Видео с ютуба Moe Cache

Внутри llama.cpp: как запускать огромные MoE-модели (новый кэш экспертов)

Worst clutch in the history of CS by mOE

llama.cpp PR #29887: GPU Cache for MoE Experts

The LRU Cache That Doubled Local MoE Speeds: llama.cpp Expert Offloading

llama.cpp's New MoE Expert Cache: Up To 2.2x Faster on One GPU

Janas: Run MoE Models Larger Than RAM from Your SSD?

Run 125B Models on 8GB Ram: llama.cpp Expert Cache

llama.cpp Just Made Huge MoE Models 4× Faster

llama.cpp Just Made 125B MoE Models 4× Faster!

Your MoE Model Is 2× Slower Without This llama.cpp Flag

How KV Cache Speeds Up LLMs for Faster AI Models on GPUs

llama.cpp's New MoE Cache Is 3× Faster on 16GB

В llama.cpp добавили Expert Cache... Победит ли он Strata на 12 ГБ?!

KV-кэш: трюк, который ускоряет LLM

The MoE Trap: Why 3B Active Doesn't Mean 3GB of VRAM (Qwen 35B)

Broken in 12 Days: How a 1993 Idea Fixed llama.cpp MoE Caching

llama.cpp увеличил скорость вашего GPU в 2 раза на больших моделях (включите это)

llama.cpp's New MoE Trick. Big AI Models on Small GPUs

freakazoid and moe rage

KV Cache - Explained