|
|
--- |
|
|
title: Yuuki - Mobile-Trained Code Generator |
|
|
emoji: πΈ |
|
|
colorFrom: purple |
|
|
colorTo: purple |
|
|
sdk: gradio |
|
|
sdk_version: 5.9.1 |
|
|
python_version: 3.11 |
|
|
app_file: app.py |
|
|
pinned: true |
|
|
license: apache-2.0 |
|
|
tags: |
|
|
- code-generation |
|
|
- mobile-training |
|
|
- zero-budget |
|
|
- edge-ml |
|
|
- experimental |
|
|
thumbnail: >- |
|
|
https://cdn-uploads.huggingface.co/production/uploads/68a8bd1d45ff88ffe886e331/M1kPiM_vnsSzo6Ee-mzG9.jpeg |
|
|
--- |
|
|
|
|
|
# πΈ Yuuki - Mobile-Trained Code Generator |
|
|
|
|
|
**First LLM trained entirely on mobile CPU with $0 budget.** |
|
|
|
|
|
Try the live demo above! |
|
|
|
|
|
## Features |
|
|
|
|
|
- β
Agda code generation (best performance: 55/100) |
|
|
- β οΈ Limited C, Assembly support |
|
|
- π± Trained on Snapdragon 685 CPU |
|
|
- π° Zero computational cost |
|
|
- π¬ Fully documented research experiment |
|
|
|
|
|
## Model |
|
|
|
|
|
- **Base:** DistilGPT-2 (82M parameters) |
|
|
- **Training:** 2,000 steps (5.3% of v0.1) |
|
|
- **Hardware:** Mobile CPU only |
|
|
- **Time:** ~50 hours continuous |
|
|
- **Cost:** $0.00 |
|
|
|
|
|
## Links |
|
|
|
|
|
- π€ [Model Card](https://huggingface.co/OpceanAI/Yuuki-best) |
|
|
- π Paper (coming soon) |
|
|
- π» [Training Code](https://github.com/YuuKi-OS/yuuki-training) |
|
|
|
|
|
--- |
|
|
|
|
|
*Proving the barrier to AI is mindset, not money* πΈ |