Autonomy-of-Experts Models
About
Mixture-of-Experts (MoE) models mostly use a router to assign tokens to specific expert modules, activating only partial parameters and often outperforming dense models. We argue that the separation between the router's decision-making and the experts' execution is a critical yet overlooked issue, leading to suboptimal expert selection and ineffective learning. To address this, we propose Autonomy-of-Experts (AoE), a novel MoE paradigm in which experts autonomously select themselves to process inputs. AoE is based on the insight that an expert is aware of its own capacity to effectively process a token, an awareness reflected in the scale of its internal activations. In AoE, routers are removed; instead, experts pre-compute internal activations for inputs and are ranked based on their activation norms. Only the top-ranking experts proceed with the forward pass, while the others abort. The overhead of pre-computing activations is reduced through a low-rank weight factorization. This self-evaluating-then-partner-comparing approach ensures improved expert selection and effective learning. We pre-train language models having 700M up to 4B parameters, demonstrating that AoE outperforms traditional MoE models with comparable efficiency.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Commonsense Reasoning | HellaSwag | -- | 1891 | |
| Commonsense Reasoning | WinoGrande | Accuracy50.2 | 1085 | |
| Question Answering | ARC Challenge | -- | 906 | |
| Question Answering | OpenBookQA | Normalized Accuracy25 | 102 | |
| Language Modeling | OpenWebText | Perplexity30 | 91 | |
| Question Answering | ARC Easy | Normalized Accuracy33.84 | 18 | |
| Commonsense Reasoning | PIQA | Normalized Accuracy56.09 | 13 | |
| Natural Language Understanding | GLUE | QQP Accuracy36.82 | 8 |