README.md

---
license: mit
library_name: kronos-mlx
pipeline_tag: time-series-forecasting
tags:
  - mlx
  - apple-silicon
  - finance
  - kronos
  - time-series
base_model: NeoQuasar/Kronos-mini
---

# Kronos-mini (MLX)

Apple [MLX](https://github.com/ml-explore/mlx) port of [`NeoQuasar/Kronos-mini`](https://huggingface.co/NeoQuasar/Kronos-mini) — the lightweight 4.1M-parameter Kronos variant with a 2048-token context window. Pair with [`gxcsoccer/kronos-mlx-tokenizer-2k`](https://huggingface.co/gxcsoccer/kronos-mlx-tokenizer-2k).

## Usage

```python
from kronos_mlx import Kronos, KronosTokenizer, KronosPredictor

tokenizer = KronosTokenizer.from_pretrained("gxcsoccer/kronos-mlx-tokenizer-2k")
model     = Kronos.from_pretrained("gxcsoccer/kronos-mlx-mini")
predictor = KronosPredictor(model, tokenizer, max_context=2048)
```

## Original

- Upstream: [shiyu-coder/Kronos](https://github.com/shiyu-coder/Kronos) (AAAI 2026)
- PyTorch weights: [NeoQuasar/Kronos-mini](https://huggingface.co/NeoQuasar/Kronos-mini)
- MLX port: [github.com/gxcsoccer/kronos-mlx](https://github.com/gxcsoccer/kronos-mlx)