Beta

Explore
Marketplace
Neural Labs
Chat
Wallet
Docs

Beta

Explore Marketplace Neural Labs Wallet Docs

Build

Manage

My assets Manage wallet Organization

Community

Discord X (Twitter)Telegram

Docs Privacy Policy Terms and Conditions

Proud member of

Proud member of

Docs Cookie Policy

Privacy Policy Terms and Conditions

© 2026 AlphaNeural AI. All rights reserved.

mortal-298k – AI Model by VoidShine | AlphaNeural AI

You can deploy this model and start earning money today!

VoidShine/
mortal-298k

like

0

reinforcement-learning

VoidShine/mortal-dataset

Views

No views yet

Mortal Training Checkpoint — 298k Steps

⚠️ WARNING: This model is NOT recommended for use in ranked matches.

Overview

A trained checkpoint of Mortal, an AI for Japanese Riichi Mahjong powered by deep reinforcement learning. This is a hanchan (south round) model.

Model

Base framework: Mortal (by Equim)
Checkpoint: mortal_298k.pth (298,000 steps)
Architecture: ResNet with 192 channels, 40 blocks
Game type: 4-player hanchan (south round)

Dataset

Training data sourced from Tenhou high-level 4-player hanchan games (2025–2026), stored under dataset/4p_hanchan/.

Note: dataset/4p_tonpuu/ (east round data) is included in the repository but was not used for training this model.

Performance

AI similarity: 83%–92% on mjai.ekyu.moe
Mahjong Soul: Average S+ level on MAKA test

Usage

Clone the Mortal repository and follow its setup instructions.
Place mortal_298k.pth at the path specified by state_file in config.toml.
Adjust paths in config.toml to match your local environment.
Refer to Mortal's documentation for inference and training details.

Configuration

See config.toml for the full training configuration. All paths are set to /path/to/ placeholders — replace them with your actual local paths before use.

License

This project is licensed under AGPL-3.0, consistent with Mortal.

The dataset consists of game logs from Tenhou. Please respect Tenhou's terms of service when using this data.