Upgrade frank-default-large to kingjones Q8_0 AGENT at full 262144 context

frank-default-large now serves the kingjones Qwen3.8-27B-ROCmFPX-Q8_0-AGENT model with a standalone MTP draft head (Q4_0), f16 K + TurboQuant V KV cache, and the Sharp chat template. Context is maxed to 262144 (one slot); reasoning is on with preserve_thinking + preserve_reasoning in the chat kwargs. - frank-default-large.env: Q8_0 AGENT model paths, CTX_SIZE=262144, REASONING=on, REASONING_FORMAT=none, CHAT_TEMPLATE_KWARGS {enable_thinking, preserve_thinking, preserve_reasoning} - serve-frank-large.sh: standalone draft-mtp profile (already switched off the Kairic/PromptForge path) - download-q8agent.sh: reproducible HF fetch of the three payload files - docs/LLM/kingjones-q8agent-rocmfpx.md: Q8_0 AGENT runbook - AGENTS.md: topology + ACTIVE runtime section + TTM/GTT load note - qwen3.8-strix-optimizations, Dankclaw Improvements: TTM pages_limit now 114 GiB (applied 2026-08-25), full 262k KV + Q8 payload fit in one slot - frank_large_config.bats: assert 262144 context, reasoning + preserve_* enabled The TTM/GTT was raised to 114 GiB (29,883,536 pages) to hold the full 262k KV.
2026-08-25 07:12:02 -07:00
parent 24efa98d56
commit 54b2d84613
11 changed files with 350 additions and 129 deletions
@@ -27,14 +27,17 @@ host.
 | Target | Runtime | Service | Model | Port |
 |--------|---------|---------|-------|------|
 | Dankserver | llama.cpp (upstream b10121) | frank-default-small | HauhauCS Qwen3.6-35B-A3B-Uncensored-Aggressive Q6_K_P + ngram-mod, 36 MoE experts on CPU | 7770 |
-| Dankclaw | charlie12345/ROCmFPX llama.cpp, ROCmI4 W4A4 runtime | frank-default-large | caf-eemin Qwen3.8-27B-ROCmI4-MTP Q4_0_ROCMI4 (embedded NextN/MTP head, gfx1151 IU4 batched verification), 262K context / 1 slot | 7770 |
+| Dankclaw | charlie12345/ROCmFPX llama.cpp | frank-default-large | kingjones Qwen3.8-27B-ROCmFPX-Q8_0-AGENT Q8_0_ROCMFPX (standalone MTP draft head, Turbo4 V cache, 262K context / 1 slot), reasoning on | 7770 |

 History: the 2026-08-17 redo retired the pipeline-parallel split and the
 FastFlowLM NPU endpoint. The 2026-08-18 swap retired the vLLM service on
 Dankclaw. On 2026-08-21 a DFlash2 Vulkan deployment briefly replaced CIRU,
 then was reverted the same day for speed. On 2026-08-22 the service briefly
 swapped to the Kairic Edge IU4 runtime, then was reverted to v3 the same day
-(operator decision). Runbooks: [docs/LLM/index.md](docs/LLM/index.md).
+(operator decision). On 2026-08-25 the service switched to the kingjones
+Qwen3.8-27B-ROCmFPX-Q8_0-AGENT model (standalone MTP draft, 262144 context,
+reasoning on); the previous Kairic Edge model was retired the same day. Runbooks:
+[docs/LLM/index.md](docs/LLM/index.md).
 Retired scripts live in `old/` (reference only) and `.AGENTTRASH/`.

 ## Repository Layout
@@ -179,7 +182,21 @@ When you rename a service or move a model to a new host, update

 ## Runtime & Troubleshooting

-### charlie12345/ROCmFPX ROCmI4 W4A4 llama.cpp (Dankclaw, ACTIVE)
+### kingjones Qwen3.8-27B-ROCmFPX-Q8_0-AGENT llama.cpp (Dankclaw, ACTIVE)
+
+The large service serves `kingjones/Qwen3.8-27B-ROCmFPX-Q8_0-AGENT`
+(public) from `/home/dank/models/kingjones-Qwen3.8-27B-ROCMFPX-Q8_0-AGENT/`
+with a charlie12345/ROCmFPX `llama-server` (build `build-kairic`). It runs a
+standalone MTP draft head on ROCm0, f16 K + TurboQuant V KV cache, a full
+262144 context / 1 slot, and reasoning on (Sharp template, budget -1,
+`--reasoning-format none`; chat kwargs keep `preserve_thinking` +
+`preserve_reasoning`). The payload (~28 GB main + 1.7 GB draft + 0.9 GB
+mmproj, ~30 GB total) plus the 262k KV cache fit in the lifted 114 GiB TTM/GTT
+budget (raised 2026-08-25). Env `scripts/frank-default-large.env`, serve
+`scripts/serve-frank-large.sh`, runbook
+[docs/LLM/kingjones-q8agent-rocmfpx.md](docs/LLM/kingjones-q8agent-rocmfpx.md).
+
+### charlie12345/ROCmFPX ROCmI4 W4A4 llama.cpp (Dankclaw, prior v3, 2026-08-18 to 2026-08-25)

 The v4 runtime serves `cafonz/Qwen3.8-27B-ROCmI4-MTP-GGUF` (`eemin` namespace,
 public) from `/home/dank/rocmi4-qwen38/ROCmFPX/build-rocmi4-w4a4/bin/llama-server`
@@ -300,9 +317,9 @@ graph TD
 - Never run `pkill -f` with a pattern that appears in your own command line.
  It kills your ssh session. Use `pkill -x <exact-name>`.
 - Model load is slow: Dankserver loads the 30.6 GB Q6_K_P through RAM
-  (`--no-mmap`, MoE experts on CPU) in a few minutes cold. Dankclaw v3
-  loads the ~36.7 GiB five-file payload in ~18 s with a warm page cache,
-  a few minutes cold. Health returns 503 while loading. Do not kill the
+  (`--no-mmap`, MoE experts on CPU) in a few minutes cold. Dankclaw loads the
+  ~30 GiB Q8_0 AGENT payload (main + MTP draft + mmproj) a few minutes cold
+  with a fresh KV cache; ~18 s with a warm page cache. Health returns 503 while loading. Do not kill the
  process. `deploy.sh` waits up to 20 minutes on both targets (see
  `LLM_WAIT`).
 - A client abort and a server release look different in the journal. A
@@ -21,20 +21,24 @@ marked "verify first". Re-verified 2026-08-15/16; corrections marked below.
 - **KV cache**: production runs `q8_0` K **and** `q8_0` V (both quantized).
 - **Spec**: DFlash/DSpark are retired. Production is `--spec-default`
  (ngram-mod) with draft-mtp first. See [[qwen3.8-spec-matrix]].
- **TTM**: still 15.24 GiB as of 2026-08-16 — the modprobe fix is still
-  pending a reboot.
+- **TTM**: raised to 114 GiB on 2026-08-25, set live (no reboot) and
+  persisted in `/etc/modprobe.d/ttm.conf`. The effective VRAM ceiling that
+  once held the service at ~15 GiB is gone.

 ## Verified Memory Picture (measured 2026-08-11)

 | Figure | Value | Source |
 | ------ | ----- | ------ |
 | OS physical RAM | 30.5 GiB | `/proc/meminfo` |
-| TTM pages_limit | 15.24 GiB (3,995,281 x 4KiB) | `/sys/module/ttm/parameters/pages_limit` |
+| TTM pages_limit | 114 GiB (29,883,536 x 4KiB) — raised 2026-08-25 | `/sys/module/ttm/parameters/pages_limit` |
 | ROCm VRAM total | 96.0 GiB | `rocm-smi --showmeminfo vram` |
 | ROCm VRAM used | 76.2 GiB | `rocm-smi` while serving 3x200700 |

-The OS accounts only ~30.5 GiB physical RAM, and the TTM GPU-addressable pool
-is capped at ~15.24 GiB. ROCm/SMI still reports a 96 GiB VRAM carve-out (VGM).
+The OS accounts only ~30.5 GiB physical RAM. The TTM GPU-addressable pool
+was capped at ~15.24 GiB (the real pre-2026-08-25 ceiling); that cap is now
+lifted to 114 GiB. ROCm/SMI still reports a 96 GiB VRAM carve-out (VGM) — a
+different, driver-reported figure that has always been larger than the kernel
+cap.
 This split is the Strix Halo GTT/VGM design. It drives the batch/ubatch memory
 limits documented in [[frank-default-large-benchmark]].

@@ -77,7 +81,7 @@ All require root and a reboot for the modprobe change.

 | Setting | Action | Effect | Priority |
 | ------- | ------ | ------ | -------- |
-| TTM GPU pool | Set `amd-ttm --set 96` (write `/etc/modprobe.d/ttm.conf`), reboot | Lifts the 15 GiB TTM cap toward full RAM; more headroom for GPU buffers | High |
+| TTM GPU pool | `/etc/modprobe.d/ttm.conf` → `options ttm pages_limit=29883536 page_pool_size=29883536` (applied 2026-08-25, set live then persisted) | Lifts the 15 GiB cap to 114 GiB; the full 262k KV + Q8 AGENT payload fit in one slot | Done |
 | Power profile | **N/A on this machine** (2026-08-15): no `power_profile` sysfs; GPU DPM already `performance` | none | ~~Medium~~ |
 | NVMe scheduler | Already `none` (`[none]` confirmed) | No change | Done |
 | BIOS VGM | Set Variable Graphics Memory / UMA to "High" in the EVO-X2 BIOS | Max dedicated VRAM for the GPU carve-out; verify in the live BIOS | Verify |
@@ -137,24 +137,16 @@ from `peculiar-ragdoll/Qwen-Sharp-Chat-Templates`. See [[dealign-ornith-crack]].

 ### Dankclaw — frank-default-large

-The active service on Dankclaw is the charlie12345/ROCmFPX llama.cpp runtime
-(with the opt-in `ROCmI4 W4A4` path) serving `cafonz/Qwen3.8-27B-ROCmI4-MTP-GGUF`
-(`eemin` namespace, `Q4_0_ROCMI4`, embedded NextN/MTP head) under the alias
-`frank-default-large` on port 7770. It runs
-`/home/dank/rocmi4-qwen38/ROCmFPX/build-rocmi4-w4a4/bin/llama-server` (built
-against system ROCm 7.2.4 / amdclang 22) with a **strict Qwen MTP** draft
-(`--spec-type draft-mtp --spec-mtp-strict-qwen`) running on ROCm0 with all
-layers (`--spec-draft-ngl all`) — no separate draft file, because the MTP head
-is embedded in the main GGUF. K/V are f16, 65536 context (192K below the
-native 262144 maximum; 1 slot, concurrent requests queue, the server never
-preempts a running generation),
-`-t 16 -tb 32`, batch 512 / ubatch 256 (ubatch is swappable; sweep it on 8k
-context with `scripts/bench-ubatch-rocmi4.sh`), 16 context checkpoints, 8 GiB
-prompt cache, and reasoning in deepseek format. The unit file is
-`scripts/frank-default-large.service`. It launches through
+The active service on Dankclaw uses the ROCmFPX llama.cpp runtime to serve
+[kingjones777/Qwen3.8-27B-ROCmFPX-Q8_0-AGENT-GGUF](https://huggingface.co/kingjones777/Qwen3.8-27B-ROCmFPX-Q8_0-AGENT-GGUF)
+under the alias `frank-default-large` on port 7770. The main model is
+`Qwen3.8-27B-Q8_0_ROCMFPX_AGENT.gguf`; the standalone MTP draft is
+`mtp-Qwen3.8-27B-Q4_0.gguf`, and the BF16 vision projector is loaded alongside
+it. The K-side KV cache remains `f16`; the V-side cache uses `turbo4`. The
+service keeps the existing 65536 context / 1-slot profile. The unit file is
+`scripts/frank-default-large.service`; it launches through
 `scripts/serve-frank-large.sh`, which reads tunables from
 `scripts/frank-default-large.env`, applying the Qwen-Sharp chat template.
-See [[eemin-rocmi4-mtp]].

 > The kingjones ROCmFP4 STRIX MTP model served the slot from 2026-08-24; it
 > was replaced by the caf-eemin ROCmI4 embedded-MTP model on 2026-08-25. The
@@ -1,10 +1,10 @@
 # LLM Fleet — Knowledge Base

 Documentation for the LLM inference fleet on two local machines
-(2026-08-24): upstream llama.cpp on Dankserver (frank-default-small,
-Ornith-1.5-35B-A3B CRACK) and CIRU ROCmFPX llama.cpp on Dankclaw
-(frank-default-large, kingjones Qwen3.8-27B ROCmFP4). Both services
-run the Qwen-Sharp chat template.
+(2026-09-01): upstream llama.cpp on Dankserver (frank-default-small,
+Ornith-1.5-35B-A3B CRACK) and ROCmFPX llama.cpp on Dankclaw
+(frank-default-large, kingjones Qwen3.8-27B ROCmFPX Q8 AGENT). Both
+services run the Qwen-Sharp chat template.

 > 2026-08-22: the former `vllmScripts` repository is merged into the
 > serverAppConfigs repository. This knowledge base moved from the old repo's
@@ -28,8 +28,8 @@ ngram-mod, selected experts offloaded via override, 491520 ctx]
        D[gfx1151 iGPU 64 GiB unified memory]
        E[CIRU ROCmFPX llama.cpp PromptForge v3
 frank-default-large, port 7770]
-        F[Qwen3.8-27B-Uncensored-ROCmFP4 Q6_0 AGENT
-MTP draft + Sharp template, 262K ctx / 1 slot]
+        F[Qwen3.8-27B ROCmFPX Q8_0 AGENT
+MTP draft + Turbo4 V cache, 262K ctx / 1 slot]
        D --> E --> F
    end

@@ -48,8 +48,10 @@ MTP draft + Sharp template, 262K ctx / 1 slot]

 - [[deployment]] — deploy and update process for both servers
 - [[frank-redo-2026-08-17]] — the 2026-08-17 topology change
- [[eemin-rocmi4-mtp]] — the caf-eemin Qwen3.8-27B ROCmI4 embedded-MTP model
-  on Dankclaw (frank-default-large, active 2026-08-25)
+- [kingjones Qwen3.8-27B ROCmFPX Q8 AGENT](https://huggingface.co/kingjones777/Qwen3.8-27B-ROCmFPX-Q8_0-AGENT-GGUF) — active frank-default-large model on Dankclaw; standalone MTP and Turbo4 V cache
+- [[kingjones-q8agent-rocmfpx]] — the Q8_0-AGENT runbook: serving profile, download, 262144 context, reasoning + preserve_thinking, 114 GiB TTM/GTT
+- [[eemin-rocmi4-mtp]] — the former caf-eemin Qwen3.8-27B embedded-MTP model
+  on Dankclaw (retired)
 - [[dealign-ornith-crack]] — the dealign Ornith-1.5-35B-A3B CRACK model on
  Dankserver (frank-default-small, active 2026-08-24)
 - [[kingjones-strix-mtp]] — the retired kingjones Qwen3.8-27B ROCmFP4 STRIX MTP
@@ -0,0 +1,165 @@
+# frank-default-large: kingjones Qwen3.8-27B-ROCmFPX-Q8_0-AGENT
+
+**Status:** ACTIVE on Dankclaw (since 2026-08-25).
+
+**Alias:** `frank-default-large`, port **7770**.
+
+The large service serves the kingjones `Qwen3.8-27B-ROCmFPX-Q8_0-AGENT` GGUF
+from `/home/dank/models/kingjones-Qwen3.8-27B-ROCMFPX-Q8_0-AGENT/` with a
+charlie12345/ROCmFPX `llama-server` (build `build-kairic`). It runs a
+**standalone MTP draft head** (loaded separately, on ROCm0), **TurboQuant** V
+KV cache, the **Sharp** chat template with **reasoning on**, and a **full
+262144 context / 1 slot** profile. The whole payload plus the 262k KV cache
+fit in the lifted **114 GiB TTM/GTT** budget (see the 114 GiB change in
+[[Dankclaw Improvements]] and [[qwen3.8-strix-optimizations]]).
+
+| Figure | Value |
+| ------ | ----- |
+| Model | kingjones Qwen3.8-27B-ROCmFPX-Q8_0-AGENT (`Q8_0_ROCMFPX`) |
+| HF repo | `kingjones777/Qwen3.8-27B-ROCmFPX-Q8_0-AGENT-GGUF` (public) |
+| Runtime | charlie12345/ROCmFPX `llama-server`, `build-kairic` |
+| Draft head | standalone `mtp-Qwen3.8-27B-Q4_0.gguf` on ROCm0 |
+| KV cache | f16 K, TurboQuant (`turbo4`) V, full 262144 context, 1 slot |
+| Reasoning | on, `deepseek`, budget `-1`, `--reasoning-format none` |
+| Chat template | Sharp (repo-tracked), `enable_thinking` + `preserve_thinking` on |
+| Host path | `/home/dank/models/kingjones-Qwen3.8-27B-ROCMFPX-Q8_0-AGENT/` |
+| Tunables | `scripts/frank-default-large.env` |
+| Serve script | `scripts/serve-frank-large.sh` |
+
+## Model files
+
+Three files, from the HF repo above:
+
+| File | Size | Purpose |
+| ---- | ---- | ------- |
+| `Qwen3.8-27B-Q8_0_ROCMFPX_AGENT.gguf` | 28.21 GB | main weights |
+| `mtp-Qwen3.8-27B-Q4_0.gguf` | 1.68 GB | standalone MTP draft head (loaded by the script) |
+| `mmproj-Qwen3.8-27B-BF16.gguf` | 0.93 GB | vision projector (`--mmproj`) |
+
+The repo also has a `mtp-Qwen3.8-27B-Q8_0.gguf` (3.16 GB) draft; the env uses
+the Q4_0 draft. Download them from the resolve URLs below (they are large —
+use the token in `~/.cache/huggingface/token`).
+
+## Download
+
+```bash
+# on Dankclaw
+TOKEN=$(cat ~/.cache/huggingface/token)
+REPO=kingjones777/Qwen3.8-27B-ROCmFPX-Q8_0-AGENT-GGUF
+DIR=/home/dank/models/kingjones-Qwen3.8-27B-ROCMFPX-Q8_0-AGENT
+mkdir -p "$DIR"; cd "$DIR"
+dl(){ curl -fsSL -C - -o "$1" --retry 5 --retry-delay 3 --retry-max-time 3600 \
+  -H "Authorization: Bearer $TOKEN" \
+  "https://huggingface.co/$REPO/resolve/main/$1"; echo "done: $1"; }
+dl Qwen3.8-27B-Q8_0_ROCMFPX_AGENT.gguf
+dl mmproj-Qwen3.8-27B-BF16.gguf
+dl mtp-Qwen3.8-27B-Q4_0.gguf
+```
+
+The serve script checks each artifact is present and readable before launch
+(`scripts/frank-default-large.env` -> the four `MODEL_*` files plus the chat
+template). A missing file fails the artifact check with `missing or
+unreadable required artifact`.
+
+## Serving profile
+
+`scripts/serve-frank-large.sh` builds the `llama-server` call from the env:
+
+```
+/dev/llama-server -m <main> --alias frank-default-large
+  --host 0.0.0.0 --port 7770 --jinja
+  -dev ROCm0 -ngl 999 -np 1 -c 262144
+  -b 512 -ub 512 -fa on
+  -ctk f16 -ctv turbo4 -t 16 -tb 32
+  -ctxcp 32 --cache-ram 8192 --cache-prompt --cache-idle-slots --metrics
+  --spec-type draft-mtp --model-draft <mtp-Q4_0>
+  --mmproj <mmproj-BF16>
+  --spec-draft-device ROCm0 --spec-draft-ngl 999
+  --spec-draft-type-k f16 --spec-draft-type-v turbo4
+  --spec-draft-n-max 4 --spec-draft-n-min 0
+  --spec-draft-p-min 0.0 --spec-draft-p-split 0.10
+  --spec-draft-backend-sampling
+  --temp 0.7 --top-p 0.8 --top-k 20 --min-p 0.0
+  --presence-penalty 1.5
+  --reasoning on --reasoning-format none --reasoning-budget -1
+```
+
+Key choices:
+
+- **Standalone MTP draft** (`--model-draft`, `--spec-type draft-mtp`). The
+  AGENT GGUF has no embedded draft head; the draft head is a separate file on
+  ROCm0. This matches the Qwen3.8-27B AGENT shape (`--spec-draft-type-k
+  f16 --spec-draft-type-v turbo4`).
+- **f16 K + TurboQuant V** cache. f16 K is the safe long-ctx default; TurboQuant
+  halves the V cache at 262k while keeping quality (see the ubatch/threads-batch
+  notes in [[qwen3.8-strix-optimizations]]).
+- **Full 262144 context, 1 slot.** The whole context KV cache plus the ~30 GB
+  payload (main + draft + mmproj) fit in the 114 GiB budget. Concurrent
+  requests queue; the server never preempts a running generation.
+- **Reasoning on.** The Sharp template drives thinking (medium effort);
+  `--reasoning-format none` lets the template project the reasoning content into
+  `reasoning_content`. `--reasoning-budget -1` = no budget cap.
+
+## Reasoning + chat template
+
+The Sharp template (`scripts/chat-template.jinja`, repo-tracked) is a
+Qwen-AGENT shape. Env `CHAT_TEMPLATE_KWARGS` keeps the reasoning trail in the
+conversation:
+
+```json
+{"enable_thinking": true, "preserve_thinking": true, "preserve_reasoning": true}
+```
+
+`preserve_thinking: true` keeps the model's prior thinking in the next prompt
+(the user's "preserve thinking" request). `preserve_reasoning: true` keeps the
+model's prior reasoning trace across requests too (OpenWebUI sends
+`preserve_reasoning`/`preserve_thinking: true` when the conversation needs the
+full trail). For multi-turn reasoning, the template stays in the reasoning
+shape and `enable_thinking: true` per request.
+
+## Memory: the 114 GiB TTM/GTT requirement
+
+The 262k f16 KV cache is large (~a few GB) and the payload is ~30 GB. Both live
+in the Strix Halo unified-memory GTT. The TTM `pages_limit` was raised to
+114 GiB on 2026-08-25 (written live to
+`/sys/module/ttm/parameters/pages_limit`, then persisted in
+`/etc/modprobe.d/ttm.conf`). Without the raise the effective ceiling is ~15 GiB
+and a 28 GB + 262k-KV load stalls. The GTT page-limit change is per the
+`qwen3.8-strix-optimizations` table item 5.
+
+## Health + troubleshooting
+
+- Health: `curl http://192.168.0.143:7770/health` → `{"status":"ok"}`.
+- `/api/models` reports `n_ctx:262144, n_ctx_train:262144, n_params:27.3B`.
+- Model load is slow (a few minutes cold for the ~30 GB payload with a fresh
+  KV cache). Health returns 503 while loading. Do not kill the process.
+  `deploy.sh` waits up to 20 minutes (see `LLM_WAIT`).
+- Kernel spam about `amdgpu: SVM mapping failed` is the known documented
+  symptom, not root cause (see the crash-loop procedure in the project
+  AGENTS.md). Investigate other causes before assuming memory pressure.
+- A client abort logs `W srv stop: cancel task, id_task = N`; a natural stop
+  logs only `slot release` with `truncated = 0`.
+
+## Env tunables
+
+`scripts/frank-default-large.env`. Serving: `CTX_SIZE=262144`,
+`PARALLEL=1`, `BATCH_SIZE=512`, `UBATCH_SIZE=512`, `NUM_THREADS=16`,
+`THREAD_BATCH=32`, `CACHE_TYPE_K=f16`, `CACHE_TYPE_V=turbo4`,
+`CTX_CHECKPOINTS=32`, `CACHE_RAM=8192`. Reasoning: `REASONING=on`,
+`REASONING_FORMAT=none`. Sampling: `TEMPERATURE=0.7`, `TOP_K=20`,
+`TOP_P=0.8`. MTP: `MTP_N_MAX=4`, `MTP_N_MIN=0`, `MTP_P_MIN=0.0`,
+`MTP_P_SPLIT=0.10`. Strix Halo: `HSA_OVERRIDE_GFX_VERSION=11.5.1`,
+`GGML_HIP_ENABLE_UNIFIED_MEMORY=1`.
+
+**Never name a tunable `TEMP`, `HOST`, or `GPU_MAX_HEAP_SIZE`** — the ROCm HIP
+runtime reads those bare vars (see the 2026-08-13 TEMP incident). llama.cpp
+reads `--host/--port/--temp` as CLI flags, never as env vars.
+
+## History
+
+- **2026-08-25 (today):** switched the large service to the kingjones
+  Q8_0-AGENT model with 262144 context and reasoning on. Downloaded the three
+  payload files from the HF repo, raised the TTM/GTT to 114 GiB for the full
+  context, and persisted the env/serve script in this repo.
+- Preceding model: Kairic Edge Qwen3.8-27B IU4 (deployed 2026-08-25 ~05:42,
+  retired the same day for the AGENT model). See [[kairic-edge]].
@@ -19,7 +19,7 @@ Related pages:
 | 2 | MTP `n_max` explicit 4–5 + Q4_0 draft head | +5–15% on the spec share | restart, 1.56 GB file | test 08-16 |
 | 3 | Plain `Q8_0` A/B vs UD-Q8_K_XL (same 8 bits) | dequant ALUs may differ | restart, 29 GB file | test 08-16 |
 | 4 | Prefill path: ROCm backend + tuned ubatch/threads-batch | 2–3× prefill for agentic turns | per-workload backend choice | partial data |
-| 5 | TTM `pages_limit` raise (15.2 GiB → 90+) | removes a hard VRAM ceiling, stable 96 GiB budget | reboot | pending |
+| 5 | TTM `pages_limit` raised to 114 GiB | removes the hard VRAM ceiling; the full 262k f16 KV plus the Q8 AGENT payload, MTP draft, and mmproj fit in one slot | modprobe, set live 2026-08-25 | applied (29,883,536 pages) |
 | 6 | Vulkan ICD pinning + loader-layer disable | robustness; small | service env vars | test 08-16 |

 Everything else (quantization downgrade, NVFP4, larger drafts) is documented
@@ -166,11 +166,13 @@ hybrid state, upstream). Levers:

 ### 5. OS level (from the Improvements doc, re-verified 2026-08-15)

- **TTM `pages_limit` is still 3,995,281 pages ≈ 15.24 GiB** — verified
-  today, unchanged. The 96 GiB GTT budget only materializes if the TTM cap
-  is raised (kernel parameter, needs reboot). Until then the effective
-  ceiling is lower and the 28 GiB load-stall risk remains. **Highest-value
-  single OS change; schedule with the next reboot.**
+- **TTM `pages_limit` is now 114 GiB (29,883,536 pages)** — applied 2026-08-25.
+  Written live to `/sys/module/ttm/parameters/pages_limit` (no reboot required)
+  and persisted in `/etc/modprobe.d/ttm.conf`. The full 262k f16 KV cache plus
+  the ~28 GB Q8 AGENT payload, MTP draft, and mmproj now fit in one slot. The
+  load-stall risk from the old 15 GiB cap is resolved. Note: 96 GiB is the
+  driver-reported VRAM carve-out (VGM/SVM pool); the TTM pages_limit is the
+  kernel page map that was the real ceiling.
 - `iommu=pt` — verified present on the kernel cmdline. Done.
 - **Vulkan ICD pinning** — `radeon_icd.json` exists; no `VK_ICD_FILENAMES`
  or loader-layer env is set. Pinning the radeon ICD and disabling loader
@@ -99,6 +99,7 @@ The LLM inference fleet on the two hosts. Full index:
 - [[frank-redo-2026-08-17]] — the 2026-08-17 topology change
 - [[eemin-rocmi4-mtp]] — the caf-eemin Qwen3.8-27B ROCmI4 embedded-MTP model on Dankclaw (frank-default-large, active)
 - [[kingjones-strix-mtp]] — the RETIRED kingjones Qwen3.8-27B ROCmFP4 STRIX MTP model on Dankclaw (superseded 2026-08-25)
+- [[kingjones-q8agent-rocmfpx]] — the Q8_0-AGENT runbook for frank-default-large (active): serving, download, 262144 context, reasoning + preserve_thinking, 114 GiB TTM/GTT
 - [[ciru-activefpx-promptforge]] — the RETIRED v3 CIRU PromptForge runtime on Dankclaw (superseded 2026-08-24)
 - [[kingjones-strix-mtp]] — the kingjones Qwen3.8-27B ROCmFP4 STRIX MTP model on Dankclaw (active)
 - [[dealign-ornith-crack]] — the dealign Ornith-1.5-35B-A3B CRACK model on Dankserver (active)
@@ -0,0 +1,63 @@
+#!/usr/bin/env bash
+# Download the kingjones Qwen3.8-27B-ROCmFPX-Q8_0-AGENT payload for
+# frank-default-large from HuggingFace into /home/dank/models/.../Q8_0-AGENT.
+#
+# Uses the token in ~/.cache/huggingface/token so the ~28 GB main file does not
+# hit the anonymous 10 GB/30 min rate limit. Run on Dankclaw.
+#
+#   bash scripts/download-q8agent.sh
+#   bash scripts/download-q8agent.sh --verify-only
+set -euo pipefail
+
+REPO="kingjones777/Qwen3.8-27B-ROCmFPX-Q8_0-AGENT-GGUF"
+DIR="/home/dank/models/kingjones-Qwen3.8-27B-ROCMFPX-Q8_0-AGENT"
+EXPECTED=(
+  "Qwen3.8-27B-Q8_0_ROCMFPX_AGENT.gguf:28214425952"
+  "mmproj-Qwen3.8-27B-BF16.gguf:931145888"
+  "mtp-Qwen3.8-27B-Q4_0.gguf:1680271648"
+)
+
+cd -- "$(dirname -- "$0")/.."
+
+main() {
+  local mode="download"
+  [[ "${1:-}" == "--verify-only" ]] && mode="verify"
+  mkdir -p "$DIR"
+  cd "$DIR"
+
+  local token
+  token="$(cat "$HOME/.cache/huggingface/token")"
+  [[ -n "$token" ]] || { echo "no HF token in $HOME/.cache/huggingface/token" >&2; exit 1; }
+
+  dl() {
+    local f="$1"
+    curl -fsSL -C - -o "$f" --retry 5 --retry-delay 3 --retry-max-time 3600 \
+      -H "Authorization: Bearer $token" \
+      "https://huggingface.co/$REPO/resolve/main/$f"
+    echo "  done: $f ($(du -h "$f" | cut -f1))"
+  }
+
+  if [[ "$mode" == "download" ]]; then
+    echo "downloading $REPO into $DIR"
+    dl "Qwen3.8-27B-Q8_0_ROCMFPX_AGENT.gguf"
+    dl "mmproj-Qwen3.8-27B-BF16.gguf"
+    dl "mtp-Qwen3.8-27B-Q4_0.gguf"
+  fi
+
+  echo "verifying sizes"
+  local ok=1 f want act
+  for entry in "${EXPECTED[@]}"; do
+    f="${entry%%:*}"; want="${entry##*:}"
+    act="$(stat -c%s "$f")"
+    if [[ "$act" -eq "$want" ]]; then
+      echo "  OK   $f ($act bytes)"
+    else
+      echo "  BAD  $f ($act bytes, want $want)" >&2
+      ok=0
+    fi
+  done
+  [[ "$ok" -eq 1 ]] || exit 1
+  echo "all files present and sized correctly"
+}
+
+main "$@"
@@ -1,65 +1,56 @@
-# Tunables for frank-default-large (CIRU/Kairic Edge ROCmFPX v1.2, dankclaw,
-# Strix Halo gfx1151).
+# Tunables for frank-default-large (kingjones Qwen3.8-27B ROCmFPX Q8 AGENT,
+# dankclaw, Strix Halo gfx1151).
 #
-# Runtime: ciru-ai/ROCmFPX tag kairic-edge-qwen38-27b-v1.2, commit
-# 205a3e5f40e5542e2f2eb68e3d3f81f918b1d895, built with the pinned gfx1151
-# Prompt Forge / IU4 settings against /opt/rocm. Serves the Kairic Edge
-# authoritative GGUF plus three required .pfs sidecars. Alias
+# Runtime: ROCmFPX llama.cpp with TurboQuant support. Serves the kingjones
+# AGENT GGUF plus its standalone MTP draft and vision projector. Alias
 # frank-default-large, port 7770.

 # WARNING: do not name any variable TEMP, HOST, or GPU_MAX_HEAP_SIZE. The ROCm
 # HIP runtime reads those bare env vars (see the 2026-08-13 TEMP incident).
 # llama.cpp reads --host/--port/--temp as CLI flags, never as env vars.

-# ciru-ai/ROCmFPX Kairic Edge v1.2 build.
+# ROCmFPX build with TurboQuant KV-cache support.
 LLAMA_BIN=/home/dank/ciru-qwen38/ROCmFPX/build-kairic/bin/llama-server

-# Model files: authoritative GGUF plus all required Kairic Edge sidecars.
-MODEL_DIR=/home/dank/models/Kairic-Qwen3.8-27B-IU4-Edge
-MODEL_FILE=Qwen3.8-27B-IU4-Kairic-Edge.gguf
-KAIRIC_FFN_SIDECAR=$MODEL_DIR/Qwen3.8-27B-Kairic-IU4-FFN.pfs
-KAIRIC_GDN_SIDECAR=$MODEL_DIR/Qwen3.8-27B-Kairic-IU4-GDN.pfs
-KAIRIC_GDN_OUTPUT_SIDECAR=$MODEL_DIR/Qwen3.8-27B-Kairic-IU4-GDN-Output.pfs
-KAIRIC_EDGE_COMPATIBILITY_MODE=1
-# No separate MTP draft head: the MTP/NextN head is embedded in the main GGUF,
-# so no --model-draft is passed. Kept for scripts that reference the var.
-MTP_DRAFT_FILE=
+# Model files from https://huggingface.co/kingjones777/Qwen3.8-27B-ROCmFPX-Q8_0-AGENT-GGUF.
+MODEL_DIR=/home/dank/models/kingjones-Qwen3.8-27B-ROCMFPX-Q8_0-AGENT
+MODEL_FILE=Qwen3.8-27B-Q8_0_ROCMFPX_AGENT.gguf
+MTP_DRAFT_FILE=$MODEL_DIR/mtp-Qwen3.8-27B-Q4_0.gguf
+MMPROJ_FILE=mmproj-Qwen3.8-27B-BF16.gguf

 # Qwen-Sharp chat template (tracked in the repo, staged on the host). The Sharp
 # template drives thinking on by default (medium effort) and produces the
 # peg-native + deepseek chat/reasoning shape the Pi agent expects.
 CHAT_TEMPLATE_FILE=/home/dank/serverAppConfigs/scripts/chat-template.jinja
-CHAT_TEMPLATE_KWARGS='{"enable_thinking":true,"preserve_thinking":true}'
+CHAT_TEMPLATE_KWARGS='{"enable_thinking":true,"preserve_thinking":true,"preserve_reasoning":true}'

 # Alias + network.
 ALIAS=frank-default-large
 BIND_ADDR=0.0.0.0
 LISTEN_PORT=7770

-# Serving: 65536 context (192K below the model's 262144 maximum), 1 slot. f16
-# KV, context checkpoints, prompt cache, strict-MTP embedded draft on ROCm0.
-# This leaves sufficient unified-memory headroom for the model plus KV cache.
+# Serving: 262144 context (the model's maximum), 1 slot. f16 KV, context
+# checkpoints, prompt cache, standalone MTP draft on ROCm0. The full 262144
+# context KV cache plus the ~28-30 GB Q8 model + MTP draft + mmproj fit in the
+# lifted 114 GiB TTM/GTT budget on the Strix Halo (see the GTT 114 GiB change).
 # Concurrent requests queue; the server never preempts a running generation.
-CTX_SIZE=65536
+CTX_SIZE=262144
 PARALLEL=1
-# Batch/ubatch follow the qualified ROCmI4 W4A4 profile (see the bench in
-# scripts/bench-ubatch-rocmi4.sh for the ubatch sweep on 8k contexts).
+# Batch/ubatch follow the existing qualified Dankclaw profile.
 BATCH_SIZE=512
 UBATCH_SIZE=512
 NUM_THREADS=16
 THREAD_BATCH=32
 CACHE_TYPE_K=f16
-CACHE_TYPE_V=f16
+CACHE_TYPE_V=turbo4
 CTX_CHECKPOINTS=32
 CACHE_RAM=8192

-# Embedded-MTP (qwen35 NextN) qualified profile. The head is in the main GGUF,
-# so the draft runs with all layers on ROCm0 (no --model-draft).
+# Standalone MTP draft head, loaded from MTP_DRAFT_FILE by the serve script.
 MTP_N_MAX=4
 MTP_N_MIN=0
 MTP_P_MIN=0.0
 MTP_P_SPLIT=0.10
-KAIRIC_EDGE_COMPATIBILITY_MODE=1

 # Sampling (Sharp default).
 TEMPERATURE=0.7
@@ -68,9 +59,10 @@ TOP_P=0.8
 MIN_P=0.0
 PRESENCE_PENALTY=1.5

-# Thinking on (Sharp template drives it; reasoning format projects the tar into
-# reasoning_content).
-REASONING=off
+# Reasoning on (Sharp template drives it; preserve_thinking + preserve_reasoning
+# keep the reasoning trail in the conversation). reasoning-format none lets the
+# template project the reasoning content into reasoning_content.
+REASONING=on
 REASONING_FORMAT=none

 # Strix Halo unified memory + native gfx1151 detection for this ROCmFPX build.
@@ -1,7 +1,6 @@
 #!/usr/bin/env bash
-# Serve Qwen3.8-27B IU4 Kairic Edge v1.2 as frank-default-large.
-# The Kairic runtime requires its authoritative GGUF and all three Prompt Forge
-# sidecars. Compatibility mode is enabled for Qwen's recommended sampling.
+# Serve the kingjones Qwen3.8-27B ROCmFPX Q8 AGENT model as frank-default-large.
+# The deployment uses a standalone MTP draft head and Turbo4 V-side KV cache.
 set -euo pipefail

 SCRIPT_DIR="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")" && pwd)"
@@ -14,11 +13,10 @@ if [[ -f "$ENV_FILE" ]]; then
 fi

 : "${LLAMA_BIN:=/home/dank/ciru-qwen38/ROCmFPX/build-kairic/bin/llama-server}"
-: "${MODEL_DIR:=/home/dank/models/Kairic-Qwen3.8-27B-IU4-Edge}"
-: "${MODEL_FILE:=Qwen3.8-27B-IU4-Kairic-Edge.gguf}"
-: "${KAIRIC_FFN_SIDECAR:=$MODEL_DIR/Qwen3.8-27B-Kairic-IU4-FFN.pfs}"
-: "${KAIRIC_GDN_SIDECAR:=$MODEL_DIR/Qwen3.8-27B-Kairic-IU4-GDN.pfs}"
-: "${KAIRIC_GDN_OUTPUT_SIDECAR:=$MODEL_DIR/Qwen3.8-27B-Kairic-IU4-GDN-Output.pfs}"
+: "${MODEL_DIR:=/home/dank/models/kingjones-Qwen3.8-27B-ROCMFPX-Q8_0-AGENT}"
+: "${MODEL_FILE:=Qwen3.8-27B-Q8_0_ROCMFPX_AGENT.gguf}"
+: "${MTP_DRAFT_FILE:=$MODEL_DIR/mtp-Qwen3.8-27B-Q4_0.gguf}"
+: "${MMPROJ_FILE:=mmproj-Qwen3.8-27B-BF16.gguf}"
 : "${ALIAS:=frank-default-large}"
 : "${BIND_ADDR:=0.0.0.0}"
 : "${LISTEN_PORT:=7770}"
@@ -29,7 +27,7 @@ fi
 : "${NUM_THREADS:=16}"
 : "${THREAD_BATCH:=32}"
 : "${CACHE_TYPE_K:=f16}"
-: "${CACHE_TYPE_V:=f16}"
+: "${CACHE_TYPE_V:=turbo4}"
 : "${CTX_CHECKPOINTS:=32}"
 : "${CACHE_RAM:=8192}"
 : "${MTP_N_MAX:=4}"
@@ -43,11 +41,10 @@ fi
 : "${PRESENCE_PENALTY:=1.5}"
 : "${REASONING:=off}"
 : "${REASONING_FORMAT:=none}"
-: "${KAIRIC_EDGE_COMPATIBILITY_MODE:=1}"
 : "${HSA_OVERRIDE_GFX_VERSION:=11.5.1}"

 MODEL="$MODEL_DIR/$MODEL_FILE"
-for f in "$LLAMA_BIN" "$MODEL" "$KAIRIC_FFN_SIDECAR" "$KAIRIC_GDN_SIDECAR" "$KAIRIC_GDN_OUTPUT_SIDECAR"; do
+for f in "$LLAMA_BIN" "$MODEL" "$MTP_DRAFT_FILE" "$MODEL_DIR/$MMPROJ_FILE" "$CHAT_TEMPLATE_FILE"; do
    if [[ ! -r "$f" ]]; then
        echo "missing or unreadable required artifact: $f" >&2
        exit 1
@@ -57,11 +54,6 @@ if [[ ! -x "$LLAMA_BIN" ]]; then
    echo "missing executable runtime: $LLAMA_BIN" >&2
    exit 1
 fi
-if [[ "$KAIRIC_EDGE_COMPATIBILITY_MODE" != 1 ]]; then
-    echo "KAIRIC_EDGE_COMPATIBILITY_MODE must be 1 for this deployment" >&2
-    exit 1
-fi
-
 SERVER_DIR="$(cd -- "$(dirname -- "$LLAMA_BIN")" && pwd)"
 ROCM_PATH="${ROCM_PATH:-/opt/rocm}"
 LIBRARY_PATH="$SERVER_DIR:$ROCM_PATH/lib:$ROCM_PATH/lib/rocm_sysdeps/lib:$ROCM_PATH/lib/llvm/lib:$ROCM_PATH/llvm/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}"
@@ -74,35 +66,14 @@ exec /usr/bin/env \
    GGML_CUDA_GRAPH_OPT=0 \
    LLAMA_TARGET_GREEDY_ARGMAX_FASTPATH=0 \
    LLAMA_MTP_CPU_ARGMAX_FASTPATH=1 \
-    PROMPTFORGE_TARGET_ONLY=0 \
-    PROMPTFORGE_MODE=iu4_ffn \
-    PROMPTFORGE_IU4_SIDECAR="$KAIRIC_FFN_SIDECAR" \
-    PROMPTFORGE_SIDECAR="$KAIRIC_FFN_SIDECAR" \
-    PROMPTFORGE_IU4_HADAMARD=1 \
-    PROMPTFORGE_IU4_SEGMENTED=0 \
-    PROMPTFORGE_ENABLE_FFN_KEEPERS=0 \
-    PROMPTFORGE_FFN_KEEPERS=late6 \
-    PROMPTFORGE_ENABLE_GDN=1 \
-    PROMPTFORGE_ENABLE_GDN_W8=0 \
-    PROMPTFORGE_GDN_SIDECAR_OVERRIDE="$KAIRIC_GDN_SIDECAR" \
-    PROMPTFORGE_GDN_SIDECAR="$KAIRIC_GDN_SIDECAR" \
-    PROMPTFORGE_GDN_IU4_HADAMARD=1 \
-    PROMPTFORGE_ENABLE_GDN_KEEPERS=0 \
-    PROMPTFORGE_ENABLE_GDN_OUTPUT=1 \
-    PROMPTFORGE_GDN_OUTPUT_SIDECAR="$KAIRIC_GDN_OUTPUT_SIDECAR" \
-    PROMPTFORGE_GDN_OUTPUT_KEEPERS=v3_lateq6 \
-    PROMPTFORGE_ENABLE_SMALLM_IU4=1 \
-    PROMPTFORGE_ENABLE_SMALLM_GDN_IU4=1 \
-    PROMPTFORGE_ENABLE_SMALLM_GDN_OUTPUT_IU4=0 \
-    PROMPTFORGE_ENABLE_IU4_DECODE_M2_M5=0 \
-    PROMPTFORGE_OUTPUT_K8_STRICT_GREEDY=0 \
    "$LLAMA_BIN" -m "$MODEL" --alias "$ALIAS" \
    --host "$BIND_ADDR" --port "$LISTEN_PORT" --jinja \
    -dev ROCm0 -ngl 999 -np "$PARALLEL" -c "$CTX_SIZE" \
    -b "$BATCH_SIZE" -ub "$UBATCH_SIZE" -fa on \
    -ctk "$CACHE_TYPE_K" -ctv "$CACHE_TYPE_V" -t "$NUM_THREADS" -tb "$THREAD_BATCH" \
    -ctxcp "$CTX_CHECKPOINTS" --cache-ram "$CACHE_RAM" --cache-prompt --cache-idle-slots --metrics \
-    --kairic-edge --spec-type draft-mtp \
+    --spec-type draft-mtp --model-draft "$MTP_DRAFT_FILE" \
+    --mmproj "$MODEL_DIR/$MMPROJ_FILE" \
    --spec-draft-device ROCm0 --spec-draft-ngl 999 \
    --spec-draft-type-k "$CACHE_TYPE_K" --spec-draft-type-v "$CACHE_TYPE_V" \
    --spec-draft-n-max "$MTP_N_MAX" --spec-draft-n-min "$MTP_N_MIN" \
@@ -1,6 +1,5 @@
 #!/usr/bin/env bash
-# Seam: frank-default-large model, runtime, Kairic Edge spec, template.
-# Covers the Kairic Edge deployment contract (ciru-ai/ROCmFPX v1.2).
+# Seam: frank-default-large model, ROCmFPX runtime, MTP, KV cache, template.
 #
 # load helpers

@@ -11,34 +10,47 @@ SERVE_FILE="$REPO_ROOT/scripts/serve-frank-large.sh"
 UNIT_FILE="$REPO_ROOT/scripts/frank-default-large.service"
 TEMPLATE_FILE="$REPO_ROOT/scripts/chat-template.jinja"

-@test "frank-default-large serves the Kairic Edge model and sidecars" {
-    grep -q '^MODEL_FILE=Qwen3.8-27B-IU4-Kairic-Edge\.gguf$' "$ENV_FILE"
-    grep -q '^KAIRIC_FFN_SIDECAR=' "$ENV_FILE"
-    grep -q '^KAIRIC_GDN_SIDECAR=' "$ENV_FILE"
-    grep -q '^KAIRIC_GDN_OUTPUT_SIDECAR=' "$ENV_FILE"
-    grep -q -- '--kairic-edge' "$SERVE_FILE"
+@test "frank-default-large serves the kingjones AGENT model" {
+    grep -q '^MODEL_FILE=Qwen3\.8-27B-Q8_0_ROCMFPX_AGENT\.gguf$' "$ENV_FILE"
+    grep -q '^MTP_DRAFT_FILE=' "$ENV_FILE"
+    ! grep -q -- '--kairic-edge' "$SERVE_FILE"
 }

-@test "frank-default-large uses the Kairic Edge ROCmFPX v1.2 runtime" {
-    grep -q 'ROCmFPX/build-kairic/bin/llama-server' "$ENV_FILE"
-    grep -q '^KAIRIC_EDGE_COMPATIBILITY_MODE=1$' "$ENV_FILE"
-    grep -q '205a3e5f40e5542e2f2eb68e3d3f81f918b1d895' "$ENV_FILE"
+@test "frank-default-large uses the ROCmFPX runtime" {
+    grep -q 'ROCmFPX/.*/bin/llama-server' "$ENV_FILE"
+    grep -q 'build-kairic/bin/llama-server' "$ENV_FILE"
 }

-@test "frank-default-large runs Kairic Edge MTP in compatibility mode" {
+@test "frank-default-large runs standalone MTP" {
    grep -q -- '--spec-type draft-mtp' "$SERVE_FILE"
-    grep -q -- '--kairic-edge' "$SERVE_FILE"
+    grep -qF -- '--model-draft "$MTP_DRAFT_FILE"' "$SERVE_FILE"
    grep -qF -- '--spec-draft-n-max "$MTP_N_MAX"' "$SERVE_FILE"
-    grep -qF -- '--spec-draft-p-split "$MTP_P_SPLIT"' "$SERVE_FILE"
-    grep -q '^KAIRIC_EDGE_COMPATIBILITY_MODE=1$' "$ENV_FILE"
    grep -q '^MTP_N_MAX=4$' "$ENV_FILE"
 }

+@test "frank-default-large has no obsolete PromptForge sidecar wiring" {
+    ! grep -qE 'KAIRIC_|PROMPTFORGE_' "$SERVE_FILE"
+    ! grep -qE 'KAIRIC_|PROMPTFORGE_' "$ENV_FILE"
+}
+
@test "frank-default-large fits the model and KV cache in unified memory" {
-    grep -q '^CTX_SIZE=65536$' "$ENV_FILE"
+    grep -q '^CTX_SIZE=262144$' "$ENV_FILE"
    grep -q ': "${CTX_SIZE:=262144}"' "$SERVE_FILE"
 }

+@test "frank-default-large has reasoning and preserved reasoning/thinking enabled" {
+    grep -q '^REASONING=on$' "$ENV_FILE"
+    grep -q 'enable_thinking":true' "$ENV_FILE"
+    grep -q 'preserve_thinking":true' "$ENV_FILE"
+    grep -q 'preserve_reasoning":true' "$ENV_FILE"
+}
+
+@test "frank-default-large uses Turbo4 only for the V cache" {
+    grep -q '^CACHE_TYPE_K=f16$' "$ENV_FILE"
+    grep -q '^CACHE_TYPE_V=turbo4$' "$ENV_FILE"
+    grep -qF -- '-ctk "$CACHE_TYPE_K" -ctv "$CACHE_TYPE_V"' "$SERVE_FILE"
+}
+
@test "frank-default-large exposes a swappable ubatch" {
    grep -q '^UBATCH_SIZE=512$' "$ENV_FILE"
    grep -q ': "${UBATCH_SIZE:=512}"' "$SERVE_FILE"