Add native CUDA weights and update the model card

Native safetensors for the scriptrs CUDA backend (Parakeet TDT v2 and v3, Silero VAD v5), a card covering every runtime, a LICENSE covering Parakeet and Silero files, and provenance for the native files.

LICENSE

@@ -1,9 +1,21 @@
-SPDX-License-Identifier: CC-BY-4.0
+Parakeet files (parakeet-v2/, parakeet-v3/)
 
+SPDX-License-Identifier: CC-BY-4.0
 Copyright (c) NVIDIA Corporation.
 
-This Core ML bundle is a conversion of NVIDIA Parakeet TDT 0.6B v3
-and is licensed under Creative Commons Attribution 4.0 International.
-You must give appropriate credit to NVIDIA.
+These files are conversions of NVIDIA Parakeet TDT 0.6B v2 and v3 to Core ML
+and to a safetensors layout for scriptrs, and are licensed under Creative
+Commons Attribution 4.0 International. You must give appropriate credit to
+NVIDIA.
 
 https://creativecommons.org/licenses/by/4.0/legalcode
+
+Silero VAD files (vad/)
+
+SPDX-License-Identifier: MIT
+Copyright (c) 2020-present Silero Team
+
+These files are conversions of Silero VAD and are licensed under the MIT
+License.
+
+https://github.com/snakers4/silero-vad/blob/master/LICENSE

PROVENANCE-native.json

@@ -0,0 +1,40 @@
+{
+  "schema_version": 1,
+  "artifact_id": "scriptrs-native-cuda",
+  "runtime": "scriptrs_cuda",
+  "format": "scriptrs-native",
+  "format_version": "2",
+  "exporter": "dev-tools/nemo/export_native.py",
+  "files": [
+    {
+      "path": "parakeet-v2/model.safetensors",
+      "size": 2471715672,
+      "sha256": "ad48c4903e9bec8ad0004959d85ab9a92ad520294f0298c7822e45dcfd0dec42",
+      "license": "CC-BY-4.0",
+      "source": "nvidia/parakeet-tdt-0.6b-v2",
+      "source_sha256": "d99e39955c9d3d0350d8fb7c75e40c64a2b2eaeb003883d7c941fd2e8747b28c",
+      "toolkit": "nemo-toolkit 3.0.0"
+    },
+    {
+      "path": "parakeet-v3/model.safetensors",
+      "size": 2508444512,
+      "sha256": "fc79fd91aca14a2b02ba591d066d4e3cd088ef7321f6966dfd0b9c85038fbd47",
+      "license": "CC-BY-4.0",
+      "source": "nvidia/parakeet-tdt-0.6b-v3",
+      "source_sha256": "3cbdc85877e668ca7b82d0d56770eb1fac76691f55d6b97545e8d61ca588d10d",
+      "toolkit": "nemo-toolkit 3.0.0"
+    },
+    {
+      "path": "vad/silero-vad-v5.safetensors",
+      "size": 1240188,
+      "sha256": "88f62b0eea97c55d6ccee086d1ca67b6f733c29ceee4263e30fe5425bdddc187",
+      "license": "MIT",
+      "source": "silero-vad 5.1.2 (16 kHz branch)",
+      "source_sha256": "85c48e1f0ecb604e5d2a268f3ccfb912d4f7e935acdc86af5a3fc5b0aea7b29a"
+    }
+  ],
+  "shared_with_coreml": [
+    "parakeet-v2/vocab.txt",
+    "parakeet-v3/parakeet_vocab.json"
+  ]
+}

README.md

@@ -1,26 +1,103 @@
 ---
+license: cc-by-4.0
+library_name: scriptrs
+pipeline_tag: automatic-speech-recognition
+base_model:
+- nvidia/parakeet-tdt-0.6b-v2
+- nvidia/parakeet-tdt-0.6b-v3
+base_model_relation: quantized
+language:
+- en
+- es
+- fr
+- de
+- bg
+- hr
+- cs
+- da
+- nl
+- et
+- fi
+- el
+- hu
+- it
+- lv
+- lt
+- mt
+- pl
+- pt
+- ro
+- sk
+- sl
+- sv
+- ru
+- uk
 tags:
-- coreml
-- automatic-speech-recognition
 - parakeet
-- macos
-license: apache-2.0
+- automatic-speech-recognition
+- coreml
+- cuda
+- safetensors
+- rust
 ---
 
 # scriptrs models
 
-Runtime model bundle for `scriptrs`.
+Model files for [scriptrs](https://github.com/avencera/scriptrs), a Rust library for fast NVIDIA Parakeet TDT v2 and v3 speech-to-text on Core ML (macOS) and native CUDA (Linux). With the default `online` feature, scriptrs downloads these files on first use from a pinned revision of this repository.
+
+These are format conversions of NVIDIA's models, not new or retrained models. The native CUDA weights are NVIDIA's FP32 weights with their original names; scriptrs converts selected layers to FP16 on the GPU at load time.
+
+## Speed
+
+RTX 5090, Parakeet v3, LibriSpeech test-clean (5.4 h), one utterance at a time:
+
+| | Speed (x real time) | WER |
+|---|---|---|
+| scriptrs CUDA (FP16 default) | 1416x | 1.93 % |
+| NVIDIA NeMo (PyTorch) | 477x | 1.93 % |
+| parakeet-rs 0.4.3 (ONNX Runtime) | 253x | 1.93 % |
+
+On the 44 Earnings-21 calls (39 h), long-form runs at 3443x with 8.24 % pooled WER. More results are in the [scriptrs README](https://github.com/avencera/scriptrs#cuda).
+
+## Usage
+
+```toml
+# macOS, Core ML
+scriptrs = "0.2"
+
+# Linux, NVIDIA GPU (not yet in a crates.io release)
+scriptrs = { git = "https://github.com/avencera/scriptrs", features = ["cuda"] }
+```
+
+```rust
+use scriptrs::{ModelBundle, ParakeetDeployment, TranscriptionPipeline};
+
+// Core ML on macOS
+let pipeline = TranscriptionPipeline::from_pretrained()?;
+
+// native CUDA on Linux
+let bundle = ModelBundle::from_pretrained_native(ParakeetDeployment::v3())?;
+let pipeline = TranscriptionPipeline::from_bundle(bundle)?;
+
+let result = pipeline.run(&audio)?; // mono 16 kHz f32 samples
+```
+
+For offline setups, download this repository and set `SCRIPTRS_MODELS_DIR` to its local path.
+
+## Contents
+
+| Path | Runtime | Source |
+|---|---|---|
+| `parakeet-v2/encoder.mlmodelc`, `decoder.mlmodelc`, `joint-decision.mlmodelc` | Core ML | [nvidia/parakeet-tdt-0.6b-v2](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v2) |
+| `parakeet-v3/encoder.mlmodelc`, `decoder.mlmodelc`, `joint-decision.mlmodelc` | Core ML | [nvidia/parakeet-tdt-0.6b-v3](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3), via [FluidInference/parakeet-tdt-0.6b-v3-coreml](https://huggingface.co/FluidInference/parakeet-tdt-0.6b-v3-coreml) |
+| `parakeet-v2/model.safetensors` | CUDA | nvidia/parakeet-tdt-0.6b-v2, exported from the NeMo checkpoint |
+| `parakeet-v3/model.safetensors` | CUDA | nvidia/parakeet-tdt-0.6b-v3, exported from the NeMo checkpoint |
+| `parakeet-v2/vocab.txt`, `parakeet-v3/parakeet_vocab.json` | both | NVIDIA tokenizers |
+| `vad/silero-vad.mlmodelc` | Core ML | [Silero VAD](https://github.com/snakers4/silero-vad) |
+| `vad/silero-vad-v5.safetensors` | CUDA | Silero VAD 5.1.2, 16 kHz branch |
 
-Current contents:
-- Parakeet TDT v2 CoreML split models
-- vocabulary file
-- Silero VAD CoreML model for the `long-form` feature
+The native weight files record their source and its SHA-256 in the safetensors metadata. `PROVENANCE.json` and `PROVENANCE-native.json` list sources, revisions and checksums.
 
-Layout:
-- `parakeet-v2/encoder.mlmodelc`
-- `parakeet-v2/decoder.mlmodelc`
-- `parakeet-v2/joint-decision.mlmodelc`
-- `parakeet-v2/vocab.txt`
-- `vad/silero-vad.mlmodelc`
+## License
 
-This repo is intended for `scriptrs` model download via the `online` feature.
+The Parakeet files are derived from NVIDIA Parakeet TDT 0.6B v2 and v3, licensed under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/). Credit: NVIDIA. Changes: converted to Core ML and to a safetensors layout for scriptrs. The Silero VAD files are derived from Silero VAD, licensed under the MIT License. See `LICENSE`.

parakeet-v2/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ad48c4903e9bec8ad0004959d85ab9a92ad520294f0298c7822e45dcfd0dec42
+size 2471715672

parakeet-v3/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fc79fd91aca14a2b02ba591d066d4e3cd088ef7321f6966dfd0b9c85038fbd47
+size 2508444512

vad/silero-vad-v5.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:88f62b0eea97c55d6ccee086d1ca67b6f733c29ceee4263e30fe5425bdddc187
+size 1240188