ginnoirandClaude Opus 4.8 4aa2cd9468 fix(llm): q8_0 V-cache (q4_0 tanked generation to 1.3 tok/s on Pascal) ...

The q4_0 V-cache + flash-attention path is pathological on the GP100: 1.28
tok/s generation at 5-8% GPU util. q8_0 V-cache gives 9.2 tok/s and still fits
64k context in 16GB (15.3GB used, ~950MB free).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

2026-06-26 17:04:17 -05:00

..

authentik

chore(authentik): upgrade to 2026.5.2 (hop 2 of 2)

2026-06-11 23:07:38 -05:00

backup

Complete TB-006 batch 1 SSO so admin apps use Authentik without double login.

2026-06-10 23:52:14 -05:00

dev

fix(code-server): scope mount to projects dir and ai extension configs

2026-06-15 20:15:38 -05:00

famapp

fix(famapp): pull image from valhalla registry

2026-06-15 00:17:28 -05:00

files

fix: improve onlyoffice container runtime

2026-06-16 00:23:36 -05:00

foundry

refactor: pure env_file stack.env convention across all stacks

2026-06-09 16:03:47 -05:00

llm

fix(llm): q8_0 V-cache (q4_0 tanked generation to 1.3 tok/s on Pascal)

2026-06-26 17:04:17 -05:00

media

fix(media): raise nofile ulimit to 65536 for *arr containers

2026-06-24 01:28:08 -05:00

monitoring

Complete TB-006 batch 1 SSO so admin apps use Authentik without double login.

2026-06-10 23:52:14 -05:00

notes

Ship homelab quick wins: Recyclarr, infra pins, jd.ginnoir.com, and LAN lockdown.

2026-06-11 02:28:06 -05:00

notify

Ship homelab quick wins: Recyclarr, infra pins, jd.ginnoir.com, and LAN lockdown.

2026-06-11 02:28:06 -05:00

owncloud

Ship homelab quick wins: Recyclarr, infra pins, jd.ginnoir.com, and LAN lockdown.

2026-06-11 02:28:06 -05:00

photos

feat(photos): add Immich stack

2026-06-15 22:15:28 -05:00

proxy

feat(proxy): add share_gate for Discord-group external access

2026-06-24 01:02:06 -05:00

remote

fix(remote): correct hbbs relay flag to lowercase -r

2026-06-18 04:54:04 -05:00

resume

fix: make portainer stacks use stack env only

2026-06-14 22:56:38 -05:00

romhacks

fix(romhacks): only link library files that exist on disk

2026-06-08 05:02:34 -05:00

roms

chore(roms): unpin RomM from 4.9.0-beta.2 to floating :4 stable

2026-06-18 02:41:11 -05:00

share

fix(share): absolute host bind for serve config

2026-06-06 23:06:30 -05:00