Research map / Architectures and efficiency / Mixture of experts
DirMoE: Dirichlet-Routed Mixture of Experts
Amirhossein Vahidi, Hesam Asadollahzadeh, Navid Akhavan Attar, Marie Moullet, Kevin Ly, Xingyi Yang, Mohammad Lotfollahi
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace Top-k+Softmax (Switch) router with DirMoE | LLaMA-185M MoE · zero-shot average (7 tasks) | accuracy (%): 39.47 -> 41.13 | Improved |
| replace ReMoE router with DirMoE | LLaMA-185M MoE · zero-shot average (7 tasks) | accuracy (%): 40.03 -> 41.13 | Improved |
| replace Switch router with DirMoE | LLaMA-185M MoE · BoolQ | accuracy (%): 57.83 -> 61.52 | Improved |
| replace Switch router with DirMoE | LLaMA-185M MoE · LAMBADA | accuracy (%): 32.97 -> 36.44 | Improved |
| replace ReMoE router with DirMoE | LLaMA-185M MoE · ARC-e | accuracy (%): 46.68 -> 46.20 | Hurt |
| replace Switch router with DirMoE | LLaMA-185M MoE · training throughput | TFLOP/s/GPU: 138.2 -> 137.2 | No clear effect |
| replace Switch router with DirMoE | LLaMA-185M MoE · iteration time | ms: 431.5 -> 437.3 | No clear effect |
| add sparsity regularization R_sparsity (lambda_sparsity > 0) (ablation) | DirMoE · sparsity of active experts | desired sparsity | Improved |
| set lambda_sparsity to zero (ablation) | DirMoE · LLM loss | LM loss | Mixed |
| add Dirichlet concentration decrease (smaller lambda) (ablation) | DirMoE · Simpson index | sparsity: E[H] decreases strictly from 1 as lambda grows | Improved |
| add expert specialization via near-binary selection vs vanilla balanced MoE | DirMoE · domain specialization | expert specialization | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ReMoE — ReMoE: Fully Differentiable Mixture-of-Experts with ReLU Routing (ICLR 2025) · better on 6 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Routing Experts: Learning to Route Dynamic Experts in Existing Multi-modal Large Language Models (ICLR 2025 · similarity 0.91)
- Advancing Expert Specialization for Better MoE (NeurIPS 2025 · similarity 0.89)
- OLMoE: Open Mixture-of-Experts Language Models (ICLR 2025 · similarity 0.89)
- dMoE: dLLMs with Learnable Block Experts (NeurIPS 2026 · similarity 0.88)
- MoDE: A Mixture-of-Experts Model with Mutual Distillation among the Experts (AAAI 2024 · similarity 0.88)
- Pruning and Distilling Mixture-of-Experts into Dense Language Models (NeurIPS 2026 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
