dinco1/moe-sparsity-scaling

MoE Sparsity Scaling — Checkpoints and Logs Artifacts for Selection Bias in Validation Loss Distorts Small-Scale Mixture-of-Experts Sparsity Studies. checkpoints/ — 42 runs. Optimizer state removed; model weights, config and metadata retained. Verified to reproduce the reported validation losses exactly. logs/ — training logs and config snapshots for every run. Naming: fineweb_edu_N{experts}_aux{coef}_cap{factor}[_frac{fraction}]_seed{seed} Code:… See the full description on the dataset page: https://huggingface.co/datasets/dinco1/moe-sparsity-scaling.

<iframe src="https://silytics.si/embed/item/aad72275-c9ed-4c43-813d-ccf63dff4555" width="420" height="190" style="border:0" loading="lazy" title="Silytics entry"></iframe>

Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 18 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.