feat(ml): lease announces detector config; agent builds proposers from it live (#134 step 2)
CI / lint (push) Successful in 3s
CI / frontend-build (push) Successful in 19s
CI / backend-lint-and-test (push) Successful in 35s
CI / integration (push) Successful in 3m34s

The GPU lease now carries the crop-proposer config from MLSettings in a per-job
'detectors' block (same pattern as embed_model_name). The agent's worker builds
its Proposers from the announced config via _effective_cfg (lease block overlaid
on env) + _proposers_for (rebuilds only when a config signature changes) — so an
operator's UI edit takes effect on the next lease with NO restart, and env is now
just the bootstrap fallback until the server announces. enabled-off maps to empty
weights (proposer skipped); dedupe_iou + max_regions also come from the effective
cfg. Test: lease announces the detectors block with the seeded default weights.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CDgx8bQS5YrGRK76v8HUnM
This commit is contained in:
2026-07-05 19:42:59 -04:00
parent 62ec70b9e4
commit a4df279343
3 changed files with 102 additions and 9 deletions
+50 -9
View File
@@ -265,6 +265,8 @@ class Worker:
self._embedder = None self._embedder = None
self._embedder_lock = threading.Lock() self._embedder_lock = threading.Lock()
self._proposers = None self._proposers = None
self._proposers_sig = None # detector-config signature the current
# proposers were built for (#134)
self._proposers_lock = threading.Lock() self._proposers_lock = threading.Lock()
# --- held-lease bookkeeping -------------------------------------------- # --- held-lease bookkeeping --------------------------------------------
@@ -796,14 +798,52 @@ class Worker:
self._embedder = CropEmbedder(model_name, self.cfg.embed_dtype) self._embedder = CropEmbedder(model_name, self.cfg.embed_dtype)
return self._embedder return self._embedder
def _ensure_proposers(self): def _effective_cfg(self, det: dict | None):
if self._proposers is not None: """The detector config for a job: the LEASE-ANNOUNCED block (the DB/UI
return self._proposers source of truth, #134) overlaid on the env cfg, or the env cfg unchanged
when the server announced nothing (bootstrap / older backend). An enabled
toggle off maps to empty weights — the 'weights == "" → proposer skipped'
contract in detectors.py. detector_level + ccip_model stay env (separate
settings, not part of the crop-proposer config)."""
if not det:
return self.cfg
def _w(p: str) -> str:
slot = det.get(p) or {}
return (slot.get("weights") or "") if slot.get("enabled", True) else ""
from dataclasses import replace
return replace(
self.cfg,
person_weights=_w("person"),
person_conf=float(det["person"]["conf"]),
anatomy_weights=_w("anatomy"),
anatomy_conf=float(det["anatomy"]["conf"]),
panel_weights=_w("panel"),
panel_conf=float(det["panel"]["conf"]),
max_figures=int(det["max_figures"]),
max_components=int(det["max_components"]),
max_panels=int(det["max_panels"]),
max_regions=int(det["max_regions"]),
dedupe_iou=float(det["dedupe_iou"]),
)
def _proposers_for(self, eff):
"""Proposers built from the effective detector config, rebuilt ONLY when
that config changes — so an operator's UI edit takes effect on the next
lease with no restart, and steady-state reuses the loaded YOLO models."""
sig = (
eff.person_weights, eff.person_conf,
eff.anatomy_weights, eff.anatomy_conf,
eff.panel_weights, eff.panel_conf,
eff.max_figures, eff.max_components, eff.max_panels,
)
with self._proposers_lock: with self._proposers_lock:
if self._proposers is None: if self._proposers is None or self._proposers_sig != sig:
from .detectors import Proposers from .detectors import Proposers
self._proposers = Proposers(self.cfg) self._proposers = Proposers(eff)
return self._proposers self._proposers_sig = sig
return self._proposers
def _consume(self, job: dict, frames: list, stop_evt: threading.Event) -> bool: def _consume(self, job: dict, frames: list, stop_evt: threading.Event) -> bool:
"""Detect + embed the decoded frames and submit the result. Returns True """Detect + embed the decoded frames and submit the result. Returns True
@@ -856,7 +896,8 @@ class Worker:
else ["figure", "face", "concept", "panel"] else ["figure", "face", "concept", "panel"]
) )
embedder = self._ensure_embedder(model_name) if want_siglip else None embedder = self._ensure_embedder(model_name) if want_siglip else None
proposers = self._ensure_proposers() eff = self._effective_cfg(job.get("detectors"))
proposers = self._proposers_for(eff)
regions = [] regions = []
ccip_ev = self.cfg.ccip_model or "ccip-default" ccip_ev = self.cfg.ccip_model or "ccip-default"
@@ -922,7 +963,7 @@ class Worker:
# we never SigLIP the same region twice — saves GPU and a # we never SigLIP the same region twice — saves GPU and a
# slot against max_regions. High-IoU + kind-aware, so # slot against max_regions. High-IoU + kind-aware, so
# intentional nested crops (figure ⊃ head) survive. # intentional nested crops (figure ⊃ head) survive.
pending = dedupe_crops(pending, self.cfg.dedupe_iou) pending = dedupe_crops(pending, eff.dedupe_iou)
vecs = embedder.embed_batch([c for c, _ in pending]) vecs = embedder.embed_batch([c for c, _ in pending])
for (_c, tmpl), vec in zip(pending, vecs, strict=True): for (_c, tmpl), vec in zip(pending, vecs, strict=True):
tmpl["siglip_embedding"] = vec tmpl["siglip_embedding"] = vec
@@ -932,7 +973,7 @@ class Worker:
# mp4, 64 sampled frames × ~32 regions) would otherwise burn # mp4, 64 sampled frames × ~32 regions) would otherwise burn
# ~38s of GPU across every frame before the submit is even # ~38s of GPU across every frame before the submit is even
# truncated. Bounds the WORK, not just the POST body. # truncated. Bounds the WORK, not just the POST body.
if len(regions) >= self.cfg.max_regions: if len(regions) >= eff.max_regions:
break break
# A Stop mid-frame-loop leaves partial regions — don't submit those; # A Stop mid-frame-loop leaves partial regions — don't submit those;
+28
View File
@@ -269,6 +269,33 @@ async def lease():
).scalars() ).scalars()
} if ids else {} } if ids else {}
await session.commit() await session.commit()
# Crop-proposer config, announced FROM THE SETTING like embed_model_name
# (#134): the agent builds its detectors from this, rebuilding live when
# it changes — so tuning is a DB/UI edit, never an agent restart. Same
# block for every job in the batch (it's global), built once. An enabled
# toggle off is carried through so the agent skips that proposer.
detectors = {
"person": {
"enabled": ml.detector_person_enabled,
"weights": ml.detector_person_weights,
"conf": ml.detector_person_conf,
},
"anatomy": {
"enabled": ml.detector_anatomy_enabled,
"weights": ml.detector_anatomy_weights,
"conf": ml.detector_anatomy_conf,
},
"panel": {
"enabled": ml.detector_panel_enabled,
"weights": ml.detector_panel_weights,
"conf": ml.detector_panel_conf,
},
"max_figures": ml.detector_max_figures,
"max_components": ml.detector_max_components,
"max_panels": ml.detector_max_panels,
"max_regions": ml.detector_max_regions,
"dedupe_iou": ml.detector_dedupe_iou,
}
out = [] out = []
for j in jobs: for j in jobs:
img = imgs.get(j.image_record_id) img = imgs.get(j.image_record_id)
@@ -290,6 +317,7 @@ async def lease():
# re-embed, never an agent change. # re-embed, never an agent change.
"embed_model_name": ml.embedder_model_name, "embed_model_name": ml.embedder_model_name,
"embed_version": ml.embedder_model_version, "embed_version": ml.embedder_model_version,
"detectors": detectors,
}) })
return jsonify({"jobs": out}) return jsonify({"jobs": out})
+24
View File
@@ -105,6 +105,30 @@ async def test_lease_announces_embed_model_then_submit_embedding(client, db):
assert img.siglip_embedding is not None and len(list(img.siglip_embedding)) == 1152 assert img.siglip_embedding is not None and len(list(img.siglip_embedding)) == 1152
@pytest.mark.asyncio
async def test_lease_announces_detector_config(client, db):
# #134: the lease carries the crop-proposer config from MLSettings, so the
# agent builds its detectors from the DB/UI (no restart). Defaults are all-on
# with the pinned working weights.
img = await _img(db, "c" * 64)
await GpuJobService(db).enqueue(img.id, "embed")
await db.commit()
token = (await (await client.post("/api/gpu/token/rotate")).get_json())["token"]
hdr = {"Authorization": f"Bearer {token}"}
leased = await client.post(
"/api/gpu/jobs/lease", json={"agent_id": "a1", "batch_size": 5}, headers=hdr,
)
det = (await leased.get_json())["jobs"][0]["detectors"]
assert det["person"]["enabled"]
assert det["anatomy"]["enabled"]
assert det["panel"]["enabled"]
assert "yolov11m_aa22" in det["anatomy"]["weights"] # booru_yolo default
assert det["panel"]["weights"].endswith("::best.pt") # mosesb default
assert det["max_regions"] == 128
assert det["dedupe_iou"] == 0.85
@pytest.mark.asyncio @pytest.mark.asyncio
async def test_submit_with_stale_lease_is_409(client, db): async def test_submit_with_stale_lease_is_409(client, db):
img = await _img(db, "b" * 64) img = await _img(db, "b" * 64)