b10514

ggml-org/llama.cppb10514Aug 20, 2026by github-actions[bot]same day as b10509

What changed since b10509

high risk

Adds GraniteSWA model architectures with Sliding Window Attention support.

New features

  • GraniteSWAForCausalLM model support
  • GraniteMoeSWAForCausalLM model support
  • Infrastructure for converting rope pattern arrays

Improvements

  • Added llama_hparams::has_rope and arch constants
  • Implemented per-layer rope determination logic
  • Added support for non-rope layers in SWA patterns

Bug fixes

  • Fixed SWA pattern logic and non-rope layer support
  • Fixed failing flake8 formatting

Upgrade notes

This release introduces new Granite model architectures utilizing Sliding Window Attention. Ensure conversion scripts are updated to handle the new rope_pattern arrays if converting models from other frameworks.

Analyzed Oct 20, 2026

Upgrading?

Release notes summary

Introduces support for new Granite model architectures featuring Sliding Window Attention (SWA), including GraniteSWAForCausalLM and GraniteMoeSWAForCausalLM, along with infrastructure for converting rope patterns.

Key Highlights

  • Added GraniteSWAForCausalLM model support
  • Added GraniteMoeSWAForCausalLM model support
  • Added conversion support for rope_pattern arrays
  • Implemented per-layer rope determination

Release Notes (excerpt)

<details open> model : GraniteSWAForCausalLM / GraniteMoeSWAForCausalLM (#25505) * feat(convert): Add conversion for GraniteSWAForCausalLM Branch: GraniteSWAForCausalLM AI-usage: full (Bob, OpenCode + Qwen3.6-35b) Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat(llama): Add granite_swa support Branch: GraniteSWAForCausalLM AI-usage: full (Bob, OpenCode + Qwen3.6-35b) Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat(conversion): Add conversion infra for rope_pattern array…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories