[bong-water-water-bong]

tests/hrx_long_prompt.py sends enable_thinking: false at the top level of the request. llama-server ignores that field, so Qwen3 spends the probe's 16 tokens thinking (finish=length) and the 7K-token prompt returns '' on every pin. The probe can't serve as a gate in that state.

This keeps the top-level field and adds chat_template_kwargs: {"enable_thinking": false}, which llama-server honours, in both requests.

Evidence from the PR #393 gate on strixhalo, 2026-10-12, Qwen3-0.6B on HRX0:

• Unmodified probe: FAIL on the PR build and on main pins (0e255d77 / 98d05d94).
• Same prompt with the template kwargs: '7341' at 7094 tokens on both builds.
• Server started with --chat-template-kwargs '{"enable_thinking":false}': the probe passes 3/3 on both.

tests/hrx_long_prompt_test.py passes.

🤖 Generated with Claude Code

[github-actions]

PR Reviewer Guide 🔍

(Review updated until commit 62ff16a)

Here are some key observations to aid the review process:

🎫 Ticket compliance analysis 🔶 393 - Partially compliant Compliant requirements: Bump HRX submodules to llama.cpp 2bdb04a0928c and hrx-system 98d05d94a9f9 Run long-prompt correctness probe and test-backend-ops -b HRX0 on Strix Halo Record the exact validated hrx-system pin in config/hrx-long-prompt-validated-pin.config Validate that the probe passes with the updated configuration Non-compliant requirements: Ensure CI builds without HRX (this requirement is not directly verifiable from the diff, but implied by the PR description) Requires further human verification: CI build verification (not directly visible in diff)

⏱️ Estimated effort to review: 2 🔵🔵⚪⚪⚪

🧪 PR contains tests

🔒 No security concerns identified

⚡ Recommended focus areas for review Inconsistent enable_thinking handling The test adds chat_template_kwargs: {"enable_thinking": false} to override the top-level enable_thinking: False field, which suggests that the top-level field is not being respected by the server. This could indicate a potential inconsistency or misconfiguration in how the server handles the enable_thinking parameter, which should ideally be consistently processed at the top level. "enable_thinking": False, "chat_template_kwargs": {"enable_thinking": False}, Inconsistent enable_thinking handling The test adds chat_template_kwargs: {"enable_thinking": false} to override the top-level enable_thinking: False field, which suggests that the top-level field is not being respected by the server. This could indicate a potential inconsistency or misconfiguration in how the server handles the enable_thinking parameter, which should ideally be consistently processed at the top level. "enable_thinking": False, "chat_template_kwargs": {"enable_thinking": False},

[github-actions]

Persistent review updated to latest commit 62ff16a