v2 manifest: 36 ships, FLEURS benchmarks, mlmodelc-only B1

manifest.json

@@ -1,10 +1,12 @@
 {
-  "name": "Nemotron 3.5 ASR Streaming Multilingual 0.6B — CoreML",
+  "name": "Nemotron 3.5 ASR Streaming Multilingual 0.6B \u2014 CoreML",
   "base_model": "nvidia/nemotron-asr-streaming-multilingual-0.6b",
+  "base_model_checkpoint": "2026-05-29 update",
   "architecture": "Conformer encoder + RNN-T decoder",
   "runtime": "CoreML / Apple Neural Engine",
   "benchmark_machine": "Apple M5 Pro / macOS 26.5",
-  "recipe": "LAYERPOS[42,13] mixed-precision encoder + per-language vocab pruning + B1 fusion + triple-stage pipelining + smart-spec K=4",
+  "benchmark_dataset": "FLEURS test (all languages)",
+  "recipe": "LAYERPOS [42,13] mixed-precision encoder (INT8 cuff + 6-bit middle) + Latin-script vocab prune (shared en/es/fr/it/pt/de) / full vocab (zh/ja/multilingual) + B1 decoder-joint fusion + triple-stage pipelining",
   "no_retraining": true,
   "no_calibration": true,
   "tiers_ms": [
@@ -14,26 +16,10 @@
     4480
   ],
   "recommended_tier_ms": 2240,
-  "models": [
-    "de",
-    "en",
-    "es",
-    "fr",
-    "it",
-    "ja",
-    "multilingual",
-    "pt",
-    "zh"
-  ],
-  "ship_count": 36,
-  "notes": [
-    "2240ms (2s) is the RTFx-per-latency sweet spot for every model (+20-44% over 1120ms, quality-neutral).",
-    "Multilingual peaks at 2240ms (74.6 RTFx); its 4480ms tier craters (19.4) because the 13088-vocab joint exceeds ANE working-set.",
-    "560ms is the lowest-latency tier but off the trained 14-frame attention tiling: lower RTFx (~57) and a small quality cost. en measured (57.2 RTFx); other 560ms ships shipped unbenched.",
-    "Portuguese 2240ms WER is +3.1pp vs 1120ms (B1-fallback path, chunk-sensitive).",
-    "de/zh/ja keep-sets were derived from FLEURS-test transcripts: in-domain/optimistic numbers + OOV risk on out-of-domain text. Rebuild keep-set from a broader corpus for production.",
-    "Each <lang>/<tier>ms/ dir is a self-contained FluidAudio model bundle (point --model-dir at it)."
-  ],
+  "vocab": {
+    "latin_shared": 2828,
+    "full": 13087
+  },
   "ships": [
     {
       "path": "en/560ms",
@@ -47,31 +33,27 @@
         42,
         13
       ],
-      "vocab_size": 988,
+      "vocab_size": 2828,
       "vocab_pruned": true,
+      "vocab_prune_method": "latin-script",
       "components": [
         "decoder",
         "decoder_joint",
-        "decoder_joint_noencproj",
         "encoder",
         "joint",
-        "joint_noencproj_batched",
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 57.2,
-        "wer_pct": 4.1,
-        "cer_pct": 1.5,
+        "rtfx": 57.5,
+        "wer_pct": 9.43,
         "metric": "WER",
-        "n": 2620,
-        "test_set": "LibriSpeech test-clean",
+        "n": 647,
+        "test_set": "FLEURS en_us",
         "benched": true
-      },
-      "recommended": false
+      }
     },
     {
       "path": "en/1120ms",
@@ -85,31 +67,27 @@
         42,
         13
       ],
-      "vocab_size": 988,
+      "vocab_size": 2828,
       "vocab_pruned": true,
+      "vocab_prune_method": "latin-script",
       "components": [
         "decoder",
         "decoder_joint",
-        "decoder_joint_noencproj",
         "encoder",
         "joint",
-        "joint_noencproj_batched",
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 101.4,
-        "wer_pct": 3.81,
-        "cer_pct": 1.4,
+        "rtfx": 102.9,
+        "wer_pct": 8.89,
         "metric": "WER",
-        "n": 2620,
-        "test_set": "LibriSpeech test-clean",
+        "n": 647,
+        "test_set": "FLEURS en_us",
         "benched": true
-      },
-      "recommended": false
+      }
     },
     {
       "path": "en/2240ms",
@@ -123,31 +101,27 @@
         42,
         13
       ],
-      "vocab_size": 988,
+      "vocab_size": 2828,
       "vocab_pruned": true,
+      "vocab_prune_method": "latin-script",
       "components": [
         "decoder",
         "decoder_joint",
-        "decoder_joint_noencproj",
         "encoder",
         "joint",
-        "joint_noencproj_batched",
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 134.7,
-        "wer_pct": 3.7,
-        "cer_pct": 1.4,
+        "rtfx": 130.2,
+        "wer_pct": 8.96,
         "metric": "WER",
-        "n": 2620,
-        "test_set": "LibriSpeech test-clean",
+        "n": 647,
+        "test_set": "FLEURS en_us",
         "benched": true
-      },
-      "recommended": true
+      }
     },
     {
       "path": "en/4480ms",
@@ -158,34 +132,30 @@
       "chunk_mel_frames": 448,
       "total_mel_frames": 457,
       "att_context": [
-        56,
+        42,
         13
       ],
-      "vocab_size": 989,
+      "vocab_size": 2828,
       "vocab_pruned": true,
+      "vocab_prune_method": "latin-script",
       "components": [
         "decoder",
         "decoder_joint",
-        "decoder_joint_noencproj",
         "encoder",
         "joint",
-        "joint_noencproj_batched",
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 136.7,
-        "wer_pct": 3.7,
-        "cer_pct": 1.4,
+        "rtfx": 122.1,
+        "wer_pct": 9.02,
         "metric": "WER",
-        "n": 2620,
-        "test_set": "LibriSpeech test-clean",
+        "n": 647,
+        "test_set": "FLEURS en_us",
         "benched": true
-      },
-      "recommended": false
+      }
     },
     {
       "path": "es/560ms",
@@ -199,8 +169,9 @@
         42,
         13
       ],
-      "vocab_size": 831,
+      "vocab_size": 2828,
       "vocab_pruned": true,
+      "vocab_prune_method": "latin-script",
       "components": [
         "decoder",
         "decoder_joint",
@@ -209,19 +180,16 @@
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": null,
-        "wer_pct": null,
-        "cer_pct": null,
+        "rtfx": 58.2,
+        "wer_pct": 4.95,
         "metric": "WER",
-        "n": 0,
-        "test_set": "MLS es_419",
-        "benched": false
-      },
-      "recommended": false
+        "n": 908,
+        "test_set": "FLEURS es_419",
+        "benched": true
+      }
     },
     {
       "path": "es/1120ms",
@@ -235,8 +203,9 @@
         42,
         13
       ],
-      "vocab_size": 831,
+      "vocab_size": 2828,
       "vocab_pruned": true,
+      "vocab_prune_method": "latin-script",
       "components": [
         "decoder",
         "decoder_joint",
@@ -245,19 +214,16 @@
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 106.8,
-        "wer_pct": 6.5,
-        "cer_pct": null,
+        "rtfx": 106.5,
+        "wer_pct": 4.76,
         "metric": "WER",
-        "n": 2385,
-        "test_set": "MLS es_419",
+        "n": 908,
+        "test_set": "FLEURS es_419",
         "benched": true
-      },
-      "recommended": false
+      }
     },
     {
       "path": "es/2240ms",
@@ -271,8 +237,9 @@
         42,
         13
       ],
-      "vocab_size": 831,
+      "vocab_size": 2828,
       "vocab_pruned": true,
+      "vocab_prune_method": "latin-script",
       "components": [
         "decoder",
         "decoder_joint",
@@ -281,19 +248,16 @@
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 153.6,
-        "wer_pct": 6.8,
-        "cer_pct": null,
+        "rtfx": 139.6,
+        "wer_pct": 4.8,
         "metric": "WER",
-        "n": 2385,
-        "test_set": "MLS es_419",
+        "n": 908,
+        "test_set": "FLEURS es_419",
         "benched": true
-      },
-      "recommended": true
+      }
     },
     {
       "path": "es/4480ms",
@@ -307,31 +271,27 @@
         42,
         13
       ],
-      "vocab_size": 831,
+      "vocab_size": 2828,
       "vocab_pruned": true,
+      "vocab_prune_method": "latin-script",
       "components": [
         "decoder",
         "decoder_joint",
-        "decoder_joint_noencproj",
         "encoder",
         "joint",
-        "joint_noencproj_batched",
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 139.0,
-        "wer_pct": 6.4,
-        "cer_pct": null,
+        "rtfx": 135.8,
+        "wer_pct": 4.77,
         "metric": "WER",
-        "n": 2385,
-        "test_set": "MLS es_419",
+        "n": 908,
+        "test_set": "FLEURS es_419",
         "benched": true
-      },
-      "recommended": false
+      }
     },
     {
       "path": "fr/560ms",
@@ -345,31 +305,27 @@
         42,
         13
       ],
-      "vocab_size": 848,
+      "vocab_size": 2828,
       "vocab_pruned": true,
+      "vocab_prune_method": "latin-script",
       "components": [
         "decoder",
         "decoder_joint",
-        "decoder_joint_noencproj",
         "encoder",
         "joint",
-        "joint_noencproj_batched",
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": null,
-        "wer_pct": null,
-        "cer_pct": null,
+        "rtfx": 57.4,
+        "wer_pct": 9.68,
         "metric": "WER",
-        "n": 0,
-        "test_set": "MLS fr_fr",
-        "benched": false
-      },
-      "recommended": false
+        "n": 676,
+        "test_set": "FLEURS fr_fr",
+        "benched": true
+      }
     },
     {
       "path": "fr/1120ms",
@@ -383,31 +339,27 @@
         42,
         13
       ],
-      "vocab_size": 848,
+      "vocab_size": 2828,
       "vocab_pruned": true,
+      "vocab_prune_method": "latin-script",
       "components": [
         "decoder",
         "decoder_joint",
-        "decoder_joint_noencproj",
         "encoder",
         "joint",
-        "joint_noencproj_batched",
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 109.5,
-        "wer_pct": 9.9,
-        "cer_pct": null,
+        "rtfx": 104.7,
+        "wer_pct": 9.44,
         "metric": "WER",
-        "n": 2426,
-        "test_set": "MLS fr_fr",
+        "n": 676,
+        "test_set": "FLEURS fr_fr",
         "benched": true
-      },
-      "recommended": false
+      }
     },
     {
       "path": "fr/2240ms",
@@ -421,31 +373,27 @@
         42,
         13
       ],
-      "vocab_size": 848,
+      "vocab_size": 2828,
       "vocab_pruned": true,
+      "vocab_prune_method": "latin-script",
       "components": [
         "decoder",
         "decoder_joint",
-        "decoder_joint_noencproj",
         "encoder",
         "joint",
-        "joint_noencproj_batched",
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 134.4,
-        "wer_pct": 10.4,
-        "cer_pct": null,
+        "rtfx": 130.4,
+        "wer_pct": 9.52,
         "metric": "WER",
-        "n": 2426,
-        "test_set": "MLS fr_fr",
+        "n": 676,
+        "test_set": "FLEURS fr_fr",
         "benched": true
-      },
-      "recommended": true
+      }
     },
     {
       "path": "fr/4480ms",
@@ -459,31 +407,27 @@
         42,
         13
       ],
-      "vocab_size": 848,
+      "vocab_size": 2828,
       "vocab_pruned": true,
+      "vocab_prune_method": "latin-script",
       "components": [
         "decoder",
         "decoder_joint",
-        "decoder_joint_noencproj",
         "encoder",
         "joint",
-        "joint_noencproj_batched",
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 130.2,
-        "wer_pct": 16.8,
-        "cer_pct": null,
+        "rtfx": 124.3,
+        "wer_pct": 9.42,
         "metric": "WER",
-        "n": 100,
-        "test_set": "MLS fr_fr",
+        "n": 676,
+        "test_set": "FLEURS fr_fr",
         "benched": true
-      },
-      "recommended": false
+      }
     },
     {
       "path": "it/560ms",
@@ -497,31 +441,27 @@
         42,
         13
       ],
-      "vocab_size": 805,
+      "vocab_size": 2828,
       "vocab_pruned": true,
+      "vocab_prune_method": "latin-script",
       "components": [
         "decoder",
         "decoder_joint",
-        "decoder_joint_noencproj",
         "encoder",
         "joint",
-        "joint_noencproj_batched",
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": null,
-        "wer_pct": null,
-        "cer_pct": null,
+        "rtfx": 59.0,
+        "wer_pct": 5.68,
         "metric": "WER",
-        "n": 0,
-        "test_set": "MLS it_it",
-        "benched": false
-      },
-      "recommended": false
+        "n": 865,
+        "test_set": "FLEURS it_it",
+        "benched": true
+      }
     },
     {
       "path": "it/1120ms",
@@ -535,31 +475,27 @@
         42,
         13
       ],
-      "vocab_size": 805,
+      "vocab_size": 2828,
       "vocab_pruned": true,
+      "vocab_prune_method": "latin-script",
       "components": [
         "decoder",
         "decoder_joint",
-        "decoder_joint_noencproj",
         "encoder",
         "joint",
-        "joint_noencproj_batched",
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 109.2,
-        "wer_pct": 23.0,
-        "cer_pct": null,
+        "rtfx": 109.0,
+        "wer_pct": 5.45,
         "metric": "WER",
-        "n": 1262,
-        "test_set": "MLS it_it",
+        "n": 865,
+        "test_set": "FLEURS it_it",
         "benched": true
-      },
-      "recommended": false
+      }
     },
     {
       "path": "it/2240ms",
@@ -573,31 +509,27 @@
         42,
         13
       ],
-      "vocab_size": 805,
+      "vocab_size": 2828,
       "vocab_pruned": true,
+      "vocab_prune_method": "latin-script",
       "components": [
         "decoder",
         "decoder_joint",
-        "decoder_joint_noencproj",
         "encoder",
         "joint",
-        "joint_noencproj_batched",
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 136.8,
-        "wer_pct": 18.0,
-        "cer_pct": null,
+        "rtfx": 146.7,
+        "wer_pct": 5.41,
         "metric": "WER",
-        "n": 1262,
-        "test_set": "MLS it_it",
+        "n": 865,
+        "test_set": "FLEURS it_it",
         "benched": true
-      },
-      "recommended": true
+      }
     },
     {
       "path": "it/4480ms",
@@ -611,31 +543,27 @@
         42,
         13
       ],
-      "vocab_size": 805,
+      "vocab_size": 2828,
       "vocab_pruned": true,
+      "vocab_prune_method": "latin-script",
       "components": [
         "decoder",
         "decoder_joint",
-        "decoder_joint_noencproj",
         "encoder",
         "joint",
-        "joint_noencproj_batched",
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 134.6,
-        "wer_pct": 25.6,
-        "cer_pct": null,
+        "rtfx": 150.5,
+        "wer_pct": 5.4,
         "metric": "WER",
-        "n": 100,
-        "test_set": "MLS it_it",
+        "n": 865,
+        "test_set": "FLEURS it_it",
         "benched": true
-      },
-      "recommended": false
+      }
     },
     {
       "path": "pt/560ms",
@@ -649,8 +577,9 @@
         42,
         13
       ],
-      "vocab_size": 870,
+      "vocab_size": 2828,
       "vocab_pruned": true,
+      "vocab_prune_method": "latin-script",
       "components": [
         "decoder",
         "decoder_joint",
@@ -659,19 +588,16 @@
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": null,
-        "wer_pct": null,
-        "cer_pct": null,
+        "rtfx": 58.7,
+        "wer_pct": 6.38,
         "metric": "WER",
-        "n": 0,
-        "test_set": "MLS pt_br",
-        "benched": false
-      },
-      "recommended": false
+        "n": 919,
+        "test_set": "FLEURS pt_br",
+        "benched": true
+      }
     },
     {
       "path": "pt/1120ms",
@@ -685,8 +611,9 @@
         42,
         13
       ],
-      "vocab_size": 870,
+      "vocab_size": 2828,
       "vocab_pruned": true,
+      "vocab_prune_method": "latin-script",
       "components": [
         "decoder",
         "decoder_joint",
@@ -695,19 +622,16 @@
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 111.4,
-        "wer_pct": 9.8,
-        "cer_pct": null,
+        "rtfx": 107.6,
+        "wer_pct": 6.11,
         "metric": "WER",
-        "n": 871,
-        "test_set": "MLS pt_br",
+        "n": 919,
+        "test_set": "FLEURS pt_br",
         "benched": true
-      },
-      "recommended": false
+      }
     },
     {
       "path": "pt/2240ms",
@@ -721,8 +645,9 @@
         42,
         13
       ],
-      "vocab_size": 870,
+      "vocab_size": 2828,
       "vocab_pruned": true,
+      "vocab_prune_method": "latin-script",
       "components": [
         "decoder",
         "decoder_joint",
@@ -731,19 +656,16 @@
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 155.2,
-        "wer_pct": 12.9,
-        "cer_pct": null,
+        "rtfx": 141.0,
+        "wer_pct": 6.14,
         "metric": "WER",
-        "n": 871,
-        "test_set": "MLS pt_br",
+        "n": 919,
+        "test_set": "FLEURS pt_br",
         "benched": true
-      },
-      "recommended": true
+      }
     },
     {
       "path": "pt/4480ms",
@@ -757,31 +679,27 @@
         42,
         13
       ],
-      "vocab_size": 870,
+      "vocab_size": 2828,
       "vocab_pruned": true,
+      "vocab_prune_method": "latin-script",
       "components": [
         "decoder",
         "decoder_joint",
-        "decoder_joint_noencproj",
         "encoder",
         "joint",
-        "joint_noencproj_batched",
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 134.9,
-        "wer_pct": 10.3,
-        "cer_pct": null,
+        "rtfx": 141.3,
+        "wer_pct": 6.18,
         "metric": "WER",
-        "n": 871,
-        "test_set": "MLS pt_br",
+        "n": 919,
+        "test_set": "FLEURS pt_br",
         "benched": true
-      },
-      "recommended": false
+      }
     },
     {
       "path": "de/560ms",
@@ -795,8 +713,9 @@
         42,
         13
       ],
-      "vocab_size": 795,
+      "vocab_size": 2828,
       "vocab_pruned": true,
+      "vocab_prune_method": "latin-script",
       "components": [
         "decoder",
         "decoder_joint",
@@ -805,19 +724,16 @@
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": null,
-        "wer_pct": null,
-        "cer_pct": null,
+        "rtfx": 58.8,
+        "wer_pct": 10.83,
         "metric": "WER",
-        "n": 0,
+        "n": 862,
         "test_set": "FLEURS de_de",
-        "benched": false
-      },
-      "recommended": false
+        "benched": true
+      }
     },
     {
       "path": "de/1120ms",
@@ -831,8 +747,9 @@
         42,
         13
       ],
-      "vocab_size": 795,
+      "vocab_size": 2828,
       "vocab_pruned": true,
+      "vocab_prune_method": "latin-script",
       "components": [
         "decoder",
         "decoder_joint",
@@ -841,19 +758,16 @@
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 107.5,
-        "wer_pct": 11.2,
-        "cer_pct": 3.6,
+        "rtfx": 107.2,
+        "wer_pct": 9.78,
         "metric": "WER",
         "n": 862,
         "test_set": "FLEURS de_de",
         "benched": true
-      },
-      "recommended": false
+      }
     },
     {
       "path": "de/2240ms",
@@ -867,8 +781,9 @@
         42,
         13
       ],
-      "vocab_size": 795,
+      "vocab_size": 2828,
       "vocab_pruned": true,
+      "vocab_prune_method": "latin-script",
       "components": [
         "decoder",
         "decoder_joint",
@@ -877,19 +792,16 @@
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 150.1,
-        "wer_pct": 11.0,
-        "cer_pct": 3.8,
+        "rtfx": 144.4,
+        "wer_pct": 9.83,
         "metric": "WER",
         "n": 862,
         "test_set": "FLEURS de_de",
         "benched": true
-      },
-      "recommended": true
+      }
     },
     {
       "path": "de/4480ms",
@@ -903,8 +815,9 @@
         42,
         13
       ],
-      "vocab_size": 795,
+      "vocab_size": 2828,
       "vocab_pruned": true,
+      "vocab_prune_method": "latin-script",
       "components": [
         "decoder",
         "decoder_joint",
@@ -913,19 +826,16 @@
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 151.9,
-        "wer_pct": 11.0,
-        "cer_pct": 3.9,
+        "rtfx": 141.5,
+        "wer_pct": 9.83,
         "metric": "WER",
         "n": 862,
         "test_set": "FLEURS de_de",
         "benched": true
-      },
-      "recommended": false
+      }
     },
     {
       "path": "zh/560ms",
@@ -939,8 +849,9 @@
         42,
         13
       ],
-      "vocab_size": 1875,
-      "vocab_pruned": true,
+      "vocab_size": 13087,
+      "vocab_pruned": false,
+      "vocab_prune_method": "none (full vocab)",
       "components": [
         "decoder",
         "decoder_joint",
@@ -949,19 +860,16 @@
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": null,
-        "wer_pct": null,
-        "cer_pct": null,
+        "rtfx": 22.5,
+        "cer_pct": 19.48,
         "metric": "CER",
-        "n": 0,
+        "n": 945,
         "test_set": "FLEURS cmn_hans_cn",
-        "benched": false
-      },
-      "recommended": false
+        "benched": true
+      }
     },
     {
       "path": "zh/1120ms",
@@ -975,8 +883,9 @@
         42,
         13
       ],
-      "vocab_size": 1875,
-      "vocab_pruned": true,
+      "vocab_size": 13087,
+      "vocab_pruned": false,
+      "vocab_prune_method": "none (full vocab)",
       "components": [
         "decoder",
         "decoder_joint",
@@ -985,19 +894,16 @@
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 106.3,
-        "wer_pct": null,
-        "cer_pct": 21.9,
+        "rtfx": 26.6,
+        "cer_pct": 18.75,
         "metric": "CER",
         "n": 945,
         "test_set": "FLEURS cmn_hans_cn",
         "benched": true
-      },
-      "recommended": false
+      }
     },
     {
       "path": "zh/2240ms",
@@ -1011,8 +917,9 @@
         42,
         13
       ],
-      "vocab_size": 1875,
-      "vocab_pruned": true,
+      "vocab_size": 13087,
+      "vocab_pruned": false,
+      "vocab_prune_method": "none (full vocab)",
       "components": [
         "decoder",
         "decoder_joint",
@@ -1021,19 +928,16 @@
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 146.0,
-        "wer_pct": null,
-        "cer_pct": 21.4,
+        "rtfx": 89.0,
+        "cer_pct": 18.57,
         "metric": "CER",
         "n": 945,
         "test_set": "FLEURS cmn_hans_cn",
         "benched": true
-      },
-      "recommended": true
+      }
     },
     {
       "path": "zh/4480ms",
@@ -1047,8 +951,9 @@
         42,
         13
       ],
-      "vocab_size": 1875,
-      "vocab_pruned": true,
+      "vocab_size": 13087,
+      "vocab_pruned": false,
+      "vocab_prune_method": "none (full vocab)",
       "components": [
         "decoder",
         "decoder_joint",
@@ -1057,19 +962,16 @@
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 140.2,
-        "wer_pct": null,
-        "cer_pct": 21.4,
+        "rtfx": 89.6,
+        "cer_pct": 18.05,
         "metric": "CER",
         "n": 945,
         "test_set": "FLEURS cmn_hans_cn",
         "benched": true
-      },
-      "recommended": false
+      }
     },
     {
       "path": "ja/560ms",
@@ -1083,8 +985,9 @@
         42,
         13
       ],
-      "vocab_size": 1403,
-      "vocab_pruned": true,
+      "vocab_size": 13087,
+      "vocab_pruned": false,
+      "vocab_prune_method": "none (full vocab)",
       "components": [
         "decoder",
         "decoder_joint",
@@ -1093,19 +996,16 @@
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": null,
-        "wer_pct": null,
-        "cer_pct": null,
+        "rtfx": 20.7,
+        "cer_pct": 14.61,
         "metric": "CER",
-        "n": 0,
+        "n": 650,
         "test_set": "FLEURS ja_jp",
-        "benched": false
-      },
-      "recommended": false
+        "benched": true
+      }
     },
     {
       "path": "ja/1120ms",
@@ -1119,8 +1019,9 @@
         42,
         13
       ],
-      "vocab_size": 1403,
-      "vocab_pruned": true,
+      "vocab_size": 13087,
+      "vocab_pruned": false,
+      "vocab_prune_method": "none (full vocab)",
       "components": [
         "decoder",
         "decoder_joint",
@@ -1129,19 +1030,16 @@
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 108.5,
-        "wer_pct": null,
-        "cer_pct": 15.6,
+        "rtfx": 25.9,
+        "cer_pct": 13.77,
         "metric": "CER",
         "n": 650,
         "test_set": "FLEURS ja_jp",
         "benched": true
-      },
-      "recommended": false
+      }
     },
     {
       "path": "ja/2240ms",
@@ -1155,8 +1053,9 @@
         42,
         13
       ],
-      "vocab_size": 1403,
-      "vocab_pruned": true,
+      "vocab_size": 13087,
+      "vocab_pruned": false,
+      "vocab_prune_method": "none (full vocab)",
       "components": [
         "decoder",
         "decoder_joint",
@@ -1165,19 +1064,16 @@
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 150.5,
-        "wer_pct": null,
-        "cer_pct": 15.4,
+        "rtfx": 84.2,
+        "cer_pct": 13.79,
         "metric": "CER",
         "n": 650,
         "test_set": "FLEURS ja_jp",
         "benched": true
-      },
-      "recommended": true
+      }
     },
     {
       "path": "ja/4480ms",
@@ -1191,8 +1087,9 @@
         42,
         13
       ],
-      "vocab_size": 1403,
-      "vocab_pruned": true,
+      "vocab_size": 13087,
+      "vocab_pruned": false,
+      "vocab_prune_method": "none (full vocab)",
       "components": [
         "decoder",
         "decoder_joint",
@@ -1201,23 +1098,20 @@
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 147.8,
-        "wer_pct": null,
-        "cer_pct": 15.3,
+        "rtfx": 89.3,
+        "cer_pct": 13.82,
         "metric": "CER",
         "n": 650,
         "test_set": "FLEURS ja_jp",
         "benched": true
-      },
-      "recommended": false
+      }
     },
     {
       "path": "multilingual/560ms",
-      "language": "Multilingual (100+ langs via prompt_id)",
+      "language": "Multilingual",
       "language_code": "auto",
       "chunk_ms": 560,
       "latency_s": 0.56,
@@ -1229,33 +1123,29 @@
       ],
       "vocab_size": 13087,
       "vocab_pruned": false,
+      "vocab_prune_method": "none (full vocab)",
       "components": [
         "decoder",
         "decoder_joint",
-        "decoder_joint_noencproj",
         "encoder",
         "joint",
-        "joint_noencproj_batched",
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": null,
-        "wer_pct": null,
-        "cer_pct": null,
+        "rtfx": 23.4,
+        "wer_pct": 9.15,
         "metric": "WER",
-        "n": 0,
-        "test_set": "LibriSpeech test-clean (en)",
-        "benched": false
-      },
-      "recommended": false
+        "n": 647,
+        "test_set": "FLEURS en_us",
+        "benched": true
+      }
     },
     {
       "path": "multilingual/1120ms",
-      "language": "Multilingual (100+ langs via prompt_id)",
+      "language": "Multilingual",
       "language_code": "auto",
       "chunk_ms": 1120,
       "latency_s": 1.12,
@@ -1267,33 +1157,29 @@
       ],
       "vocab_size": 13087,
       "vocab_pruned": false,
+      "vocab_prune_method": "none (full vocab)",
       "components": [
         "decoder",
         "decoder_joint",
-        "decoder_joint_noencproj",
         "encoder",
         "joint",
-        "joint_noencproj_batched",
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 62.2,
-        "wer_pct": 3.8,
-        "cer_pct": 1.4,
+        "rtfx": 70.9,
+        "wer_pct": 8.64,
         "metric": "WER",
-        "n": 2620,
-        "test_set": "LibriSpeech test-clean (en)",
+        "n": 647,
+        "test_set": "FLEURS en_us",
         "benched": true
-      },
-      "recommended": false
+      }
     },
     {
       "path": "multilingual/2240ms",
-      "language": "Multilingual (100+ langs via prompt_id)",
+      "language": "Multilingual",
       "language_code": "auto",
       "chunk_ms": 2240,
       "latency_s": 2.24,
@@ -1305,33 +1191,29 @@
       ],
       "vocab_size": 13087,
       "vocab_pruned": false,
+      "vocab_prune_method": "none (full vocab)",
       "components": [
         "decoder",
         "decoder_joint",
-        "decoder_joint_noencproj",
         "encoder",
         "joint",
-        "joint_noencproj_batched",
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 74.6,
-        "wer_pct": 3.7,
-        "cer_pct": 1.4,
+        "rtfx": 80.4,
+        "wer_pct": 8.76,
         "metric": "WER",
-        "n": 2620,
-        "test_set": "LibriSpeech test-clean (en)",
+        "n": 647,
+        "test_set": "FLEURS en_us",
         "benched": true
-      },
-      "recommended": true
+      }
     },
     {
       "path": "multilingual/4480ms",
-      "language": "Multilingual (100+ langs via prompt_id)",
+      "language": "Multilingual",
       "language_code": "auto",
       "chunk_ms": 4480,
       "latency_s": 4.48,
@@ -1343,29 +1225,25 @@
       ],
       "vocab_size": 13087,
       "vocab_pruned": false,
+      "vocab_prune_method": "none (full vocab)",
       "components": [
         "decoder",
         "decoder_joint",
-        "decoder_joint_noencproj",
         "encoder",
         "joint",
-        "joint_noencproj_batched",
         "preprocessor"
       ],
       "formats": [
-        "mlpackage",
         "mlmodelc"
       ],
       "benchmark": {
-        "rtfx": 19.4,
-        "wer_pct": 3.7,
-        "cer_pct": 1.4,
+        "rtfx": 78.0,
+        "wer_pct": 8.78,
         "metric": "WER",
-        "n": 2620,
-        "test_set": "LibriSpeech test-clean (en)",
+        "n": 647,
+        "test_set": "FLEURS en_us",
         "benched": true
-      },
-      "recommended": false
+      }
     }
   ]
 }