vh/esh-pfi-infrastructure
1
0
Fork 0
Code Issues 7 Pull Requests Actions Packages Projects Releases Wiki Activity

fix(scriberr): patch 0004 — chat context window for gateway model names (LLM_DEFAULT_CONTEXT_WINDOW=131072); image ctxwin1 live, idle-gated deploy; svos v0.0.20 staged

Browse Source
This commit is contained in:
vh
2026-10-19 00:55:37 -07:00
parent 94ba3c659c
commit 5be01a8744
7 changed files with 139 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
persistent-memory.md
+1 -1
View File
@@ -157,7 +157,7 @@ Every vLLM seat on a shared card (GPU 0, GPU 1) has byte-pinned KV (`--kv-cache-
### Scriberr (fv-ml1 GPU 3)
- **LIVE `scriberr:local-blackwell-a353078-dropout2`:** upstream a353078 plus patch 0001 (overlap slicer) and patch 0002 (gap retry, `PARAKEET_MODEL_PATH`), carried LOCALLY ONLY (Prime 2026-10-01: no upstream). `patches/0003` (pyannote env pins) is committed but in no image yet. `scripts/scriberr-rebuild` re-applies all. → `persistent-memory.d/2026-09-30-scriberr-slicer-gap-retry-gpu3.md`
- **LIVE `scriberr:local-blackwell-a353078-ctxwin1` since 10-19 0051** (was `-dropout2`, kept for rollback via `.env.bak-ctxwin1`): upstream a353078 + patches 0001 (overlap slicer), 0002 (gap retry, `PARAKEET_MODEL_PATH`), 0003 (pyannote env pins; the live env was already hand-pinned and did not re-sync) and **0004 (chat: `LLM_DEFAULT_CONTEXT_WINDOW`, compose default 131072; upstream returned 4096 for every non-OpenAI model name, so chat refused any transcript over ~3.6k est. tokens — Prime 10-19 "fix the 4k issue")**, carried LOCALLY ONLY (Prime 2026-10-01: no upstream). Rebuild verdict PASS; deploy idle-GATED (`playbooks/scriberr-deploy-ctxwin1.yaml`). Post-deploy: all 7 backends initialized; pyannote smoke on the new container: AMI 19.5 s clip 2 speakers ×3 runs identical, SCOTUS 30 min 8 speakers / 196 segments in 52 s (functional check, not a quality comparison). Chat end-to-end not re-tried by infra-ops (Prime's account). ⚠ Prime uses Scriberr LIVE IN MEETINGS (diarized notes): every restart must pass the idle gate. Prime is weighing a from-scratch rebuild with Scriberr as the reference design (10-19; no decision). `scripts/scriberr-rebuild` re-applies all. → `persistent-memory.d/2026-09-30-scriberr-slicer-gap-retry-gpu3.md`
- **Diarization works (fixed 2026-10-06):** model caches on /tank, `HF_TOKEN` a placeholder, saved profiles `WhisperX large-v3 · GPU · speakers` and `Parakeet · GPU · speakers`. A NEW gated model needs one download with vault `fv-ml1/hf-token`. → `persistent-memory.d/2026-10-06-scriberr-diarization-fixed.md`
- **URL is https://scriberr.nh3.phasefinal.com since 2026-10-01** (fronted by the fleet TLS caddy on nh3-dev; Caddyfile backup `.bak-20261001-scriberr`). `getUserMedia` is refused on http://, so recording needs the HTTPS name, which is in `ALLOWED_ORIGINS` (`playbooks/scriberr-https-origin.yaml`). ⚠ `/opt/docker/compose/scriberr` is lkraven-owned, so `deploy-stack.sh` as infra-ops FAILS there; upload single files with `elway --upload --sudo --owner lkraven:lkraven`.