From e5a07e25e31e34c70171136fdba45728e261ea32 Mon Sep 17 00:00:00 2001 From: bong-water-water-bong <277547417+bong-water-water-bong@users.noreply.github.com> Date: Wed, 11 Nov 2026 17:51:35 -0400 Subject: [PATCH] docs/hrx: the dispatch-flag paragraph the other two pages already carry, so this page stops teaching the flag bare (engine#2143) My first version of this branch said the opposite -- that GGML_HRX_LOG_DISPATCH is "not a behaviour-neutral instrument", with 271.75 vs 118.86 t/s as the evidence. That reading was refuted while the branch was open and engine#2143 is closed: at the engine's pin the flag only switches on two `fprintf` calls (dispatch_log_match() and dispatch_log_requested_output() in dispatch-registry.cpp), and the readback check plus its decline run in both arms, so the dispatches selected are the same either way. Those two numbers are one run per arm, box load moves a rate by about 2x here, and #1536 then failed in 4 of 6 runs -- so neither of them shows a flag effect. I would rather correct this in the branch than let a third page publish a warning the other two had already walked back. The paragraph here is docs/determinism.md's and docs/zyphra.md's text copied verbatim, not a paraphrase, so the three pages agree word for word and the claim in the intro sentence is true by construction. What remains true and useful: use the flag in BOTH arms of any A/B, and record box load beside any rate. --- docs/hrx.md | 11 +++++++++++ 1 file changed, 11 insertions(+) diff --git a/docs/hrx.md b/docs/hrx.md index 78d0acd20..85ffd0bfb 100644 --- a/docs/hrx.md +++ b/docs/hrx.md @@ -1107,6 +1107,17 @@ load per arm). the device leaves **no line at all**. Measured: two arms differing only in one tensor type's format had a **near-identical** registration set (one extra line, none missing) while `pp512` differed **5.4x** and `pp8` **8.1x**. So the histogram answers *"which kernel won"*, never *"did something fall back"*. +**The flag gates logging, not arithmetic (engine#2143), so set it in both arms of any A/B.** The two pages that also teach +this flag carry the same paragraph; this is the copy a reader here needs: + +> **Set it in both arms of an A/B, and record box load beside any rate (engine#2143).** #2143 asked whether this +> flag changes behaviour: one 18,782-token request failed without it and completed with it, and one 18.6k prefill +> measured 271.75 t/s unset against 118.86 t/s set. At the engine's pin the flag only gates `fprintf` calls +> (`dispatch_log_match()` and `dispatch_log_requested_output()` in `dispatch-registry.cpp`): the readback check and +> its decline run in both arms, so the dispatches selected are the same either way. Each of those measurements is +> one run per arm, #1536 then failed in 4 of 6 runs, and box load moves a rate by about 2x here, so neither shows +> a flag effect. Setting it in both arms is right whichever way #2143 settles. + Use the right instrument for each question — three are needed and the histogram is only the first: | question | instrument |