---
title: GAME-MLX-medium
canonical_url: "https://www.modelscope.cn/models/baicai1145/GAME-MLX-medium"
md_url: "https://www.modelscope.cn/models/baicai1145/GAME-MLX-medium.md"
repository: baicai1145/GAME-MLX-medium
last_updated: 2026-10-23
license: mit
base_model:
  - openvpi/GAME
base_model_relation: finetune
parameters: 99.8M
tensor_type:
  - F32
  - F16
library_name:
  - mlx
  - safetensors
language:
  - en
  - zh
  - ja
  - yue
downloads: 4
stars: 0
tags:
  - mlx
  - midi
  - singing-voice
  - music-transcription
  - audio
---

# GAME-MLX-medium

> GAME-MLX-medium - baicai1145 在 ModelScope 开源的模型。GAME (Generative Adaptive MIDI Extractor) — MLX version of openvpi/GAME GAME-1.0-medium checkpoint, for Apple Silicon GPU inference.

baicai1145/GAME-MLX-medium 是 ModelScope 魔搭社区上的 99.8M 参数机器学习模型，采用 mit 许可，基于 openvpi/GAME 构建。

- **Repository**: baicai1145/GAME-MLX-medium
- **License**: mit
- **Parameters**: 99.8M
- **Base model**: openvpi/GAME
- **Tags**: mlx, midi, singing-voice, music-transcription, audio
- **Downloads**: 4
- **Stars**: 0
- **Last updated**: 2026-10-23

Source: https://www.modelscope.cn/models/baicai1145/GAME-MLX-medium

---

# GAME-MLX-medium

**GAME (Generative Adaptive MIDI Extractor)** — MLX version of [openvpi/GAME](https://github.com/openvpi/GAME) `GAME-1.0-medium` checkpoint, for Apple Silicon GPU inference.

| File | Description |
|---|---|
| `weights.fp16.safetensors` | fp16 weights (low memory) |
| `weights.safetensors` | fp32 weights (bit-aligned with torch) |
| `config.yaml` | architecture & features config (same as upstream) |
| `lang_map.json` | language map (en/ja/yue/zh) |

Parameters: 49,887,251 (49.89M) / 671 tensors. Key names preserved from the torch checkpoint (including `model.` Lightning prefix).

## Usage

```bash
pip install gamemlx
gamemlx extract vocals.wav -m medium --language zh
```

```python
import gamemlx
model, lang_map = gamemlx.load_model("medium")
gamemlx.extract(model, "vocals.wav", language="zh")
```

## Alignment & performance

Verified on 20 test files against the official PyTorch implementation (CPU, 32-true, identical injected d3pm randomness): **fp32 = 20/20 files with all onset/offset/pitch diffs 0.0000**; fp16 = 19/20 (pitch diff ≤ 0.0011). MLX fp32 is ~9.4× faster than torch CPU (0.172s vs 1.609s avg per file), peak wired GPU memory 206MB (fp16: 111MB).

Also available on ModelScope: [baicai1145/GAME-MLX-medium](https://modelscope.cn/models/baicai1145/GAME-MLX-medium)

## License

MIT (following upstream openvpi/GAME).

Code: [baicai-1145/game-mlx](https://github.com/baicai-1145/game-mlx) — `pip install gamemlx`