Layout: GPU 0 = cyberprev (KV 16 -> 15 GiB, 1.29x), erp-seat (moved from
GPU 1, KV 8.49 -> 5.5 GiB, 1.32x), voices (ctx 8,192 -> 40,960, its native
maximum, KV 6 GiB, 1.07x), coder (moved from GPU 1, ctx 16,384 -> 32,768,
the card maximum, 2.35x), parakeet-nemo. GPU 1 = gen-small (moved from
GPU 0, 5.5 GiB, 1.76x), meromero-rp (22 GiB, 1.32x), intern-decision.
Sizing rule: KV >= 1.25x max context rounded up to 0.5 GiB, checked against
7 days of LiteLLM traffic and the engines' own KV-usage logs. Each card keeps
spare memory for its bursty tenant (parakeet's cap on GPU 0, intern-decision's
32k call on GPU 1). YaRN extensions are not enabled: they do not fit, and the
Qwen cards warn static YaRN degrades short prompts.
gen-small gets --override-generation-config max_new_tokens 32768. It applies
only to requests that omit max_tokens (vLLM get_max_tokens) and answers 9
runaway ~260k-token replies in 7 days; the longest normal reply was 6,357.
Applied with playbooks/fv-ml1-rebalance-20261005.yaml (boot-check guard and
health/KV verification per seat); all nine seats healthy, gateway smoke
tests pass.
meromero-rp sized its KV from --gpu-memory-utilization, and vLLM's profile
counts co-booting neighbours as its own non-torch memory. Three boots gave
20.37 / 26.74 / 29.04 GiB of KV, and the 2026-10-05 Docker-move restart cut
GPU 1 free memory to 3,981 MiB, below intern-decision's 32k-token budget.
Coder read negative KV on every co-boot. Both are now pinned, so util only
gates the boot check (meromero util 0.52 -> 0.47, just above its footprint).
GPU 0 plateaus at ~60 MiB free 1-2 days after a boot (Beszel), so
gen-small's KV drops to 5.5 GiB (461,215 tokens, 1.76x) and parakeet-nemo
stays on GPU 0 with room for its 3,840 MiB cap. GPU 3 stays clear.
Applied with playbooks/fv-ml1-kv-pin-20261005.yaml: GPU 1 free 3,981 ->
11,801 MiB, GPU 0 495 -> 5,039 MiB right after boot; all nine seats healthy.
The db/ and runs/ bind mounts are root:docker 2775 since the 2026-09-14 tree
normalisation; the app runs as 1000:1000, so SQLite could not create its journal
and every write returned 500 (found at v0.40.2).
HunyuanVideo-Foley, MMAudio and ControlFoley load their Hugging Face dependencies
from pinned snapshots staged on the model store. Without this they look in the
container's ephemeral cache and download again on every recreate. Requested by
comfy-dev; HF_HUB_OFFLINE deliberately not set.
Since skaldsong v0.39.0 the structured pipeline passes call the LiteLLM
gateway directly; with no key every story generation fails 401 at its first
pass. Operator ruling 2026-10-05: the shared all-agents key until Skaldsong
v1.0 final, then a project-scoped key. The stack passes an explicit
environment list, so the .env line alone does not reach the container.
Prime's ruling 2026-10-05 (relayed by albok-dev, confirmed directly) replaces
the 2026-09-26 simple-shuffle load-share for embeddings only; the reranker
groups are unchanged. Routing uses LiteLLM per-deployment `order` (semantics
read at source in v1.97.0): lowest order among healthy deployments, and a
failed request retries at the next order.
Verified after the restart (43 s gateway outage): 6/6 calls per alias landed
on its primary. With nh3-ml1 tei-embed stopped, qwen3-embedding-bulk served 6/6
from esh-ml1 with 200s; once nh3-ml1 was back, bulk returned 6/6 to nh3-ml1.
The albok-service key now covers qwen3-embedding-bulk (200); the Nemi key does
not (403).
albok-service leaves nh3-docker before unit 4 puts the personal vault and
licensed books into restricted wings. On nh3-docker, portainer, portainer_agent,
dockge (socket rw) and dockerd on plaintext tcp/2375 were each root over
/srv/albok. The new host runs only albok-service: no socket containers and no
TCP Docker API, so its Homepage tile is a static services.yaml entry.
- nh3-pve-2: vmbr0 made VLAN-aware (applied by reboot, no guests running);
pbs-ana storage in the new namespace nh3-pve-2, plus a 21:30 vzdump job
(pbs-nh3's recursive sync mirrors it)
- VM 110 nh3-albok, 10.100.50.44 on VLAN 50, Debian 13 cloud-init, 4 vCPU
x86-64-v3 / 12 GiB / 100 GB; built to fleet conventions (vh 1000,
infra-ops 850, docker 851)
- compose: mem_limit 4g -> 8g, homepage labels removed; host port 8392 kept
- cutover by tar stream, verified by digest (contents plus owner, group, mode,
mtime and type identical); albok.nh3.internal now points at nh3-albok
- old /srv/albok stays on nh3-docker (container removed, compose renamed)
until acceptance and the operator's approval to delete it
- sudo-io archive hosts and ops-log STACK_HOSTS include nh3-albok
albok-service 0.1.0 (vh/albok 0b37431), image pfi/albok-service pinned by
digest, published on host port 8392 because 8390 is the post office.
Host prep playbook creates the fixed ids (albok 1500, albok-read 1510,
albok-personal 1511) and the local store/private roots; the container
gets a mounted /etc/group and group_add so the service can resolve and
chgrp its wing dirs. The config carries a LiteLLM key scoped to
qwen3-embedding and lives outside the deploy-synced conf dir. DNS name
albok.nh3.internal.
The in-browser recorder calls getUserMedia, which browsers refuse on http://
origins, so it sat at "Initializing recorder...". scriberr.nh3.phasefinal.com
is now fronted by the fleet TLS caddy on nh3-dev (wildcard cert) and added to
Scriberr's ALLOWED_ORIGINS; the Homepage link points at it. The plain
http://10.251.50.54:8080 URL keeps working except for recording.
Image local/mia:0.1.0 built from stacks/mia: MIA v2 @ bbd8b158 (MIT) with
its pinned submodules, dread-dev's proven Python lock with the torch family
swapped to cu129, and a driver adapted from dread-dev's run_mia.py that
seeds every mesh (fix_random + trimesh's module RNG) and writes
weights_effective into the npz. Weights stay in fv-ml1's shared HF cache
at pinned revisions, mounted read-only.
scripts/mia-run mirrors blender-run: --job DIR is shipped to
fv-ml1:/tank/mia/jobs, one docker run --rm rigs every mesh, out/ comes back.
Acceptance on the four Dread Naught characters: 3.9-4.7 s a mesh (median
of 3) plus 12.7 s model load, peak 3,394 MiB; seeded runs bit-identical
across rotated mesh order; GPU-vs-CPU distances the same size as sampling
noise, with an unseeded GPU run as the positive control.
The gen-small EngineCore OOM (04:21 PT): our parked 3,582 MiB window cache left
no room for vLLM's runtime workspace. Seat-side fix, three controls:
- windowed path wraps every window in torch.cuda.empty_cache(), so the seat
returns to ~2,108 MiB rest after a 12-min file instead of parking at the
peak (measured: peak 3,028 MiB during, rest after, restarts=0);
- MEM_CAP_MIB=3840 hard set_per_process_memory_fraction: over-cap requests
answer 503 with the seat alive (proved at cap=2000), so the failure lands
on us, never on a neighbour;
- CUDA_GRAPHS=0: the graph decoder pins cache blocks that empty_cache must
free (illegal-memory-access wedge when both were on first try). Cost:
12-min file 3.0 s vs 1.2 s, short bins 35-62 ms vs 33-42 ms -- still 4-15x
under the sherpa seat.
Measured, not computed: gen-small moved ZERO from 36,116 MiB across three
realistic requests (1,351 in / ~180 out) -- its workspace lands at engine
init; the growth window is restart-relative, matching infra-ops's observation.
WINDOW_S is now a real compose tunable. README memory section rewritten.
Prime-approved switch of the fleet STT seat (fv-ml1 :8300, LiteLLM ext-stt/
whisper-1, caller talk) from the sherpa-onnx int8 seat to arm B-bf16w of the
2026-09-30 A/B (docs/pfi/parakeet-seat-ab-2026-09-30.md): p50 33/36/42/71 ms
vs the old seat's 187/308/626 measured on the same card today, WER 1.965/3.026
vs the A/B floor 1.97/3.09. All three seat defects fixed: 12-min file 200s
(windowed at 360 s after a GPU 0 OOM on one whole request; the A/B's own
long-form method), no pause truncation, no long-form dropout.
GPU 0 room: gen-small --gpu-memory-utilization 0.48 -> 0.36 (0.46 and 0.40
refuse their boot check; cyberprev+voices hold the card). Its KV is byte-
pinned, so the boot log is token-identical: 670,142 tokens / 2.56x before
and after. Seat rests 2,088 MiB; GPU 0 keeps ~1.9 GB free.
Two runtime landmines documented in the README: NeMo's attention mask is
materialised T x T even under local attention (hence the window), and
httptools 0.8.0 writes a NUL into the HTTP status line that httpx — i.e.
LiteLLM — rejects, so the image ships plain uvicorn with --http h11.
Old seat stopped, not removed: docker stop parakeet-nemo && docker start
parakeet is the rollback.
License: NVIDIA Open Model License (accepted by Prime 2026-09-30); note in
stacks/parakeet-nemo/README.md.
Prime: Scriberr gets the basic fix, v3 stays (no NeMo 3.0.0 surgery). 0002 moves
from proposed/ into the carried set; scriberr-rebuild now applies 0001+0002 by
default (suffix dropout2) and its memory budget becomes a 5,600 MiB regression
guard (Scriberr is on GPU 3). Live on fv-ml1 1602: scripts rewritten from the
patched embed, a 20-min file at 5,502 MiB with retried_gaps reported.
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's ask (via the coordinator): investigate the "Parakeet skips
stretches of speech" finding, including other Parakeet weights.
Investigation only; nothing deployed.
Against ground truth (official SCOTUS transcript, Gutenberg #38916) the
drops are real: production v3 loses 140 / 66 clean words per transcript on
the two public files and ~50 on each private one (Whisper-referenced,
Canary-confirmed; adjudicator 129/129 correct on the calibration). Cause:
the v2/v3 0.6B weights collapse deep inside long full-attention windows;
the encoder output is degraded, the audio alone transcribes fine, and
1.1B TDT/RNNT/CTC and CTC-0.6B never do it. Decoding (CUDA graphs, greedy
variants, max_symbols, beam), slice length, local attention, loudness,
resampling and a noise floor do not fix it. Controls: A-vs-A, silence
positive control (>=15 words 36/36), null control, bootstrap floor.
Proposed patch 0002 re-transcribes >=3 s stretches where the audio holds
speech but no word came out (-80 to -90 % lost words on all four
recordings, lower WER, no invented text, +10 MiB) and adds an explicit
PARAKEET_MODEL_PATH with the loaded model recorded in JSON and ModelUsed.
Reviewed at high effort, all findings fixed; built and tested as
scriberr:local-blackwell-a353078-dropout2, not deployed.
scriberr-rebuild: --patches takes DIR[:DIR...]; embeds and seam-checks
both Parakeet scripts (seam-check --standard for the short-audio one).
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.
Scriberr moved to fv-ml1 GPU 3 on 2026-09-30, so the card is no longer idle when the rebuild runs. The memory stage now requires >= 20 GB free instead of an idle card (GPUs 0-2 still fail that) and attributes the peak only to the host PIDs of its own container, captured with docker top alongside the 0.2 s nvidia-smi samples. Verified with five processes on the card: peak 5,496 MiB, identical to the exclusive-card figure.
Deployed 2026-09-30 1211 PT by pointing SCRIBERR_IMAGE at the patched tag (.env backed up as .env.bak-20260930-pre-slicer1; rollback is the unpatched scriberr:local-blackwell). PrepareEnvironment rewrote the env's parakeet_transcribe_buffered.py from the embed (sha256 matches the patched source). One live run on GPU 1 beside intern-decision peaked at 5,496 MiB. Memory records the open Parakeet mid-chunk dropout finding and the held upstream PR.
Carry patches/0001 on our Scriberr build (upstream a353078): adjacent
buffered chunks overlap by 4 s inside --chunk-len and hand over at a word
both chunks transcribed alike, instead of cutting at fixed marks with no
overlap. Pause-aware cutting is included as an opt-in (--pause-search);
it measured neutral once the stitch was right. The Go<->Python CLI and
JSON seam is unchanged.
Bench (4 recordings, 118 min, 3 cut placements each, against a no-cut
whole-file reference; metrics only, private audio stays on fv-ml1):
cuts with an error within +-3 s fall from 52% (93/179) to 22% (41/184)
against a 19% background; floor +-0.08. Positive control: upstream's
cutter +0.33 over background. A-vs-A byte-identical in-process and
across CLI processes. Peak GPU memory unchanged at 5,496 MiB (n=3).
Also found: Parakeet skips runs of >=10 words mid-chunk with any
slicer, upstream's included; not addressed here.
scripts/scriberr-rebuild clones a pinned upstream sha into a new
/opt/docker/src dir, git-apply-checks the patches, builds a distinct
tag, and checks embed, unit tests, the JSON seam (scriberr-seam-check.py)
and the memory budget on idle GPU 3. Deploy stays manual. The upstream
PR is prepared under patches/upstream-pr/ and not opened.
intern-decision deployed 0941 PT (cap 9.0 GiB, MAX_TOKENS 7168). Live acceptance: positive
control 240/259 and Wyrd 79/84, bit-identical to the bench (0/560 rows, Δp 0); negative control
10/122/14; largest accepted requests 200 with no 503; per-process 8,812 MiB at rest and 9,866 peak;
GPU 1 Free 15,442 before and 6,581 after (lowest 5,569 under load). Latency from nh3-dev:
21 criteria 114 ms, 16 over ~3,900 tokens 238 ms. semif README banner now REPLACED with the
rollback; fv-ml1 GPU 1 note updated.
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.
Measured Parakeet peak on a 35-min file (n=3 each, deterministic): 300 s
9,384 MiB, 120 s 6,510, 60 s 5,976, 10 s 5,634 (fixed floor); with
expandable_segments 120 s 5,496 and 60 s 5,502. Verified 5,496 under the
recreated container's own env. Transcripts: 95.8% word-sequence similarity
vs 300 s, diffs mostly casing/punctuation.
vh