there's more details under the Too narrow and too wide tests heading. It would be interesting to see a deeper investigation, into how the models are dealing wi

there's more details under the Too narrow and too wide tests heading. It would be interesting to see a deeper investigation, into how the models are dealing with this and whether the successful ones seemed to be trained on the benchmark.