Commit Graph

  • ace839c768 memory: snapshot — the Ada inference server is a stripped used R750xa main vh 2026-09-01 17:01:14 -07:00
  • 7142657749 memory: snapshot — gx10 unracked and next up as an inference+training box vh 2026-09-01 16:32:21 -07:00
  • 73866f6a7e feat(synapse): restrict /_synapse/admin to LAN, track the stack vh 2026-09-01 16:09:42 -07:00
  • a0c5fc6ed5 feat(pfi-gx10): rack-move network playbook — VLAN 50, static 10.100.50.60 vh 2026-09-01 11:41:43 -07:00
  • 8fb8cc87ca chore(pfi-gx10): first inventory snapshot vh 2026-09-01 11:37:09 -07:00
  • 1b596c8c30 feat(pfi-gx10): register the ASUS Ascent GX10 and convert it to headless vh 2026-09-01 11:36:49 -07:00
  • 931bac8f68 docs(matrix): current state, upgrade procedure, alias convention, push findings vh 2026-09-01 11:16:07 -07:00
  • 9e986d8ee8 feat(phasefinal-web): cloudflare edge config — cache ruleset + always online vh 2026-08-29 23:02:49 -07:00
  • 524aa4d860 fix(phasefinal-web): healthcheck targeted ::1, so traefik skipped the container vh 2026-08-29 22:33:21 -07:00
  • 7ffbee6f09 feat(phasefinal-web): corporate site stack on ana-docker vh 2026-08-29 22:25:20 -07:00
  • c488eadc31 memory: snapshot — althing v3 fleet-wide at 3.1.1, sec on GPU0 vh 2026-08-28 22:22:32 -07:00
  • 583f329d00 fix(playbooks): 3.1.1 deploy — and why the markers match presence, not count vh 2026-08-28 16:15:34 -07:00
  • 8a04d6f1bb fix(statusline): resolve the handle from the v3 binding, not the v2 map vh 2026-08-28 16:01:11 -07:00
  • c648a40b68 fix(playbooks): 3.1.0 herald deploy; the marker check takes a LIST now vh 2026-08-28 15:59:06 -07:00
  • 590b55f7d8 feat(playbooks): potrace/agg headers, with the two traps that mislead vh 2026-08-28 13:38:12 -07:00
  • cdeb57c18b fix(playbooks): 3.0.3 herald deploy, and a content check that survives releases vh 2026-08-28 12:46:35 -07:00
  • 9f87ff87e1 feat(althing): surface the post office on Homepage under Toolchain vh 2026-08-28 10:37:12 -07:00
  • dbb930d546 fix(playbooks): 3.0.1 herald reinstall, and two guards that were release-hostile vh 2026-08-28 10:27:03 -07:00
  • 22da609053 feat: registry-push the post office image; version the statusline vh 2026-08-28 10:08:05 -07:00
  • 9d4e7bd34a feat(althing): move the post office to nh3-docker vh 2026-08-28 10:02:13 -07:00
  • e58360668e feat: althing v3.0.0 cutover (U9b) and the sec seat onto GPU0 vh 2026-08-28 07:19:18 -07:00
  • f875f746b8 feat(playbooks): nh3-dev memory forensics — and the OOM hog is Claude Code vh 2026-08-28 06:03:43 -07:00
  • ea818380ff memory: the rack is one circuit — my blast-radius objection was wrong vh 2026-08-27 11:04:42 -07:00
  • 3cc55b4b40 memory: separate the measured breaker trip from the load hypothesis vh 2026-08-27 11:01:44 -07:00
  • 88d79375f7 memory: run 3c had TWO launches — the third was an untimestamped report vh 2026-08-27 10:59:29 -07:00
  • 98e7d4886a memory: snapshot — run 3 gated DO-NOT-SERVE, run 3c held on a tripped breaker vh 2026-08-27 07:27:41 -07:00
  • 1a36e60d3a docs(quant-playbook): §3.7's APC-off mitigation was reverted nine days ago and the section never said so vh 2026-08-26 16:35:43 -07:00
  • 5a51e76135 fix(litellm)!: withdraw merged-run03 — measured self-harm guardrail regression vh 2026-08-26 14:52:20 -07:00
  • c577d69e2d feat(litellm): expose run-3's merged tune for parallel hand-testing vh 2026-08-26 13:39:26 -07:00
  • b6ce22ddcb feat(litellm): register the run-3 gate base arm at operator request vh 2026-08-26 13:21:31 -07:00
  • 71e44176e9 memory: snapshot — run 3 corpus built and held on a megamix containment defect vh 2026-08-26 09:40:18 -07:00
  • 1a4ef5c7a1 docs(training-playbook): 4.6.3 was wrong twice — correct it, and keep the retraction visible vh 2026-08-26 09:29:14 -07:00
  • 37d3189622 docs(erp-dpo): the clip hypothesis is falsified — the distribution is bimodal vh 2026-08-26 06:22:21 -07:00
  • 1e4d827c5d memory: erp-tune-v2 registered in the LiteLLM gateway at operator request vh 2026-08-26 06:21:09 -07:00
  • b5bbc29b91 memory: gate verdict FAIL — and the T6/T3 trade is what the pair of runs bought vh 2026-08-26 02:25:35 -07:00
  • 5171f19e16 docs(erp-dpo): the PIPPA length clip, measured — DPO pairs would inherit it vh 2026-08-26 02:25:08 -07:00
  • 0bb9ee7777 docs(training-playbook): 4.6.3 — a short-answer gate cannot see a long-form defect vh 2026-08-26 02:24:42 -07:00
  • 3ae32ddc7f memory: base set complete, tuned arm live with digests verified identical vh 2026-08-26 02:09:14 -07:00
  • a0f59d2778 docs(training-playbook): 4.6.2 — a null result needs a positive control vh 2026-08-26 02:08:59 -07:00
  • 3df8707e28 memory: base arm live, tuned arm down — battery running sequentially vh 2026-08-26 01:53:47 -07:00
  • 62f01a02da memory: snapshot — run 2 trained, merged, coherence-gated and serving as erp-tune-v2 vh 2026-08-26 01:37:53 -07:00
  • d54f25605f docs(training-playbook): 4.4.1 sample identity at launch; 4.6.1 calibrate gates against correct input vh 2026-08-26 01:35:27 -07:00
  • bcf63db527 docs(erp-dpo): readiness survey for the DPO stage vh 2026-08-26 01:18:41 -07:00
  • dbca9a3c66 docs(training-playbook): audit the whole manifest against the pairing rule vh 2026-08-25 21:30:45 -07:00
  • c1db188e6a docs(training-playbook): §4.3 records an OBSERVED consequence, not just a config string vh 2026-08-25 20:30:29 -07:00
  • dae6ede8e2 docs(training-playbook): §4 — when the artifact lies about itself vh 2026-08-25 18:12:27 -07:00
  • 2656196f47 memory: snapshot — the tune is trained, gated, and serving vh 2026-08-25 16:54:49 -07:00
  • 2a05ae91af feat(training-probes): counted-not-surfaced classifier scaffold vh 2026-08-25 13:10:03 -07:00
  • 64bf9d313f docs(training-playbook): measure refusal retention on the abliteration's OWN axis vh 2026-08-25 12:58:27 -07:00
  • a696b49e2a docs(training-playbook): merging a tune back toward stock can UNDO an abliteration vh 2026-08-25 08:46:59 -07:00
  • 2ec8f42297 docs(training-playbook): base-viability pre-flight, three greps before you pick vh 2026-08-25 01:51:06 -07:00
  • 96731bb090 docs(training-playbook): prove the serving path before spending the window vh 2026-08-25 01:36:01 -07:00
  • 8de5f7a73c docs(gemma4-erp-tune): merged weights are mandatory — vLLM cannot LoRA any Gemma-4 vh 2026-08-25 01:32:51 -07:00
  • ab980e9345 fix(erp-tune-serve): four defects the end-to-end dry run found, all silent vh 2026-08-24 23:48:25 -07:00
  • 6a8582936e feat(erp-tune): NVFP4A16 serving pipeline, and the MoE landmine it uncovered vh 2026-08-24 23:11:33 -07:00
  • 7b5fd91d3c docs(gemma4-erp-tune): root-cause the 8.6% MFU — attention on Ampere kernels, 29.9% padding vh 2026-08-24 22:10:51 -07:00
  • 872c2c562f memory: the MFU hunt — two hypotheses measured and killed, consult dispatched vh 2026-08-24 21:38:15 -07:00
  • 07743c6aff memory: snapshot — tune training unattended, MFU root-caused to a Python expert loop vh 2026-08-24 21:20:58 -07:00
  • d6dfd61c91 memory: the ERP tune is running — override granted, 12 defects fixed first vh 2026-08-24 21:00:17 -07:00
  • 33433e0d1e docs(gemma4-erp-tune): replace the estimates with measurements — they were 3x optimistic vh 2026-08-24 19:01:49 -07:00
  • 47ec3d1a97 memory: the ERP tune is blocked on a corpus gate only the operator can clear vh 2026-08-24 18:51:52 -07:00
  • c9943b1507 docs(gemma4-erp-tune): whole-card placement — gen moves to GPU1, sec stands down vh 2026-08-24 18:40:42 -07:00
  • 9d70100867 feat(ana-ml2): elway playbooks to open and close the ERP/RP tune window vh 2026-08-24 18:39:13 -07:00
  • c507db9ac0 docs(gemma4-erp-tune): size the run against the checkpoint — QLoRA is structurally unavailable vh 2026-08-24 18:28:30 -07:00
  • 9d0e628643 memory: correct the vLLM version claim — ana-ml2 runs a spread, and 0.27.1 is on disk vh 2026-08-24 18:12:35 -07:00
  • 668e590e7d memory: snapshot — char-rp on the Gemma-4 MoE, abliterated trainee staged, QLoRA sizing next vh 2026-08-24 15:54:30 -07:00
  • 5415fd4b30 docs(gemma4-charrp): abliteration measured in isolation — close to free, but it MOVES capability vh 2026-08-24 15:39:43 -07:00
  • 019ccff7e8 feat(gemma4-trainee-bench): BF16 bench stack; record that gen's footprint grows with uptime vh 2026-08-24 14:19:49 -07:00
  • 14ff4a3f57 docs(gemma4-charrp): stage two abliterated trainee bases; record the endemic stale-template trap vh 2026-08-24 13:52:00 -07:00
  • 8d6a9390de docs(gemma4-charrp): RETRACT the contradiction-deficit claim — the item was ill-posed vh 2026-08-24 13:07:58 -07:00
  • 3446367d5e feat(gemma4-charrp): pin the chat template; A16 control run executed and reverted vh 2026-08-24 13:03:19 -07:00
  • 1bd90eaacc docs(gemma4-charrp): stack README — the three model dirs, the A16 control runbook vh 2026-08-24 12:34:50 -07:00
  • 24e8826219 docs(gemma4-charrp): the A16 control needs a chat-template override, not just a path swap vh 2026-08-24 12:32:03 -07:00
  • f509668e45 docs(gemma4-charrp): record the A16 activation-axis control and how to run it vh 2026-08-24 12:28:17 -07:00
  • 27155c0f3b feat(char-rp): swap the seat to the Gemma-4 26B-A4B MoE, NVFP4, same port vh 2026-08-24 12:02:59 -07:00
  • 850e0c3351 fix(meromero-charrp): drop GPU0 budget to 0.47 — the seat was OOM crash-looping vh 2026-08-24 11:42:36 -07:00
  • 35adc4a043 feat(homepage): rebuild on Australis Skyfall — dual theme, light mode shipped vh 2026-08-24 09:44:45 -07:00
  • 39da1d4a97 feat(homepage): recategorise on "do I open this?", collapse the API groups vh 2026-08-24 08:54:06 -07:00
  • f6f2f69649 fix(homepage): uniform 4-column grid, hold the status gutter, unleak AI Systems vh 2026-08-24 08:15:02 -07:00
  • 32349b7653 memory: snapshot — Anaheim tunnels on AES-128, ana-gw admin closed, Scriberr live, ESH DNS fixed vh 2026-08-24 07:42:30 -07:00
  • d419b11d43 docs(ipv6): close the NH3 multi-prefix question by operator ruling vh 2026-08-24 07:05:49 -07:00
  • bd209951ac docs(ipv6): correct the NH3 delegation finding — eight /64s, not one vh 2026-08-24 07:02:53 -07:00
  • d127e29fac docs(ipv6): retire the next-candidates line now that all three are done vh 2026-08-23 22:39:51 -07:00
  • 4e83395ddf docs(ipv6): all three esh-server hosts now carry the segment name vh 2026-08-23 22:39:39 -07:00
  • ffb7fba346 docs: give the ESH IPv6 naming scheme a home, and make it real on one host vh 2026-08-23 22:31:13 -07:00
  • 41091eef8f docs: restore the ESH IPv6 naming scheme, dropped by a snapshot vh 2026-08-23 22:26:39 -07:00
  • 6217d3993e fix(scriberr): force uv to copy rather than reflink when building envs vh 2026-08-23 19:31:54 -07:00
  • efddb4e511 feat(scriberr): stand up transcription on ana-ml2, pinned to GPU1 vh 2026-08-23 19:28:33 -07:00
  • 22ae9cd480 ops(ana-gw): disable ACME and retract the all-port VIP claim vh 2026-08-23 16:02:20 -07:00
  • f39b66d2e1 ops(ana-gw): port 80 on the WAN address is the FortiOS ACME listener vh 2026-08-23 15:53:55 -07:00
  • 7bc9754e40 ops: adopt AES-128 on both Anaheim tunnels and close public admin vh 2026-08-23 15:44:32 -07:00
  • 6edebe4864 ops: forwarding through a downstream WireGuard terminator is free vh 2026-08-23 15:25:05 -07:00
  • 062215e81a ops: the tunnel's per-flow limit is the UDM's software AES-CBC vh 2026-08-23 15:11:51 -07:00
  • 8ecffa1fab ops: WireGuard over the same path carries one stream at 767 Mbit/s vh 2026-08-23 14:57:29 -07:00
  • d42e9d8712 ops(ana-gw): establish the per-flow cap is IPsec-specific, not capacity vh 2026-08-23 14:48:34 -07:00
  • cf0cb2cbb3 ops(ana-gw): pin down what limits a single stream across the tunnel vh 2026-08-23 14:42:17 -07:00
  • e41d19f1cb ops(ana-gw): close out the Anaheim tunnel cipher question vh 2026-08-23 14:29:17 -07:00
  • 5af362e9d0 ops(ana-gw): restore WAN admin access ahead of the FortiGate cutover vh 2026-08-23 14:03:40 -07:00
  • b6340519bc memory: snapshot — Anaheim tunnel at 25% of circuit; selene retired; hrafn CI fixed vh 2026-08-23 13:35:48 -07:00
  • 0ad332bb4a feat(booth): per-row link removal + render the link board as real UI vh 2026-08-23 12:55:33 -07:00