[reedyalouh2]

Summary

An explicit thinking: {"type":"disabled"} on /v1/messages loses its false value during Anthropic-to-chat conversion. The handler then skips auto-enabling thinking, but a model such as Qwen3 still uses its thinking-enabled template default. This change carries enable_thinking: false through the existing mapping.

Details

Dynamo owns this translation; the selected model template consumes the control. Enabled, omitted, null, adaptive and unknown values keep their existing conversion behavior. History, tools, thinking budgets and response formatting are unchanged.

Six converter tests cover explicit controls, defaults and preserving the other request fields. An HTTP regression sends omitted, enabled and disabled controls through streaming and non-streaming /v1/messages and checks the canonical requests delivered to the scripted chat engine.

The October 7 Qwen3-8B experiment used an earlier frontend with this converter change: all 24 stock disabled requests generated reasoning; patched requests generated none. Those GPU results are separate from the current-source tests below. Disabling thinking can change answer quality.

Where should the reviewer start?

lib/llm/src/protocols/anthropic/types.rs, then lib/llm/tests/anthropic_http_replay.rs. The HTTP regression checks request delivery; it does not run a model template.

Validation

Base: c2c3338874cd170d80e5d5e9922c5f8f753e0fce, macOS arm64, Rust 1.96.1, --no-default-features.

• Stock plus the regressions: the two disabled-conversion checks fail; the HTTP check fails for disabled in both streaming and non-streaming requests.
• With the fix: all 114 Anthropic protocol tests and all 14 HTTP replay tests pass.
• cargo fmt --all --check and git diff --check pass.

cargo test --locked --offline -p dynamo-llm --no-default-features --lib protocols::anthropic:: -- --nocapture
cargo test --locked --offline -p dynamo-llm --no-default-features --test anthropic_http_replay -- --nocapture

Related Issues

Closes #16600.

#8782 previously addressed this toggle in a broader, unmerged change. #8485 covers response representation.

Summary by CodeRabbit

• New Features
  • Anthropic Messages requests now honor explicit thinking settings: enabled turns thinking on, and disabled turns it off.
  • When the setting is omitted, null, or unrecognized, no explicit thinking override is applied. The behavior is consistent for streaming and non-streaming requests.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[github-actions]

👋 Hi reedyalouh2! Thank you for contributing to ai-dynamo/dynamo.

Just a reminder: The NVIDIA Test Github Validation CI runs an essential subset of the testing framework to quickly catch errors.Your PR reviewers may elect to test the changes comprehensively before approving your changes.

🚀

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

• Configuration used: Repository: ai-dynamo/dynamo/.coderabbit.yaml
• Review profile: CHILL
• Plan: Enterprise
• Run ID: 288656d1-81c6-4317-ad01-54e257866351

📥 Commits

Reviewing files that changed from the base of the PR and between cdec3bd and f9be3c3.

📒 Files selected for processing: 2

• lib/llm/src/protocols/anthropic/types.rs
• lib/llm/tests/anthropic_http_replay.rs

Included review availability: This review used your included allowance. Your plan provides up to 12 included reviews per hour; 11 remain after this review.

Walkthrough

The Anthropic request converter now maps thinking.type values enabled and disabled to enable_thinking values true and false. New unit and HTTP replay tests cover these values and omitted or other configurations.

Changes

Anthropic thinking controls

Layer / File(s)	Summary
Convert thinking configuration lib/llm/src/protocols/anthropic/types.rs	The converter sets enable_thinking for enabled and disabled. Unit tests cover absent, null, and other values, and verify that disabling thinking does not change other converted request fields.
Verify HTTP request delivery lib/llm/tests/anthropic_http_replay.rs	A replay test checks omitted, enabled, and disabled thinking values in streaming and non-streaming requests. It also verifies that all six request scripts are consumed.

Priority: ➖ Normal

Estimated code review effort: 2 (Simple) | ~10 minutes

Severity of issue fixed: Medium

Merge Risk: ⚪ Minimal · up to f9be3

No actionable merge-blocking issue is identified; the change is ready for normal checks.

Pre-merge checks: 5

Check name	Status	Explanation
Title check		The title clearly and concisely identifies the main change: preserving explicit Anthropic thinking disablement.
Description check		The description explains the issue, implementation, scope, reviewer entry points, validation, and related issue. It uses “Summary” instead of the template’s “Overview” heading, but the required inform…
Linked Issues check		The PR satisfies the coding requirements in directly linked issue #16600. lib/llm/src/protocols/anthropic/types.rs preserves enable_thinking: false for thinking.type == "disabled" and retains th…
Out of Scope Changes check		The reported changes are limited to the Anthropic converter and its converter and HTTP regression tests. The tests directly verify issue #16600 behavior and preservation of unrelated request fields. N…
Docstring Coverage		Docstring coverage is 81.82% which is sufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 11 functions across 2 files.

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

_{Comment @coderabbitai help to get the list of available commands.}

[rmccorm4]

/ok to test 62406c0

[rmccorm4]

Thanks for the contribution @reedyalouh2 !