-
c9c42e7489
build: bundle the rocBLAS device libraries at the depth the runtime looks for
Frank
2026-09-16 15:15:02 -07:00
-
05c0d3a00a
frank-medium: llama.cpp layer split across both R9700s, DFlash2 drafting
Frank
2026-09-16 14:59:23 -07:00
-
49e8dfff66
docs: the vault and AGENTS.md describe two Dankclaw endpoints
Frank
2026-09-16 14:52:51 -07:00
-
40850bbc8f
frank-medium: add the RDNA4 llama.cpp build and the GGUF fetch
Frank
2026-09-16 14:52:11 -07:00
-
f367a5a856
deploy: gate the port wait after stopping this service's own quadlet
Frank
2026-09-16 14:06:51 -07:00
-
7403bd3d50
deploy: never start the replacement while the predecessor holds the port
Frank
2026-09-16 14:06:03 -07:00
-
79cca19919
observability: the LLM surfaces follow the two-endpoint split
Frank
2026-09-16 13:59:49 -07:00
-
43ec77fee6
inference: frank-large returns to halogen, frank-small takes the 3090
Frank
2026-09-16 13:58:00 -07:00
-
cc29875842
research: how to make the two R9700s work together (and one correction to my own claim)
Frank
2026-09-16 08:59:12 -07:00
-
f134cc8876
docs+units: fix every claim my own changes made false
Frank
2026-09-16 07:46:23 -07:00
-
2e33ec36c1
docs: three dangling links, and a gate so they cannot accumulate
Frank
2026-09-16 07:43:33 -07:00
-
80a070d71b
homepage: three cards could never be green, and one of them was lying
Frank
2026-09-16 07:36:11 -07:00
-
25f387dd76
tests: gate the hermes vhost, and stop specs from silently not running
Frank
2026-09-16 07:33:38 -07:00
-
9c8a62f9fe
hermes: restore the LAN bind — reverting my own regression
Frank
2026-09-16 07:24:54 -07:00
-
aaa1f82313
hermes: the webui was LAN-facing, not loopback — and Dankclaw's compose is gone
Frank
2026-09-16 07:22:20 -07:00
-
06d3502a2b
docs: ComfyUI is a unit now, and three scrape jobs can never be green
Frank
2026-09-16 07:17:59 -07:00
-
03d44d3308
comfyui: replace the stale compose container with a Quadlet unit
Frank
2026-09-16 07:13:27 -07:00
-
d8835e69c6
docs: six observability units across two hosts, and what the reboot cost
Frank
2026-09-16 07:09:55 -07:00
-
da9cff046e
observability: give Dankserver its node-exporter back as a Quadlet
Frank
2026-09-16 07:07:18 -07:00
-
d104eef4e0
bench: commit the harnesses the frank-medium numbers came from
Frank
2026-09-16 06:59:44 -07:00
-
ef7755f2b3
chore: retire the unreferenced ComfyUI Containerfile
Frank
2026-09-16 06:57:25 -07:00
-
3d5b1f44a6
tests: fix the comfyui Containerfile test instead of living with it red
Frank
2026-09-16 06:57:22 -07:00
-
56edfabb19
homepage: the Dankserver LLM card was labelled with a retired name
Frank
2026-09-16 06:53:20 -07:00
-
823814920b
docs: record the alerting rules and what they still cannot do
Frank
2026-09-16 06:51:23 -07:00
-
af05c890fd
observability: LLM alerting rules, evaluated but not yet delivered
Frank
2026-09-16 06:45:03 -07:00
-
c272407e6e
docs: observability state after the Grafana provisioning
Frank
2026-09-16 06:42:24 -07:00
-
6a191430ca
observability: provision Grafana from git — datasource plus the LLM dashboard
Frank
2026-09-16 06:39:15 -07:00
-
3037240f4f
docs: point the frank-medium runbook and observability page at the new scrape
Frank
2026-09-16 06:37:39 -07:00
-
0f291115bd
observability: name the prefix-cache series that actually exist
Frank
2026-09-16 06:34:53 -07:00
-
42b47063ec
observability: scrape the LLM fleet into Prometheus
Frank
2026-09-16 06:32:33 -07:00
-
4f2f49d2d0
frank-medium: measure admission control and cache health, keep neither change
Frank
2026-09-16 06:31:01 -07:00
-
36007666cf
frank-medium: measure load-balanced placement, find it does not help, revert
Frank
2026-09-16 06:20:19 -07:00
-
99ed1aca9a
docs: reboot-tested the frank-medium trio
Frank
2026-09-16 06:00:15 -07:00
-
24687f561f
docs: frank-medium capacity budget and the CHUNK sweet spot
Frank
2026-09-16 05:49:54 -07:00
-
d8189b65f4
docs: record the proxy's failover behaviour and how to verify it
Frank
2026-09-16 05:29:45 -07:00
-
3946ce9a3a
frank-medium: fail over to the other replica instead of returning 502
Frank
2026-09-16 05:29:08 -07:00
-
4d37f8ca65
frank-medium: measure speculative depth on the repo's own code corpus
Frank
2026-09-16 05:23:09 -07:00
-
8b79766d81
docs: record why CHUNK stays at 8192 on the single-GPU replicas
Frank
2026-09-16 04:58:47 -07:00
-
83625e84a7
frank-medium: route by workload — spread batches, pin agent loops
Frank
2026-09-16 04:45:21 -07:00
-
7ce6e96bc5
docs: record the KV_MEM=0 passthrough trap in the frank-medium runbook
Frank
2026-09-16 03:58:48 -07:00
-
df260dd419
frank-medium: pass KV_MEM=0 through as --kv-cache-memory 0
Frank
2026-09-16 03:54:18 -07:00
-
4fca739bf8
frank-medium: deploy two single-GPU replicas — TP=2 is blocked by PCIe atomics
Frank
2026-09-16 03:47:11 -07:00
-
ad7a9c42f9
log: record the 2026-09-15/16 ik cutover and decode campaign
Frank
2026-09-16 02:44:21 -07:00
-
795b0b3a5e
frank-large: document the ik_llama.cpp deployment and the decode performance campaign
Frank
2026-09-16 02:42:41 -07:00
-
0c1e10a5ef
frank-large: retire the halogen quadlet test seam (ik_llama.cpp cutover)
Frank
2026-09-16 00:38:34 -07:00
-
f2277863ae
docs: rewrite AGENTS.md — dedupe fleet history into lessons + runbook index
Frank
2026-09-16 00:20:58 -07:00
-
f7cda17518
frank-large: restore serve-script defaults eaten by the guard edit
Frank
2026-09-16 00:06:46 -07:00
-
0022ba96ad
frank-large: mmap the CPU-overridden weights (no CUDA_Host pinning)
Frank
2026-09-15 23:58:34 -07:00
-
fe3fc62c41
frank-large: swap halogen Quadlet for ik_llama.cpp (3090 + Strix Halo CPU path)
Frank
2026-09-15 23:20:19 -07:00
-
4183e31fd1
frank-medium: default RADIANCE_NORMQUANT_FUSION under set -u
Frank
2026-09-15 21:02:50 -07:00
-
6b647c1558
frank-medium: swap ROCmFPX llama.cpp runtime for vllm-radiance MXFP4
Frank
2026-09-15 20:59:55 -07:00
-
db29c4d227
hermes: sync sanitized config example to frank-medium cutover
Frank
2026-09-15 19:31:41 -07:00
-
d4024a9180
docs: frank-medium runbook, fleet topology and client cutover
Frank
2026-09-15 19:27:04 -07:00
-
d53962fec5
frank-medium: serve the new R9700s, retire frank-small
Frank
2026-09-15 19:03:30 -07:00
-
6085a99dd8
frank-medium: serve the new R9700s, retire frank-small
Frank
2026-09-15 18:53:20 -07:00
-
27e63359d6
frank-large: correct docs to measured 401408 ledger figures
Frank
2026-09-13 23:02:16 -07:00
-
91406af2a1
frank-large: trim KV pool 524288 -> 401408 for co-tenant headroom
Frank
2026-09-13 22:57:25 -07:00
-
025216fc84
docs(frank-small): document that FreeToken samples by default from the checkpoint
Frank
2026-09-12 23:40:34 -07:00
-
4e3d79388b
docs(frank-large): resync slots=3 across AGENTS and runbook
Frank
2026-09-12 23:26:03 -07:00
-
6169278911
frank-large: lower KV slots 4 -> 3
Frank
2026-09-12 23:20:16 -07:00
-
5dfcde6c8b
docs(dankclaw): record the 64 GiB host swapfile and its safe resize ordering
Frank
2026-09-12 22:03:24 -07:00
-
a45edbcc6d
docs(halogen): record the reserve ceiling and that the reserve is not an allocation
Frank
2026-09-12 21:32:08 -07:00
-
5c4e8a9600
frank-large: revert reserve to 15 — the knob cannot buy host headroom
Frank
2026-09-12 21:30:15 -07:00
-
b87f18569a
frank-large: reserve 15 -> 16 for co-tenant headroom
Frank
2026-09-12 21:27:58 -07:00
-
2969e08616
frank-large: document reserve 15 result — both features in force, stall risk logged
Frank
2026-09-12 21:22:36 -07:00
-
8a79f9a709
frank-large: reserve 24 -> 15, buying the doubled pool AND full prefill speed
Frank
2026-09-12 21:17:16 -07:00
-
65d1be06ac
frank-large: record the 32k outcome — the doubling is clamped away, docs resynced
Frank
2026-09-12 20:12:03 -07:00
-
75ef47c342
frank-large: EXPERIMENT - HALOGEN_MAX_TOK back to upstream default 32768
Frank
2026-09-12 20:06:25 -07:00
-
41ac1f83f6
docs(halogen): record the measured arena model; the 65536 experiment was worth doing
Frank
2026-09-12 19:38:13 -07:00
-
258cccbdcf
Revert "frank-large: EXPERIMENT - raise HALOGEN_MAX_TOK to 65536"
Frank
2026-09-12 19:34:56 -07:00
-
6851940b4d
frank-large: EXPERIMENT - raise HALOGEN_MAX_TOK to 65536
Frank
2026-09-12 19:33:29 -07:00
-
d1544f3191
docs(halogen): correct the memory model to the measured ledger
Frank
2026-09-12 18:04:02 -07:00
-
c489b204ea
frank-large: halve the prefill arena so the doubled pool actually fits
Frank
2026-09-12 17:58:08 -07:00
-
5ace29e7a0
frank-large: double the KV pool to 524288 positions
Frank
2026-09-12 17:51:24 -07:00
-
dd44223f3f
docs: record marreta login-wall fix (proxy provider needs linked Application)
Frank
2026-09-12 13:34:02 -07:00
-
57dbcc5459
e2e: add Marreta homepage-link regression test
Frank
2026-09-12 13:31:00 -07:00
-
6e6bdf7451
homepage: add Marreta as a user-category service
Frank
2026-09-12 12:06:55 -07:00
-
09c54113b1
deploy: make Playwright e2e a mandatory post-deploy gate
Frank
2026-09-12 11:50:40 -07:00
-
476b21b85d
docs(marreta): record live deploy + the base64: APP_KEY gotcha
Frank
2026-09-12 10:31:08 -07:00
-
91e9410c5e
deploy.sh: allowlist dankclaw:marreta in the top-level service gate
Frank
2026-09-12 10:10:01 -07:00
-
73e46a1d85
marreta: content-cleaning proxy on Dankclaw, gluetun netns, guard-chained
Frank
2026-09-12 10:09:12 -07:00
-
9530bffad6
frank-large: ship HALOGEN_VERBOSE=1 for permanent startup evidence
Frank
2026-09-12 10:08:57 -07:00
-
701ab14f60
docs(observability): record the quadlet cutover, the exposure posture, and the unconfigured state
Frank
2026-09-11 16:10:27 -07:00
-
25c93984ad
fix(build): build the loki/alloy/comfyui images in docker format
Frank
2026-09-11 16:02:47 -07:00
-
b96359e628
fix(observability): give loki an exec-form healthcheck in its image
Frank
2026-09-11 16:01:21 -07:00
-
fe4eeaa03a
fix(observability): seed grafana's OAuth secret into a per-service env file
Frank
2026-09-11 15:59:15 -07:00
-
dab81b91a3
feat(observability): make the five Dankclaw observability containers quadlets
Frank
2026-09-11 15:56:39 -07:00
-
549bc57df2
fix(hermes): give title generation a working route on frank-small
Frank
2026-09-11 15:56:38 -07:00
-
351fdf3a05
docs(agents): record the compose-only observability gap found by the reboot
Frank
2026-09-11 15:17:42 -07:00
-
301914b1c0
feat(llm): swap frank-large to halogen-flash-server, retire frank-npu
Frank
2026-09-11 14:48:28 -07:00
-
184df9aa0d
fix(hermes): pin hermes-webui HERMES_HOME to the shared /data home
Frank
2026-09-05 15:41:45 -07:00
-
f75d78a98c
docs(hermes): sync config backup with live post alias-rename propagation
Frank
2026-09-05 14:24:54 -07:00
-
b6f5b2bcf5
test(e2e): pin unauthenticated wget-style denial for copyparty and homepage
Frank
2026-09-04 19:34:59 -07:00
-
32b4016e55
feat(exposure): sonarr to admin side, internal-only by construction
Frank
2026-09-04 18:34:02 -07:00
-
df4339a82a
docs(arr): fleet consolidated on FrankBook-Pro, VPN egress proven; E2E counts derived from registry
Frank
2026-09-04 13:17:39 -07:00
-
4c7465d00d
arr fleet: FrankBook-Pro is the sole deployment, all egress VPN-gated
Frank
2026-09-04 12:54:28 -07:00
-
8a6362401a
docs(arr): dankserver compose stack reactivated 2026-09-04; rootless identity model
Frank
2026-09-04 12:30:16 -07:00
-
e4541598ff
revert(dankserver): drop userns_mode from arr services
Frank
2026-09-04 12:26:16 -07:00
-
df10a33ead
fix(dankserver): arr stack runs with userns_mode host — stack never booted rootless
Frank
2026-09-04 12:14:09 -07:00
-
19b2589362
Add arr stack to homepage admin section — sonarr on user side
Frank
2026-09-04 01:03:36 -07:00