Hugging Face's logo

FluidInference
/
pocket-tts-coreml

Text-to-Speech
Core ML
English
tts
apple
on-device
Model card Files
xet
 Community
4
alexwengg commited on
Commit
943c1eb
·
verified ·
1 Parent(s): 84dc743

Upload manifest.json

Browse files
Files changed (1) hide show
  1. manifest.json +1285 -0
manifest.json ADDED
@@ -0,0 +1,1285 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "schema_version": "2",
3
+ "repo": "FluidInference/pocket-tts-coreml",
4
+ "repo_url": "https://huggingface.co/FluidInference/pocket-tts-coreml",
5
+ "generated": "2026-04-26T05:04:10+00:00",
6
+ "design": "Per-submodel quantization selection. The four PocketTTS submodels (cond_step, flowlm_step, flow_decoder, mimi_decoder) are independently swappable between fp16 and int8 \u2014 the int8 .mlpackage has the exact same input/output signature as the fp16 one, so a user can mix-and-match any subset (e.g. flow_decoder=fp16, the other three=int8) by loading different paths into the same PocketTTS pipeline.",
7
+ "quantization": {
8
+ "scheme": "CoreML weight-only int8",
9
+ "granularity": "per-channel symmetric",
10
+ "weight_threshold": 100000,
11
+ "activation_precision": "fp16",
12
+ "tool": "coremltools.optimize.coreml.linear_quantize_weights",
13
+ "notes": "Only body weights with \u2265100K elements are quantized; small head/tail linears stay fp16. Activations are NOT quantized (W8A16 weight-only, not W8A8). Audio quality A/B was run on English 6L only \u2014 non-English not yet A/B'd."
14
+ },
15
+ "ab_test": {
16
+ "language": "english",
17
+ "layer_count": "6L",
18
+ "prompt": "The quick brown fox jumps over the lazy dog. Pack my box with five dozen liquor jugs.",
19
+ "voice": "alba",
20
+ "seed": 42,
21
+ "metrics": [
22
+ "Pearson r (waveform-level correlation vs fp16 baseline)",
23
+ "Speaker similarity (256-dim embedding cosine; 0.65 = identity threshold)"
24
+ ]
25
+ },
26
+ "per_submodel": {
27
+ "cond_step": {
28
+ "audio_quality_int8_only": {
29
+ "speaker_sim": 0.984,
30
+ "pearson": 0.94,
31
+ "verdict": "safe",
32
+ "summary": "minor drift, well within identity threshold"
33
+ },
34
+ "swap_independent": true
35
+ },
36
+ "flowlm_step": {
37
+ "audio_quality_int8_only": {
38
+ "speaker_sim": 0.989,
39
+ "pearson": 0.94,
40
+ "verdict": "safe",
41
+ "summary": "minor drift, well within identity threshold"
42
+ },
43
+ "swap_independent": true
44
+ },
45
+ "flow_decoder": {
46
+ "audio_quality_int8_only": {
47
+ "speaker_sim": 0.981,
48
+ "pearson": 0.78,
49
+ "verdict": "risky",
50
+ "summary": "audible drift; 8-step LSD inner loop compounds quantization error"
51
+ },
52
+ "swap_independent": true
53
+ },
54
+ "mimi_decoder": {
55
+ "audio_quality_int8_only": {
56
+ "speaker_sim": 0.998,
57
+ "pearson": 1.0,
58
+ "verdict": "transparent",
59
+ "summary": "no audible difference vs fp16"
60
+ },
61
+ "swap_independent": true
62
+ }
63
+ },
64
+ "selection_profiles": {
65
+ "all_fp16_baseline": {
66
+ "description": "Reference; everything fp16. No int8 used.",
67
+ "size_english_mlmodelc": "\u2248601 MiB (existing fp16 ship)",
68
+ "models": {
69
+ "cond_step": "fp16",
70
+ "flowlm_step": "fp16",
71
+ "flow_decoder": "fp16",
72
+ "mimi_decoder": "fp16"
73
+ },
74
+ "expected_quality": "reference"
75
+ },
76
+ "safe_int8": {
77
+ "description": "Quantize only the three transparent/safe submodels. Recommended for production.",
78
+ "models": {
79
+ "cond_step": "int8",
80
+ "flowlm_step": "int8",
81
+ "flow_decoder": "fp16",
82
+ "mimi_decoder": "int8"
83
+ },
84
+ "english_mlmodelc_size": "\u2248147 MiB (vs 601 MiB fp16; -75%)",
85
+ "expected_quality": "speaker similarity \u22650.98 vs fp16"
86
+ },
87
+ "aggressive_int8": {
88
+ "description": "All four int8. Max compression but flow_decoder introduces audible drift.",
89
+ "models": {
90
+ "cond_step": "int8",
91
+ "flowlm_step": "int8",
92
+ "flow_decoder": "int8",
93
+ "mimi_decoder": "int8"
94
+ },
95
+ "english_mlmodelc_size": "\u2248156 MiB (vs 601 MiB fp16; -74%)",
96
+ "expected_quality": "speaker similarity 0.940 (still above 0.65 threshold)"
97
+ },
98
+ "user_custom": {
99
+ "description": "Any of the 2^4 = 16 combinations is valid. Pick per submodel based on the audio_quality_int8_only field above and your own ear / target binary size.",
100
+ "example": {
101
+ "cond_step": "int8",
102
+ "flowlm_step": "fp16",
103
+ "flow_decoder": "fp16",
104
+ "mimi_decoder": "int8"
105
+ }
106
+ }
107
+ },
108
+ "languages": {
109
+ "english": {
110
+ "layer_count": "6L",
111
+ "int8_dir": "languages/english/int8",
112
+ "manifest_txt": "languages/english/english.txt",
113
+ "size_bytes": 328626045,
114
+ "size_human": "313.4 MiB",
115
+ "submodels": {
116
+ "cond_step": {
117
+ "fp16": {
118
+ "path_in_repo": "cond_step.mlmodelc",
119
+ "format": "mlmodelc",
120
+ "note": "shipped separately; not part of this int8 upload"
121
+ },
122
+ "int8": {
123
+ "mlpackage": {
124
+ "present": true,
125
+ "path_in_repo": "languages/english/int8/cond_step.mlpackage",
126
+ "size_bytes": 66955176,
127
+ "size_human": "63.9 MiB"
128
+ },
129
+ "mlmodelc": {
130
+ "present": true,
131
+ "path_in_repo": "languages/english/int8/cond_step.mlmodelc",
132
+ "size_bytes": 66995543,
133
+ "size_human": "63.9 MiB"
134
+ }
135
+ },
136
+ "audio_quality": {
137
+ "speaker_sim": 0.984,
138
+ "pearson": 0.94,
139
+ "verdict": "safe",
140
+ "summary": "minor drift, well within identity threshold"
141
+ }
142
+ },
143
+ "flowlm_step": {
144
+ "fp16": {
145
+ "path_in_repo": "flowlm_step.mlmodelc",
146
+ "format": "mlmodelc",
147
+ "note": "shipped separately; not part of this int8 upload"
148
+ },
149
+ "int8": {
150
+ "mlpackage": {
151
+ "present": true,
152
+ "path_in_repo": "languages/english/int8/flowlm_step.mlpackage",
153
+ "size_bytes": 76498561,
154
+ "size_human": "73.0 MiB"
155
+ },
156
+ "mlmodelc": {
157
+ "present": true,
158
+ "path_in_repo": "languages/english/int8/flowlm_step.mlmodelc",
159
+ "size_bytes": 76542686,
160
+ "size_human": "73.0 MiB"
161
+ }
162
+ },
163
+ "audio_quality": {
164
+ "speaker_sim": 0.989,
165
+ "pearson": 0.94,
166
+ "verdict": "safe",
167
+ "summary": "minor drift, well within identity threshold"
168
+ }
169
+ },
170
+ "flow_decoder": {
171
+ "fp16": {
172
+ "path_in_repo": "flow_decoder.mlmodelc",
173
+ "format": "mlmodelc",
174
+ "note": "shipped separately; not part of this int8 upload"
175
+ },
176
+ "int8": {
177
+ "mlpackage": {
178
+ "present": true,
179
+ "path_in_repo": "languages/english/int8/flow_decoder.mlpackage",
180
+ "size_bytes": 9923605,
181
+ "size_human": "9.5 MiB"
182
+ },
183
+ "mlmodelc": {
184
+ "present": true,
185
+ "path_in_repo": "languages/english/int8/flow_decoder.mlmodelc",
186
+ "size_bytes": 9939641,
187
+ "size_human": "9.5 MiB"
188
+ }
189
+ },
190
+ "audio_quality": {
191
+ "speaker_sim": 0.981,
192
+ "pearson": 0.78,
193
+ "verdict": "risky",
194
+ "summary": "audible drift; 8-step LSD inner loop compounds quantization error"
195
+ }
196
+ },
197
+ "mimi_decoder": {
198
+ "fp16": {
199
+ "path_in_repo": "mimi_decoder.mlmodelc",
200
+ "format": "mlmodelc",
201
+ "note": "shipped separately; not part of this int8 upload"
202
+ },
203
+ "int8": {
204
+ "mlpackage": {
205
+ "present": true,
206
+ "path_in_repo": "languages/english/int8/mimi_decoder.mlpackage",
207
+ "size_bytes": 10867723,
208
+ "size_human": "10.4 MiB"
209
+ },
210
+ "mlmodelc": {
211
+ "present": true,
212
+ "path_in_repo": "languages/english/int8/mimi_decoder.mlmodelc",
213
+ "size_bytes": 10903110,
214
+ "size_human": "10.4 MiB"
215
+ }
216
+ },
217
+ "audio_quality": {
218
+ "speaker_sim": 0.998,
219
+ "pearson": 1.0,
220
+ "verdict": "transparent",
221
+ "summary": "no audible difference vs fp16"
222
+ }
223
+ }
224
+ }
225
+ },
226
+ "german": {
227
+ "layer_count": "6L",
228
+ "int8_dir": "languages/german/int8",
229
+ "manifest_txt": "languages/german/german.txt",
230
+ "size_bytes": 328626042,
231
+ "size_human": "313.4 MiB",
232
+ "submodels": {
233
+ "cond_step": {
234
+ "fp16": {
235
+ "path_in_repo": "languages/german/cond_step.mlmodelc",
236
+ "format": "mlmodelc",
237
+ "note": "shipped separately; not part of this int8 upload"
238
+ },
239
+ "int8": {
240
+ "mlpackage": {
241
+ "present": true,
242
+ "path_in_repo": "languages/german/int8/cond_step.mlpackage",
243
+ "size_bytes": 66955175,
244
+ "size_human": "63.9 MiB"
245
+ },
246
+ "mlmodelc": {
247
+ "present": true,
248
+ "path_in_repo": "languages/german/int8/cond_step.mlmodelc",
249
+ "size_bytes": 66995541,
250
+ "size_human": "63.9 MiB"
251
+ }
252
+ },
253
+ "audio_quality": {
254
+ "speaker_sim": 0.984,
255
+ "pearson": 0.94,
256
+ "verdict": "safe",
257
+ "summary": "minor drift, well within identity threshold"
258
+ }
259
+ },
260
+ "flowlm_step": {
261
+ "fp16": {
262
+ "path_in_repo": "languages/german/flowlm_step.mlmodelc",
263
+ "format": "mlmodelc",
264
+ "note": "shipped separately; not part of this int8 upload"
265
+ },
266
+ "int8": {
267
+ "mlpackage": {
268
+ "present": true,
269
+ "path_in_repo": "languages/german/int8/flowlm_step.mlpackage",
270
+ "size_bytes": 76498561,
271
+ "size_human": "73.0 MiB"
272
+ },
273
+ "mlmodelc": {
274
+ "present": true,
275
+ "path_in_repo": "languages/german/int8/flowlm_step.mlmodelc",
276
+ "size_bytes": 76542686,
277
+ "size_human": "73.0 MiB"
278
+ }
279
+ },
280
+ "audio_quality": {
281
+ "speaker_sim": 0.989,
282
+ "pearson": 0.94,
283
+ "verdict": "safe",
284
+ "summary": "minor drift, well within identity threshold"
285
+ }
286
+ },
287
+ "flow_decoder": {
288
+ "fp16": {
289
+ "path_in_repo": "languages/german/flow_decoder.mlmodelc",
290
+ "format": "mlmodelc",
291
+ "note": "shipped separately; not part of this int8 upload"
292
+ },
293
+ "int8": {
294
+ "mlpackage": {
295
+ "present": true,
296
+ "path_in_repo": "languages/german/int8/flow_decoder.mlpackage",
297
+ "size_bytes": 9923605,
298
+ "size_human": "9.5 MiB"
299
+ },
300
+ "mlmodelc": {
301
+ "present": true,
302
+ "path_in_repo": "languages/german/int8/flow_decoder.mlmodelc",
303
+ "size_bytes": 9939641,
304
+ "size_human": "9.5 MiB"
305
+ }
306
+ },
307
+ "audio_quality": {
308
+ "speaker_sim": 0.981,
309
+ "pearson": 0.78,
310
+ "verdict": "risky",
311
+ "summary": "audible drift; 8-step LSD inner loop compounds quantization error"
312
+ }
313
+ },
314
+ "mimi_decoder": {
315
+ "fp16": {
316
+ "path_in_repo": "languages/german/mimi_decoder.mlmodelc",
317
+ "format": "mlmodelc",
318
+ "note": "shipped separately; not part of this int8 upload"
319
+ },
320
+ "int8": {
321
+ "mlpackage": {
322
+ "present": true,
323
+ "path_in_repo": "languages/german/int8/mimi_decoder.mlpackage",
324
+ "size_bytes": 10867723,
325
+ "size_human": "10.4 MiB"
326
+ },
327
+ "mlmodelc": {
328
+ "present": true,
329
+ "path_in_repo": "languages/german/int8/mimi_decoder.mlmodelc",
330
+ "size_bytes": 10903110,
331
+ "size_human": "10.4 MiB"
332
+ }
333
+ },
334
+ "audio_quality": {
335
+ "speaker_sim": 0.998,
336
+ "pearson": 1.0,
337
+ "verdict": "transparent",
338
+ "summary": "no audible difference vs fp16"
339
+ }
340
+ }
341
+ }
342
+ },
343
+ "italian": {
344
+ "layer_count": "6L",
345
+ "int8_dir": "languages/italian/int8",
346
+ "manifest_txt": "languages/italian/italian.txt",
347
+ "size_bytes": 328626042,
348
+ "size_human": "313.4 MiB",
349
+ "submodels": {
350
+ "cond_step": {
351
+ "fp16": {
352
+ "path_in_repo": "languages/italian/cond_step.mlmodelc",
353
+ "format": "mlmodelc",
354
+ "note": "shipped separately; not part of this int8 upload"
355
+ },
356
+ "int8": {
357
+ "mlpackage": {
358
+ "present": true,
359
+ "path_in_repo": "languages/italian/int8/cond_step.mlpackage",
360
+ "size_bytes": 66955175,
361
+ "size_human": "63.9 MiB"
362
+ },
363
+ "mlmodelc": {
364
+ "present": true,
365
+ "path_in_repo": "languages/italian/int8/cond_step.mlmodelc",
366
+ "size_bytes": 66995541,
367
+ "size_human": "63.9 MiB"
368
+ }
369
+ },
370
+ "audio_quality": {
371
+ "speaker_sim": 0.984,
372
+ "pearson": 0.94,
373
+ "verdict": "safe",
374
+ "summary": "minor drift, well within identity threshold"
375
+ }
376
+ },
377
+ "flowlm_step": {
378
+ "fp16": {
379
+ "path_in_repo": "languages/italian/flowlm_step.mlmodelc",
380
+ "format": "mlmodelc",
381
+ "note": "shipped separately; not part of this int8 upload"
382
+ },
383
+ "int8": {
384
+ "mlpackage": {
385
+ "present": true,
386
+ "path_in_repo": "languages/italian/int8/flowlm_step.mlpackage",
387
+ "size_bytes": 76498561,
388
+ "size_human": "73.0 MiB"
389
+ },
390
+ "mlmodelc": {
391
+ "present": true,
392
+ "path_in_repo": "languages/italian/int8/flowlm_step.mlmodelc",
393
+ "size_bytes": 76542686,
394
+ "size_human": "73.0 MiB"
395
+ }
396
+ },
397
+ "audio_quality": {
398
+ "speaker_sim": 0.989,
399
+ "pearson": 0.94,
400
+ "verdict": "safe",
401
+ "summary": "minor drift, well within identity threshold"
402
+ }
403
+ },
404
+ "flow_decoder": {
405
+ "fp16": {
406
+ "path_in_repo": "languages/italian/flow_decoder.mlmodelc",
407
+ "format": "mlmodelc",
408
+ "note": "shipped separately; not part of this int8 upload"
409
+ },
410
+ "int8": {
411
+ "mlpackage": {
412
+ "present": true,
413
+ "path_in_repo": "languages/italian/int8/flow_decoder.mlpackage",
414
+ "size_bytes": 9923605,
415
+ "size_human": "9.5 MiB"
416
+ },
417
+ "mlmodelc": {
418
+ "present": true,
419
+ "path_in_repo": "languages/italian/int8/flow_decoder.mlmodelc",
420
+ "size_bytes": 9939641,
421
+ "size_human": "9.5 MiB"
422
+ }
423
+ },
424
+ "audio_quality": {
425
+ "speaker_sim": 0.981,
426
+ "pearson": 0.78,
427
+ "verdict": "risky",
428
+ "summary": "audible drift; 8-step LSD inner loop compounds quantization error"
429
+ }
430
+ },
431
+ "mimi_decoder": {
432
+ "fp16": {
433
+ "path_in_repo": "languages/italian/mimi_decoder.mlmodelc",
434
+ "format": "mlmodelc",
435
+ "note": "shipped separately; not part of this int8 upload"
436
+ },
437
+ "int8": {
438
+ "mlpackage": {
439
+ "present": true,
440
+ "path_in_repo": "languages/italian/int8/mimi_decoder.mlpackage",
441
+ "size_bytes": 10867723,
442
+ "size_human": "10.4 MiB"
443
+ },
444
+ "mlmodelc": {
445
+ "present": true,
446
+ "path_in_repo": "languages/italian/int8/mimi_decoder.mlmodelc",
447
+ "size_bytes": 10903110,
448
+ "size_human": "10.4 MiB"
449
+ }
450
+ },
451
+ "audio_quality": {
452
+ "speaker_sim": 0.998,
453
+ "pearson": 1.0,
454
+ "verdict": "transparent",
455
+ "summary": "no audible difference vs fp16"
456
+ }
457
+ }
458
+ }
459
+ },
460
+ "portuguese": {
461
+ "layer_count": "6L",
462
+ "int8_dir": "languages/portuguese/int8",
463
+ "manifest_txt": "languages/portuguese/portuguese.txt",
464
+ "size_bytes": 328626042,
465
+ "size_human": "313.4 MiB",
466
+ "submodels": {
467
+ "cond_step": {
468
+ "fp16": {
469
+ "path_in_repo": "languages/portuguese/cond_step.mlmodelc",
470
+ "format": "mlmodelc",
471
+ "note": "shipped separately; not part of this int8 upload"
472
+ },
473
+ "int8": {
474
+ "mlpackage": {
475
+ "present": true,
476
+ "path_in_repo": "languages/portuguese/int8/cond_step.mlpackage",
477
+ "size_bytes": 66955175,
478
+ "size_human": "63.9 MiB"
479
+ },
480
+ "mlmodelc": {
481
+ "present": true,
482
+ "path_in_repo": "languages/portuguese/int8/cond_step.mlmodelc",
483
+ "size_bytes": 66995541,
484
+ "size_human": "63.9 MiB"
485
+ }
486
+ },
487
+ "audio_quality": {
488
+ "speaker_sim": 0.984,
489
+ "pearson": 0.94,
490
+ "verdict": "safe",
491
+ "summary": "minor drift, well within identity threshold"
492
+ }
493
+ },
494
+ "flowlm_step": {
495
+ "fp16": {
496
+ "path_in_repo": "languages/portuguese/flowlm_step.mlmodelc",
497
+ "format": "mlmodelc",
498
+ "note": "shipped separately; not part of this int8 upload"
499
+ },
500
+ "int8": {
501
+ "mlpackage": {
502
+ "present": true,
503
+ "path_in_repo": "languages/portuguese/int8/flowlm_step.mlpackage",
504
+ "size_bytes": 76498561,
505
+ "size_human": "73.0 MiB"
506
+ },
507
+ "mlmodelc": {
508
+ "present": true,
509
+ "path_in_repo": "languages/portuguese/int8/flowlm_step.mlmodelc",
510
+ "size_bytes": 76542686,
511
+ "size_human": "73.0 MiB"
512
+ }
513
+ },
514
+ "audio_quality": {
515
+ "speaker_sim": 0.989,
516
+ "pearson": 0.94,
517
+ "verdict": "safe",
518
+ "summary": "minor drift, well within identity threshold"
519
+ }
520
+ },
521
+ "flow_decoder": {
522
+ "fp16": {
523
+ "path_in_repo": "languages/portuguese/flow_decoder.mlmodelc",
524
+ "format": "mlmodelc",
525
+ "note": "shipped separately; not part of this int8 upload"
526
+ },
527
+ "int8": {
528
+ "mlpackage": {
529
+ "present": true,
530
+ "path_in_repo": "languages/portuguese/int8/flow_decoder.mlpackage",
531
+ "size_bytes": 9923605,
532
+ "size_human": "9.5 MiB"
533
+ },
534
+ "mlmodelc": {
535
+ "present": true,
536
+ "path_in_repo": "languages/portuguese/int8/flow_decoder.mlmodelc",
537
+ "size_bytes": 9939641,
538
+ "size_human": "9.5 MiB"
539
+ }
540
+ },
541
+ "audio_quality": {
542
+ "speaker_sim": 0.981,
543
+ "pearson": 0.78,
544
+ "verdict": "risky",
545
+ "summary": "audible drift; 8-step LSD inner loop compounds quantization error"
546
+ }
547
+ },
548
+ "mimi_decoder": {
549
+ "fp16": {
550
+ "path_in_repo": "languages/portuguese/mimi_decoder.mlmodelc",
551
+ "format": "mlmodelc",
552
+ "note": "shipped separately; not part of this int8 upload"
553
+ },
554
+ "int8": {
555
+ "mlpackage": {
556
+ "present": true,
557
+ "path_in_repo": "languages/portuguese/int8/mimi_decoder.mlpackage",
558
+ "size_bytes": 10867723,
559
+ "size_human": "10.4 MiB"
560
+ },
561
+ "mlmodelc": {
562
+ "present": true,
563
+ "path_in_repo": "languages/portuguese/int8/mimi_decoder.mlmodelc",
564
+ "size_bytes": 10903110,
565
+ "size_human": "10.4 MiB"
566
+ }
567
+ },
568
+ "audio_quality": {
569
+ "speaker_sim": 0.998,
570
+ "pearson": 1.0,
571
+ "verdict": "transparent",
572
+ "summary": "no audible difference vs fp16"
573
+ }
574
+ }
575
+ }
576
+ },
577
+ "spanish": {
578
+ "layer_count": "6L",
579
+ "int8_dir": "languages/spanish/int8",
580
+ "manifest_txt": "languages/spanish/spanish.txt",
581
+ "size_bytes": 328626043,
582
+ "size_human": "313.4 MiB",
583
+ "submodels": {
584
+ "cond_step": {
585
+ "fp16": {
586
+ "path_in_repo": "languages/spanish/cond_step.mlmodelc",
587
+ "format": "mlmodelc",
588
+ "note": "shipped separately; not part of this int8 upload"
589
+ },
590
+ "int8": {
591
+ "mlpackage": {
592
+ "present": true,
593
+ "path_in_repo": "languages/spanish/int8/cond_step.mlpackage",
594
+ "size_bytes": 66955175,
595
+ "size_human": "63.9 MiB"
596
+ },
597
+ "mlmodelc": {
598
+ "present": true,
599
+ "path_in_repo": "languages/spanish/int8/cond_step.mlmodelc",
600
+ "size_bytes": 66995541,
601
+ "size_human": "63.9 MiB"
602
+ }
603
+ },
604
+ "audio_quality": {
605
+ "speaker_sim": 0.984,
606
+ "pearson": 0.94,
607
+ "verdict": "safe",
608
+ "summary": "minor drift, well within identity threshold"
609
+ }
610
+ },
611
+ "flowlm_step": {
612
+ "fp16": {
613
+ "path_in_repo": "languages/spanish/flowlm_step.mlmodelc",
614
+ "format": "mlmodelc",
615
+ "note": "shipped separately; not part of this int8 upload"
616
+ },
617
+ "int8": {
618
+ "mlpackage": {
619
+ "present": true,
620
+ "path_in_repo": "languages/spanish/int8/flowlm_step.mlpackage",
621
+ "size_bytes": 76498561,
622
+ "size_human": "73.0 MiB"
623
+ },
624
+ "mlmodelc": {
625
+ "present": true,
626
+ "path_in_repo": "languages/spanish/int8/flowlm_step.mlmodelc",
627
+ "size_bytes": 76542686,
628
+ "size_human": "73.0 MiB"
629
+ }
630
+ },
631
+ "audio_quality": {
632
+ "speaker_sim": 0.989,
633
+ "pearson": 0.94,
634
+ "verdict": "safe",
635
+ "summary": "minor drift, well within identity threshold"
636
+ }
637
+ },
638
+ "flow_decoder": {
639
+ "fp16": {
640
+ "path_in_repo": "languages/spanish/flow_decoder.mlmodelc",
641
+ "format": "mlmodelc",
642
+ "note": "shipped separately; not part of this int8 upload"
643
+ },
644
+ "int8": {
645
+ "mlpackage": {
646
+ "present": true,
647
+ "path_in_repo": "languages/spanish/int8/flow_decoder.mlpackage",
648
+ "size_bytes": 9923605,
649
+ "size_human": "9.5 MiB"
650
+ },
651
+ "mlmodelc": {
652
+ "present": true,
653
+ "path_in_repo": "languages/spanish/int8/flow_decoder.mlmodelc",
654
+ "size_bytes": 9939641,
655
+ "size_human": "9.5 MiB"
656
+ }
657
+ },
658
+ "audio_quality": {
659
+ "speaker_sim": 0.981,
660
+ "pearson": 0.78,
661
+ "verdict": "risky",
662
+ "summary": "audible drift; 8-step LSD inner loop compounds quantization error"
663
+ }
664
+ },
665
+ "mimi_decoder": {
666
+ "fp16": {
667
+ "path_in_repo": "languages/spanish/mimi_decoder.mlmodelc",
668
+ "format": "mlmodelc",
669
+ "note": "shipped separately; not part of this int8 upload"
670
+ },
671
+ "int8": {
672
+ "mlpackage": {
673
+ "present": true,
674
+ "path_in_repo": "languages/spanish/int8/mimi_decoder.mlpackage",
675
+ "size_bytes": 10867723,
676
+ "size_human": "10.4 MiB"
677
+ },
678
+ "mlmodelc": {
679
+ "present": true,
680
+ "path_in_repo": "languages/spanish/int8/mimi_decoder.mlmodelc",
681
+ "size_bytes": 10903111,
682
+ "size_human": "10.4 MiB"
683
+ }
684
+ },
685
+ "audio_quality": {
686
+ "speaker_sim": 0.998,
687
+ "pearson": 1.0,
688
+ "verdict": "transparent",
689
+ "summary": "no audible difference vs fp16"
690
+ }
691
+ }
692
+ }
693
+ },
694
+ "french_24l": {
695
+ "layer_count": "24L",
696
+ "int8_dir": "languages/french_24l/int8",
697
+ "manifest_txt": "languages/french_24l/french_24l.txt",
698
+ "size_bytes": 1239353135,
699
+ "size_human": "1.2 GiB",
700
+ "submodels": {
701
+ "cond_step": {
702
+ "fp16": {
703
+ "path_in_repo": "languages/french_24l/cond_step.mlmodelc",
704
+ "format": "mlmodelc",
705
+ "note": "shipped separately; not part of this int8 upload"
706
+ },
707
+ "int8": {
708
+ "mlpackage": {
709
+ "present": true,
710
+ "path_in_repo": "languages/french_24l/int8/cond_step.mlpackage",
711
+ "size_bytes": 294576395,
712
+ "size_human": "280.9 MiB"
713
+ },
714
+ "mlmodelc": {
715
+ "present": true,
716
+ "path_in_repo": "languages/french_24l/int8/cond_step.mlmodelc",
717
+ "size_bytes": 294737549,
718
+ "size_human": "281.1 MiB"
719
+ }
720
+ },
721
+ "audio_quality": {
722
+ "speaker_sim": 0.984,
723
+ "pearson": 0.94,
724
+ "verdict": "safe",
725
+ "summary": "minor drift, well within identity threshold"
726
+ }
727
+ },
728
+ "flowlm_step": {
729
+ "fp16": {
730
+ "path_in_repo": "languages/french_24l/flowlm_step.mlmodelc",
731
+ "format": "mlmodelc",
732
+ "note": "shipped separately; not part of this int8 upload"
733
+ },
734
+ "int8": {
735
+ "mlpackage": {
736
+ "present": true,
737
+ "path_in_repo": "languages/french_24l/int8/flowlm_step.mlpackage",
738
+ "size_bytes": 304120099,
739
+ "size_human": "290.0 MiB"
740
+ },
741
+ "mlmodelc": {
742
+ "present": true,
743
+ "path_in_repo": "languages/french_24l/int8/flowlm_step.mlmodelc",
744
+ "size_bytes": 304285013,
745
+ "size_human": "290.2 MiB"
746
+ }
747
+ },
748
+ "audio_quality": {
749
+ "speaker_sim": 0.989,
750
+ "pearson": 0.94,
751
+ "verdict": "safe",
752
+ "summary": "minor drift, well within identity threshold"
753
+ }
754
+ },
755
+ "flow_decoder": {
756
+ "fp16": {
757
+ "path_in_repo": "languages/french_24l/flow_decoder.mlmodelc",
758
+ "format": "mlmodelc",
759
+ "note": "shipped separately; not part of this int8 upload"
760
+ },
761
+ "int8": {
762
+ "mlpackage": {
763
+ "present": true,
764
+ "path_in_repo": "languages/french_24l/int8/flow_decoder.mlpackage",
765
+ "size_bytes": 9923605,
766
+ "size_human": "9.5 MiB"
767
+ },
768
+ "mlmodelc": {
769
+ "present": true,
770
+ "path_in_repo": "languages/french_24l/int8/flow_decoder.mlmodelc",
771
+ "size_bytes": 9939641,
772
+ "size_human": "9.5 MiB"
773
+ }
774
+ },
775
+ "audio_quality": {
776
+ "speaker_sim": 0.981,
777
+ "pearson": 0.78,
778
+ "verdict": "risky",
779
+ "summary": "audible drift; 8-step LSD inner loop compounds quantization error"
780
+ }
781
+ },
782
+ "mimi_decoder": {
783
+ "fp16": {
784
+ "path_in_repo": "languages/french_24l/mimi_decoder.mlmodelc",
785
+ "format": "mlmodelc",
786
+ "note": "shipped separately; not part of this int8 upload"
787
+ },
788
+ "int8": {
789
+ "mlpackage": {
790
+ "present": true,
791
+ "path_in_repo": "languages/french_24l/int8/mimi_decoder.mlpackage",
792
+ "size_bytes": 10867723,
793
+ "size_human": "10.4 MiB"
794
+ },
795
+ "mlmodelc": {
796
+ "present": true,
797
+ "path_in_repo": "languages/french_24l/int8/mimi_decoder.mlmodelc",
798
+ "size_bytes": 10903110,
799
+ "size_human": "10.4 MiB"
800
+ }
801
+ },
802
+ "audio_quality": {
803
+ "speaker_sim": 0.998,
804
+ "pearson": 1.0,
805
+ "verdict": "transparent",
806
+ "summary": "no audible difference vs fp16"
807
+ }
808
+ }
809
+ }
810
+ },
811
+ "german_24l": {
812
+ "layer_count": "24L",
813
+ "int8_dir": "languages/german_24l/int8",
814
+ "manifest_txt": "languages/german_24l/german_24l.txt",
815
+ "size_bytes": 1239353135,
816
+ "size_human": "1.2 GiB",
817
+ "submodels": {
818
+ "cond_step": {
819
+ "fp16": {
820
+ "path_in_repo": "languages/german_24l/cond_step.mlmodelc",
821
+ "format": "mlmodelc",
822
+ "note": "shipped separately; not part of this int8 upload"
823
+ },
824
+ "int8": {
825
+ "mlpackage": {
826
+ "present": true,
827
+ "path_in_repo": "languages/german_24l/int8/cond_step.mlpackage",
828
+ "size_bytes": 294576395,
829
+ "size_human": "280.9 MiB"
830
+ },
831
+ "mlmodelc": {
832
+ "present": true,
833
+ "path_in_repo": "languages/german_24l/int8/cond_step.mlmodelc",
834
+ "size_bytes": 294737549,
835
+ "size_human": "281.1 MiB"
836
+ }
837
+ },
838
+ "audio_quality": {
839
+ "speaker_sim": 0.984,
840
+ "pearson": 0.94,
841
+ "verdict": "safe",
842
+ "summary": "minor drift, well within identity threshold"
843
+ }
844
+ },
845
+ "flowlm_step": {
846
+ "fp16": {
847
+ "path_in_repo": "languages/german_24l/flowlm_step.mlmodelc",
848
+ "format": "mlmodelc",
849
+ "note": "shipped separately; not part of this int8 upload"
850
+ },
851
+ "int8": {
852
+ "mlpackage": {
853
+ "present": true,
854
+ "path_in_repo": "languages/german_24l/int8/flowlm_step.mlpackage",
855
+ "size_bytes": 304120099,
856
+ "size_human": "290.0 MiB"
857
+ },
858
+ "mlmodelc": {
859
+ "present": true,
860
+ "path_in_repo": "languages/german_24l/int8/flowlm_step.mlmodelc",
861
+ "size_bytes": 304285013,
862
+ "size_human": "290.2 MiB"
863
+ }
864
+ },
865
+ "audio_quality": {
866
+ "speaker_sim": 0.989,
867
+ "pearson": 0.94,
868
+ "verdict": "safe",
869
+ "summary": "minor drift, well within identity threshold"
870
+ }
871
+ },
872
+ "flow_decoder": {
873
+ "fp16": {
874
+ "path_in_repo": "languages/german_24l/flow_decoder.mlmodelc",
875
+ "format": "mlmodelc",
876
+ "note": "shipped separately; not part of this int8 upload"
877
+ },
878
+ "int8": {
879
+ "mlpackage": {
880
+ "present": true,
881
+ "path_in_repo": "languages/german_24l/int8/flow_decoder.mlpackage",
882
+ "size_bytes": 9923605,
883
+ "size_human": "9.5 MiB"
884
+ },
885
+ "mlmodelc": {
886
+ "present": true,
887
+ "path_in_repo": "languages/german_24l/int8/flow_decoder.mlmodelc",
888
+ "size_bytes": 9939641,
889
+ "size_human": "9.5 MiB"
890
+ }
891
+ },
892
+ "audio_quality": {
893
+ "speaker_sim": 0.981,
894
+ "pearson": 0.78,
895
+ "verdict": "risky",
896
+ "summary": "audible drift; 8-step LSD inner loop compounds quantization error"
897
+ }
898
+ },
899
+ "mimi_decoder": {
900
+ "fp16": {
901
+ "path_in_repo": "languages/german_24l/mimi_decoder.mlmodelc",
902
+ "format": "mlmodelc",
903
+ "note": "shipped separately; not part of this int8 upload"
904
+ },
905
+ "int8": {
906
+ "mlpackage": {
907
+ "present": true,
908
+ "path_in_repo": "languages/german_24l/int8/mimi_decoder.mlpackage",
909
+ "size_bytes": 10867723,
910
+ "size_human": "10.4 MiB"
911
+ },
912
+ "mlmodelc": {
913
+ "present": true,
914
+ "path_in_repo": "languages/german_24l/int8/mimi_decoder.mlmodelc",
915
+ "size_bytes": 10903110,
916
+ "size_human": "10.4 MiB"
917
+ }
918
+ },
919
+ "audio_quality": {
920
+ "speaker_sim": 0.998,
921
+ "pearson": 1.0,
922
+ "verdict": "transparent",
923
+ "summary": "no audible difference vs fp16"
924
+ }
925
+ }
926
+ }
927
+ },
928
+ "italian_24l": {
929
+ "layer_count": "24L",
930
+ "int8_dir": "languages/italian_24l/int8",
931
+ "manifest_txt": "languages/italian_24l/italian_24l.txt",
932
+ "size_bytes": 1239353135,
933
+ "size_human": "1.2 GiB",
934
+ "submodels": {
935
+ "cond_step": {
936
+ "fp16": {
937
+ "path_in_repo": "languages/italian_24l/cond_step.mlmodelc",
938
+ "format": "mlmodelc",
939
+ "note": "shipped separately; not part of this int8 upload"
940
+ },
941
+ "int8": {
942
+ "mlpackage": {
943
+ "present": true,
944
+ "path_in_repo": "languages/italian_24l/int8/cond_step.mlpackage",
945
+ "size_bytes": 294576395,
946
+ "size_human": "280.9 MiB"
947
+ },
948
+ "mlmodelc": {
949
+ "present": true,
950
+ "path_in_repo": "languages/italian_24l/int8/cond_step.mlmodelc",
951
+ "size_bytes": 294737549,
952
+ "size_human": "281.1 MiB"
953
+ }
954
+ },
955
+ "audio_quality": {
956
+ "speaker_sim": 0.984,
957
+ "pearson": 0.94,
958
+ "verdict": "safe",
959
+ "summary": "minor drift, well within identity threshold"
960
+ }
961
+ },
962
+ "flowlm_step": {
963
+ "fp16": {
964
+ "path_in_repo": "languages/italian_24l/flowlm_step.mlmodelc",
965
+ "format": "mlmodelc",
966
+ "note": "shipped separately; not part of this int8 upload"
967
+ },
968
+ "int8": {
969
+ "mlpackage": {
970
+ "present": true,
971
+ "path_in_repo": "languages/italian_24l/int8/flowlm_step.mlpackage",
972
+ "size_bytes": 304120099,
973
+ "size_human": "290.0 MiB"
974
+ },
975
+ "mlmodelc": {
976
+ "present": true,
977
+ "path_in_repo": "languages/italian_24l/int8/flowlm_step.mlmodelc",
978
+ "size_bytes": 304285013,
979
+ "size_human": "290.2 MiB"
980
+ }
981
+ },
982
+ "audio_quality": {
983
+ "speaker_sim": 0.989,
984
+ "pearson": 0.94,
985
+ "verdict": "safe",
986
+ "summary": "minor drift, well within identity threshold"
987
+ }
988
+ },
989
+ "flow_decoder": {
990
+ "fp16": {
991
+ "path_in_repo": "languages/italian_24l/flow_decoder.mlmodelc",
992
+ "format": "mlmodelc",
993
+ "note": "shipped separately; not part of this int8 upload"
994
+ },
995
+ "int8": {
996
+ "mlpackage": {
997
+ "present": true,
998
+ "path_in_repo": "languages/italian_24l/int8/flow_decoder.mlpackage",
999
+ "size_bytes": 9923605,
1000
+ "size_human": "9.5 MiB"
1001
+ },
1002
+ "mlmodelc": {
1003
+ "present": true,
1004
+ "path_in_repo": "languages/italian_24l/int8/flow_decoder.mlmodelc",
1005
+ "size_bytes": 9939641,
1006
+ "size_human": "9.5 MiB"
1007
+ }
1008
+ },
1009
+ "audio_quality": {
1010
+ "speaker_sim": 0.981,
1011
+ "pearson": 0.78,
1012
+ "verdict": "risky",
1013
+ "summary": "audible drift; 8-step LSD inner loop compounds quantization error"
1014
+ }
1015
+ },
1016
+ "mimi_decoder": {
1017
+ "fp16": {
1018
+ "path_in_repo": "languages/italian_24l/mimi_decoder.mlmodelc",
1019
+ "format": "mlmodelc",
1020
+ "note": "shipped separately; not part of this int8 upload"
1021
+ },
1022
+ "int8": {
1023
+ "mlpackage": {
1024
+ "present": true,
1025
+ "path_in_repo": "languages/italian_24l/int8/mimi_decoder.mlpackage",
1026
+ "size_bytes": 10867723,
1027
+ "size_human": "10.4 MiB"
1028
+ },
1029
+ "mlmodelc": {
1030
+ "present": true,
1031
+ "path_in_repo": "languages/italian_24l/int8/mimi_decoder.mlmodelc",
1032
+ "size_bytes": 10903110,
1033
+ "size_human": "10.4 MiB"
1034
+ }
1035
+ },
1036
+ "audio_quality": {
1037
+ "speaker_sim": 0.998,
1038
+ "pearson": 1.0,
1039
+ "verdict": "transparent",
1040
+ "summary": "no audible difference vs fp16"
1041
+ }
1042
+ }
1043
+ }
1044
+ },
1045
+ "portuguese_24l": {
1046
+ "layer_count": "24L",
1047
+ "int8_dir": "languages/portuguese_24l/int8",
1048
+ "manifest_txt": "languages/portuguese_24l/portuguese_24l.txt",
1049
+ "size_bytes": 1239353135,
1050
+ "size_human": "1.2 GiB",
1051
+ "submodels": {
1052
+ "cond_step": {
1053
+ "fp16": {
1054
+ "path_in_repo": "languages/portuguese_24l/cond_step.mlmodelc",
1055
+ "format": "mlmodelc",
1056
+ "note": "shipped separately; not part of this int8 upload"
1057
+ },
1058
+ "int8": {
1059
+ "mlpackage": {
1060
+ "present": true,
1061
+ "path_in_repo": "languages/portuguese_24l/int8/cond_step.mlpackage",
1062
+ "size_bytes": 294576395,
1063
+ "size_human": "280.9 MiB"
1064
+ },
1065
+ "mlmodelc": {
1066
+ "present": true,
1067
+ "path_in_repo": "languages/portuguese_24l/int8/cond_step.mlmodelc",
1068
+ "size_bytes": 294737549,
1069
+ "size_human": "281.1 MiB"
1070
+ }
1071
+ },
1072
+ "audio_quality": {
1073
+ "speaker_sim": 0.984,
1074
+ "pearson": 0.94,
1075
+ "verdict": "safe",
1076
+ "summary": "minor drift, well within identity threshold"
1077
+ }
1078
+ },
1079
+ "flowlm_step": {
1080
+ "fp16": {
1081
+ "path_in_repo": "languages/portuguese_24l/flowlm_step.mlmodelc",
1082
+ "format": "mlmodelc",
1083
+ "note": "shipped separately; not part of this int8 upload"
1084
+ },
1085
+ "int8": {
1086
+ "mlpackage": {
1087
+ "present": true,
1088
+ "path_in_repo": "languages/portuguese_24l/int8/flowlm_step.mlpackage",
1089
+ "size_bytes": 304120099,
1090
+ "size_human": "290.0 MiB"
1091
+ },
1092
+ "mlmodelc": {
1093
+ "present": true,
1094
+ "path_in_repo": "languages/portuguese_24l/int8/flowlm_step.mlmodelc",
1095
+ "size_bytes": 304285013,
1096
+ "size_human": "290.2 MiB"
1097
+ }
1098
+ },
1099
+ "audio_quality": {
1100
+ "speaker_sim": 0.989,
1101
+ "pearson": 0.94,
1102
+ "verdict": "safe",
1103
+ "summary": "minor drift, well within identity threshold"
1104
+ }
1105
+ },
1106
+ "flow_decoder": {
1107
+ "fp16": {
1108
+ "path_in_repo": "languages/portuguese_24l/flow_decoder.mlmodelc",
1109
+ "format": "mlmodelc",
1110
+ "note": "shipped separately; not part of this int8 upload"
1111
+ },
1112
+ "int8": {
1113
+ "mlpackage": {
1114
+ "present": true,
1115
+ "path_in_repo": "languages/portuguese_24l/int8/flow_decoder.mlpackage",
1116
+ "size_bytes": 9923605,
1117
+ "size_human": "9.5 MiB"
1118
+ },
1119
+ "mlmodelc": {
1120
+ "present": true,
1121
+ "path_in_repo": "languages/portuguese_24l/int8/flow_decoder.mlmodelc",
1122
+ "size_bytes": 9939641,
1123
+ "size_human": "9.5 MiB"
1124
+ }
1125
+ },
1126
+ "audio_quality": {
1127
+ "speaker_sim": 0.981,
1128
+ "pearson": 0.78,
1129
+ "verdict": "risky",
1130
+ "summary": "audible drift; 8-step LSD inner loop compounds quantization error"
1131
+ }
1132
+ },
1133
+ "mimi_decoder": {
1134
+ "fp16": {
1135
+ "path_in_repo": "languages/portuguese_24l/mimi_decoder.mlmodelc",
1136
+ "format": "mlmodelc",
1137
+ "note": "shipped separately; not part of this int8 upload"
1138
+ },
1139
+ "int8": {
1140
+ "mlpackage": {
1141
+ "present": true,
1142
+ "path_in_repo": "languages/portuguese_24l/int8/mimi_decoder.mlpackage",
1143
+ "size_bytes": 10867723,
1144
+ "size_human": "10.4 MiB"
1145
+ },
1146
+ "mlmodelc": {
1147
+ "present": true,
1148
+ "path_in_repo": "languages/portuguese_24l/int8/mimi_decoder.mlmodelc",
1149
+ "size_bytes": 10903110,
1150
+ "size_human": "10.4 MiB"
1151
+ }
1152
+ },
1153
+ "audio_quality": {
1154
+ "speaker_sim": 0.998,
1155
+ "pearson": 1.0,
1156
+ "verdict": "transparent",
1157
+ "summary": "no audible difference vs fp16"
1158
+ }
1159
+ }
1160
+ }
1161
+ },
1162
+ "spanish_24l": {
1163
+ "layer_count": "24L",
1164
+ "int8_dir": "languages/spanish_24l/int8",
1165
+ "manifest_txt": "languages/spanish_24l/spanish_24l.txt",
1166
+ "size_bytes": 1239353136,
1167
+ "size_human": "1.2 GiB",
1168
+ "submodels": {
1169
+ "cond_step": {
1170
+ "fp16": {
1171
+ "path_in_repo": "languages/spanish_24l/cond_step.mlmodelc",
1172
+ "format": "mlmodelc",
1173
+ "note": "shipped separately; not part of this int8 upload"
1174
+ },
1175
+ "int8": {
1176
+ "mlpackage": {
1177
+ "present": true,
1178
+ "path_in_repo": "languages/spanish_24l/int8/cond_step.mlpackage",
1179
+ "size_bytes": 294576395,
1180
+ "size_human": "280.9 MiB"
1181
+ },
1182
+ "mlmodelc": {
1183
+ "present": true,
1184
+ "path_in_repo": "languages/spanish_24l/int8/cond_step.mlmodelc",
1185
+ "size_bytes": 294737549,
1186
+ "size_human": "281.1 MiB"
1187
+ }
1188
+ },
1189
+ "audio_quality": {
1190
+ "speaker_sim": 0.984,
1191
+ "pearson": 0.94,
1192
+ "verdict": "safe",
1193
+ "summary": "minor drift, well within identity threshold"
1194
+ }
1195
+ },
1196
+ "flowlm_step": {
1197
+ "fp16": {
1198
+ "path_in_repo": "languages/spanish_24l/flowlm_step.mlmodelc",
1199
+ "format": "mlmodelc",
1200
+ "note": "shipped separately; not part of this int8 upload"
1201
+ },
1202
+ "int8": {
1203
+ "mlpackage": {
1204
+ "present": true,
1205
+ "path_in_repo": "languages/spanish_24l/int8/flowlm_step.mlpackage",
1206
+ "size_bytes": 304120099,
1207
+ "size_human": "290.0 MiB"
1208
+ },
1209
+ "mlmodelc": {
1210
+ "present": true,
1211
+ "path_in_repo": "languages/spanish_24l/int8/flowlm_step.mlmodelc",
1212
+ "size_bytes": 304285013,
1213
+ "size_human": "290.2 MiB"
1214
+ }
1215
+ },
1216
+ "audio_quality": {
1217
+ "speaker_sim": 0.989,
1218
+ "pearson": 0.94,
1219
+ "verdict": "safe",
1220
+ "summary": "minor drift, well within identity threshold"
1221
+ }
1222
+ },
1223
+ "flow_decoder": {
1224
+ "fp16": {
1225
+ "path_in_repo": "languages/spanish_24l/flow_decoder.mlmodelc",
1226
+ "format": "mlmodelc",
1227
+ "note": "shipped separately; not part of this int8 upload"
1228
+ },
1229
+ "int8": {
1230
+ "mlpackage": {
1231
+ "present": true,
1232
+ "path_in_repo": "languages/spanish_24l/int8/flow_decoder.mlpackage",
1233
+ "size_bytes": 9923605,
1234
+ "size_human": "9.5 MiB"
1235
+ },
1236
+ "mlmodelc": {
1237
+ "present": true,
1238
+ "path_in_repo": "languages/spanish_24l/int8/flow_decoder.mlmodelc",
1239
+ "size_bytes": 9939641,
1240
+ "size_human": "9.5 MiB"
1241
+ }
1242
+ },
1243
+ "audio_quality": {
1244
+ "speaker_sim": 0.981,
1245
+ "pearson": 0.78,
1246
+ "verdict": "risky",
1247
+ "summary": "audible drift; 8-step LSD inner loop compounds quantization error"
1248
+ }
1249
+ },
1250
+ "mimi_decoder": {
1251
+ "fp16": {
1252
+ "path_in_repo": "languages/spanish_24l/mimi_decoder.mlmodelc",
1253
+ "format": "mlmodelc",
1254
+ "note": "shipped separately; not part of this int8 upload"
1255
+ },
1256
+ "int8": {
1257
+ "mlpackage": {
1258
+ "present": true,
1259
+ "path_in_repo": "languages/spanish_24l/int8/mimi_decoder.mlpackage",
1260
+ "size_bytes": 10867723,
1261
+ "size_human": "10.4 MiB"
1262
+ },
1263
+ "mlmodelc": {
1264
+ "present": true,
1265
+ "path_in_repo": "languages/spanish_24l/int8/mimi_decoder.mlmodelc",
1266
+ "size_bytes": 10903111,
1267
+ "size_human": "10.4 MiB"
1268
+ }
1269
+ },
1270
+ "audio_quality": {
1271
+ "speaker_sim": 0.998,
1272
+ "pearson": 1.0,
1273
+ "verdict": "transparent",
1274
+ "summary": "no audible difference vs fp16"
1275
+ }
1276
+ }
1277
+ }
1278
+ }
1279
+ },
1280
+ "totals": {
1281
+ "size_bytes": 7839895890,
1282
+ "size_human": "7.3 GiB",
1283
+ "file_count": 80
1284
+ }
1285
+ }