Why we are featuring it
Treating markets like a language.
Kronos applies hierarchical tokenizers and an autoregressive Transformer to OHLCV sequences. Its models come in different capacities for experimenting with financial forecasting and fine-tuning.
What is inside
- Trained on data from more than 45 global exchanges.
- Quantizes OHLCV K-line data into hierarchical discrete tokens.
- A model family from 4.1M to 499.2M parameters.
- Models and tokenizers are available on Hugging Face.