Tsinghua students' open-source financial prediction model can actually predict tomorrow's trend using K-line!

Kronos, the first open-source foundational model for financial K-line charts, trained on data from over 45 global exchanges. It’s not some mystical "predict ups and downs" thing; it genuinely treats K-line as a language to understand.
The core idea is really interesting:
Convert opening price, closing price, highest price, and lowest price into digital tokens, and use a Transformer large model to learn market patterns. K-line is the "language" of the financial market; once the model learns this language, it can understand trends.
It offers three versions: mini, small, and base, with the smallest only 4.1 million parameters, able to run on a single graphics card. There’s also an online demo where you input historical K-lines to directly predict the future 24-hour trend, including volatility probability.
Check out the real-time prediction of BTC/USDT, with historical trend (blue line) and predicted range (orange) for a pretty intuitive comparison. For quantitative trading or trading enthusiasts, this is worth exploring.
Previous GitHub collection 👉
Project link in the comments 👇