Lingua's high detection accuracy comes at the cost of being noticeably slower than other language detectors. The large language models also consume significant amounts of memory. These requirements might not be feasible for systems running low on resources.
For users who want to classify mostly long texts or need to save resources, a so-called low accuracy mode will be implemented that loads only a small subset of the language models into memory. The API will be as follows:
The downside of this approach is that detection accuracy for short texts consisting of less than 120 characters will drop significantly. However, detection accuracy for texts which are longer than 120 characters will remain mostly unaffected.
Lingua's high detection accuracy comes at the cost of being noticeably slower than other language detectors. The large language models also consume significant amounts of memory. These requirements might not be feasible for systems running low on resources.
For users who want to classify mostly long texts or need to save resources, a so-called low accuracy mode will be implemented that loads only a small subset of the language models into memory. The API will be as follows:
The downside of this approach is that detection accuracy for short texts consisting of less than 120 characters will drop significantly. However, detection accuracy for texts which are longer than 120 characters will remain mostly unaffected.