Format Requirements for Other Datasets

In addition to text, image, video, and audio datasets, the platform also supports the import of other types of datasets, that is, custom datasets used for model training.

• Import from OBS: The size of a single file or compressed package cannot exceed 20 GB. If multiple files are imported, the total file size cannot exceed 20 GB.
• Local upload: The size of a single file cannot exceed 1 GB, and a maximum of 20 files can be uploaded at a time.

The platform does not impose any rigid requirements on file content or nesting hierarchies. Users can choose to import data this way for specific or cutting-edge service scenarios outside of the platform's presets.

For example, in meteorological and weather data scenarios, datasets are primarily used for scientific training and fluid dynamics simulation of global or regional meteorological models. This scenario typically supports multi-dimensional grid scientific computing files such as .nc, .cdf, .netcdf, .gr, and .grib to house complex meteorological feature variables across different surface elevations and upper-air pressure levels within a global spatial grid. As another example, in time-series classification and time-series regression forecasting scenarios, datasets are deeply adapted for structured data tasks such as industrial sensor monitoring, IoT time-series signal analysis, and economic trend prediction. This scenario typically uses the .csv format for storage, where a core requirement is that the data must include at least one time-series baseline column with a fixed time interval, paired with continuous or discrete target prediction columns. All these scenarios can achieve data connectivity through custom methods.