FAQ

Cloud service: input up to 11,520 time steps with up to 50 covariates; output up to 720 forecast points.

Private deployment: supports larger-scale input; the upper limit depends on your deployment configuration.

The Timer model family is continuously improved — newer versions deliver higher accuracy. We recommend the latest version, Timer-3.5, which performs best across a wide range of time-series tasks.

No. Timer is a deterministic model — given the same input, the output is stable and variations across runs are negligible. Forecast quality depends primarily on the quality and completeness of your input data. Ensure there are no missing values or significant outliers.

Yes. We provide fine-tuning services using scenario-specific data to improve forecasting performance in vertical domains or custom business contexts. Contact us if you need fine-tuning support.

401 means authentication failed. Check:

1. Whether your API key is correct (it is case-sensitive)
2. Whether your request includes Authorization: Bearer {API-Key}
3. Whether the API key has expired or been revoked

429 means you’ve hit the rate limit. Reduce your request frequency, or contact us to learn about higher-quota plans.

Try the following to improve accuracy:

1. Provide more historical data — a longer input window helps the model capture periodic patterns
2. Add relevant covariates — external variables correlated with the target series can improve precision
3. Check data quality — look for missing values, outliers, or irregular timestamps