|
|
--- |
|
|
license: other |
|
|
language: |
|
|
- en |
|
|
tags: |
|
|
- on-device |
|
|
- mobile |
|
|
- android |
|
|
- ios |
|
|
- melange |
|
|
- zetic |
|
|
- whisper |
|
|
pipeline_tag: automatic-speech-recognition |
|
|
base_model: |
|
|
- openai/whisper-base |
|
|
--- |
|
|
|
|
|
<!-- ============================================== --> |
|
|
<!-- FILL THIS IN BEFORE PUBLISHING --> |
|
|
<!-- ============================================== --> |
|
|
<!-- |
|
|
|
|
|
🚨 Model License Rule (One Rule Applies to All) |
|
|
|
|
|
**Never claim or imply commercial usage rights unless the base model license explicitly allows commercial use.** |
|
|
|
|
|
- If the base model license allows commercial use → you may state “commercial use allowed (subject to the base model license)”. |
|
|
- If the base model license restricts commercial use, is AGPL/GPL, research-only, or dual-licensed → **do NOT mention commercial use**. Always defer to the base model license. |
|
|
|
|
|
ZETIC licenses only the runtime and optimization artifacts. |
|
|
Model licensing responsibility always remains with the original model owner. |
|
|
|
|
|
---> |
|
|
|
|
|
--- |
|
|
|
|
|
 |
|
|
|
|
|
## Melange: On-device AI for Mobile — *Select. Benchmark. Deploy.* |
|
|
<div align="center"> |
|
|
|
|
|
[**🌐 Website**](https://zetic.ai) | [**📚 Docs**](https://docs.zetic.ai) | [**🎯 Melange Dashboard**](https://mlange.zetic.ai) | [**💻 GitHub Apps Repo**](https://github.com/zetic-ai/ZETIC_Melange_apps/issues) | [**💬 Discord**](https://discord.gg/q6vW4UscRY) |
|
|
|
|
|
</div> |
|
|
|
|
|
--- |
|
|
|
|
|
## ⚡️ What is ZETIC-ai/whisper-tiny-decoder? |
|
|
|
|
|
ZETIC-ai/whisper-tiny-encoder is a highly optimized version of **[Whisper Tiny by OpenAI](https://huggingface.co/openai/whisper-tiny)**, built to run **on-device** (Android, iOS) with **Melange SDK**. |
|
|
<br> |
|
|
<br> |
|
|
For the encoder model that runs locally: [Check model card](https://huggingface.co/ZETIC-ai/whisper-tiny-encoder) |
|
|
|
|
|
**Key Details:** |
|
|
- 🎯 **Task**: Automatic Speech Recognition |
|
|
- 🚀 **Engine**: ZETIC Melange Runtime |
|
|
- 💡 **Best For**: Speech to Text, Speech Translation |
|
|
|
|
|
Unlike standard GGUF or ONNX exports where you manually download `.tflite` or `.gguf` files and bundle them into your APK/IPA, **this model is fetched automatically** by the Melange SDK at runtime: no asset management, no file hosting, no bloated app sizes. |
|
|
|
|
|
--- |
|
|
|
|
|
## 🚀 How to Use |
|
|
|
|
|
Deploy in seconds. The Melange SDK handles model fetching, NPU delegation, and hardware acceleration automatically. |
|
|
|
|
|
### Android (Kotlin) |
|
|
```kotlin |
|
|
val model = ZeticMLangeHFModel(context, "ZETIC-ai/whisper-tiny-decoder") |
|
|
val output = model.run(inputData) |
|
|
``` |
|
|
|
|
|
|
|
|
### iOS (Swift) |
|
|
```swift |
|
|
let model = try await ZeticMLangeHFModel("ZETIC-ai/whisper-tiny-decoder") |
|
|
let outputs = try model.run(inputs: inputs) |
|
|
``` |
|
|
|
|
|
💡 **Looking for Java, React Native, or Flutter?** Visit [Melange Dashboard](https://mlange.zetic.ai). |
|
|
|
|
|
**Installation**: [View docs](https://docs.zetic.ai/hugging-face) |
|
|
|
|
|
--- |
|
|
|
|
|
## 📊 Benchmarks & Compatibility |
|
|
|
|
|
**Optimized for Flagships.** This model release is specifically tuned for the latest flagship smartphones. |
|
|
|
|
|
| Device | Chipset | Inference Time (ms) | |
|
|
| :--- | :--- | :--- | |
|
|
| **Samsung Galaxy S25 Ultra** | Snapdragon 8 Elite | TBD | |
|
|
| **Samsung Galaxy S25** | Snapdragon 8 Elite | TBD | |
|
|
| **iPhone 16 Pro** | A18 Pro | TBD | |
|
|
| **iPhone 16** | A18 | TBD | |
|
|
|
|
|
🚀 **Beyond the Flagships:** Melange supports a vast ecosystem of **100+ devices**. <br><br> |
|
|
|
|
|
Melange SDK automatically handles optimization and fallback for the wider ecosystem, including budget & legacy smartphones. <br> |
|
|
**[Explore the full ecosystem optimization on Melange Dashboard]([mlange.zetic.ai])** |
|
|
|
|
|
--- |
|
|
|
|
|
## 🤔 Why Melange? |
|
|
|
|
|
<div> |
|
|
|
|
|
### **1. The Old Way (Slow & Manual)** |
|
|
Separate iOS/Android Workflows | Manual Updates Require Full Re-Deployment |
|
|
|
|
|
 |
|
|
|
|
|
### **2. Semi-Auto SDKs (Hardware-Locked)** |
|
|
Hardware-Locked Model Cards | Manual File Management (Drag & Drop) |
|
|
|
|
|
 |
|
|
|
|
|
|
|
|
### **3. The Melange Way (Fast & Automatic)** |
|
|
Device-Aware Auto-Routing | Hardware-Agnostic NPU API | Small App Size |
|
|
|
|
|
 |
|
|
|
|
|
</div> |
|
|
|
|
|
✅ **Zero Asset Management**: We host the model & the complexity. Your app calls '[MODEL_ID]', and we deliver. <br/> |
|
|
✅ **Automatic Fetching**: The SDK streams the model at initialization, no manual file uploads.<br/> |
|
|
✅ **Device-Aware Auto-Routing**: The SDK auto-detects and routes the best-fit runtime variant for that device.<br/> |
|
|
✅ **Unified Workflow**: Maintain a single, identical deployment path for both iOS and Android. <br/> |
|
|
✅ **Instant Updates**: Push model updates instantly without waiting for App Store approvals. <br/> |
|
|
✅ **Global Scale**: Optimized paths for mid-range and low-end chips, not just flagships.<br/> |
|
|
<br/> |
|
|
|
|
|
--- |
|
|
|
|
|
## 🚀 Unlock Full Performance |
|
|
|
|
|
**Advanced Benchmark Reports • Extended Model Catalog • 100+ Optimized Devices** |
|
|
|
|
|
<a href="https://mlange.zetic.ai"> |
|
|
<img src="https://img.shields.io/badge/🚀_Open_Dashboard-34A9A3?style=for-the-badge" alt="MeLange Dashboard" height="30" /> |
|
|
</a> |
|
|
|
|
|
--- |
|
|
|
|
|
## 📄 License & Terms |
|
|
|
|
|
**Base Model** |
|
|
Model weights are licensed under the original license provided by the model author. |
|
|
See the upstream model card for full terms. |
|
|
|
|
|
**ZETIC Hugging Face Runtime Bundle** |
|
|
This release includes a pre-optimized runtime + model bundle enabling local, on-device inference across compatible devices. |
|
|
|
|
|
The runtime is optimized and benchmarked for a limited set of flagship devices. |
|
|
On other compatible devices, the model may still run, but performance may vary. |
|
|
|
|
|
Advanced device coverage and performance guarantees are available via the **Melange Dashboard (Pro)**. |
|
|
|
|
|
**You may** |
|
|
- Run the model locally on compatible devices |
|
|
- Embed it in applications distributed to end users |
|
|
- Use it in accordance with the base model license |
|
|
|
|
|
**You may not** |
|
|
- Extract or reuse the runtime outside this package |
|
|
- Reverse engineer the optimization format |
|
|
- Redistribute the runtime as a standalone SDK or service |
|
|
- Bypass license verification or telemetry (if/when enabled) |
|
|
|
|
|
**More devices, custom models, and performance guarantees** |
|
|
Get started at the **Melange Dashboard**: https://mlange.zetic.ai |
|
|
|
|
|
Enterprise or custom licensing: contact@zetic.ai |
|
|
|
|
|
--- |
|
|
## 📞 Support & Community |
|
|
|
|
|
- **📚 Documentation**: [https://docs.zetic.ai](https://docs.zetic.ai) |
|
|
- **🐛 Melange Apps Repo**: [https://github.com/zetic-ai/ZETIC_Melange_apps/](https://github.com/zetic-ai/ZETIC_Melange_apps/) |
|
|
- **💬 Discord**: [https://discord.gg/q6vW4UscRY](https://discord.gg/q6vW4UscRY) |
|
|
|
|
|
--- |
|
|
|
|
|
<div align="center"> |
|
|
|
|
|
**By ZETIC.ai** |
|
|
*On-Device AI for Everything — for any model, on any device, in any framework* |
|
|
|
|
|
[⭐ Star us on GitHub](https://github.com/zetic-ai/ZETIC_Melange_apps) • [👾 Discord](https://discord.gg/q6vW4UscRY) • [🚀 Try Melange](https://mlange.zetic.ai) • [📖 Documentation](https://docs.zetic.ai) |
|
|
|
|
|
</div>](https://www.notion.so/Melange-Dashboard-2-0-Release-Checklist-2fe933a34b9880bd9494c31ea06d4f08?source=copy_link) |
