Is Loss-free balancing superseded?

Loss-free balancing (Mixture-of-experts routing): superseded — cited as a baseline and beaten by newer methods. 1 paper(s) critique it, 1 beat it on benchmarks — #150 of 1370 most-superseded. Sub-problem: cluster led by MoE-LLaVA. Newer alternatives in the same sub-problem include MoE-GRPO, Input Domain Aware MoE.

Mixture-of-experts routing

Loss-free balancing

Superseded baseline#150 of 1,370 most-superseded

Superseded — cited as a baseline and beaten by newer methods

1 papers critique it · 1 beat it on benchmarks

What papers say

Verbatim critique sentences, each from a paper that cites Loss-free balancing as a baseline.

Loss-free routing exhibits the largest imbalance in activation frequencies, with a few experts over-selected and many under-used, indicating the strongest capacity imbalance and likely explaining its weaker accuracy

— Hierarchical Mixture-of-Experts with Two-Stage Optimization

Beaten on benchmarks

Head-to-head results where a newer method reports beating Loss-free balancing. Values are copied from the source paper's tables — verify against the cited paper.

Hi-MoE beats Loss-free balancing
2.947 vs 2.979
PPL · [nanoGPT on OpenWebText]
Hierarchical Mixture-of-Experts with Two-Stage Optimization

What to use instead

Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.