From 735b064f3ba2ceaae90b615da7a62bd5bf371661 Mon Sep 17 00:00:00 2001 From: Onur Tirtir Date: Fri, 16 Aug 2024 01:38:23 +0300 Subject: [PATCH] Correctly generate dataset filename in DatasetCompetitionFormat Since the way that BillionScaleDatasetCompetitionFormat and DatasetCompetitionFormat generates the dataset filename differs, we got the following error during the benchmarks: --- benchmark/datasets.py | 3 +++ 1 file changed, 3 insertions(+) diff --git a/benchmark/datasets.py b/benchmark/datasets.py index 76e7efdb..76bfe1e7 100644 --- a/benchmark/datasets.py +++ b/benchmark/datasets.py @@ -177,6 +177,9 @@ def prepare(self, skip_data=False, original_size=10**9): def get_dataset_fn(self): fn = os.path.join(self.basedir, self.ds_fn) + if self.nb != 10**9: + fn += '.crop_nb_%d' % self.nb + if os.path.exists(fn): return fn else: