| --- |
| license: mit |
| base_model: openai/whisper-small |
| tags: [snapdragon, hexagon, npu, qualcomm, qnn, onnxruntime, whisper, asr, on-device, edge] |
| --- |
| |
| # Whisper‑small — QNN bundle for Snapdragon X2 Elite (Hexagon **V81**) |
|
|
| OpenAI **Whisper‑small** compiled to **precompiled QNN‑ONNX** (encoder + decoder) |
| for the **Hexagon NPU of the Snapdragon X2 Elite** (`dsp_arch v81`). Intended for |
| **on‑device, low‑power streaming speech‑to‑text** via **ONNX Runtime + QNN EP**. |
|
|
| ## Specs |
| | | | |
| |---|---| |
| | Base model | `openai/whisper-small` | |
| | Format | `precompiled_qnn_onnx` (ONNX wrappers + QNN context binaries) | |
| | Components | encoder, decoder | |
| | Target | Snapdragon **X2 Elite**, Hexagon **V81** | |
| | Use via | ONNX Runtime with the QNN execution provider | |
|
|
| ## How it was made |
| Compiled via [`qai-hub-models`](https://github.com/quic/ai-hub-models) on |
| **Qualcomm AI Hub**, targeting `Snapdragon X2 Elite CRD`, from stock Whisper‑small. |
|
|
| ## ⚠️ Runtime note |
| Like all QNN bundles, this is version‑coupled to the QAIRT/ONNX‑Runtime‑QNN and the |
| Hexagon NPU driver it was built against — keep the NPU driver current. |
|
|
| ## License |
| MIT (inherits OpenAI Whisper's license). |
|
|
