TimeBraid

Unifying time series and language for understanding and forecasting — paper · model · GitHub

Paste one series per line (comma-separated numbers), write what you want to know, and choose whether the model should explain the observed data or forecast its future. TimeBraid-2.5B braids a Qwen3-1.7B language backbone with a TimesFM 2.5 time-series expert, so the same numbers can read very differently depending on the context you give them.

Time series — one series per line, comma-separated values

Question or instruction

Forecast horizon

0 = explain the observed series; N > 0 = predict the next N values

0 64

Model response

Series and forecast

Example requests (from the TimeBraid repository and model card)

Time series — one series per line, comma-separated values	Question or instruction	Forecast horizon	System prompt (optional)

Pages:

All series and forecasts are plotted on their original scale. Forecasts come from greedy decoding on a fresh prompt, exactly as in the released inference recipe.