Research
UrduQA: why we built a benchmark that models cannot have memorised
Static benchmarks decay the moment they enter training data. Here is how we built one that stays honest, and what it tells us about the state of Urdu language technology.
Dr. Bilal Ahmed