bvandeusen/FabledCurator
1
0
Fork 0
Code Issues 1 Pull Requests Actions Packages Projects Releases 35 Wiki Activity

feat(agent): idle-unload GPU models to free VRAM when the queue is idle
CI / lint (push) Successful in 3s
Details
CI / frontend-build (push) Successful in 22s
Details
CI / backend-lint-and-test (push) Successful in 38s
Details
CI / integration (push) Successful in 4m3s
Details

Browse Source 
The SigLIP embedder + YOLO proposers load lazily then stay resident for the
container's whole lifetime — a 24/7 agent with an empty queue squats on ~5GB of
VRAM doing nothing (operator-observed: 4900MiB held at GPU-util 8% / P8). Sleep
mode only sheds downloaders + poll cadence; even a UI Stop left the models loaded.

Add a monitor thread that unloads the torch-owned models after
cfg.idle_unload_seconds (env IDLE_UNLOAD_SECONDS, default 300; 0 disables) with
the GPU genuinely idle (active==0, buffer drained, no job completed in the
window), then torch.cuda.empty_cache() to hand the blocks back to the driver.
They reload lazily on the next job via the existing _ensure_embedder /
_proposers_for. Covers both sleep-mode idle and a full Stop. Surfaced in
/status (models_loaded) and the agent UI pipe line; the VRAM meter drops too.

Residual: imgutils CCIP/person ONNX sessions + the CUDA context stay resident
(no clean unload API) — idle VRAM drops substantially, not to zero.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TbrA36zNczjVhrM6cWThQa
This commit is contained in:
Bryan Van Deusen
2026-07-17 12:57:31 -04:00
parent ec66ea5f83
commit 57e52433d0
5 changed files with 119 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
agent/fc_agent/app.py
+4 -1
View File
@@ -21,7 +21,7 @@ log = logging.getLogger("fc_agent.app")
# Bump on every agent change. The page embeds this and /status reports it; the UI
# warns to reload when they differ — so a stale browser-cached page can't be
# mistaken for "the new image didn't deploy". (Belt-and-braces with no-store.)
VERSION = "2026-07-02.6 · sleep mode: an empty queue sheds to one downloader and backs the lease poll off to 15 min"
VERSION = "2026-07-17.1 · idle model-unload: after ~5 min idle the GPU models release their VRAM and reload on the next job (env IDLE_UNLOAD_SECONDS, 0=off) · sleep mode sheds to one downloader"
logbuf.install()
cfg = Config.from_env()
@@ -334,9 +334,12 @@ _PAGE = """<!doctype html><html><head><meta charset=utf-8>
waited.textContent=s.transient||0
// Instantaneous pool state → demoted to the sub-line, where its jumpiness reads
// as live churn rather than a "broken" headline metric.
// '=== false' (not falsy) so a stale page that doesn't send models_loaded shows
// nothing; when the idle monitor unloads, the VRAM meter drops alongside this.
pipe.textContent='downloaders '+(s.downloaders!=null?s.downloaders:'—')+' · consumers '+(s.consumers!=null?s.consumers:'—')+' · on GPU '+(s.active||0)
+' · net '+(s.net_mb_s!=null?s.net_mb_s.toFixed(1):'—')+' MB/s'
+(s.bandwidth_limit_mb_s>0?(' / cap '+s.bandwidth_limit_mb_s):'')
+(s.models_loaded===false?' · GPU models unloaded (idle — reload on next job)':'')
if(document.activeElement!==bw && s.bandwidth_limit_mb_s!=null) bw.value=s.bandwidth_limit_mb_s
// Buffer occupancy bar (also driven here so it tracks the /status cadence).
if(s.buffer!=null && s.buffer_max){ const p=Math.round(100*s.buffer/s.buffer_max)