Release: GPU agent autoscaler (throughput hill-climb, Auto default-on) #158

2026-06-30T18:23:10-04:00

bvandeusen commented

2026-06-30 18:23:10 -04:00

Agent-only. Auto mode (default on) finds the worker count via a throughput hill-climb + VRAM back-off — apt now that each job runs 3 detectors + several SigLIP embeds (GPU-bound). Manual dial still available; "Auto" toggle in the control UI. Ships in the agent image. CI green on dev (run 1751).

🤖 Generated with Claude Code

Agent-only. Auto mode (default on) finds the worker count via a throughput hill-climb + VRAM back-off — apt now that each job runs 3 detectors + several SigLIP embeds (GPU-bound). Manual dial still available; "Auto" toggle in the control UI. Ships in the agent image. CI green on dev (run 1751). 🤖 Generated with [Claude Code](https://claude.com/claude-code)

bvandeusen added 1 commit 2026-06-30 18:23:10 -04:00

feat(agent): autoscale the worker count (throughput hill-climb), Auto default-on

CI / lint (push) Successful in 3s

Details

CI / integration (push) Successful in 3m23s

Details

CI / frontend-build (push) Successful in 18s

Details

CI / backend-lint-and-test (push) Successful in 26s

Details

6d7b17b0b5

The new per-job workload (3 detectors + several SigLIP embeds) is far more
GPU-bound than the old I/O-bound CCIP pass, so the right worker count shifted and
is hard to guess. Add an Auto mode (default ON) that finds it:

- _control_loop samples jobs/sec + GPU util/VRAM every ~6s and hill-climbs the
  target: grow while throughput keeps improving and VRAM stays under budget,
  revert a step that doesn't help, back off under memory pressure (VRAM >= 90%),
  then settle and periodically re-probe (the GPU/IO balance shifts over a run).
- A manual concurrency set is an override → leaves Auto; an "Auto" toggle in the
  control UI re-enables it. status() reports `auto`; the dial reflects the
  auto-chosen count (read-only) while Auto is on.
- AUTO_SCALE env (default on) + compose doc. Agent py-compiled (outside CI).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa

bvandeusen merged commit bdfc17477c into main

2026-06-30 18:23:15 -04:00

bvandeusen referenced this issue from a commit

2026-07-13 13:25:18 -04:00

chore(deps): pinia 2→3 + vue-router 4→5 (batch 4 phase A, #1449)

bvandeusen referenced this issue from a commit

2026-07-13 13:35:03 -04:00

feat(deps): vuetify 3→4 (batch 4 phase B, #1449)

bvandeusen referenced this pull request

2026-07-13 13:40:15 -04:00

Ship: UI framework batch 4 — Vuetify 4 + vue-router 5 + pinia 3 (#1449) #224

Sign in to join this conversation.