Kronos: A foundation model for the language of financial markets.
What problem it solves
Financial candlestick sequences are continuous, multidimensional, and high-noise, making them difficult to represent with general-purpose time-series models. Kronos addresses this by quantizing OHLCV data into hierarchical discrete tokens and pre-training an autoregressive Transformer for quantitative forecasting tasks.
Key features
Financial candlestick tokenization — The tokenizer converts continuous OHLCV candlestick data into hierarchical discrete tokens for autoregressive modeling.
Source: README
Pre-trained model family — Kronos provides pre-trained models with different capacities, including Kronos-mini, Kronos-small, and Kronos-base, which are available through the Hugging Face Hub.
Source: README
Unified forecasting pipeline — The `KronosPredictor` class handles preprocessing, normalization, prediction, and inverse normalization from raw K-line data to forecasts.
Source: README
Batch and probabilistic forecasting — Users can forecast multiple time series in parallel and control sampling parameters such as temperature, nucleus sampling probability, and sample count.
Source: README
Fine-tuning and evaluation pipeline — The repository includes an A-share example covering data preparation, tokenizer and predictor fine-tuning, and backtesting, along with dedicated training and evaluation scripts.
Source: README, finetune, examples
Quick start
Copied verbatim from the project's README.
pip install -r requirements.txtfrom model import Kronos, KronosTokenizer, KronosPredictor
# Load from Hugging Face Hub
tokenizer = KronosTokenizer.from_pretrained("NeoQuasar/Kronos-Tokenizer-base")
model = Kronos.from_pretrained("NeoQuasar/Kronos-small")# Generate predictions
pred_df = predictor.predict(
df=x_df,
x_timestamp=x_timestamp,
y_timestamp=y_timestamp,
pred_len=pred_len,
T=1.0, # Temperature for sampling
top_p=0.9, # Nucleus sampling probability
sample_count=1 # Number of forecast paths to generate and average
)
print("Forecasted Data Head:")
print(pred_df.head())Who it may suit
Inference: May suit quantitative finance researchers and developers building K-line forecasting workflows or adapting a pre-trained model to domain-specific market data.
Topics to discuss
Which financial markets and candlestick frequencies should be prioritized for additional evaluation?
What applications and domain-specific tasks are best suited to hierarchical K-line tokenization?
How should the fine-tuning and backtesting example be extended for research beyond the A-share demonstration?
Written by ReadmeX AI from the project's README, source and public GitHub data; it may contain mistakes.
Read the original README https://github.com/shiyu-coder/Kronos#readme
ReadmeX opened this community for a trending project; it is not run by the project. Maintainers can claim it from the community page.
Comments
…