airfRANS-model-exploration/configs/remote_smoke.toml

[run]
name = "airfrans-aggressive-smoke"
timeout_minutes = 360
local_artifact_dir = "artifacts/remote_runs"
max_attempts = 2

[provider]
kind = "vastai"
disk_gb = 128
max_price_per_hour = 0.80
image = "vastai/base:0.0.2"

[provider.gpu]
name = "RTX 4090"
count = 1
min_vram_gb = 20

[selection]
min_reliability = 0.95
min_down_mbps = 100
min_up_mbps = 25
require_verified = true
blocked_geos = ["CN"]
blacklist_hosts = [59017]
drop_cheap_frac = 0.30
image_size_gb = 5.0
base_url = "https://cloud.vast.ai"

[workspace]
workdir = "."
exclude = [
  "/artifacts",
  "/data/raw",
  "/.venv",
  "/notebooks",
  "__pycache__",
  "*.pyc",
]

[bootstrap]
command = """
uv sync --no-dev
uv run --no-dev python -c "import torch; assert torch.cuda.is_available(); print(torch.cuda.get_device_name(0))"
"""

[data]
validation_command = """
uv run --no-dev python -c "from pathlib import Path; files=sorted(Path('data/processed/full').glob('*.npz')); assert len(files) >= 50; print(f'processed_full_cases={len(files)}')"
"""

[job]
command = """
uv run --no-dev remote-run smoke-train configs/aggressive_smoke.toml --artifact-dir artifacts/current_run --run-id "$AIRFRANS_REMOTE_RUN_ID"
"""
artifact_dir = "artifacts/current_run"
heartbeat_file = "artifacts/current_run/heartbeat.json"
metrics_file = "artifacts/current_run/metrics.jsonl"

[artifacts]
mode = "rsync"
required = [
  "config.toml",
  "metrics.jsonl",
  "latest_metrics.json",
  "heartbeat.json",
  "checkpoint_latest.pt",
  "checkpoint_best.pt",
  "checkpoint_final.pt",
  "final_metrics.json",
  "split_manifest.json",
  "data_manifest.json",
  "normalization.json",
  "run_manifest.json",
  "environment_manifest.json",
  "artifact_manifest.json",
  "checksums.txt",
]

[cleanup]
on_success = "sky_down"
on_failure = "collect_then_keep"