-
33438f223e
deploy.sh: fix LLM unit relink (unquote source path so ${unit} expands on the remote)
Frank
2026-08-22 22:33:16 -07:00
-
fbfbc5c6db
Move frank-default-large from :52625 to :7770 (dankclaw)
Frank
2026-08-22 22:11:53 -07:00
-
e367f0f43c
docs: one knowledge base for both workloads
Frank
2026-08-22 15:50:39 -07:00
-
0461988884
scripts: unified deploy.sh; units + bench point at the merged checkout
Frank
2026-08-22 15:50:39 -07:00
-
2b115ad378
README + .gitignore: one repo, two workloads
Frank
2026-08-22 15:49:44 -07:00
-
99c7255db5
AGENTS.md: merge the vllmScripts directives into one rulebook
Frank
2026-08-22 15:49:44 -07:00
-
8001132744
tests: LLM unit integrity, sudo-password leak guard, deploy LLM coverage
Frank
2026-08-22 15:49:44 -07:00
-
02f92386f8
Reorganize: drop llm/ root files (absorbed into repo root in follow-up)
Frank
2026-08-22 15:07:31 -07:00
-
1f565c08c3
Reorganize: tracked DFlash2 archive files -> .AGENTTRASH (repo root)
Frank
2026-08-22 15:07:31 -07:00
-
aa410f4439
Reorganize: llm/scripts -> scripts (LLM serve scripts, units, env tunables)
Frank
2026-08-22 15:07:31 -07:00
-
506614525b
Reorganize: llm/old -> old (retired PP/NPU-era scripts, reference only)
Frank
2026-08-22 15:06:51 -07:00
-
017d3704d2
Reorganize: llm/docs -> docs/LLM (LLM fleet knowledge base)
Frank
2026-08-22 15:06:51 -07:00
-
918fe4eaec
Merge vllmScripts under llm/ (history-preserving import)
Frank
2026-08-22 15:03:03 -07:00
-
-
0746303c33
AGENTS.md: add mandatory Deletion & Trash Policy (mv to ~/.AGENTTRASH, never rm)
Frank
2026-08-22 14:38:26 -07:00
-
c45f40a9b3
docs: reflect rollback to v3 on frank-default-large (Kairic experiment retired)
Frank
2026-08-22 14:26:32 -07:00
-
c60156f3c6
rollback frank-default-large to v3 CIRU ActiveFPX PromptForge (operator decision 2026-08-22)
Frank
2026-08-22 14:20:11 -07:00
-
ddcd54e13f
Make FASTGREEDY=1 compatible with non-greedy requests (argmax fastpath always off)
Frank
2026-08-22 13:45:57 -07:00
-
c66273033b
open-webui: reach the LLM via the host LAN IP, not the bridge gateway
Frank
2026-08-22 13:44:48 -07:00
-
0951fc165c
deploy.sh: restart caddy when the reload cannot be applied
Frank
2026-08-22 13:35:31 -07:00
-
fe71f7516d
Flip production to FASTGREEDY=1 (operator decision 2026-08-22)
Frank
2026-08-22 13:26:52 -07:00
-
bb1ef9bfb3
caddy: keep admin on for hot reload; deploy.sh reloads after render
Frank
2026-08-22 13:22:01 -07:00
-
da68b5ce1a
fix: basicauth double-prefix render; open-webui off slirp4netns
Frank
2026-08-22 13:07:43 -07:00
-
7f617b9ee6
Add bench-fast-greedy.sh for A/B fast-greedy vs production testing
Frank
2026-08-22 13:03:21 -07:00
-
f347d5336f
Set FASTGREEDY=0 default in env; add deployment notes
Frank
2026-08-22 12:54:28 -07:00
-
e0329caef8
Add FASTGREEDY mode to serve script: reasoning off + argmax fastpath on for fast-greedy bench
Frank
2026-08-22 12:54:03 -07:00
-
5c9ade4e7a
deploy.sh: fix quoting of the cfhash label template
Frank
2026-08-22 04:16:18 -07:00
-
e0a25415a1
deploy.sh: rebuild local images when their Containerfile hash changes
Frank
2026-08-22 04:14:43 -07:00
-
7d6a285371
copy /bin/busybox.static (the -static package does not overwrite /bin/busybox)
Frank
2026-08-22 04:13:55 -07:00
-
824d39a29e
busybox-static for healthchecks; alloy: rely on the default UI listener
Frank
2026-08-22 04:09:04 -07:00
-
dc421efea4
deploy.sh: escape the image-build loop vars for the remote string
Frank
2026-08-22 04:04:24 -07:00
-
1f4766f233
observability: fix alloy ui component, busybox healthchecks for loki/alloy
Frank
2026-08-22 03:59:14 -07:00
-
816986139f
observability: run as container root 0:0, not 1000:1000
Frank
2026-08-22 03:45:06 -07:00
-
8b3cd2ff79
large: batch profile flipped to the runner's qualified 2048/512 (fastest measured prefill, no decode change)
Frank
2026-08-22 03:30:58 -07:00
-
11a713e05d
observability: run prometheus/loki/grafana as 1000:1000 (rootless uid mapping)
Frank
2026-08-22 03:22:50 -07:00
-
3c3f90e4c5
observability: run prometheus/loki/grafana as 1000:1000 (rootless uid mapping)
Frank
2026-08-22 03:19:18 -07:00
-
3eb29de6ca
bench: add prod2048 config (runner batch sizes on chat profile); docs: ubatch 4096 results
Frank
2026-08-22 03:17:41 -07:00
-
765fd2902a
searxng limiter: fix stale .local name in comment
Frank
2026-08-22 03:05:25 -07:00
-
37f05a5f9e
observability configs: loki delete_request_store, prometheus targets, grafana root URL
Frank
2026-08-22 03:04:36 -07:00
-
bd9d75af3f
large: ubatch 4096 (match v3), add --no-mmap; sync bench harness prod config
Frank
2026-08-22 03:02:16 -07:00
-
0f2bb636d7
bench: add Kairic A/B config harness; make v3 spec-stack harness self-contained (survives the Kairic env swap)
Frank
2026-08-22 02:53:32 -07:00
-
4360551938
docs: Kairic Edge reference page (build runbook, config.h gap, byte-identical rebuild, perf data); mark v3 doc retired; update topology
Frank
2026-08-22 02:51:39 -07:00
-
d5e9c63802
deploy.sh: drop --pull=missing (podman-compose rejects it)
Frank
2026-08-22 02:51:36 -07:00
-
5caa2178c3
deploy.sh: extract CADDY_BASICAUTH_HASH via grep/cut, not by sourcing .env
Frank
2026-08-22 02:51:13 -07:00
-
4572a7739d
Caddy: keep plex.franksdank.casa HTTP-only (not in the Porkbun account)
Frank
2026-08-22 02:44:06 -07:00
-
7f4ce06927
Caddy phase 2: TLS via Porkbun DNS-01 wildcard certs
Frank
2026-08-22 02:39:41 -07:00
-
40d50fed06
large: Kairic chat profile — reasoning on/deepseek, batch 4096/2048, preserve_thinking
Frank
2026-08-22 01:03:08 -07:00
-
7aee1df7c3
large: disable target greedy argmax fast path for chat (Kairic)
Frank
2026-08-22 00:28:01 -07:00
-
4c438390dd
large: swap to Kairic Edge IU4 runtime (pinned tag kairic-edge-qwen38-27b-v1)
Frank
2026-08-22 00:21:34 -07:00
-
e9277fa8af
deploy.sh: declare the per-target arrays (COMPOSE_UP, COPY_SHARED)
Frank
2026-08-21 23:17:04 -07:00
-
724f186414
Caddyfile: transport is a reverse_proxy subdirective (jellyfin h2c)
Frank
2026-08-21 21:50:03 -07:00
-
6ea490d64c
deploy: sed delimiter '|' for the Caddyfile hash render (bcrypt contains '/')
Frank
2026-08-21 21:48:15 -07:00
-
4a679fce80
Caddy consolidation: single Caddy on dankclaw fronts both hosts
Frank
2026-08-21 21:44:18 -07:00
-
3b9379bb5b
docs: benchmark ngram-mod stacking under MTP on v3 stack (8k/16k/64k); doc CLI order does not matter
Frank
2026-08-21 20:43:47 -07:00
-
f1074ade61
fix bench harness: derive PromptForge sidecar paths and set PROMPTFORGE_MODE; correct env sourcing
Frank
2026-08-21 20:25:49 -07:00
-
fb49202e59
bench: add spec-stack comparison harness (MTP vs MTP+ngram-mod) for v3 CIRU stack
Frank
2026-08-21 20:25:00 -07:00
-
8652116cec
docs: document frank-default-large v3 upgrade (Output-K8 runtime, full-W8 views, BF16 vision); update AGENTS/deploy payload sizes
Frank
2026-08-21 20:18:57 -07:00
-
132e1478cf
frank-default-large: upgrade to CIRU ActiveFPX PromptForge v3 (Output-K8 runtime, full-W8 views, BF16 vision)
Frank
2026-08-21 20:00:31 -07:00
-
9c8298df05
dnsmasq: add foreground CMD to the image
Frank
2026-08-21 19:38:08 -07:00
-
577003bed1
docs: 2026-08-21 incident (distro upgrade dropped CUDA 12 runtime) + Dankserver now Ubuntu 26.04
Frank
2026-08-21 19:31:05 -07:00
-
1dee2b737e
dnsmasq: serve config from /home/dank/Containers/dnsmasq (rootless-correct)
Frank
2026-08-21 16:41:49 -07:00
-
55d062ea7b
DNS names: flat franksdank.cloud scheme + dnsmasq LAN DNS on dankclaw
Frank
2026-08-21 16:31:42 -07:00
-
833f2365f1
env: add Porkbun API token placeholders (DNS-01 cert issuance)
Frank
2026-08-21 16:00:16 -07:00
-
583efd8cf2
Revert frank-default-large to CIRU runtime (DFlash2 too slow)
Frank
2026-08-21 10:15:23 -07:00
-
fd0d9dd3cb
Swap frank-default-large to DFlash2 on Vulkan (PR #27342)
Frank
2026-08-21 09:40:08 -07:00
-
e8fd009419
docs: fact-check sweep, Jellyfin, *arr stack, and ROM scraping guides
Frank
2026-08-20 23:40:07 -07:00
-
34918bf309
hermes: add web dashboard service (basic-auth, LAN port 9119)
Frank
2026-08-20 22:54:35 -07:00
-
701e742f9f
docs: fix stale context sizes, batch/ubatch, MoE, and service names across docs
Frank
2026-08-20 22:48:25 -07:00
-
afdb686c7e
hermes: point Signal adapter at host LAN IP (slirp alias drops at runtime)
Frank
2026-08-20 21:34:48 -07:00
-
cd6202f4a0
frank-default-large: batch=4096, ubatch=4096 (OOM fix)
Frank
2026-08-20 19:58:45 -07:00
-
e09733ee62
frank-default-small: ncpu-moe=38 (2 experts on GPU)
Frank
2026-08-20 18:02:47 -07:00
-
99c7ec531f
frank-default-small: ncpu-moe=32(8GPU, ~51.1 t/s gen)
Frank
2026-08-20 17:09:23 -07:00
-
90a8b1c562
frank-default-small: ncpu-moe=35(5GPU), ubatch=2048, ctx-checkpoints=16
Frank
2026-08-20 16:57:03 -07:00
-
f8deaa798a
frank-default-small: ubatch=1536, ctx-checkpoints=2 (24GB VRAM safety)
Frank
2026-08-20 16:51:40 -07:00
-
87dd4224b7
frank-default-small: ubatch=1024 (tight 24GB VRAM, 786k ctx pool)
Frank
2026-08-20 16:45:20 -07:00
-
26b42d2d6e
frank-default-small: batch=4096, ubatch=2048
Frank
2026-08-20 15:53:41 -07:00
-
e777d3cb56
frank-default-small: batch=5120, ubatch=2560
Frank
2026-08-20 15:38:18 -07:00
-
635e074af8
frank-default-small: ctx=786432 max, parallel=3, batch=6144
Frank
2026-08-20 15:27:37 -07:00
-
ed9e1cabfd
frank-default-npu: deploy gemma4-it:e4b on NPU, port 52626, 64k context
Frank
2026-08-20 15:12:58 -07:00
-
b38e384f6e
frank-default-large: raise ubatch back to 8192
Frank
2026-08-20 14:56:14 -07:00
-
eb3f26f747
frank-default-large: reduce batch=8192, ubatch=4096
Frank
2026-08-20 14:52:31 -07:00
-
8bf9462f21
frank-default-large: add --no-mmap to avoid teardown OOM
Frank
2026-08-20 14:49:45 -07:00
-
d9bf09e144
frank-default-large: raise batch=16384, ubatch=8192
Frank
2026-08-20 14:41:41 -07:00
-
b5211845da
Hermes: drop the :8090 healthcheck (messaging gateway has no HTTP endpoint)
Frank
2026-08-20 12:30:42 -07:00
-
5e71b0584f
Hermes: run the headless gateway (not the TUI) in the container
Frank
2026-08-20 12:14:18 -07:00
-
b167a07fc0
Hermes: wire the Signal adapter to the host signal-cli daemon
Frank
2026-08-20 10:51:43 -07:00
-
bec56d2a72
ComfyUI docs: document ComfyUI-Manager custom node
Frank
2026-08-19 17:10:26 -07:00
-
a6b3d4a522
ComfyUI docs: reflect the running GPU (gfx1151) container
Frank
2026-08-19 16:16:50 -07:00
-
bb188becb0
ComfyUI: persist models/data under /home/dank/comfyui-data
Frank
2026-08-19 15:55:00 -07:00
-
88dcd3cc0f
ComfyUI: rootless-correct GPU group config (video only)
Frank
2026-08-19 15:50:28 -07:00
-
354ddc473f
ComfyUI: GPU (gfx1151 ROCm) build via container
Frank
2026-08-19 15:20:46 -07:00
-
aa38c556b2
SearXNG docs: document the 429 root cause + pass_ip fix
Frank
2026-08-19 12:48:53 -07:00
-
4fa9fdf36e
SearXNG: stop rate-limiting local subagents (pass_ip covers slirp/loopback)
Frank
2026-08-19 12:29:27 -07:00
-
048e7eaaba
frank-default-large: full 262144 context, 1 slot
Frank
2026-08-19 11:17:31 -07:00
-
8e4d63b68d
frank-default-large: 200000 context per slot (400000 total, 2 slots)
Frank
2026-08-19 10:38:53 -07:00
-
3d4604c19f
Stand up open-webui on dankclaw (quadlet, boot-persistent)
Frank
2026-08-19 01:04:50 -07:00
-
9f4c0d4175
SearXNG deploy: do not systemctl enable the generated units
Frank
2026-08-19 00:05:25 -07:00
-
06ac2a84cf
SearXNG quadlet: use AddDevice= (the generator rejects Device=)
Frank
2026-08-19 00:00:42 -07:00
-
e036296fd1
SearXNG quadlet: use AddCapability= (the generator rejects CapAdd=)
Frank
2026-08-18 23:57:25 -07:00
-
d7c66b0253
SearXNG deploy: do not enable the generated .network unit
Frank
2026-08-18 23:46:35 -07:00
-
f488382b08
SearXNG docs: document quadlet boot-persistence for the trio
Frank
2026-08-18 23:42:45 -07:00