Add `NerModel.tokenizer`

2 jobs for main
in 9 minutes and 12 seconds, using 0.0 compute credits, and was queued for 5 seconds