b10273

ggml-org/llama.cppb10273Aug 4, 2026by github-actions[bot]same day as b10271

What changed since b10271

high risk

History-based samplers drop the full-context windows parameter, defaulting to a fixed 64-window size instead of the context length.

Breaking changes

  • The 'full-context windows' / context_size parameter is removed from history-based samplers, so configs or CLI flags that set it will no longer apply.
  • A value of -1 now resolves to 1024 rather than the model's context length, changing effective sampler behavior for existing setups.

Improvements

  • History-based samplers now share a single default of 64 windows, simplifying configuration.
  • Sampler construction no longer depends on knowing the resolved context length, which suits backend sampling where samplers are built before the full llama_context exists.

Bug fixes

  • Fixed sampler initialization failing when the complete context was not yet available at construction time.

Upgrade notes

Remove any full-context windows / context-size settings from sampler configuration, as the parameter no longer exists. Expect history-based samplers to use 64 windows by default, and note that -1 now maps to 1024 instead of the context length. If you relied on context-length-sized windows, set an explicit value to preserve prior behavior.

Analyzed Oct 31, 2026

Upgrading?

Release notes summary

Removes the 'full-context windows' parameter from history-based samplers.

Key Highlights

  • Removed context size parameter from samplers
  • Simplified sampler defaults to 64
  • Fixed initialization before complete context is available

Release Notes (excerpt)

<details open> sampler : remove "full-context windows" from history-based samplers (#26524) * Resolve -1 to 1024 instead of ctx-len for samplers Because of backend-sampling we initialize samplers before the complete llama_context is there. Therefore, we cannot infer the resolved context length yet at the time we construct the samplers. * Shared default of 64 for history-based samplers, remove context_size </details> **Website:** - <https://llama.app> **macOS/iOS:**…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories