118
+ "Ios18.sliceUpdate" : 10,
119
+ "Ios18.sliceByIndex" : 40,
120
+ "Ios18.layerNorm" : 10,
121
+ "Ios18.transpose" : 45,
122
+ "Tile" : 10,
123
+ "Ios18.writeState" : 10,
124
+ "Ios18.reshape" : 40,
125
+ "Ios18.softmax" : 5,
126
+ "Ios18.squeeze" : 5,
127
+ "Ios18.mul" : 70
128
+ }
129
+ },
130
+ {
131
+ "inputSchema" : [
132
+ {
133
+ "hasShapeFlexibility" : "0",
134
+ "isOptional" : "0",
135
+ "dataType" : "Float16",
136
+ "formattedType" : "MultiArray (Float16 1 × 32 × 2560)",
137
+ "shortDescription" : "",
138
+ "shape" : "[1, 32, 2560]",
139
+ "name" : "input_embedding",
140
+ "type" : "MultiArray"
141
+ },
142
+ {
143
+ "hasShapeFlexibility" : "0",
144
+ "isOptional" : "0",
145
+ "dataType" : "Float16",
146
+ "formattedType" : "MultiArray (Float16 1 × 32 × 64)",
147
+ "shortDescription" : "",
148
+ "shape" : "[1, 32, 64]",
149
+ "name" : "cos",
150
+ "type" : "MultiArray"
151
+ },
152
+ {
153
+ "hasShapeFlexibility" : "0",
154
+ "isOptional" : "0",
155
+ "dataType" : "Float16",
156
+ "formattedType" : "MultiArray (Float16 1 × 32 × 64)",
157
+ "shortDescription" : "",
158
+ "shape" : "[1, 32, 64]",
159
+ "name" : "sin",
160
+ "type" : "MultiArray"
161
+ },
162
+ {
163
+ "hasShapeFlexibility" : "0",
164
+ "isOptional" : "0",
165
+ "dataType" : "Float16",
166
+ "formattedType" : "MultiArray (Float16 1 × 1 × 32 × 1024)",
167
+ "shortDescription" : "",
168
+ "shape" : "[1, 1, 32, 1024]",
169
+ "name" : "attention_mask",
170
+ "type" : "MultiArray"
171
+ },
172
+ {
173
+ "hasShapeFlexibility" : "0",
174
+ "isOptional" : "0",
175
+ "dataType" : "Int32",
176
+ "formattedType" : "MultiArray (Int32 1)",
177
+ "shortDescription" : "",
178
+ "shape" : "[1]",
179
+ "name" : "n_past",
180
+ "type" : "MultiArray"
181
+ }
182
+ ],
183
+ "computePrecision" : "Mixed (Float16, Int32)",
184
+ "storagePrecision" : "Float16",
185
+ "stateSchema" : [
186
+ {
187
+ "dataType" : "Float16",
188
+ "isOptional" : "0",
189
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
190
+ "shortDescription" : "",
191
+ "shape" : "[5, 8, 1024, 64]",
192
+ "name" : "key_cache",
193
+ "type" : "State"
194
+ },
195
+ {
196
+ "dataType" : "Float16",
197
+ "isOptional" : "0",
198
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
199
+ "shortDescription" : "",
200
+ "shape" : "[5, 8, 1024, 64]",
201
+ "name" : "value_cache",
202
+ "type" : "State"
203
+ }
204
+ ],
205
+ "outputSchema" : [
206
+ {
207
+ "hasShapeFlexibility" : "0",
208
+ "isOptional" : "0",
209
+ "dataType" : "Float16",
210
+ "formattedType" : "MultiArray (Float16 1 × 32 × 2560)",
211
+ "shortDescription" : "",
212
+ "shape" : "[1, 32, 2560]",
213
+ "name" : "output",
214
+ "type" : "MultiArray"
215
+ }
216
+ ],
217
+ "name" : "ar32",
218
+ "mlProgramOperationTypeHistogram" : {
219
+ "Ios18.readState" : 12,
220
+ "Ios18.conv" : 35,
221
+ "Ios18.expandDims" : 32,
222
+ "Ios18.matmul" : 10,
223
+ "Ios18.concat" : 26,
224
+ "Ios18.add" : 26,
225
+ "Ios18.silu" : 5,
226
+ "Ios18.sliceUpdate" : 10,
227
+ "Ios18.sliceByIndex" : 40,
228
+ "Ios18.layerNorm" : 10,
229
+ "Ios18.transpose" : 45,
230
+ "Tile" : 10,
231
+ "Ios18.writeState" : 10,
232
+ "Ios18.reshape" : 40,
233
+ "Ios18.softmax" : 5,
234
+ "Ios18.squeeze" : 5,
235
+ "Ios18.mul" : 70
236
+ }
237
+ }
238
+ ],
239
+ "mlProgramOperationTypeHistogram" : {
240
+ "Ios18.readState" : 12,
241
+ "Ios18.conv" : 35,
242
+ "Ios18.expandDims" : 32,
243
+ "Ios18.matmul" : 10,
244
+ "Ios18.concat" : 26,
245
+ "Ios18.add" : 26,
246
+ "Ios18.silu" : 5,
247
+ "Ios18.sliceUpdate" : 10,
248
+ "Ios18.sliceByIndex" : 40,
249
+ "Ios18.layerNorm" : 10,
250
+ "Ios18.transpose" : 45,
251
+ "Tile" : 10,
252
+ "Ios18.writeState" : 10,
253
+ "Ios18.reshape" : 40,
254
+ "Ios18.softmax" : 5,
255
+ "Ios18.squeeze" : 5,
256
+ "Ios18.mul" : 70
257
+ },
258
+ "isUpdatable" : "0",
259
+ "stateSchema" : [
260
+ {
261
+ "dataType" : "Float16",
262
+ "isOptional" : "0",
263
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
264
+ "shortDescription" : "",
265
+ "shape" : "[5, 8, 1024, 64]",
266
+ "name" : "key_cache",
267
+ "type" : "State"
268
+ },
269
+ {
270
+ "dataType" : "Float16",
271
+ "isOptional" : "0",
272
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
273
+ "shortDescription" : "",
274
+ "shape" : "[5, 8, 1024, 64]",
275
+ "name" : "value_cache",
276
+ "type" : "State"
277
+ }
278
+ ],
279
+ "availability" : {
280
+ "macOS" : "15.0",
281
+ "tvOS" : "18.0",
282
+ "visionOS" : "2.0",
283
+ "watchOS" : "11.0",
284
+ "iOS" : "18.0",
285
+ "macCatalyst" : "18.0"
286
+ },
287
+ "computePrecision" : "Mixed (Float16, Int32)",
288
+ "modelType" : {
289
+ "name" : "MLModelType_mlProgram"
290
+ },
291
+ "inputSchema" : [
292
+ {
293
+ "hasShapeFlexibility" : "0",
294
+ "isOptional" : "0",
295
+ "dataType" : "Float16",
296
+ "formattedType" : "MultiArray (Float16 1 × 1 × 2560)",
297
+ "shortDescription" : "",
298
+ "shape" : "[1, 1, 2560]",
299
+ "name" : "input_embedding",
300
+ "type" : "MultiArray"
301
+ },
302
+ {
303
+ "hasShapeFlexibility" : "0",
304
+ "isOptional" : "0",
305
+ "dataType" : "Float16",
306
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
307
+ "shortDescription" : "",
308
+ "shape" : "[1, 1, 64]",
309
+ "name" : "cos",
310
+ "type" : "MultiArray"
311
+ },
312
+ {
313
+ "hasShapeFlexibility" : "0",
314
+ "isOptional" : "0",
315
+ "dataType" : "Float16",
316
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
317
+ "shortDescription" : "",
318
+ "shape" : "[1, 1, 64]",
319
+ "name" : "sin",
320
+ "type" : "MultiArray"
321
+ },
322
+ {
323
+ "hasShapeFlexibility" : "0",
324
+ "isOptional" : "0",
325
+ "dataType" : "Float16",
326
+ "formattedType" : "MultiArray (Float16 1 × 1 × 1 × 1024)",
327
+ "shortDescription" : "",
328
+ "shape" : "[1, 1, 1, 1024]",
329
+ "name" : "attention_mask",
330
+ "type" : "MultiArray"
331
+ },
332
+ {
333
+ "hasShapeFlexibility" : "0",
334
+ "isOptional" : "0",
335
+ "dataType" : "Int32",
336
+ "formattedType" : "MultiArray (Int32 1)",
337
+ "shortDescription" : "",
338
+ "shape" : "[1]",
339
+ "name" : "n_past",
340
+ "type" : "MultiArray"
341
+ }
342
+ ],
343
+ "defaultFunctionName" : "ar1",
344
+ "generatedClassName" : "granite_3B_part0_ar1_32",
345
+ "userDefinedMetadata" : {
346
+
347
+ },
348
+ "method" : "predict"
349
+ }
350
+ ]
granite-3B-part0-ar1-32.mlmodelc/model.mil ADDED
The diff for this file is too large to render. See raw diff
 
granite-3B-part0-ar1-32.mlmodelc/weights/weight.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:65b59d0d1a355d7d5922a879cf86c4b1d4e45579e020eaf620b187572de2a01e
3
+ size 786486144
granite-3B-part1-ar1-32.mlmodelc/analytics/coremldata.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c124ee22bb353127161251e0c4f467faf8714d5d4fee30eefa230413a20a96e4
3
+ size 243
granite-3B-part1-ar1-32.mlmodelc/coremldata.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:61c212ece12e17d28f26f8124a14e9b0d5292b3f86c46a60a3d93b3e371f5926
3
+ size 569
granite-3B-part1-ar1-32.mlmodelc/metadata.json ADDED
@@ -0,0 +1,350 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 × 1 × 2560)",
11
+ "shortDescription" : "",
12
+ "shape" : "[1, 1, 2560]",
13
+ "name" : "output",
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" : "Float16",
28
+ "formattedType" : "MultiArray (Float16 1 × 1 × 2560)",
29
+ "shortDescription" : "",
30
+ "shape" : "[1, 1, 2560]",
31
+ "name" : "input_embedding",
32
+ "type" : "MultiArray"
33
+ },
34
+ {
35
+ "hasShapeFlexibility" : "0",
36
+ "isOptional" : "0",
37
+ "dataType" : "Float16",
38
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
39
+ "shortDescription" : "",
40
+ "shape" : "[1, 1, 64]",
41
+ "name" : "cos",
42
+ "type" : "MultiArray"
43
+ },
44
+ {
45
+ "hasShapeFlexibility" : "0",
46
+ "isOptional" : "0",
47
+ "dataType" : "Float16",
48
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
49
+ "shortDescription" : "",
50
+ "shape" : "[1, 1, 64]",
51
+ "name" : "sin",
52
+ "type" : "MultiArray"
53
+ },
54
+ {
55
+ "hasShapeFlexibility" : "0",
56
+ "isOptional" : "0",
57
+ "dataType" : "Float16",
58
+ "formattedType" : "MultiArray (Float16 1 × 1 × 1 × 1024)",
59
+ "shortDescription" : "",
60
+ "shape" : "[1, 1, 1, 1024]",
61
+ "name" : "attention_mask",
62
+ "type" : "MultiArray"
63
+ },
64
+ {
65
+ "hasShapeFlexibility" : "0",
66
+ "isOptional" : "0",
67
+ "dataType" : "Int32",
68
+ "formattedType" : "MultiArray (Int32 1)",
69
+ "shortDescription" : "",
70
+ "shape" : "[1]",
71
+ "name" : "n_past",
72
+ "type" : "MultiArray"
73
+ }
74
+ ],
75
+ "computePrecision" : "Mixed (Float16, Int32)",
76
+ "storagePrecision" : "Float16",
77
+ "stateSchema" : [
78
+ {
79
+ "dataType" : "Float16",
80
+ "isOptional" : "0",
81
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
82
+ "shortDescription" : "",
83
+ "shape" : "[5, 8, 1024, 64]",
84
+ "name" : "key_cache",
85
+ "type" : "State"
86
+ },
87
+ {
88
+ "dataType" : "Float16",
89
+ "isOptional" : "0",
90
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
91
+ "shortDescription" : "",
92
+ "shape" : "[5, 8, 1024, 64]",
93
+ "name" : "value_cache",
94
+ "type" : "State"
95
+ }
96
+ ],
97
+ "outputSchema" : [
98
+ {
99
+ "hasShapeFlexibility" : "0",
100
+ "isOptional" : "0",
101
+ "dataType" : "Float16",
102
+ "formattedType" : "MultiArray (Float16 1 × 1 × 2560)",
103
+ "shortDescription" : "",
104
+ "shape" : "[1, 1, 2560]",
105
+ "name" : "output",
106
+ "type" : "MultiArray"
107
+ }
108
+ ],
109
+ "name" : "ar1",
110
+ "mlProgramOperationTypeHistogram" : {
111
+ "Ios18.readState" : 12,
112
+ "Ios18.conv" : 35,
113
+ "Ios18.expandDims" : 32,
114
+ "Ios18.matmul" : 10,
115
+ "Ios18.concat" : 26,
116
+ "Ios18.add" : 26,
117
+ "Ios18.silu" : 5,
118
+ "Ios18.sliceUpdate" : 10,
119
+ "Ios18.sliceByIndex" : 40,
120
+ "Ios18.layerNorm" : 10,
121
+ "Ios18.transpose" : 45,
122
+ "Tile" : 10,
123
+ "Ios18.writeState" : 10,
124
+ "Ios18.reshape" : 40,
125
+ "Ios18.softmax" : 5,
126
+ "Ios18.squeeze" : 5,
127
+ "Ios18.mul" : 70
128
+ }
129
+ },
130
+ {
131
+ "inputSchema" : [
132
+ {
133
+ "hasShapeFlexibility" : "0",
134
+ "isOptional" : "0",
135
+ "dataType" : "Float16",
136
+ "formattedType" : "MultiArray (Float16 1 × 32 × 2560)",
137
+ "shortDescription" : "",
138
+ "shape" : "[1, 32, 2560]",
139
+ "name" : "input_embedding",
140
+ "type" : "MultiArray"
141
+ },
142
+ {
143
+ "hasShapeFlexibility" : "0",
144
+ "isOptional" : "0",
145
+ "dataType" : "Float16",
146
+ "formattedType" : "MultiArray (Float16 1 × 32 × 64)",
147
+ "shortDescription" : "",
148
+ "shape" : "[1, 32, 64]",
149
+ "name" : "cos",
150
+ "type" : "MultiArray"
151
+ },
152
+ {
153
+ "hasShapeFlexibility" : "0",
154
+ "isOptional" : "0",
155
+ "dataType" : "Float16",
156
+ "formattedType" : "MultiArray (Float16 1 × 32 × 64)",
157
+ "shortDescription" : "",
158
+ "shape" : "[1, 32, 64]",
159
+ "name" : "sin",
160
+ "type" : "MultiArray"
161
+ },
162
+ {
163
+ "hasShapeFlexibility" : "0",
164
+ "isOptional" : "0",
165
+ "dataType" : "Float16",
166
+ "formattedType" : "MultiArray (Float16 1 × 1 × 32 × 1024)",
167
+ "shortDescription" : "",
168
+ "shape" : "[1, 1, 32, 1024]",
169
+ "name" : "attention_mask",
170
+ "type" : "MultiArray"
171
+ },
172
+ {
173
+ "hasShapeFlexibility" : "0",
174
+ "isOptional" : "0",
175
+ "dataType" : "Int32",
176
+ "formattedType" : "MultiArray (Int32 1)",
177
+ "shortDescription" : "",
178
+ "shape" : "[1]",
179
+ "name" : "n_past",
180
+ "type" : "MultiArray"
181
+ }
182
+ ],
183
+ "computePrecision" : "Mixed (Float16, Int32)",
184
+ "storagePrecision" : "Float16",
185
+ "stateSchema" : [
186
+ {
187
+ "dataType" : "Float16",
188
+ "isOptional" : "0",
189
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
190
+ "shortDescription" : "",
191
+ "shape" : "[5, 8, 1024, 64]",
192
+ "name" : "key_cache",
193
+ "type" : "State"
194
+ },
195
+ {
196
+ "dataType" : "Float16",
197
+ "isOptional" : "0",
198
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
199
+ "shortDescription" : "",
200
+ "shape" : "[5, 8, 1024, 64]",
201
+ "name" : "value_cache",
202
+ "type" : "State"
203
+ }
204
+ ],
205
+ "outputSchema" : [
206
+ {
207
+ "hasShapeFlexibility" : "0",
208
+ "isOptional" : "0",
209
+ "dataType" : "Float16",
210
+ "formattedType" : "MultiArray (Float16 1 × 32 × 2560)",
211
+ "shortDescription" : "",
212
+ "shape" : "[1, 32, 2560]",
213
+ "name" : "output",
214
+ "type" : "MultiArray"
215
+ }
216
+ ],
217
+ "name" : "ar32",
218
+ "mlProgramOperationTypeHistogram" : {
219
+ "Ios18.readState" : 12,
220
+ "Ios18.conv" : 35,
221
+ "Ios18.expandDims" : 32,
222
+ "Ios18.matmul" : 10,
223
+ "Ios18.concat" : 26,
224
+ "Ios18.add" : 26,
225
+ "Ios18.silu" : 5,
226
+ "Ios18.sliceUpdate" : 10,
227
+ "Ios18.sliceByIndex" : 40,
228
+ "Ios18.layerNorm" : 10,
229
+ "Ios18.transpose" : 45,
230
+ "Tile" : 10,
231
+ "Ios18.writeState" : 10,
232
+ "Ios18.reshape" : 40,
233
+ "Ios18.softmax" : 5,
234
+ "Ios18.squeeze" : 5,
235
+ "Ios18.mul" : 70
236
+ }
237
+ }
238
+ ],
239
+ "mlProgramOperationTypeHistogram" : {
240
+ "Ios18.readState" : 12,
241
+ "Ios18.conv" : 35,
242
+ "Ios18.expandDims" : 32,
243
+ "Ios18.matmul" : 10,
244
+ "Ios18.concat" : 26,
245
+ "Ios18.add" : 26,
246
+ "Ios18.silu" : 5,
247
+ "Ios18.sliceUpdate" : 10,
248
+ "Ios18.sliceByIndex" : 40,
249
+ "Ios18.layerNorm" : 10,
250
+ "Ios18.transpose" : 45,
251
+ "Tile" : 10,
252
+ "Ios18.writeState" : 10,
253
+ "Ios18.reshape" : 40,
254
+ "Ios18.softmax" : 5,
255
+ "Ios18.squeeze" : 5,
256
+ "Ios18.mul" : 70
257
+ },
258
+ "isUpdatable" : "0",
259
+ "stateSchema" : [
260
+ {
261
+ "dataType" : "Float16",
262
+ "isOptional" : "0",
263
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
264
+ "shortDescription" : "",
265
+ "shape" : "[5, 8, 1024, 64]",
266
+ "name" : "key_cache",
267
+ "type" : "State"
268
+ },
269
+ {
270
+ "dataType" : "Float16",
271
+ "isOptional" : "0",
272
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
273
+ "shortDescription" : "",
274
+ "shape" : "[5, 8, 1024, 64]",
275
+ "name" : "value_cache",
276
+ "type" : "State"
277
+ }
278
+ ],
279
+ "availability" : {
280
+ "macOS" : "15.0",
281
+ "tvOS" : "18.0",
282
+ "visionOS" : "2.0",
283
+ "watchOS" : "11.0",
284
+ "iOS" : "18.0",
285
+ "macCatalyst" : "18.0"
286
+ },
287
+ "computePrecision" : "Mixed (Float16, Int32)",
288
+ "modelType" : {
289
+ "name" : "MLModelType_mlProgram"
290
+ },
291
+ "inputSchema" : [
292
+ {
293
+ "hasShapeFlexibility" : "0",
294
+ "isOptional" : "0",
295
+ "dataType" : "Float16",
296
+ "formattedType" : "MultiArray (Float16 1 × 1 × 2560)",
297
+ "shortDescription" : "",
298
+ "shape" : "[1, 1, 2560]",
299
+ "name" : "input_embedding",
300
+ "type" : "MultiArray"
301
+ },
302
+ {
303
+ "hasShapeFlexibility" : "0",
304
+ "isOptional" : "0",
305
+ "dataType" : "Float16",
306
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
307
+ "shortDescription" : "",
308
+ "shape" : "[1, 1, 64]",
309
+ "name" : "cos",
310
+ "type" : "MultiArray"
311
+ },
312
+ {
313
+ "hasShapeFlexibility" : "0",
314
+ "isOptional" : "0",
315
+ "dataType" : "Float16",
316
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
317
+ "shortDescription" : "",
318
+ "shape" : "[1, 1, 64]",
319
+ "name" : "sin",
320
+ "type" : "MultiArray"
321
+ },
322
+ {
323
+ "hasShapeFlexibility" : "0",
324
+ "isOptional" : "0",
325
+ "dataType" : "Float16",
326
+ "formattedType" : "MultiArray (Float16 1 × 1 × 1 × 1024)",
327
+ "shortDescription" : "",
328
+ "shape" : "[1, 1, 1, 1024]",
329
+ "name" : "attention_mask",
330
+ "type" : "MultiArray"
331
+ },
332
+ {
333
+ "hasShapeFlexibility" : "0",
334
+ "isOptional" : "0",
335
+ "dataType" : "Int32",
336
+ "formattedType" : "MultiArray (Int32 1)",
337
+ "shortDescription" : "",
338
+ "shape" : "[1]",
339
+ "name" : "n_past",
340
+ "type" : "MultiArray"
341
+ }
342
+ ],
343
+ "defaultFunctionName" : "ar1",
344
+ "generatedClassName" : "granite_3B_part1_ar1_32",
345
+ "userDefinedMetadata" : {
346
+
347
+ },
348
+ "method" : "predict"
349
+ }
350
+ ]
granite-3B-part1-ar1-32.mlmodelc/model.mil ADDED
The diff for this file is too large to render. See raw diff
 
granite-3B-part1-ar1-32.mlmodelc/weights/weight.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:41af9594a88d93de0f6720083a928550f712d447dc9e732c615aefc876b98711
3
+ size 786486144
granite-3B-part2-ar1-32.mlmodelc/analytics/coremldata.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c124ee22bb353127161251e0c4f467faf8714d5d4fee30eefa230413a20a96e4
3
+ size 243
granite-3B-part2-ar1-32.mlmodelc/coremldata.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:61c212ece12e17d28f26f8124a14e9b0d5292b3f86c46a60a3d93b3e371f5926
3
+ size 569
granite-3B-part2-ar1-32.mlmodelc/metadata.json ADDED
@@ -0,0 +1,350 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 × 1 × 2560)",
11
+ "shortDescription" : "",
12
+ "shape" : "[1, 1, 2560]",
13
+ "name" : "output",
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" : "Float16",
28
+ "formattedType" : "MultiArray (Float16 1 × 1 × 2560)",
29
+ "shortDescription" : "",
30
+ "shape" : "[1, 1, 2560]",
31
+ "name" : "input_embedding",
32
+ "type" : "MultiArray"
33
+ },
34
+ {
35
+ "hasShapeFlexibility" : "0",
36
+ "isOptional" : "0",
37
+ "dataType" : "Float16",
38
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
39
+ "shortDescription" : "",
40
+ "shape" : "[1, 1, 64]",
41
+ "name" : "cos",
42
+ "type" : "MultiArray"
43
+ },
44
+ {
45
+ "hasShapeFlexibility" : "0",
46
+ "isOptional" : "0",
47
+ "dataType" : "Float16",
48
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
49
+ "shortDescription" : "",
50
+ "shape" : "[1, 1, 64]",
51
+ "name" : "sin",
52
+ "type" : "MultiArray"
53
+ },
54
+ {
55
+ "hasShapeFlexibility" : "0",
56
+ "isOptional" : "0",
57
+ "dataType" : "Float16",
58
+ "formattedType" : "MultiArray (Float16 1 × 1 × 1 × 1024)",
59
+ "shortDescription" : "",
60
+ "shape" : "[1, 1, 1, 1024]",
61
+ "name" : "attention_mask",
62
+ "type" : "MultiArray"
63
+ },
64
+ {
65
+ "hasShapeFlexibility" : "0",
66
+ "isOptional" : "0",
67
+ "dataType" : "Int32",
68
+ "formattedType" : "MultiArray (Int32 1)",
69
+ "shortDescription" : "",
70
+ "shape" : "[1]",
71
+ "name" : "n_past",
72
+ "type" : "MultiArray"
73
+ }
74
+ ],
75
+ "computePrecision" : "Mixed (Float16, Int32)",
76
+ "storagePrecision" : "Float16",
77
+ "stateSchema" : [
78
+ {
79
+ "dataType" : "Float16",
80
+ "isOptional" : "0",
81
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
82
+ "shortDescription" : "",
83
+ "shape" : "[5, 8, 1024, 64]",
84
+ "name" : "key_cache",
85
+ "type" : "State"
86
+ },
87
+ {
88
+ "dataType" : "Float16",
89
+ "isOptional" : "0",
90
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
91
+ "shortDescription" : "",
92
+ "shape" : "[5, 8, 1024, 64]",
93
+ "name" : "value_cache",
94
+ "type" : "State"
95
+ }
96
+ ],
97
+ "outputSchema" : [
98
+ {
99
+ "hasShapeFlexibility" : "0",
100
+ "isOptional" : "0",
101
+ "dataType" : "Float16",
102
+ "formattedType" : "MultiArray (Float16 1 × 1 × 2560)",
103
+ "shortDescription" : "",
104
+ "shape" : "[1, 1, 2560]",
105
+ "name" : "output",
106
+ "type" : "MultiArray"
107
+ }
108
+ ],
109
+ "name" : "ar1",
110
+ "mlProgramOperationTypeHistogram" : {
111
+ "Ios18.readState" : 12,
112
+ "Ios18.conv" : 35,
113
+ "Ios18.expandDims" : 32,
114
+ "Ios18.matmul" : 10,
115
+ "Ios18.concat" : 26,
116
+ "Ios18.add" : 26,
117
+ "Ios18.silu" : 5,
118
+ "Ios18.sliceUpdate" : 10,
119
+ "Ios18.sliceByIndex" : 40,
120
+ "Ios18.layerNorm" : 10,
121
+ "Ios18.transpose" : 45,
122
+ "Tile" : 10,
123
+ "Ios18.writeState" : 10,
124
+ "Ios18.reshape" : 40,
125
+ "Ios18.softmax" : 5,
126
+ "Ios18.squeeze" : 5,
127
+ "Ios18.mul" : 70
128
+ }
129
+ },
130
+ {
131
+ "inputSchema" : [
132
+ {
133
+ "hasShapeFlexibility" : "0",
134
+ "isOptional" : "0",
135
+ "dataType" : "Float16",
136
+ "formattedType" : "MultiArray (Float16 1 × 32 × 2560)",
137
+ "shortDescription" : "",
138
+ "shape" : "[1, 32, 2560]",
139
+ "name" : "input_embedding",
140
+ "type" : "MultiArray"
141
+ },
142
+ {
143
+ "hasShapeFlexibility" : "0",
144
+ "isOptional" : "0",
145
+ "dataType" : "Float16",
146
+ "formattedType" : "MultiArray (Float16 1 × 32 × 64)",
147
+ "shortDescription" : "",
148
+ "shape" : "[1, 32, 64]",
149
+ "name" : "cos",
150
+ "type" : "MultiArray"
151
+ },
152
+ {
153
+ "hasShapeFlexibility" : "0",
154
+ "isOptional" : "0",
155
+ "dataType" : "Float16",
156
+ "formattedType" : "MultiArray (Float16 1 × 32 × 64)",
157
+ "shortDescription" : "",
158
+ "shape" : "[1, 32, 64]",
159
+ "name" : "sin",
160
+ "type" : "MultiArray"
161
+ },
162
+ {
163
+ "hasShapeFlexibility" : "0",
164
+ "isOptional" : "0",
165
+ "dataType" : "Float16",
166
+ "formattedType" : "MultiArray (Float16 1 × 1 × 32 × 1024)",
167
+ "shortDescription" : "",
168
+ "shape" : "[1, 1, 32, 1024]",
169
+ "name" : "attention_mask",
170
+ "type" : "MultiArray"
171
+ },
172
+ {
173
+ "hasShapeFlexibility" : "0",
174
+ "isOptional" : "0",
175
+ "dataType" : "Int32",
176
+ "formattedType" : "MultiArray (Int32 1)",
177
+ "shortDescription" : "",
178
+ "shape" : "[1]",
179
+ "name" : "n_past",
180
+ "type" : "MultiArray"
181
+ }
182
+ ],
183
+ "computePrecision" : "Mixed (Float16, Int32)",
184
+ "storagePrecision" : "Float16",
185
+ "stateSchema" : [
186
+ {
187
+ "dataType" : "Float16",
188
+ "isOptional" : "0",
189
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
190
+ "shortDescription" : "",
191
+ "shape" : "[5, 8, 1024, 64]",
192
+ "name" : "key_cache",
193
+ "type" : "State"
194
+ },
195
+ {
196
+ "dataType" : "Float16",
197
+ "isOptional" : "0",
198
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
199
+ "shortDescription" : "",
200
+ "shape" : "[5, 8, 1024, 64]",
201
+ "name" : "value_cache",
202
+ "type" : "State"
203
+ }
204
+ ],
205
+ "outputSchema" : [
206
+ {
207
+ "hasShapeFlexibility" : "0",
208
+ "isOptional" : "0",
209
+ "dataType" : "Float16",
210
+ "formattedType" : "MultiArray (Float16 1 × 32 × 2560)",
211
+ "shortDescription" : "",
212
+ "shape" : "[1, 32, 2560]",
213
+ "name" : "output",
214
+ "type" : "MultiArray"
215
+ }
216
+ ],
217
+ "name" : "ar32",
218
+ "mlProgramOperationTypeHistogram" : {
219
+ "Ios18.readState" : 12,
220
+ "Ios18.conv" : 35,
221
+ "Ios18.expandDims" : 32,
222
+ "Ios18.matmul" : 10,
223
+ "Ios18.concat" : 26,
224
+ "Ios18.add" : 26,
225
+ "Ios18.silu" : 5,
226
+ "Ios18.sliceUpdate" : 10,
227
+ "Ios18.sliceByIndex" : 40,
228
+ "Ios18.layerNorm" : 10,
229
+ "Ios18.transpose" : 45,
230
+ "Tile" : 10,
231
+ "Ios18.writeState" : 10,
232
+ "Ios18.reshape" : 40,
233
+ "Ios18.softmax" : 5,
234
+ "Ios18.squeeze" : 5,
235
+ "Ios18.mul" : 70
236
+ }
237
+ }
238
+ ],
239
+ "mlProgramOperationTypeHistogram" : {
240
+ "Ios18.readState" : 12,
241
+ "Ios18.conv" : 35,
242
+ "Ios18.expandDims" : 32,
243
+ "Ios18.matmul" : 10,
244
+ "Ios18.concat" : 26,
245
+ "Ios18.add" : 26,
246
+ "Ios18.silu" : 5,
247
+ "Ios18.sliceUpdate" : 10,
248
+ "Ios18.sliceByIndex" : 40,
249
+ "Ios18.layerNorm" : 10,
250
+ "Ios18.transpose" : 45,
251
+ "Tile" : 10,
252
+ "Ios18.writeState" : 10,
253
+ "Ios18.reshape" : 40,
254
+ "Ios18.softmax" : 5,
255
+ "Ios18.squeeze" : 5,
256
+ "Ios18.mul" : 70
257
+ },
258
+ "isUpdatable" : "0",
259
+ "stateSchema" : [
260
+ {
261
+ "dataType" : "Float16",
262
+ "isOptional" : "0",
263
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
264
+ "shortDescription" : "",
265
+ "shape" : "[5, 8, 1024, 64]",
266
+ "name" : "key_cache",
267
+ "type" : "State"
268
+ },
269
+ {
270
+ "dataType" : "Float16",
271
+ "isOptional" : "0",
272
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
273
+ "shortDescription" : "",
274
+ "shape" : "[5, 8, 1024, 64]",
275
+ "name" : "value_cache",
276
+ "type" : "State"
277
+ }
278
+ ],
279
+ "availability" : {
280
+ "macOS" : "15.0",
281
+ "tvOS" : "18.0",
282
+ "visionOS" : "2.0",
283
+ "watchOS" : "11.0",
284
+ "iOS" : "18.0",
285
+ "macCatalyst" : "18.0"
286
+ },
287
+ "computePrecision" : "Mixed (Float16, Int32)",
288
+ "modelType" : {
289
+ "name" : "MLModelType_mlProgram"
290
+ },
291
+ "inputSchema" : [
292
+ {
293
+ "hasShapeFlexibility" : "0",
294
+ "isOptional" : "0",
295
+ "dataType" : "Float16",
296
+ "formattedType" : "MultiArray (Float16 1 × 1 × 2560)",
297
+ "shortDescription" : "",
298
+ "shape" : "[1, 1, 2560]",
299
+ "name" : "input_embedding",
300
+ "type" : "MultiArray"
301
+ },
302
+ {
303
+ "hasShapeFlexibility" : "0",
304
+ "isOptional" : "0",
305
+ "dataType" : "Float16",
306
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
307
+ "shortDescription" : "",
308
+ "shape" : "[1, 1, 64]",
309
+ "name" : "cos",
310
+ "type" : "MultiArray"
311
+ },
312
+ {
313
+ "hasShapeFlexibility" : "0",
314
+ "isOptional" : "0",
315
+ "dataType" : "Float16",
316
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
317
+ "shortDescription" : "",
318
+ "shape" : "[1, 1, 64]",
319
+ "name" : "sin",
320
+ "type" : "MultiArray"
321
+ },
322
+ {
323
+ "hasShapeFlexibility" : "0",
324
+ "isOptional" : "0",
325
+ "dataType" : "Float16",
326
+ "formattedType" : "MultiArray (Float16 1 × 1 × 1 × 1024)",
327
+ "shortDescription" : "",
328
+ "shape" : "[1, 1, 1, 1024]",
329
+ "name" : "attention_mask",
330
+ "type" : "MultiArray"
331
+ },
332
+ {
333
+ "hasShapeFlexibility" : "0",
334
+ "isOptional" : "0",
335
+ "dataType" : "Int32",
336
+ "formattedType" : "MultiArray (Int32 1)",
337
+ "shortDescription" : "",
338
+ "shape" : "[1]",
339
+ "name" : "n_past",
340
+ "type" : "MultiArray"
341
+ }
342
+ ],
343
+ "defaultFunctionName" : "ar1",
344
+ "generatedClassName" : "granite_3B_part2_ar1_32",
345
+ "userDefinedMetadata" : {
346
+
347
+ },
348
+ "method" : "predict"
349
+ }
350
+ ]
granite-3B-part2-ar1-32.mlmodelc/model.mil ADDED
The diff for this file is too large to render. See raw diff
 
granite-3B-part2-ar1-32.mlmodelc/weights/weight.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6447d3858dafc6bae1ad7ac4cef5801b13199dd47d50912e5e96061140740e3b
3
+ size 786486144
granite-3B-part3-ar1-32.mlmodelc/analytics/coremldata.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c124ee22bb353127161251e0c4f467faf8714d5d4fee30eefa230413a20a96e4
3
+ size 243
granite-3B-part3-ar1-32.mlmodelc/coremldata.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:61c212ece12e17d28f26f8124a14e9b0d5292b3f86c46a60a3d93b3e371f5926
3
+ size 569
granite-3B-part3-ar1-32.mlmodelc/metadata.json ADDED
@@ -0,0 +1,350 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 × 1 × 2560)",
11
+ "shortDescription" : "",
12
+ "shape" : "[1, 1, 2560]",
13
+ "name" : "output",
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" : "Float16",
28
+ "formattedType" : "MultiArray (Float16 1 × 1 × 2560)",
29
+ "shortDescription" : "",
30
+ "shape" : "[1, 1, 2560]",
31
+ "name" : "input_embedding",
32
+ "type" : "MultiArray"
33
+ },
34
+ {
35
+ "hasShapeFlexibility" : "0",
36
+ "isOptional" : "0",
37
+ "dataType" : "Float16",
38
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
39
+ "shortDescription" : "",
40
+ "shape" : "[1, 1, 64]",
41
+ "name" : "cos",
42
+ "type" : "MultiArray"
43
+ },
44
+ {
45
+ "hasShapeFlexibility" : "0",
46
+ "isOptional" : "0",
47
+ "dataType" : "Float16",
48
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
49
+ "shortDescription" : "",
50
+ "shape" : "[1, 1, 64]",
51
+ "name" : "sin",
52
+ "type" : "MultiArray"
53
+ },
54
+ {
55
+ "hasShapeFlexibility" : "0",
56
+ "isOptional" : "0",
57
+ "dataType" : "Float16",
58
+ "formattedType" : "MultiArray (Float16 1 × 1 × 1 × 1024)",
59
+ "shortDescription" : "",
60
+ "shape" : "[1, 1, 1, 1024]",
61
+ "name" : "attention_mask",
62
+ "type" : "MultiArray"
63
+ },
64
+ {
65
+ "hasShapeFlexibility" : "0",
66
+ "isOptional" : "0",
67
+ "dataType" : "Int32",
68
+ "formattedType" : "MultiArray (Int32 1)",
69
+ "shortDescription" : "",
70
+ "shape" : "[1]",
71
+ "name" : "n_past",
72
+ "type" : "MultiArray"
73
+ }
74
+ ],
75
+ "computePrecision" : "Mixed (Float16, Int32)",
76
+ "storagePrecision" : "Float16",
77
+ "stateSchema" : [
78
+ {
79
+ "dataType" : "Float16",
80
+ "isOptional" : "0",
81
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
82
+ "shortDescription" : "",
83
+ "shape" : "[5, 8, 1024, 64]",
84
+ "name" : "key_cache",
85
+ "type" : "State"
86
+ },
87
+ {
88
+ "dataType" : "Float16",
89
+ "isOptional" : "0",
90
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
91
+ "shortDescription" : "",
92
+ "shape" : "[5, 8, 1024, 64]",
93
+ "name" : "value_cache",
94
+ "type" : "State"
95
+ }
96
+ ],
97
+ "outputSchema" : [
98
+ {
99
+ "hasShapeFlexibility" : "0",
100
+ "isOptional" : "0",
101
+ "dataType" : "Float16",
102
+ "formattedType" : "MultiArray (Float16 1 × 1 × 2560)",
103
+ "shortDescription" : "",
104
+ "shape" : "[1, 1, 2560]",
105
+ "name" : "output",
106
+ "type" : "MultiArray"
107
+ }
108
+ ],
109
+ "name" : "ar1",
110
+ "mlProgramOperationTypeHistogram" : {
111
+ "Ios18.readState" : 12,
112
+ "Ios18.conv" : 35,
113
+ "Ios18.expandDims" : 32,
114
+ "Ios18.matmul" : 10,
115
+ "Ios18.concat" : 26,
116
+ "Ios18.add" : 26,
117
+ "Ios18.silu" : 5,
118
+ "Ios18.sliceUpdate" : 10,
119
+ "Ios18.sliceByIndex" : 40,
120
+ "Ios18.layerNorm" : 10,
121
+ "Ios18.transpose" : 45,
122
+ "Tile" : 10,
123
+ "Ios18.writeState" : 10,
124
+ "Ios18.reshape" : 40,
125
+ "Ios18.softmax" : 5,
126
+ "Ios18.squeeze" : 5,
127
+ "Ios18.mul" : 70
128
+ }
129
+ },
130
+ {
131
+ "inputSchema" : [
132
+ {
133
+ "hasShapeFlexibility" : "0",
134
+ "isOptional" : "0",
135
+ "dataType" : "Float16",
136
+ "formattedType" : "MultiArray (Float16 1 × 32 × 2560)",
137
+ "shortDescription" : "",
138
+ "shape" : "[1, 32, 2560]",
139
+ "name" : "input_embedding",
140
+ "type" : "MultiArray"
141
+ },
142
+ {
143
+ "hasShapeFlexibility" : "0",
144
+ "isOptional" : "0",
145
+ "dataType" : "Float16",
146
+ "formattedType" : "MultiArray (Float16 1 × 32 × 64)",
147
+ "shortDescription" : "",
148
+ "shape" : "[1, 32, 64]",
149
+ "name" : "cos",
150
+ "type" : "MultiArray"
151
+ },
152
+ {
153
+ "hasShapeFlexibility" : "0",
154
+ "isOptional" : "0",
155
+ "dataType" : "Float16",
156
+ "formattedType" : "MultiArray (Float16 1 × 32 × 64)",
157
+ "shortDescription" : "",
158
+ "shape" : "[1, 32, 64]",
159
+ "name" : "sin",
160
+ "type" : "MultiArray"
161
+ },
162
+ {
163
+ "hasShapeFlexibility" : "0",
164
+ "isOptional" : "0",
165
+ "dataType" : "Float16",
166
+ "formattedType" : "MultiArray (Float16 1 × 1 × 32 × 1024)",
167
+ "shortDescription" : "",
168
+ "shape" : "[1, 1, 32, 1024]",
169
+ "name" : "attention_mask",
170
+ "type" : "MultiArray"
171
+ },
172
+ {
173
+ "hasShapeFlexibility" : "0",
174
+ "isOptional" : "0",
175
+ "dataType" : "Int32",
176
+ "formattedType" : "MultiArray (Int32 1)",
177
+ "shortDescription" : "",
178
+ "shape" : "[1]",
179
+ "name" : "n_past",
180
+ "type" : "MultiArray"
181
+ }
182
+ ],
183
+ "computePrecision" : "Mixed (Float16, Int32)",
184
+ "storagePrecision" : "Float16",
185
+ "stateSchema" : [
186
+ {
187
+ "dataType" : "Float16",
188
+ "isOptional" : "0",
189
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
190
+ "shortDescription" : "",
191
+ "shape" : "[5, 8, 1024, 64]",
192
+ "name" : "key_cache",
193
+ "type" : "State"
194
+ },
195
+ {
196
+ "dataType" : "Float16",
197
+ "isOptional" : "0",
198
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
199
+ "shortDescription" : "",
200
+ "shape" : "[5, 8, 1024, 64]",
201
+ "name" : "value_cache",
202
+ "type" : "State"
203
+ }
204
+ ],
205
+ "outputSchema" : [
206
+ {
207
+ "hasShapeFlexibility" : "0",
208
+ "isOptional" : "0",
209
+ "dataType" : "Float16",
210
+ "formattedType" : "MultiArray (Float16 1 × 32 × 2560)",
211
+ "shortDescription" : "",
212
+ "shape" : "[1, 32, 2560]",
213
+ "name" : "output",
214
+ "type" : "MultiArray"
215
+ }
216
+ ],
217
+ "name" : "ar32",
218
+ "mlProgramOperationTypeHistogram" : {
219
+ "Ios18.readState" : 12,
220
+ "Ios18.conv" : 35,
221
+ "Ios18.expandDims" : 32,
222
+ "Ios18.matmul" : 10,
223
+ "Ios18.concat" : 26,
224
+ "Ios18.add" : 26,
225
+ "Ios18.silu" : 5,
226
+ "Ios18.sliceUpdate" : 10,
227
+ "Ios18.sliceByIndex" : 40,
228
+ "Ios18.layerNorm" : 10,
229
+ "Ios18.transpose" : 45,
230
+ "Tile" : 10,
231
+ "Ios18.writeState" : 10,
232
+ "Ios18.reshape" : 40,
233
+ "Ios18.softmax" : 5,
234
+ "Ios18.squeeze" : 5,
235
+ "Ios18.mul" : 70
236
+ }
237
+ }
238
+ ],
239
+ "mlProgramOperationTypeHistogram" : {
240
+ "Ios18.readState" : 12,
241
+ "Ios18.conv" : 35,
242
+ "Ios18.expandDims" : 32,
243
+ "Ios18.matmul" : 10,
244
+ "Ios18.concat" : 26,
245
+ "Ios18.add" : 26,
246
+ "Ios18.silu" : 5,
247
+ "Ios18.sliceUpdate" : 10,
248
+ "Ios18.sliceByIndex" : 40,
249
+ "Ios18.layerNorm" : 10,
250
+ "Ios18.transpose" : 45,
251
+ "Tile" : 10,
252
+ "Ios18.writeState" : 10,
253
+ "Ios18.reshape" : 40,
254
+ "Ios18.softmax" : 5,
255
+ "Ios18.squeeze" : 5,
256
+ "Ios18.mul" : 70
257
+ },
258
+ "isUpdatable" : "0",
259
+ "stateSchema" : [
260
+ {
261
+ "dataType" : "Float16",
262
+ "isOptional" : "0",
263
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
264
+ "shortDescription" : "",
265
+ "shape" : "[5, 8, 1024, 64]",
266
+ "name" : "key_cache",
267
+ "type" : "State"
268
+ },
269
+ {
270
+ "dataType" : "Float16",
271
+ "isOptional" : "0",
272
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
273
+ "shortDescription" : "",
274
+ "shape" : "[5, 8, 1024, 64]",
275
+ "name" : "value_cache",
276
+ "type" : "State"
277
+ }
278
+ ],
279
+ "availability" : {
280
+ "macOS" : "15.0",
281
+ "tvOS" : "18.0",
282
+ "visionOS" : "2.0",
283
+ "watchOS" : "11.0",
284
+ "iOS" : "18.0",
285
+ "macCatalyst" : "18.0"
286
+ },
287
+ "computePrecision" : "Mixed (Float16, Int32)",
288
+ "modelType" : {
289
+ "name" : "MLModelType_mlProgram"
290
+ },
291
+ "inputSchema" : [
292
+ {
293
+ "hasShapeFlexibility" : "0",
294
+ "isOptional" : "0",
295
+ "dataType" : "Float16",
296
+ "formattedType" : "MultiArray (Float16 1 × 1 × 2560)",
297
+ "shortDescription" : "",
298
+ "shape" : "[1, 1, 2560]",
299
+ "name" : "input_embedding",
300
+ "type" : "MultiArray"
301
+ },
302
+ {
303
+ "hasShapeFlexibility" : "0",
304
+ "isOptional" : "0",
305
+ "dataType" : "Float16",
306
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
307
+ "shortDescription" : "",
308
+ "shape" : "[1, 1, 64]",
309
+ "name" : "cos",
310
+ "type" : "MultiArray"
311
+ },
312
+ {
313
+ "hasShapeFlexibility" : "0",
314
+ "isOptional" : "0",
315
+ "dataType" : "Float16",
316
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
317
+ "shortDescription" : "",
318
+ "shape" : "[1, 1, 64]",
319
+ "name" : "sin",
320
+ "type" : "MultiArray"
321
+ },
322
+ {
323
+ "hasShapeFlexibility" : "0",
324
+ "isOptional" : "0",
325
+ "dataType" : "Float16",
326
+ "formattedType" : "MultiArray (Float16 1 × 1 × 1 × 1024)",
327
+ "shortDescription" : "",
328
+ "shape" : "[1, 1, 1, 1024]",
329
+ "name" : "attention_mask",
330
+ "type" : "MultiArray"
331
+ },
332
+ {
333
+ "hasShapeFlexibility" : "0",
334
+ "isOptional" : "0",
335
+ "dataType" : "Int32",
336
+ "formattedType" : "MultiArray (Int32 1)",
337
+ "shortDescription" : "",
338
+ "shape" : "[1]",
339
+ "name" : "n_past",
340
+ "type" : "MultiArray"
341
+ }
342
+ ],
343
+ "defaultFunctionName" : "ar1",
344
+ "generatedClassName" : "granite_3B_part3_ar1_32",
345
+ "userDefinedMetadata" : {
346
+
347
+ },
348
+ "method" : "predict"
349
+ }
350
+ ]
granite-3B-part3-ar1-32.mlmodelc/model.mil ADDED
The diff for this file is too large to render. See raw diff
 
granite-3B-part3-ar1-32.mlmodelc/weights/weight.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a148e6e903f1d6e27889a499d7264c3bdecf7f3f793ea679a2eca5e9631929ef
3
+ size 786486144
granite-3B-part4-ar1-32.mlmodelc/analytics/coremldata.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c124ee22bb353127161251e0c4f467faf8714d5d4fee30eefa230413a20a96e4
3
+ size 243
granite-3B-part4-ar1-32.mlmodelc/coremldata.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:61c212ece12e17d28f26f8124a14e9b0d5292b3f86c46a60a3d93b3e371f5926
3
+ size 569
granite-3B-part4-ar1-32.mlmodelc/metadata.json ADDED
@@ -0,0 +1,350 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 × 1 × 2560)",
11
+ "shortDescription" : "",
12
+ "shape" : "[1, 1, 2560]",
13
+ "name" : "output",
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" : "Float16",
28
+ "formattedType" : "MultiArray (Float16 1 × 1 × 2560)",
29
+ "shortDescription" : "",
30
+ "shape" : "[1, 1, 2560]",
31
+ "name" : "input_embedding",
32
+ "type" : "MultiArray"
33
+ },
34
+ {
35
+ "hasShapeFlexibility" : "0",
36
+ "isOptional" : "0",
37
+ "dataType" : "Float16",
38
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
39
+ "shortDescription" : "",
40
+ "shape" : "[1, 1, 64]",
41
+ "name" : "cos",
42
+ "type" : "MultiArray"
43
+ },
44
+ {
45
+ "hasShapeFlexibility" : "0",
46
+ "isOptional" : "0",
47
+ "dataType" : "Float16",
48
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
49
+ "shortDescription" : "",
50
+ "shape" : "[1, 1, 64]",
51
+ "name" : "sin",
52
+ "type" : "MultiArray"
53
+ },
54
+ {
55
+ "hasShapeFlexibility" : "0",
56
+ "isOptional" : "0",
57
+ "dataType" : "Float16",
58
+ "formattedType" : "MultiArray (Float16 1 × 1 × 1 × 1024)",
59
+ "shortDescription" : "",
60
+ "shape" : "[1, 1, 1, 1024]",
61
+ "name" : "attention_mask",
62
+ "type" : "MultiArray"
63
+ },
64
+ {
65
+ "hasShapeFlexibility" : "0",
66
+ "isOptional" : "0",
67
+ "dataType" : "Int32",
68
+ "formattedType" : "MultiArray (Int32 1)",
69
+ "shortDescription" : "",
70
+ "shape" : "[1]",
71
+ "name" : "n_past",
72
+ "type" : "MultiArray"
73
+ }
74
+ ],
75
+ "computePrecision" : "Mixed (Float16, Int32)",
76
+ "storagePrecision" : "Float16",
77
+ "stateSchema" : [
78
+ {
79
+ "dataType" : "Float16",
80
+ "isOptional" : "0",
81
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
82
+ "shortDescription" : "",
83
+ "shape" : "[5, 8, 1024, 64]",
84
+ "name" : "key_cache",
85
+ "type" : "State"
86
+ },
87
+ {
88
+ "dataType" : "Float16",
89
+ "isOptional" : "0",
90
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
91
+ "shortDescription" : "",
92
+ "shape" : "[5, 8, 1024, 64]",
93
+ "name" : "value_cache",
94
+ "type" : "State"
95
+ }
96
+ ],
97
+ "outputSchema" : [
98
+ {
99
+ "hasShapeFlexibility" : "0",
100
+ "isOptional" : "0",
101
+ "dataType" : "Float16",
102
+ "formattedType" : "MultiArray (Float16 1 × 1 × 2560)",
103
+ "shortDescription" : "",
104
+ "shape" : "[1, 1, 2560]",
105
+ "name" : "output",
106
+ "type" : "MultiArray"
107
+ }
108
+ ],
109
+ "name" : "ar1",
110
+ "mlProgramOperationTypeHistogram" : {
111
+ "Ios18.readState" : 12,
112
+ "Ios18.conv" : 35,
113
+ "Ios18.expandDims" : 32,
114
+ "Ios18.matmul" : 10,
115
+ "Ios18.concat" : 26,
116
+ "Ios18.add" : 26,
117
+ "Ios18.silu" : 5,
118
+ "Ios18.sliceUpdate" : 10,
119
+ "Ios18.sliceByIndex" : 40,
120
+ "Ios18.layerNorm" : 10,
121
+ "Ios18.transpose" : 45,
122
+ "Tile" : 10,
123
+ "Ios18.writeState" : 10,
124
+ "Ios18.reshape" : 40,
125
+ "Ios18.softmax" : 5,
126
+ "Ios18.squeeze" : 5,
127
+ "Ios18.mul" : 70
128
+ }
129
+ },
130
+ {
131
+ "inputSchema" : [
132
+ {
133
+ "hasShapeFlexibility" : "0",
134
+ "isOptional" : "0",
135
+ "dataType" : "Float16",
136
+ "formattedType" : "MultiArray (Float16 1 × 32 × 2560)",
137
+ "shortDescription" : "",
138
+ "shape" : "[1, 32, 2560]",
139
+ "name" : "input_embedding",
140
+ "type" : "MultiArray"
141
+ },
142
+ {
143
+ "hasShapeFlexibility" : "0",
144
+ "isOptional" : "0",
145
+ "dataType" : "Float16",
146
+ "formattedType" : "MultiArray (Float16 1 × 32 × 64)",
147
+ "shortDescription" : "",
148
+ "shape" : "[1, 32, 64]",
149
+ "name" : "cos",
150
+ "type" : "MultiArray"
151
+ },
152
+ {
153
+ "hasShapeFlexibility" : "0",
154
+ "isOptional" : "0",
155
+ "dataType" : "Float16",
156
+ "formattedType" : "MultiArray (Float16 1 × 32 × 64)",
157
+ "shortDescription" : "",
158
+ "shape" : "[1, 32, 64]",
159
+ "name" : "sin",
160
+ "type" : "MultiArray"
161
+ },
162
+ {
163
+ "hasShapeFlexibility" : "0",
164
+ "isOptional" : "0",
165
+ "dataType" : "Float16",
166
+ "formattedType" : "MultiArray (Float16 1 × 1 × 32 × 1024)",
167
+ "shortDescription" : "",
168
+ "shape" : "[1, 1, 32, 1024]",
169
+ "name" : "attention_mask",
170
+ "type" : "MultiArray"
171
+ },
172
+ {
173
+ "hasShapeFlexibility" : "0",
174
+ "isOptional" : "0",
175
+ "dataType" : "Int32",
176
+ "formattedType" : "MultiArray (Int32 1)",
177
+ "shortDescription" : "",
178
+ "shape" : "[1]",
179
+ "name" : "n_past",
180
+ "type" : "MultiArray"
181
+ }
182
+ ],
183
+ "computePrecision" : "Mixed (Float16, Int32)",
184
+ "storagePrecision" : "Float16",
185
+ "stateSchema" : [
186
+ {
187
+ "dataType" : "Float16",
188
+ "isOptional" : "0",
189
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
190
+ "shortDescription" : "",
191
+ "shape" : "[5, 8, 1024, 64]",
192
+ "name" : "key_cache",
193
+ "type" : "State"
194
+ },
195
+ {
196
+ "dataType" : "Float16",
197
+ "isOptional" : "0",
198
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
199
+ "shortDescription" : "",
200
+ "shape" : "[5, 8, 1024, 64]",
201
+ "name" : "value_cache",
202
+ "type" : "State"
203
+ }
204
+ ],
205
+ "outputSchema" : [
206
+ {
207
+ "hasShapeFlexibility" : "0",
208
+ "isOptional" : "0",
209
+ "dataType" : "Float16",
210
+ "formattedType" : "MultiArray (Float16 1 × 32 × 2560)",
211
+ "shortDescription" : "",
212
+ "shape" : "[1, 32, 2560]",
213
+ "name" : "output",
214
+ "type" : "MultiArray"
215
+ }
216
+ ],
217
+ "name" : "ar32",
218
+ "mlProgramOperationTypeHistogram" : {
219
+ "Ios18.readState" : 12,
220
+ "Ios18.conv" : 35,
221
+ "Ios18.expandDims" : 32,
222
+ "Ios18.matmul" : 10,
223
+ "Ios18.concat" : 26,
224
+ "Ios18.add" : 26,
225
+ "Ios18.silu" : 5,
226
+ "Ios18.sliceUpdate" : 10,
227
+ "Ios18.sliceByIndex" : 40,
228
+ "Ios18.layerNorm" : 10,
229
+ "Ios18.transpose" : 45,
230
+ "Tile" : 10,
231
+ "Ios18.writeState" : 10,
232
+ "Ios18.reshape" : 40,
233
+ "Ios18.softmax" : 5,
234
+ "Ios18.squeeze" : 5,
235
+ "Ios18.mul" : 70
236
+ }
237
+ }
238
+ ],
239
+ "mlProgramOperationTypeHistogram" : {
240
+ "Ios18.readState" : 12,
241
+ "Ios18.conv" : 35,
242
+ "Ios18.expandDims" : 32,
243
+ "Ios18.matmul" : 10,
244
+ "Ios18.concat" : 26,
245
+ "Ios18.add" : 26,
246
+ "Ios18.silu" : 5,
247
+ "Ios18.sliceUpdate" : 10,
248
+ "Ios18.sliceByIndex" : 40,
249
+ "Ios18.layerNorm" : 10,
250
+ "Ios18.transpose" : 45,
251
+ "Tile" : 10,
252
+ "Ios18.writeState" : 10,
253
+ "Ios18.reshape" : 40,
254
+ "Ios18.softmax" : 5,
255
+ "Ios18.squeeze" : 5,
256
+ "Ios18.mul" : 70
257
+ },
258
+ "isUpdatable" : "0",
259
+ "stateSchema" : [
260
+ {
261
+ "dataType" : "Float16",
262
+ "isOptional" : "0",
263
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
264
+ "shortDescription" : "",
265
+ "shape" : "[5, 8, 1024, 64]",
266
+ "name" : "key_cache",
267
+ "type" : "State"
268
+ },
269
+ {
270
+ "dataType" : "Float16",
271
+ "isOptional" : "0",
272
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
273
+ "shortDescription" : "",
274
+ "shape" : "[5, 8, 1024, 64]",
275
+ "name" : "value_cache",
276
+ "type" : "State"
277
+ }
278
+ ],
279
+ "availability" : {
280
+ "macOS" : "15.0",
281
+ "tvOS" : "18.0",
282
+ "visionOS" : "2.0",
283
+ "watchOS" : "11.0",
284
+ "iOS" : "18.0",
285
+ "macCatalyst" : "18.0"
286
+ },
287
+ "computePrecision" : "Mixed (Float16, Int32)",
288
+ "modelType" : {
289
+ "name" : "MLModelType_mlProgram"
290
+ },
291
+ "inputSchema" : [
292
+ {
293
+ "hasShapeFlexibility" : "0",
294
+ "isOptional" : "0",
295
+ "dataType" : "Float16",
296
+ "formattedType" : "MultiArray (Float16 1 × 1 × 2560)",
297
+ "shortDescription" : "",
298
+ "shape" : "[1, 1, 2560]",
299
+ "name" : "input_embedding",
300
+ "type" : "MultiArray"
301
+ },
302
+ {
303
+ "hasShapeFlexibility" : "0",
304
+ "isOptional" : "0",
305
+ "dataType" : "Float16",
306
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
307
+ "shortDescription" : "",
308
+ "shape" : "[1, 1, 64]",
309
+ "name" : "cos",
310
+ "type" : "MultiArray"
311
+ },
312
+ {
313
+ "hasShapeFlexibility" : "0",
314
+ "isOptional" : "0",
315
+ "dataType" : "Float16",
316
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
317
+ "shortDescription" : "",
318
+ "shape" : "[1, 1, 64]",
319
+ "name" : "sin",
320
+ "type" : "MultiArray"
321
+ },
322
+ {
323
+ "hasShapeFlexibility" : "0",
324
+ "isOptional" : "0",
325
+ "dataType" : "Float16",
326
+ "formattedType" : "MultiArray (Float16 1 × 1 × 1 × 1024)",
327
+ "shortDescription" : "",
328
+ "shape" : "[1, 1, 1, 1024]",
329
+ "name" : "attention_mask",
330
+ "type" : "MultiArray"
331
+ },
332
+ {
333
+ "hasShapeFlexibility" : "0",
334
+ "isOptional" : "0",
335
+ "dataType" : "Int32",
336
+ "formattedType" : "MultiArray (Int32 1)",
337
+ "shortDescription" : "",
338
+ "shape" : "[1]",
339
+ "name" : "n_past",
340
+ "type" : "MultiArray"
341
+ }
342
+ ],
343
+ "defaultFunctionName" : "ar1",
344
+ "generatedClassName" : "granite_3B_part4_ar1_32",
345
+ "userDefinedMetadata" : {
346
+
347
+ },
348
+ "method" : "predict"
349
+ }
350
+ ]
granite-3B-part4-ar1-32.mlmodelc/model.mil ADDED
The diff for this file is too large to render. See raw diff
 
granite-3B-part4-ar1-32.mlmodelc/weights/weight.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c5aac2afbefc91ea099ef1ca2dbf9d16b145ac61ce60a4941b6b45572bcfe82f
3
+ size 786486144
granite-3B-part5-ar1-32.mlmodelc/analytics/coremldata.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c124ee22bb353127161251e0c4f467faf8714d5d4fee30eefa230413a20a96e4
3
+ size 243
granite-3B-part5-ar1-32.mlmodelc/coremldata.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:61c212ece12e17d28f26f8124a14e9b0d5292b3f86c46a60a3d93b3e371f5926
3
+ size 569
granite-3B-part5-ar1-32.mlmodelc/metadata.json ADDED
@@ -0,0 +1,350 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 × 1 × 2560)",
11
+ "shortDescription" : "",
12
+ "shape" : "[1, 1, 2560]",
13
+ "name" : "output",
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" : "Float16",
28
+ "formattedType" : "MultiArray (Float16 1 × 1 × 2560)",
29
+ "shortDescription" : "",
30
+ "shape" : "[1, 1, 2560]",
31
+ "name" : "input_embedding",
32
+ "type" : "MultiArray"
33
+ },
34
+ {
35
+ "hasShapeFlexibility" : "0",
36
+ "isOptional" : "0",
37
+ "dataType" : "Float16",
38
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
39
+ "shortDescription" : "",
40
+ "shape" : "[1, 1, 64]",
41
+ "name" : "cos",
42
+ "type" : "MultiArray"
43
+ },
44
+ {
45
+ "hasShapeFlexibility" : "0",
46
+ "isOptional" : "0",
47
+ "dataType" : "Float16",
48
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
49
+ "shortDescription" : "",
50
+ "shape" : "[1, 1, 64]",
51
+ "name" : "sin",
52
+ "type" : "MultiArray"
53
+ },
54
+ {
55
+ "hasShapeFlexibility" : "0",
56
+ "isOptional" : "0",
57
+ "dataType" : "Float16",
58
+ "formattedType" : "MultiArray (Float16 1 × 1 × 1 × 1024)",
59
+ "shortDescription" : "",
60
+ "shape" : "[1, 1, 1, 1024]",
61
+ "name" : "attention_mask",
62
+ "type" : "MultiArray"
63
+ },
64
+ {
65
+ "hasShapeFlexibility" : "0",
66
+ "isOptional" : "0",
67
+ "dataType" : "Int32",
68
+ "formattedType" : "MultiArray (Int32 1)",
69
+ "shortDescription" : "",
70
+ "shape" : "[1]",
71
+ "name" : "n_past",
72
+ "type" : "MultiArray"
73
+ }
74
+ ],
75
+ "computePrecision" : "Mixed (Float16, Int32)",
76
+ "storagePrecision" : "Float16",
77
+ "stateSchema" : [
78
+ {
79
+ "dataType" : "Float16",
80
+ "isOptional" : "0",
81
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
82
+ "shortDescription" : "",
83
+ "shape" : "[5, 8, 1024, 64]",
84
+ "name" : "key_cache",
85
+ "type" : "State"
86
+ },
87
+ {
88
+ "dataType" : "Float16",
89
+ "isOptional" : "0",
90
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
91
+ "shortDescription" : "",
92
+ "shape" : "[5, 8, 1024, 64]",
93
+ "name" : "value_cache",
94
+ "type" : "State"
95
+ }
96
+ ],
97
+ "outputSchema" : [
98
+ {
99
+ "hasShapeFlexibility" : "0",
100
+ "isOptional" : "0",
101
+ "dataType" : "Float16",
102
+ "formattedType" : "MultiArray (Float16 1 × 1 × 2560)",
103
+ "shortDescription" : "",
104
+ "shape" : "[1, 1, 2560]",
105
+ "name" : "output",
106
+ "type" : "MultiArray"
107
+ }
108
+ ],
109
+ "name" : "ar1",
110
+ "mlProgramOperationTypeHistogram" : {
111
+ "Ios18.readState" : 12,
112
+ "Ios18.conv" : 35,
113
+ "Ios18.expandDims" : 32,
114
+ "Ios18.matmul" : 10,
115
+ "Ios18.concat" : 26,
116
+ "Ios18.add" : 26,
117
+ "Ios18.silu" : 5,
118
+ "Ios18.sliceUpdate" : 10,
119
+ "Ios18.sliceByIndex" : 40,
120
+ "Ios18.layerNorm" : 10,
121
+ "Ios18.transpose" : 45,
122
+ "Tile" : 10,
123
+ "Ios18.writeState" : 10,
124
+ "Ios18.reshape" : 40,
125
+ "Ios18.softmax" : 5,
126
+ "Ios18.squeeze" : 5,
127
+ "Ios18.mul" : 70
128
+ }
129
+ },
130
+ {
131
+ "inputSchema" : [
132
+ {
133
+ "hasShapeFlexibility" : "0",
134
+ "isOptional" : "0",
135
+ "dataType" : "Float16",
136
+ "formattedType" : "MultiArray (Float16 1 × 32 × 2560)",
137
+ "shortDescription" : "",
138
+ "shape" : "[1, 32, 2560]",
139
+ "name" : "input_embedding",
140
+ "type" : "MultiArray"
141
+ },
142
+ {
143
+ "hasShapeFlexibility" : "0",
144
+ "isOptional" : "0",
145
+ "dataType" : "Float16",
146
+ "formattedType" : "MultiArray (Float16 1 × 32 × 64)",
147
+ "shortDescription" : "",
148
+ "shape" : "[1, 32, 64]",
149
+ "name" : "cos",
150
+ "type" : "MultiArray"
151
+ },
152
+ {
153
+ "hasShapeFlexibility" : "0",
154
+ "isOptional" : "0",
155
+ "dataType" : "Float16",
156
+ "formattedType" : "MultiArray (Float16 1 × 32 × 64)",
157
+ "shortDescription" : "",
158
+ "shape" : "[1, 32, 64]",
159
+ "name" : "sin",
160
+ "type" : "MultiArray"
161
+ },
162
+ {
163
+ "hasShapeFlexibility" : "0",
164
+ "isOptional" : "0",
165
+ "dataType" : "Float16",
166
+ "formattedType" : "MultiArray (Float16 1 × 1 × 32 × 1024)",
167
+ "shortDescription" : "",
168
+ "shape" : "[1, 1, 32, 1024]",
169
+ "name" : "attention_mask",
170
+ "type" : "MultiArray"
171
+ },
172
+ {
173
+ "hasShapeFlexibility" : "0",
174
+ "isOptional" : "0",
175
+ "dataType" : "Int32",
176
+ "formattedType" : "MultiArray (Int32 1)",
177
+ "shortDescription" : "",
178
+ "shape" : "[1]",
179
+ "name" : "n_past",
180
+ "type" : "MultiArray"
181
+ }
182
+ ],
183
+ "computePrecision" : "Mixed (Float16, Int32)",
184
+ "storagePrecision" : "Float16",
185
+ "stateSchema" : [
186
+ {
187
+ "dataType" : "Float16",
188
+ "isOptional" : "0",
189
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
190
+ "shortDescription" : "",
191
+ "shape" : "[5, 8, 1024, 64]",
192
+ "name" : "key_cache",
193
+ "type" : "State"
194
+ },
195
+ {
196
+ "dataType" : "Float16",
197
+ "isOptional" : "0",
198
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
199
+ "shortDescription" : "",
200
+ "shape" : "[5, 8, 1024, 64]",
201
+ "name" : "value_cache",
202
+ "type" : "State"
203
+ }
204
+ ],
205
+ "outputSchema" : [
206
+ {
207
+ "hasShapeFlexibility" : "0",
208
+ "isOptional" : "0",
209
+ "dataType" : "Float16",
210
+ "formattedType" : "MultiArray (Float16 1 × 32 × 2560)",
211
+ "shortDescription" : "",
212
+ "shape" : "[1, 32, 2560]",
213
+ "name" : "output",
214
+ "type" : "MultiArray"
215
+ }
216
+ ],
217
+ "name" : "ar32",
218
+ "mlProgramOperationTypeHistogram" : {
219
+ "Ios18.readState" : 12,
220
+ "Ios18.conv" : 35,
221
+ "Ios18.expandDims" : 32,
222
+ "Ios18.matmul" : 10,
223
+ "Ios18.concat" : 26,
224
+ "Ios18.add" : 26,
225
+ "Ios18.silu" : 5,
226
+ "Ios18.sliceUpdate" : 10,
227
+ "Ios18.sliceByIndex" : 40,
228
+ "Ios18.layerNorm" : 10,
229
+ "Ios18.transpose" : 45,
230
+ "Tile" : 10,
231
+ "Ios18.writeState" : 10,
232
+ "Ios18.reshape" : 40,
233
+ "Ios18.softmax" : 5,
234
+ "Ios18.squeeze" : 5,
235
+ "Ios18.mul" : 70
236
+ }
237
+ }
238
+ ],
239
+ "mlProgramOperationTypeHistogram" : {
240
+ "Ios18.readState" : 12,
241
+ "Ios18.conv" : 35,
242
+ "Ios18.expandDims" : 32,
243
+ "Ios18.matmul" : 10,
244
+ "Ios18.concat" : 26,
245
+ "Ios18.add" : 26,
246
+ "Ios18.silu" : 5,
247
+ "Ios18.sliceUpdate" : 10,
248
+ "Ios18.sliceByIndex" : 40,
249
+ "Ios18.layerNorm" : 10,
250
+ "Ios18.transpose" : 45,
251
+ "Tile" : 10,
252
+ "Ios18.writeState" : 10,
253
+ "Ios18.reshape" : 40,
254
+ "Ios18.softmax" : 5,
255
+ "Ios18.squeeze" : 5,
256
+ "Ios18.mul" : 70
257
+ },
258
+ "isUpdatable" : "0",
259
+ "stateSchema" : [
260
+ {
261
+ "dataType" : "Float16",
262
+ "isOptional" : "0",
263
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
264
+ "shortDescription" : "",
265
+ "shape" : "[5, 8, 1024, 64]",
266
+ "name" : "key_cache",
267
+ "type" : "State"
268
+ },
269
+ {
270
+ "dataType" : "Float16",
271
+ "isOptional" : "0",
272
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
273
+ "shortDescription" : "",
274
+ "shape" : "[5, 8, 1024, 64]",
275
+ "name" : "value_cache",
276
+ "type" : "State"
277
+ }
278
+ ],
279
+ "availability" : {
280
+ "macOS" : "15.0",
281
+ "tvOS" : "18.0",
282
+ "visionOS" : "2.0",
283
+ "watchOS" : "11.0",
284
+ "iOS" : "18.0",
285
+ "macCatalyst" : "18.0"
286
+ },
287
+ "computePrecision" : "Mixed (Float16, Int32)",
288
+ "modelType" : {
289
+ "name" : "MLModelType_mlProgram"
290
+ },
291
+ "inputSchema" : [
292
+ {
293
+ "hasShapeFlexibility" : "0",
294
+ "isOptional" : "0",
295
+ "dataType" : "Float16",
296
+ "formattedType" : "MultiArray (Float16 1 × 1 × 2560)",
297
+ "shortDescription" : "",
298
+ "shape" : "[1, 1, 2560]",
299
+ "name" : "input_embedding",
300
+ "type" : "MultiArray"
301
+ },
302
+ {
303
+ "hasShapeFlexibility" : "0",
304
+ "isOptional" : "0",
305
+ "dataType" : "Float16",
306
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
307
+ "shortDescription" : "",
308
+ "shape" : "[1, 1, 64]",
309
+ "name" : "cos",
310
+ "type" : "MultiArray"
311
+ },
312
+ {
313
+ "hasShapeFlexibility" : "0",
314
+ "isOptional" : "0",
315
+ "dataType" : "Float16",
316
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
317
+ "shortDescription" : "",
318
+ "shape" : "[1, 1, 64]",
319
+ "name" : "sin",
320
+ "type" : "MultiArray"
321
+ },
322
+ {
323
+ "hasShapeFlexibility" : "0",
324
+ "isOptional" : "0",
325
+ "dataType" : "Float16",
326
+ "formattedType" : "MultiArray (Float16 1 × 1 × 1 × 1024)",
327
+ "shortDescription" : "",
328
+ "shape" : "[1, 1, 1, 1024]",
329
+ "name" : "attention_mask",
330
+ "type" : "MultiArray"
331
+ },
332
+ {
333
+ "hasShapeFlexibility" : "0",
334
+ "isOptional" : "0",
335
+ "dataType" : "Int32",
336
+ "formattedType" : "MultiArray (Int32 1)",
337
+ "shortDescription" : "",
338
+ "shape" : "[1]",
339
+ "name" : "n_past",
340
+ "type" : "MultiArray"
341
+ }
342
+ ],
343
+ "defaultFunctionName" : "ar1",
344
+ "generatedClassName" : "granite_3B_part5_ar1_32",
345
+ "userDefinedMetadata" : {
346
+
347
+ },
348
+ "method" : "predict"
349
+ }
350
+ ]
granite-3B-part5-ar1-32.mlmodelc/model.mil ADDED
The diff for this file is too large to render. See raw diff
 
granite-3B-part5-ar1-32.mlmodelc/weights/weight.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3d06df9d51bed053d10c38867b686daf606c4b2268030f936842a3e85738647c
3
+ size 786486144
granite-3B-part6-ar1-32.mlmodelc/analytics/coremldata.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c124ee22bb353127161251e0c4f467faf8714d5d4fee30eefa230413a20a96e4
3
+ size 243
granite-3B-part6-ar1-32.mlmodelc/coremldata.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:61c212ece12e17d28f26f8124a14e9b0d5292b3f86c46a60a3d93b3e371f5926
3
+ size 569
granite-3B-part6-ar1-32.mlmodelc/metadata.json ADDED
@@ -0,0 +1,350 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 × 1 × 2560)",
11
+ "shortDescription" : "",
12
+ "shape" : "[1, 1, 2560]",
13
+ "name" : "output",
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" : "Float16",
28
+ "formattedType" : "MultiArray (Float16 1 × 1 × 2560)",
29
+ "shortDescription" : "",
30
+ "shape" : "[1, 1, 2560]",
31
+ "name" : "input_embedding",
32
+ "type" : "MultiArray"
33
+ },
34
+ {
35
+ "hasShapeFlexibility" : "0",
36
+ "isOptional" : "0",
37
+ "dataType" : "Float16",
38
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
39
+ "shortDescription" : "",
40
+ "shape" : "[1, 1, 64]",
41
+ "name" : "cos",
42
+ "type" : "MultiArray"
43
+ },
44
+ {
45
+ "hasShapeFlexibility" : "0",
46
+ "isOptional" : "0",
47
+ "dataType" : "Float16",
48
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
49
+ "shortDescription" : "",
50
+ "shape" : "[1, 1, 64]",
51
+ "name" : "sin",
52
+ "type" : "MultiArray"
53
+ },
54
+ {
55
+ "hasShapeFlexibility" : "0",
56
+ "isOptional" : "0",
57
+ "dataType" : "Float16",
58
+ "formattedType" : "MultiArray (Float16 1 × 1 × 1 × 1024)",
59
+ "shortDescription" : "",
60
+ "shape" : "[1, 1, 1, 1024]",
61
+ "name" : "attention_mask",
62
+ "type" : "MultiArray"
63
+ },
64
+ {
65
+ "hasShapeFlexibility" : "0",
66
+ "isOptional" : "0",
67
+ "dataType" : "Int32",
68
+ "formattedType" : "MultiArray (Int32 1)",
69
+ "shortDescription" : "",
70
+ "shape" : "[1]",
71
+ "name" : "n_past",
72
+ "type" : "MultiArray"
73
+ }
74
+ ],
75
+ "computePrecision" : "Mixed (Float16, Int32)",
76
+ "storagePrecision" : "Float16",
77
+ "stateSchema" : [
78
+ {
79
+ "dataType" : "Float16",
80
+ "isOptional" : "0",
81
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
82
+ "shortDescription" : "",
83
+ "shape" : "[5, 8, 1024, 64]",
84
+ "name" : "key_cache",
85
+ "type" : "State"
86
+ },
87
+ {
88
+ "dataType" : "Float16",
89
+ "isOptional" : "0",
90
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
91
+ "shortDescription" : "",
92
+ "shape" : "[5, 8, 1024, 64]",
93
+ "name" : "value_cache",
94
+ "type" : "State"
95
+ }
96
+ ],
97
+ "outputSchema" : [
98
+ {
99
+ "hasShapeFlexibility" : "0",
100
+ "isOptional" : "0",
101
+ "dataType" : "Float16",
102
+ "formattedType" : "MultiArray (Float16 1 × 1 × 2560)",
103
+ "shortDescription" : "",
104
+ "shape" : "[1, 1, 2560]",
105
+ "name" : "output",
106
+ "type" : "MultiArray"
107
+ }
108
+ ],
109
+ "name" : "ar1",
110
+ "mlProgramOperationTypeHistogram" : {
111
+ "Ios18.readState" : 12,
112
+ "Ios18.conv" : 35,
113
+ "Ios18.expandDims" : 32,
114
+ "Ios18.matmul" : 10,
115
+ "Ios18.concat" : 26,
116
+ "Ios18.add" : 26,
117
+ "Ios18.silu" : 5,
118
+ "Ios18.sliceUpdate" : 10,
119
+ "Ios18.sliceByIndex" : 40,
120
+ "Ios18.layerNorm" : 10,
121
+ "Ios18.transpose" : 45,
122
+ "Tile" : 10,
123
+ "Ios18.writeState" : 10,
124
+ "Ios18.reshape" : 40,
125
+ "Ios18.softmax" : 5,
126
+ "Ios18.squeeze" : 5,
127
+ "Ios18.mul" : 70
128
+ }
129
+ },
130
+ {
131
+ "inputSchema" : [
132
+ {
133
+ "hasShapeFlexibility" : "0",
134
+ "isOptional" : "0",
135
+ "dataType" : "Float16",
136
+ "formattedType" : "MultiArray (Float16 1 × 32 × 2560)",
137
+ "shortDescription" : "",
138
+ "shape" : "[1, 32, 2560]",
139
+ "name" : "input_embedding",
140
+ "type" : "MultiArray"
141
+ },
142
+ {
143
+ "hasShapeFlexibility" : "0",
144
+ "isOptional" : "0",
145
+ "dataType" : "Float16",
146
+ "formattedType" : "MultiArray (Float16 1 × 32 × 64)",
147
+ "shortDescription" : "",
148
+ "shape" : "[1, 32, 64]",
149
+ "name" : "cos",
150
+ "type" : "MultiArray"
151
+ },
152
+ {
153
+ "hasShapeFlexibility" : "0",
154
+ "isOptional" : "0",
155
+ "dataType" : "Float16",
156
+ "formattedType" : "MultiArray (Float16 1 × 32 × 64)",
157
+ "shortDescription" : "",
158
+ "shape" : "[1, 32, 64]",
159
+ "name" : "sin",
160
+ "type" : "MultiArray"
161
+ },
162
+ {
163
+ "hasShapeFlexibility" : "0",
164
+ "isOptional" : "0",
165
+ "dataType" : "Float16",
166
+ "formattedType" : "MultiArray (Float16 1 × 1 × 32 × 1024)",
167
+ "shortDescription" : "",
168
+ "shape" : "[1, 1, 32, 1024]",
169
+ "name" : "attention_mask",
170
+ "type" : "MultiArray"
171
+ },
172
+ {
173
+ "hasShapeFlexibility" : "0",
174
+ "isOptional" : "0",
175
+ "dataType" : "Int32",
176
+ "formattedType" : "MultiArray (Int32 1)",
177
+ "shortDescription" : "",
178
+ "shape" : "[1]",
179
+ "name" : "n_past",
180
+ "type" : "MultiArray"
181
+ }
182
+ ],
183
+ "computePrecision" : "Mixed (Float16, Int32)",
184
+ "storagePrecision" : "Float16",
185
+ "stateSchema" : [
186
+ {
187
+ "dataType" : "Float16",
188
+ "isOptional" : "0",
189
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
190
+ "shortDescription" : "",
191
+ "shape" : "[5, 8, 1024, 64]",
192
+ "name" : "key_cache",
193
+ "type" : "State"
194
+ },
195
+ {
196
+ "dataType" : "Float16",
197
+ "isOptional" : "0",
198
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
199
+ "shortDescription" : "",
200
+ "shape" : "[5, 8, 1024, 64]",
201
+ "name" : "value_cache",
202
+ "type" : "State"
203
+ }
204
+ ],
205
+ "outputSchema" : [
206
+ {
207
+ "hasShapeFlexibility" : "0",
208
+ "isOptional" : "0",
209
+ "dataType" : "Float16",
210
+ "formattedType" : "MultiArray (Float16 1 × 32 × 2560)",
211
+ "shortDescription" : "",
212
+ "shape" : "[1, 32, 2560]",
213
+ "name" : "output",
214
+ "type" : "MultiArray"
215
+ }
216
+ ],
217
+ "name" : "ar32",
218
+ "mlProgramOperationTypeHistogram" : {
219
+ "Ios18.readState" : 12,
220
+ "Ios18.conv" : 35,
221
+ "Ios18.expandDims" : 32,
222
+ "Ios18.matmul" : 10,
223
+ "Ios18.concat" : 26,
224
+ "Ios18.add" : 26,
225
+ "Ios18.silu" : 5,
226
+ "Ios18.sliceUpdate" : 10,
227
+ "Ios18.sliceByIndex" : 40,
228
+ "Ios18.layerNorm" : 10,
229
+ "Ios18.transpose" : 45,
230
+ "Tile" : 10,
231
+ "Ios18.writeState" : 10,
232
+ "Ios18.reshape" : 40,
233
+ "Ios18.softmax" : 5,
234
+ "Ios18.squeeze" : 5,
235
+ "Ios18.mul" : 70
236
+ }
237
+ }
238
+ ],
239
+ "mlProgramOperationTypeHistogram" : {
240
+ "Ios18.readState" : 12,
241
+ "Ios18.conv" : 35,
242
+ "Ios18.expandDims" : 32,
243
+ "Ios18.matmul" : 10,
244
+ "Ios18.concat" : 26,
245
+ "Ios18.add" : 26,
246
+ "Ios18.silu" : 5,
247
+ "Ios18.sliceUpdate" : 10,
248
+ "Ios18.sliceByIndex" : 40,
249
+ "Ios18.layerNorm" : 10,
250
+ "Ios18.transpose" : 45,
251
+ "Tile" : 10,
252
+ "Ios18.writeState" : 10,
253
+ "Ios18.reshape" : 40,
254
+ "Ios18.softmax" : 5,
255
+ "Ios18.squeeze" : 5,
256
+ "Ios18.mul" : 70
257
+ },
258
+ "isUpdatable" : "0",
259
+ "stateSchema" : [
260
+ {
261
+ "dataType" : "Float16",
262
+ "isOptional" : "0",
263
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
264
+ "shortDescription" : "",
265
+ "shape" : "[5, 8, 1024, 64]",
266
+ "name" : "key_cache",
267
+ "type" : "State"
268
+ },
269
+ {
270
+ "dataType" : "Float16",
271
+ "isOptional" : "0",
272
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
273
+ "shortDescription" : "",
274
+ "shape" : "[5, 8, 1024, 64]",
275
+ "name" : "value_cache",
276
+ "type" : "State"
277
+ }
278
+ ],
279
+ "availability" : {
280
+ "macOS" : "15.0",
281
+ "tvOS" : "18.0",
282
+ "visionOS" : "2.0",
283
+ "watchOS" : "11.0",
284
+ "iOS" : "18.0",
285
+ "macCatalyst" : "18.0"
286
+ },
287
+ "computePrecision" : "Mixed (Float16, Int32)",
288
+ "modelType" : {
289
+ "name" : "MLModelType_mlProgram"
290
+ },
291
+ "inputSchema" : [
292
+ {
293
+ "hasShapeFlexibility" : "0",
294
+ "isOptional" : "0",
295
+ "dataType" : "Float16",
296
+ "formattedType" : "MultiArray (Float16 1 × 1 × 2560)",
297
+ "shortDescription" : "",
298
+ "shape" : "[1, 1, 2560]",
299
+ "name" : "input_embedding",
300
+ "type" : "MultiArray"
301
+ },
302
+ {
303
+ "hasShapeFlexibility" : "0",
304
+ "isOptional" : "0",
305
+ "dataType" : "Float16",
306
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
307
+ "shortDescription" : "",
308
+ "shape" : "[1, 1, 64]",
309
+ "name" : "cos",
310
+ "type" : "MultiArray"
311
+ },
312
+ {
313
+ "hasShapeFlexibility" : "0",
314
+ "isOptional" : "0",
315
+ "dataType" : "Float16",
316
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
317
+ "shortDescription" : "",
318
+ "shape" : "[1, 1, 64]",
319
+ "name" : "sin",
320
+ "type" : "MultiArray"
321
+ },
322
+ {
323
+ "hasShapeFlexibility" : "0",
324
+ "isOptional" : "0",
325
+ "dataType" : "Float16",
326
+ "formattedType" : "MultiArray (Float16 1 × 1 × 1 × 1024)",
327
+ "shortDescription" : "",
328
+ "shape" : "[1, 1, 1, 1024]",
329
+ "name" : "attention_mask",
330
+ "type" : "MultiArray"
331
+ },
332
+ {
333
+ "hasShapeFlexibility" : "0",
334
+ "isOptional" : "0",
335
+ "dataType" : "Int32",
336
+ "formattedType" : "MultiArray (Int32 1)",
337
+ "shortDescription" : "",
338
+ "shape" : "[1]",
339
+ "name" : "n_past",
340
+ "type" : "MultiArray"
341
+ }
342
+ ],
343
+ "defaultFunctionName" : "ar1",
344
+ "generatedClassName" : "granite_3B_part6_ar1_32",
345
+ "userDefinedMetadata" : {
346
+
347
+ },
348
+ "method" : "predict"
349
+ }
350
+ ]
granite-3B-part6-ar1-32.mlmodelc/model.mil ADDED
The diff for this file is too large to render. See raw diff
 
granite-3B-part6-ar1-32.mlmodelc/weights/weight.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e25bae3dc241559d0a5dee5438dcde6e1871eccba6fa141137a5f55742e8eea4
3
+ size 786486144
granite-3B-part7-ar1-32.mlmodelc/analytics/coremldata.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ba1433907b3aee8fc85e1f4900413f795e54b54b845ed429137cfe9e980e3454
3
+ size 243
granite-3B-part7-ar1-32.mlmodelc/coremldata.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2e98d891c9bc35f9515aaac8966fa5648eba485949c9b5feb666eda70c68119d
3
+ size 571
granite-3B-part7-ar1-32.mlmodelc/metadata.json ADDED
@@ -0,0 +1,350 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 × 1 × 100352)",
11
+ "shortDescription" : "",
12
+ "shape" : "[1, 1, 100352]",
13
+ "name" : "output",
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" : "Float16",
28
+ "formattedType" : "MultiArray (Float16 1 × 1 × 2560)",
29
+ "shortDescription" : "",
30
+ "shape" : "[1, 1, 2560]",
31
+ "name" : "input_embedding",
32
+ "type" : "MultiArray"
33
+ },
34
+ {
35
+ "hasShapeFlexibility" : "0",
36
+ "isOptional" : "0",
37
+ "dataType" : "Float16",
38
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
39
+ "shortDescription" : "",
40
+ "shape" : "[1, 1, 64]",
41
+ "name" : "cos",
42
+ "type" : "MultiArray"
43
+ },
44
+ {
45
+ "hasShapeFlexibility" : "0",
46
+ "isOptional" : "0",
47
+ "dataType" : "Float16",
48
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
49
+ "shortDescription" : "",
50
+ "shape" : "[1, 1, 64]",
51
+ "name" : "sin",
52
+ "type" : "MultiArray"
53
+ },
54
+ {
55
+ "hasShapeFlexibility" : "0",
56
+ "isOptional" : "0",
57
+ "dataType" : "Float16",
58
+ "formattedType" : "MultiArray (Float16 1 × 1 × 1 × 1024)",
59
+ "shortDescription" : "",
60
+ "shape" : "[1, 1, 1, 1024]",
61
+ "name" : "attention_mask",
62
+ "type" : "MultiArray"
63
+ },
64
+ {
65
+ "hasShapeFlexibility" : "0",
66
+ "isOptional" : "0",
67
+ "dataType" : "Int32",
68
+ "formattedType" : "MultiArray (Int32 1)",
69
+ "shortDescription" : "",
70
+ "shape" : "[1]",
71
+ "name" : "n_past",
72
+ "type" : "MultiArray"
73
+ }
74
+ ],
75
+ "computePrecision" : "Mixed (Float16, Int32)",
76
+ "storagePrecision" : "Float16",
77
+ "stateSchema" : [
78
+ {
79
+ "dataType" : "Float16",
80
+ "isOptional" : "0",
81
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
82
+ "shortDescription" : "",
83
+ "shape" : "[5, 8, 1024, 64]",
84
+ "name" : "key_cache",
85
+ "type" : "State"
86
+ },
87
+ {
88
+ "dataType" : "Float16",
89
+ "isOptional" : "0",
90
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
91
+ "shortDescription" : "",
92
+ "shape" : "[5, 8, 1024, 64]",
93
+ "name" : "value_cache",
94
+ "type" : "State"
95
+ }
96
+ ],
97
+ "outputSchema" : [
98
+ {
99
+ "hasShapeFlexibility" : "0",
100
+ "isOptional" : "0",
101
+ "dataType" : "Float16",
102
+ "formattedType" : "MultiArray (Float16 1 × 1 × 100352)",
103
+ "shortDescription" : "",
104
+ "shape" : "[1, 1, 100352]",
105
+ "name" : "output",
106
+ "type" : "MultiArray"
107
+ }
108
+ ],
109
+ "name" : "ar1",
110
+ "mlProgramOperationTypeHistogram" : {
111
+ "Ios18.readState" : 12,
112
+ "Ios18.conv" : 51,
113
+ "Ios18.expandDims" : 33,
114
+ "Ios18.matmul" : 10,
115
+ "Ios18.concat" : 28,
116
+ "Ios18.add" : 26,
117
+ "Ios18.silu" : 5,
118
+ "Ios18.sliceUpdate" : 10,
119
+ "Ios18.sliceByIndex" : 41,
120
+ "Ios18.layerNorm" : 11,
121
+ "Ios18.transpose" : 47,
122
+ "Tile" : 10,
123
+ "Ios18.writeState" : 10,
124
+ "Ios18.reshape" : 40,
125
+ "Ios18.softmax" : 5,
126
+ "Ios18.squeeze" : 21,
127
+ "Ios18.mul" : 73
128
+ }
129
+ },
130
+ {
131
+ "inputSchema" : [
132
+ {
133
+ "hasShapeFlexibility" : "0",
134
+ "isOptional" : "0",
135
+ "dataType" : "Float16",
136
+ "formattedType" : "MultiArray (Float16 1 × 32 × 2560)",
137
+ "shortDescription" : "",
138
+ "shape" : "[1, 32, 2560]",
139
+ "name" : "input_embedding",
140
+ "type" : "MultiArray"
141
+ },
142
+ {
143
+ "hasShapeFlexibility" : "0",
144
+ "isOptional" : "0",
145
+ "dataType" : "Float16",
146
+ "formattedType" : "MultiArray (Float16 1 × 32 × 64)",
147
+ "shortDescription" : "",
148
+ "shape" : "[1, 32, 64]",
149
+ "name" : "cos",
150
+ "type" : "MultiArray"
151
+ },
152
+ {
153
+ "hasShapeFlexibility" : "0",
154
+ "isOptional" : "0",
155
+ "dataType" : "Float16",
156
+ "formattedType" : "MultiArray (Float16 1 × 32 × 64)",
157
+ "shortDescription" : "",
158
+ "shape" : "[1, 32, 64]",
159
+ "name" : "sin",
160
+ "type" : "MultiArray"
161
+ },
162
+ {
163
+ "hasShapeFlexibility" : "0",
164
+ "isOptional" : "0",
165
+ "dataType" : "Float16",
166
+ "formattedType" : "MultiArray (Float16 1 × 1 × 32 × 1024)",
167
+ "shortDescription" : "",
168
+ "shape" : "[1, 1, 32, 1024]",
169
+ "name" : "attention_mask",
170
+ "type" : "MultiArray"
171
+ },
172
+ {
173
+ "hasShapeFlexibility" : "0",
174
+ "isOptional" : "0",
175
+ "dataType" : "Int32",
176
+ "formattedType" : "MultiArray (Int32 1)",
177
+ "shortDescription" : "",
178
+ "shape" : "[1]",
179
+ "name" : "n_past",
180
+ "type" : "MultiArray"
181
+ }
182
+ ],
183
+ "computePrecision" : "Mixed (Float16, Int32)",
184
+ "storagePrecision" : "Float16",
185
+ "stateSchema" : [
186
+ {
187
+ "dataType" : "Float16",
188
+ "isOptional" : "0",
189
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
190
+ "shortDescription" : "",
191
+ "shape" : "[5, 8, 1024, 64]",
192
+ "name" : "key_cache",
193
+ "type" : "State"
194
+ },
195
+ {
196
+ "dataType" : "Float16",
197
+ "isOptional" : "0",
198
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
199
+ "shortDescription" : "",
200
+ "shape" : "[5, 8, 1024, 64]",
201
+ "name" : "value_cache",
202
+ "type" : "State"
203
+ }
204
+ ],
205
+ "outputSchema" : [
206
+ {
207
+ "hasShapeFlexibility" : "0",
208
+ "isOptional" : "0",
209
+ "dataType" : "Float16",
210
+ "formattedType" : "MultiArray (Float16 1 × 32 × 100352)",
211
+ "shortDescription" : "",
212
+ "shape" : "[1, 32, 100352]",
213
+ "name" : "output",
214
+ "type" : "MultiArray"
215
+ }
216
+ ],
217
+ "name" : "ar32",
218
+ "mlProgramOperationTypeHistogram" : {
219
+ "Ios18.readState" : 12,
220
+ "Ios18.conv" : 51,
221
+ "Ios18.expandDims" : 33,
222
+ "Ios18.matmul" : 10,
223
+ "Ios18.concat" : 28,
224
+ "Ios18.add" : 26,
225
+ "Ios18.silu" : 5,
226
+ "Ios18.sliceUpdate" : 10,
227
+ "Ios18.sliceByIndex" : 41,
228
+ "Ios18.layerNorm" : 11,
229
+ "Ios18.transpose" : 47,
230
+ "Tile" : 10,
231
+ "Ios18.writeState" : 10,
232
+ "Ios18.reshape" : 40,
233
+ "Ios18.softmax" : 5,
234
+ "Ios18.squeeze" : 21,
235
+ "Ios18.mul" : 73
236
+ }
237
+ }
238
+ ],
239
+ "mlProgramOperationTypeHistogram" : {
240
+ "Ios18.readState" : 12,
241
+ "Ios18.conv" : 51,
242
+ "Ios18.expandDims" : 33,
243
+ "Ios18.matmul" : 10,
244
+ "Ios18.concat" : 28,
245
+ "Ios18.add" : 26,
246
+ "Ios18.silu" : 5,
247
+ "Ios18.sliceUpdate" : 10,
248
+ "Ios18.sliceByIndex" : 41,
249
+ "Ios18.layerNorm" : 11,
250
+ "Ios18.transpose" : 47,
251
+ "Tile" : 10,
252
+ "Ios18.writeState" : 10,
253
+ "Ios18.reshape" : 40,
254
+ "Ios18.softmax" : 5,
255
+ "Ios18.squeeze" : 21,
256
+ "Ios18.mul" : 73
257
+ },
258
+ "isUpdatable" : "0",
259
+ "stateSchema" : [
260
+ {
261
+ "dataType" : "Float16",
262
+ "isOptional" : "0",
263
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
264
+ "shortDescription" : "",
265
+ "shape" : "[5, 8, 1024, 64]",
266
+ "name" : "key_cache",
267
+ "type" : "State"
268
+ },
269
+ {
270
+ "dataType" : "Float16",
271
+ "isOptional" : "0",
272
+ "formattedType" : "State (Float16 5 × 8 × 1024 × 64)",
273
+ "shortDescription" : "",
274
+ "shape" : "[5, 8, 1024, 64]",
275
+ "name" : "value_cache",
276
+ "type" : "State"
277
+ }
278
+ ],
279
+ "availability" : {
280
+ "macOS" : "15.0",
281
+ "tvOS" : "18.0",
282
+ "visionOS" : "2.0",
283
+ "watchOS" : "11.0",
284
+ "iOS" : "18.0",
285
+ "macCatalyst" : "18.0"
286
+ },
287
+ "computePrecision" : "Mixed (Float16, Int32)",
288
+ "modelType" : {
289
+ "name" : "MLModelType_mlProgram"
290
+ },
291
+ "inputSchema" : [
292
+ {
293
+ "hasShapeFlexibility" : "0",
294
+ "isOptional" : "0",
295
+ "dataType" : "Float16",
296
+ "formattedType" : "MultiArray (Float16 1 × 1 × 2560)",
297
+ "shortDescription" : "",
298
+ "shape" : "[1, 1, 2560]",
299
+ "name" : "input_embedding",
300
+ "type" : "MultiArray"
301
+ },
302
+ {
303
+ "hasShapeFlexibility" : "0",
304
+ "isOptional" : "0",
305
+ "dataType" : "Float16",
306
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
307
+ "shortDescription" : "",
308
+ "shape" : "[1, 1, 64]",
309
+ "name" : "cos",
310
+ "type" : "MultiArray"
311
+ },
312
+ {
313
+ "hasShapeFlexibility" : "0",
314
+ "isOptional" : "0",
315
+ "dataType" : "Float16",
316
+ "formattedType" : "MultiArray (Float16 1 × 1 × 64)",
317
+ "shortDescription" : "",
318
+ "shape" : "[1, 1, 64]",
319
+ "name" : "sin",
320
+ "type" : "MultiArray"
321
+ },
322
+ {
323
+ "hasShapeFlexibility" : "0",
324
+ "isOptional" : "0",
325
+ "dataType" : "Float16",
326
+ "formattedType" : "MultiArray (Float16 1 × 1 × 1 × 1024)",
327
+ "shortDescription" : "",
328
+ "shape" : "[1, 1, 1, 1024]",
329
+ "name" : "attention_mask",
330
+ "type" : "MultiArray"
331
+ },
332
+ {
333
+ "hasShapeFlexibility" : "0",
334
+ "isOptional" : "0",
335
+ "dataType" : "Int32",
336
+ "formattedType" : "MultiArray (Int32 1)",
337
+ "shortDescription" : "",
338
+ "shape" : "[1]",
339
+ "name" : "n_past",
340
+ "type" : "MultiArray"
341
+ }
342
+ ],
343
+ "defaultFunctionName" : "ar1",
344
+ "generatedClassName" : "granite_3B_part7_ar1_32",
345
+ "userDefinedMetadata" : {
346
+
347
+ },
348
+ "method" : "predict"
349
+ }
350
+ ]
granite-3B-part7-ar1-32.mlmodelc/model.mil ADDED
The diff for this file is too large to render. See raw diff
 
granite-3B-part7-ar1-32.mlmodelc/weights/weight.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2c701674db5da87b40a6f9419c68b2f7414438d40fd0084550a2cc999af1fd80
3
+ size 1300294592
utils/.cache/huggingface/.gitignore ADDED
@@ -0,0 +1 @@
 
 
1
+ *
utils/.cache/huggingface/download/utils/chat_template.jinja.lock ADDED
File without changes
utils/.cache/huggingface/download/utils/chat_template.jinja.metadata ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ d5d3069acd55cade34a52ea2a7f7a83dd55068e4
2
+ 82e3ebf57b6de47c0c4e722dde0e33915f73bfe8
3
+ 1763350054.430618
utils/.cache/huggingface/download/utils/config.json.lock ADDED
File without changes
utils/.cache/huggingface/download/utils/config.json.metadata ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ d5d3069acd55cade34a52ea2a7f7a83dd55068e4
2
+ c85e9b542219a1b84c51bc8ff54a8b1e0d799727
3
+ 1763350054.4349592
utils/.cache/huggingface/download/utils/embed.pt.lock ADDED
File without changes
utils/.cache/huggingface/download/utils/embed.pt.metadata ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ d5d3069acd55cade34a52ea2a7f7a83dd55068e4
2
+ a0660658c184d5a7017b2bad9186f27057f08dc2860393241827f6327879960b
3
+ 1763350062.448552
utils/.cache/huggingface/download/utils/merges.txt.lock ADDED
File without changes
utils/.cache/huggingface/download/utils/merges.txt.metadata ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ d5d3069acd55cade34a52ea2a7f7a83dd55068e4
2
+ 354558edcdbd64ca7abd407b8be3d5d09d39d781
3
+ 1763350054.652089
utils/.cache/huggingface/download/utils/special_tokens_map.json.lock ADDED
File without changes