docs: frank-large vLLM MiMo-BF16 runbook; retire the gufo runbook
New runbook carries the roof math, the measured three-arm spec table, the greedy-text numerics trap, the reasoning-field contract change and the rollback tag. LLM index and AGENTS.md fleet table, bench policy and runbook index follow; the frank-small index entries catch up to the SGLang cutover (entry was missed on 2026-10-01). Work log records the cutover.
This commit is contained in:
Frank
@@ -12,7 +12,7 @@ merge absorbed the former `vllmScripts` repo, history preserved).
|
||||
Calibre-Web, Jellyfin, Plex) plus the `frank-primary` LLM service.
|
||||
- **Dankclaw** (192.168.0.143) — AI/search/dev/observability stack (SearXNG
|
||||
via Gluetun, ComfyUI, OpenWebUI, Hermes, Gitea, Grafana, Prometheus, Loki,
|
||||
Alloy, Pi-hole) plus TWO LLM services: `frank-large` (stock llama.cpp on
|
||||
Alloy, Pi-hole) plus TWO LLM services: `frank-large` (vLLM TheRock on
|
||||
the gfx1151 iGPU) and `frank-small` (SGLang on the RTX 3090). Caddy on
|
||||
Dankclaw reverse-proxies both hosts.
|
||||
- **FrankBook-Pro** (192.168.0.112) — macOS mini running the *arr stack
|
||||
@@ -25,7 +25,7 @@ merge absorbed the former `vllmScripts` repo, history preserved).
|
||||
| Target | Service | Runtime | Model | Port |
|
||||
|--------|---------|---------|-------|------|
|
||||
| Dankserver | `frank-primary` | tcclaviger RDNA4 vLLM fork (`docker.io/tcclaviger/vllm@sha256:da4029666…`, digest-pinned, plain `podman run`, root systemd unit) | `orcarouter/Qwen3.8-Flash-Next-Uncensored-FP8` (~172.8 GiB, 131 shards), TP=2, expert offload automatic (fork plans it), prefetch OFF, **bf16 KV** (`--kv-cache-dtype auto`), **no speculative decode**, MNBT 2048, MoE gather 32×64, **2 seqs × 262144 ctx (the model's full native window)** at KV_BYTES 7,300,000,000/rank (bf16 measures 13,839 B/token), with **`--expert-precise-memory`** — the fitted 5.25 GiB overhead instead of the 7.60 GiB default, which hands the compile/graph-capture headroom to the expert arena (~7,278 slots vs 6,248) — util 0.98, and the client contract wired: `--reasoning-parser qwen3` (thinking in `reasoning`, never inline), `--enable-auto-tool-choice --tool-call-parser qwen3_xml`, default chat-template kwargs `enable_thinking` + `preserve_thinking`. Capacity history: five full-length contexts boot but leave ~0 bytes free at runtime (a real request with `max_tokens=261149` killed the worker: `Tried to allocate 50.00 MiB ... 0 bytes free`); six never fit (6×241664 asked 10.70 GiB vs 9.18 GiB free). DCP cannot help here — it requires TP > the model's 2 KV heads, i.e. four GPUs. fp8+MTP4 shape restorable from git tag `frank-primary-fp8-mtp-20261001` | 7770 |
|
||||
| Dankclaw | `frank-large` | gufo `gufo-org/gufo` v0.1.1, the Strix Halo engine, BUILT FROM SOURCE (root systemd unit; links the host ROCm 7.2.4) | `unsloth/Qwen3.8-27B-GGUF` **UD-Q4_K_XL** (27B dense, **vanilla — no longer abliterated**, 16.35 GiB, Apache-2.0) with the z-lab DFlash2 Q4_K_M drafter (adaptive block policy, cap 7) on the gfx1151 iGPU (Strix Halo LPDDR, 114 GiB GTT), **4 sessions x 163840 ctx, unquantized (f16) KV** — gufo has no KV-quant lever — greedy, gufo's compiled-in Qwen renderer (NO Qwen-Sharp), text only. Swapped 2026-10-01 from ajgazin's abliterated Swift-1.5 UD-Q8_K_XL (29.3 GiB): ~13 GiB less weights is what funds four f16 sessions, at the cost of the uncensored behaviour | 7770 |
|
||||
| Dankclaw | `frank-large` | vLLM 0.30 in the digest-pinned `docker.io/kyuz0/vllm-therock-gfx1151` image (TheRock: ROCm 10 / torch 2.11, Triton Gated DeltaNet kernels; plain `podman run`, root systemd unit) | `XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B` **full BF16** (9B dense, 17.53 GiB, 4 shards — the arch frank-small serves as INT8 on the 3090), **cross-model MTP spec decode from `Qwen/Qwen3.5-9B` (2 draft tokens; 19.8 t/s decode measured, +51% over the 13.1 t/s base, which is ~94% of the LPDDR5x BF16 roof)**, 4 seqs x 262144 native ctx at util 0.45, qwen3 reasoning + qwen3_coder tool parsers, text-only (`--language-model-only`). Cut over 2026-10-01 from gufo; rollback tag `frank-large-gufo-q4kxl-20261001` | 7770 |
|
||||
| Dankclaw | `frank-small` | SGLang in a self-contained python3.12 venv (root systemd unit; the venv carries its own CUDA 13 stack — see the runbook) | letechlead MiMo-V2.6-Distill-Qwen-9B-INT8-W8A16-AutoRound (qwen3_5 hybrid, 9B dense, 12.0 GiB, INT8 W8A16 AutoRound, in-checkpoint vision tower, NO MTP weights shipped so no spec decode) on the RTX 3090, fp8 KV, pinned 606,208-token pool / 2 slots of 262,144 (native window; 2026-10-01 reslice from 4 x 151,552 for compaction headroom), bf16 mamba states, eager prefill, 83.9 tok/s decode | 7771 |
|
||||
|
||||
Retired endpoints (`frank-npu`, and every prior runtime) are in
|
||||
@@ -209,14 +209,14 @@ archive.
|
||||
**STOP the production service BEFORE benchmarking** — these servers serve
|
||||
real users, and a bench beside a live service starves the GPU and corrupts
|
||||
results. Use the runtime's own harness: `llama-bench` for the llama.cpp
|
||||
endpoints, `gufo bench` for frank-large. Rolling your own bench
|
||||
endpoints; frank-large's vLLM shape is measured by replaying real text through the live server and reading `Server-Timing`. Rolling your own bench
|
||||
requires asking first.
|
||||
Scratch ports are not used on LLM hosts — a
|
||||
test process cannot share the single GPU anyway; validate against the service
|
||||
itself.
|
||||
|
||||
**`gufo bench` cannot rank speculative-decode settings, and will invert the
|
||||
ranking.** Measured 2026-09-28 on frank-large at depth 32768: its corpus
|
||||
(A gufo-era law; frank-large is vLLM since 2026-10-01, but the rollback tag keeps it live.) **`gufo bench` cannot rank speculative-decode settings, and will invert the ranking.**
|
||||
Measured 2026-09-28 on frank-large at depth 32768: its corpus
|
||||
reports `acceptance=0.933 ± 0.000` with an **identical trace sha256** across
|
||||
draft widths, i.e. the same highly-predictable text every run, and 93 % of
|
||||
every proposal accepted. It reported 29–37 t/s decode where real traffic
|
||||
@@ -241,7 +241,7 @@ one: `ip neigh show <ip>` from a peer — FAILED means off/NIC dead, while a
|
||||
merely wedged kernel usually still answers ARP.
|
||||
|
||||
1. Stop and mask (self-healing would restart it):
|
||||
- Dankclaw `frank-large` (root unit, gufo on the iGPU):
|
||||
- Dankclaw `frank-large` (root unit, vLLM podman on the iGPU):
|
||||
`sudo systemctl stop frank-large && sudo systemctl mask frank-large`
|
||||
- Dankclaw `frank-small` (root unit, SGLang on the 3090):
|
||||
`sudo systemctl stop frank-small && sudo systemctl mask frank-small`
|
||||
@@ -281,8 +281,8 @@ of 2026-10-01 both `frank-small` and `frank-primary` accept images
|
||||
(`input: [text, image]`) — frank-primary loads the BF16 mmproj explicitly
|
||||
(this build never autoloads projectors), and frank-small's MiMo payload
|
||||
bakes its vision tower INTO the checkpoint (no projector file at all);
|
||||
`frank-large` remains text-only because its gufo deployment loads no
|
||||
vision projector. Note also that
|
||||
`frank-large` stays text-only because its vLLM deployment passes
|
||||
`--language-model-only` — the MiMo vision tower is in the checkpoint; the flag is the lever. Note also that
|
||||
the reasoning channel differs by runtime: llama.cpp AND SGLang (qwen3
|
||||
parser, live-verified) emit `reasoning_content`, vLLM emits `reasoning` —
|
||||
models.yml sets `reasoningContentField` per endpoint to match.
|
||||
@@ -292,111 +292,75 @@ endpoint on 2026-09-11, returned 2026-09-17 with the Gemma4 projector's
|
||||
audio encoder, and left again with the Gemma4 retirement on 2026-10-01
|
||||
(the MiMo tower is vision-only).
|
||||
|
||||
### frank-large — gufo, Swift-1.5-Qwen3.8-27B-Uncensored UD-Q8_K_XL (Dankclaw, ACTIVE)
|
||||
### frank-large — vLLM TheRock, MiMo-V2.6-Distill-Qwen-9B BF16 + cross-model MTP (Dankclaw, ACTIVE)
|
||||
|
||||
**gufo** (`gufo-org/gufo`) v0.1.1 — "the Strix Halo inference engine", a
|
||||
vertical engine built for exactly this hardware (Ryzen AI MAX+ 395 /
|
||||
Radeon 8060S, `gfx1151`) — as a **root systemd unit**
|
||||
(`scripts/frank-large.service`, `User=dank`), serving
|
||||
`ajgazin/Swift-1.5-Qwen3.8-27B-Uncensored-Dynamic-MTP-GGUF`'s `UD-Q8_K_XL`
|
||||
(27B dense, 29.3 GiB, sha256-verified; **abliterated/uncensored**, and under
|
||||
the *non-OSI* SWIFT OPEN LICENSE 1.0) on port 7770 from the gfx1151 iGPU.
|
||||
One session, full 262,144 ctx (the model's native window, `SESSIONS=1`),
|
||||
DFlash2 spec decode (the z-lab `Qwen3.8-27B-DFlash2-Q4_K_M` drafter, gufo's
|
||||
**adaptive** block policy, `--draft-tokens 7`), greedy sampling, reasoning
|
||||
extracted to `reasoning_content`, and **no vision projector**.
|
||||
**vLLM 0.30** in the **digest-pinned** `docker.io/kyuz0/vllm-therock-gfx1151`
|
||||
image (image ID `2d24a8c5a3fd`, digest `sha256:98c1f3ac9833…` — TheRock
|
||||
build: ROCm 10.0.0 / torch 2.11.0, Triton Gated DeltaNet kernels,
|
||||
`vllm/v1/spec_decode/dflash.py` present) — as a **root systemd unit**
|
||||
(`scripts/frank-large.service`, `User=dank`, `Restart=always`,
|
||||
`ExecStop=podman stop -t 120`, `TimeoutStopSec` 180 > the stop grace),
|
||||
serving `XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B` in **full BF16** (9B dense
|
||||
qwen3_5 hybrid, 17.53 GiB, 4 sha256-verified shards) on port 7770 from the
|
||||
gfx1151 iGPU. Official vLLM images are CUDA-only; this fork is the fleet's
|
||||
gfx1151 path. `scripts/serve-frank-large.sh` gates image-by-repo-digest →
|
||||
payload (config + index + exact shard count) → spec draft (accepts BOTH
|
||||
weight layouts: single `model.safetensors` for dflash, sharded index for
|
||||
the MTP head) → whole `/dev/dri` (never individual renderD nodes), sweeps
|
||||
the bench identity `vllm-mimo`, then `exec podman wait` so a container
|
||||
death trips `Restart=always`. Boot ~200–240 s.
|
||||
|
||||
**Two swaps landed on 2026-09-28**, both same day: the engine (from the
|
||||
unsloth llama.cpp b11030 prebuilt, which had replaced Kairic Edge ROCmFPX)
|
||||
and then the payload (from `unsloth/Qwen3.8-27B-GGUF:UD-Q8_K_XL`, still on
|
||||
disk and the payload-only rollback). The two targets are structurally
|
||||
interchangeable for gufo — same `qwen35` architecture, the same 866 tensors
|
||||
in the same order, the same per-tensor Q8_0/F32/BF16 layout, the same MTP
|
||||
head at `blk.64` — verified by parsing and diffing both GGUF headers before
|
||||
either was loaded. **The DFlash2 drafter did not move**, so it has its own
|
||||
`DRAFTER_DIR` and the deployer's stray-run sweep keys on the drafter's
|
||||
filename, not the target's.
|
||||
**The shape is chosen by roof math, and the spec decode is the
|
||||
performance feature.** BF16 decode is LPDDR5x-bandwidth-bound here:
|
||||
13.1 t/s measured = ~94% of the roof for 17.5 GiB of weights — a quant,
|
||||
not a bug. The only way past it is multiple tokens per weight pass.
|
||||
Measured 2026-10-01 (one session, temp 0, 28,647-token prefill / 512
|
||||
decode): base **17,876 t/s prefill / 13.1 decode**; +DFlash-8 7,976 /
|
||||
17.4 (acceptance 59%/26% per position); **+MTP-2 7,380 / 19.8
|
||||
(acceptance 71%/44%, token rate 57.4%)** — deployed arm.
|
||||
`num_speculative_tokens` beyond 2 is UNMEASURED; both spec arms pay
|
||||
~55–59% of prefill. KV grows only on the 8 full-attention layers
|
||||
(32 KiB/token bf16); the 24 GDN layers hold fixed ~25 MiB state per
|
||||
sequence — so 4 x 262,144 at util 0.45 fits the GTT aperture easily.
|
||||
|
||||
**The runtime is built from source on the host** —
|
||||
`/home/dank/GitPackages/gufo-v0.1.1/`, reproducible with
|
||||
`scripts/build-gufo.sh` (`--verify` re-checks an installed tree). gufo ships
|
||||
no prebuilt Linux binaries and the v0.1.0/v0.1.1 release assets are empty.
|
||||
Unlike the retired llama.cpp prebuilt, which bundled its own ROCm, **gufo
|
||||
links the host ROCm 7.2.4 through its RPATH**, so the HIP dev packages must
|
||||
stay installed. Two host workarounds the build script encodes:
|
||||
**The MTP head is CROSS-MODEL — the fleet's first.** MiMo ships ZERO
|
||||
`mtp.*` tensors (the reason frank-small's SGLang NEXTN is a 0.00 ghost:
|
||||
SGLang reads the head from the TARGET checkpoint); vLLM reads the draft
|
||||
from a separate path, so `--speculative-config
|
||||
'{"method":"mtp","model":"/models/Qwen3.5-9B","num_speculative_tokens":2}'`
|
||||
drafts from the distill's SOURCE model, and 71%/44% acceptance is the
|
||||
measured proof that the distillation preserved Qwen3.5 hidden-state
|
||||
structure. The alternative drafter (`z-lab/Qwen3.5-9B-DFlash`, fetched;
|
||||
its "PR #40898 unmerged" card claim is false — dflash ships in vLLM 0.30)
|
||||
stays on disk: `FRANK_LARGE_SPEC=dflash` + repoint `SPEC_DIR`.
|
||||
|
||||
- **Ubuntu's `librocwmma-dev` must be absent.** It installs a truncated
|
||||
rocWMMA header set into `/usr/include` that shadows AMD's complete one
|
||||
and breaks the build with `'internal/accessors.hpp' file not found`. A
|
||||
`-I` cannot fix it — in HIP mode the host pass searches the system
|
||||
directories first. The package has no reverse dependencies.
|
||||
- **`--gcc-install-dir` must name GCC 15.** Ubuntu 26.04 ships
|
||||
`/usr/include/c++/16`, which the ROCm clang prefers; its `<atomic_wait.h>`
|
||||
needs a `std::__detail::__notify_impl` symbol version the installed
|
||||
`libstdc++.so.6` (GLIBCXX_3.4.35) does not export, and the link fails.
|
||||
**Spec decode does NOT preserve greedy text here (measured).**
|
||||
Temperature-0 completions differ across arms (bf16 batched verification
|
||||
reorders reductions; near-tie argmaxes diverge and cascade). vLLM spec
|
||||
decode is distribution-preserving, not bitwise-greedy-preserving — do not
|
||||
chase text drift between arms as a regression.
|
||||
|
||||
`gufo diagnose` is the preflight and must report `[PASS]` with `gfx1151`.
|
||||
It also self-caps compute queues (`GPU_MAX_HW_QUEUES`, 2 resident for
|
||||
`serve llm`) after reading `/sys/class/kfd/kfd/proc/*/queues/*/type`, which
|
||||
matters because ComfyUI holds queues on the same device; four HIP processes
|
||||
is the device ceiling.
|
||||
**Client contract differs from every llama.cpp-era endpoint:** vLLM emits
|
||||
thinking in **`reasoning`**, not `reasoning_content` — `models.yml` sets
|
||||
`reasoningContentField: "reasoning"` for this provider (miss it and
|
||||
thinking vanishes from transcripts). Tools ride
|
||||
`--enable-auto-tool-choice --tool-call-parser qwen3_coder`. The harness
|
||||
extraBody moved to the Qwen3.5-family recipe (0.7/0.8) with
|
||||
`chat_template_kwargs.{enable_thinking,preserve_thinking}` — a top-level
|
||||
`preserve_thinking` is ignored by vLLM. Env keys stay
|
||||
`FRANK_LARGE_`-prefixed (HIP reads bare `TEMP`/`HOST`/`GPU_MAX_HEAP_SIZE`).
|
||||
|
||||
**The client contract changed, and the chat template is the sharp edge.**
|
||||
**gufo has no Jinja engine at all** — `src/models/qwen/chat_template.cpp` is
|
||||
587 lines of hand-written C++ that appends strings, with no
|
||||
`--chat-template-file`, no env override, and no template body to point at.
|
||||
`scripts/qwen-sharp-v22.4.0.jinja` therefore **cannot** be supplied to
|
||||
frank-large: that is a port, not a config line. Given the choice on
|
||||
2026-09-28 the decision was to ship on gufo's stock renderer and accept the
|
||||
loss — the `terse` system-prompt injection, the in-band `<|think_*|>`
|
||||
control, the `json|xml` `tool_call_format` switch, the tool-argument
|
||||
truncation and `add_vision_id` prefixes are all gone. gufo's own renderer
|
||||
does keep the effort instruction strings and `preserve_thinking`, which are
|
||||
what the harness depends on. **This is the thing to watch**: the endpoint
|
||||
now runs an *abliterated* model without the prompt that tells it to stop
|
||||
waffling. The cheap fix, if verbosity shows up, is to send the `terse` text
|
||||
as an ordinary system message from `models.yml` — not a template change.
|
||||
`reasoning_effort` accepts the full OpenAI set (all six levels verified
|
||||
200), so `models.yml`'s `reasoningEffortMap` fold is no longer *required* —
|
||||
it is kept deliberately, so that removing it does not silently move every
|
||||
`high` role from `xhigh` to `high` effort inside a swap. And there is no
|
||||
KV-quantization or prompt-cache-RAM flag; prompt reuse is on by default and
|
||||
`--cache-disk` is opt-in and off.
|
||||
Fetch: `scripts/download-frank-large-mimo.sh` (target),
|
||||
`scripts/download-frank-large-qwen35.sh` (MTP source),
|
||||
`scripts/download-frank-large-dflash.sh` (dflash alternative) — all
|
||||
`-f` + byte-count + LFS sha256 verified, `--verify-only` supported, no
|
||||
token on the command line.
|
||||
|
||||
**Read the perf trade before assuming a win.** gufo's own Q8_K_XL tables
|
||||
(vs llama.cpp b11069, DFlash2) show prefill +37 % to +80 %, cold load 4.5×
|
||||
faster, decode +7 % at zero depth and +40 % at 4 users — but **−14 % at
|
||||
65,536 tokens of depth and −24 % at 131,072**. On the 8-bit target the
|
||||
crossover is ~32k of cached context, and frank-large runs one 262,144-token
|
||||
conversation, so long agent sessions sit on the losing side of that line.
|
||||
The Q4_K_XL target shows no such regression; revisiting the quantization is
|
||||
a separate decision from the runtime. Peak memory is 50.58 GiB at the full
|
||||
262,144 context against the iGPU's 114 GiB GTT aperture.
|
||||
|
||||
The serve script (`scripts/serve-frank-large.sh`) validates runtime → target
|
||||
GGUF → drafter and exits with a pointer message if any is missing, then
|
||||
`exec`s `gufo serve --host 0.0.0.0 --port 7770 --sessions 1 llm --model
|
||||
…--served-model-name frank-large --context 262144 --speculative dflash2 …`.
|
||||
The deployer's `deploy_llm_large` gates on the **runtime binary** as well
|
||||
as both payload paths, and returns before it stops anything.
|
||||
|
||||
Fetch: `scripts/download-frank-large-swift.sh` (the target; per-file byte
|
||||
count + sha256 against the Hub LFS oid, no auth header — the repo is public
|
||||
and a token on the curl command line is world-readable) and
|
||||
`scripts/download-frank-large.sh` (the DFlash2 drafter, which did not move).
|
||||
Both take `--verify-only`.
|
||||
|
||||
Runbook: `serverAppConfigs-docs/LLM/frank-large-gufo-qwen38-27b.md`.
|
||||
Rollback, two ways: **payload only** — re-fetch with
|
||||
`scripts/download-frank-large.sh` and repoint `MODEL_FILE` at
|
||||
`/home/dank/models/Qwen3.8-27B-UD-Q8_K_XL/Qwen3.8-27B-UD-Q8_K_XL.gguf`; that
|
||||
29.3 GiB target was **reclaimed on 2026-09-28** (staged to
|
||||
`~/.AGENTTRASH/20260928-retired-models/`, restorable from there), so it is
|
||||
no longer a one-minute flip. **Engine** —
|
||||
[[frank-large-unsloth-qwen38-27b-q8xl]], the b11030 prebuilt at
|
||||
`/home/dank/GitPackages/unsloth-llama-b11030/`; the only way to get
|
||||
Qwen-Sharp back. The DFlash2 drafter was never reclaimed and still lives in
|
||||
`Qwen3.8-27B-UD-Q8_K_XL/`.
|
||||
Runbook: `serverAppConfigs-docs/LLM/frank-large-vllm-mimo-v26-bf16.md`.
|
||||
Rollback: git tag **`frank-large-gufo-q4kxl-20261001`** — gufo v0.1.1
|
||||
(`scripts/build-gufo.sh` stays on disk for it), the Q4_K_XL Qwen3.8-27B
|
||||
payload and the DFlash2 drafter all remain on the host; restore the tag's
|
||||
unit/env/serve trio and redeploy.
|
||||
|
||||
### frank-small — SGLang on the RTX 3090 (Dankclaw, ACTIVE)
|
||||
|
||||
@@ -827,7 +791,7 @@ Active:
|
||||
|
||||
| Service | Runbook |
|
||||
|---------|---------|
|
||||
| frank-large (gufo Qwen3.8-27B UD-Q8_K_XL, Dankclaw) | `serverAppConfigs-docs/LLM/frank-large-gufo-qwen38-27b.md` |
|
||||
| frank-large (vLLM TheRock MiMo-V2.6 BF16 + cross-model MTP, Dankclaw) | `serverAppConfigs-docs/LLM/frank-large-vllm-mimo-v26-bf16.md` |
|
||||
| frank-small (SGLang MiMo-V2.6 INT8, Dankclaw) | `serverAppConfigs-docs/LLM/frank-small-sglang-mimo-v26.md` |
|
||||
| frank-primary (tcclaviger vLLM, Qwen3.8-Flash-Next FP8 + MTP4, Dankserver) | `serverAppConfigs-docs/LLM/dankserver-qwen38-fp8-expert-offload-tuning.md` |
|
||||
| Dankclaw kernel posture (`amd_iommu=off`) | `serverAppConfigs-docs/LLM/dankclaw-iommu-translation-off.md` |
|
||||
@@ -837,6 +801,7 @@ Retired (rollback targets and history):
|
||||
|
||||
| Runtime | Host | Window | Runbook |
|
||||
|---------|------|--------|---------|
|
||||
| gufo v0.1.1 Qwen3.8-27B UD-Q8_K_XL/UD-Q4_K_XL + DFlash2 (`frank-large`) | Dankclaw | 2026-09-28 → 10-01 | `frank-large-gufo-qwen38-27b.md` |
|
||||
|tcclaviger vLLM GLM-5.3-Flash W4A16 int4, +/- DFlash2 drafter (`frank-primary`)|Dankserver|2026-09-30 → 10-01|`frank-primary-glm53-flash-w4a16.md`|
|
||||
| llama.cpp b11030 Qwen3.8-27B UD-Q8_K_XL + DFlash2 (`frank-large`) | Dankclaw | 2026-09-21 → 09-28 | `frank-large-unsloth-qwen38-27b-q8xl.md` |
|
||||
| R9V radiance-vLLM Qwen3.8-Flash-Next-Uncensored IQ4_XS (`frank-primary`) | Dankserver | 2026-09-27 (same-day swap to Q8_0) | `frank-primary-r9v-radiance-qwen38-uncensored.md` |
|
||||
|
||||
@@ -1,7 +1,7 @@
|
||||
---
|
||||
type: reference
|
||||
title: "gufo, Swift-1.5-Qwen3.8-27B-Uncensored UD-Q8_K_XL (DFlash2) on Dankclaw (frank-large)"
|
||||
description: "The active frank-large runbook — gufo (gufo-org/gufo) v0.1.1, the Strix Halo inference engine, built from source on Dankclaw and serving the abliterated ajgazin/Swift-1.5-Qwen3.8-27B-Uncensored UD-Q8_K_XL with the z-lab DFlash2 Q4_K_M drafter on the gfx1151 iGPU: the from-source build and its two host workarounds, the GGUF header evidence that made the payload swap safe, the one-session / full-window shape, why the fleet's Qwen-Sharp chat template CANNOT be used on this runtime and what that costs, the client-contract delta, the measured prefill/decode trade, and the failure modes worth knowing."
|
||||
description: "The RETIRED frank-large runbook (2026-09-28 → 2026-10-01, rollback target via tag frank-large-gufo-q4kxl-20261001) — gufo (gufo-org/gufo) v0.1.1, the Strix Halo inference engine, built from source on Dankclaw and serving the abliterated ajgazin/Swift-1.5-Qwen3.8-27B-Uncensored UD-Q8_K_XL with the z-lab DFlash2 Q4_K_M drafter on the gfx1151 iGPU: the from-source build and its two host workarounds, the GGUF header evidence that made the payload swap safe, the one-session / full-window shape, why the fleet's Qwen-Sharp chat template CANNOT be used on this runtime and what that costs, the client-contract delta, the measured prefill/decode trade, and the failure modes worth knowing."
|
||||
tags:
|
||||
- llm
|
||||
- gufo
|
||||
@@ -18,7 +18,11 @@ timestamp: 2026-09-28
|
||||
|
||||
# frank-large — gufo, Swift-1.5-Qwen3.8-27B-Uncensored UD-Q8_K_XL (DFlash2)
|
||||
|
||||
> **ACTIVE since 2026-09-28.** Replaces
|
||||
> **RETIRED 2026-10-01** — replaced by [[frank-large-vllm-mimo-v26-bf16]]
|
||||
> (vLLM TheRock + MiMo-V2.6 BF16). Rollback: git tag
|
||||
> `frank-large-gufo-q4kxl-20261001`; the gufo build (`build-gufo.sh`), the
|
||||
> Q4_K_XL payload and the DFlash2 drafter all remain on the host.
|
||||
> **Originally ACTIVE 2026-09-28 → 2026-10-01.** Replaces
|
||||
> [[frank-large-unsloth-qwen38-27b-q8xl]] (stock llama.cpp, the unsloth
|
||||
> b11030 prebuilt), which had replaced
|
||||
> [[frank-large-kairic-edge-qwen38-27b]]. Same host, same port (7770), same
|
||||
|
||||
@@ -0,0 +1,176 @@
|
||||
---
|
||||
type: reference
|
||||
title: "frank-large on Dankclaw — vLLM TheRock, MiMo-V2.6-Distill-Qwen-9B BF16 + cross-model MTP"
|
||||
description: "The active frank-large runbook — the digest-pinned kyuz0 TheRock gfx1151 vLLM image serving XiaomiMiMo MiMo-V2.6-Distill-Qwen-9B in full BF16 on the Strix Halo iGPU: identity, the measured roof math, the cross-model Qwen3.5-9B MTP drafter and its acceptance curve, the greedy-text numerics trap, the client-contract field change, deployed flags, and what to check when it misbehaves."
|
||||
tags:
|
||||
- llm
|
||||
- frank-large
|
||||
- dankclaw
|
||||
- vllm
|
||||
- strix-halo
|
||||
- mimo
|
||||
- mtp
|
||||
- spec-decode
|
||||
- bf16
|
||||
timestamp: 2026-10-01
|
||||
---
|
||||
|
||||
# frank-large — vLLM MiMo-V2.6-Distill-Qwen-9B BF16 (cross-model MTP)
|
||||
|
||||
> **ACTIVE since 2026-10-01.** Replaces [[frank-large-gufo-qwen38-27b]]
|
||||
> (gufo + Qwen3.8-27B GGUF, now the rollback target, git tag
|
||||
> `frank-large-gufo-q4kxl-20261001`). Same port (7770), same served alias
|
||||
> (`frank-large`). This is the fleet's second vLLM endpoint and the first
|
||||
> CROSS-MODEL speculative-decode deployment: the target is Xiaomi's distill,
|
||||
> the draft head is Qwen's.
|
||||
|
||||
## Identity
|
||||
|
||||
- **Endpoint:** `frank-large` at `192.168.0.143:7770` (Dankclaw).
|
||||
- **Runtime:** vLLM 0.30.0 in the **digest-pinned** `docker.io/kyuz0/vllm-therock-gfx1151`
|
||||
image (image ID `2d24a8c5a3fd`, digest `sha256:98c1f3ac9833…` — ROCm 10.0.0 /
|
||||
torch 2.11.0 / TheRock, Triton Gated DeltaNet kernels). The official vLLM
|
||||
images are CUDA-only; this fork is the fleet's gfx1151 path. Plain
|
||||
`podman run` driven by `scripts/serve-frank-large.sh` under a root unit
|
||||
(`scripts/frank-large.service`, `User=dank`, `Restart=always`,
|
||||
`ExecStop=podman stop -t 120`, `TimeoutStopSec=180` > the stop grace).
|
||||
- **Model:** `XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B` in **full BF16**
|
||||
(17.53 GiB, 4 shards) — the same checkpoint frank-small serves as an INT8
|
||||
W8A16 AutoRound conversion on the 3090, here unquantized.
|
||||
- **Draft:** `Qwen/Qwen3.5-9B` (18 GiB on disk; only the 15 `mtp.*` tensors
|
||||
are loaded into the draft pass).
|
||||
|
||||
## Why this shape (the roof math)
|
||||
|
||||
Decode speed for a dense model on Strix Halo is bandwidth-bound:
|
||||
BF16 weights x one pass per token against the LPDDR5x aperture. Measured
|
||||
2026-10-01 on this host: **13.1 t/s is ~94% of the roof** — the payload, not
|
||||
a misconfiguration. The same INT8 payload reaches 83.9 t/s on the 3090
|
||||
because it moves ~12 GiB through ~936 GB/s. The only way past a bandwidth
|
||||
wall is generating more than one token per weight pass: **speculative decode
|
||||
is the performance feature of this endpoint**, not a garnish.
|
||||
|
||||
The memory side is the opposite of a squeeze: KV grows ONLY on the 8 full
|
||||
attention layers (32 KiB/token bf16: 8 x 2 x 4 heads x 256 dim x 2 B);
|
||||
the 24 Gated DeltaNet layers hold a FIXED ~25 MiB state per running
|
||||
sequence. `--max-model-len 262144` (the native window) with
|
||||
`--max-num-seqs 4` at `--gpu-memory-utilization 0.45` boots clean — the
|
||||
four-session contract the gufo shape ran, with the pool shared by the
|
||||
scheduler instead of carved per session.
|
||||
|
||||
## Spec method: cross-model MTP — measured, not assumed
|
||||
|
||||
| arm | prefill | decode | acceptance |
|
||||
|---|---|---|---|
|
||||
| BF16 base | 17,876 t/s | 13.1 t/s | — |
|
||||
| BF16 + DFlash 8 tok | 7,976 t/s | 17.4 t/s | pos0 59% pos1 26% (token rate 14.8%) |
|
||||
| **BF16 + MTP 2 tok** | 7,380 t/s | **19.8 t/s** | pos0 71% pos1 44% (token rate 57.4%) |
|
||||
|
||||
One session, temperature 0, 28,647-token prefill / 512-token decode, same
|
||||
client for all three arms (2026-10-01 bench window, frank-large stopped per
|
||||
Benchmarking Policy). **MTP wins (+51% vs +33%) and needs only 2 draft
|
||||
tokens** — acceptance past position 1 is 44% and falling; DFlash's own curve
|
||||
decays 59/26/14/7, so 8 proposals are mostly wasted verification.
|
||||
`num_speculative_tokens` beyond 2 is UNMEASURED; do not raise it without a
|
||||
fresh A/B. Both spec arms pay ~55–59% of prefill (the draft passes share
|
||||
the compute queues) — this endpoint's thinking-heavy traffic spends its wall
|
||||
clock in decode, which is the regime where that trade wins.
|
||||
|
||||
**Why a Qwen head accepts on a MiMo target:** MiMo-V2.6 is a distillation of
|
||||
Qwen3.5-9B that preserves the hidden-state geometry — identical
|
||||
4096-hidden/248320-vocab/32-layer/interval-4 structure, and the 71%/44%
|
||||
per-position acceptance is the direct measurement that the distillation
|
||||
preserved more than geometry. MiMo itself ships **ZERO `mtp.*` tensors**
|
||||
(verified in both the BF16 base and the INT8 conversion) — which is exactly
|
||||
why frank-small's SGLang NEXTN draft is a 0.00-acceptance ghost: SGLang
|
||||
reads the head from the TARGET checkpoint, vLLM reads the draft from a
|
||||
separate path. That difference is the whole trick:
|
||||
|
||||
```
|
||||
--speculative-config '{"method":"mtp","model":"/models/Qwen3.5-9B","num_speculative_tokens":2}'
|
||||
```
|
||||
|
||||
The alternative drafter (`z-lab/Qwen3.5-9B-DFlash`, 6 layers, target_layer_ids
|
||||
[1,5,…,29]) stays fetched on disk; flip `FRANK_LARGE_SPEC=dflash` and repoint
|
||||
`FRANK_LARGE_SPEC_DIR` to use it. Its card's "vLLM PR #40898 unmerged" claim
|
||||
is FALSE — `vllm/v1/spec_decode/dflash.py` ships in 0.30.
|
||||
|
||||
## The numerics trap (greedy text is NOT preserved by spec decode)
|
||||
|
||||
Measured: temperature-0 completions DIFFER across arms (sha `1318aeb6` base,
|
||||
`0208b45d` dflash, `20c41c38` mtp). Batched verification changes the bf16
|
||||
reduction order, argmax near-ties diverge, and the divergence cascades.
|
||||
vLLM spec decode is distribution-preserving, not bitwise-greedy-preserving.
|
||||
The gufo-era "greedy == autoregressive text" law does not carry over; do not
|
||||
chase text drift between spec and no-spec arms as a regression.
|
||||
|
||||
## Payload
|
||||
|
||||
- Target: 4 safetensors shards (17.53 GiB) + config/tokenizer under
|
||||
`/home/dank/models/MiMo-V2.6-Distill-Qwen-9B-BF16/`, each shard
|
||||
byte-count + LFS sha256-verified by `scripts/download-frank-large-mimo.sh`
|
||||
(`--verify-only` re-checks; no token on the curl command line).
|
||||
- MTP source: `scripts/download-frank-large-qwen35.sh` →
|
||||
`/home/dank/models/Qwen3.5-9B/` (4 shards behind
|
||||
`model.safetensors.index.json`; the shards carry the odd HF asset names
|
||||
`model.safetensors-0000N-of-00004.safetensors` — the index resolves them,
|
||||
the engine boots from exactly this tree).
|
||||
- DFlash drafter: `scripts/download-frank-large-dflash.sh` →
|
||||
`/home/dank/models/Qwen3.5-9B-DFlash/`.
|
||||
- The serve script gates the image by REPO DIGEST (rootless podman resolves
|
||||
tags from the network; the pin exists to boot exactly these bytes
|
||||
offline), then config + index + exact shard count, then the spec draft
|
||||
(either weight layout: single `model.safetensors` or a sharded index),
|
||||
then `/dev/kfd` + whole `/dev/dri` (never individual renderD nodes —
|
||||
Dankserver 2026-09-29 lesson).
|
||||
|
||||
## Client contract
|
||||
|
||||
- **Alias:** `frank-large` (the `id` at `/v1/models`).
|
||||
- **Reasoning:** `--reasoning-parser qwen3`; vLLM emits the thinking text in
|
||||
**`reasoning`**, NOT llama.cpp/gufo's `reasoning_content` — the harness
|
||||
`models.yml` compat block for this provider sets
|
||||
`reasoningContentField: "reasoning"`. Miss that and thinking silently
|
||||
vanishes from transcripts.
|
||||
- **Tools:** `--enable-auto-tool-choice --tool-call-parser qwen3_coder`.
|
||||
- **Vision:** the checkpoint carries the `qwen3_5_vision` tower but the
|
||||
serve script passes `--language-model-only` — the endpoint stays
|
||||
text-only by decision; dropping the flag brings vision back.
|
||||
- **Sampling:** request-side; the harness sends the Qwen3.5-family recipe
|
||||
(temp 0.7, top_p 0.8) via `compat.extraBody`, and
|
||||
`chat_template_kwargs.{enable_thinking,preserve_thinking}` — the
|
||||
`chat_template_kwargs` form is what vLLM reads; a top-level
|
||||
`preserve_thinking` is ignored there.
|
||||
- **Env hygiene:** every tunable is `FRANK_LARGE_`-prefixed because HIP
|
||||
reads bare `TEMP`/`HOST`/`GPU_MAX_HEAP_SIZE` and the serve script exports
|
||||
the env file with `set -a`.
|
||||
|
||||
## Ops
|
||||
|
||||
- Boot ~200–240 s warm; health 503 while loading (do not kill it).
|
||||
- The serve script sweeps stale container identities before starting:
|
||||
`frank-large-mimo` itself and the bench-window container `vllm-mimo`.
|
||||
- Container death → `podman wait` returns non-zero from ExecStart →
|
||||
`Restart=always` self-heals (a clean exit-0 drain also restarts, the
|
||||
frank-primary lesson).
|
||||
- Bench windows: `systemctl stop frank-large` is enough (mask fails on the
|
||||
relinked unit; the deploy restores).
|
||||
- First-deploy fact to record here: the boot-time KV pool line
|
||||
(`Maximum concurrency for 262,144 tokens per prompt: N.Nx`).
|
||||
|
||||
## Rollback
|
||||
|
||||
Git tag **`frank-large-gufo-q4kxl-20261001`** — gufo v0.1.1 (built from
|
||||
source, `scripts/build-gufo.sh` stays on disk) + the Q4_K_XL Qwen3.8-27B
|
||||
payload + DFlash2 drafter, all still on the host. Restoring the tag's
|
||||
unit/env/serve trio and redeploying returns the 27B endpoint.
|
||||
|
||||
# Citations
|
||||
|
||||
- Model: https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B
|
||||
- Draft source: https://huggingface.co/Qwen/Qwen3.5-9B
|
||||
- DFlash drafter: https://huggingface.co/z-lab/Qwen3.5-9B-DFlash
|
||||
- Engine image: https://hub.docker.com/r/kyuz0/vllm-therock-gfx1151
|
||||
- Serve script: `scripts/serve-frank-large.sh`; tunables: `scripts/frank-large.env`
|
||||
- Predecessor: [[frank-large-gufo-qwen38-27b]]; sibling on the same arch:
|
||||
[[frank-small-sglang-mimo-v26]]
|
||||
@@ -10,12 +10,12 @@ headroom; 6 do not fit), no speculative decode, MNBT 1024, MoE gather 32x64, uti
|
||||
reasoning/tool parsers wired (qwen3 / qwen3_xml) with preserve_thinking on,
|
||||
measured 36.4-39.4 tok/s decode — the 2026-10-01 bf16 no-spec cutover plus
|
||||
the night ladder; rollback frank-primary-fp8-mtp-20261001), and Dankclaw
|
||||
runs two endpoints —
|
||||
gufo v0.1.1 (the Strix Halo engine, built from source; frank-large,
|
||||
unsloth Qwen3.8-27B UD-Q4_K_XL (vanilla) + the z-lab DFlash2 Q4_K_M drafter
|
||||
on the gfx1151 iGPU, 4 x 163840 ctx with f16 KV, port 7770) and mainline llama.cpp (frank-small,
|
||||
HauhauCS Gemma4-12B QAT Q4_K_M with MTP and vision on the RTX 3090,
|
||||
port 7771).
|
||||
runs two endpoints on the same arch —
|
||||
vLLM (digest-pinned kyuz0 TheRock gfx1151 image; frank-large,
|
||||
XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B in FULL BF16 with cross-model MTP
|
||||
spec decode from the Qwen3.5-9B head on the gfx1151 iGPU, 4 x 262144 ctx,
|
||||
port 7770) and SGLang (frank-small, the same model as an INT8 W8A16
|
||||
AutoRound conversion on the RTX 3090, 2 x 262144 ctx, port 7771).
|
||||
|
||||
> 2026-08-22: the former `vllmScripts` repository is merged into the
|
||||
> serverAppConfigs repository. This knowledge base moved from the old repo's
|
||||
@@ -39,14 +39,14 @@ fp8 KV, MNBT 1024, gather 32x64, qwen3 parsers]
|
||||
subgraph Dankclaw[AMD Strix Halo - Ubuntu 26.04]
|
||||
E2[RTX 3090 (PCIe x4 dock) +
|
||||
gfx1151 iGPU 114 GiB GTT]
|
||||
E[stock llama.cpp (unsloth b11030, gfx1151)
|
||||
E[vLLM TheRock gfx1151, digest-pinned
|
||||
frank-large, port 7770]
|
||||
F[Qwen3.8-27B UD-Q8_K_XL + DFlash2 Q4_K_M drafter
|
||||
262144 ctx / 1 slot, DFlash2 depth 4, q8_0 KV]
|
||||
G[mainline llama.cpp 37b53fd
|
||||
F[MiMo-V2.6-Distill-Qwen-9B BF16 + Qwen3.5-9B MTP head
|
||||
4 x 262144 ctx, spec 2 tok, qwen3/qwen3_coder parsers]
|
||||
G[SGLang venv cu130
|
||||
frank-small, port 7771]
|
||||
H[Gemma4-12B QAT Q4_K_M
|
||||
4 x 262144 ctx, MTP + vision + audio]
|
||||
H[MiMo-V2.6 INT8 W8A16 AutoRound
|
||||
2 x 262144 ctx, pinned pool 606208, in-checkpoint vision]
|
||||
E2 --> E --> F
|
||||
E2 --> G --> H
|
||||
end
|
||||
@@ -162,14 +162,23 @@ frank-small, port 7771]
|
||||
(bs1/bs3/bs4/bs6, 8k–700k KV) on Dankserver RTX 3090
|
||||
- [[deployment]] — deploy and update process for both servers
|
||||
- [[frank-redo-2026-08-17]] — the 2026-08-17 topology change
|
||||
- [[frank-large-gufo-qwen38-27b]] — ACTIVE frank-large runbook since
|
||||
2026-09-28 (root systemd unit, port 7770): gufo (gufo-org/gufo) v0.1.1, the
|
||||
Strix Halo inference engine, built from source on Dankclaw, serving
|
||||
unsloth/Qwen3.8-27B-GGUF:UD-Q8_K_XL (27B dense) with the z-lab DFlash2
|
||||
Q4_K_M drafter (adaptive block policy) on the gfx1151 iGPU — one full
|
||||
262,144-token conversation, greedy, text-only, the two from-source build
|
||||
workarounds, the client-contract delta, the measured prefill/decode trade
|
||||
against llama.cpp, troubleshooting
|
||||
- [[frank-large-vllm-mimo-v26-bf16]] — ACTIVE frank-large runbook since
|
||||
2026-10-01 (root systemd unit, port 7770): the digest-pinned
|
||||
kyuz0/vllm-therock-gfx1151 image serving
|
||||
XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B in FULL BF16 on the gfx1151 iGPU,
|
||||
4 x 262144 ctx, with CROSS-MODEL MTP spec decode from Qwen/Qwen3.5-9B
|
||||
(the fleet's first; 71%/44% per-position acceptance measured) —
|
||||
the roof math (13.1 t/s base is ~94% of LPDDR5x BF16), the measured
|
||||
MTP-vs-DFlash table (19.8 vs 17.4 t/s decode), the greedy-text numerics
|
||||
trap, the `reasoning` field change, troubleshooting
|
||||
- [[frank-large-gufo-qwen38-27b]] — RETIRED 2026-09-28 → 2026-10-01
|
||||
(replaced by [[frank-large-vllm-mimo-v26-bf16]]; rollback target via git
|
||||
tag frank-large-gufo-q4kxl-20261001, engine + payload + drafter still on
|
||||
the host): gufo (gufo-org/gufo) v0.1.1, the Strix Halo inference engine,
|
||||
built from source on Dankclaw, serving Qwen3.8-27B UD-Q8_K_XL/UD-Q4_K_XL
|
||||
with the z-lab DFlash2 Q4_K_M drafter on the gfx1151 iGPU — the two
|
||||
from-source build workarounds, the client-contract delta, the measured
|
||||
prefill/decode trade against llama.cpp
|
||||
- [[frank-large-unsloth-qwen38-27b-q8xl]] — RETIRED 2026-09-28 (replaced by
|
||||
[[frank-large-gufo-qwen38-27b]]; rollback target): stock llama.cpp
|
||||
(unsloth b11030, ROCm gfx1151) serving
|
||||
@@ -203,12 +212,18 @@ frank-small, port 7771]
|
||||
.hgn + quality overlay) on the gfx1151 iGPU — two 200k conversations over one
|
||||
400,384-position pool, deployed flags, memory model, client contract,
|
||||
troubleshooting (2026-09-16 → 09-19)
|
||||
- [[frank-small-gemma4-12b]] — ACTIVE frank-small runbook since 2026-09-17
|
||||
(root systemd unit, port 7771): mainline llama.cpp serving
|
||||
- [[frank-small-sglang-mimo-v26]] — ACTIVE frank-small runbook since
|
||||
2026-10-01 (root systemd unit, port 7771): SGLang in a self-contained
|
||||
python3.12 venv serving leteschlead MiMo-V2.6-Distill-Qwen-9B-INT8-W8A16-
|
||||
AutoRound on the RTX 3090 — the same architecture frank-large now serves
|
||||
in BF16 on the iGPU; 2 x 262144 ctx on a pinned 606,208-token fp8 pool,
|
||||
NO spec decode (zero mtp.* tensors shipped), in-checkpoint vision.
|
||||
- [[frank-small-gemma4-12b]] — RETIRED 2026-09-17 → 2026-10-01 (replaced by
|
||||
[[frank-small-sglang-mimo-v26]]; rollback target, payload and binary
|
||||
intact on disk): mainline llama.cpp serving
|
||||
HauhauCS/Gemma4-12B-QAT-Uncensored-HauhauCS-Balanced Q4_K_M on the RTX
|
||||
3090, with the separate MTP assistant sidecar chained behind ngram-mod,
|
||||
q8_0 KV, and `--fit` sizing the full 262k window into ~12 GiB. A same-day
|
||||
vLLM detour (2026-09-19) was reverted the same day; see below.
|
||||
q8_0 KV, and `--fit` sizing the full 262k window into ~12 GiB.
|
||||
- [[frank-small-vllm-gemma4]] — RETIRED 2026-09-19 (same-day revert,
|
||||
rollback target): vLLM serving google/gemma-4-12B-it-qat-w4a16-ct (w4a16
|
||||
Marlin) with the bf16 MTP assistant on the RTX 3090 — bf16 KV, 289,647-
|
||||
|
||||
@@ -875,3 +875,40 @@ any repo service.
|
||||
verified (KV cache size, concurrency, MNBT 1024, gather 32x64) plus live
|
||||
contract probes (reasoning split, tool_calls). Rollback target for the
|
||||
2026-09-29 fp8+MTP4 shape: git tag `frank-primary-fp8-mtp-20261001`.
|
||||
|
||||
## 2026-10-01 — frank-large cutover: gufo -> vLLM TheRock, MiMo-V2.6 BF16 + cross-model MTP
|
||||
|
||||
- **Question that started it:** can a Qwen3.5 MTP drafter attach to
|
||||
frank-small, and how does vLLM run the FULL BF16 MiMo checkpoint? Answer 1:
|
||||
not on frank-small — SGLang reads the MTP head from the TARGET checkpoint
|
||||
and MiMo ships ZERO `mtp.*` tensors (base and INT8 alike). Answer 2: the
|
||||
digest-pinned `kyuz0/vllm-therock-gfx1151` image (vLLM 0.30 / torch 2.11 /
|
||||
TheRock ROCm 10) boots BF16 MiMo on the gfx1151 iGPU in ~200-240 s, and
|
||||
vLLM reads the draft from a SEPARATE path — so cross-model MTP from
|
||||
`Qwen/Qwen3.5-9B` (the distill's source, identical 4096/248320/32-layer
|
||||
geometry) works. Frank's call: ship that as frank-large.
|
||||
- **Bench window (frank-large stopped, one session, temp 0, 28,647-token
|
||||
prefill / 512 decode):** base 17,876 t/s prefill / 13.1 t/s decode (13.1
|
||||
is ~94% of the LPDDR5x BF16 roof - decode here is bandwidth, not a bug);
|
||||
+DFlash-8 7,976 / 17.4 (acceptance 59%/26% per position); +MTP-2 7,380 /
|
||||
19.8 (71%/44%, token rate 57.4%). Deployed MTP-2: +51% decode. Both spec
|
||||
arms pay ~55-59% of prefill. `num_speculative_tokens` beyond 2 unmeasured.
|
||||
- **Measured numerics law:** spec arms do NOT reproduce the no-spec greedy
|
||||
text (bf16 batched verification reorders reductions; near-tie argmaxes
|
||||
cascade). vLLM spec decode is distribution-preserving, not
|
||||
bitwise-greedy-preserving.
|
||||
- **Client-contract delta wired:** vLLM emits `reasoning`, not
|
||||
`reasoning_content` - `models.yml` frank-large block now sets
|
||||
`reasoningContentField: "reasoning"`, and the extraBody moved to the
|
||||
Qwen3.5-family recipe (0.7/0.8) with `chat_template_kwargs`
|
||||
(a top-level `preserve_thinking` is ignored by vLLM).
|
||||
- Deployer gap found pre-deploy: the spec-draft gate demanded a single
|
||||
`model.safetensors` (the dflash layout) and would have rejected the
|
||||
deployed MTP arm (4 shards behind an index); fixed to accept both
|
||||
layouts, gated by bats. `z-lab/Qwen3.5-9B-DFlash`'s "vLLM PR #40898
|
||||
unmerged" card claim is FALSE - `vllm/v1/spec_decode/dflash.py` ships in
|
||||
0.30 (verified in-image); the drafter stays on disk as the alternative.
|
||||
- Host state: +38 GiB fetched (MiMo BF16 17.53 GiB, Qwen3.5-9B 18 GiB,
|
||||
DFlash 2.58 GiB), disk down to ~26 GiB free - reclaim pass due. gufo
|
||||
rollback pinned at git tag `frank-large-gufo-q4kxl-20261001`; engine,
|
||||
Q4_K_XL payload and DFlash2 drafter all remain on the host.
|
||||
|
||||
Reference in New Issue
Block a user