I want library users to be able to construct supported scene-text recognizer models by public model id…

I want library users to be able to construct supported scene-text recognizer models by public model id…: a task in MiMo-V2.6-RL-harbor-code: MiMo-V2.6-RL Code (Harbor) (Harbor dataset). For create model('parseq', pretrained=False) , return a PARSeq model initialized from the shared defaults plus…

The task

For `create_model('parseq', pretrained=False)`, return a PARSeq model initialized from the shared defaults plus the `parseq` experiment/model config, with runtime defaults such as `decode_ar=True` and `refine_iters=1`. For `create_model('parseq-tiny', pretrained=False)`, return a PARSeq model whose resolved config…

Part of FineEnvs/MiMo-V2.6-RL-harbor-code.