vh 750675e391 feat(intern-decision): cap 9.0 GiB with MAX_TOKENS 7168, the largest call measured to fit ...

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).

2026-09-30 09:40:08 -07:00

..

althing-alert-bridge

feat(alerts): generalize the althing bridge, wire Kuma to it, retire chamber

2026-09-21 23:11:54 -07:00

althing-notify-failure

docs(notify-failure): a stop that exits non-zero pages as a failure

2026-09-23 02:15:13 -07:00

booth

chore: extract the Booth to its own repo (vh/booth)

2026-09-21 21:54:47 -07:00

coldfusion-abliteration

feat(coldfusion-abliteration): Heretic-300 — 8/100 refusals at KL 0.0136, beats the heresy bar 3.6x

2026-08-20 22:51:56 -07:00

draupnir-slicer

fix(draupnir-slicer): warn that every invocation writes result.json to CWD

2026-09-21 11:33:27 -07:00

erp-seat-quant

Swap the MeroMero A4B onto the erp-seat seat as char-rp-fast, retire the Pfish-6 alias

2026-09-10 11:34:05 -07:00

esh-vm-docker-watchdog

feat(esh): harden esh-vm-docker against the NFS D-state wedge (Tier 1 + watchdog)

2026-08-16 10:23:24 -07:00

flash-next-mtp-bench

feat(fv-ml1): generate the seat inventory from the live box instead of maintaining it by hand

2026-09-13 23:01:44 -07:00

gen-seat-mixed-quant

fix(quant): stop baking the calibration truncation cap into the shipped tokenizer

2026-08-22 00:32:21 -07:00

grok-token-broker

grok-token-broker: shelved — operator kept the jail, so the renewal feature has no consumer

2026-09-16 16:04:29 -07:00

headscale-ddns

fix(headscale-ddns): retry Cloudflare, validate every response, never write blind

2026-09-23 16:27:58 -07:00

heretic2-nvfp4-quant

feat(heretic2-nvfp4): WORKING modelopt NVFP4+MTP seat + full recipe runbook

2026-07-14 14:41:48 -07:00

intern-decision-serve

feat(intern-decision): cap 9.0 GiB with MAX_TOKENS 7168, the largest call measured to fit

2026-09-30 09:40:08 -07:00

irv-ml1

fix(irv-ml1): tailscaled could never add its IPv6 mesh address, and the README described a topology two cutovers old

2026-09-22 08:21:10 -07:00

lora-training-worker

feat(lora-worker): add optional train_id to POST /train (explicit publish-path namespace)

2026-07-07 01:49:24 -07:00

meromero-quant

Land the MeroMero v2-31B NVFP4A16 quant and record the pinned-transformers trap

2026-09-10 10:56:16 -07:00

mog-sec-tuning

Set mog-sec's context to a measured ceiling so it refuses instead of dying

2026-09-10 16:04:16 -07:00

refusal-probe

feat(gen-seat): promote absolute-heresy to the live gen seat

2026-08-17 17:23:35 -07:00

searxng-mcp

feat(searxng): move to nh3-docker, update, and expose as an MCP tool

2026-09-03 14:07:06 -07:00

secrets-broker

fix(secrets-broker): bw is not concurrency-safe — serialise, and never return an empty secret with exit 0

2026-09-15 08:49:11 -07:00

semif-serve

docs: Jev candidate bench vs SemIf (fv-ml1 GPU 3) — Intern-Decision-4B is the replacement if SemIf is displaced

2026-09-30 05:00:45 -07:00

soong-lab-ci

fix(soong-lab-ci): webhook auto-deploy real root cause = gitea ALLOWED_HOST_LIST + add listener logging

2026-07-14 09:19:38 -07:00

zellij-fleet

feat(nh3-dev): zellij-fleet@.service — fleet zellij server in its own cgroup (for SVOS seat_up); installed, not enabled

2026-09-25 23:14:48 -07:00