llama.cpp Release b11026 Improves Qwen3.5-MoE Handling

Core summary

One sentence to understand this update

llama.cpp's latest release, b11026, includes an update to skip `gate_up_exps` when `TENSOR_SKIP` is set, specifically benefiting Qwen3.5-MoE models with fused but unloaded MTP tensors.

Impact & opportunity

What this could mean

Developers using llama.cpp with Qwen3.5-MoE models might see improved stability or performance due to this specific tensor handling fix.