Upload 35 files

README.md

@@ -0,0 +1,131 @@
+# Parakeet Realtime EOU 120M - CoreML
+
+CoreML conversion of [nvidia/parakeet_realtime_eou_120m-v1](https://huggingface.co/nvidia/parakeet_realtime_eou_120m-v1) for Apple platforms.
+
+## Model Overview
+
+| Property | Value |
+|----------|-------|
+| **Architecture** | FastConformer-RNNT (Cache-aware streaming) |
+| **Parameters** | 120M |
+| **Sample Rate** | 16kHz mono |
+| **Vocab Size** | 1026 (+ blank = 1027) |
+| **Max Audio** | 15 seconds (240,000 samples) |
+| **Latency** | 80-160ms streaming |
+
+## Key Features
+
+- **End-of-Utterance Detection**: Emits `<EOU>` token (ID 1024) when speaker finishes
+- **Low Latency**: Optimized for real-time streaming with 80-160ms latency
+- **Compact**: 120M parameters vs 600M for Parakeet TDT v3
+
+## Components
+
+| Component | File | Size | Description |
+|-----------|------|------|-------------|
+| Preprocessor | `parakeet_eou_preprocessor.mlmodelc` | 600KB | Audio waveform to mel spectrogram |
+| Encoder | `parakeet_eou_encoder.mlmodelc` | 207MB | Mel spectrogram to encoder features |
+| Decoder | `parakeet_eou_decoder.mlmodelc` | 7.5MB | RNNT prediction network (LSTM) |
+| Joint | `parakeet_eou_joint_decision_single_step.mlmodelc` | 2.7MB | Joint network with argmax decision |
+
+## Input/Output Shapes
+
+### Preprocessor
+- **Input**: `audio_signal` [1, S], `audio_length` [1]
+- **Output**: `mel` [1, 128, T], `mel_length` [1]
+
+### Encoder
+- **Input**: `mel` [1, 128, 1501], `mel_length` [1]
+- **Output**: `encoder` [1, 512, 189], `encoder_length` [1], `frame_times` [1, 189]
+
+### Decoder
+- **Input**: `targets` [1, 1], `target_length` [1], `h_in` [1, 1, 640], `c_in` [1, 1, 640]
+- **Output**: `decoder` [1, 640, 1], `h_out` [1, 1, 640], `c_out` [1, 1, 640]
+
+### Joint Decision Single Step
+- **Input**: `encoder_step` [1, 512, 1], `decoder_step` [1, 640, 1]
+- **Output**: `token_id` [1, 1, 1], `token_prob` [1, 1, 1], `top_k_ids` [1, 1, 1, 64], `top_k_logits` [1, 1, 1, 64]
+
+## RNNT Decoding Loop
+
+```swift
+// Initialize decoder state
+var h = zeros([1, 1, 640])
+var c = zeros([1, 1, 640])
+var lastToken = blankId  // 1026
+
+// Process each encoder frame
+for frameIdx in 0..<encoderLength {
+    let encoderStep = encoder[:, :, frameIdx:frameIdx+1]
+
+    while true {
+        // Run decoder with last token
+        let (decoderOut, hNew, cNew) = decoder(lastToken, h, c)
+
+        // Get token decision
+        let (tokenId, prob, _, _) = joint(encoderStep, decoderOut)
+
+        if tokenId == blankId {
+            break  // Move to next frame
+        }
+
+        // Emit token
+        tokens.append(tokenId)
+        lastToken = tokenId
+        h = hNew
+        c = cNew
+
+        // Check for end-of-utterance
+        if tokenId == eouId {  // 1024
+            // Utterance complete
+        }
+    }
+}
+```
+
+## Special Tokens
+
+| Token | ID | Description |
+|-------|-----|-------------|
+| `<EOU>` | 1024 | End-of-utterance marker |
+| `<EOB>` | 1025 | End-of-block marker |
+| `<blank>` | 1026 | RNNT blank token |
+
+## Platform Requirements
+
+- iOS 17.0+
+- macOS 14.0+
+- Compute Units: CPU_ONLY (compatible with all devices)
+
+## Files
+
+```
+parakeet_eou_coreml/
+├── metadata.json                                    # Model configuration
+├── vocab.json                                       # Tokenizer vocabulary
+├── parakeet_eou_preprocessor.mlmodelc/             # Compiled preprocessor
+├── parakeet_eou_encoder.mlmodelc/                  # Compiled encoder
+├── parakeet_eou_decoder.mlmodelc/                  # Compiled decoder
+├── parakeet_eou_joint_decision_single_step.mlmodelc/ # Compiled joint
+├── parakeet_eou_preprocessor.mlpackage/            # Source preprocessor
+├── parakeet_eou_encoder.mlpackage/                 # Source encoder
+├── parakeet_eou_decoder.mlpackage/                 # Source decoder
+└── parakeet_eou_joint_decision_single_step.mlpackage/ # Source joint
+```
+
+## License
+
+This model conversion follows the license of the original NVIDIA model.
+See [nvidia/parakeet_realtime_eou_120m-v1](https://huggingface.co/nvidia/parakeet_realtime_eou_120m-v1) for details.
+
+## Citation
+
+```bibtex
+@misc{nvidia_parakeet_eou,
+  title={Parakeet Realtime EOU 120M},
+  author={NVIDIA},
+  year={2024},
+  publisher={HuggingFace},
+  url={https://huggingface.co/nvidia/parakeet_realtime_eou_120m-v1}
+}
+```

metadata.json

@@ -0,0 +1,88 @@
+{
+  "model_id": "nvidia/parakeet_realtime_eou_120m-v1",
+  "model_name": "parakeet_realtime_eou_120m-v1",
+  "model_class": "EncDecRNNTBPEModel",
+  "encoder_class": "ConformerEncoder",
+  "sample_rate": 16000,
+  "max_audio_seconds": 15.0,
+  "max_audio_samples": 240000,
+  "max_symbol_steps": 1,
+  "vocab_size": 1026,
+  "vocab_with_blank": 1027,
+  "decoder_hidden": 640,
+  "decoder_layers": 1,
+  "num_extra_outputs": 0,
+  "has_eou_token": true,
+  "checkpoint": {
+    "type": "pretrained",
+    "model_id": "nvidia/parakeet_realtime_eou_120m-v1"
+  },
+  "coreml": {
+    "compute_units": "CPU_ONLY",
+    "compute_precision": "FLOAT32"
+  },
+  "components": {
+    "preprocessor": {
+      "inputs": {
+        "audio_signal": [1, 240000],
+        "audio_length": [1]
+      },
+      "outputs": {
+        "mel": [1, 128, 1501],
+        "mel_length": [1]
+      },
+      "path": "parakeet_eou_preprocessor.mlpackage"
+    },
+    "encoder": {
+      "inputs": {
+        "mel": [1, 128, 1501],
+        "mel_length": [1]
+      },
+      "outputs": {
+        "encoder": [1, 512, 189],
+        "encoder_length": [1],
+        "frame_times": [1, 189]
+      },
+      "path": "parakeet_eou_encoder.mlpackage"
+    },
+    "decoder": {
+      "inputs": {
+        "targets": [1, 1],
+        "target_length": [1],
+        "h_in": [1, 1, 640],
+        "c_in": [1, 1, 640]
+      },
+      "outputs": {
+        "decoder": [1, 640, 1],
+        "h_out": [1, 1, 640],
+        "c_out": [1, 1, 640]
+      },
+      "path": "parakeet_eou_decoder.mlpackage"
+    },
+    "joint_decision_single_step": {
+      "inputs": {
+        "encoder_step": [1, 512, 1],
+        "decoder_step": [1, 640, 1]
+      },
+      "outputs": {
+        "token_id": [1, 1, 1],
+        "token_prob": [1, 1, 1],
+        "top_k_ids": [1, 1, 1, 64],
+        "top_k_logits": [1, 1, 1, 64]
+      },
+      "path": "parakeet_eou_joint_decision_single_step.mlpackage"
+    }
+  },
+  "tokenizer": {
+    "blank_id": 1026,
+    "vocab_size": 1026,
+    "eou_token": {
+      "token": "<EOU>",
+      "id": 1024
+    },
+    "eob_token": {
+      "token": "<EOB>",
+      "id": 1025
+    }
+  }
+}

parakeet_eou_decoder.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b5608d90a0bd35f0abbc10bd19588ff5689e9c30bb663d37962c03720eb76d70
+size 243

parakeet_eou_decoder.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:87a8d3145e2702a6999ede333e8155164121a391e531193b18eff88312b01479
+size 556

parakeet_eou_decoder.mlmodelc/metadata.json

@@ -0,0 +1,122 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "shortDescription" : "Parakeet EOU decoder (RNNT prediction network)",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 640 × 1)",
+        "shortDescription" : "",
+        "shape" : "[1, 640, 1]",
+        "name" : "decoder",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 1 × 640)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 640]",
+        "name" : "h_out",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 1 × 640)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 640]",
+        "name" : "c_out",
+        "type" : "MultiArray"
+      }
+    ],
+    "storagePrecision" : "Float16",
+    "modelParameters" : [
+
+    ],
+    "author" : "Fluid Inference",
+    "specificationVersion" : 8,
+    "mlProgramOperationTypeHistogram" : {
+      "Select" : 1,
+      "Ios17.squeeze" : 2,
+      "Ios17.gather" : 1,
+      "Ios17.cast" : 8,
+      "Ios17.lstm" : 1,
+      "Ios17.transpose" : 2,
+      "Ios17.add" : 1,
+      "Identity" : 1,
+      "Ios17.greaterEqual" : 1,
+      "Ios17.expandDims" : 2
+    },
+    "computePrecision" : "Mixed (Float16, Float32, Int16, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "14.0",
+      "tvOS" : "17.0",
+      "visionOS" : "1.0",
+      "watchOS" : "10.0",
+      "iOS" : "17.0",
+      "macCatalyst" : "17.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "inputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1 × 1)",
+        "shortDescription" : "",
+        "shape" : "[1, 1]",
+        "name" : "targets",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "target_length",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 1 × 640)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 640]",
+        "name" : "h_in",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 1 × 640)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 640]",
+        "name" : "c_in",
+        "type" : "MultiArray"
+      }
+    ],
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2025-11-27",
+      "com.github.apple.coremltools.source" : "torch==2.9.1",
+      "com.github.apple.coremltools.version" : "9.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "generatedClassName" : "parakeet_eou_decoder",
+    "method" : "predict"
+  }
+]

parakeet_eou_decoder.mlmodelc/model.mil

@@ -0,0 +1,54 @@
+program(1.0)
+[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3500.14.1"}, {"coremlc-version", "3500.32.1"}, {"coremltools-component-torch", "2.9.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})]
+{
+    func main<ios17>(tensor<fp32, [1, 1, 640]> c_in, tensor<fp32, [1, 1, 640]> h_in, tensor<int32, [1]> target_length, tensor<int32, [1, 1]> targets) {
+            tensor<int32, []> y_batch_dims_0 = const()[name = tensor<string, []>("y_batch_dims_0"), val = tensor<int32, []>(0)];
+            tensor<bool, []> y_validate_indices_0 = const()[name = tensor<string, []>("y_validate_indices_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1027, 640]> module_prediction_embed_weight_to_fp16 = const()[name = tensor<string, []>("module_prediction_embed_weight_to_fp16"), val = tensor<fp16, [1027, 640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
+            tensor<string, []> targets_to_int16_dtype_0 = const()[name = tensor<string, []>("targets_to_int16_dtype_0"), val = tensor<string, []>("int16")];
+            tensor<string, []> cast_1_dtype_0 = const()[name = tensor<string, []>("cast_1_dtype_0"), val = tensor<string, []>("int32")];
+            tensor<int32, []> greater_equal_0_y_0 = const()[name = tensor<string, []>("greater_equal_0_y_0"), val = tensor<int32, []>(0)];
+            tensor<int16, [1, 1]> targets_to_int16 = cast(dtype = targets_to_int16_dtype_0, x = targets)[name = tensor<string, []>("cast_9")];
+            tensor<int32, [1, 1]> cast_1 = cast(dtype = cast_1_dtype_0, x = targets_to_int16)[name = tensor<string, []>("cast_8")];
+            tensor<bool, [1, 1]> greater_equal_0 = greater_equal(x = cast_1, y = greater_equal_0_y_0)[name = tensor<string, []>("greater_equal_0")];
+            tensor<int32, []> slice_by_index_0 = const()[name = tensor<string, []>("slice_by_index_0"), val = tensor<int32, []>(1027)];
+            tensor<int32, [1, 1]> add_1 = add(x = cast_1, y = slice_by_index_0)[name = tensor<string, []>("add_1")];
+            tensor<int32, [1, 1]> select_0 = select(a = cast_1, b = add_1, cond = greater_equal_0)[name = tensor<string, []>("select_0")];
+            tensor<int32, []> y_cast_fp16_cast_uint16_axis_0 = const()[name = tensor<string, []>("y_cast_fp16_cast_uint16_axis_0"), val = tensor<int32, []>(0)];
+            tensor<string, []> select_0_to_int16_dtype_0 = const()[name = tensor<string, []>("select_0_to_int16_dtype_0"), val = tensor<string, []>("int16")];
+            tensor<int16, [1, 1]> select_0_to_int16 = cast(dtype = select_0_to_int16_dtype_0, x = select_0)[name = tensor<string, []>("cast_7")];
+            tensor<fp16, [1, 1, 640]> y_cast_fp16_cast_uint16_cast_uint16 = gather(axis = y_cast_fp16_cast_uint16_axis_0, batch_dims = y_batch_dims_0, indices = select_0_to_int16, validate_indices = y_validate_indices_0, x = module_prediction_embed_weight_to_fp16)[name = tensor<string, []>("y_cast_fp16_cast_uint16_cast_uint16")];
+            tensor<int32, [3]> input_3_perm_0 = const()[name = tensor<string, []>("input_3_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
+            tensor<int32, [1]> input_lstm_h0_squeeze_axes_0 = const()[name = tensor<string, []>("input_lstm_h0_squeeze_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<string, []> h_in_to_fp16_dtype_0 = const()[name = tensor<string, []>("h_in_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
+            tensor<fp16, [1, 1, 640]> h_in_to_fp16 = cast(dtype = h_in_to_fp16_dtype_0, x = h_in)[name = tensor<string, []>("cast_6")];
+            tensor<fp16, [1, 640]> input_lstm_h0_squeeze_cast_fp16 = squeeze(axes = input_lstm_h0_squeeze_axes_0, x = h_in_to_fp16)[name = tensor<string, []>("input_lstm_h0_squeeze_cast_fp16")];
+            tensor<int32, [1]> input_lstm_c0_squeeze_axes_0 = const()[name = tensor<string, []>("input_lstm_c0_squeeze_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<string, []> c_in_to_fp16_dtype_0 = const()[name = tensor<string, []>("c_in_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
+            tensor<fp16, [1, 1, 640]> c_in_to_fp16 = cast(dtype = c_in_to_fp16_dtype_0, x = c_in)[name = tensor<string, []>("cast_5")];
+            tensor<fp16, [1, 640]> input_lstm_c0_squeeze_cast_fp16 = squeeze(axes = input_lstm_c0_squeeze_axes_0, x = c_in_to_fp16)[name = tensor<string, []>("input_lstm_c0_squeeze_cast_fp16")];
+            tensor<string, []> input_direction_0 = const()[name = tensor<string, []>("input_direction_0"), val = tensor<string, []>("forward")];
+            tensor<bool, []> input_output_sequence_0 = const()[name = tensor<string, []>("input_output_sequence_0"), val = tensor<bool, []>(true)];
+            tensor<string, []> input_recurrent_activation_0 = const()[name = tensor<string, []>("input_recurrent_activation_0"), val = tensor<string, []>("sigmoid")];
+            tensor<string, []> input_cell_activation_0 = const()[name = tensor<string, []>("input_cell_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<string, []> input_activation_0 = const()[name = tensor<string, []>("input_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<fp16, [2560, 640]> concat_1_to_fp16 = const()[name = tensor<string, []>("concat_1_to_fp16"), val = tensor<fp16, [2560, 640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1314688)))];
+            tensor<fp16, [2560, 640]> concat_2_to_fp16 = const()[name = tensor<string, []>("concat_2_to_fp16"), val = tensor<fp16, [2560, 640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4591552)))];
+            tensor<fp16, [2560]> concat_0_to_fp16 = const()[name = tensor<string, []>("concat_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7868416)))];
+            tensor<fp16, [1, 1, 640]> input_3_cast_fp16 = transpose(perm = input_3_perm_0, x = y_cast_fp16_cast_uint16_cast_uint16)[name = tensor<string, []>("transpose_2")];
+            tensor<fp16, [1, 1, 640]> input_cast_fp16_0, tensor<fp16, [1, 640]> input_cast_fp16_1, tensor<fp16, [1, 640]> input_cast_fp16_2 = lstm(activation = input_activation_0, bias = concat_0_to_fp16, cell_activation = input_cell_activation_0, direction = input_direction_0, initial_c = input_lstm_c0_squeeze_cast_fp16, initial_h = input_lstm_h0_squeeze_cast_fp16, output_sequence = input_output_sequence_0, recurrent_activation = input_recurrent_activation_0, weight_hh = concat_2_to_fp16, weight_ih = concat_1_to_fp16, x = input_3_cast_fp16)[name = tensor<string, []>("input_cast_fp16")];
+            tensor<int32, [1]> obj_3_axes_0 = const()[name = tensor<string, []>("obj_3_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp16, [1, 1, 640]> obj_3_cast_fp16 = expand_dims(axes = obj_3_axes_0, x = input_cast_fp16_1)[name = tensor<string, []>("obj_3_cast_fp16")];
+            tensor<string, []> obj_3_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("obj_3_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<int32, [1]> obj_axes_0 = const()[name = tensor<string, []>("obj_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp16, [1, 1, 640]> obj_cast_fp16 = expand_dims(axes = obj_axes_0, x = input_cast_fp16_2)[name = tensor<string, []>("obj_cast_fp16")];
+            tensor<string, []> obj_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("obj_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<int32, [3]> transpose_0_perm_0 = const()[name = tensor<string, []>("transpose_0_perm_0"), val = tensor<int32, [3]>([1, 2, 0])];
+            tensor<string, []> transpose_0_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("transpose_0_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<fp16, [1, 640, 1]> transpose_0_cast_fp16 = transpose(perm = transpose_0_perm_0, x = input_cast_fp16_0)[name = tensor<string, []>("transpose_1")];
+            tensor<fp32, [1, 640, 1]> decoder = cast(dtype = transpose_0_cast_fp16_to_fp32_dtype_0, x = transpose_0_cast_fp16)[name = tensor<string, []>("cast_2")];
+            tensor<fp32, [1, 1, 640]> c_out = cast(dtype = obj_cast_fp16_to_fp32_dtype_0, x = obj_cast_fp16)[name = tensor<string, []>("cast_3")];
+            tensor<fp32, [1, 1, 640]> h_out = cast(dtype = obj_3_cast_fp16_to_fp32_dtype_0, x = obj_3_cast_fp16)[name = tensor<string, []>("cast_4")];
+            tensor<int32, [1]> target_length_tmp = identity(x = target_length)[name = tensor<string, []>("target_length_tmp")];
+        } -> (decoder, h_out, c_out);
+}

parakeet_eou_decoder.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b4cacecdcd9df79ab1e56de67230baf5a8664d2afe0bb8f3408eefa972cb2f4
+size 7873600

parakeet_eou_decoder.mlpackage/Data/com.apple.CoreML/model.mlmodel

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f2eaacb49715e293622dccba047d292672022d0d76f2dd30cef1a520b36d93eb
+size 7965

parakeet_eou_decoder.mlpackage/Data/com.apple.CoreML/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b4cacecdcd9df79ab1e56de67230baf5a8664d2afe0bb8f3408eefa972cb2f4
+size 7873600

parakeet_eou_decoder.mlpackage/Manifest.json

@@ -0,0 +1,18 @@
+{
+    "fileFormatVersion": "1.0.0",
+    "itemInfoEntries": {
+        "16235E9F-F8AF-40EA-9B35-6BD5D7D815A4": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Weights",
+            "name": "weights",
+            "path": "com.apple.CoreML/weights"
+        },
+        "D463F5AE-B901-4E3E-8383-71929943FCB8": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Specification",
+            "name": "model.mlmodel",
+            "path": "com.apple.CoreML/model.mlmodel"
+        }
+    },
+    "rootModelIdentifier": "D463F5AE-B901-4E3E-8383-71929943FCB8"
+}

parakeet_eou_encoder.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f1690bb93a34cae20083921a8ce0d494f11158d85b0534f064a676f54c864b4d
+size 243

parakeet_eou_encoder.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e44cbe49df7a1b0e6e2f8457fc1f6775b071ff5fedb8c1d2d9e40c65173381a3
+size 508

parakeet_eou_encoder.mlmodelc/metadata.json

@@ -0,0 +1,116 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "shortDescription" : "Parakeet EOU encoder (15.0s window)",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 512 × 189)",
+        "shortDescription" : "",
+        "shape" : "[1, 512, 189]",
+        "name" : "encoder",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "encoder_length",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 189)",
+        "shortDescription" : "",
+        "shape" : "[189]",
+        "name" : "frame_times",
+        "type" : "MultiArray"
+      }
+    ],
+    "storagePrecision" : "Float16",
+    "modelParameters" : [
+
+    ],
+    "author" : "Fluid Inference",
+    "specificationVersion" : 8,
+    "mlProgramOperationTypeHistogram" : {
+      "Ios17.logicalAnd" : 2,
+      "Ios17.reshape" : 103,
+      "Ios16.softmax" : 17,
+      "Ios17.matmul" : 51,
+      "Ios17.transpose" : 157,
+      "Split" : 17,
+      "Ios17.expandDims" : 17,
+      "Select" : 51,
+      "Ios17.add" : 128,
+      "Tile" : 8,
+      "Ios17.sliceByIndex" : 34,
+      "Ios16.sigmoid" : 17,
+      "Pad" : 37,
+      "Ios17.logicalNot" : 2,
+      "Ios17.layerNorm" : 102,
+      "Ios17.less" : 5,
+      "Ios17.sub" : 3,
+      "Ios16.silu" : 51,
+      "Ios17.conv" : 56,
+      "Ios16.relu" : 3,
+      "Ios17.linear" : 137,
+      "Ios17.cast" : 11,
+      "Ios17.floorDiv" : 3,
+      "Ios17.mul" : 77
+    },
+    "computePrecision" : "Mixed (Float16, Float32, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "14.0",
+      "tvOS" : "17.0",
+      "visionOS" : "1.0",
+      "watchOS" : "10.0",
+      "iOS" : "17.0",
+      "macCatalyst" : "17.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "inputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 128 × 1501)",
+        "shortDescription" : "",
+        "shape" : "[1, 128, 1501]",
+        "name" : "mel",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "mel_length",
+        "type" : "MultiArray"
+      }
+    ],
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2025-11-27",
+      "com.github.apple.coremltools.source" : "torch==2.9.1",
+      "com.github.apple.coremltools.version" : "9.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "generatedClassName" : "parakeet_eou_encoder",
+    "method" : "predict"
+  }
+]

parakeet_eou_encoder.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1dabf1813f088d044c4c0c8a845e0f7435aa35fcae752a521880359c3b724ed2
+size 216730996

parakeet_eou_encoder.mlpackage/Data/com.apple.CoreML/model.mlmodel

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:efd885bd12d20b3227aa4347f6a5f462a68c45ffca3c04b29179d6669e4aff30
+size 489535

parakeet_eou_encoder.mlpackage/Data/com.apple.CoreML/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1dabf1813f088d044c4c0c8a845e0f7435aa35fcae752a521880359c3b724ed2
+size 216730996

parakeet_eou_encoder.mlpackage/Manifest.json

@@ -0,0 +1,18 @@
+{
+    "fileFormatVersion": "1.0.0",
+    "itemInfoEntries": {
+        "97C60A74-0AA4-45C6-93EE-2AACECCB99B9": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Specification",
+            "name": "model.mlmodel",
+            "path": "com.apple.CoreML/model.mlmodel"
+        },
+        "E456E92A-5C50-4543-8CA1-AACB7EEED81E": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Weights",
+            "name": "weights",
+            "path": "com.apple.CoreML/weights"
+        }
+    },
+    "rootModelIdentifier": "97C60A74-0AA4-45C6-93EE-2AACECCB99B9"
+}

parakeet_eou_joint_decision_single_step.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fb134f4700955b507e141ba8399563ffa74756b528cb5c690e8dc11564ddfbe0
+size 243

parakeet_eou_joint_decision_single_step.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8a895494acdf45aaf05f932cb7170d97f307b9755abd06a03d4f4e59fcf3882e
+size 568

parakeet_eou_joint_decision_single_step.mlmodelc/metadata.json

@@ -0,0 +1,113 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "shortDescription" : "Parakeet EOU single-step joint decision (current frame)",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1 × 1 × 1)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 1]",
+        "name" : "token_id",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 1 × 1)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 1]",
+        "name" : "token_prob",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1 × 1 × 1 × 64)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 1, 64]",
+        "name" : "top_k_ids",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 1 × 1 × 64)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 1, 64]",
+        "name" : "top_k_logits",
+        "type" : "MultiArray"
+      }
+    ],
+    "storagePrecision" : "Float16",
+    "modelParameters" : [
+
+    ],
+    "author" : "Fluid Inference",
+    "specificationVersion" : 8,
+    "mlProgramOperationTypeHistogram" : {
+      "Ios17.reduceArgmax" : 1,
+      "Ios17.squeeze" : 1,
+      "Ios17.cast" : 6,
+      "Ios17.linear" : 3,
+      "Ios17.transpose" : 2,
+      "Ios17.add" : 1,
+      "Ios16.relu" : 1,
+      "Ios16.softmax" : 1,
+      "Ios17.gatherAlongAxis" : 1,
+      "Ios17.topk" : 1,
+      "Ios17.expandDims" : 3
+    },
+    "computePrecision" : "Mixed (Float16, Float32, Int16, Int32, UInt16)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "14.0",
+      "tvOS" : "17.0",
+      "visionOS" : "1.0",
+      "watchOS" : "10.0",
+      "iOS" : "17.0",
+      "macCatalyst" : "17.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "inputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 512 × 1)",
+        "shortDescription" : "",
+        "shape" : "[1, 512, 1]",
+        "name" : "encoder_step",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 640 × 1)",
+        "shortDescription" : "",
+        "shape" : "[1, 640, 1]",
+        "name" : "decoder_step",
+        "type" : "MultiArray"
+      }
+    ],
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2025-11-27",
+      "com.github.apple.coremltools.source" : "torch==2.9.1",
+      "com.github.apple.coremltools.version" : "9.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "generatedClassName" : "parakeet_eou_joint_decision_single_step",
+    "method" : "predict"
+  }
+]

parakeet_eou_joint_decision_single_step.mlmodelc/model.mil

@@ -0,0 +1,57 @@
+program(1.0)
+[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3500.14.1"}, {"coremlc-version", "3500.32.1"}, {"coremltools-component-torch", "2.9.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})]
+{
+    func main<ios17>(tensor<fp32, [1, 640, 1]> decoder_step, tensor<fp32, [1, 512, 1]> encoder_step) {
+            tensor<int32, [3]> input_1_perm_0 = const()[name = tensor<string, []>("input_1_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<string, []> encoder_step_to_fp16_dtype_0 = const()[name = tensor<string, []>("encoder_step_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
+            tensor<int32, [3]> input_3_perm_0 = const()[name = tensor<string, []>("input_3_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<string, []> decoder_step_to_fp16_dtype_0 = const()[name = tensor<string, []>("decoder_step_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
+            tensor<fp16, [640, 512]> joint_module_enc_weight_to_fp16 = const()[name = tensor<string, []>("joint_module_enc_weight_to_fp16"), val = tensor<fp16, [640, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
+            tensor<fp16, [640]> joint_module_enc_bias_to_fp16 = const()[name = tensor<string, []>("joint_module_enc_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(655488)))];
+            tensor<fp16, [1, 512, 1]> encoder_step_to_fp16 = cast(dtype = encoder_step_to_fp16_dtype_0, x = encoder_step)[name = tensor<string, []>("cast_5")];
+            tensor<fp16, [1, 1, 512]> input_1_cast_fp16 = transpose(perm = input_1_perm_0, x = encoder_step_to_fp16)[name = tensor<string, []>("transpose_1")];
+            tensor<fp16, [1, 1, 640]> linear_0_cast_fp16 = linear(bias = joint_module_enc_bias_to_fp16, weight = joint_module_enc_weight_to_fp16, x = input_1_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")];
+            tensor<fp16, [640, 640]> joint_module_pred_weight_to_fp16 = const()[name = tensor<string, []>("joint_module_pred_weight_to_fp16"), val = tensor<fp16, [640, 640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(656832)))];
+            tensor<fp16, [640]> joint_module_pred_bias_to_fp16 = const()[name = tensor<string, []>("joint_module_pred_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1476096)))];
+            tensor<fp16, [1, 640, 1]> decoder_step_to_fp16 = cast(dtype = decoder_step_to_fp16_dtype_0, x = decoder_step)[name = tensor<string, []>("cast_4")];
+            tensor<fp16, [1, 1, 640]> input_3_cast_fp16 = transpose(perm = input_3_perm_0, x = decoder_step_to_fp16)[name = tensor<string, []>("transpose_0")];
+            tensor<fp16, [1, 1, 640]> linear_1_cast_fp16 = linear(bias = joint_module_pred_bias_to_fp16, weight = joint_module_pred_weight_to_fp16, x = input_3_cast_fp16)[name = tensor<string, []>("linear_1_cast_fp16")];
+            tensor<int32, [1]> var_23_axes_0 = const()[name = tensor<string, []>("op_23_axes_0"), val = tensor<int32, [1]>([2])];
+            tensor<fp16, [1, 1, 1, 640]> var_23_cast_fp16 = expand_dims(axes = var_23_axes_0, x = linear_0_cast_fp16)[name = tensor<string, []>("op_23_cast_fp16")];
+            tensor<int32, [1]> var_24_axes_0 = const()[name = tensor<string, []>("op_24_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1, 1, 1, 640]> var_24_cast_fp16 = expand_dims(axes = var_24_axes_0, x = linear_1_cast_fp16)[name = tensor<string, []>("op_24_cast_fp16")];
+            tensor<fp16, [1, 1, 1, 640]> input_5_cast_fp16 = add(x = var_23_cast_fp16, y = var_24_cast_fp16)[name = tensor<string, []>("input_5_cast_fp16")];
+            tensor<fp16, [1, 1, 1, 640]> input_7_cast_fp16 = relu(x = input_5_cast_fp16)[name = tensor<string, []>("input_7_cast_fp16")];
+            tensor<fp16, [1027, 640]> joint_module_joint_net_2_weight_to_fp16 = const()[name = tensor<string, []>("joint_module_joint_net_2_weight_to_fp16"), val = tensor<fp16, [1027, 640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1477440)))];
+            tensor<fp16, [1027]> joint_module_joint_net_2_bias_to_fp16 = const()[name = tensor<string, []>("joint_module_joint_net_2_bias_to_fp16"), val = tensor<fp16, [1027]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2792064)))];
+            tensor<fp16, [1, 1, 1, 1027]> linear_2_cast_fp16 = linear(bias = joint_module_joint_net_2_bias_to_fp16, weight = joint_module_joint_net_2_weight_to_fp16, x = input_7_cast_fp16)[name = tensor<string, []>("linear_2_cast_fp16")];
+            tensor<int32, []> var_33_axis_0 = const()[name = tensor<string, []>("op_33_axis_0"), val = tensor<int32, []>(-1)];
+            tensor<bool, []> var_33_keep_dims_0 = const()[name = tensor<string, []>("op_33_keep_dims_0"), val = tensor<bool, []>(false)];
+            tensor<string, []> var_33_output_dtype_0 = const()[name = tensor<string, []>("op_33_output_dtype_0"), val = tensor<string, []>("int32")];
+            tensor<int32, [1, 1, 1]> token_id = reduce_argmax(axis = var_33_axis_0, keep_dims = var_33_keep_dims_0, output_dtype = var_33_output_dtype_0, x = linear_2_cast_fp16)[name = tensor<string, []>("op_33_cast_fp16")];
+            tensor<int32, []> var_39 = const()[name = tensor<string, []>("op_39"), val = tensor<int32, []>(-1)];
+            tensor<fp16, [1, 1, 1, 1027]> token_probs_all_cast_fp16 = softmax(axis = var_39, x = linear_2_cast_fp16)[name = tensor<string, []>("token_probs_all_cast_fp16")];
+            tensor<int32, [1]> var_48_axes_0 = const()[name = tensor<string, []>("op_48_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<int32, [1, 1, 1, 1]> var_48 = expand_dims(axes = var_48_axes_0, x = token_id)[name = tensor<string, []>("op_48")];
+            tensor<int32, []> var_49 = const()[name = tensor<string, []>("op_49"), val = tensor<int32, []>(-1)];
+            tensor<bool, []> var_51_validate_indices_0 = const()[name = tensor<string, []>("op_51_validate_indices_0"), val = tensor<bool, []>(false)];
+            tensor<string, []> var_48_to_int16_dtype_0 = const()[name = tensor<string, []>("op_48_to_int16_dtype_0"), val = tensor<string, []>("int16")];
+            tensor<int16, [1, 1, 1, 1]> var_48_to_int16 = cast(dtype = var_48_to_int16_dtype_0, x = var_48)[name = tensor<string, []>("cast_3")];
+            tensor<fp16, [1, 1, 1, 1]> var_51_cast_fp16_cast_int16 = gather_along_axis(axis = var_49, indices = var_48_to_int16, validate_indices = var_51_validate_indices_0, x = token_probs_all_cast_fp16)[name = tensor<string, []>("op_51_cast_fp16_cast_int16")];
+            tensor<int32, [1]> var_53_axes_0 = const()[name = tensor<string, []>("op_53_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1, 1, 1]> var_53_cast_fp16 = squeeze(axes = var_53_axes_0, x = var_51_cast_fp16_cast_int16)[name = tensor<string, []>("op_53_cast_fp16")];
+            tensor<string, []> var_53_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("op_53_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<int32, []> var_54 = const()[name = tensor<string, []>("op_54"), val = tensor<int32, []>(64)];
+            tensor<int32, []> var_58_axis_0 = const()[name = tensor<string, []>("op_58_axis_0"), val = tensor<int32, []>(-1)];
+            tensor<bool, []> var_58_ascending_0 = const()[name = tensor<string, []>("op_58_ascending_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_58_sort_0 = const()[name = tensor<string, []>("op_58_sort_0"), val = tensor<bool, []>(true)];
+            tensor<bool, []> var_58_return_indices_0 = const()[name = tensor<string, []>("op_58_return_indices_0"), val = tensor<bool, []>(true)];
+            tensor<string, []> var_58_cast_fp16_cast_int16_output_indices_dtype_0 = const()[name = tensor<string, []>("op_58_cast_fp16_cast_int16_output_indices_dtype_0"), val = tensor<string, []>("uint16")];
+            tensor<fp16, [1, 1, 1, 64]> var_58_cast_fp16_cast_int16_0, tensor<uint16, [1, 1, 1, 64]> var_58_cast_fp16_cast_int16_1 = topk(ascending = var_58_ascending_0, axis = var_58_axis_0, k = var_54, output_indices_dtype = var_58_cast_fp16_cast_int16_output_indices_dtype_0, return_indices = var_58_return_indices_0, sort = var_58_sort_0, x = linear_2_cast_fp16)[name = tensor<string, []>("op_58_cast_fp16_cast_int16")];
+            tensor<string, []> var_58_cast_fp16_cast_int16_1_to_int32_dtype_0 = const()[name = tensor<string, []>("op_58_cast_fp16_cast_int16_1_to_int32_dtype_0"), val = tensor<string, []>("int32")];
+            tensor<string, []> var_58_cast_fp16_0_to_fp32_dtype_0 = const()[name = tensor<string, []>("op_58_cast_fp16_0_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<fp32, [1, 1, 1, 64]> top_k_logits = cast(dtype = var_58_cast_fp16_0_to_fp32_dtype_0, x = var_58_cast_fp16_cast_int16_0)[name = tensor<string, []>("cast_0")];
+            tensor<int32, [1, 1, 1, 64]> top_k_ids = cast(dtype = var_58_cast_fp16_cast_int16_1_to_int32_dtype_0, x = var_58_cast_fp16_cast_int16_1)[name = tensor<string, []>("cast_1")];
+            tensor<fp32, [1, 1, 1]> token_prob = cast(dtype = var_53_cast_fp16_to_fp32_dtype_0, x = var_53_cast_fp16)[name = tensor<string, []>("cast_2")];
+        } -> (token_id, token_prob, top_k_ids, top_k_logits);
+}

parakeet_eou_joint_decision_single_step.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7039b2010a269153f5a96edf28637f921a86ef8822f248f2d6712f7a6bce84b4
+size 2794182

parakeet_eou_joint_decision_single_step.mlpackage/Data/com.apple.CoreML/model.mlmodel

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7975f5a9b707b5aa7307121dc73defe7772a83db89e42135d4146ec921252be8
+size 8758

parakeet_eou_joint_decision_single_step.mlpackage/Data/com.apple.CoreML/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7039b2010a269153f5a96edf28637f921a86ef8822f248f2d6712f7a6bce84b4
+size 2794182

parakeet_eou_joint_decision_single_step.mlpackage/Manifest.json

@@ -0,0 +1,18 @@
+{
+    "fileFormatVersion": "1.0.0",
+    "itemInfoEntries": {
+        "661CE870-1A41-4769-9A11-E8D34CB67A36": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Weights",
+            "name": "weights",
+            "path": "com.apple.CoreML/weights"
+        },
+        "BD55C909-51AF-42F7-8CE5-B7C05B80A700": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Specification",
+            "name": "model.mlmodel",
+            "path": "com.apple.CoreML/model.mlmodel"
+        }
+    },
+    "rootModelIdentifier": "BD55C909-51AF-42F7-8CE5-B7C05B80A700"
+}

parakeet_eou_preprocessor.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d5a335c7c3414656bca7c4d47e9718a4617665516546f734525eb7359fe0703
+size 243

parakeet_eou_preprocessor.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4f97a8d4d3d1d4f2410e8382f3445d92551d55a62664fe6509c4d73d0c5a4ff3
+size 496

parakeet_eou_preprocessor.mlmodelc/metadata.json

@@ -0,0 +1,109 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "shortDescription" : "Parakeet EOU preprocessor (15.0s window)",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32)",
+        "shortDescription" : "",
+        "shape" : "[]",
+        "name" : "mel",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "mel_length",
+        "type" : "MultiArray"
+      }
+    ],
+    "storagePrecision" : "Float16",
+    "modelParameters" : [
+
+    ],
+    "author" : "Fluid Inference",
+    "specificationVersion" : 8,
+    "mlProgramOperationTypeHistogram" : {
+      "Range1d" : 2,
+      "Ios17.equal" : 1,
+      "Ios17.reshape" : 2,
+      "Identity" : 1,
+      "Ios17.matmul" : 1,
+      "Select" : 3,
+      "Ios17.expandDims" : 7,
+      "Ios17.add" : 2,
+      "Ios17.sliceByIndex" : 3,
+      "Ios16.reduceSum" : 1,
+      "Shape" : 2,
+      "Ios17.gather" : 2,
+      "Ios17.logicalNot" : 1,
+      "Pad" : 1,
+      "Ios17.log" : 1,
+      "Ios17.less" : 1,
+      "Ios17.sub" : 2,
+      "Ios17.conv" : 2,
+      "Ios17.pow" : 1,
+      "Ios17.cast" : 6,
+      "Ios17.concat" : 1,
+      "Stack" : 1,
+      "Ios17.floorDiv" : 1,
+      "Ios17.greaterEqual" : 1,
+      "Ios17.mul" : 1
+    },
+    "computePrecision" : "Mixed (Float16, Float32, Int16, Int32, UInt16)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "14.0",
+      "tvOS" : "17.0",
+      "visionOS" : "1.0",
+      "watchOS" : "10.0",
+      "iOS" : "17.0",
+      "macCatalyst" : "17.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "inputSchema" : [
+      {
+        "dataType" : "Float32",
+        "hasShapeFlexibility" : "1",
+        "isOptional" : "0",
+        "shapeFlexibility" : "1 × 1...240000",
+        "shapeRange" : "[[1, 1], [1, 240000]]",
+        "formattedType" : "MultiArray (Float32 1 × 1)",
+        "type" : "MultiArray",
+        "shape" : "[1, 1]",
+        "name" : "audio_signal",
+        "shortDescription" : ""
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "audio_length",
+        "type" : "MultiArray"
+      }
+    ],
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2025-11-27",
+      "com.github.apple.coremltools.source" : "torch==2.9.1",
+      "com.github.apple.coremltools.version" : "9.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "generatedClassName" : "parakeet_eou_preprocessor",
+    "method" : "predict"
+  }
+]

parakeet_eou_preprocessor.mlmodelc/model.mil

@@ -0,0 +1,123 @@
+program(1.0)
+[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3500.14.1"}, {"coremlc-version", "3500.32.1"}, {"coremltools-component-torch", "2.9.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})]
+{
+    func main<ios17>(tensor<int32, [1]> audio_length, tensor<fp32, [1, ?]> audio_signal) [FlexibleShapeInformation = tuple<tuple<tensor<string, []>, dict<tensor<string, []>, tensor<int32, [?]>>>, tuple<tensor<string, []>, dict<tensor<string, []>, list<tensor<int32, [2]>, ?>>>>((("DefaultShapes", {{"audio_signal", [1, 1]}}), ("RangeDims", {{"audio_signal", [[1, 1], [1, 240000]]}})))] {
+            tensor<int32, []> var_9 = const()[name = tensor<string, []>("op_9"), val = tensor<int32, []>(1)];
+            tensor<int32, []> var_10 = const()[name = tensor<string, []>("op_10"), val = tensor<int32, []>(160)];
+            tensor<int32, []> var_12 = const()[name = tensor<string, []>("op_12"), val = tensor<int32, []>(0)];
+            tensor<int32, []> var_33 = const()[name = tensor<string, []>("op_33"), val = tensor<int32, []>(512)];
+            tensor<int32, [1]> var_34 = add(x = audio_length, y = var_33)[name = tensor<string, []>("op_34")];
+            tensor<int32, []> var_35 = const()[name = tensor<string, []>("op_35"), val = tensor<int32, []>(512)];
+            tensor<int32, [1]> var_36 = sub(x = var_34, y = var_35)[name = tensor<string, []>("op_36")];
+            tensor<int32, [1]> floor_div_0 = floor_div(x = var_36, y = var_10)[name = tensor<string, []>("floor_div_0")];
+            tensor<bool, [1]> var_39 = equal(x = audio_length, y = var_12)[name = tensor<string, []>("op_39")];
+            tensor<int32, [1]> var_40 = const()[name = tensor<string, []>("op_40"), val = tensor<int32, [1]>([0])];
+            tensor<int32, [1]> mel_length = select(a = var_40, b = floor_div_0, cond = var_39)[name = tensor<string, []>("seq_len")];
+            tensor<string, []> audio_signal_to_fp16_dtype_0 = const()[name = tensor<string, []>("audio_signal_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
+            tensor<fp16, [1, ?]> audio_signal_to_fp16 = cast(dtype = audio_signal_to_fp16_dtype_0, x = audio_signal)[name = tensor<string, []>("cast_11")];
+            tensor<int32, [2]> var_42_shape_cast_fp16 = shape(x = audio_signal_to_fp16)[name = tensor<string, []>("op_42_shape_cast_fp16")];
+            tensor<int32, []> gather_0_axis_0 = const()[name = tensor<string, []>("gather_0_axis_0"), val = tensor<int32, []>(0)];
+            tensor<int32, []> gather_0_batch_dims_0 = const()[name = tensor<string, []>("gather_0_batch_dims_0"), val = tensor<int32, []>(0)];
+            tensor<bool, []> gather_0_validate_indices_0 = const()[name = tensor<string, []>("gather_0_validate_indices_0"), val = tensor<bool, []>(false)];
+            tensor<string, []> var_42_shape_cast_fp16_to_int16_dtype_0 = const()[name = tensor<string, []>("op_42_shape_cast_fp16_to_int16_dtype_0"), val = tensor<string, []>("int16")];
+            tensor<uint16, []> gather_0_indices_0_to_uint16 = const()[name = tensor<string, []>("gather_0_indices_0_to_uint16"), val = tensor<uint16, []>(1)];
+            tensor<int16, [2]> var_42_shape_cast_fp16_to_int16 = cast(dtype = var_42_shape_cast_fp16_to_int16_dtype_0, x = var_42_shape_cast_fp16)[name = tensor<string, []>("cast_10")];
+            tensor<int16, []> gather_0_cast_uint16 = gather(axis = gather_0_axis_0, batch_dims = gather_0_batch_dims_0, indices = gather_0_indices_0_to_uint16, validate_indices = gather_0_validate_indices_0, x = var_42_shape_cast_fp16_to_int16)[name = tensor<string, []>("gather_0_cast_uint16")];
+            tensor<string, []> gather_0_cast_uint16_to_int32_dtype_0 = const()[name = tensor<string, []>("gather_0_cast_uint16_to_int32_dtype_0"), val = tensor<string, []>("int32")];
+            tensor<int32, []> const_0 = const()[name = tensor<string, []>("const_0"), val = tensor<int32, []>(0)];
+            tensor<int32, []> const_1 = const()[name = tensor<string, []>("const_1"), val = tensor<int32, []>(1)];
+            tensor<int32, []> gather_0_cast_uint16_to_int32 = cast(dtype = gather_0_cast_uint16_to_int32_dtype_0, x = gather_0_cast_uint16)[name = tensor<string, []>("cast_9")];
+            tensor<int32, [?]> var_43 = range_1d(end = gather_0_cast_uint16_to_int32, start = const_0, step = const_1)[name = tensor<string, []>("op_43")];
+            tensor<int32, [1]> var_44_axes_0 = const()[name = tensor<string, []>("op_44_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<int32, [1, ?]> var_44 = expand_dims(axes = var_44_axes_0, x = var_43)[name = tensor<string, []>("op_44")];
+            tensor<int32, [1]> var_45_axes_0 = const()[name = tensor<string, []>("op_45_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1, 1]> var_45 = expand_dims(axes = var_45_axes_0, x = audio_length)[name = tensor<string, []>("op_45")];
+            tensor<bool, [1, ?]> timemask = less(x = var_44, y = var_45)[name = tensor<string, []>("timemask")];
+            tensor<int32, [2]> var_48_begin_0 = const()[name = tensor<string, []>("op_48_begin_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [2]> var_48_end_0 = const()[name = tensor<string, []>("op_48_end_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<bool, [2]> var_48_end_mask_0 = const()[name = tensor<string, []>("op_48_end_mask_0"), val = tensor<bool, [2]>([true, false])];
+            tensor<bool, [2]> var_48_squeeze_mask_0 = const()[name = tensor<string, []>("op_48_squeeze_mask_0"), val = tensor<bool, [2]>([false, true])];
+            tensor<fp16, [1]> var_48_cast_fp16 = slice_by_index(begin = var_48_begin_0, end = var_48_end_0, end_mask = var_48_end_mask_0, squeeze_mask = var_48_squeeze_mask_0, x = audio_signal_to_fp16)[name = tensor<string, []>("op_48_cast_fp16")];
+            tensor<int32, [1]> var_49_axes_0 = const()[name = tensor<string, []>("op_49_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1, 1]> var_49_cast_fp16 = expand_dims(axes = var_49_axes_0, x = var_48_cast_fp16)[name = tensor<string, []>("op_49_cast_fp16")];
+            tensor<int32, [2]> var_51_begin_0 = const()[name = tensor<string, []>("op_51_begin_0"), val = tensor<int32, [2]>([0, 1])];
+            tensor<int32, [2]> var_51_end_0 = const()[name = tensor<string, []>("op_51_end_0"), val = tensor<int32, [2]>([1, 0])];
+            tensor<bool, [2]> var_51_end_mask_0 = const()[name = tensor<string, []>("op_51_end_mask_0"), val = tensor<bool, [2]>([true, true])];
+            tensor<fp16, [1, ?]> var_51_cast_fp16 = slice_by_index(begin = var_51_begin_0, end = var_51_end_0, end_mask = var_51_end_mask_0, x = audio_signal_to_fp16)[name = tensor<string, []>("op_51_cast_fp16")];
+            tensor<int32, [2]> var_53_begin_0 = const()[name = tensor<string, []>("op_53_begin_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [2]> var_53_end_0 = const()[name = tensor<string, []>("op_53_end_0"), val = tensor<int32, [2]>([1, -1])];
+            tensor<bool, [2]> var_53_end_mask_0 = const()[name = tensor<string, []>("op_53_end_mask_0"), val = tensor<bool, [2]>([true, false])];
+            tensor<fp16, [1, ?]> var_53_cast_fp16 = slice_by_index(begin = var_53_begin_0, end = var_53_end_0, end_mask = var_53_end_mask_0, x = audio_signal_to_fp16)[name = tensor<string, []>("op_53_cast_fp16")];
+            tensor<fp16, []> var_54_to_fp16 = const()[name = tensor<string, []>("op_54_to_fp16"), val = tensor<fp16, []>(0x1.f0cp-1)];
+            tensor<fp16, [1, ?]> var_55_cast_fp16 = mul(x = var_53_cast_fp16, y = var_54_to_fp16)[name = tensor<string, []>("op_55_cast_fp16")];
+            tensor<fp16, [1, ?]> var_56_cast_fp16 = sub(x = var_51_cast_fp16, y = var_55_cast_fp16)[name = tensor<string, []>("op_56_cast_fp16")];
+            tensor<bool, []> x_3_interleave_0 = const()[name = tensor<string, []>("x_3_interleave_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, ?]> x_3_cast_fp16 = concat(axis = var_9, interleave = x_3_interleave_0, values = (var_49_cast_fp16, var_56_cast_fp16))[name = tensor<string, []>("x_3_cast_fp16")];
+            tensor<bool, [1, ?]> var_59 = logical_not(x = timemask)[name = tensor<string, []>("op_59")];
+            tensor<fp16, []> var_16_to_fp16 = const()[name = tensor<string, []>("op_16_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, ?]> input_1_cast_fp16 = select(a = var_16_to_fp16, b = x_3_cast_fp16, cond = var_59)[name = tensor<string, []>("input_1_cast_fp16")];
+            tensor<int32, [3]> concat_1x = const()[name = tensor<string, []>("concat_1x"), val = tensor<int32, [3]>([1, 1, -1])];
+            tensor<fp16, [1, 1, ?]> input_3_cast_fp16 = reshape(shape = concat_1x, x = input_1_cast_fp16)[name = tensor<string, []>("input_3_cast_fp16")];
+            tensor<int32, [6]> input_5_pad_0 = const()[name = tensor<string, []>("input_5_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 256, 256])];
+            tensor<string, []> input_5_mode_0 = const()[name = tensor<string, []>("input_5_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_3_to_fp16 = const()[name = tensor<string, []>("const_3_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1, ?]> input_5_cast_fp16 = pad(constant_val = const_3_to_fp16, mode = input_5_mode_0, pad = input_5_pad_0, x = input_3_cast_fp16)[name = tensor<string, []>("input_5_cast_fp16")];
+            tensor<int32, [2]> concat_2x = const()[name = tensor<string, []>("concat_2x"), val = tensor<int32, [2]>([1, -1])];
+            tensor<fp16, [1, ?]> input_cast_fp16 = reshape(shape = concat_2x, x = input_5_cast_fp16)[name = tensor<string, []>("input_cast_fp16")];
+            tensor<int32, [1]> expand_dims_3 = const()[name = tensor<string, []>("expand_dims_3"), val = tensor<int32, [1]>([160])];
+            tensor<int32, [1]> expand_dims_4_axes_0 = const()[name = tensor<string, []>("expand_dims_4_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1, 1, ?]> expand_dims_4_cast_fp16 = expand_dims(axes = expand_dims_4_axes_0, x = input_cast_fp16)[name = tensor<string, []>("expand_dims_4_cast_fp16")];
+            tensor<string, []> conv_0_pad_type_0 = const()[name = tensor<string, []>("conv_0_pad_type_0"), val = tensor<string, []>("valid")];
+            tensor<int32, [2]> conv_0_pad_0 = const()[name = tensor<string, []>("conv_0_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> conv_0_dilations_0 = const()[name = tensor<string, []>("conv_0_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> conv_0_groups_0 = const()[name = tensor<string, []>("conv_0_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [257, 1, 512]> expand_dims_1_to_fp16 = const()[name = tensor<string, []>("expand_dims_1_to_fp16"), val = tensor<fp16, [257, 1, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
+            tensor<fp16, [1, 257, ?]> conv_0_cast_fp16 = conv(dilations = conv_0_dilations_0, groups = conv_0_groups_0, pad = conv_0_pad_0, pad_type = conv_0_pad_type_0, strides = expand_dims_3, weight = expand_dims_1_to_fp16, x = expand_dims_4_cast_fp16)[name = tensor<string, []>("conv_0_cast_fp16")];
+            tensor<string, []> conv_1_pad_type_0 = const()[name = tensor<string, []>("conv_1_pad_type_0"), val = tensor<string, []>("valid")];
+            tensor<int32, [2]> conv_1_pad_0 = const()[name = tensor<string, []>("conv_1_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> conv_1_dilations_0 = const()[name = tensor<string, []>("conv_1_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> conv_1_groups_0 = const()[name = tensor<string, []>("conv_1_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [257, 1, 512]> expand_dims_2_to_fp16 = const()[name = tensor<string, []>("expand_dims_2_to_fp16"), val = tensor<fp16, [257, 1, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(263296)))];
+            tensor<fp16, [1, 257, ?]> conv_1_cast_fp16 = conv(dilations = conv_1_dilations_0, groups = conv_1_groups_0, pad = conv_1_pad_0, pad_type = conv_1_pad_type_0, strides = expand_dims_3, weight = expand_dims_2_to_fp16, x = expand_dims_4_cast_fp16)[name = tensor<string, []>("conv_1_cast_fp16")];
+            tensor<int32, []> stack_0_axis_0 = const()[name = tensor<string, []>("stack_0_axis_0"), val = tensor<int32, []>(-1)];
+            tensor<fp16, [1, 257, ?, 2]> stack_0_cast_fp16 = stack(axis = stack_0_axis_0, values = (conv_0_cast_fp16, conv_1_cast_fp16))[name = tensor<string, []>("stack_0_cast_fp16")];
+            tensor<fp16, []> var_19_promoted_to_fp16 = const()[name = tensor<string, []>("op_19_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
+            tensor<fp16, [1, 257, ?, 2]> var_74_cast_fp16 = pow(x = stack_0_cast_fp16, y = var_19_promoted_to_fp16)[name = tensor<string, []>("op_74_cast_fp16")];
+            tensor<int32, [1]> var_76_axes_0 = const()[name = tensor<string, []>("op_76_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<bool, []> var_76_keep_dims_0 = const()[name = tensor<string, []>("op_76_keep_dims_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 257, ?]> var_76_cast_fp16 = reduce_sum(axes = var_76_axes_0, keep_dims = var_76_keep_dims_0, x = var_74_cast_fp16)[name = tensor<string, []>("op_76_cast_fp16")];
+            tensor<fp16, [1, 257, ?]> x_11_cast_fp16 = identity(x = var_76_cast_fp16)[name = tensor<string, []>("x_11_cast_fp16")];
+            tensor<bool, []> x_13_transpose_x_0 = const()[name = tensor<string, []>("x_13_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_13_transpose_y_0 = const()[name = tensor<string, []>("x_13_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 128, 257]> const_4_to_fp16 = const()[name = tensor<string, []>("const_4_to_fp16"), val = tensor<fp16, [1, 128, 257]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(526528)))];
+            tensor<fp16, [1, 128, ?]> x_13_cast_fp16 = matmul(transpose_x = x_13_transpose_x_0, transpose_y = x_13_transpose_y_0, x = const_4_to_fp16, y = x_11_cast_fp16)[name = tensor<string, []>("x_13_cast_fp16")];
+            tensor<fp16, []> var_83_to_fp16 = const()[name = tensor<string, []>("op_83_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
+            tensor<fp16, [1, 128, ?]> var_84_cast_fp16 = add(x = x_13_cast_fp16, y = var_83_to_fp16)[name = tensor<string, []>("op_84_cast_fp16")];
+            tensor<fp32, []> x_epsilon_0 = const()[name = tensor<string, []>("x_epsilon_0"), val = tensor<fp32, []>(0x1p-149)];
+            tensor<fp16, [1, 128, ?]> x_cast_fp16 = log(epsilon = x_epsilon_0, x = var_84_cast_fp16)[name = tensor<string, []>("x_cast_fp16")];
+            tensor<int32, [3]> var_86_shape_cast_fp16 = shape(x = x_cast_fp16)[name = tensor<string, []>("op_86_shape_cast_fp16")];
+            tensor<int32, []> gather_5_batch_dims_0 = const()[name = tensor<string, []>("gather_5_batch_dims_0"), val = tensor<int32, []>(0)];
+            tensor<bool, []> gather_5_validate_indices_0 = const()[name = tensor<string, []>("gather_5_validate_indices_0"), val = tensor<bool, []>(false)];
+            tensor<string, []> var_86_shape_cast_fp16_to_uint16_dtype_0 = const()[name = tensor<string, []>("op_86_shape_cast_fp16_to_uint16_dtype_0"), val = tensor<string, []>("uint16")];
+            tensor<int32, []> gather_5_cast_uint16_axis_0 = const()[name = tensor<string, []>("gather_5_cast_uint16_axis_0"), val = tensor<int32, []>(0)];
+            tensor<uint16, []> select_0_to_uint16 = const()[name = tensor<string, []>("select_0_to_uint16"), val = tensor<uint16, []>(2)];
+            tensor<uint16, [3]> var_86_shape_cast_fp16_to_uint16 = cast(dtype = var_86_shape_cast_fp16_to_uint16_dtype_0, x = var_86_shape_cast_fp16)[name = tensor<string, []>("cast_8")];
+            tensor<uint16, []> gather_5_cast_uint16_cast_uint16 = gather(axis = gather_5_cast_uint16_axis_0, batch_dims = gather_5_batch_dims_0, indices = select_0_to_uint16, validate_indices = gather_5_validate_indices_0, x = var_86_shape_cast_fp16_to_uint16)[name = tensor<string, []>("gather_5_cast_uint16_cast_uint16")];
+            tensor<string, []> gather_5_cast_uint16_to_int32_dtype_0 = const()[name = tensor<string, []>("gather_5_cast_uint16_to_int32_dtype_0"), val = tensor<string, []>("int32")];
+            tensor<int32, []> const_5 = const()[name = tensor<string, []>("const_5"), val = tensor<int32, []>(0)];
+            tensor<int32, []> const_6 = const()[name = tensor<string, []>("const_6"), val = tensor<int32, []>(1)];
+            tensor<int32, []> gather_5_cast_uint16_to_int32 = cast(dtype = gather_5_cast_uint16_to_int32_dtype_0, x = gather_5_cast_uint16_cast_uint16)[name = tensor<string, []>("cast_7")];
+            tensor<int32, [?]> mask_1 = range_1d(end = gather_5_cast_uint16_to_int32, start = const_5, step = const_6)[name = tensor<string, []>("mask_1")];
+            tensor<int32, [1]> expand_dims_0_axes_0 = const()[name = tensor<string, []>("expand_dims_0_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<int32, [1, ?]> expand_dims_0 = expand_dims(axes = expand_dims_0_axes_0, x = mask_1)[name = tensor<string, []>("expand_dims_0")];
+            tensor<int32, [1]> var_91_axes_0 = const()[name = tensor<string, []>("op_91_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1, 1]> var_91 = expand_dims(axes = var_91_axes_0, x = mel_length)[name = tensor<string, []>("op_91")];
+            tensor<bool, [1, ?]> mask = greater_equal(x = expand_dims_0, y = var_91)[name = tensor<string, []>("mask")];
+            tensor<int32, [1]> var_93_axes_0 = const()[name = tensor<string, []>("op_93_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<bool, [1, 1, ?]> var_93 = expand_dims(axes = var_93_axes_0, x = mask)[name = tensor<string, []>("op_93")];
+            tensor<fp16, []> cast_2_to_fp16 = const()[name = tensor<string, []>("cast_2_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 128, ?]> processed_signal_cast_fp16 = select(a = cast_2_to_fp16, b = x_cast_fp16, cond = var_93)[name = tensor<string, []>("processed_signal_cast_fp16")];
+            tensor<string, []> processed_signal_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("processed_signal_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<fp32, [1, 128, ?]> mel = cast(dtype = processed_signal_cast_fp16_to_fp32_dtype_0, x = processed_signal_cast_fp16)[name = tensor<string, []>("cast_6")];
+        } -> (mel, mel_length);
+}

parakeet_eou_preprocessor.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f257ad1ac11575d73a6ffda555319b2c96b0a224f0dc03ddd8c62950e9b18e53
+size 592384

parakeet_eou_preprocessor.mlpackage/Data/com.apple.CoreML/model.mlmodel

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d7f5e54cf5aba93ba74b293ec05dd5bf5d84f59441fbabb3d9dc7de3666effc2
+size 16267

parakeet_eou_preprocessor.mlpackage/Data/com.apple.CoreML/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f257ad1ac11575d73a6ffda555319b2c96b0a224f0dc03ddd8c62950e9b18e53
+size 592384

parakeet_eou_preprocessor.mlpackage/Manifest.json

@@ -0,0 +1,18 @@
+{
+    "fileFormatVersion": "1.0.0",
+    "itemInfoEntries": {
+        "1D2D0892-1547-4727-AF4E-DBC70A712151": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Weights",
+            "name": "weights",
+            "path": "com.apple.CoreML/weights"
+        },
+        "DD2AE790-0DE3-468D-B5D0-ED1BEEACA7C6": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Specification",
+            "name": "model.mlmodel",
+            "path": "com.apple.CoreML/model.mlmodel"
+        }
+    },
+    "rootModelIdentifier": "DD2AE790-0DE3-468D-B5D0-ED1BEEACA7C6"
+}

vocab.json

@@ -0,0 +1,1037 @@
+{
+  "vocab": [
+    "<unk>",
+    "▁t",
+    "▁th",
+    "▁a",
+    "▁i",
+    "▁the",
+    "▁s",
+    "re",
+    "▁w",
+    "▁o",
+    "in",
+    "at",
+    "er",
+    "nd",
+    "ou",
+    "▁c",
+    "▁b",
+    "▁h",
+    "en",
+    "on",
+    "▁m",
+    "▁f",
+    "ing",
+    "▁p",
+    "▁to",
+    "▁and",
+    "▁d",
+    "an",
+    "or",
+    "es",
+    "▁y",
+    "▁l",
+    "▁of",
+    "ll",
+    "▁in",
+    "ed",
+    "it",
+    "▁g",
+    "is",
+    "▁you",
+    "▁n",
+    "ar",
+    "om",
+    "as",
+    "ve",
+    "▁e",
+    "ic",
+    "▁it",
+    "al",
+    "us",
+    "▁wh",
+    "▁we",
+    "▁be",
+    "ion",
+    "ow",
+    "le",
+    "▁is",
+    "et",
+    "ent",
+    "ot",
+    "ut",
+    "▁re",
+    "▁on",
+    "ay",
+    "▁ha",
+    "ig",
+    "▁so",
+    "ct",
+    "▁he",
+    "▁for",
+    "ver",
+    "ke",
+    "ro",
+    "▁st",
+    "id",
+    "▁go",
+    "all",
+    "se",
+    "ly",
+    "▁u",
+    "ch",
+    "st",
+    "ld",
+    "▁k",
+    "ce",
+    "ur",
+    "▁li",
+    "am",
+    "▁r",
+    "ht",
+    "▁j",
+    "ith",
+    "▁se",
+    "ir",
+    "▁as",
+    "▁an",
+    "im",
+    "▁do",
+    "ad",
+    "▁was",
+    "ight",
+    "th",
+    "▁are",
+    "▁but",
+    "▁sh",
+    "ust",
+    "ally",
+    "▁not",
+    "▁or",
+    "▁com",
+    "▁can",
+    "▁me",
+    "op",
+    "▁mo",
+    "▁at",
+    "ill",
+    "▁ch",
+    "▁ne",
+    "ant",
+    "▁de",
+    "▁kn",
+    "▁one",
+    "il",
+    "ol",
+    "▁con",
+    "ter",
+    "▁ab",
+    "▁fr",
+    "ere",
+    "ck",
+    "▁al",
+    "▁all",
+    "qu",
+    "▁pro",
+    "▁som",
+    "ould",
+    "▁tw",
+    "ul",
+    "ra",
+    "od",
+    "ers",
+    "▁su",
+    "ive",
+    "▁v",
+    "use",
+    "ate",
+    "ge",
+    "if",
+    "▁ex",
+    "ess",
+    "pp",
+    "▁lo",
+    "out",
+    "▁if",
+    "est",
+    "ain",
+    "ist",
+    "and",
+    "ea",
+    "very",
+    "art",
+    "▁wor",
+    "▁my",
+    "ab",
+    "ment",
+    "▁bec",
+    "un",
+    "ity",
+    "ri",
+    "pe",
+    "ions",
+    "▁by",
+    "ok",
+    "our",
+    "ort",
+    "ind",
+    "ink",
+    "nt",
+    "▁up",
+    "um",
+    "▁don",
+    "▁get",
+    "red",
+    "▁out",
+    "el",
+    "ause",
+    "res",
+    "▁ma",
+    "ich",
+    "▁us",
+    "rou",
+    "▁int",
+    "em",
+    "os",
+    "ies",
+    "ie",
+    "▁pl",
+    "▁tr",
+    "ven",
+    "ous",
+    "▁le",
+    "▁two",
+    "ard",
+    "ine",
+    "▁co",
+    "een",
+    "▁now",
+    "ty",
+    "her",
+    "ack",
+    "▁pe",
+    "ame",
+    "▁how",
+    "▁who",
+    "▁see",
+    "▁tim",
+    "ect",
+    "ast",
+    "▁our",
+    "ci",
+    "ree",
+    "ople",
+    "gh",
+    "▁no",
+    "▁had",
+    "▁man",
+    "▁qu",
+    "▁en",
+    "ide",
+    "ure",
+    "ud",
+    "so",
+    "▁his",
+    "▁sa",
+    "▁sp",
+    "▁say",
+    "ose",
+    "ther",
+    "▁act",
+    "▁ta",
+    "▁cl",
+    "ings",
+    "pt",
+    "king",
+    "▁any",
+    "▁has",
+    "▁un",
+    "iv",
+    "▁im",
+    "▁ag",
+    "▁te",
+    "▁fe",
+    "one",
+    "per",
+    "ong",
+    "▁po",
+    "▁ad",
+    "ff",
+    "ore",
+    "itt",
+    "ans",
+    "iz",
+    "eah",
+    "reat",
+    "act",
+    "own",
+    "hing",
+    "enty",
+    "age",
+    "ber",
+    "ice",
+    "▁am",
+    "ple",
+    "are",
+    "▁per",
+    "und",
+    "ite",
+    "ix",
+    "pl",
+    "▁way",
+    "▁did",
+    "▁pr",
+    "▁got",
+    "ars",
+    "▁she",
+    "▁let",
+    "ag",
+    "▁ac",
+    "int",
+    "▁ar",
+    "ry",
+    "ign",
+    "ish",
+    "▁fir",
+    "ace",
+    "ble",
+    "og",
+    "ue",
+    "▁ye",
+    "ap",
+    "iff",
+    "▁ro",
+    "▁her",
+    "nder",
+    "▁ok",
+    "▁res",
+    "▁gu",
+    "ence",
+    "▁may",
+    "ated",
+    "ip",
+    "▁bo",
+    "▁him",
+    "way",
+    "ac",
+    "ical",
+    "ass",
+    "ase",
+    "▁dis",
+    "able",
+    "ick",
+    "▁app",
+    "ance",
+    "▁pre",
+    "▁six",
+    "▁off",
+    "▁new",
+    "ia",
+    "orm",
+    "ank",
+    "▁lot",
+    "ach",
+    "▁fo",
+    "inet",
+    "ire",
+    "ary",
+    "ult",
+    "▁tal",
+    "▁mu",
+    "▁bl",
+    "ount",
+    "sel",
+    "vel",
+    "▁br",
+    "▁imp",
+    "ep",
+    "cess",
+    "ord",
+    "▁sc",
+    "▁inc",
+    "ound",
+    "ang",
+    "be",
+    "ress",
+    "uct",
+    "▁ind",
+    "▁af",
+    "ving",
+    "▁oh",
+    "▁bet",
+    "▁use",
+    "ome",
+    "ens",
+    "ys",
+    "▁bu",
+    "co",
+    "ory",
+    "ater",
+    "ild",
+    "ght",
+    "ial",
+    "▁day",
+    "ning",
+    "na",
+    "ile",
+    "▁spe",
+    "▁mar",
+    "ody",
+    "ough",
+    "ade",
+    "vers",
+    "xt",
+    "▁fl",
+    "▁ke",
+    "ian",
+    "▁sy",
+    "▁put",
+    "fore",
+    "ub",
+    "▁ph",
+    "fe",
+    "▁em",
+    "▁ser",
+    "form",
+    "ting",
+    "te",
+    "av",
+    "ious",
+    "▁rec",
+    "ks",
+    "▁gr",
+    "ces",
+    "wn",
+    "ors",
+    "▁jo",
+    "ents",
+    "▁des",
+    "▁try",
+    "▁equ",
+    "▁z",
+    "▁rem",
+    "▁str",
+    "self",
+    "▁bit",
+    "ph",
+    "ved",
+    "▁why",
+    "▁bas",
+    "▁hel",
+    "▁rel",
+    "ath",
+    "ject",
+    "ail",
+    "▁la",
+    "ual",
+    "▁god",
+    "▁nat",
+    "erm",
+    "day",
+    "▁id",
+    "ft",
+    "▁wr",
+    "▁min",
+    "ates",
+    "▁gen",
+    "tain",
+    "▁ob",
+    "ull",
+    "ict",
+    "▁tra",
+    "▁end",
+    "▁hig",
+    "▁fif",
+    "oth",
+    "tern",
+    "▁its",
+    "vent",
+    "▁sm",
+    "ons",
+    "▁add",
+    "iss",
+    "▁bel",
+    "ful",
+    "get",
+    "▁ele",
+    "▁rep",
+    "ak",
+    "▁ho",
+    "▁pos",
+    "▁num",
+    "ange",
+    "ves",
+    "ific",
+    "urn",
+    "ise",
+    "▁cr",
+    "▁um",
+    "ward",
+    "▁reg",
+    "ady",
+    "ower",
+    "uc",
+    "▁dec",
+    "lic",
+    "▁set",
+    "▁gon",
+    "▁op",
+    "▁ear",
+    "▁sub",
+    "▁sl",
+    "les",
+    "stem",
+    "cial",
+    "olog",
+    "atch",
+    "ily",
+    "body",
+    "nds",
+    "ular",
+    "ren",
+    "▁own",
+    "▁too",
+    "cent",
+    "ible",
+    "pect",
+    "ered",
+    "ways",
+    "teen",
+    "▁uh",
+    "▁big",
+    "▁mod",
+    "▁att",
+    "▁car",
+    "gr",
+    "▁acc",
+    "ied",
+    "mun",
+    "ib",
+    "▁mon",
+    "▁sch",
+    "▁pol",
+    "▁dat",
+    "▁fin",
+    "▁sim",
+    "▁inv",
+    "▁def",
+    "ked",
+    "▁ent",
+    "▁yes",
+    "ows",
+    "ics",
+    "ited",
+    "ute",
+    "ism",
+    "ps",
+    "▁ed",
+    "▁el",
+    "ably",
+    "ppen",
+    "als",
+    "▁ten",
+    "ract",
+    "ss",
+    "▁ass",
+    "▁met",
+    "gan",
+    "▁eng",
+    "▁stu",
+    "ween",
+    "arch",
+    "▁gl",
+    "▁cor",
+    "▁dr",
+    "vern",
+    "▁ty",
+    "▁run",
+    "hip",
+    "cus",
+    "cond",
+    "▁ins",
+    "irty",
+    "▁pub",
+    "lud",
+    "llow",
+    "▁cou",
+    "ew",
+    "iew",
+    "▁sur",
+    "ero",
+    "ood",
+    "ness",
+    "▁fun",
+    "▁eff",
+    "cept",
+    "▁ca",
+    "▁exp",
+    "duct",
+    "▁sw",
+    "ize",
+    "ope",
+    "▁par",
+    "kes",
+    "cy",
+    "▁ev",
+    "▁ref",
+    "ell",
+    "▁bus",
+    "ug",
+    "rib",
+    "▁cur",
+    "mo",
+    "ock",
+    "ures",
+    "air",
+    "▁war",
+    "str",
+    "▁med",
+    "▁wa",
+    "▁val",
+    "▁sin",
+    "blem",
+    "▁fam",
+    "li",
+    "▁far",
+    "▁cle",
+    "▁col",
+    "mon",
+    "▁gra",
+    "led",
+    "ense",
+    "tin",
+    "ues",
+    "its",
+    "▁mem",
+    "▁inf",
+    "▁eas",
+    "ideo",
+    "▁top",
+    "io",
+    "pan",
+    "▁hum",
+    "▁old",
+    "ead",
+    "▁ord",
+    "ric",
+    "ants",
+    "oy",
+    "esn",
+    "uck",
+    "ason",
+    "ced",
+    "ool",
+    "rat",
+    "ouse",
+    "▁lar",
+    "▁art",
+    "▁wee",
+    "▁cer",
+    "ized",
+    "▁mat",
+    "con",
+    "erg",
+    "land",
+    "ines",
+    "▁chr",
+    "▁aut",
+    "▁lea",
+    "▁sou",
+    "oney",
+    "tty",
+    "▁ple",
+    "ulat",
+    "oks",
+    "▁few",
+    "▁sol",
+    "▁che",
+    "chn",
+    "ird",
+    "▁bre",
+    "▁dur",
+    "▁wom",
+    "me",
+    "izat",
+    "eric",
+    "ote",
+    "▁uni",
+    "eren",
+    "arn",
+    "ross",
+    "ices",
+    "ten",
+    "eral",
+    "ever",
+    "ieve",
+    "lish",
+    "ash",
+    "▁opp",
+    "alth",
+    "ger",
+    "▁sk",
+    "▁red",
+    "peri",
+    "▁det",
+    "▁ext",
+    "ner",
+    "ah",
+    "▁var",
+    "▁loc",
+    "gram",
+    "ists",
+    "ives",
+    "▁es",
+    "▁nor",
+    "tro",
+    "ale",
+    "▁iss",
+    "▁pri",
+    "gin",
+    "az",
+    "oc",
+    "▁pop",
+    "ern",
+    "▁sit",
+    "ket",
+    "▁pa",
+    "▁law",
+    "ages",
+    "br",
+    "▁cam",
+    "▁mom",
+    "osed",
+    "▁bro",
+    "ne",
+    "bs",
+    "▁cre",
+    "erat",
+    "▁sec",
+    "▁cap",
+    "▁vis",
+    "▁pat",
+    "ield",
+    "iet",
+    "▁tri",
+    "up",
+    "▁bra",
+    "ts",
+    "▁mot",
+    "▁unt",
+    "put",
+    "bo",
+    "ork",
+    "mer",
+    "ital",
+    "▁air",
+    "ined",
+    "▁beh",
+    "▁adv",
+    "▁ret",
+    "imes",
+    "▁tea",
+    "ural",
+    "sid",
+    "ters",
+    "▁pur",
+    "▁sci",
+    "bers",
+    "ient",
+    "ier",
+    "cc",
+    "sw",
+    "▁av",
+    "reen",
+    "ode",
+    "ont",
+    "▁dra",
+    "ann",
+    "nect",
+    "▁x",
+    "▁eu",
+    "ton",
+    "inat",
+    "ene",
+    "ared",
+    "els",
+    "▁mor",
+    "▁rat",
+    "cri",
+    "▁men",
+    "▁ah",
+    "ames",
+    "▁arm",
+    "eak",
+    "▁pay",
+    "▁hal",
+    "ins",
+    "ilit",
+    "stit",
+    "▁ra",
+    "▁leg",
+    "cl",
+    "pr",
+    "▁wal",
+    "▁bad",
+    "▁ge",
+    "roup",
+    "▁mus",
+    "man",
+    "▁gi",
+    "eds",
+    "▁aw",
+    "po",
+    "ark",
+    "row",
+    "▁dep",
+    "ully",
+    "ral",
+    "lect",
+    "pend",
+    "▁sev",
+    "ime",
+    "gest",
+    "here",
+    "▁yet",
+    "ted",
+    "▁rev",
+    "ds",
+    "▁ask",
+    "less",
+    "▁di",
+    "ets",
+    "line",
+    "▁aff",
+    "ired",
+    "▁est",
+    "ken",
+    "vid",
+    "most",
+    "ivid",
+    "unch",
+    "par",
+    "med",
+    "rop",
+    "ased",
+    "eone",
+    "▁ve",
+    "▁abs",
+    "ergy",
+    "ret",
+    "▁saw",
+    "▁ey",
+    "▁cal",
+    "uat",
+    "▁mid",
+    "vat",
+    "ream",
+    "vice",
+    "ians",
+    "rent",
+    "ctor",
+    "err",
+    "ush",
+    "ases",
+    "▁suc",
+    "erms",
+    "ave",
+    "angu",
+    "ries",
+    "▁wo",
+    "arts",
+    "▁fil",
+    "▁fat",
+    "▁cho",
+    "orts",
+    "▁fre",
+    "ee",
+    "ught",
+    "eng",
+    "ump",
+    "▁bar",
+    "ying",
+    "ane",
+    "▁tem",
+    "anks",
+    "ury",
+    "iat",
+    "mit",
+    "trol",
+    "▁net",
+    "▁maj",
+    "▁cra",
+    "ling",
+    "▁fig",
+    "orn",
+    "icat",
+    "pany",
+    "▁occ",
+    "ott",
+    "ands",
+    "▁exc",
+    "▁mr",
+    "ency",
+    "rope",
+    "itch",
+    "▁lit",
+    "abil",
+    "not",
+    "ma",
+    "▁typ",
+    "▁opt",
+    "ob",
+    "ser",
+    "ety",
+    "ms",
+    "peci",
+    "aces",
+    "aut",
+    "▁hon",
+    "cuss",
+    "▁sal",
+    "▁sor",
+    "att",
+    "▁lab",
+    "▁har",
+    "urch",
+    "nded",
+    "uce",
+    "ids",
+    "▁hy",
+    "▁fut",
+    "▁ste",
+    "ours",
+    "ems",
+    "utes",
+    "ng",
+    "ta",
+    "▁won",
+    "▁fa",
+    "▁env",
+    "ards",
+    "▁job",
+    "ium",
+    "▁dot",
+    "▁obv",
+    "ina",
+    "side",
+    "elve",
+    "cu",
+    "▁jes",
+    "▁pot",
+    "▁pie",
+    "▁tre",
+    "▁hey",
+    "▁mag",
+    "ron",
+    "▁key",
+    "swer",
+    "▁win",
+    "ucat",
+    "work",
+    "ides",
+    "▁low",
+    "▁vol",
+    "▁oth",
+    "atic",
+    "lf",
+    "ads",
+    "inds",
+    "com",
+    "ths",
+    "▁ver",
+    "ised",
+    "lo",
+    "▁squ",
+    "▁cut",
+    "oked",
+    "irit",
+    "ateg",
+    "ppy",
+    "mitt",
+    "come",
+    "hn",
+    "igin",
+    "mand",
+    "▁dam",
+    "ho",
+    "▁da",
+    "▁fur",
+    "iron",
+    "ilar",
+    "▁fac",
+    "▁neg",
+    "▁ago",
+    "ged",
+    "miss",
+    "enth",
+    "▁dou",
+    "▁hit",
+    "▁guy",
+    "▁bi",
+    "ove",
+    "fess",
+    "ples",
+    "owed",
+    "ured",
+    "▁ris",
+    "ints",
+    "rew",
+    "▁sum",
+    "▁hu",
+    "ploy",
+    "ude",
+    "ried",
+    "▁cir",
+    "▁dev",
+    "ear",
+    "▁tot",
+    "▁ann",
+    "duc",
+    "ik",
+    "pon",
+    "sted",
+    "▁ide",
+    "▁'",
+    "ipp",
+    "▁eat",
+    "▁dom",
+    "▁",
+    "e",
+    "t",
+    "o",
+    "a",
+    "i",
+    "n",
+    "s",
+    "r",
+    "h",
+    "l",
+    "d",
+    "u",
+    "c",
+    "m",
+    "y",
+    "g",
+    "w",
+    "f",
+    "p",
+    "b",
+    "v",
+    "k",
+    "'",
+    "j",
+    "x",
+    "q",
+    "z",
+    "<EOU>",
+    "<EOB>"
+  ],
+  "vocab_size": 1026,
+  "blank_id": 1026,
+  "eou_token": "<EOU>",
+  "eou_id": 1024,
+  "bos_id": -1,
+  "eos_id": -1,
+  "pad_id": -1
+}