S-MoE
Datasets
All datasets matching “S-MoE”smoea-15ood-benchmark
SMoEA fixed 15-OOD benchmark data
The evaluation data for scripts/run_rejection_benchmark.py. Not needed for
interactive or batch mode — only the benchmark reads it.
Getting it
scripts/setup_workspace.sh downloads this for you into dataset/ood_data/,
which is what --benchmark-root defaults to. To fetch it on its own:
python scripts/fetch_benchmark.py --repo Tincan0325/smoea-15ood-benchmark
That verifies the per-group counts on arrival. The plain hf command works… See the full description on the dataset page: https://huggingface.co/datasets/Tincan0325/smoea-15ood-benchmark.SMoE-TrainingMagpie-Pro-300K-Filtered-CTC-512-1024-MoEPretrain_language_model-1BL3-smoePretrain_language_model-1BL3-smoe_sigmoidPretrain_language_model-1BL3-smoe_perturbed
