GitHub - shiyu-coder/Kronos: Kronos: A Foundation Model for the Language of Financial Markets
6 hours ago
- #Time Series Forecasting
- #Financial AI
- #Open Source
- Kronos is an open-source foundation model for financial candlesticks (K-lines), trained on data from over 45 global exchanges.
- It features a two-stage framework: a tokenizer quantizes OHLCV data into hierarchical discrete tokens, and an autoregressive Transformer is pre-trained on these tokens.
- The model is designed for diverse quantitative tasks and supports forecasting and fine-tuning, with a live demo available for BTC/USDT predictions.
- Pre-trained models (mini, small, base, large) are released on Hugging Face Hub, with varying capacities and context lengths.
- Forecasting is performed via the KronosPredictor class, handling preprocessing, normalization, and prediction, including batch predictions for multiple time series.
- Fine-tuning involves a complete pipeline using Qlib for data preparation and backtesting, with steps for tokenizer and predictor adaptation.
- The project includes scripts for data preprocessing, training, and evaluation, with a focus on Chinese A-share market data as an example.
- Research citations and licensing (MIT License) are provided, with acknowledgments of AI-generated code comments.