From adfc6a8421b686758ab26d4abcd0e896f2f8a803 Mon Sep 17 00:00:00 2001 From: agent Date: Sat, 7 Nov 2026 21:41:01 -0400 Subject: [PATCH] docs: stop citing the deleted import knob as the Zyphra page's evidence (Refs #1339) The paragraph in docs/zyphra-models.md that cleared up the CPU-side input GET_ROWS ended with "it is not the #856 host-pointer import, which GGML_HRX_WEIGHT_HOST_PLACEMENT=0 shows changes nothing here". That was true when written and is stale at this pin: the import was replaced (#234, pin ed43da0b4) and the variable no longer exists - docs/hrx.md already records "GGML_HRX_WEIGHT_HOST_PLACEMENT does nothing". A reader following the citation would look for a knob that is not in the tree. The claim itself still holds and is now stated so it survives the removal: while the import existed, the variable reproduced the placement identically, and the import is no longer in the tree at all, with hrx.md as the record. This is the qualifier the #1356 review asked for, taken as a rewrite rather than a "(at pin X)" footnote because the honest reading at this pin is that there is nothing left to toggle. Refs #1339 --- docs/zyphra-models.md | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/docs/zyphra-models.md b/docs/zyphra-models.md index a5a912d5e..a1dd8786e 100644 --- a/docs/zyphra-models.md +++ b/docs/zyphra-models.md @@ -123,5 +123,6 @@ input measured ~6 % slower than CPU input on a 0.6B while removing one node and So at default settings a census of `1 of 3024` nodes with `GET_ROWS x 1` and 2 graph splits is the floor for **every** model at `-ngl 99`, Zamba2 forms included, and it is not residue these changes added; `LLAMA_INPUT_LAYER_ON_DEV=1` -is how you leave it. (Engine **#1316** is where both halves were measured — it is not the #856 host-pointer import, -which `GGML_HRX_WEIGHT_HOST_PLACEMENT=0` shows changes nothing here.) +is how you leave it. (Engine **#1316** is where both halves were measured, and it is not the #856 host-pointer import: +while that import existed, `GGML_HRX_WEIGHT_HOST_PLACEMENT=0` reproduced the placement identically — and it has since +been replaced outright, so at this pin the import is not in the tree at all. [hrx.md](hrx.md) records that removal.)