b10275

ggml-org/llama.cppb10275Aug 4, 2026by github-actions[bot]same day as b10274

What changed since b10274

low risk

b10275 fixes Windows server built-in tools losing accented characters by decoding OEM code page output to UTF-8.

Improvements

  • Built-in server tools now decode child-process output using the Windows OEM code page and convert it to UTF-8 before it reaches the JSON layer.
  • Text that already decodes as valid UTF-8 is passed through untouched, avoiding double-decoding for children that emit UTF-8.
  • Streamed chunks ending mid-character are handled by trimming incomplete trailing byte sequences before validation.

Bug fixes

  • Accented and other non-ASCII characters from built-in tool output on western Windows installs are no longer silently replaced with invalid bytes.
  • Decoding uses MB_ERR_INVALID_CHARS so a mismatched code page returns the original text rather than emitting replacement characters.

Upgrade notes

This is a narrow server-side fix with no API, config, or model-format changes, so a straight binary swap is sufficient. Windows users of the built-in tools are the main beneficiaries; other platforms should see no behavioral difference. The prior release's multi-modal audio chunk fix remains included.

Analyzed Oct 31, 2026

Upgrading?

Release notes summary

Fixes a bug where accented characters were silently lost on Windows due to OEM code page encoding issues in built-in tools.

Key Highlights

  • Decodes Windows OEM output to UTF-8 in built-in tools
  • Prevents invalid bytes from being silently replaced in JSON
  • Handles incomplete trailing byte sequences in streamed chunks

Release Notes (excerpt)

<details open> server: decode Windows OEM output to UTF-8 in built-in tools (#26597) a child process writes in the OEM code page, which is not UTF-8 on a western Windows install, so accented output reaches the JSON layer as invalid bytes and gets replaced there, silently losing the characters run() spawns without a console, so the child never inherits the console code page and GetOEMCP is the one that applies decode with MB_ERR_INVALID_CHARS so a wrong code page returns the text…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories