From 8e89ddfce2a08bd7fff48c37ae1eef05911d9029 Mon Sep 17 00:00:00 2001 From: agent Date: Tue, 10 Nov 2026 21:31:35 -0400 Subject: [PATCH] bump llama.cpp to 50d28600a: the pin on main was five commits BACKWARDS (restores #2033's regression) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit `origin/main` records gitlink `2293f95ea`, and that commit is an **ancestor** of `50d28600a`: $ gh api repos/1bit-MONSTER/llama.cpp/compare/2293f95ea...50d28600a status=behind ahead_by=0 behind_by=5 The gitlink history says how: 0f305ed bump llama.cpp to 50d28600a (fixes #1467) (#2011) 153b7b6 bump llama.cpp to 2293f95ea (fixes #1989) (#2033) so #2033 pinned five commits *before* #2011's pin, silently dropping fixes that had already landed: 50d2860 hrx: the multi-section rotation of Qwen3-VL runs on the device (engine#1467) fa533bd hrx: a view into storage the graph does not hold is skipped dfbf2b1 hrx: keep dense matmul activations inside the f16 range (engine#1962) a5dc597 ggml-hrx: a shared pool_tally pair (engine#1856's ledger arm) c47eae4 hrx: read the host term before classifying it (engine#1942) This restores the pin to `50d28600a`, which is the value `tools/check_pins.py` asks for — it is the check's own prescription ("point it at a commit that contains the old one"), not a new route decision, and it puts the five fixes back. Why this belongs to engine#1945 as well as to the pin: `tools/check_pins.py` already produced the right message on `main` before this PR, and **nothing gates a merge on it**, so a backwards pin lands green and only surfaces on the next scheduled `bump-hrx` sync — up to a day later, which is that issue's own title. Running it in CI on any PR that touches `third_party/llama.cpp` is the durable half and is not in this PR. Checks, run locally on this branch (`HEAD` = origin/main + these two changes): python3 tools/check_pins.py HEAD -> pass (fails on origin/main) python3 tools/registry_build.py --check-pins -> "records the committed pins" python3 tools/copyright.py --check -> pass The registry regenerated to a **one-line** change (recorded pin only; 298 architectures unchanged), because the five dropped commits changed behaviour, not the class list — which is also why a `--check-pins`-only review could not see the regression. --- registry/architectures.json | 2 +- third_party/llama.cpp | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/registry/architectures.json b/registry/architectures.json index ef1ebe7c4..031dafa8d 100644 --- a/registry/architectures.json +++ b/registry/architectures.json @@ -1,7 +1,7 @@ { "about": "HF architecture -> GGUF architecture and the backends whose code accepts it. Generated by tools/registry_build.py from the pinned sources; do not edit.", "sources": { - "llama.cpp (hrx)": "2293f95ea49dafec71a2460739d95ea31cd27bfc" + "llama.cpp (hrx)": "50d28600a5f7a38df390709b94c463df02785a54" }, "counts": { "hrx": 298, diff --git a/third_party/llama.cpp b/third_party/llama.cpp index 2293f95ea..50d28600a 160000 --- a/third_party/llama.cpp +++ b/third_party/llama.cpp @@ -1 +1 @@ -Subproject commit 2293f95ea49dafec71a2460739d95ea31cd27bfc +Subproject commit 50d28600a5f7a38df390709b94c463df02785a54