llama: add Maple 20B-A1B ternary MoE architecture (CPU) by AlexGabbia · Pull Request #27000 · ggml-org/llama.cpp

A new Maple 20B-A1B ternary MoE architecture for CPU is being added to llama.cpp, as seen in Pull Request #27000 by AlexGabbia. This development, which includes a preview available on Hugging Face at deepgrove/maple-preview, suggests a potential benefit for users with low VRAM.

This pull request marks the first time a ternary MoE architecture, specifically the Maple 20B-A1B, is being integrated into llama.cpp for CPU, unlike previous additions focused on other architectures.