b10080

ggml-org/llama.cppb10080Jul 22, 2026by github-actions[bot]same day as b10079

What changed since b10079

low risk

Server now returns HTTP 400 instead of 500 for X-Conversation-Id validation failures, plus a UI fix for the backend_sampling placeholder.

Improvements

  • Failed requests carrying the X-Conversation-Id header no longer leave an empty finalized session behind for discovery or replay.
  • Session cleanup now also covers valid requests that include the header but never start streaming, avoiding stale entries until GC TTL.
  • UI settings sync now filters the backend_sampling placeholder the same way as neighboring fields, so default-profile completions validate correctly.

Bug fixes

  • Validation errors on the X-Conversation-Id header now surface as the intended 400 JSON response rather than a generic 500 caused by a bad_function_call crash in on_complete().
  • on_complete() treats an empty next_orig as 'streaming never started' and evicts the session installed by set_req().
  • Explicit false values for backend_sampling still reach the server, preserving the send-both-true-and-false behavior while only dropping the placeholder.

Upgrade notes

Clients that specifically matched on HTTP 500 for malformed X-Conversation-Id requests should now expect 400, which is the correct validation status. No config or migration steps are required; the session-eviction change is transparent to callers. Rebuild the web UI assets to pick up the backend_sampling filtering fix.

Analyzed Oct 29, 2026

Upgrading?

Release notes summary

Updates server error handling to return HTTP 400 instead of 500 for validation errors involving the X-Conversation-Id header, and fixes the UI to properly filter the backend_sampling placeholder.

Key Highlights

  • Server returns 400 on validation error instead of 500
  • UI fixes backend_sampling placeholder filtering
  • Prevents empty session artifacts on failed requests

New Features

  • Server error handling improvement
  • UI settings validation fix

Release Notes (excerpt)

<details open> server: return 400 instead of 500 on validation error with X-Conversation-Id (#25760) * server: return 400 instead of 500 on validation error with X-Conversation-Id set_req() attaches the spipe as soon as the header is present, before the request body is parsed. When params validation throws, set_next() never runs and next_orig stays empty, so on_complete() called it and crashed with std::bad_function_call, turning the prepared 400 JSON into a generic 500.…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories