Skip to content

Preserve thinking content in Tinker server chat templates#685

Merged
bradhilton merged 2 commits into
mainfrom
feat/preserve_thinking
May 19, 2026
Merged

Preserve thinking content in Tinker server chat templates#685
bradhilton merged 2 commits into
mainfrom
feat/preserve_thinking

Conversation

@bradhilton
Copy link
Copy Markdown
Collaborator

Summary

  • Pass preserve_thinking=True when the Tinker server renders a chat template that supports it.
  • Preserve the existing behavior of disabling generation-time thinking via enable_thinking=False when supported.

Test plan

  • uv run prek run --all-files passes ruff, ruff format, and uv.lock sync; ty fails only on unresolved optional Megatron/Triton dependencies (transformer_engine, triton, causal_conv1d, fla) plus the existing redundant-cast warning.
  • uv run ty check src/art/tinker/server.py

Made with Cursor

@bradhilton bradhilton merged commit 8d8ce60 into main May 19, 2026
5 checks passed
@bradhilton bradhilton deleted the feat/preserve_thinking branch May 19, 2026 18:21
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant