# llama.cpp b12549 fixes attention handling for Inkling2

> The update gives Inkling2 a dedicated relative-position attention input and fixes mask reuse for hybrid-memory models including LFM2 and LFM3.

Oossa · 2026-11-19 · https://oossa.com/en/llama-cpp-b12549-fixes-attention-handling-for-inkling2

The ggml-org team has released llama.cpp b12549 with fixes to attention-mask handling for Inkling2, a model supported by the project. The update gives Inkling2 a dedicated input that builds its dense attention mask and relative-position bias from the same cells. It also fixes hybrid-memory mask construction that prevented graph reuse with a unified cache and flash attention, affecting LFM2, LFM3 and Inkling2. The release additionally allows Inkling2’s expert-weight scale and final-logit softcap to be zero.

## The facts

- Release: llama.cpp b12549, published November 19, 2026.
- The mask-reuse fix covers hybrid-memory models LFM2, LFM3 and Inkling2.

## Why it matters

For people running these hybrid-memory models in llama.cpp, the fixes address a condition that stopped graph reuse with a unified cache and flash attention; the release notes do not quantify any speed improvement.

## Sources & references

1. [ggml-org/llama.cpp b12549](https://github.com/ggml-org/llama.cpp/releases/tag/b12549) – llama.cpp, 2026-11-19

Last updated: 2026-11-19
