b9945

ggml-org/llama.cppb9945Jul 9, 2026by github-actions[bot]same day as b9941

What changed since b9941

low risk

b9945 fixes a SIGABRT crash when a chat template parses at init but fails at apply time.

New features

  • Chat-template thinking probe now runs inside the init try/catch, so apply-time template errors surface as load failures rather than aborts.

Improvements

  • Chat template error handling is now consistent between init-time parse failures and apply-time parser generation failures.

Bug fixes

  • Fixed an uncaught std::invalid_argument from common_chat_templates_support_enable_thinking() that caused llama-cli to abort (SIGABRT) instead of failing to load.
  • Models whose templates use constructs like {% call %} no longer crash the process during load.

Upgrade notes

No config or API changes are required; this is a drop-in replacement for b9941. Models with templates that fail at apply time will now report a load error instead of terminating the process, so scripts that relied on the abort behavior should handle the error path.

Analyzed Oct 24, 2026

Upgrading?

Release notes summary

Fixed crash when using chat templates that fail at apply time but pass at init time.

Key Highlights

  • Moved chat-template thinking probe inside init try/catch block
  • Prevents SIGABRT crash when template parsing fails at apply time
  • Ensures consistent error handling for chat templates

Release Notes (excerpt)

<details open> server : move chat-template thinking probe inside the init try/catch (#24093) A model whose chat template parses at init but fails parser generation at apply time (e.g. uses {% call %}) throws std::invalid_argument from common_chat_templates_support_enable_thinking(), which ran outside the try/catch guarding common_chat_templates_init(). The throw was uncaught and llama-cli aborted (SIGABRT) instead of failing to load. Moved the…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories