Prime ruled the cutover ("do it", 16:40, via albok-dev) after the judge's
note question was re-fitted (albok-nemi 0.1.18). switch-to-h2o.sh ran 17:02:23,
the GPU 1 gate passed, h2o healthy 17:04:54. Checks on :8033: /health at the
pin, /decide/shared 200, 422 over the cap, inline image answers, image URL
refused by the shim, vLLM media-domain backstop refuses (400). compose.yaml is
h2o again; rollback-to-inc.sh stays the one-step way back.
albok-dev withdrew the judge sign-off on fleet-scale evidence and brokkr
concurred: over 10,841 texts judged by both models, h2o flagged 242 adult at
0.5 against INC's 88, and no threshold reproduces INC's set. rollback-to-inc.sh
ran 14:03:28, INC healthy 14:03:56, /decide/shared smoke 200 x3.
compose.yaml is INC again; the h2o compose is kept as compose.h2o.yaml with
switch-to-h2o.sh (GPU 1 free-memory gate, falls back to INC). It now carries
Prime's inline-images-only setting: LIGHTNING_IMAGE_URLS=off on the front
(verified live: an https image URL is refused before any fetch, inline images
still answer) and vLLM --allowed-media-domains media.invalid as a backstop.
rollback-to-inc.sh no longer leaves timestamped compose copies behind.
The :8033 seat now serves h2oai/h2o-lightning-4b @ b43f759b on vLLM 0.30.0 with
byte-pinned KV (4.48 GiB, 135,558 tokens) behind services/h2o-decision-front:
32k refuse-not-trim admission, the seat's bearer token, and /decide +
/decide/shared translated onto /v1/systemone for albok-nemi's judge. URL, port,
token and stack name are unchanged. INC down 13:08:18, h2o healthy 13:11:27.
Post-checks on the live seat: /health at the pinned revision; JevBench public
205/231 (= bench); 422 over the cap; 16q at the cap warm 2.0-2.16 s; seat peak
16,164 MiB under 4 concurrent 16q x 32k calls against a 16,847 budget.
compose.inc.yaml + rollback-to-inc.sh restore Intern-Decision-4B in one step.
The temporary GPU 3 eval seat (stacks/h2o-eval) is down.
The ~6.5-9 s first-call-per-bucket autotune lived in the container's writable
layer and died on every recreate. 0.1.3 creates /tmp/triton-cache in the image
owned by 10001 so the named volume intern-decision_triton-cache inherits a
writable mount point, and compose mounts it.
Bucket model PROVEN, not inferred: 2,048-token buckets, 16 up to 32,768. After
one warmed call per bucket, 12 random sizes across 8k-32k were all warm (worst
2.09 s); cold entries cost 6.5-9 s. Full cold warm-up 109 s; warm re-run 17 s.
scripts/intern-decision-warmup: one noul call per bucket, MAX_TOKENS from
/health, two-point live calibration of the tokenizer's linear token model (a
single probe overcorrects and the aim oscillates around the bucket edge),
per-bucket wall times, non-zero exit on a missed bucket. Run it after an IMAGE
CHANGE only; the volume carries ordinary recreates (measured: force-recreate,
then a warmed 32k call answered in 2.11 s).
Acceptance on 0.1.3: JevBench 202/231, hard 83/111, 0 diffs / 924; warm 32k GPU
1 peak 15,218 MiB (budget 15,220; a COLD autotune touched 15,224 once, README
caveat); /decide answers. Artifacts in the acceptance dir.
Prime 2026-09-30: move scriberr to GPU 3 and extend the Jev endpoint to 32k tokens.
Scriberr holds 0 VRAM idle; verified a 20-min job on GPU 3 at 5,496 MiB. With GPU 1
freed, intern-decision's measured card peak at MAX_TOKENS=32768 is 15,220 MiB against
a 15,437 MiB budget (n=3, 1 and 16 questions); 32,769 tokens is refused 422 up front.
JevBench v1.2.16 via /v1/systemone unchanged: 202/231, 0 diffs vs the bench.
Both are required in compose because they are coupled: MAX_TOKENS is checked before the forward
pass, so an oversized call is a clear 422 instead of reaching the cap as a 503. Pre-deploy floor
is nvidia-smi Free >= 15,400 MiB on GPU 1 (card peak 9,876 + scriberr 5,496).
stacks/intern-decision: compose (GPU 1, :8033, hard VRAM cap as the single .env knob,
healthcheck, Homepage group 'AI - Eval & Retrieval'), .env.example and README.
dns: intern-decision.fv.internal -> fv-ml1 (synced to ana/esh/nh3).
acceptance on fv-ml1 GPU 3, 3 fresh processes: bit-identical to the Jev bench's native rows
(pooled 240/259, Wyrd 79/84, 0/560 flips, Δp 0), negative control 10/122/14, 0 flips across
restarts; largest accepted request 200 at a 10,134 MiB card peak under a 9.25 GiB cap; 503 and
recovery proven at a tight cap. GPU 1 deploy held: nvidia-smi Free on GPU 1 is 15,442 MiB.
vh