Smarter routing speeds up sparse AI models on memory-limited devices

Original: ReMoE: Boosting Expert Reuse through Router Fine-Tuning in Memory-Constrained MoE LLM Inference

Source: arxiv.org ↗