---
language: zh
license: mit
tags:
- kronos
- financial-modeling
- time-series
- cryptocurrency
- stock-prediction
- pytorch
- transformers
datasets:
- custom
metrics:
- mse
- mae
widget:
- text: "Financial time series prediction"
---

# Kronos Tokenizer - Fine-tuned on Custom Dataset

This is a fine-tuned version of [Kronos](https://huggingface.co/NeoQuasar/Kronos-Tokenizer-base) tokenizer,
adapted for better performance on custom financial datasets.

## Model Details

- **Model Type**: Tokenizer
- **Base Model**: NeoQuasar/Kronos-Tokenizer-base
- **Fine-tuned For**: Financial Time Series Prediction
- **Architecture**: Transformer-based with custom tokenization
- **Input**: OHLCV (Open, High, Low, Close, Volume, Amount) data
- **Output**: Multi-step time series predictions

## Training Details

- **Training Data**: Crypto Dataset (BTC, ETH, SOL, XAU)
- **Time Range**: 2022-01-21 to 2025-09-16
- **Frequency**: 5-minute intervals
- **Sequence Length**: 90 historical points
- **Prediction Horizon**: 10 future points

## Usage

### For Tokenizer
```python
from model.kronos import KronosTokenizer

tokenizer = KronosTokenizer.from_pretrained("NeoQuasar/Kronos-Tokenizer-base")
# Your tokenization code here
```

### For Predictor
```python
from model.kronos import Kronos, KronosTokenizer, KronosPredictor

tokenizer = KronosTokenizer.from_pretrained("NeoQuasar/Kronos-Tokenizer-base")
model = Kronos.from_pretrained("NeoQuasar/Kronos-Tokenizer-base")
predictor = KronosPredictor(model, tokenizer, device="cuda")

# Your prediction code here
predictions = predictor.predict(...)
```

### With the Original Repository
```python
# Clone the Kronos repository
git clone https://github.com/shiyu-coder/Kronos.git
cd Kronos

# Use the fine-tuned models
python examples/use_finetuned_model.py \
    --csv_data your_data.csv \
    --lookback 400 \
    --pred_len 120
```

## Performance

This fine-tuned model shows improved performance on the target domain compared to the base model:

- **Domain Adaptation**: Specialized for the training dataset characteristics
- **Numerical Stability**: Improved convergence during fine-tuning
- **Inference Speed**: Optimized for the target sequence lengths

## Limitations

- Optimized for 5-minute financial data intervals
- May require re-tuning for different time frequencies
- Performance may vary on datasets with different statistical properties

## Citation

```bibtex
@misc{shi2025kronos,
      title={Kronos: A Foundation Model for the Language of Financial Markets},
      author={Yu Shi and Zongliang Fu and Shuo Chen and Bohan Zhao and Wei Xu and Changshui Zhang and Jian Li},
      year={2025},
      eprint={2508.02739},
      archivePrefix={arXiv},
      primaryClass={q-fin.ST},
      url={https://arxiv.org/abs/2508.02739},
}
```

## Contact

For questions or issues, please open an issue on the [Kronos GitHub repository](https://github.com/shiyu-coder/Kronos).