| --- |
| license: apache-2.0 |
| tags: |
| - speech-recognition |
| - cuda |
| language: |
| - en |
| base_model: |
| - nvidia/parakeet-tdt-0.6b-v2 |
| --- |
| |
| # Paraketto |
|
|
| Weights for [paraketto](https://github.com/lfrati/paraketto) — a C++/CUDA speech-to-text engine for NVIDIA's Parakeet TDT 0.6B V2. |
|
|
| ## Files |
|
|
| | File | Size | Description | |
| |------|------|-------------| |
| | `weights.bin` | 1.2 GB | FP16 model weights | |
| | `weights_fp8.bin` | 604 MB | FP8 E4M3 quantized weights (Blackwell) | |
| | `bench-data.tar.gz` | 222 MB | Benchmark WAV files | |
|
|
| ## Usage |
|
|
| ```bash |
| make paraketto.cuda |
| ./paraketto.cuda audio.wav # auto-downloads weights on first run |
| ``` |
|
|
| See the [GitHub repo](https://github.com/lfrati/paraketto) for build instructions and benchmarks. |
