README / README.md
shgsousa's picture
Updated Project Card.
34311e6
|
Raw History Blame Contribute Delete
1.85 kB
---
title: README
emoji: πŸ‘
colorFrom: green
colorTo: yellow
sdk: static
pinned: false
license: mit
short_description: High-performance, NPU-accelerated AI and speaker diarization
---
# Snapdragon-AI-Stack πŸš€
**Optimized SOTA AI for the Windows ARM64 Ecosystem.**
Welcome to the official Hugging Face home of the **Snapdragon-AI-Stack**. Our mission is to transform the **Qualcomm Snapdragon X Elite/Plus** into a powerhouse for Artificial Intelligence by providing NPU-accelerated models and native binaries.
---
## πŸ—οΈ Our Focus Areas
- **πŸŽ™οΈ SnapNote:** High-performance transcription and diarization pipelines (Whisper Turbo + Pyannote 3.1) running natively on the Hexagon NPU.
- **πŸ“¦ NPU-Ready Model Zoo:** Quantized ONNX weights specifically calibrated for the `QNNExecutionProvider`.
- **🎑 Wheel Factory:** Pre-compiled `win_arm64` wheels for PyTorch, Torchaudio, and hf-transfer to ensure seamless installation on Windows 11 ARM.
---
## πŸ“ˆ The NPU Advantage
By offloading tensor operations from the CPU to the NPU on Snapdragon laptops, we achieve:
* **Efficiency:** Up to 80% reduction in CPU power draw.
* **Speed:** Faster-than-real-time diarization and inference.
* **Mobility:** Professional-grade AI workflows that don't sacrifice battery life.
---
## πŸ› οΈ Quick Links
* **Main Organization:** [Snapdragon-AI-Stack on GitHub](https://github.com/Snapdragon-AI-Stack)
* **Binary Builds:** [PyTorch QNN Wheels](https://github.com/Snapdragon-AI-Stack/pytorch-qnn)
* **Fast Downloads:** [hf-transfer-qnn](https://github.com/Snapdragon-AI-Stack/hf-transfer-qnn)
---
## 🀝 Community & Collaboration
We are actively seeking contributors with expertise in the **Qualcomm AI Stack**, **ONNX Runtime quantization**, and **Windows ARM64 development**.
*"Making the Hexagon NPU the heart of your AI workflow."*