Skip to content

v2.3.1

Choose a tag to compare

View all tags
@gaabrielrd gaabrielrd released this 04 Aug 21:10
· 20 commits to main since this release
v2.3.1
785d7ec

Fixes

  • Per-response token cap now follows the model context window. The LM Studio provider clamps max_tokens to the model's context (loaded instance context_length, falling back to max_context_length via /api/v1/models and /api/v0/models). Requests no longer ask for more output tokens than the model can produce, which previously caused responses to be rejected during processing (incomplete_response) before the server replied.
  • Actionable truncation diagnostics. An incomplete_response error now reports finish_reason, the requested max_tokens, and completion_tokens.

See docs/decisions/0012-derive-max-tokens-from-model-context.md. Full npm run validate green (411 tests).

Assets 3
Loading