File size: 1,417 Bytes
d015030 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 | ---
license: apache-2.0
tags:
- chess
- chess-engine
- numba
- nnue
- aichessathon
- game-ai
language:
- en
pipeline_tag: other
---
# Hyperion-Chess
Chess engine for [AI Chessathon 2026](https://aichessathon.com).
## Overview
Hyperion-Chess is a competition-grade Python chess agent combining:
- **Iterative deepening alpha-beta** with quiescence search
- **Numba-jitted PST evaluation** (Phase 1)
- **Micro-NNUE** int8 weights (Phase 3, trained locally on SF-labelled FENs)
- **4-man Syzygy tablebases** for perfect endgames within the 50 MB zip cap
- **Dynamic time management** for 120s + 0.5s control
## Source code
GitHub: [shahaman098/hyperion-chess](https://github.com/shahaman098/hyperion-chess)
## Weights (coming soon)
| File | Description |
|---|---|
| `nnue.npz` | Quantized Micro-NNUE (768→128×2→32→1) |
| Syzygy 4-man | Endgame tablebase (too large for HF; shipped in competition zip) |
## Usage
This is a **competition submission**, not a pip package. The platform imports `agent.get_move(fen, time_left_ms)`.
```python
import agent
move = agent.get_move("rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1", 120_000)
# → "e2e4"
```
## Training
Training scripts live in the GitHub repo under `training/`. Positions may be labelled with Stockfish at train time (allowed by competition rules); only self-trained weights ship in the zip.
## License
Apache 2.0
|