Kronos-Tokenizer-base is the tokenizer component of Kronos, a foundation model developed for financial time series forecasting. It forms part of an open-source system that treats candlestick or K-line sequences as a specialized language for financial markets. The tokenizer quantizes continuous multi-dimensional financial data to enable modeling of high-noise market patterns.
The model was trained on data drawn from over 45 global exchanges. It employs a decoder-only architecture within a two-stage framework that begins with this specialized tokenizer. Documentation on the associated Hugging Face repository outlines steps for loading the tokenizer alongside the model, instantiating a predictor, preparing input data, generating forecasts, and performing finetuning on custom datasets such as A-Share market examples. An arXiv paper reference (2508.02739) accompanies the release.
It is delivered as a PyTorch and Safetensors model hosted on Hugging Face under the repository NeoQuasar/Kronos-Tokenizer-base. The project includes supporting materials such as a live demo, model zoo, getting-started instructions, and visualization examples. The license is MIT.
Kronos Tokenizer Base sits in PulseGate's Data science & ML workbench category. It provides a specialized model and tokenizer for forecasting and analyzing financial time series data. Kronos Tokenizer Base is an open-source project aimed at quantitative analysts. The project is open source (Open Source). Kronos Tokenizer Base is available on the command line, and it can be self-hosted.
It is developed by NeoQuasar, and the product first shipped in 2025. Among its 5 catalogued features are time series forecasting, financial data support, and pyTorch support.
Latest indexed changes and source events
NeoQuasar/Kronos-Tokenizer-base verified by the PulseGate indexer
Other apps tracked under the same category.
