feat(vibevoice-asr-seat): switch to Q8_0 (Prime); WER 2/69 vs 3/69 on the bundled clips, +1.1 GB VRAM

persistent-memory.md

@@ -218,6 +218,7 @@ in `servers/pfi-gx10/README.md`. → `persistent-memory.d/2026-09-24-gx10-ac-res
 
 ## Recent decisions
 
+- `[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.
 - `[2026-09-26]` **esh-matter LIVE: a Matter server (matter.js 1.4.0) on CT 111 @ 10.0.90.20, VLAN 90 only**, for ha-dev (operator-approved, relayed). It is on the IoT link because Matter/Thread IPv6 (Echo ULA + RIO routes) is link-only. Kernel RA with rt_info_max_plen=64, forwarding off; Docker ip-forward/iptables off. The in-CT nftables admits :5580 from HA 10.0.50.46 only and SSH from mgmt ranges only, and accepts everything else so Matter UDP never meets conntrack. The CT is in esh-pve vzdump (fabric creds). Acceptance all green: fdad:: SLAAC; ping6 thermostat; 2 Thread routes learned (Aqara + Echo); ws OK from inside the HA container; refused from 10.0.50.45, nh3-dev and a temp VLAN 90 netns. → `servers/esh-matter/README.md`
 - `[2026-09-26]` **Embed/rerank LOAD-SHARED across esh-ml1 + nh3-ml1 (Prime).** Second deployments were added for qwen3-embedding and reranker (config) and for reranker-a3-bge-v2-m3 (DB, `/model/new`). Without failover, a dead backend failed 7/40 embeds; `router_settings.enable_weighted_failover: true` (affects only multi-deployment groups) brought it to rerank 80/80 and embed 38/40 at onset, 60/60 sustained. In LiteLLM 1.97 the embedding path does not stamp the failed deployment, so it relies on cooldown. LiteLLM restarted 3× tonight, ~36 s each.
 - `[2026-09-26]` **Two dataset-foundry utility seats LIVE on nh3-ml1 (brokkr; operator approval relayed):** LFM2.5-VL-3B on llama.cpp `:8030` (gateway `lfm25-vl-3b`, LiteLLM restarted 36 s at 0039) and VibeVoice-ASR-Streaming-1.5B on **audio.cpp** `:8031`, not vibevoice.cpp as specced; the GGUF card names audio.cpp. Controls: VL read a synthetic image exactly, but hallucinates when no image is sent; ASR WER 3/69 on the bundled LibriSpeech clips. libsoxr added (linear resampling misheard a word). First ASR request ~31 s cold. Answered brokkr: the old 27B image seat (10.250.50.54:8015) was retired 09-14; baseline = gateway `image-judge` (Flash-Next).

servers/nh3-ml1/README.md

@@ -56,7 +56,7 @@ VRAM ~2.7 GB for both, so ~13 GB is free.
 - `stacks/lfm-vl-uncensored-seat`: an abliterated LFM2.5-VL-3B running in
   parallel for brokkr's NSFW-caption A/B, `:8032`, direct only.
 
-GPU total with TEI is ~10.2 of 16 GB. All of these are batch workloads, which
+GPU total with TEI is ~11.3 of 16 GB (the ASR seat moved to Q8_0 on 2026-09-26). All of these are batch workloads, which
 suits this card.
 
 **Tried and removed (2026-09-25):** a copy of the code-completion seat

stacks/vibevoice-asr-seat/README.md

@@ -1,6 +1,6 @@
 # vibevoice-asr-seat
 
-**Microsoft VibeVoice-ASR-Streaming-1.5B** (Q4_K) on **nh3-ml1**, served by
+**Microsoft VibeVoice-ASR-Streaming-1.5B** (**Q8_0** since 2026-09-26; was Q4_K) on **nh3-ml1**, served by
 **audio.cpp** (`0xShug0/audio.cpp` v0.8.2-audio8-perf-hotfix, Linux CUDA 12.8
 build) on `:8031`, with direct access only. A utility seat for brokkr's dataset
 foundry (speech → text with inline speaker labels). Operator-approved 2026-09-26,
@@ -25,6 +25,10 @@ is 3/69 = 4.35%, identical across 3 reps. Two of the three errors are "I'm" vs
 "I am" normalization. RTF is 0.07–0.14. ⚠ **The first request after a start
 takes ~31 s** (CUDA graph warmup); later ones take 0.3–1 s. VRAM ~2.0 GB.
 
-The model card recommends Q8_0 (3.3 GB; its CUDA WER is 5.80% vs Q4_K's 7.25%,
-on 69 words). To switch, change `path` in `conf/server.json`, download the file,
-and recreate the container.
+**Switched to Q8_0 (Prime, 2026-09-26 1616).** It is the card's recommended
+package: 3.3 GB on disk, ~1.1 GB more VRAM. On the same 4 clips, WER went from
+3/69 to **2/69 = 2.90%**, identical over 3 reps. The remaining 2 are the
+"I'm"/"I am" normalization artifact, so no real errors are left; Q4_K's
+"forwarded→folded" is gone. RTF is 0.09–0.17 (Q4_K was 0.07–0.14). This is 69
+words, so it separates working from broken, not a fine ranking. The Q4_K file was
+removed. To roll back, download it again and point `conf/server.json` at it.

stacks/vibevoice-asr-seat/conf/server.json

@@ -7,7 +7,7 @@
     {
       "id": "vibevoice-asr-streaming-1.5b",
       "family": "vibevoice_asr_streaming",
-      "path": "/models/vibevoice-asr-streaming-1.5b-q4_k.gguf",
+      "path": "/models/vibevoice-asr-streaming-1.5b-q8_0.gguf",
       "task": "asr",
       "mode": "streaming"
     }