seba commited on
Commit
b85f3a9
·
verified ·
1 Parent(s): 99441fc

Upload residual_lm_fused_multifunction.mlmodelc

Browse files
residual_lm_fused_multifunction.mlmodelc/analytics/coremldata.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e1dd7cb3c97bc674697e875d6674480f8234f0ee08b5c97199e8713d3114b5d7
3
+ size 243
residual_lm_fused_multifunction.mlmodelc/coremldata.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7c3935cfc9117e95167c40d483e3cb260e73dc62199628ae0eb1abad051f2164
3
+ size 1784
residual_lm_fused_multifunction.mlmodelc/metadata.json ADDED
@@ -0,0 +1,788 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 1 × 2048 × 1 × 1)",
11
+ "shortDescription" : "",
12
+ "shape" : "[1, 2048, 1, 1]",
13
+ "name" : "hidden_states",
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 × 4096 × 1 × 1)",
29
+ "shortDescription" : "",
30
+ "shape" : "[1, 4096, 1, 1]",
31
+ "name" : "inputs_embeds",
32
+ "type" : "MultiArray"
33
+ },
34
+ {
35
+ "hasShapeFlexibility" : "0",
36
+ "isOptional" : "0",
37
+ "dataType" : "Int32",
38
+ "formattedType" : "MultiArray (Int32 1)",
39
+ "shortDescription" : "",
40
+ "shape" : "[1]",
41
+ "name" : "position_id",
42
+ "type" : "MultiArray"
43
+ },
44
+ {
45
+ "hasShapeFlexibility" : "0",
46
+ "isOptional" : "0",
47
+ "dataType" : "Int32",
48
+ "formattedType" : "MultiArray (Int32 1)",
49
+ "shortDescription" : "",
50
+ "shape" : "[1]",
51
+ "name" : "position_index_seed",
52
+ "type" : "MultiArray"
53
+ }
54
+ ],
55
+ "computePrecision" : "Mixed (Float16, Int32)",
56
+ "storagePrecision" : "Mixed (Float16, Int8)",
57
+ "stateSchema" : [
58
+ {
59
+ "dataType" : "Float16",
60
+ "isOptional" : "0",
61
+ "formattedType" : "State (Float16 8 × 2 × 2048 × 128)",
62
+ "shortDescription" : "",
63
+ "shape" : "[8, 2, 2048, 128]",
64
+ "name" : "key_cache",
65
+ "type" : "State"
66
+ },
67
+ {
68
+ "dataType" : "Float16",
69
+ "isOptional" : "0",
70
+ "formattedType" : "State (Float16 8 × 2 × 2048 × 128)",
71
+ "shortDescription" : "",
72
+ "shape" : "[8, 2, 2048, 128]",
73
+ "name" : "value_cache",
74
+ "type" : "State"
75
+ }
76
+ ],
77
+ "outputSchema" : [
78
+ {
79
+ "hasShapeFlexibility" : "0",
80
+ "isOptional" : "0",
81
+ "dataType" : "Float16",
82
+ "formattedType" : "MultiArray (Float16 1 × 2048 × 1 × 1)",
83
+ "shortDescription" : "",
84
+ "shape" : "[1, 2048, 1, 1]",
85
+ "name" : "hidden_states",
86
+ "type" : "MultiArray"
87
+ }
88
+ ],
89
+ "name" : "main",
90
+ "mlProgramOperationTypeHistogram" : {
91
+ "Ios18.expandDims" : 3,
92
+ "Ios18.mul" : 41,
93
+ "Ios18.softmax" : 16,
94
+ "Ios18.matmul" : 32,
95
+ "Ios18.equal" : 1,
96
+ "Split" : 41,
97
+ "Select" : 1,
98
+ "Ios18.greater" : 1,
99
+ "Ios18.readState" : 18,
100
+ "Ios18.add" : 34,
101
+ "Ios18.layerNorm" : 17,
102
+ "Ios16.fillLike" : 1,
103
+ "Ios18.reshape" : 32,
104
+ "Ios18.sliceUpdate" : 16,
105
+ "Ios18.writeState" : 16,
106
+ "Ios18.conv" : 57,
107
+ "Ios18.concat" : 41,
108
+ "Ios18.transpose" : 25,
109
+ "Ios18.sub" : 1,
110
+ "Ios18.cast" : 3,
111
+ "Ios18.silu" : 8,
112
+ "Ios18.sliceByIndex" : 16,
113
+ "Ios16.cumsum" : 1,
114
+ "Ios18.constexprBlockwiseShiftScale" : 22
115
+ }
116
+ },
117
+ {
118
+ "inputSchema" : [
119
+ {
120
+ "hasShapeFlexibility" : "0",
121
+ "isOptional" : "0",
122
+ "dataType" : "Float32",
123
+ "formattedType" : "MultiArray (Float32 1 × 4096 × 1 × 1)",
124
+ "shortDescription" : "",
125
+ "shape" : "[1, 4096, 1, 1]",
126
+ "name" : "inputs_embeds",
127
+ "type" : "MultiArray"
128
+ },
129
+ {
130
+ "hasShapeFlexibility" : "0",
131
+ "isOptional" : "0",
132
+ "dataType" : "Int32",
133
+ "formattedType" : "MultiArray (Int32 1)",
134
+ "shortDescription" : "",
135
+ "shape" : "[1]",
136
+ "name" : "position_id",
137
+ "type" : "MultiArray"
138
+ },
139
+ {
140
+ "hasShapeFlexibility" : "0",
141
+ "isOptional" : "0",
142
+ "dataType" : "Int32",
143
+ "formattedType" : "MultiArray (Int32 1)",
144
+ "shortDescription" : "",
145
+ "shape" : "[1]",
146
+ "name" : "position_index_seed",
147
+ "type" : "MultiArray"
148
+ }
149
+ ],
150
+ "computePrecision" : "Mixed (Float16, Int32)",
151
+ "storagePrecision" : "Mixed (Float16, Int8)",
152
+ "stateSchema" : [
153
+ {
154
+ "dataType" : "Float16",
155
+ "isOptional" : "0",
156
+ "formattedType" : "State (Float16 8 × 2 × 2048 × 128)",
157
+ "shortDescription" : "",
158
+ "shape" : "[8, 2, 2048, 128]",
159
+ "name" : "key_cache",
160
+ "type" : "State"
161
+ },
162
+ {
163
+ "dataType" : "Float16",
164
+ "isOptional" : "0",
165
+ "formattedType" : "State (Float16 8 × 2 × 2048 × 128)",
166
+ "shortDescription" : "",
167
+ "shape" : "[8, 2, 2048, 128]",
168
+ "name" : "value_cache",
169
+ "type" : "State"
170
+ }
171
+ ],
172
+ "outputSchema" : [
173
+ {
174
+ "hasShapeFlexibility" : "0",
175
+ "isOptional" : "0",
176
+ "dataType" : "Float16",
177
+ "formattedType" : "MultiArray (Float16 1 × 2048 × 1 × 1)",
178
+ "shortDescription" : "",
179
+ "shape" : "[1, 2048, 1, 1]",
180
+ "name" : "hidden_states",
181
+ "type" : "MultiArray"
182
+ }
183
+ ],
184
+ "name" : "length_1",
185
+ "mlProgramOperationTypeHistogram" : {
186
+ "Ios18.expandDims" : 3,
187
+ "Ios18.mul" : 41,
188
+ "Ios18.softmax" : 16,
189
+ "Ios18.matmul" : 32,
190
+ "Ios18.equal" : 1,
191
+ "Split" : 41,
192
+ "Select" : 1,
193
+ "Ios18.greater" : 1,
194
+ "Ios18.readState" : 18,
195
+ "Ios18.add" : 34,
196
+ "Ios18.layerNorm" : 17,
197
+ "Ios16.fillLike" : 1,
198
+ "Ios18.reshape" : 32,
199
+ "Ios18.sliceUpdate" : 16,
200
+ "Ios18.writeState" : 16,
201
+ "Ios18.conv" : 57,
202
+ "Ios18.concat" : 41,
203
+ "Ios18.transpose" : 25,
204
+ "Ios18.sub" : 1,
205
+ "Ios18.cast" : 3,
206
+ "Ios18.silu" : 8,
207
+ "Ios18.sliceByIndex" : 16,
208
+ "Ios16.cumsum" : 1,
209
+ "Ios18.constexprBlockwiseShiftScale" : 22
210
+ }
211
+ },
212
+ {
213
+ "inputSchema" : [
214
+ {
215
+ "hasShapeFlexibility" : "0",
216
+ "isOptional" : "0",
217
+ "dataType" : "Float32",
218
+ "formattedType" : "MultiArray (Float32 1 × 4096 × 1 × 8)",
219
+ "shortDescription" : "",
220
+ "shape" : "[1, 4096, 1, 8]",
221
+ "name" : "inputs_embeds",
222
+ "type" : "MultiArray"
223
+ },
224
+ {
225
+ "hasShapeFlexibility" : "0",
226
+ "isOptional" : "0",
227
+ "dataType" : "Int32",
228
+ "formattedType" : "MultiArray (Int32 1)",
229
+ "shortDescription" : "",
230
+ "shape" : "[1]",
231
+ "name" : "position_id",
232
+ "type" : "MultiArray"
233
+ },
234
+ {
235
+ "hasShapeFlexibility" : "0",
236
+ "isOptional" : "0",
237
+ "dataType" : "Int32",
238
+ "formattedType" : "MultiArray (Int32 8)",
239
+ "shortDescription" : "",
240
+ "shape" : "[8]",
241
+ "name" : "position_index_seed",
242
+ "type" : "MultiArray"
243
+ }
244
+ ],
245
+ "computePrecision" : "Mixed (Float16, Int32)",
246
+ "storagePrecision" : "Mixed (Float16, Int8)",
247
+ "stateSchema" : [
248
+ {
249
+ "dataType" : "Float16",
250
+ "isOptional" : "0",
251
+ "formattedType" : "State (Float16 8 × 2 × 2048 × 128)",
252
+ "shortDescription" : "",
253
+ "shape" : "[8, 2, 2048, 128]",
254
+ "name" : "key_cache",
255
+ "type" : "State"
256
+ },
257
+ {
258
+ "dataType" : "Float16",
259
+ "isOptional" : "0",
260
+ "formattedType" : "State (Float16 8 × 2 × 2048 × 128)",
261
+ "shortDescription" : "",
262
+ "shape" : "[8, 2, 2048, 128]",
263
+ "name" : "value_cache",
264
+ "type" : "State"
265
+ }
266
+ ],
267
+ "outputSchema" : [
268
+ {
269
+ "hasShapeFlexibility" : "0",
270
+ "isOptional" : "0",
271
+ "dataType" : "Float16",
272
+ "formattedType" : "MultiArray (Float16 1 × 2048 × 1 × 8)",
273
+ "shortDescription" : "",
274
+ "shape" : "[1, 2048, 1, 8]",
275
+ "name" : "hidden_states",
276
+ "type" : "MultiArray"
277
+ }
278
+ ],
279
+ "name" : "length_8",
280
+ "mlProgramOperationTypeHistogram" : {
281
+ "Ios18.expandDims" : 3,
282
+ "Ios18.mul" : 41,
283
+ "Ios18.softmax" : 16,
284
+ "Ios18.matmul" : 32,
285
+ "Ios18.equal" : 1,
286
+ "Split" : 41,
287
+ "Select" : 1,
288
+ "Ios18.greater" : 1,
289
+ "Ios18.readState" : 18,
290
+ "Ios18.add" : 34,
291
+ "Ios18.layerNorm" : 17,
292
+ "Ios16.fillLike" : 1,
293
+ "Ios18.reshape" : 32,
294
+ "Ios18.sliceUpdate" : 16,
295
+ "Ios18.writeState" : 16,
296
+ "Ios18.conv" : 57,
297
+ "Ios18.concat" : 41,
298
+ "Ios18.transpose" : 25,
299
+ "Ios18.sub" : 1,
300
+ "Ios18.cast" : 3,
301
+ "Ios18.silu" : 8,
302
+ "Ios18.sliceByIndex" : 16,
303
+ "Ios16.cumsum" : 1,
304
+ "Ios18.constexprBlockwiseShiftScale" : 22
305
+ }
306
+ },
307
+ {
308
+ "inputSchema" : [
309
+ {
310
+ "hasShapeFlexibility" : "0",
311
+ "isOptional" : "0",
312
+ "dataType" : "Float32",
313
+ "formattedType" : "MultiArray (Float32 1 × 4096 × 1 × 16)",
314
+ "shortDescription" : "",
315
+ "shape" : "[1, 4096, 1, 16]",
316
+ "name" : "inputs_embeds",
317
+ "type" : "MultiArray"
318
+ },
319
+ {
320
+ "hasShapeFlexibility" : "0",
321
+ "isOptional" : "0",
322
+ "dataType" : "Int32",
323
+ "formattedType" : "MultiArray (Int32 1)",
324
+ "shortDescription" : "",
325
+ "shape" : "[1]",
326
+ "name" : "position_id",
327
+ "type" : "MultiArray"
328
+ },
329
+ {
330
+ "hasShapeFlexibility" : "0",
331
+ "isOptional" : "0",
332
+ "dataType" : "Int32",
333
+ "formattedType" : "MultiArray (Int32 16)",
334
+ "shortDescription" : "",
335
+ "shape" : "[16]",
336
+ "name" : "position_index_seed",
337
+ "type" : "MultiArray"
338
+ }
339
+ ],
340
+ "computePrecision" : "Mixed (Float16, Int32)",
341
+ "storagePrecision" : "Mixed (Float16, Int8)",
342
+ "stateSchema" : [
343
+ {
344
+ "dataType" : "Float16",
345
+ "isOptional" : "0",
346
+ "formattedType" : "State (Float16 8 × 2 × 2048 × 128)",
347
+ "shortDescription" : "",
348
+ "shape" : "[8, 2, 2048, 128]",
349
+ "name" : "key_cache",
350
+ "type" : "State"
351
+ },
352
+ {
353
+ "dataType" : "Float16",
354
+ "isOptional" : "0",
355
+ "formattedType" : "State (Float16 8 × 2 × 2048 × 128)",
356
+ "shortDescription" : "",
357
+ "shape" : "[8, 2, 2048, 128]",
358
+ "name" : "value_cache",
359
+ "type" : "State"
360
+ }
361
+ ],
362
+ "outputSchema" : [
363
+ {
364
+ "hasShapeFlexibility" : "0",
365
+ "isOptional" : "0",
366
+ "dataType" : "Float16",
367
+ "formattedType" : "MultiArray (Float16 1 × 2048 × 1 × 16)",
368
+ "shortDescription" : "",
369
+ "shape" : "[1, 2048, 1, 16]",
370
+ "name" : "hidden_states",
371
+ "type" : "MultiArray"
372
+ }
373
+ ],
374
+ "name" : "length_16",
375
+ "mlProgramOperationTypeHistogram" : {
376
+ "Ios18.expandDims" : 3,
377
+ "Ios18.mul" : 41,
378
+ "Ios18.softmax" : 16,
379
+ "Ios18.matmul" : 32,
380
+ "Ios18.equal" : 1,
381
+ "Split" : 41,
382
+ "Select" : 1,
383
+ "Ios18.greater" : 1,
384
+ "Ios18.readState" : 18,
385
+ "Ios18.add" : 34,
386
+ "Ios18.layerNorm" : 17,
387
+ "Ios16.fillLike" : 1,
388
+ "Ios18.reshape" : 32,
389
+ "Ios18.sliceUpdate" : 16,
390
+ "Ios18.writeState" : 16,
391
+ "Ios18.conv" : 57,
392
+ "Ios18.concat" : 41,
393
+ "Ios18.transpose" : 25,
394
+ "Ios18.sub" : 1,
395
+ "Ios18.cast" : 3,
396
+ "Ios18.silu" : 8,
397
+ "Ios18.sliceByIndex" : 16,
398
+ "Ios16.cumsum" : 1,
399
+ "Ios18.constexprBlockwiseShiftScale" : 22
400
+ }
401
+ },
402
+ {
403
+ "inputSchema" : [
404
+ {
405
+ "hasShapeFlexibility" : "0",
406
+ "isOptional" : "0",
407
+ "dataType" : "Float32",
408
+ "formattedType" : "MultiArray (Float32 1 × 4096 × 1 × 32)",
409
+ "shortDescription" : "",
410
+ "shape" : "[1, 4096, 1, 32]",
411
+ "name" : "inputs_embeds",
412
+ "type" : "MultiArray"
413
+ },
414
+ {
415
+ "hasShapeFlexibility" : "0",
416
+ "isOptional" : "0",
417
+ "dataType" : "Int32",
418
+ "formattedType" : "MultiArray (Int32 1)",
419
+ "shortDescription" : "",
420
+ "shape" : "[1]",
421
+ "name" : "position_id",
422
+ "type" : "MultiArray"
423
+ },
424
+ {
425
+ "hasShapeFlexibility" : "0",
426
+ "isOptional" : "0",
427
+ "dataType" : "Int32",
428
+ "formattedType" : "MultiArray (Int32 32)",
429
+ "shortDescription" : "",
430
+ "shape" : "[32]",
431
+ "name" : "position_index_seed",
432
+ "type" : "MultiArray"
433
+ }
434
+ ],
435
+ "computePrecision" : "Mixed (Float16, Int32)",
436
+ "storagePrecision" : "Mixed (Float16, Int8)",
437
+ "stateSchema" : [
438
+ {
439
+ "dataType" : "Float16",
440
+ "isOptional" : "0",
441
+ "formattedType" : "State (Float16 8 × 2 × 2048 × 128)",
442
+ "shortDescription" : "",
443
+ "shape" : "[8, 2, 2048, 128]",
444
+ "name" : "key_cache",
445
+ "type" : "State"
446
+ },
447
+ {
448
+ "dataType" : "Float16",
449
+ "isOptional" : "0",
450
+ "formattedType" : "State (Float16 8 × 2 × 2048 × 128)",
451
+ "shortDescription" : "",
452
+ "shape" : "[8, 2, 2048, 128]",
453
+ "name" : "value_cache",
454
+ "type" : "State"
455
+ }
456
+ ],
457
+ "outputSchema" : [
458
+ {
459
+ "hasShapeFlexibility" : "0",
460
+ "isOptional" : "0",
461
+ "dataType" : "Float16",
462
+ "formattedType" : "MultiArray (Float16 1 × 2048 × 1 × 32)",
463
+ "shortDescription" : "",
464
+ "shape" : "[1, 2048, 1, 32]",
465
+ "name" : "hidden_states",
466
+ "type" : "MultiArray"
467
+ }
468
+ ],
469
+ "name" : "length_32",
470
+ "mlProgramOperationTypeHistogram" : {
471
+ "Ios18.expandDims" : 3,
472
+ "Ios18.mul" : 41,
473
+ "Ios18.softmax" : 16,
474
+ "Ios18.matmul" : 32,
475
+ "Ios18.equal" : 1,
476
+ "Split" : 41,
477
+ "Select" : 1,
478
+ "Ios18.greater" : 1,
479
+ "Ios18.readState" : 18,
480
+ "Ios18.add" : 34,
481
+ "Ios18.layerNorm" : 17,
482
+ "Ios16.fillLike" : 1,
483
+ "Ios18.reshape" : 32,
484
+ "Ios18.sliceUpdate" : 16,
485
+ "Ios18.writeState" : 16,
486
+ "Ios18.conv" : 57,
487
+ "Ios18.concat" : 41,
488
+ "Ios18.transpose" : 25,
489
+ "Ios18.sub" : 1,
490
+ "Ios18.cast" : 3,
491
+ "Ios18.silu" : 8,
492
+ "Ios18.sliceByIndex" : 16,
493
+ "Ios16.cumsum" : 1,
494
+ "Ios18.constexprBlockwiseShiftScale" : 22
495
+ }
496
+ },
497
+ {
498
+ "inputSchema" : [
499
+ {
500
+ "hasShapeFlexibility" : "0",
501
+ "isOptional" : "0",
502
+ "dataType" : "Float32",
503
+ "formattedType" : "MultiArray (Float32 1 × 4096 × 1 × 64)",
504
+ "shortDescription" : "",
505
+ "shape" : "[1, 4096, 1, 64]",
506
+ "name" : "inputs_embeds",
507
+ "type" : "MultiArray"
508
+ },
509
+ {
510
+ "hasShapeFlexibility" : "0",
511
+ "isOptional" : "0",
512
+ "dataType" : "Int32",
513
+ "formattedType" : "MultiArray (Int32 1)",
514
+ "shortDescription" : "",
515
+ "shape" : "[1]",
516
+ "name" : "position_id",
517
+ "type" : "MultiArray"
518
+ },
519
+ {
520
+ "hasShapeFlexibility" : "0",
521
+ "isOptional" : "0",
522
+ "dataType" : "Int32",
523
+ "formattedType" : "MultiArray (Int32 64)",
524
+ "shortDescription" : "",
525
+ "shape" : "[64]",
526
+ "name" : "position_index_seed",
527
+ "type" : "MultiArray"
528
+ }
529
+ ],
530
+ "computePrecision" : "Mixed (Float16, Int32)",
531
+ "storagePrecision" : "Mixed (Float16, Int8)",
532
+ "stateSchema" : [
533
+ {
534
+ "dataType" : "Float16",
535
+ "isOptional" : "0",
536
+ "formattedType" : "State (Float16 8 × 2 × 2048 × 128)",
537
+ "shortDescription" : "",
538
+ "shape" : "[8, 2, 2048, 128]",
539
+ "name" : "key_cache",
540
+ "type" : "State"
541
+ },
542
+ {
543
+ "dataType" : "Float16",
544
+ "isOptional" : "0",
545
+ "formattedType" : "State (Float16 8 × 2 × 2048 × 128)",
546
+ "shortDescription" : "",
547
+ "shape" : "[8, 2, 2048, 128]",
548
+ "name" : "value_cache",
549
+ "type" : "State"
550
+ }
551
+ ],
552
+ "outputSchema" : [
553
+ {
554
+ "hasShapeFlexibility" : "0",
555
+ "isOptional" : "0",
556
+ "dataType" : "Float16",
557
+ "formattedType" : "MultiArray (Float16 1 × 2048 × 1 × 64)",
558
+ "shortDescription" : "",
559
+ "shape" : "[1, 2048, 1, 64]",
560
+ "name" : "hidden_states",
561
+ "type" : "MultiArray"
562
+ }
563
+ ],
564
+ "name" : "length_64",
565
+ "mlProgramOperationTypeHistogram" : {
566
+ "Ios18.expandDims" : 3,
567
+ "Ios18.mul" : 41,
568
+ "Ios18.softmax" : 16,
569
+ "Ios18.matmul" : 32,
570
+ "Ios18.equal" : 1,
571
+ "Split" : 41,
572
+ "Select" : 1,
573
+ "Ios18.greater" : 1,
574
+ "Ios18.readState" : 18,
575
+ "Ios18.add" : 34,
576
+ "Ios18.layerNorm" : 17,
577
+ "Ios16.fillLike" : 1,
578
+ "Ios18.reshape" : 32,
579
+ "Ios18.sliceUpdate" : 16,
580
+ "Ios18.writeState" : 16,
581
+ "Ios18.conv" : 57,
582
+ "Ios18.concat" : 41,
583
+ "Ios18.transpose" : 25,
584
+ "Ios18.sub" : 1,
585
+ "Ios18.cast" : 3,
586
+ "Ios18.silu" : 8,
587
+ "Ios18.sliceByIndex" : 16,
588
+ "Ios16.cumsum" : 1,
589
+ "Ios18.constexprBlockwiseShiftScale" : 22
590
+ }
591
+ },
592
+ {
593
+ "inputSchema" : [
594
+ {
595
+ "hasShapeFlexibility" : "0",
596
+ "isOptional" : "0",
597
+ "dataType" : "Float32",
598
+ "formattedType" : "MultiArray (Float32 1 × 4096 × 1 × 128)",
599
+ "shortDescription" : "",
600
+ "shape" : "[1, 4096, 1, 128]",
601
+ "name" : "inputs_embeds",
602
+ "type" : "MultiArray"
603
+ },
604
+ {
605
+ "hasShapeFlexibility" : "0",
606
+ "isOptional" : "0",
607
+ "dataType" : "Int32",
608
+ "formattedType" : "MultiArray (Int32 1)",
609
+ "shortDescription" : "",
610
+ "shape" : "[1]",
611
+ "name" : "position_id",
612
+ "type" : "MultiArray"
613
+ },
614
+ {
615
+ "hasShapeFlexibility" : "0",
616
+ "isOptional" : "0",
617
+ "dataType" : "Int32",
618
+ "formattedType" : "MultiArray (Int32 128)",
619
+ "shortDescription" : "",
620
+ "shape" : "[128]",
621
+ "name" : "position_index_seed",
622
+ "type" : "MultiArray"
623
+ }
624
+ ],
625
+ "computePrecision" : "Mixed (Float16, Int32)",
626
+ "storagePrecision" : "Mixed (Float16, Int8)",
627
+ "stateSchema" : [
628
+ {
629
+ "dataType" : "Float16",
630
+ "isOptional" : "0",
631
+ "formattedType" : "State (Float16 8 × 2 × 2048 × 128)",
632
+ "shortDescription" : "",
633
+ "shape" : "[8, 2, 2048, 128]",
634
+ "name" : "key_cache",
635
+ "type" : "State"
636
+ },
637
+ {
638
+ "dataType" : "Float16",
639
+ "isOptional" : "0",
640
+ "formattedType" : "State (Float16 8 × 2 × 2048 × 128)",
641
+ "shortDescription" : "",
642
+ "shape" : "[8, 2, 2048, 128]",
643
+ "name" : "value_cache",
644
+ "type" : "State"
645
+ }
646
+ ],
647
+ "outputSchema" : [
648
+ {
649
+ "hasShapeFlexibility" : "0",
650
+ "isOptional" : "0",
651
+ "dataType" : "Float16",
652
+ "formattedType" : "MultiArray (Float16 1 × 2048 × 1 × 128)",
653
+ "shortDescription" : "",
654
+ "shape" : "[1, 2048, 1, 128]",
655
+ "name" : "hidden_states",
656
+ "type" : "MultiArray"
657
+ }
658
+ ],
659
+ "name" : "length_128",
660
+ "mlProgramOperationTypeHistogram" : {
661
+ "Ios18.expandDims" : 3,
662
+ "Ios18.mul" : 41,
663
+ "Ios18.softmax" : 16,
664
+ "Ios18.matmul" : 32,
665
+ "Ios18.equal" : 1,
666
+ "Split" : 41,
667
+ "Select" : 1,
668
+ "Ios18.greater" : 1,
669
+ "Ios18.readState" : 18,
670
+ "Ios18.add" : 34,
671
+ "Ios18.layerNorm" : 17,
672
+ "Ios16.fillLike" : 1,
673
+ "Ios18.reshape" : 32,
674
+ "Ios18.sliceUpdate" : 16,
675
+ "Ios18.writeState" : 16,
676
+ "Ios18.conv" : 57,
677
+ "Ios18.concat" : 41,
678
+ "Ios18.transpose" : 25,
679
+ "Ios18.sub" : 1,
680
+ "Ios18.cast" : 3,
681
+ "Ios18.silu" : 8,
682
+ "Ios18.sliceByIndex" : 16,
683
+ "Ios16.cumsum" : 1,
684
+ "Ios18.constexprBlockwiseShiftScale" : 22
685
+ }
686
+ }
687
+ ],
688
+ "mlProgramOperationTypeHistogram" : {
689
+ "Ios18.expandDims" : 3,
690
+ "Ios18.mul" : 41,
691
+ "Ios18.softmax" : 16,
692
+ "Ios18.matmul" : 32,
693
+ "Ios18.equal" : 1,
694
+ "Split" : 41,
695
+ "Select" : 1,
696
+ "Ios18.greater" : 1,
697
+ "Ios18.readState" : 18,
698
+ "Ios18.add" : 34,
699
+ "Ios18.layerNorm" : 17,
700
+ "Ios16.fillLike" : 1,
701
+ "Ios18.reshape" : 32,
702
+ "Ios18.sliceUpdate" : 16,
703
+ "Ios18.writeState" : 16,
704
+ "Ios18.conv" : 57,
705
+ "Ios18.concat" : 41,
706
+ "Ios18.transpose" : 25,
707
+ "Ios18.sub" : 1,
708
+ "Ios18.cast" : 3,
709
+ "Ios18.silu" : 8,
710
+ "Ios18.sliceByIndex" : 16,
711
+ "Ios16.cumsum" : 1,
712
+ "Ios18.constexprBlockwiseShiftScale" : 22
713
+ },
714
+ "isUpdatable" : "0",
715
+ "stateSchema" : [
716
+ {
717
+ "dataType" : "Float16",
718
+ "isOptional" : "0",
719
+ "formattedType" : "State (Float16 8 × 2 × 2048 × 128)",
720
+ "shortDescription" : "",
721
+ "shape" : "[8, 2, 2048, 128]",
722
+ "name" : "key_cache",
723
+ "type" : "State"
724
+ },
725
+ {
726
+ "dataType" : "Float16",
727
+ "isOptional" : "0",
728
+ "formattedType" : "State (Float16 8 × 2 × 2048 × 128)",
729
+ "shortDescription" : "",
730
+ "shape" : "[8, 2, 2048, 128]",
731
+ "name" : "value_cache",
732
+ "type" : "State"
733
+ }
734
+ ],
735
+ "availability" : {
736
+ "macOS" : "15.0",
737
+ "tvOS" : "18.0",
738
+ "visionOS" : "2.0",
739
+ "watchOS" : "11.0",
740
+ "iOS" : "18.0",
741
+ "macCatalyst" : "18.0"
742
+ },
743
+ "computePrecision" : "Mixed (Float16, Int32)",
744
+ "modelType" : {
745
+ "name" : "MLModelType_mlProgram"
746
+ },
747
+ "inputSchema" : [
748
+ {
749
+ "hasShapeFlexibility" : "0",
750
+ "isOptional" : "0",
751
+ "dataType" : "Float32",
752
+ "formattedType" : "MultiArray (Float32 1 × 4096 × 1 × 1)",
753
+ "shortDescription" : "",
754
+ "shape" : "[1, 4096, 1, 1]",
755
+ "name" : "inputs_embeds",
756
+ "type" : "MultiArray"
757
+ },
758
+ {
759
+ "hasShapeFlexibility" : "0",
760
+ "isOptional" : "0",
761
+ "dataType" : "Int32",
762
+ "formattedType" : "MultiArray (Int32 1)",
763
+ "shortDescription" : "",
764
+ "shape" : "[1]",
765
+ "name" : "position_id",
766
+ "type" : "MultiArray"
767
+ },
768
+ {
769
+ "hasShapeFlexibility" : "0",
770
+ "isOptional" : "0",
771
+ "dataType" : "Int32",
772
+ "formattedType" : "MultiArray (Int32 1)",
773
+ "shortDescription" : "",
774
+ "shape" : "[1]",
775
+ "name" : "position_index_seed",
776
+ "type" : "MultiArray"
777
+ }
778
+ ],
779
+ "defaultFunctionName" : "main",
780
+ "generatedClassName" : "residual_lm_fused_multifunction",
781
+ "userDefinedMetadata" : {
782
+ "com.github.apple.coremltools.version" : "9.0",
783
+ "com.github.apple.coremltools.conversion_date" : "2026-05-18",
784
+ "com.github.apple.coremltools.source" : "milinternal"
785
+ },
786
+ "method" : "predict"
787
+ }
788
+ ]
residual_lm_fused_multifunction.mlmodelc/model.mil ADDED
The diff for this file is too large to render. See raw diff
 
residual_lm_fused_multifunction.mlmodelc/weights/weight.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:86a86badc6712de17986139445dfcafd8aab9db5d8ed71a6d837b70534048095
3
+ size 613231424