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

memory: scriberr pause-aware slicer build in flight

Browse Source
This commit is contained in:
vh
2026-09-30 10:53:03 -07:00
parent 128d1d847c
commit ab62644315
1 changed files with 5 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
persistent-memory.md
+5
View File
@@ -179,6 +179,11 @@ _As of 2026-09-30 ~0120 PT._
### SemIf on fv-ml1 GPU 1 (2026-09-27, Prime) ### SemIf on fv-ml1 GPU 1 (2026-09-27, Prime)
- **OFFLINE since 2026-09-30 0135 PT (Prime: "take semif offline for now; we'll optimize scriberr later").** Stopped with `docker compose stop`, not removed, to give scriberr back its GPU 1 room. Scriberr's Parakeet path hardcodes `--chunk-len 300`, and the attention memory grows with the square of the slice, so it needs over 6 GB; it hit CUDA OOM at 0124 on a 35-min file with ~6.7 GB free. Stopping SemIf moved GPU 1 from 91,052 to 81,806 MiB used. The same job re-run at 0137 finished clean: 35m17s of audio in 44 s. That is n=1, and the peak memory was not captured. **Deferred fix (Prime: later):** shorten scriberr's slice to ~120 s in our local build, then SemIf can come back. Embedding cards were ruled out: esh-ml1 has ~4.4 GB free and nh3-ml1 ~5.1 GB. A replacement bench (brokkr's Jev candidates) is running on GPU 3 under a separate harness. - **OFFLINE since 2026-09-30 0135 PT (Prime: "take semif offline for now; we'll optimize scriberr later").** Stopped with `docker compose stop`, not removed, to give scriberr back its GPU 1 room. Scriberr's Parakeet path hardcodes `--chunk-len 300`, and the attention memory grows with the square of the slice, so it needs over 6 GB; it hit CUDA OOM at 0124 on a 35-min file with ~6.7 GB free. Stopping SemIf moved GPU 1 from 91,052 to 81,806 MiB used. The same job re-run at 0137 finished clean: 35m17s of audio in 44 s. That is n=1, and the peak memory was not captured. **Deferred fix (Prime: later):** shorten scriberr's slice to ~120 s in our local build, then SemIf can come back. Embedding cards were ruled out: esh-ml1 has ~4.4 GB free and nh3-ml1 ~5.1 GB. A replacement bench (brokkr's Jev candidates) is running on GPU 3 under a separate harness.
- **Scriberr pause-aware slicer: BUILD IN FLIGHT (Prime, 2026-09-30: "build the slicer").** A background agent owns it.
- The design: pause-aware cuts plus a small overlap with timestamp stitching, carried in `stacks/scriberr/patches/` and rebuilt by `scripts/scriberr-rebuild` (pinned sha, `git apply --check`, a distinct image tag, then a seam, memory and quality check).
- Constraints: the peak must stay ≤ 5,496 MiB (the GPU 1 budget beside intern-decision); the Go↔Python seam (CLI + JSON) is unchanged; it is judged on ≥3 recordings against a no-cut reference. Prime's transcripts stay private and off git.
- An upstream PR is PREPARED, not opened: that needs Prime's yes.
- Upstream cadence: 1 commit in 90 days, and the maintainer is restarting (a353078, 2026-09-20), so the patch burden is low.
- **intern-decision LIVE on fv-ml1 GPU 1 since 0941 2026-09-30, REPLACING SemIf (Prime: "replace semif with intern-decision now", with Scriberr fixed alongside).** - **intern-decision LIVE on fv-ml1 GPU 1 since 0941 2026-09-30, REPLACING SemIf (Prime: "replace semif with intern-decision now", with Scriberr fixed alongside).**
- Where: `http://intern-decision.fv.internal:8033`, image `intern-decision-serve:0.1.0`, token `intern-decision/api-token`. Code and contract are in `services/intern-decision-serve/`, the stack in `stacks/intern-decision`. - Where: `http://intern-decision.fv.internal:8033`, image `intern-decision-serve:0.1.0`, token `intern-decision/api-token`. Code and contract are in `services/intern-decision-serve/`, the stack in `stacks/intern-decision`.
- Surface: semif-compatible `/decide`, `/decide/shared`, `/health`. It has 12 documented deltas; the main one is that the questions in one call share a prompt, in calls of at most 16. - Surface: semif-compatible `/decide`, `/decide/shared`, `/health`. It has 12 documented deltas; the main one is that the questions in one call share a prompt, in calls of at most 16.