File size: 5,725 Bytes
e0265b9 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 | from __future__ import annotations
from pathlib import Path
import logging
from types import SimpleNamespace
from PIL import Image
from adam.atlas import AtlasSupervisor
from adam.job_manager import JobManager
from adam.models import ExecutionPlan, Job, JobStatus, PlanStep, SystemSnapshot
from adam.nova import evaluate_job_output
from adam.orion import apply_orion_review, recommend_training_settings
from adam.ui.main_window import CommandCenterPage
def _training_job(tmp_path: Path) -> Job:
return Job(
plan=ExecutionPlan(
request="train",
summary="Train.",
steps=[PlanStep("ddpm_trainer", "Train DDPM", "Train", {
"dataset_dir": str(tmp_path), "epochs": 10, "batch_size": 2,
"resolution": 128,
})],
),
status=JobStatus.RUNNING,
current_step=0,
progress=10,
)
def test_orion_flags_small_dataset_preset_applied_to_large_dataset(tmp_path: Path) -> None:
dataset = tmp_path / "dataset"
dataset.mkdir()
for index in range(1_000):
(dataset / f"{index}.png").touch()
plan = ExecutionPlan(
request="train",
summary="Train a model.",
steps=[PlanStep("ddpm_trainer", "Train DDPM", "Train", {
"dataset_dir": str(dataset), "epochs": 600, "batch_size": 1,
"resolution": 128,
})],
)
report = apply_orion_review(plan)
assert report["level"] == "warning"
assert report["settings_changed"] is False
assert plan.requires_confirmation is True
assert "ORION" in plan.summary
assert "600,000 image exposures" in plan.summary
def test_orion_uses_planned_collection_size_before_dataset_exists(tmp_path: Path) -> None:
dataset = tmp_path / "future-dataset"
plan = ExecutionPlan(
request="collect and train",
summary="Collect and train.",
steps=[
PlanStep("dataset_collector", "Collect", "Collect", {
"output_dir": str(dataset), "image_count": 2_000,
}),
PlanStep("ddpm_trainer", "Train", "Train", {
"dataset_dir": str(dataset), "epochs": 600, "batch_size": 1,
"resolution": 128,
}),
],
)
report = apply_orion_review(plan)
assert report["level"] == "warning"
assert "1,200,000 image exposures" in plan.summary
def test_orion_warning_blocks_trusted_automation() -> None:
class Config:
def get(self, key: str, default=None):
return key == "trusted_dataset_ddpm_automation"
page = SimpleNamespace(config=Config())
plan = ExecutionPlan(
request="train", summary="Train.", requires_confirmation=True,
steps=[PlanStep("dataset_collector", "Collect", "Collect"), PlanStep("ddpm_trainer", "Train", "Train")],
orion_review={"level": "warning"},
)
assert CommandCenterPage._can_trusted_start(page, plan) is False
def test_orion_recipe_scales_epochs_and_batch_to_dataset_size_and_resolution() -> None:
recipe = recommend_training_settings("ddpm", 1_000, 128, vram_gb=12)
assert recipe["epochs"] == 180
assert recipe["settings"]["batch_size"] == 12
assert recipe["settings"]["preview_every"] == 18
assert "180,000 image exposures" in recipe["summary"]
def test_orion_recipe_becomes_more_conservative_at_high_resolution() -> None:
recipe = recommend_training_settings("flow", 300, 512, vram_gb=6)
assert recipe["settings"]["batch_size"] == 1
assert recipe["settings"]["gradient_checkpointing"] is True
assert recipe["settings"]["workers"] == recipe["settings"]["dataloader_num_workers"]
def test_atlas_pauses_for_a_new_non_finite_loss_message(tmp_path: Path) -> None:
job = _training_job(tmp_path)
job.logs.append("loss = NaN")
atlas = AtlasSupervisor()
decision = atlas.observe(job, SystemSnapshot())
assert decision.severity == "critical"
assert decision.action == "pause"
def test_atlas_requires_repeated_critical_temperature_samples(tmp_path: Path) -> None:
job = _training_job(tmp_path)
atlas = AtlasSupervisor()
hot = SystemSnapshot(gpu_temperature=91)
assert atlas.observe(job, hot).action == "none"
assert atlas.observe(job, hot).action == "none"
assert atlas.observe(job, hot).action == "pause"
def test_job_manager_applies_an_atlas_critical_pause(tmp_path: Path) -> None:
class Worker:
paused = False
def pause(self) -> None:
self.paused = True
manager = JobManager(tmp_path, None, logging.getLogger("test.atlas")) # type: ignore[arg-type]
job = _training_job(tmp_path)
job.logs.append("loss: inf")
worker = Worker()
manager.jobs = [job]
manager._active_job = job
manager._worker = worker # type: ignore[assignment]
manager.supervise(SystemSnapshot())
assert worker.paused is True
assert job.status == JobStatus.PAUSED
assert job.atlas_report["severity"] == "critical"
def test_nova_reports_duplicate_sample_collapse(tmp_path: Path) -> None:
output = tmp_path / "output"
output.mkdir()
for index in range(4):
Image.new("RGB", (32, 32), (80, 90, 100)).save(output / f"sample_{index}.png")
job = _training_job(tmp_path)
job.output_folder = str(output)
report = evaluate_job_output(job)
assert report["status"] == "NEEDS REVIEW"
assert report["sample_count"] == 4
assert report["duplicate_count"] == 3
def test_nova_requests_samples_when_training_has_no_generations(tmp_path: Path) -> None:
job = _training_job(tmp_path)
report = evaluate_job_output(job)
assert report["status"] == "NEEDS SAMPLES"
assert report["sample_count"] == 0
|