memory: snapshot — 2026-10-11 U13 cutover executed + accepted (one card + svos-dev answers open), fv-ml1 hang + evidence capture, stale snapshots gone; 6 entries archived

2026-10-11 13:54:24 -07:00
parent df7161bbe0
commit 5202847abb
12 changed files with 88 additions and 40 deletions
@@ -15292,3 +15292,25 @@ latency-sensitive.

 - `[2026-09-25]` **elway against an unpinned host-key name** — every `when:` hit ssh rc 255, so all 9 steps showed SKIPPED, not failed. Only the verify phase went red. Pin the name (the IP key matched) or use the IP. Untracked elway defect: rc 255 in `when:` should fail.
  _Archived 2026-10-09._
+
+## Recent decisions (archived 2026-10-11 batch)
+
+- `[2026-09-26]` **GPU-LXC Temperature alerts watch the GPU, not the host CPU** (`SENSORS=-coretemp_*,acpitz`); hypervisor CPU alerts at >95 °C on nh3-pve/esh-pve. The first nh3-ml1 alert was nh3-pve's CPU during vzdump. `6d901c0`.
+  _Archived 2026-10-11._
+
+- `[2026-09-26]` **Abliterated LFM2.5-VL-3B parallel seat (:8032) passed brokkr's 6-image eval and stays** (direct only; stock seat kept for SFW A/B). `9cc3824`, `944bb36`.
+  _Archived 2026-10-11._
+
+- `[2026-09-26]` **VibeVoice ASR → Q8_0 (Prime).** WER on the 4 bundled LibriSpeech clips 3/69 → 2/69 (only the I'm/I am artifact left), ×3 identical; RTF 0.09–0.17; +1.1 GB VRAM (nh3-ml1 ~11.3/16 GB). Q4_K file removed.
+  _Archived 2026-10-11._
+
+- **Coder seat STAYS on fv-ml1 (Prime).** The nh3-ml1 copy gave the same quality (teacher-forced true-code logprob diff +0.008 ± 0.019) but ran ~5× slower (64-tok FIM ~1.0 s vs 0.2 s; 63 vs 338 tok/s), and freeing 6.3 GB on fv-ml1 GPU 1 (20 GB spare) bought little. Copy removed; recipe kept in `stacks/coder-seat/`. The RTX 2000E suits embed/rerank/classify, not latency-sensitive generation. Only coder was a candidate: parakeet stays (earlier ruling), voices is a generation seat.
+  _Archived 2026-10-11._
+
+## Tried and abandoned (archived 2026-10-11 batch)
+
+- **Greedy exact-match as parity for a small generative seat (coder)** — same-host repeats matched only 30–38% (run B hit the prefix cache, so a different numeric path; near-tie tokens cascade). Use teacher-forced logprobs of real text with `cache_salt`: each host bit-exact with itself. `stacks/coder-seat/README.md`.
+  _Archived 2026-10-11._
+
+- `[2026-09-26]` **zsh traps in ad-hoc test loops:** `set -- $a` does NOT word-split, so alert POSTs went out with empty fields; and `local path=…` inside a function CLOBBERS `$PATH` (zsh's tied array), so the failover test ran nothing while TEI sat stopped. Use other names and `${=var}`.
+  _Archived 2026-10-11._