Two faces of the same verification round.
12-crawl returned to the homepage between links with a bare page.goto, and Plex
Web (signed out, so it renders its own sign-in iframe for its Plex-account flow)
keeps redirecting the top frame: Chromium cancelled our goto with
net::ERR_ABORTED, outside the try, so the whole crawl died — once as a 420 s
timeout in the deploy gate, once at 5.4 s on the same link. backHome() tolerates
that one error class and re-issues; any other error still propagates, and the
per-app verdict (guard settled, hostname, no authentik uidField, status < 500)
is unchanged. Plex was healthy throughout (container up 2 days, 32400 answering
HTTPS 401 / plain-HTTP empty reply), so this was app behaviour, not a fault —
recorded as gotcha 6 on the E2E page.
The 200,704 boot printed 5.5 GiB reserved and 74.1 GiB in all (not the ~5.4/~74
my arithmetic predicted), so every prose figure is re-keyed to the ledger, and
the runbook gains the second production-boot section: marker revision, pool
reservation, shape line, prompt-cache entry size, and the refusal contract at the
new ceiling (400 in 0.86 s naming 200704 and the 96,405-token overage, before any
prefill). Cost stated (two maximum-length conversations never coexist; the 180k
compaction target is unreachable inside a 155,648 window) and purchase stated
(1.7 GiB of page cache, 9.1 GiB against the 2026-10-06 shape).
Frank's call the same evening as the native-payload cutover: HALOGEN_CTX and
HALOGEN_KV_POOL_POSITIONS both 262144 -> 200704 (a multiple of 4096, the rung
frank-zealot already runs). The checkpoint's native window is 262,144 and the
engine serves anything below it, so this is a RAM decision, not a capability
limit: the KV reservation drops 7.2 GiB -> ~5.4 (the 2026-10-06 pool measured
14.6), and by the FLAGS.md law -- RAM the pool takes is RAM that n-gram table
loses -- the freed positions become page cache for the 47.68 GiB table that
long prompts read. Engine-side total ~74 GiB instead of 75.8.
The 400-zone law is invariant to the value, so the harness moves WITH the
ceiling: models.yml contextWindow 217088 -> 155648 = ceiling - 45056, keeping
the gap that absorbs the chars/4 under-count (worst-case request ~156.2k vs
the 200,704 ceiling, ~44k margin). Compaction now fires at 85% = 132,300 est
(~136k measured); the 180k target of the 262,144 era is unreachable because
the window itself is 155,648. maxTokens 45056 and the coupled
cap/timeout 65536/7200 are untouched.
Cost stated plainly: with pool == ctx, ONE maximum-length request fills the
pool, so two long conversations never coexist (two of <=100k do). Acceptable
because frank-dragoon carries only the slow role and the security-reviewer
subagent, not the interactive workhorse.
Seam re-keyed: the fit-law test now proves the invariant at the new pair, and
the pool law (ctx <= pool <= slots * ctx) still holds with pool == ctx.
Deploy verification for fcd7630 (22:20): healthy in ~12 s, e2e 33/33, and the
boot ledger settles the shape — 62.9 GiB weights locked (0.8 vision) + 7.2 GiB
KV in ONE 262,144-position pool (6,656 MiB, two slots) + 5.7 GiB working =
75.8 GiB in all, n-gram 47.68 GiB read from disk, prompt cache ON (12 entries),
MTP depth-2, no clamp line. Live probes: 27*59 -> 1593, tool call, red PNG ->
Red. The two KV figures are reconciled (engine preflight FORMULA 7.4 GiB
includes block scratch + per-slot O(1) state; the RESERVATION is 7.2).
Also corrects two stale claims: the retired swift15 rows named 200,192 ctx
(raised to 262,144 on 2026-10-07 inside that same era), and the OpenWebUI
frank-dragoon row's description is now re-keyed in the live DB (backup
webui.db.bak-20261017b).
New LLM/frank-dragoon-flash-server-v2-native.md carries the shape with the
three deliberate deviations from the card (served id, SPEC_ADAPT=0, engine
pin 0.16.2), the memory ledger with the PIN_TRUNK rejection, the full
measured A/B table, and the companion law (hardlinked stock peonist files,
engine-tool certified). The swift15 page is re-banned RETIRED 2026-10-06 →
10-17 with its rollback terms; both indexes and AGENTS.md (topology table,
service section, runbook index) are re-keyed, and the root index also stops
describing the halogen-27B page as ACTIVE.
New runbook frank-archon-radiance-qwen38-iq4r.md (publisher container,
payload chain, shape-verbatim cutover record with the boot-ledger and
measurement slots left open for the first boot, rollback). LLM index,
root vault index, and AGENTS.md (fleet table row, section, runbook
index: iq4r ACTIVE, w4dn8 RETIRED/on-disk, fp8-native deleted). Full
suite green.
Measured A/B on the shipped pair 25600/5120: A-cold 43.5k prefill 3,290 ->
3,584 t/s (+8.9 %) and the 3-way 197k aggregate 3,051 -> 3,589 t/s (+17.6 %,
wall 193.7 -> 164.5 s). The prediction I shipped with the arm -- slab is a
threshold below ~26 GiB, so 1,600 MiB buys ~0 -- is falsified by the measurement
it asked for. Two slab points inside noise of EACH OTHER (22,208 / 24,000) are
not a plateau. The cliff is above 25,600: 26,071 measured 4,379 t/s.
Ceiling re-derived from the engine's own budget table (card 32,624 MiB, pair at
29,120 settling 2,058 MiB free against its 209.19 MiB floor => 30,969 claimable).
The 29,120 bound was my 1 GiB \`keep free\` reserve; the seam now derives the
number, so the next resize starts from the engine's arithmetic.
Ledger for the record: slab 16,235 -> 17,548 slots while the pinned host pool
hands the same experts back (8,893 -> 7,580) -- the transfer is the mechanism,
the pinned pool is the slow path. Capacity contract live-verified at 3 slots:
3 x 199,984-token prompts accepted (599,952 tokens), fourth refused as too
large to ever fit.
Probe law learned: the oversized-prompt refusal names the prompt's token count
BEFORE prefill (free calibration) and fires on prompt tokens ALONE, not on
prompt + max_tokens; the window is ~514,286 chars at 2.5714 chars/token.
Runbook gains the 2026-10-13 section (symptom signature, the runtime-knob law,
the shape arithmetic, the boot ledger that confirms it: 843,276 tokens =
4 x 200,000, 12 all-reduce rows, 5.87 GiB of KV lent to the slab). AGENTS.md
gains the fleet-row update, the KV dtype law at four slots, the sysfs-raise
lesson under Key Lessons, and the pinned-RAM paragraph now points at
scripts/ttm-preflight.sh as the repair. LLM index entry + topology node match.
Runbook section `2026-10-11 — unquantized KV at 2 x 200,000`: the dtype law
(only bf16|fp8, no K/V split; the QSA block key is dtype-fixed, which is why
bf16 costs 1.89x and not 2x), the pool formula that reproduces both live
boots within 0.1%, the seqs-3-to-2 provenance (the tcclaviger fork's
torch.compile constraint, not radiance's), the boot ledger, and the A/B.
The A/B, fp8 control arm through a transient FRANK_ARCHON_ENV_FILE drop-in:
prefill -29% (4,379 -> 3,126 cold), TTFT at the full window +25% (the 2 x 197k
pair: 110.5 s -> 138.1 s), decode inside the +-3% band on two of three passes,
draft acceptance UP 0.661 -> 0.740. The price is prefill, not decode. Open
call for Frank: keep exact KV, or take the cheap fp8 2 x 200k fit (pool 3,648).
Also corrects two stale index entries (w4dn8 said STAGED, fp8-native said
ACTIVE), adds the worklog entry, and commits the verification tool as
scripts/probe-frank-archon-kv.py — it carries the probe law that cost two
false negatives: radiance answers in reasoning_content, so a chat probe with a
small budget reads as empty content + finish_reason stop on a healthy server.
Tests 368/368.
The 2026-10-09 wire aliases become the service names everywhere:
systemd units, serve scripts, env files and FRANK_* variables,
deploy-plan/deploy-remote service keys, container names, Prometheus
job names + alert rules, Grafana dashboard queries, bats seams,
runbook filenames, homepage/AGENTS lore, and historical prose.
Rollback tags renamed in lockstep. Residue in the tree: zero.
Retired frank-medium/small/npu names are history and stay.
Known casualty: scripts/patches/*.patch context lines were rewritten;
those retired-era vLLM-fork patches no longer apply byte-exact.
Capacity trade: the 9B dense-hybrid (qwen35: 32 layers = 8 full-attention
+ 24 Gated DeltaNet, 4 KV heads x 256 dim, native 262,144, MTP-less,
natively multimodal) costs ~7 GiB of weights where the 26B MoE cost
~16.6 GiB. The freed room buys back the second conversation slot and
exact KV: --ctx-size 401408 --parallel 2 (two 200,704 windows), f16 K +
f16 V, MTP off (the file carries no nextn/mtp tensors - flag-only enable
is impossible), vision via the publisher's BF16 mmproj, ubatch 1024
(the CUDA decode-optimal value), verbosity 4, publisher sampler recipe
(0.6/0.95/20/0) in models.yml.
Hosts with the MiMo-V2.6-Distill Q6_K boots at this geometry: 21,276 MiB
used / 2,849 free at ubatch 1024, 40,960 tokens of fp16 KV at ubatch
2048. Compaction stays routed here; window math is unchanged (maxTokens
16,384 keeps the worst-case harness request inside the slot).
Gemma4 payload (Q4_K_M + mmproj + MTP assistant) stays on disk as the
rollback; its runbook is retired in place, keeping the attention_k_eq_v
pair-billing law.
Co-Authored-By: omp <omp@franksdank.cloud>
Payload: HauhauCS/Gemma4-26B-A4B-QAT-Uncensored Q4_K_M (16,796,015,520 B)
+ the publisher's BF16 mmproj, /home/dank/models/Gemma4-26B-A4B/.
Shape: ONE 200,704-token slot (--parallel 1; 2x200,704 does not fit the
24 GiB card), f16 K + q8_0 V under flash attention, batch 4096 / ubatch
1024, --log-verbosity 4, MTP OFF (the gemma4-assistant file is on disk,
not loaded: its draft KV is built against the target's full window,
measured +1.9 GiB -> 23,806/319 free, no margin beside the f16-K raise).
Measured boots (nvidia-smi): q8_0 K + q8_0 V / ubatch 2048 / MTP off =
21,890 used / 2,235 free; with the MTP assistant = 23,806 / 319.
Engine facts recorded: gemma4 hybrid attention (25 sliding-window + 5
full-attention layers) so KV scales with 5 layers only, and
attention_k_eq_v is honoured by build 5fc4f3c (pool bills K-only; the
measured q8_0 pool = 2.05 GiB = the K-only figure).
Serve script gained VERBOSITY (+ --log-verbosity in the argv), the
SPEC_DRAFT_FILE drafter path, and defaults/lore retargeted to Gemma4.
deploy-remote.sh payload gates follow the env. The frank-secondary bats
seam is rewritten for the new contract (argv-based) and fixes a
pre-existing vacuity: serve_argv regenerates the test env, so
mutations made before it were silently undone - tests that mutate now
exec through run_stage. frank-secondary is the fleet compaction hub
(200,704 is the only window that fits a worst-case compaction input).
23,120 used / 1,005 free at boot, 763 settled after a vision probe
(encoder buffers resident). Rung cost 746 MiB, not the transferable
616 prediction. Prefill baseline 1,209 t/s @ 3,760 recorded; decode
unchanged as the law says. 2048 = ceiling for this shape.
Cross-cutting rename of the current llama.cpp endpoint on the RTX 3090
(port 7771) from frank-small to frank-secondary:
- Service trio: scripts/{frank-small.env,frank-small.service,
serve-frank-small.sh,download-frank-small.sh} → frank-secondary.*
- Test: tests/frank_small_unit.bats → tests/frank_secondary_unit.bats
- Runbooks: serverAppConfigs-docs/LLM/frank-small-*.md →
frank-secondary-*.md (9 files); the FreeToken runbook stays
freetoken-frank-small.md (it documents the retired CUDA-era service)
- Identifiers: FRANK_SMALL→FRANK_SECONDARY, llm_small→llm_secondary,
LLM_SMALL_*→LLM_SECONDARY_*, deploy_llm_small→deploy_llm_secondary
- Historical references to the OLD frank-small (FreeToken/CUDA on the
3090, and the Dankserver service that became frank-medium) are
preserved verbatim — they describe what the service was called
at the time.
Stop-window A/B, same 80k filler prompt, 128-token essays, runtime
timings: HALOGEN_FA_TQ=256 (upstream issue #12's carve workaround) moves
decode ±0.3% (noise, draft acceptance identical) and costs 9.5% prefill
on this topology. The depth collapse sits deeper than the FA tile knob;
the default stands. The harness's slow/advisor roles moved to
frank-primary (fp8 KV 7.32 KiB/token = 8.7x less KV read at depth).
Probe on the live endpoint (2026-10-06): 17.3k cold prefill 27.0 s ->
warm turn resumes at cache_n 16,384, prefill 1.7 s. The boot WARNING
'smaller than ONE full-context entry' sizes against the native 262,144
window, not our 200,700 slot — documented as expected in the checklist.
Tuning notes record the depth law (decode 5.5 t/s @ 46k -> 3.3-3.8 @ 80k
vs 30-50 shallow; KV reads ~8 GB/s effective; no supported flag touches
decode) and the AGENTS.md unit paragraph regains the unit-facts
parenthetical the reflow dropped.
Runbook reshaped: 2 x 131072 no-spec identity, slot-2 arithmetic (drafter ~0.4 GiB + ubatch ~0.3 GiB bought the second 4,352 MiB KV slot), ngram-mod skipped on the recurrent-rollback law, MTP references kept (44.8-52.1% acceptance, 58.9-66.9 t/s), rollback = one revert of 812d9eb. AGENTS fleet row + section, LLM/index.md intro + entry, root index entry all flipped to the 2026-10-06 shape. 360/360 suite green; e2e gate 33/33 green at deploy.
The parallel session's uncommitted swap, committed at Frank's direction
(commit whatever's in the working tree):
- llama.cpp master on the RTX 3090 now serves ukisai's
Swift-1.5-Qwen3.8-27B GSQ-RCO IQ3_S WITH the in-file MTP head, ONE
131,072-token slot at Frank's 128k cap, q8_0 K + q8_0 V, vision through
the base-model projector; MiMo-V2.6 Q6_K (2x200704 f16 KV) retired to
rollback target.
- env/serve/download + seam twins updated together (deployer payload gates
rode the previous commit's deploy-remote.sh).
- docs: new swift15 runbook; MiMo Q6K runbook retired; AGENTS.md topology
and both vault indexes follow (root index ACTIVE/RETIRED flip, test 135
seam green).
- comfyui: two qwen-image t2i workflow drops under dankclaw/comfyui/
workflows/ (12 KiB).
Both indexes stated the pre-doubling VRAM as current. Now 2,279 MiB free with
the 4096/2048 pair named, so no index contradicts the runbook it links to.
- new active runbook frank-small-llama-mimo-q6k-2x200704.md: 21,276 used /
2,849 true free (inside the 2-3 GiB floor), decode 94.44 t/s vs 104.84 on
Q5_K_M - the rung costs a measured -10 % (weight-bandwidth-bound card),
512px red-PNG vision verified, and the new probe-size law (>=256 px: a
10x10 image undersupplies image tokens and the model answers 'video'),
nvidia-smi free-field law (driver holds ~450 MiB the subtraction hides)
- frank-small-llama-mimo-q5km.md retired; it keeps the boot-pair table and
becomes the Q5 rollback reference (3,571 free, 104.84 t/s measured)
- AGENTS.md fleet row, endpoint section, --parallel lesson, runbook index
(active row + two retired rows); LLM + root index entries
- Q6_K GGUF-vs-GGUF diff proven: template and all 46 KV keys byte-identical
to Q5_K_M - the vision probe scare was probe size, not payload
- new active runbook frank-large-rocmfpx-q6agent-uncensored-mtp.md: the
Q6_0_ROCMFPX_AGENT file (ftype 114, 339 type-102 + 159 type-103 tensors),
27.6-28.1 t/s at 86.8% draft acceptance, the measured ngram-mod chain
rejection (-15% prose: recurrent-state rollback disabled), the
bench-window hardening laws
- frank-large-rocmfpx-q8agent-mtp.md retired; rollback section corrected
(gufo payloads removed from disk 2026-10-03 - that rollback is now a
re-download; the aligned Q8 file is the direct one, still on disk)
- frank-small-llama-mimo-q5km.md reshaped to 2 x 200704 with fully
unquantized f16 KV: boot-pair table gains the measured row (20,530 used /
4,046 free), 104.84 tok/s, slot-window contract at 200704, rollback to
the 3x180224 + q8_0-V shape documented
- AGENTS.md: fleet table rows, both endpoint sections, the chain law
extended with the recurrent-model measurement, the --parallel law with
the reslice, runbook index (active + retired)
- LLM/index.md + root index entries updated
Measured by boot probe on the shape-D base (production stopped, candidate
booted AS the service), plan numbers read off the engine's own log:
runtime overhead 7.60 -> 5.25 GiB (precise gfx1201 TP2 registry)
expert cache 14.87 -> 17.22 GiB
slots 6,496 -> 7,522 (+1,026, +16%)
swap region 13.7 GiB / 5,986 slots
boot healthy in 420 s, no OOM
'--expert-precise-memory': True also appears in non-default args, so a silent
argparse drop is visible. The pool is NOT cut (option A) because the pool is
not an arena lever: 3.15e9/rank keeps shape D's measured 1.99x concurrency,
and a 2.2e9 arm probed in the same window (277,686 tokens, 1.38x) is recorded
as the conservative fallback.
Boot-cost bug found while diffing Frank's 3xR9700 recipe: the image ships no
/root/.cache at all, our boot writes 1.1 GB of vLLM AOT/torch_compile
artifacts (4,665 files) plus 229 MB of comgr JIT kernels into the ephemeral
layer, and podman rm -f discards them — so every deploy/restart/reboot re-pays
'torch.compile took 181.20 s in total' (173.19 s of it one graph). Mounting
FRANK_PRIMARY_CACHE_DIR at /root/.cache persists both and hides nothing.
GPU_MAX_HW_QUEUES=2 added at Frank's direction (real HSA knob); not A/B
measured. Retracted my own recipe-diff claim that cudagraph_capture_sizes
needed pinning: capture is 3 PIECEWISE + 2 FULL graphs, 3 s, 0.31 GiB.
Tests: 367/367.
index.md intro and mermaid nodes still described 4 x 262144 fp8/bf16 and
frank-small at 2 x 262144; the active-deployment list had no frank-primary
entry at all. Fixed both, plus a malformed GLM list item.
log.md gains the shape-D entry (the SEQS=2 constraint, the arena-lever
measurement, the 2x2 probe table, the measured 1.99x), and the frank-small
entry's stale consequence bullet now points at the resolution.
frank-small runbook: the compaction-fit note is corrected - the 200704 window
moved frank-primary's trigger to 140,492, which fits a 180,224 slot again.
Runbook, both indexes, AGENTS.md (topology row, endpoint section, three new
Key Lessons, one stale SGLang label in the bench list) and the bundle log now
describe the deployed shape: pool 540,672, 180,224 tokens per conversation,
f16 K + q8_0 V, mmproj vision, 21,684 used / 2,441 free of 24,576 MiB, 106.8
tok/s decode, 3,814 t/s prefill.
The claims that mattered:
- three contexts buy 180,224 tokens per conversation, not the native 262,144,
and a longer prompt is a hard 400 - so models.yml contextWindow is 180224
and frank-primary's 183,501-token compaction request no longer fits here
(forward pointer added to the retired SGLang runbook, which asserted it)
- the vision tower bills ~1.4 GiB for a 918 MB file, established by subtracting
two measured boots, not by a formula
- /v1/chat/completions ignores the legacy top-level 'images' field: the model
invents a color; only the image_url content part reaches it
Staged 59 GiB of retired LLM payloads on Dankclaw to
~dank/.AGENTTRASH/20261002-retired-llm-payloads/ (vLLM MiMo BF16, Qwen3.5-9B
MTP source, both DFlash dirs, Gemma4-12B-QAT, SGLang MiMo INT8) and corrected
every doc claim about what is on disk: gufo's rollback pair (Q4_K_XL weights +
the Q8_K_XL DFlash2 drafter) and frank-small's Q6_K stay in place, so the
rollback paths in the tags' env files still resolve. The vLLM-era frank-large
fetch trio moves to old/ with its runtime; runbook links updated.
frank-large: new runbook frank-large-rocmfpx-q8agent-mtp.md — the ftype-115
tensor-type lock-in that forced the fork, the pinned build (set_rows history),
the measured MTP A/B (17.0 vs 7.6 t/s decode, acceptance 0.462), the live
client contract (reasoning_content, effort fold, vision restored) and rollback.
frank-small: new runbook frank-small-llama-mimo-q5km.md; q6k marked retired
(rollback target). gufo runbook marked retired (rollback target). Both indexes,
the fleet topology table, the vision/models.yml notes, the benchmarking law and
three new Key Lessons updated.
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.
Frank's call. Payload: ajgazin's abliterated Swift-1.5 UD-Q8_K_XL (29.3 GiB,
SWIFT OPEN LICENSE) -> unsloth/Qwen3.8-27B-GGUF UD-Q4_K_XL (16.35 GiB,
sha256 3f227079...b01e, vanilla/Apache-2.0). NOTE: the endpoint is no longer
uncensored — that was one of the reasons for the Swift payload on 2026-09-28.
New fetcher scripts/download-frank-large-q4.sh (no auth header: repo is public
and ungated; size + sha256 verified after transfer). The speculative drafter
and its fetcher are unchanged.
Shape: --sessions 4 --context 163840 with unquantized KV. gufo exposes no KV
quantization lever (execution_policy.hpp pins kv_cache_storage to kFp16), so
f16 is inherent.
KV arithmetic: f16 KV = 64 KiB/token (16 full-attention layers x 4 KV heads x
256 head_dim x 2 x 2B; the other 48 layers are linear-attention with a
per-sequence fp32 state). 4 x 163840 = 655,360 tokens = 40.0 GiB KV, each
session exactly 10 GiB. Budget ~62 GiB of the 114 GiB GTT.
Seam retargeted (FETCH -> the q4 script, TARGET_STUB, context/sessions pins).
Five 262144-token contexts passed every boot gate (KV 1,350,182 / 5.15x) but
left 0 bytes free at runtime: a real client request (995-token prompt,
max_tokens=261149, sampling params straight from models.yml) killed Worker_TP1
with torch.OutOfMemoryError 'Tried to allocate 50.00 MiB ... 0 bytes free'.
Boot is not proof of runtime survivability.
Four slots x 262144 fp8 (KV_BYTES 8,311,013,376/rank) returns ~1.93 GiB per
rank for activations/graph capture.
Unit: Restart=always instead of on-failure — a runtime worker fault makes
vLLM's API server exit clean, so ExecStart returned 0 and on-failure never
fired, leaving the endpoint down until a human noticed. Also
StartLimitIntervalSec=0 and OOMScoreAdjust=-500.
Frank's call after the night ladder: five contexts at the model's full native
262,144 window beat six shortened ones. Six full-length contexts do NOT fit at
any KV dtype (6 x 241664 fp8 asked 10.70 GiB against 9.18 GiB free — the
6-seq graph/activation footprint leaves too little for the pool); 5 x 262144
at fp8 PASSED on the first rung: KV_BYTES 10,388,766,720/rank, measured
'GPU KV cache size: 1,350,182 tokens, Maximum concurrency for 262,144 tokens
per request: 5.15x' (39,462 tokens of margin).
Also documented: the SEQS=3 bf16 max (241,664, KV 732,359 / 3.03x) measured
earlier the same night, and the root-cause read that the 7.2 GiB KV ceiling
law was the MTP4 workspace, not VRAM. Seam and harness (contextWindow 262144)
re-pinned.
Night ladder result (Frank's subtract-16k method): 262144 and 245760 both OOM
at CUDA graph capture ('Tried to allocate 76.00 MiB ... 0 bytes free'), 237568
passed (KV 718,606 tokens), and the +4096 bonus step PASSED -> 241664 is max
possible and DEPLOYED at util 0.98 (KV 732,359 tokens / 3.03x concurrency at
241,664-token requests). Util stays 0.98: lowering it shrinks the engine's own
budget instead of buying slack.
Client contract restored (the GLM rollback d91379d dropped all three and the
qwen rollback inherited the gap): --reasoning-parser qwen3 fixes the malformed
thinking block (thinking was landing inline in content with a dangling
</think> and a null reasoning field), --enable-auto-tool-choice with
--tool-call-parser qwen3_xml makes tools work, and
--default-chat-template-kwargs enables enable_thinking + preserve_thinking.
Harness: contextWindow 241664 (compaction gap recomputed), preserve_thinking
moved into chat_template_kwargs.
Payload: Intel/GLM-5.3-Flash-W4A16-AutoRound (AutoRound symmetric int4 group
128, auto_round:auto_gptq, 34 shards + 2 extras, 113,074 tensors, ~170 GiB).
It is the only W4A16 checkpoint this fork can serve with expert offload: its
quant_method "auto-round" is overridden to "inc", which routes the MoE to
libr4d's grouped W4A16 GEMM (R4dW4a16MoEMethod). A compressed-tensors W4A16
build of the same model resolves to TritonWNA16Experts, which has no offload
recipe, and refuses to serve at all (170 GiB of experts cannot fit 64 GiB of
VRAM, so offload is mandatory).
Shape: tcclaviger RDNA4 vLLM fork (same digest-pinned image as the 2026-09-29
qwen cutover), TP=2, 131,072 ctx x 2 seqs, bf16 KV (fp8 is ILLEGAL here — it
disables the sparse MLA backend's rope-free Triton path, the only MLA decode
gfx12 has), util 0.98, NBATCH 2048, --enable-expert-offload
--expert-precise-memory, prefix caching, glm47 tool + reasoning parsers,
MTP and expert prefetch absent by measurement.
Measured: decode 9.9-10.0 tok/s, prefill 918 tok/s (warm), arena 14.51 GiB =
2,383 slots/rank over 12,096 host-backed experts (73.64 GiB pinned/rank).
THE 35 tok/s TARGET IS NOT MET and the shortfall is measured: live telemetry
(R4D_LRU_TELEMETRY=1) gives a 0.53-0.66 per-layer hit rate, a step moves
0.82 GiB of misses in ~100 ms = 8.2 GB/s effective against a link measured at
22.74 GB/s — the same structural gather-path gap the qwen campaign found.
Growing the arena 13.08 -> 14.51 GiB changed decode by nothing.
Rejected by measurement: MTP4 (head loads, acceptance 3.76-4.90, but decode
unchanged and prefill collapses 918 -> 255 — a verification forward gathers the
union of the batch's experts), NBATCH 8192 (identical prefill).
Also: scripts/patches/gpu_worker-kv-cap.patch (fork-local fix for a KV-reserve
model that under-counts GLM's per-sequence KDA state; not needed by this shape),
runbook, AGENTS.md row + section, both indexes, work log.
Rollback: 2339fae (qwen FP8 + MTP4).
gufo is a vertical inference engine built for exactly this hardware
(Ryzen AI MAX+ 395 / Radeon 8060S, gfx1151), and Dankclaw's frank-large
has swapped runtimes seven times in search of a better one. This one
ships no prebuilt Linux binary, so the runtime is BUILT FROM SOURCE on
the host and the payload, port, alias and client contract stay exactly
where they are.
Runtime: gufo v0.1.1 (b0f8673) at /home/dank/GitPackages/gufo-v0.1.1,
built by the new scripts/build-gufo.sh. It links the host ROCm 7.2.4
through its RPATH — the opposite of the unsloth prebuilt it replaces,
which carried its own — so two host workarounds are encoded in that
script: Ubuntu's librocwmma-dev installs a truncated rocWMMA header set
that shadows AMD's and cannot be overridden with -I in HIP mode, and the
ROCm clang prefers Ubuntu 26.04's GCC 16 headers over the GCC 15 whose
libstdc++ is actually installed.
Shape is unchanged: one session, full 262144 ctx, the same UD-Q8_K_XL
target and the same DFlash2 drafter, greedy, text-only. Speculation
moves from llama.cpp's fixed depth 4 to gufo's adaptive block policy
capped at 7 proposals.
Client-contract deltas, all documented: the chat template is compiled
into the binary (no --chat-template-file, so the Qwen-Sharp 73k-token
failure mode is gone), reasoning_effort now takes the full OpenAI set
(so models.yml's reasoningEffortMap fold is no longer required — kept
anyway, so this commit does not also move every high role from xhigh to
high effort), and there is no KV-quant or prompt-cache-RAM flag.
The test seam now runs the serve script against a stub engine and
asserts on the argv it actually execs, instead of grepping the script's
text. The deploy handler gates on the runtime binary and returns before
it stops the serving unit.
Measured trade, from gufo's own Q8_K_XL tables against llama.cpp
b11069: prefill +37% to +80%, cold load 4.5x faster, decode +7% at zero
depth — but -14% at 65,536 tokens of depth and -24% at 131,072. On the
8-bit target the crossover is ~32k, and frank-large runs one 262,144
conversation. The Q4_K_XL target shows no such regression.
Answers Frank's question: given the 2x R9700 box, which of DeepSeek-V4.1-Flash,
MiMo-V2.6-Flash-MOPD and GLM-5.3-Flash runs best, on what runtime, and can
Affinity / vllm-radiance / FreeToken-AMD be modified to run them fast.
Sizing, measured from the HF API rather than the model cards:
- DS4.1 payload 475.27 GiB, of which 189.13 GiB is Engram (two 94.56 GiB
shards, the layer 1 and 14 lookup tables). Lucebox never maps them, so
they do not evict the expert page cache.
- 240 routed experts x ~11 MB per token at the 37-of-384-per-layer residency
the hot stack can hold = ~2.4 GB across PCIe per token. At 55 GB/s real on
Gen5 x8 that is a ~23 tok/s ceiling, under the 35 tok/s goal before any
attention or dense work.
- The backend refuses layer-major prefill for V4.1, so the only legal
prefill is token-by-token; 1000 t/s is ~20x away.
Affinity (Codeberg StillDeadcode) builds clean here (hipcc 23, gfx1201, zero
errors) and claims 110 tok/s for the 284B V4-Flash, because the 101 GiB
container leaves the cold expert set in a pinned host pool in RAM. V4.1 is
2.5x too big to repeat that. Port scope, from both checkpoints' real
safetensors indices: expert machinery ports unchanged (identical tensor
naming, 15,360 tensors at 40x384), 26/32 per-layer tensors and 34/35 config
keys present. Five gaps - Engram, the disjoint {0,1,2} vs {0,4,128} ratio
vocabulary, CED, the DSpark draft geometry, vision. num_hash_layers is the
only key V4.1 lacks, so model.h:36's default of 3 silently survives and
would route layers 0-2 through a table that does not exist.
Also corrects: Dankserver RAM 60 -> 246 GiB (stale since the 2026-09-20 board
swap) and storage; the top-level index credited V4.1 with V4's geometry
(43 layers / 256 experts vs the actual 40 / 384).
New ACTIVE runbook (setup/doctor PASS=34/FAIL=0, UVA transport verified
in serving workers, wire-verified alias/reasoning/metrics), index +
vault entries flipped: llama.cpp b11030 endpoint RETIRED 2026-09-27 to
rollback target. Topology diagram updated.
Discovery, not guesswork: dump the GGUF v3 headers of the fetched parts
(scripts/dump-gguf.py, added — a dependency-free reader that never
touches tensor data, so a 50 GiB part scans in milliseconds) before
the first load.
What the dump says (parts 2-3, 2026-09-26):
- The 51B-param PLE n-gram table is ONE tensor: per_layer_token_embd.
weight, dims [160, 320001536], Q8_0, 54.4 GiB — exactly the whole of
part 3. It is the same tensor the shared MTP head borrows, so the
=CPU override is required for placement AND for the draft load
(GPU-resident borrow: "invalid vector subscript").
- The model is qwen4exp, 512 experts, top-10 routing
(qwen4exp.ple.ngram_size=3, heads_per_ngram=8).
- Part 1 is metadata-only (ntensors=0); parts 2-6 carry tensors.
The placeholder pattern ngram_embeddings matched nothing — dropped from
OT_OVERRIDES (env, serve-script default, seam, AGENTS.md, runbook,
LLM index) before it could masquerade as a working override at load
time. Deployed set: exps=CPU,per_layer_token_embd.weight=CPU.
Suite: 346 green.
The 2026-09-26 same-day cutover that ended the Lucebox DS4.1 detour:
Dankserver's :7770 endpoint is now unsloth llama.cpp (the b11030-mix
prebuilt, ROCm gfx120X, bundled runtime) serving
unsloth/Qwen3.8-Flash-Next-GGUF at Q8_0 — 175.30 GiB across six sha256-
verified split parts — on both Radeon AI PRO R9700s, as a ROOT unit (the
frank-large shape: no container, no generated unit).
The deployed contract:
- Placement law: layer split -ts 1,1 with -ngl 999; the routed experts
and the 51B-param n-gram (PLE) table are COMPUTED in host RAM via
-ot exps=CPU,ngram_embeddings=CPU (the x4-pair measurement: streaming
over PCIe loses at 3.7 t/s, computing in RAM does not care).
per_layer_token_embd.weight=CPU is required by the shared MTP head's
borrow path (a GPU-resident borrow crashes the draft load).
mmap stays on; --cache-ram 16384.
- Spec chain draft-mtp,ngram-mod: the shared Q8_0 MTP head
(mtp-...-shared-Q8_0.gguf, 2,786,568,256 B) at depth 2, with the
draftless ngram-mod matcher chained behind it — the only legal way to
layer two spec implementations in one process.
- Vision explicit: --mmproj names the BF16 projector (this build never
autoloads; a server without it comes up healthy and text-only).
- Reasoning: --reasoning-format deepseek → reasoning_content; the
Qwen3.8 template's {xhigh, medium, low} reasoning_effort gate folded
by reasoningEffortMap in models.yml.
- One stream, the model's full 262,144 context; greedy server default
with Qwen-recommended sampling riding compat.extraBody.
The deployer's deploy_llm_primary is rewritten for the root unit: it
gates every artifact at its declared byte count (stat -c %s, never wc -c
— which read the whole 377 GiB lucebox file for minutes on 2026-09-26),
retires the lucebox user quadlet to dated home trash, waits out port
7770, relinks/enables/restarts the root unit and waits up to 1800 s for
/health. The quadlet-era tree under dankserver/frank-primary/ is
archived (dankserver/.AGENTTRASH/20260926-llamacpp-cutover/ on this
box; git history everywhere else) and its planner mapping replaced by
the scripts/ trio.
Observability flips back to the llamacpp namespace: a frank-primary
Prometheus job at 192.168.0.235:7770 (cross-host LAN — the netavark
published-port quirk that forces host.containers.internal for the two
Dankclaw endpoints is same-host only), the LlmEndpointDown runbook
text, and the dashboard description for three llama.cpp endpoints.
Docs: the new runbook (frank-primary-llamacpp-qwen38-flash-next-q8.md),
RETIRED banners on the lucebox and radiance runbooks and their index
entries, the AGENTS.md fleet section rewritten for the root-unit shape,
and the homepage frank-medium card renamed to frank-primary with the
Q8_0 Flash-Next spec (docs.html #models and the modality note updated:
images on frank-primary and frank-small, audio on frank-small only).
models.yml: the dankserver provider is "llamacpp (dankserver)" serving
frank-primary, contextWindow 262144, maxTokens 65536 (compaction-gap
law: 183,501 + 65,536 <= 262,144), input [text, image].
Suite: 346 green (frank_primary_unit.bats rewritten for the root unit —
16 tests pinning the placement law, the shared-head requirement, the
explicit projector, HIP env hygiene (BIND_HOST, never bare HOST), and
the deployer's byte-count gates; the quadlet-era planner and
changed-tests mappings rewritten for the trio; the observability scrape
test asserts three llamacpp jobs again).
Rollback: the lucebox image (localhost/lucebox-rocm:gfx1201) and the
378 GiB payload stay on the host; the Qwen payload stays fetched, so a
re-swap in either direction is a redeploy, not a refetch.
Frank
