Changes since UDPipe 1.1.0:
- On-demand loading of models in REST server, with a pool of least recently used models.
- Make GRU tokenizer dimension configurable (16, 24, 64 supported).
- Track paragraph boundaries even under
- Support experimental sentence segmentation using jointly both the tokenizer and the parser.
- Add EPE output format.
- Make default model in REST server explicit.
- Support pre-filling according to URL params in the webapp.