Preserve FP32 ReDimNet frontend and embedding head

Keep the learned backbone in FP16; publish compiled artifacts, export metadata, SHA-256 checksums and numerical validation.

README.md

@@ -22,8 +22,8 @@ samples. It does not diarize audio or assign names by itself.
 | Property | Value |
 |---|---:|
 | Parameters | 12.3 million |
-| Format | Compiled Core ML, Float16 weights |
-| Compiled size | 24.7 MiB |
+| Format | Compiled Core ML; FP32 frontend/head, FP16 backbone |
+| Compiled size | 28.6 MiB |
 | Input | 96,000 mono Float32 samples |
 | Sample rate | 16 kHz |
 | Window | 6 seconds |
@@ -35,29 +35,37 @@ shape avoids the slow Core ML fallback observed with a flexible waveform shape.
 Applications should repeat clean two-to-six-second speech to fill the input and
 center-crop longer samples.
 
+Export revision `frontend-fp32-v1` preserves waveform normalization, spectral
+and mel computation, log/feature normalization, pooling and output projection
+in FP32. The learned backbone remains FP16. This prevents numerical overflow
+in preprocessing without changing the checkpoint or input/output contract.
+
 ## Files
 
 | File | Size | Description |
 |---|---:|---|
-| `ReDimNet2B6.mlmodelc/` | 24.7 MiB | Precompiled Core ML model |
-| `config.json` | <2 KiB | Input, output, source revision, checksum, and validation metadata |
-| `README.md` | <4 KiB | This model card |
+| `ReDimNet2B6.mlmodelc/` | 28.6 MiB | Precompiled Core ML model |
+| `config.json` | <16 KiB | Contract, export revision, compiled-file SHA-256 and numerical validation |
+| `checksums.json` | <2 KiB | SHA-256 of every other published artifact |
+| `README.md` | <8 KiB | This model card |
 | `LICENSE` | 1.0 KiB | MIT license from the upstream implementation |
 
 ## Performance
 
-Measured on an Apple M2 Max after two warm-up predictions:
+Measured on an Apple M5 Pro after two warm-up predictions:
 
 | Measurement | Result | Meaning |
 |---|---:|---|
-| Warm six-second inference | 13.8 ms | One voice-profile embedding |
-| Warm throughput | 72.6 embeddings/s | Repeated six-second windows after warm-up |
-| Meeting pilot equal-error rate, 2-second clips | 1.50% | Lower is better; WeSpeaker Core ML was 5.17% |
-| Meeting pilot equal-error rate, 3-second clips | 0.00% | Lower is better; WeSpeaker Core ML was 1.50% |
-| LibriSpeech test-clean equal-error rate, 40 speakers | 0.00% | Two- and three-second controls |
-
-The meeting pilot contains five recurring speakers and is not a universal
-quality claim. Thresholds must be calibrated for the intended microphones,
+| Warm six-second inference | 13.6 ms | One voice-profile embedding |
+| Warm throughput | 73.6 embeddings/s | Repeated six-second windows after warm-up |
+| Numerical validation | 20/20 configurations and controls | Five deterministic controls across four allowed-device settings |
+
+Numerical validation checks finite, normalized outputs and cosine >= 0.999
+against the original FP32 model. Controls cover a noisy two-tone waveform,
+sparse burst, quiet waveform, repeated 600 ms waveform and silence. Allowed-device
+settings do not identify actual per-operation placement. These checks do not
+measure diarization, recognition accuracy, or speaker-verification error rate.
+Thresholds must be calibrated for the intended microphones,
 languages, and acoustic conditions. Speaker embeddings are useful for labeling;
 they are not biometric authentication and do not protect against voice spoofing.
 

ReDimNet2B6.mlmodelc/analytics/coremldata.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1fa08bf1dda00e0fa91ba683fdd0665880bf8f8a0bb4e42f40d844ad81b0e839
+oid sha256:05fe3e096940a0a13f1a4f2382108872e4a5c448cc552cefcfdebee93f096322
 size 243

ReDimNet2B6.mlmodelc/coremldata.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:81a52a3b06ef39f27a242e47b51d6d00b0dce4e57809d6c8b68677d89bf3e66b
+oid sha256:8b23b5a85f79b16a5798121193254e18b43096264c63afdbed90911a64ba1294
 size 518

ReDimNet2B6.mlmodelc/weights/weight.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dc2144871b34862429a4c82b28a413e201b52c89e6e66414b61fe9ba0e4ad7a2
-size 25423808
+oid sha256:780eb60151995b33d6a967deac32b9fe22ebd119e8288417f6d51ac7b856bcfc
+size 29519872

checksums.json

@@ -0,0 +1,9 @@
+{
+  "LICENSE": "3d5a4bf9936a7d09dd770588c0d7cb116fc16916a0310d71fb1fff5d64562101",
+  "README.md": "d799fb856ad8d9c04f34579c999ccac4d5f5abc56630fc70fd790ef3af76f1bd",
+  "ReDimNet2B6.mlmodelc/analytics/coremldata.bin": "05fe3e096940a0a13f1a4f2382108872e4a5c448cc552cefcfdebee93f096322",
+  "ReDimNet2B6.mlmodelc/coremldata.bin": "8b23b5a85f79b16a5798121193254e18b43096264c63afdbed90911a64ba1294",
+  "ReDimNet2B6.mlmodelc/model.mil": "8b138359651f98bcd876eab2377ed9a5ddaf5f726745775ce1670ac5103fc79d",
+  "ReDimNet2B6.mlmodelc/weights/weight.bin": "780eb60151995b33d6a967deac32b9fe22ebd119e8288417f6d51ac7b856bcfc",
+  "config.json": "7c4e9a6afcd5322f5bf6e5c7d2ec1d555221504c87fdaed3bc50fa197628c630"
+}

config.json

@@ -12,14 +12,165 @@
   "checkpoint_sha256": "e0a7d340a92f798720d1208949aa6a6bd0cddcb0ba7d4cec33596a17a484e6a2",
   "source_repository": "https://github.com/PalabraAI/redimnet2.git",
   "source_revision": "2a8d15f65b1dfb5d73fede2f11ee42bcccca3035",
-  "compute_precision": "float16",
+  "compute_precision": "fp32-frontend-head-fp16-backbone",
+  "artifact_revision": "frontend-fp32-v1",
+  "compiled_files_sha256": {
+    "analytics/coremldata.bin": "05fe3e096940a0a13f1a4f2382108872e4a5c448cc552cefcfdebee93f096322",
+    "coremldata.bin": "8b23b5a85f79b16a5798121193254e18b43096264c63afdbed90911a64ba1294",
+    "model.mil": "8b138359651f98bcd876eab2377ed9a5ddaf5f726745775ce1670ac5103fc79d",
+    "weights/weight.bin": "780eb60151995b33d6a967deac32b9fe22ebd119e8288417f6d51ac7b856bcfc"
+  },
   "compiled_model": "ReDimNet2B6.mlmodelc",
-  "compiled_size_bytes": 25931707,
+  "compiled_size_bytes": 30024704,
   "validation": {
-    "pytorch_coreml_cosine": 0.9999898663545027,
-    "max_absolute_error": 0.0011521875858306885,
-    "output_l2_norm": 0.9999155402183533,
-    "warm_latency_median_ms": 13.766916505119298,
-    "warm_latency_min_ms": 13.575500008300878
+    "pytorch_coreml_cosine": 0.9999944567677055,
+    "max_absolute_error": 0.0007698908448219299,
+    "output_l2_norm": 0.9999999403953552,
+    "numerical_trials": [
+      {
+        "control": "representative",
+        "configured_compute": "all",
+        "pytorch_coreml_cosine": 0.9999944567677055,
+        "max_absolute_error": 0.0007698908448219299,
+        "output_l2_norm": 0.9999999403953552
+      },
+      {
+        "control": "sparse_burst",
+        "configured_compute": "all",
+        "pytorch_coreml_cosine": 0.9999972581863403,
+        "max_absolute_error": 0.0005928315222263336,
+        "output_l2_norm": 0.9999999403953552
+      },
+      {
+        "control": "quiet",
+        "configured_compute": "all",
+        "pytorch_coreml_cosine": 0.9999944567677055,
+        "max_absolute_error": 0.0007817968726158142,
+        "output_l2_norm": 0.9999999403953552
+      },
+      {
+        "control": "short_600ms_repeated",
+        "configured_compute": "all",
+        "pytorch_coreml_cosine": 0.9999915361394329,
+        "max_absolute_error": 0.0008646771311759949,
+        "output_l2_norm": 0.9999999403953552
+      },
+      {
+        "control": "silence",
+        "configured_compute": "all",
+        "pytorch_coreml_cosine": 0.9999983906745911,
+        "max_absolute_error": 0.0004066694527864456,
+        "output_l2_norm": 1.0
+      },
+      {
+        "control": "representative",
+        "configured_compute": "cpuOnly",
+        "pytorch_coreml_cosine": 0.9996381401800001,
+        "max_absolute_error": 0.004830073099583387,
+        "output_l2_norm": 0.9999999403953552
+      },
+      {
+        "control": "sparse_burst",
+        "configured_compute": "cpuOnly",
+        "pytorch_coreml_cosine": 0.9994637966156006,
+        "max_absolute_error": 0.006543740630149841,
+        "output_l2_norm": 0.9999999403953552
+      },
+      {
+        "control": "quiet",
+        "configured_compute": "cpuOnly",
+        "pytorch_coreml_cosine": 0.9996398687362671,
+        "max_absolute_error": 0.004818218294531107,
+        "output_l2_norm": 1.0
+      },
+      {
+        "control": "short_600ms_repeated",
+        "configured_compute": "cpuOnly",
+        "pytorch_coreml_cosine": 0.999514043302181,
+        "max_absolute_error": 0.006073140539228916,
+        "output_l2_norm": 1.0
+      },
+      {
+        "control": "silence",
+        "configured_compute": "cpuOnly",
+        "pytorch_coreml_cosine": 0.9995816946029663,
+        "max_absolute_error": 0.004980172961950302,
+        "output_l2_norm": 1.0
+      },
+      {
+        "control": "representative",
+        "configured_compute": "cpuAndGPU",
+        "pytorch_coreml_cosine": 0.9999944567677055,
+        "max_absolute_error": 0.0007698908448219299,
+        "output_l2_norm": 0.9999999403953552
+      },
+      {
+        "control": "sparse_burst",
+        "configured_compute": "cpuAndGPU",
+        "pytorch_coreml_cosine": 0.9999972581863403,
+        "max_absolute_error": 0.0005928315222263336,
+        "output_l2_norm": 0.9999999403953552
+      },
+      {
+        "control": "quiet",
+        "configured_compute": "cpuAndGPU",
+        "pytorch_coreml_cosine": 0.9999944567677055,
+        "max_absolute_error": 0.0007817968726158142,
+        "output_l2_norm": 0.9999999403953552
+      },
+      {
+        "control": "short_600ms_repeated",
+        "configured_compute": "cpuAndGPU",
+        "pytorch_coreml_cosine": 0.9999915361394329,
+        "max_absolute_error": 0.0008646771311759949,
+        "output_l2_norm": 0.9999999403953552
+      },
+      {
+        "control": "silence",
+        "configured_compute": "cpuAndGPU",
+        "pytorch_coreml_cosine": 0.9999983906745911,
+        "max_absolute_error": 0.0004066694527864456,
+        "output_l2_norm": 1.0
+      },
+      {
+        "control": "representative",
+        "configured_compute": "cpuAndNeuralEngine",
+        "pytorch_coreml_cosine": 0.9997856616973877,
+        "max_absolute_error": 0.004807561635971069,
+        "output_l2_norm": 1.0
+      },
+      {
+        "control": "sparse_burst",
+        "configured_compute": "cpuAndNeuralEngine",
+        "pytorch_coreml_cosine": 0.999923825263977,
+        "max_absolute_error": 0.002673111855983734,
+        "output_l2_norm": 0.9999999403953552
+      },
+      {
+        "control": "quiet",
+        "configured_compute": "cpuAndNeuralEngine",
+        "pytorch_coreml_cosine": 0.9997879862785339,
+        "max_absolute_error": 0.004699621349573135,
+        "output_l2_norm": 1.0
+      },
+      {
+        "control": "short_600ms_repeated",
+        "configured_compute": "cpuAndNeuralEngine",
+        "pytorch_coreml_cosine": 0.9998489022074821,
+        "max_absolute_error": 0.0035514049232006073,
+        "output_l2_norm": 0.9999999403953552
+      },
+      {
+        "control": "silence",
+        "configured_compute": "cpuAndNeuralEngine",
+        "pytorch_coreml_cosine": 0.9998403787612915,
+        "max_absolute_error": 0.0037407726049423218,
+        "output_l2_norm": 1.0
+      }
+    ],
+    "numerical_trial_count": 20,
+    "minimum_pytorch_coreml_cosine": 0.9994637966156006,
+    "warm_latency_median_ms": 13.595416501630098,
+    "warm_latency_min_ms": 13.451875012833625
   }
 }