Hot Expert Reload on GPU is what this community needs

A Reddit user suggests that Llama maintainers implement "Hot Expert Reload on GPU" to improve decode speed for Mixture-of-Experts (MOE) models. This feature would significantly enhance performance on GPUs like the 3090, making models such as Qwen3.8-Flash-Next, Deepseek V4/V4.1 Flash, and GLM 5.3 Flash more usable locally, with speeds approaching full VRAM offload when using multiple cards.