From 4278db4b20951a8a31422bddf99d77f2635a81a5 Mon Sep 17 00:00:00 2001
From: Andrew Mello <andrew@88plug.com>
Date: Thu, 29 Oct 2026 04:05:09 -0700
Subject: [PATCH 1/4] agent: vLLM XPU v0.32.0, and the model optimizer reads
 0.32.0

The vllm backend on an Intel GPU moves from vllm/vllm-openai-xpu v0.31.0 to
v0.32.0, the newest vLLM release on PyPI, GitHub and Docker Hub on 2026-10-29,
pinned to the tag's index digest 20509fab (linux/amd64 9dd913e9). The image
still ships Level Zero 1.32 and compute-runtime 26.27 (docker/Dockerfile.xpu at
v0.32.0) and still starts with an exec-form `vllm serve`.

The vLLM version the model optimizer is told follows the image: agent_optimize's
vver, plan.py's AGENT_VLLM_VERSION for a GPU and its fallback, and model_opt.py's
default. Every flag the agent manifest and the optimizer emit, the parsers they
name and the env vars in VLLM_XPU_ENV are present in v0.32.0's source. The
optimizer always sets num_speculative_tokens. Its vLLM source links now point at
v0.32.0, at the lines re-read there (--language-model-only is arg_utils.py L1460).

The README no longer says this backend is validated on Arc Pro B-series. It now
says what vLLM's XPU guide lists, and that v0.32.0 has not yet been run here on
a B-series card.
---
 README.md                  |  2 +-
 console/model_opt.py       | 10 +++++-----
 console/plan.py            |  4 ++--
 docs/intel/upstream-eat.md |  2 +-
 docs/platforms.md          |  2 +-
 images.lock                |  2 +-
 scripts/k3d-gpu            |  6 +++---
 7 files changed, 14 insertions(+), 14 deletions(-)

diff --git a/README.md b/README.md
index 3ad6475c..42b25cc7 100644
--- a/README.md
+++ b/README.md
@@ -944,7 +944,7 @@ OpenAI API at `/v1` from `svc/vllm`:
 
 | Backend | Image (pinned by digest) | Notes |
 |---------|--------------------------|-------|
-| `vllm` | `vllm/vllm-openai-xpu:v0.31.0` | upstream vLLM for Intel GPUs; validated on Arc Pro B-series |
+| `vllm` | `vllm/vllm-openai-xpu:v0.32.0` | upstream vLLM for Intel GPUs (Data Center GPU and Arc, per [vLLM's XPU guide](https://github.com/vllm-project/vllm/blob/v0.32.0/docs/getting_started/installation/gpu.xpu.inc.md)); v0.32.0 not yet run here on an Arc B-series card |
 | `llm-scaler` | `intel/llm-scaler-vllm:1.5` (vLLM 0.26.0-b2) | Intel's build for Arc B60/B70 |
 
 ```bash
diff --git a/console/model_opt.py b/console/model_opt.py
index ef20ea35..b6903bfa 100644
--- a/console/model_opt.py
+++ b/console/model_opt.py
@@ -94,9 +94,9 @@
 OVMS_SPEC_DEMO = OVMS_DOC + "demos/continuous_batching/speculative_decoding/README.md"
 OVMS_TOOL_PARSERS = ("llama3", "hermes3", "phi4", "mistral", "gptoss", "qwen3coder", "devstral", "lfm2", "gemma4", "onyx", "minicpm5")
 OVMS_REASONING_PARSERS = ("qwen3", "gemma4", "gptoss", "minicpm5", "lfm2", "onyx")
-VLLM_SPEC_DOC = "https://github.com/vllm-project/vllm/blob/v0.31.0/vllm/config/speculative.py"
-VLLM_MTP_DOC = "https://github.com/vllm-project/vllm/blob/v0.31.0/docs/features/speculative_decoding/mtp.md"
-VLLM_HW_TABLE = "https://github.com/vllm-project/vllm/blob/v0.31.0/docs/features/README.md (hardware table, L67)"
+VLLM_SPEC_DOC = "https://github.com/vllm-project/vllm/blob/v0.32.0/vllm/config/speculative.py"
+VLLM_MTP_DOC = "https://github.com/vllm-project/vllm/blob/v0.32.0/docs/features/speculative_decoding/mtp.md"
+VLLM_HW_TABLE = "https://github.com/vllm-project/vllm/blob/v0.32.0/docs/features/README.md (hardware table, L67)"
 VLLM_GAUDI_SD = "https://github.com/vllm-project/vllm-gaudi/blob/v0.30.0/docs/features/supported_features.md (L25, L27)"
 LLAMACPP_ARGS = "https://github.com/ggml-org/llama.cpp/blob/v0.8.0/common/arg.cpp (L4322 -md)"
 LLAMACPP_SPEC = "https://github.com/ggml-org/llama.cpp/blob/v0.8.0/common/arg.cpp (L4322 --spec-draft-model, L4330 --spec-type)"
@@ -709,7 +709,7 @@ def optimize(bundle, engine, device, opts=None):
     o = dict(opts or {})
     thinking_want = o.get("thinking") or "off"
     draft_want = o.get("draft") or "auto"
-    vllm_version = tuple(int(x) for x in (o.get("vllm_version") or "0.31.0").split(".")[:3])
+    vllm_version = tuple(int(x) for x in (o.get("vllm_version") or "0.32.0").split(".")[:3])
     prefix_caching = o.get("prefix_caching")
     hw_class = o.get("hw_class")
     target = bundle["target"]
@@ -976,7 +976,7 @@ def choose(fact, choice, source):
             doc["text_only"] = True
             doc["engine_args"].append("--language-model-only")
             choose("vision%s tower in config.json; agents send text" % ("/audio" if facts["audio"] else ""),
-                   "--language-model-only (no encoder memory, larger KV pool)", "https://github.com/vllm-project/vllm/blob/v0.31.0/vllm/engine/arg_utils.py (L1429, v0.17.0+)")
+                   "--language-model-only (no encoder memory, larger KV pool)", "https://github.com/vllm-project/vllm/blob/v0.32.0/vllm/engine/arg_utils.py (L1460, v0.17.0+)")
         elif engine == "ovms":
             choose("vision tower in the IR", "loaded as a VLM pipeline: OVMS refuses LM on a VLM directory",
                    OVMS_DOC + "src/llm/servable_initializer.cpp (L588-641)")
diff --git a/console/plan.py b/console/plan.py
index 0803c172..4a3dcda9 100755
--- a/console/plan.py
+++ b/console/plan.py
@@ -255,7 +255,7 @@ def plan(facts, matrix, overrides=None):
     return doc
 
 
-AGENT_VLLM_VERSION = {"HPU": "0.30.0", "GPU": "0.31.0"}  # the agent's pinned vLLM images
+AGENT_VLLM_VERSION = {"HPU": "0.30.0", "GPU": "0.32.0"}  # the agent's pinned vLLM images
 
 
 def agent_target(facts, doc, backend=None):
@@ -312,7 +312,7 @@ def add_optimizer(doc, facts, meta, opts):
     o = dict(opts or {})
     backend = o.pop("backend", None)
     engine, device, t = agent_target(facts, doc, backend)
-    o.setdefault("vllm_version", "0.26.0" if backend == "llm-scaler" else AGENT_VLLM_VERSION.get(device, "0.31.0"))
+    o.setdefault("vllm_version", "0.26.0" if backend == "llm-scaler" else AGENT_VLLM_VERSION.get(device, "0.32.0"))
     o["hw_class"] = hw_class(t)
     # No backend set: as `agent up`, the CPU engine follows the ledger (model_opt.py H59).
     o.setdefault("engine_from_ledger", backend is None)
diff --git a/docs/intel/upstream-eat.md b/docs/intel/upstream-eat.md
index f48ed990..14a76fe8 100644
--- a/docs/intel/upstream-eat.md
+++ b/docs/intel/upstream-eat.md
@@ -51,7 +51,7 @@ targets #19. What it changes for a k3d or k3d-gpu cluster, read from source only
 
 | #21 adds | Already covered by k3d-gpu | Expected on a k3d-gpu cluster |
 |---|---|---|
-| Arc B60/B70 XPU path (`compute_platform=xpu`, `vllm-openai-xpu:v0.28.0`, TP4 `gemma-4-31B-it`) | B70 detection, `gpu.intel.com/xe` asserted by `up`, vLLM XPU `v0.31.0` by digest through `agent up` | Host Level Zero and OpenCL libraries mounted by `hostPath` resolve inside the k3d node container, not on the host |
+| Arc B60/B70 XPU path (`compute_platform=xpu`, `vllm-openai-xpu:v0.28.0`, TP4 `gemma-4-31B-it`) | B70 detection, `gpu.intel.com/xe` asserted by `up`, vLLM XPU `v0.32.0` by digest through `agent up` | Host Level Zero and OpenCL libraries mounted by `hostPath` resolve inside the k3d node container, not on the host |
 | Intel GPU plugin from intel-device-plugins `main` (`intel_gpu_plugin_ref`, unpinned) | Plugin `0.37.1` baked into the node image | Two GPU plugin DaemonSets on one node |
 | NUMA-aware NRI balloons, one `BalloonsPolicy` per node | Baked balloons for hybrid P/E and multi-NUMA hosts; `K3D_GPU_BALLOONS=external` [ledger:H14] | Two policies if `deploy_nri_balloon_policy=yes`; `eat-up.sh` sets `off` |
 | Kata sandboxes (`kata-fc`, `kata-qemu`) | `doctor --perf` reports `/dev/kvm` on the host | Kata needs `/dev/kvm` inside the k3d node, which `up` does not pass through |
diff --git a/docs/platforms.md b/docs/platforms.md
index 380742f4..206d173b 100644
--- a/docs/platforms.md
+++ b/docs/platforms.md
@@ -323,7 +323,7 @@ B70 runs land:
   `/usr/lib/x86_64-linux-gnu/intel-opencl`).
 - What k3d-gpu already covers: B70 detection and the Intel GPU plugin `0.37.1`, baked
   and pinned; `gpu.intel.com/xe` advertised and asserted by `up`; the label through
-  `K3D_GPU_NODE_LABELS`; vLLM XPU through `K3D_GPU_AGENT_BACKEND=vllm` (`v0.31.0`, by digest).
+  `K3D_GPU_NODE_LABELS`; vLLM XPU through `K3D_GPU_AGENT_BACKEND=vllm` (`v0.32.0`, by digest).
 - Expected collision: a second GPU plugin. `deploy-intel-gpu-plugin.yml` applies
   `deployments/gpu_plugin/base/intel-gpu-plugin.yaml` from intel-device-plugins `main`
   (`intel_gpu_plugin_ref` defaults to `main`, unpinned) beside the baked DaemonSet.
diff --git a/images.lock b/images.lock
index 375cfbcb..41bfff5f 100644
--- a/images.lock
+++ b/images.lock
@@ -291,7 +291,7 @@
       "tag": "v0.32.0"
     },
     {
-      "digest": "sha256:95ac815038c4b3537b173798bfe3b89bdf525a52bd653cd14d4deb01f1062be7",
+      "digest": "sha256:20509fabe50dda62c09c0786d1a6effa299d3dbdddcc49331b93e65144652a8c",
       "name": "docker.io/vllm/vllm-openai-xpu",
       "sources": [
         "scripts/k3d-gpu AGENT_DEFAULT_IMAGE"
diff --git a/scripts/k3d-gpu b/scripts/k3d-gpu
index 760cd64e..bc1e60f1 100755
--- a/scripts/k3d-gpu
+++ b/scripts/k3d-gpu
@@ -7488,8 +7488,8 @@ agent_config() {
                 # vllm-0.30.0-ptupstream-2.13.0, pulled anonymously.
                 AGENT_DEFAULT_IMAGE="vault.habana.ai/gaudi-docker/1.24.2/ubuntu24.04/habanalabs/vllm-0.30.0-ptupstream-2.13.0@sha256:5d064465d751a73a729e7ecdc32c0ba5de0df1d305debb60a3d9a85b4fd3a967"
             elif [ "${AGENT_BACKEND}" = vllm ]; then
-                # vllm/vllm-openai-xpu:v0.31.0 (Level Zero 1.32 and compute-runtime 26.27 inside).
-                AGENT_DEFAULT_IMAGE="vllm/vllm-openai-xpu@sha256:95ac815038c4b3537b173798bfe3b89bdf525a52bd653cd14d4deb01f1062be7"
+                # vllm/vllm-openai-xpu:v0.32.0 (Level Zero 1.32 and compute-runtime 26.27 inside).
+                AGENT_DEFAULT_IMAGE="vllm/vllm-openai-xpu@sha256:20509fabe50dda62c09c0786d1a6effa299d3dbdddcc49331b93e65144652a8c"
             else
                 # intel/llm-scaler-vllm:1.5 = 0.26.0-b2 (vLLM 0.26 baseline, compute-runtime 26.14).
                 AGENT_DEFAULT_IMAGE="intel/llm-scaler-vllm@sha256:52218ad85513ab6686d4c090c83c2bd8c5b02423c63aa4dabd41837fe641fe3b"
@@ -11531,7 +11531,7 @@ agent_model_format() {
 # Applies the optimizer's choices for AGENT_OPT_MODEL to the agent settings.
 agent_optimize() {
     [ -n "${AGENT_OPT_MODEL}" ] || return 0
-    local dir bundle out errf engine=${AGENT_APP/#llamacpp/llama.cpp} device=${AGENT_SEL%% *} vver=0.31.0 draft=auto ledger meas
+    local dir bundle out errf engine=${AGENT_APP/#llamacpp/llama.cpp} device=${AGENT_SEL%% *} vver=0.32.0 draft=auto ledger meas
     local follow=() OPT_ENGINE="" OPT_MODEL="" OPT_REVISION="" OPT_TOOL_PARSER="" OPT_REASONING_PARSER="" OPT_SPEC=""
     local OPT_CONVERSION="" OPT_ARGS_JSON="[]" OPT_SUMMARY="" OPT_FALLBACK_MODEL="" OPT_FALLBACK_REVISION="" OPT_WARNING=""
     AGENT_MODEL=${AGENT_OPT_MODEL}

From 4dbb0cd9f0444e3fe59457b93f602c73dae0e7c6 Mon Sep 17 00:00:00 2001
From: Andrew Mello <andrew@88plug.com>
Date: Thu, 29 Oct 2026 04:05:22 -0700
Subject: [PATCH 2/4] facts: the llama.cpp Vulkan pin is release v0.8.0
 (b11960)

LLAMA_VULKAN_IMAGE, the llama.cpp pin facts/v1 reports and plan compares a
Vulkan cell's image against, moves from server-vulkan-v0.5.0 (be586c64) to
server-vulkan-b11960 (c9aba1a6). That is the multi-arch index (amd64, arm64)
built from release v0.8.0, the latest llama.cpp release: tag b11960, commit
8642541b, which the image's revision label also names. ghcr has published no
semver image tags since v0.5.0.

The fixture facts.json files, the console fallback goldens and the report.html
goldens are regenerated with each suite's --update mode. Each diff is the pin
line, plus each report's facts sha256. The synthetic Vulkan cells in
matrix-fallback.json and the xeon-b70 facts fixture move to the same pin, so the
fallback cells stay green.
---
 README.md                                          | 2 +-
 images.lock                                        | 2 +-
 scripts/k3d-gpu                                    | 6 +++---
 tests/console/fallback-json.golden                 | 6 +++---
 tests/console/fallback-vulkan.golden               | 2 +-
 tests/console/green-report.html                    | 4 ++--
 tests/console/matrix-fallback.json                 | 4 ++--
 tests/console/model-opt/xeon-b70.facts.json        | 2 +-
 tests/fixtures/alderlake-n/facts.json              | 2 +-
 tests/fixtures/alderlake-n/report.html             | 4 ++--
 tests/fixtures/arrowlake-h/facts.json              | 2 +-
 tests/fixtures/arrowlake-h/report.html             | 4 ++--
 tests/fixtures/arrowlake-s/facts.json              | 2 +-
 tests/fixtures/arrowlake-s/report.html             | 4 ++--
 tests/fixtures/arrowlake-u/facts.json              | 2 +-
 tests/fixtures/arrowlake-u/report.html             | 4 ++--
 tests/fixtures/battlemage-b580/facts.json          | 2 +-
 tests/fixtures/battlemage-b580/report.html         | 4 ++--
 tests/fixtures/battlemage-pro-b60/facts.json       | 2 +-
 tests/fixtures/battlemage-pro-b60/report.html      | 4 ++--
 tests/fixtures/battlemage-pro-b70-k616/facts.json  | 2 +-
 tests/fixtures/battlemage-pro-b70-k616/report.html | 4 ++--
 tests/fixtures/dg2-a770-i915/facts.json            | 2 +-
 tests/fixtures/dg2-a770-i915/report.html           | 4 ++--
 tests/fixtures/flex140-unbound/facts.json          | 2 +-
 tests/fixtures/flex140-unbound/report.html         | 4 ++--
 tests/fixtures/gaudi3-hls/facts.json               | 2 +-
 tests/fixtures/gaudi3-hls/report.html              | 4 ++--
 tests/fixtures/gaudi3-no-driver/facts.json         | 2 +-
 tests/fixtures/gaudi3-no-driver/report.html        | 4 ++--
 tests/fixtures/lunarlake-k6.8/facts.json           | 2 +-
 tests/fixtures/lunarlake-k6.8/report.html          | 4 ++--
 tests/fixtures/lunarlake/facts.json                | 2 +-
 tests/fixtures/lunarlake/report.html               | 4 ++--
 tests/fixtures/meteorlake/facts.json               | 2 +-
 tests/fixtures/meteorlake/report.html              | 4 ++--
 tests/fixtures/novalake/facts.json                 | 2 +-
 tests/fixtures/novalake/report.html                | 4 ++--
 tests/fixtures/npu-unserved/facts.json             | 2 +-
 tests/fixtures/npu-unserved/report.html            | 4 ++--
 tests/fixtures/pantherlake-k6.14/facts.json        | 2 +-
 tests/fixtures/pantherlake-k6.14/report.html       | 4 ++--
 tests/fixtures/pantherlake/facts.json              | 2 +-
 tests/fixtures/pantherlake/report.html             | 4 ++--
 tests/fixtures/pvc-max1550-unbound/facts.json      | 2 +-
 tests/fixtures/pvc-max1550-unbound/report.html     | 4 ++--
 tests/fixtures/qat-420xx/facts.json                | 2 +-
 tests/fixtures/qat-420xx/report.html               | 4 ++--
 tests/fixtures/raptorlake-nvidia/facts.json        | 2 +-
 tests/fixtures/raptorlake-nvidia/report.html       | 4 ++--
 tests/fixtures/raptorlake-stale-guc/facts.json     | 2 +-
 tests/fixtures/raptorlake-stale-guc/report.html    | 4 ++--
 tests/fixtures/wildcatlake/facts.json              | 2 +-
 tests/fixtures/wildcatlake/report.html             | 4 ++--
 tests/fixtures/xeon-6741p/facts.json               | 2 +-
 tests/fixtures/xeon-6741p/report.html              | 4 ++--
 tests/fixtures/xeon-dmr/facts.json                 | 2 +-
 tests/fixtures/xeon-dmr/report.html                | 4 ++--
 tests/fixtures/xeon-gnr/facts.json                 | 2 +-
 tests/fixtures/xeon-gnr/report.html                | 4 ++--
 tests/fixtures/xeon-gnrd/facts.json                | 2 +-
 tests/fixtures/xeon-gnrd/report.html               | 4 ++--
 tests/fixtures/xeon-max-hbm-cache/facts.json       | 2 +-
 tests/fixtures/xeon-max-hbm-cache/report.html      | 4 ++--
 tests/fixtures/xeon-max-hbm-devdax/facts.json      | 2 +-
 tests/fixtures/xeon-max-hbm-devdax/report.html     | 4 ++--
 tests/fixtures/xeon-max-hbm/facts.json             | 2 +-
 tests/fixtures/xeon-max-hbm/report.html            | 4 ++--
 tests/fixtures/xeon-spr-accel/facts.json           | 2 +-
 tests/fixtures/xeon-spr-accel/report.html          | 4 ++--
 tests/fixtures/xeon-srf/facts.json                 | 2 +-
 tests/fixtures/xeon-srf/report.html                | 4 ++--
 72 files changed, 110 insertions(+), 110 deletions(-)

diff --git a/README.md b/README.md
index 42b25cc7..79b213da 100644
--- a/README.md
+++ b/README.md
@@ -1193,7 +1193,7 @@ and submits through execlists, so a GuC fault there warns instead. A failed HuC
 only warns: it runs media decode and encode, not inference. Without root the GuC is reported as not checked. The refused GPU's
 facts/v1 target reads `usable: false` with the `reason` and `fix`; `plan` then
 falls back to the next green cell on a usable target (OVMS on the CPU, or
-llama.cpp on Vulkan, pinned as `ghcr.io/ggml-org/llama.cpp:server-vulkan-v0.5.0`
+llama.cpp on Vulkan, pinned as `ghcr.io/ggml-org/llama.cpp:server-vulkan-b11960` (release v0.8.0)
 by digest) and lists the fix. With nothing left to pick the decision is `fix`,
 not an engine; a Gaudi without its driver plans the same way. `agent up` follows
 the same verdict: `auto` uses OVMS on the CPU and says why, and
diff --git a/images.lock b/images.lock
index 41bfff5f..70e45a7f 100644
--- a/images.lock
+++ b/images.lock
@@ -316,7 +316,7 @@
       "tag": "v0.15.0"
     },
     {
-      "digest": "sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "digest": "sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "name": "ghcr.io/ggml-org/llama.cpp",
       "sources": [
         "scripts/k3d-gpu K3D_GPU_LLAMA_VULKAN_IMAGE"
diff --git a/scripts/k3d-gpu b/scripts/k3d-gpu
index bc1e60f1..8f4f50f1 100755
--- a/scripts/k3d-gpu
+++ b/scripts/k3d-gpu
@@ -7459,9 +7459,9 @@ AGENT_BACKEND="${AGENT_BACKEND,,}"
 AGENT_HPU=""    # set by agent_select when the device is a Gaudi
 AGENT_SEL=""    # "<device> <resource>", set by agent_select
 # llama.cpp server on Vulkan, the engine plan() may pick for a GPU cell other
-# than OVMS: ghcr.io/ggml-org/llama.cpp:server-vulkan-v0.5.0, the multi-arch
-# index built from release v0.5.0 (commit 7fe450e19305b828c199d602c23a8337aaa1f03b).
-LLAMA_VULKAN_IMAGE="${K3D_GPU_LLAMA_VULKAN_IMAGE:-ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744}"
+# than OVMS: ghcr.io/ggml-org/llama.cpp:server-vulkan-b11960, the multi-arch
+# index built from release v0.8.0 (tag b11960, commit 8642541b255bfb5dc39cd351a178be09013a55d6).
+LLAMA_VULKAN_IMAGE="${K3D_GPU_LLAMA_VULKAN_IMAGE:-ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e}"
 # The llama.cpp agent backend: release v0.8.0 (tag b11960, commit
 # 8642541b255bfb5dc39cd351a178be09013a55d6), the latest release on 2026-10-28; its image
 # tags are server-intel-b11960 (SYCL, Level Zero and compute-runtime 26.18 inside) for an
diff --git a/tests/console/fallback-json.golden b/tests/console/fallback-json.golden
index 6dedff0b..7b587eee 100644
--- a/tests/console/fallback-json.golden
+++ b/tests/console/fallback-json.golden
@@ -39,7 +39,7 @@
   "decision": "green",
   "engine": {
     "device": "Vulkan",
-    "image": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+    "image": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
     "name": "llama.cpp"
   },
   "explain": [
@@ -55,7 +55,7 @@
     "  synthetic.uhd770.ovms on gpu:0000:00:02.0: green (report synthetic-fallback-4, decode 5 tok/s); excluded: target gpu:0000:00:02.0 not usable: GuC i915/tgl_guc_70.1.1.bin 70.1.1 is older than the 70.5.1 i915 wants",
     "decision:",
     "  pick: synthetic.nvidia.llama-cpp-vulkan (decode 3 tok/s, highest of 2 green cell(s) on this host)",
-    "  engine: llama.cpp on Vulkan (ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744)",
+    "  engine: llama.cpp on Vulkan (ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e)",
     "  model: Qwen/Qwen3-4B-GGUF, q4_k_m",
     "  placement: gpu:nvidia via nvidia.com/gpu",
     "  endpoint: 127.0.0.1:8080, auth loopback-key (no login, gateway key in a 0600 file)",
@@ -88,7 +88,7 @@
   "schema": "k3d-gpu.plan/v1",
   "summary": [
     "pick: synthetic.nvidia.llama-cpp-vulkan (decode 3 tok/s, highest of 2 green cell(s) on this host)",
-    "engine: llama.cpp on Vulkan (ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744)",
+    "engine: llama.cpp on Vulkan (ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e)",
     "model: Qwen/Qwen3-4B-GGUF, q4_k_m",
     "placement: gpu:nvidia via nvidia.com/gpu",
     "endpoint: 127.0.0.1:8080, auth loopback-key (no login, gateway key in a 0600 file)",
diff --git a/tests/console/fallback-vulkan.golden b/tests/console/fallback-vulkan.golden
index f563a196..d147904f 100644
--- a/tests/console/fallback-vulkan.golden
+++ b/tests/console/fallback-vulkan.golden
@@ -10,7 +10,7 @@ matrix: 4 cells, 4 on a target of this host
   synthetic.uhd770.ovms on gpu:0000:00:02.0: green (report synthetic-fallback-4, decode 5 tok/s); excluded: target gpu:0000:00:02.0 not usable: GuC i915/tgl_guc_70.1.1.bin 70.1.1 is older than the 70.5.1 i915 wants
 decision:
   pick: synthetic.nvidia.llama-cpp-vulkan (decode 3 tok/s, highest of 2 green cell(s) on this host)
-  engine: llama.cpp on Vulkan (ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744)
+  engine: llama.cpp on Vulkan (ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e)
   model: Qwen/Qwen3-4B-GGUF, q4_k_m
   placement: gpu:nvidia via nvidia.com/gpu
   endpoint: 127.0.0.1:8080, auth loopback-key (no login, gateway key in a 0600 file)
diff --git a/tests/console/green-report.html b/tests/console/green-report.html
index 3cdbc5b3..44fee0e8 100644
--- a/tests/console/green-report.html
+++ b/tests/console/green-report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVXVNNI</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.CETIBT</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.CETSS</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.GFNI</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.SHA</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.VAES</th><td>true</td></tr><tr><th>gpu.intel.com/media-vcs</th><td>2</td></tr><tr><th>gpu.intel.com/media-vecs</th><td>1</td></tr><tr><th>gpu.intel.com/xmx</th><td>false</td></tr><tr><th>k3d-gpu.io/hybrid-cpu</th><td>true</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>a6264aff1b2e697cc0a9869e200366da82775053947dc046a2377e29f715f86b</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>054c57a66f59acbf44280b23d693d17a634b1a9ab433ea016de6812ffd8909da</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/console/matrix-fallback.json b/tests/console/matrix-fallback.json
index dc6d2c6b..d1986faf 100644
--- a/tests/console/matrix-fallback.json
+++ b/tests/console/matrix-fallback.json
@@ -44,7 +44,7 @@
       "device": "NVIDIA GPU, Vulkan",
       "engine": {
         "device": "Vulkan",
-        "image": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+        "image": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
         "name": "llama.cpp"
       },
       "evidence": {
@@ -83,7 +83,7 @@
       "device": "UHD 770, Vulkan",
       "engine": {
         "device": "Vulkan",
-        "image": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+        "image": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
         "name": "llama.cpp"
       },
       "evidence": {
diff --git a/tests/console/model-opt/xeon-b70.facts.json b/tests/console/model-opt/xeon-b70.facts.json
index f8a54702..4e168057 100644
--- a/tests/console/model-opt/xeon-b70.facts.json
+++ b/tests/console/model-opt/xeon-b70.facts.json
@@ -332,7 +332,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/alderlake-n/facts.json b/tests/fixtures/alderlake-n/facts.json
index 3e1a4f3b..84e6cc79 100644
--- a/tests/fixtures/alderlake-n/facts.json
+++ b/tests/fixtures/alderlake-n/facts.json
@@ -317,7 +317,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/alderlake-n/report.html b/tests/fixtures/alderlake-n/report.html
index 61d47ec4..c10c381e 100644
--- a/tests/fixtures/alderlake-n/report.html
+++ b/tests/fixtures/alderlake-n/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVXVNNI</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.CETIBT</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.GFNI</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.SHA</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.VAES</th><td>true</td></tr><tr><th>gpu.intel.com/xmx</th><td>false</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>997222876f770cd7d6a441e8136ec89ac5ea0edca44291c633bf8f9f040d6218</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>2b930cbab2911420b480d2b923c3dbb95c88201b46725e043e68ed9cf13952c1</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/arrowlake-h/facts.json b/tests/fixtures/arrowlake-h/facts.json
index 7f2f8163..6d5f132f 100644
--- a/tests/fixtures/arrowlake-h/facts.json
+++ b/tests/fixtures/arrowlake-h/facts.json
@@ -415,7 +415,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/arrowlake-h/report.html b/tests/fixtures/arrowlake-h/report.html
index 6f2c7af3..2c67e60a 100644
--- a/tests/fixtures/arrowlake-h/report.html
+++ b/tests/fixtures/arrowlake-h/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVXVNNI</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.GFNI</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.SHA</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.VAES</th><td>true</td></tr><tr><th>gpu.intel.com/xmx</th><td>true</td></tr><tr><th>k3d-gpu.io/hybrid-cpu</th><td>true</td></tr><tr><th>k3d-gpu.io/npu.gen</th><td>37xx</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>943e9246218f51063d59c9d64a48511772b55a3057d910ccd43d22bef6df5cd3</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>8532aa7e7bd7b67ce97a3755cb09013288658046f8157b5bc36f252af64c7ccb</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/arrowlake-s/facts.json b/tests/fixtures/arrowlake-s/facts.json
index d659ec56..4a35f064 100644
--- a/tests/fixtures/arrowlake-s/facts.json
+++ b/tests/fixtures/arrowlake-s/facts.json
@@ -418,7 +418,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/arrowlake-s/report.html b/tests/fixtures/arrowlake-s/report.html
index 8baba00b..a30d9b97 100644
--- a/tests/fixtures/arrowlake-s/report.html
+++ b/tests/fixtures/arrowlake-s/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVXVNNI</th><td>true</td></tr><tr><th>gpu.intel.com/media-vcs</th><td>2</td></tr><tr><th>gpu.intel.com/media-vecs</th><td>1</td></tr><tr><th>gpu.intel.com/xmx</th><td>false</td></tr><tr><th>k3d-gpu.io/hybrid-cpu</th><td>true</td></tr><tr><th>k3d-gpu.io/npu.gen</th><td>37xx</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>6729fb7553028e1e7c44140bfe9ccf06d54b6982c0d55c0238c881a49a54d2ff</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>ddb625631dd4de5919780361ffa6fd879f38947329be0186865e98b877ba1b74</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/arrowlake-u/facts.json b/tests/fixtures/arrowlake-u/facts.json
index 104dafed..186481b6 100644
--- a/tests/fixtures/arrowlake-u/facts.json
+++ b/tests/fixtures/arrowlake-u/facts.json
@@ -317,7 +317,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/arrowlake-u/report.html b/tests/fixtures/arrowlake-u/report.html
index 1921c477..ebbcdbcd 100644
--- a/tests/fixtures/arrowlake-u/report.html
+++ b/tests/fixtures/arrowlake-u/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVXVNNI</th><td>true</td></tr><tr><th>gpu.intel.com/xmx</th><td>false</td></tr><tr><th>k3d-gpu.io/hybrid-cpu</th><td>true</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>5b72a29804c7e7a62430e5728653e3e7b6e5d22410e2f326d5bc4f8b29a6110c</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>d40f851699cb171012acbe6a5511e9a1d83ccefbe2a36546c5ccedefeab67325</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/battlemage-b580/facts.json b/tests/fixtures/battlemage-b580/facts.json
index 7ffbd4f2..bb98b189 100644
--- a/tests/fixtures/battlemage-b580/facts.json
+++ b/tests/fixtures/battlemage-b580/facts.json
@@ -288,7 +288,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/battlemage-b580/report.html b/tests/fixtures/battlemage-b580/report.html
index fa740cea..ea7a601a 100644
--- a/tests/fixtures/battlemage-b580/report.html
+++ b/tests/fixtures/battlemage-b580/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512BF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512F</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512VNNI</th><td>true</td></tr><tr><th>gpu.intel.com/xmx</th><td>true</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>efa87c47c15e625248de2dfd10ab769bc1084c3e7bf3da13e9e0f1c0d1f7b17d</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>ec0ab6e41e8662ac11862a2beaa021cdecc4d15db3861849e7e4113d0eadc6c6</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/battlemage-pro-b60/facts.json b/tests/fixtures/battlemage-pro-b60/facts.json
index 2a29c0e5..c2824e9a 100644
--- a/tests/fixtures/battlemage-pro-b60/facts.json
+++ b/tests/fixtures/battlemage-pro-b60/facts.json
@@ -297,7 +297,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/battlemage-pro-b60/report.html b/tests/fixtures/battlemage-pro-b60/report.html
index ed88bd39..3dd4cac5 100644
--- a/tests/fixtures/battlemage-pro-b60/report.html
+++ b/tests/fixtures/battlemage-pro-b60/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512BF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512F</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512VNNI</th><td>true</td></tr><tr><th>gpu.intel.com/xmx</th><td>true</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>aefe022a88af3c9b7964f1d1e6598c783d96d5d2375a8f2b34779912944fe632</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>26a17082890ca4743e794b8d4bf1bae8458d83debe6d1d73af5a20328958d1e2</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/battlemage-pro-b70-k616/facts.json b/tests/fixtures/battlemage-pro-b70-k616/facts.json
index 6816b427..a766e272 100644
--- a/tests/fixtures/battlemage-pro-b70-k616/facts.json
+++ b/tests/fixtures/battlemage-pro-b70-k616/facts.json
@@ -233,7 +233,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/battlemage-pro-b70-k616/report.html b/tests/fixtures/battlemage-pro-b70-k616/report.html
index 87950992..f8d261ef 100644
--- a/tests/fixtures/battlemage-pro-b70-k616/report.html
+++ b/tests/fixtures/battlemage-pro-b70-k616/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512BF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512F</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512VNNI</th><td>true</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>95fa24025c25f57c8f5f64a620906a053e9b2f76080ea88b5347926117076bc4</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>bd676f6c5297b327cac5efe2eaa30cb446ac281ca3bad518f044341cff2021df</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/dg2-a770-i915/facts.json b/tests/fixtures/dg2-a770-i915/facts.json
index 3e1f8e81..5fb730a7 100644
--- a/tests/fixtures/dg2-a770-i915/facts.json
+++ b/tests/fixtures/dg2-a770-i915/facts.json
@@ -288,7 +288,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/dg2-a770-i915/report.html b/tests/fixtures/dg2-a770-i915/report.html
index b176765c..076b2472 100644
--- a/tests/fixtures/dg2-a770-i915/report.html
+++ b/tests/fixtures/dg2-a770-i915/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512BF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512F</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512VNNI</th><td>true</td></tr><tr><th>gpu.intel.com/xmx</th><td>true</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>a9e9041462ae7136e6ec5fe95df3dc53b3fd4f75b31d62695684be8e1b2d7a36</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>a6da1f00739e5d05847c5b0c2eb579e51746510b28e1a3ae7cce7aaa3986672b</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/flex140-unbound/facts.json b/tests/fixtures/flex140-unbound/facts.json
index a44e1f09..11b628ef 100644
--- a/tests/fixtures/flex140-unbound/facts.json
+++ b/tests/fixtures/flex140-unbound/facts.json
@@ -233,7 +233,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/flex140-unbound/report.html b/tests/fixtures/flex140-unbound/report.html
index 26c401d8..7129f903 100644
--- a/tests/fixtures/flex140-unbound/report.html
+++ b/tests/fixtures/flex140-unbound/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512BF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512F</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512VNNI</th><td>true</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>97f0218a901e1a285db0701d67c7ed56c056805153edc9fe7a9c47949f6d8cd9</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>c983e76aa4c8929ee6a71d53ddfc8948ac59b6fc37f793fdd64ca51d18e6e216</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/gaudi3-hls/facts.json b/tests/fixtures/gaudi3-hls/facts.json
index 812d9b65..50e62ac8 100644
--- a/tests/fixtures/gaudi3-hls/facts.json
+++ b/tests/fixtures/gaudi3-hls/facts.json
@@ -327,7 +327,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/gaudi3-hls/report.html b/tests/fixtures/gaudi3-hls/report.html
index 0f858f22..a7d864cd 100644
--- a/tests/fixtures/gaudi3-hls/report.html
+++ b/tests/fixtures/gaudi3-hls/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXBF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXINT8</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXTILE</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512BF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512F</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512VNNI</th><td>true</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>9889bf72233b7b3ee72880ba08e2f11d13cbe1650fefd77cc948270e36ef1270</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>d65aee54a4011e536eeb8b16d997fa9d72363403f7e2c1c0ad07d4e41801032f</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/gaudi3-no-driver/facts.json b/tests/fixtures/gaudi3-no-driver/facts.json
index 9a22fd5f..25ae844e 100644
--- a/tests/fixtures/gaudi3-no-driver/facts.json
+++ b/tests/fixtures/gaudi3-no-driver/facts.json
@@ -307,7 +307,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/gaudi3-no-driver/report.html b/tests/fixtures/gaudi3-no-driver/report.html
index ab4ccff2..5fc1fe7d 100644
--- a/tests/fixtures/gaudi3-no-driver/report.html
+++ b/tests/fixtures/gaudi3-no-driver/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXBF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXINT8</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXTILE</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512BF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512F</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512VNNI</th><td>true</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>2948b3e0b529a25d910108c126b2c2d72aec667c52c628df21aa9af35cb6d34f</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>966833f96e2ede1ebed22a48779de01a462adfc3b037b09ea56b4f0699a9e159</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/lunarlake-k6.8/facts.json b/tests/fixtures/lunarlake-k6.8/facts.json
index 94d32764..fdef4a29 100644
--- a/tests/fixtures/lunarlake-k6.8/facts.json
+++ b/tests/fixtures/lunarlake-k6.8/facts.json
@@ -324,7 +324,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/lunarlake-k6.8/report.html b/tests/fixtures/lunarlake-k6.8/report.html
index 159d1f14..cdbd5047 100644
--- a/tests/fixtures/lunarlake-k6.8/report.html
+++ b/tests/fixtures/lunarlake-k6.8/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVXVNNI</th><td>true</td></tr><tr><th>k3d-gpu.io/hybrid-cpu</th><td>true</td></tr><tr><th>k3d-gpu.io/npu.gen</th><td>40xx</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>529323b3c979d4e1dc015b737477f4c02fe21b66033e6819f90d6d390a9bb84c</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>668d338fa97bc522d22f9f4894e9172986f6b4b63619c1175c85f796a6291e55</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/lunarlake/facts.json b/tests/fixtures/lunarlake/facts.json
index 2fe2a39e..8c0e46f2 100644
--- a/tests/fixtures/lunarlake/facts.json
+++ b/tests/fixtures/lunarlake/facts.json
@@ -391,7 +391,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/lunarlake/report.html b/tests/fixtures/lunarlake/report.html
index d2f4ad11..76df00e1 100644
--- a/tests/fixtures/lunarlake/report.html
+++ b/tests/fixtures/lunarlake/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVXVNNI</th><td>true</td></tr><tr><th>gpu.intel.com/xmx</th><td>true</td></tr><tr><th>k3d-gpu.io/hybrid-cpu</th><td>true</td></tr><tr><th>k3d-gpu.io/npu.gen</th><td>40xx</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>114f92d1dd2e71352f1b077d283d28429d9b26796793eeceb96078957db357e6</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>6dad913789b0eb1ae10cfd3bf161eb4f53ed8d940817d6b7b04c40a4f7b91987</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/meteorlake/facts.json b/tests/fixtures/meteorlake/facts.json
index d739b945..8298d157 100644
--- a/tests/fixtures/meteorlake/facts.json
+++ b/tests/fixtures/meteorlake/facts.json
@@ -241,7 +241,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/meteorlake/report.html b/tests/fixtures/meteorlake/report.html
index fd7fad9e..6e5dcbf6 100644
--- a/tests/fixtures/meteorlake/report.html
+++ b/tests/fixtures/meteorlake/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVXVNNI</th><td>true</td></tr><tr><th>k3d-gpu.io/hybrid-cpu</th><td>true</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>7ef265013af2457d0afca4bab276fb0029bf74726e0602b6f08a2a9c38f523b3</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>5260ecb4bda99ec68bf1b2e9f63726d38b140d66ed4fa430322da3bada11e1a4</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/novalake/facts.json b/tests/fixtures/novalake/facts.json
index 020026c8..31d1d456 100644
--- a/tests/fixtures/novalake/facts.json
+++ b/tests/fixtures/novalake/facts.json
@@ -255,7 +255,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/novalake/report.html b/tests/fixtures/novalake/report.html
index d616372b..62248b68 100644
--- a/tests/fixtures/novalake/report.html
+++ b/tests/fixtures/novalake/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVXVNNI</th><td>true</td></tr><tr><th>k3d-gpu.io/hybrid-cpu</th><td>true</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>ffe94616ba43fed0424c4c2bfe86fb34909db880b746a30315e5e83d4c940a3d</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>cac5ad3082233ecad4b069c780e046f78b273b5921f4df98667ccf8a49bb2a2f</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/npu-unserved/facts.json b/tests/fixtures/npu-unserved/facts.json
index 3c517bed..a32050ff 100644
--- a/tests/fixtures/npu-unserved/facts.json
+++ b/tests/fixtures/npu-unserved/facts.json
@@ -399,7 +399,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/npu-unserved/report.html b/tests/fixtures/npu-unserved/report.html
index b7263238..0e262939 100644
--- a/tests/fixtures/npu-unserved/report.html
+++ b/tests/fixtures/npu-unserved/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVXVNNI</th><td>true</td></tr><tr><th>gpu.intel.com/xmx</th><td>true</td></tr><tr><th>k3d-gpu.io/hybrid-cpu</th><td>true</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>4740160249111b28e3efc55e4a31dbf8411a0099c423a6a3a68f0dbfa1103d38</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>e28d58bbb5d22637ec1745bc1341da9fb24e2bdfb60dc70318a4f251565083a7</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/pantherlake-k6.14/facts.json b/tests/fixtures/pantherlake-k6.14/facts.json
index a1ea8f42..9b6d2d87 100644
--- a/tests/fixtures/pantherlake-k6.14/facts.json
+++ b/tests/fixtures/pantherlake-k6.14/facts.json
@@ -333,7 +333,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/pantherlake-k6.14/report.html b/tests/fixtures/pantherlake-k6.14/report.html
index ad4539a0..56a9665e 100644
--- a/tests/fixtures/pantherlake-k6.14/report.html
+++ b/tests/fixtures/pantherlake-k6.14/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVXVNNI</th><td>true</td></tr><tr><th>k3d-gpu.io/hybrid-cpu</th><td>true</td></tr><tr><th>k3d-gpu.io/npu.gen</th><td>50xx</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>993f3b1d9481d312c9e9b6146d6e5195fa1c48149e56d18798f31c327dc6ff58</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>6f47e914244eb2826ee5b52720ab279f6a388cf90e7fff5d53720cd07500e5e7</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/pantherlake/facts.json b/tests/fixtures/pantherlake/facts.json
index 63df4ff4..50ee9b46 100644
--- a/tests/fixtures/pantherlake/facts.json
+++ b/tests/fixtures/pantherlake/facts.json
@@ -424,7 +424,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/pantherlake/report.html b/tests/fixtures/pantherlake/report.html
index edb4563e..9e4d05ab 100644
--- a/tests/fixtures/pantherlake/report.html
+++ b/tests/fixtures/pantherlake/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVXVNNI</th><td>true</td></tr><tr><th>gpu.intel.com/xmx</th><td>true</td></tr><tr><th>k3d-gpu.io/hybrid-cpu</th><td>true</td></tr><tr><th>k3d-gpu.io/npu.gen</th><td>50xx</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>dce345a289f7733100245e3495c8a2cbced52bccfb916abab3dec14b749fae6a</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>996be87a6dd3893eb0f3480f94908259a065a7bd013ae42a58b0a4733464e975</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/pvc-max1550-unbound/facts.json b/tests/fixtures/pvc-max1550-unbound/facts.json
index 1b2b5f6f..9e8af6fb 100644
--- a/tests/fixtures/pvc-max1550-unbound/facts.json
+++ b/tests/fixtures/pvc-max1550-unbound/facts.json
@@ -233,7 +233,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/pvc-max1550-unbound/report.html b/tests/fixtures/pvc-max1550-unbound/report.html
index 26b557e0..819c6e1f 100644
--- a/tests/fixtures/pvc-max1550-unbound/report.html
+++ b/tests/fixtures/pvc-max1550-unbound/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512BF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512F</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512VNNI</th><td>true</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>fb198b1e8cb4b6819a7f371d01c126371dae2face7a68f4dbfb1d40df5d1feed</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>c83b978ce6ddd252191e63c6ea1df8c1455fed1ba485a0d8a6d14b78398a4be5</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/qat-420xx/facts.json b/tests/fixtures/qat-420xx/facts.json
index e45b1265..b7a29ed2 100644
--- a/tests/fixtures/qat-420xx/facts.json
+++ b/tests/fixtures/qat-420xx/facts.json
@@ -301,7 +301,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/qat-420xx/report.html b/tests/fixtures/qat-420xx/report.html
index 33465ad1..5f392888 100644
--- a/tests/fixtures/qat-420xx/report.html
+++ b/tests/fixtures/qat-420xx/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXBF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXINT8</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXTILE</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512BF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512F</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512VNNI</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVXVNNI</th><td>true</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>49f130b6969839e1d7f99f973c56f0c6f73cf9f013bad8914190a43e2772f752</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>69cda846109fa15861ae5e640ba9c0a7f2b165b9e58c86d3223a179e38bc5e2c</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/raptorlake-nvidia/facts.json b/tests/fixtures/raptorlake-nvidia/facts.json
index 6ac8f4de..5c048252 100644
--- a/tests/fixtures/raptorlake-nvidia/facts.json
+++ b/tests/fixtures/raptorlake-nvidia/facts.json
@@ -937,7 +937,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/raptorlake-nvidia/report.html b/tests/fixtures/raptorlake-nvidia/report.html
index 05ef4aef..b7eb7297 100644
--- a/tests/fixtures/raptorlake-nvidia/report.html
+++ b/tests/fixtures/raptorlake-nvidia/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVXVNNI</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.CETIBT</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.CETSS</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.GFNI</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.SHA</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.VAES</th><td>true</td></tr><tr><th>gpu.intel.com/media-vcs</th><td>2</td></tr><tr><th>gpu.intel.com/media-vecs</th><td>1</td></tr><tr><th>gpu.intel.com/xmx</th><td>false</td></tr><tr><th>k3d-gpu.io/hybrid-cpu</th><td>true</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>a6264aff1b2e697cc0a9869e200366da82775053947dc046a2377e29f715f86b</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>054c57a66f59acbf44280b23d693d17a634b1a9ab433ea016de6812ffd8909da</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/raptorlake-stale-guc/facts.json b/tests/fixtures/raptorlake-stale-guc/facts.json
index f4f23e40..3f09d63a 100644
--- a/tests/fixtures/raptorlake-stale-guc/facts.json
+++ b/tests/fixtures/raptorlake-stale-guc/facts.json
@@ -943,7 +943,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/raptorlake-stale-guc/report.html b/tests/fixtures/raptorlake-stale-guc/report.html
index bbab3c46..61a0710c 100644
--- a/tests/fixtures/raptorlake-stale-guc/report.html
+++ b/tests/fixtures/raptorlake-stale-guc/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVXVNNI</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.CETIBT</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.CETSS</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.GFNI</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.SHA</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.VAES</th><td>true</td></tr><tr><th>gpu.intel.com/media-vcs</th><td>2</td></tr><tr><th>gpu.intel.com/media-vecs</th><td>1</td></tr><tr><th>gpu.intel.com/xmx</th><td>false</td></tr><tr><th>k3d-gpu.io/hybrid-cpu</th><td>true</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>21466bfd9dd3333f7ecbb7ad396faa5defbc7d4db37976bdac44cf81fd442989</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>49896a2c68669c84b35febc66b3c9735917e1344d96ad42ea06855f87c6b1377</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/wildcatlake/facts.json b/tests/fixtures/wildcatlake/facts.json
index bccc80dd..60d1aff5 100644
--- a/tests/fixtures/wildcatlake/facts.json
+++ b/tests/fixtures/wildcatlake/facts.json
@@ -324,7 +324,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/wildcatlake/report.html b/tests/fixtures/wildcatlake/report.html
index 4fd280e2..93b338d3 100644
--- a/tests/fixtures/wildcatlake/report.html
+++ b/tests/fixtures/wildcatlake/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVXVNNI</th><td>true</td></tr><tr><th>k3d-gpu.io/hybrid-cpu</th><td>true</td></tr><tr><th>k3d-gpu.io/npu.gen</th><td>50xx</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>04c3391372d9cb24b09c25eea2ab93973312c3a67a9cb6add1d4d7a5bea92329</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>7df3056a544fc9016caf3591f3cfe7b7c6ffae5519eb37bc53e8961a54706252</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/xeon-6741p/facts.json b/tests/fixtures/xeon-6741p/facts.json
index 34895546..e9b7a53d 100644
--- a/tests/fixtures/xeon-6741p/facts.json
+++ b/tests/fixtures/xeon-6741p/facts.json
@@ -700,7 +700,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/xeon-6741p/report.html b/tests/fixtures/xeon-6741p/report.html
index da7b177c..1e949e2b 100644
--- a/tests/fixtures/xeon-6741p/report.html
+++ b/tests/fixtures/xeon-6741p/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXBF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXFP16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXINT8</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXTILE</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512BF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512F</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512VNNI</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVXVNNI</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.CETIBT</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.CETSS</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.GFNI</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.SHA</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.VAES</th><td>true</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>62688db250cf0d52cf945ebad71ab4d8dba0cecc04ff1f4f57722d6dcbff8293</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>60ec986919cd95d11ba63f06ba046f83e5aadaba18131fd1fa8afcf64dffb295</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/xeon-dmr/facts.json b/tests/fixtures/xeon-dmr/facts.json
index 50ea97b9..73e79d8f 100644
--- a/tests/fixtures/xeon-dmr/facts.json
+++ b/tests/fixtures/xeon-dmr/facts.json
@@ -357,7 +357,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/xeon-dmr/report.html b/tests/fixtures/xeon-dmr/report.html
index 881a2919..e47fa548 100644
--- a/tests/fixtures/xeon-dmr/report.html
+++ b/tests/fixtures/xeon-dmr/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXBF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXCOMPLEX</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXFP16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXINT8</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXTILE</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512BF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512F</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512VNNI</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVXVNNI</th><td>true</td></tr><tr><th>k3d-gpu.io/numa-nodes</th><td>6</td></tr><tr><th>k3d-gpu.io/snc</th><td>3</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>2703565114bbde919ff02657014e870ca0db36afb54e42d761bf312d473b476c</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>cda4f02db0079e13cec29eb5fcaf42f2097d3697cd5e8aee855f3052ddacf2e2</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/xeon-gnr/facts.json b/tests/fixtures/xeon-gnr/facts.json
index d7599783..bf3a407f 100644
--- a/tests/fixtures/xeon-gnr/facts.json
+++ b/tests/fixtures/xeon-gnr/facts.json
@@ -341,7 +341,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/xeon-gnr/report.html b/tests/fixtures/xeon-gnr/report.html
index c71026bb..2a593561 100644
--- a/tests/fixtures/xeon-gnr/report.html
+++ b/tests/fixtures/xeon-gnr/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXBF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXFP16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXINT8</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXTILE</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512BF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512F</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512VNNI</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVXVNNI</th><td>true</td></tr><tr><th>k3d-gpu.io/numa-nodes</th><td>6</td></tr><tr><th>k3d-gpu.io/snc</th><td>3</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>dffc0c3645de800b96e086bf6b987463dbce58ba0d8ed8745deb89e331665355</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>922bb3ba1b40901aad7237fbdefc6fa9e2bf084813ee38858255d60084cb9e3f</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/xeon-gnrd/facts.json b/tests/fixtures/xeon-gnrd/facts.json
index 2138a41b..742481e6 100644
--- a/tests/fixtures/xeon-gnrd/facts.json
+++ b/tests/fixtures/xeon-gnrd/facts.json
@@ -283,7 +283,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/xeon-gnrd/report.html b/tests/fixtures/xeon-gnrd/report.html
index 3b9faa9e..acbdf474 100644
--- a/tests/fixtures/xeon-gnrd/report.html
+++ b/tests/fixtures/xeon-gnrd/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXBF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXCOMPLEX</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXFP16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXINT8</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXTILE</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512BF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512F</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512VNNI</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVXVNNI</th><td>true</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>82492c489cc27cb08bcf85e92798fc9eceb7f4dfb59fbf6c94d14b9d520b0a95</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>f85133b66c65ba456cfa2e54ffebb478b900b5fde57bcf0b88a45897c2f66d30</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/xeon-max-hbm-cache/facts.json b/tests/fixtures/xeon-max-hbm-cache/facts.json
index 0a6ff99b..ec317f91 100644
--- a/tests/fixtures/xeon-max-hbm-cache/facts.json
+++ b/tests/fixtures/xeon-max-hbm-cache/facts.json
@@ -289,7 +289,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/xeon-max-hbm-cache/report.html b/tests/fixtures/xeon-max-hbm-cache/report.html
index 6e57d19f..ca45f5c0 100644
--- a/tests/fixtures/xeon-max-hbm-cache/report.html
+++ b/tests/fixtures/xeon-max-hbm-cache/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXBF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXINT8</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXTILE</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512BF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512F</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512VNNI</th><td>true</td></tr><tr><th>k3d-gpu.io/numa-nodes</th><td>4</td></tr><tr><th>k3d-gpu.io/snc</th><td>4</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>b7c3c93a281556c1f7bf62b038922647952d0d73bf877366e2656bc43a89a044</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>e91bd3cfb1e9fbe450ee21042931894e3f99659092ebb1cd615b8bdc0071995a</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/xeon-max-hbm-devdax/facts.json b/tests/fixtures/xeon-max-hbm-devdax/facts.json
index 83798a2f..0b1d1a25 100644
--- a/tests/fixtures/xeon-max-hbm-devdax/facts.json
+++ b/tests/fixtures/xeon-max-hbm-devdax/facts.json
@@ -280,7 +280,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/xeon-max-hbm-devdax/report.html b/tests/fixtures/xeon-max-hbm-devdax/report.html
index bd6c1a8b..12534665 100644
--- a/tests/fixtures/xeon-max-hbm-devdax/report.html
+++ b/tests/fixtures/xeon-max-hbm-devdax/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXBF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXINT8</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXTILE</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512BF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512F</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512VNNI</th><td>true</td></tr><tr><th>k3d-gpu.io/numa-nodes</th><td>4</td></tr><tr><th>k3d-gpu.io/snc</th><td>4</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>0f5fa785c7fa71806cd93b3e83e878760d3aa8f3f3353ebc8cefaa59c8f692fb</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>3c25569f4da4ea55051255d0126460d91ccefb9a433edbfcfca7ec129fd85589</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/xeon-max-hbm/facts.json b/tests/fixtures/xeon-max-hbm/facts.json
index 5924238b..df2a2584 100644
--- a/tests/fixtures/xeon-max-hbm/facts.json
+++ b/tests/fixtures/xeon-max-hbm/facts.json
@@ -280,7 +280,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/xeon-max-hbm/report.html b/tests/fixtures/xeon-max-hbm/report.html
index 2628b379..b4d9222f 100644
--- a/tests/fixtures/xeon-max-hbm/report.html
+++ b/tests/fixtures/xeon-max-hbm/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXBF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXINT8</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXTILE</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512BF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512F</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512VNNI</th><td>true</td></tr><tr><th>k3d-gpu.io/numa-nodes</th><td>8</td></tr><tr><th>k3d-gpu.io/snc</th><td>4</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>48132fe8bf0c1d50900f2bd35888fc19be767e2b3a8f2cd7206225458e4dccf8</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>652e964a1a73480e6c1d2b64e770173433890efe380a4637590fb35918a9540f</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/xeon-spr-accel/facts.json b/tests/fixtures/xeon-spr-accel/facts.json
index 4d312d0d..fc30d70c 100644
--- a/tests/fixtures/xeon-spr-accel/facts.json
+++ b/tests/fixtures/xeon-spr-accel/facts.json
@@ -350,7 +350,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/xeon-spr-accel/report.html b/tests/fixtures/xeon-spr-accel/report.html
index 5330a26b..caaaeeac 100644
--- a/tests/fixtures/xeon-spr-accel/report.html
+++ b/tests/fixtures/xeon-spr-accel/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXBF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXINT8</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AMXTILE</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512BF16</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512F</th><td>true</td></tr><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVX512VNNI</th><td>true</td></tr><tr><th>k3d-gpu.io/numa-nodes</th><td>2</td></tr><tr><th>k3d-gpu.io/snc</th><td>1</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>ae88c40d35ba8d59f7ed8acb7c9fc9000bbd503779ea76df31f1b56568888170</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>d575e4efe5d42a9f117402d4ba92c9408fca8aa178ee492fe7a00068fbb4190f</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>
diff --git a/tests/fixtures/xeon-srf/facts.json b/tests/fixtures/xeon-srf/facts.json
index e044a244..53c7abf0 100644
--- a/tests/fixtures/xeon-srf/facts.json
+++ b/tests/fixtures/xeon-srf/facts.json
@@ -264,7 +264,7 @@
   "release": {
     "k3d_gpu_version": "<version>",
     "pins": {
-      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744",
+      "llama.cpp": "ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e",
       "ovms": "openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58"
     }
   },
diff --git a/tests/fixtures/xeon-srf/report.html b/tests/fixtures/xeon-srf/report.html
index 2ad735a4..8d37a6f9 100644
--- a/tests/fixtures/xeon-srf/report.html
+++ b/tests/fixtures/xeon-srf/report.html
@@ -67,9 +67,9 @@ <h2>Detected</h2>
 <h2>Node labels</h2>
 <div class="wrap"><table><tbody><tr><th>feature.node.kubernetes.io/cpu-cpuid.AVXVNNI</th><td>true</td></tr></tbody></table></div>
 <h2>Release pins</h2>
-<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:be586c648cb79d2cd75d0c38a6541e1f16c47a7c6621d3e711470a467fc38744</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
+<div class="wrap"><table><tbody><tr><th>llama.cpp</th><td>ghcr.io/ggml-org/llama.cpp@sha256:c9aba1a673382ad356a72fb63e6cd73e8c021797e904b6c099bfe130217df62e</td></tr><tr><th>ovms</th><td>openvino/model_server@sha256:a928903fe429f1e3e27107cc45932d30d95febfd2d1159dc50869c3eac76fc58</td></tr></tbody></table></div>
 <footer>
-<p>Rendered from facts/v1, sha256 of its canonical bytes <code>0e26d0a8e3e193601c75823ad2d2a338e2a4cd5e3f1fd198e0acffb91e5dfaf7</code>, and plan/v1.</p>
+<p>Rendered from facts/v1, sha256 of its canonical bytes <code>99404eb97e388bf6f49faa070d8723f9ebdfa30643b43b51a12cd2d30b6fbbaa</code>, and plan/v1.</p>
 <p>This page runs no script and loads nothing: its Content-Security-Policy is <code>default-src 'none'</code>. The links are references to follow, not resources it fetches.</p>
 <p><a href="https://github.com/88plug/k3d-gpu" rel="noreferrer">https://github.com/88plug/k3d-gpu</a></p>
 </footer></main></body></html>

From 8d246a90525cb1cafa44fabf894cc049a3a9d924 Mon Sep 17 00:00:00 2001
From: Andrew Mello <andrew@88plug.com>
Date: Thu, 29 Oct 2026 04:05:22 -0700
Subject: [PATCH 3/4] plan --predict: llama.cpp at release v0.8.0 (b11960)

moe_predict's llama.cpp engine names the agent's own llama.cpp images,
server-b11960 (8052c13e) and server-intel-b11960 (068cd41c), in place of the
b11749 build the H54 engine scout read. b11749 to b11960 is ahead only.
---
 console/moe_predict.py | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/console/moe_predict.py b/console/moe_predict.py
index ebf94bbe..e88420af 100644
--- a/console/moe_predict.py
+++ b/console/moe_predict.py
@@ -55,6 +55,8 @@
 
 # Engines per weight container, as the H54 engine scout verified them at the
 # pinned versions (EXPERIMENTS.md H54 pre-registration). Order is preference.
+# llama.cpp names the agent's own llama.cpp pins (release v0.8.0, tag b11960),
+# newer than the b11749 build the scout read.
 ENGINES = [
     {"name": "ovms", "container": "ov", "devices": ("cpu", "gpu"), "tp": False, "formats": None,
      "images": ("openvino/model_server:2026.4.1@sha256:148e3fe802261ed0b5b634de3bbccdefe7176a218ec34a88e382b40704c34a1c",
@@ -66,8 +68,8 @@
     {"name": "sglang-cpu", "container": "safetensors", "devices": ("cpu",), "tp": False,
      "formats": ("bf16", "fp8", "w4a16"), "images": ("lmsysorg/sglang:v0.5.22-xeon@sha256:4e5c413ed9a31f38f74e18aa9b645d65a0b0a023d42f63edb4e2b412ea1a230f",)},
     {"name": "llama.cpp", "container": "gguf", "devices": ("cpu", "gpu", "hybrid"), "tp": False,
-     "formats": None, "images": ("ghcr.io/ggml-org/llama.cpp:server-b11749@sha256:9fbaa06d7f4449e2ae0d02b40e0431f7c212dcf231c9e3a7777a6afc2cc8f312",
-                "ghcr.io/ggml-org/llama.cpp:server-intel-b11749@sha256:ae2d41ba73574a1e12cd237c88b512c8bb39ba6f703e873912b76cb8988a81bd")},
+     "formats": None, "images": ("ghcr.io/ggml-org/llama.cpp:server-b11960@sha256:8052c13edd987de8858f8150031e025ac3d66823f09cf3ec8220899c4abce220",
+                "ghcr.io/ggml-org/llama.cpp:server-intel-b11960@sha256:068cd41cdc292bed89b4a6b7740ea7a33df8ba26bf880d0e12ac75972dce1700")},
 ]
 # Architectures with a narrower engine set than their container suggests.
 ARCH_ENGINES = {"Qwen4ExpForConditionalGeneration": ("llama.cpp",), "qwen4exp": ("llama.cpp",)}

From 9ddb333da44e0263a5f4752fc1abe9fd9ed32a01 Mon Sep 17 00:00:00 2001
From: Andrew Mello <andrew@88plug.com>
Date: Thu, 29 Oct 2026 04:05:22 -0700
Subject: [PATCH 4/4] bench --vs: llama.cpp v0.8.0 and Ollama v0.40.13

The competitor lock moves llama.cpp from server-b11706 to server-b11960
(8052c13e, release v0.8.0) and Ollama from 0.40.5 to 0.40.13 (b2f0122c, the
latest release on 2026-10-28). Both digests were resolved with docker buildx
imagetools inspect and cover amd64 and arm64.

At v0.8.0, llama-server still takes -mu, -m, --alias, --host and --port
(common/arg.cpp) and still serves /health. Ollama's openai/openai.go still
streams thinking in delta.reasoning, which llm_load.py counts. The model pins
are unchanged. The docs links move to the new versions.

The finished rows that used this lock (H30, H31) record their own pins in their
run sections.
---
 bench/competitors.lock | 10 +++++-----
 tests/bench_vs_test.sh |  4 ++--
 2 files changed, 7 insertions(+), 7 deletions(-)

diff --git a/bench/competitors.lock b/bench/competitors.lock
index 2d61ac5d..e209a841 100644
--- a/bench/competitors.lock
+++ b/bench/competitors.lock
@@ -1,6 +1,6 @@
 {
   "schema": "k3d-gpu.competitors-lock/v1",
-  "note": "Competitor stacks for 'k3d-gpu bench --vs'. Every image is pinned by registry digest (verified with docker buildx imagetools inspect on 2026-10-17). Each stack serves the same model family and 4-bit tier as the k3d-gpu arm (Qwen3-4B, int4 / Q4_K_M), with its own vendor defaults. Weights are pinned too: after the model loads, 'bench' hashes artifact_path inside the container and refuses to measure unless it equals artifact_sha256 (the Hugging Face LFS sha256 at model_revision, or, for Ollama, the model-layer blob digest listed in the pinned registry manifest; Ollama rewrites its on-disk manifest copy, so that file does not hash to the registry digest).",
+  "note": "Competitor stacks for 'k3d-gpu bench --vs'. Every image is pinned by registry digest (verified with docker buildx imagetools inspect on 2026-10-17; the llama.cpp and Ollama pins moved to llama.cpp v0.8.0 and Ollama v0.40.13 on 2026-10-29, verified the same way). Each stack serves the same model family and 4-bit tier as the k3d-gpu arm (Qwen3-4B, int4 / Q4_K_M), with its own vendor defaults. Weights are pinned too: after the model loads, 'bench' hashes artifact_path inside the container and refuses to measure unless it equals artifact_sha256 (the Hugging Face LFS sha256 at model_revision, or, for Ollama, the model-layer blob digest listed in the pinned registry manifest; Ollama rewrites its on-disk manifest copy, so that file does not hash to the registry digest).",
   "stacks": {
     "llamacpp": {
       "api": "/v1",
@@ -9,9 +9,9 @@
       "artifact_sha256": "7485fe6f11af29433bc51cab58009521f205840f5b4ae3a32fa7f92e8534fdf5",
       "cache": "/root/.cache/llama.cpp",
       "device": "cpu",
-      "docs": "https://github.com/ggml-org/llama.cpp/blob/b11706/common/arg.cpp",
+      "docs": "https://github.com/ggml-org/llama.cpp/blob/v0.8.0/common/arg.cpp",
       "health": "/health",
-      "image": "ghcr.io/ggml-org/llama.cpp:server-b11706@sha256:f41d47eee678c02656eefc9ed8cce76cab584aa7032805eeef0e6639fcfda485",
+      "image": "ghcr.io/ggml-org/llama.cpp:server-b11960@sha256:8052c13edd987de8858f8150031e025ac3d66823f09cf3ec8220899c4abce220",
       "model": "Qwen/Qwen3-4B-GGUF",
       "model_revision": "Qwen/Qwen3-4B-GGUF@bc640142c66e1fdd12af0bd68f40445458f3869b",
       "port": 8080,
@@ -26,10 +26,10 @@
       "artifact_sha256": "163553aea1b1de62de7c5eb2ef5afb756b4b3133308d9ae7e42e951d8d696ef5",
       "cache": "/root/.ollama",
       "device": "cpu",
-      "docs": "https://github.com/ollama/ollama/blob/v0.40.5/openai/openai.go",
+      "docs": "https://github.com/ollama/ollama/blob/v0.40.13/openai/openai.go",
       "health": "/api/version",
       "ignore_eos": false,
-      "image": "ollama/ollama:0.40.5@sha256:5d3a2f25f9825628daba94dc0b1d421c7d57d6c03ae5faa9ba050e058bab8241",
+      "image": "ollama/ollama:0.40.13@sha256:b2f0122c476bc9fe7ce679d8bb4bd555351e25e2554eefb156f61cb0a89cf429",
       "model": "qwen3:4b-q4_K_M",
       "model_revision": "registry.ollama.ai/library/qwen3:4b-q4_K_M manifest sha256:2bfd38a7daaf4b1037efe517ccb73d1a3bbd4822cf89f1a82be1569050a114e0, model layer sha256:163553aea1b1de62de7c5eb2ef5afb756b4b3133308d9ae7e42e951d8d696ef5",
       "port": 11434,
diff --git a/tests/bench_vs_test.sh b/tests/bench_vs_test.sh
index c671db16..379ed3af 100755
--- a/tests/bench_vs_test.sh
+++ b/tests/bench_vs_test.sh
@@ -45,7 +45,7 @@ dry_ollama() {
     local o
     o=$(dry --vs ollama --id H10 --cpuset 0-15) || return 1
     grep -q '^docker run -d --name k3dgpu-g-vs-ollama -p 127.0.0.1:[0-9]*:11434 ' <<< "${o}" \
-        && grep -q -- '--cpuset-cpus 0-15 ollama/ollama:0.40.5@sha256:5d3a2f25' <<< "${o}" \
+        && grep -q -- '--cpuset-cpus 0-15 ollama/ollama:0.40.13@sha256:b2f0122c' <<< "${o}" \
         && grep -q -- '--id H10 ' <<< "${o}" \
         && grep -q -- '--arm ollama=http://127.0.0.1:[0-9]*/v1,qwen3:4b-q4_K_M$' <<< "${o}" \
         && grep -q '^verify: sha256 /root/.ollama/models/blobs/sha256-163553aea1b1de62de7c5eb2ef5afb756b4b3133308d9ae7e42e951d8d696ef5 = 163553ae' <<< "${o}"
@@ -54,7 +54,7 @@ dry_ollama() {
 dry_llamacpp() {
     local o
     o=$(dry --vs llamacpp) || return 1
-    grep -q 'llama.cpp:server-b11706@sha256:f41d47ee.* -mu https://huggingface.co/Qwen/Qwen3-4B-GGUF/resolve/bc640142c66e1fdd12af0bd68f40445458f3869b/Qwen3-4B-Q4_K_M.gguf -m /root/.cache/llama.cpp/Qwen3-4B-Q4_K_M.gguf --alias Qwen3-4B --host 0.0.0.0 --port 8080$' <<< "${o}" \
+    grep -q 'llama.cpp:server-b11960@sha256:8052c13e.* -mu https://huggingface.co/Qwen/Qwen3-4B-GGUF/resolve/bc640142c66e1fdd12af0bd68f40445458f3869b/Qwen3-4B-Q4_K_M.gguf -m /root/.cache/llama.cpp/Qwen3-4B-Q4_K_M.gguf --alias Qwen3-4B --host 0.0.0.0 --port 8080$' <<< "${o}" \
         && ! grep -q -- ' -hf ' <<< "${o}" \
         && grep -q -- '--arm llamacpp=http://127.0.0.1:[0-9]*/v1,Qwen3-4B$' <<< "${o}" \
         && grep -q -- '--id H0 --out bench-H0-llamacpp/result.json' <<< "${o}"