Datasets:
Sync EventXBench.py loading script: t3_test -> t3_train + t3_gold
Companion fix for the data update in discussion #8 (train.jsonl/gold.jsonl). The loading script hosted in this dataset repo still referenced t3_test, which no longer exists after #8 deleted the stale test.jsonl - load_dataset("t3") would currently fail. This syncs the script from GitHub PR mlsys-io/EventXBench#8 (code repo), which adds a NamedSplit("gold") split generator and points T3 at t3_train/t3_gold instead of t3_test.
Verified this only affects T3: ran _split_generators() directly against all 7 task configs with a stub downloader - t1/t2/t4/t5/t6/t7 splits are byte-identical to before (unchanged _URLS keys), only t3 changes (train+gold instead of test), since the new gold-split branch keys off "{config}_gold" which only exists in _URLS for t3.