d4data/bias-detection-model yielding large number of false positives

[AlejandroEsquivel]

There was a recent change to the validator as the previous version was broken and unmaintained

• Update validator, fixing a handful of imports that were incompatible with new guardrails. #1

As a result the underlying model was switched out in favour of d4data/bias-detection-model.
The new underlying model doesn't seem to classify adequately, albeit the sample size is negligible I suspect that benchmarks would support this claim

[JosephCatrambone]

For what it's worth, the underlying model is the same. DBias uses d4data/bias-detection-model internally. I suspect there's some kind of compatibility issue at play with how tokenization is working or maybe something else.