Possible missing files in Hugging Face train_dataset/mcut/mcut_ba_small

[53mins]

Hi Jiale,

Thanks for open-sourcing ML4CO-Bench-101 — it has been very helpful for reproduction.

I noticed a possible issue with the Hugging Face training data for MCut.

In the repository README, users are asked to download train_dataset from Hugging Face. However, under:

ML4CO-Bench-101-SL / train_dataset / mcut / mcut_ba_small

I can currently only find one file:

• mcut_ba-small_64k_1.txt

This seems incomplete, because the ML4CO-Bench-101 paper lists the MCut BA-SMALL training set size as 128,000 instances. Based on the current Hugging Face folder, only a single 64K shard appears to be available.

For comparison, mcut_ba_large seems to be uploaded as multiple shards (mcut_ba-large_16k_1.txt to mcut_ba-large_16k_8.txt), which looks consistent with a complete 128K training set. So I suspect that mcut_ba_small may be missing the remaining shard(s), such as a second 64K file, or another equivalent complete upload.

Could you please check whether the mcut_ba_small training dataset on Hugging Face is incomplete, and if so, upload the full version?

This would be very helpful for reproducing the MCut BA-SMALL experiments in the benchmark.

Thanks again for releasing the benchmark and the code.

Best,
53mins