2026-07-25 08:27:22 +00:00
|
|
|
[run]
|
|
|
|
|
name = "airfrans-hf-smoke"
|
|
|
|
|
timeout_minutes = 30
|
|
|
|
|
local_artifact_dir = "artifacts/remote_runs"
|
|
|
|
|
max_attempts = 1
|
|
|
|
|
artifact_sync_interval_seconds = 300
|
|
|
|
|
|
|
|
|
|
[provider]
|
|
|
|
|
kind = "vastai"
|
|
|
|
|
disk_gb = 32
|
|
|
|
|
max_price_per_hour = 0.08
|
|
|
|
|
image = "vastai/base:0.0.2"
|
|
|
|
|
|
|
|
|
|
[provider.gpu]
|
|
|
|
|
name = "RTX 3060"
|
|
|
|
|
count = 1
|
|
|
|
|
min_vram_gb = 12
|
|
|
|
|
|
|
|
|
|
[selection]
|
|
|
|
|
min_reliability = 0.90
|
|
|
|
|
min_down_mbps = 50
|
|
|
|
|
min_up_mbps = 5
|
|
|
|
|
require_verified = true
|
|
|
|
|
blocked_geos = ["CN"]
|
|
|
|
|
blacklist_hosts = [59017, 187467, 155385]
|
|
|
|
|
drop_cheap_frac = 0.0
|
|
|
|
|
image_size_gb = 5.0
|
|
|
|
|
base_url = "https://cloud.vast.ai"
|
|
|
|
|
|
|
|
|
|
[workspace]
|
|
|
|
|
workdir = "."
|
|
|
|
|
exclude = [
|
|
|
|
|
"/artifacts",
|
|
|
|
|
"/data/raw",
|
|
|
|
|
"/.venv",
|
|
|
|
|
"/notebooks",
|
|
|
|
|
"HF_TOKEN",
|
|
|
|
|
"__pycache__",
|
|
|
|
|
"*.pyc",
|
|
|
|
|
]
|
|
|
|
|
|
|
|
|
|
[bootstrap]
|
|
|
|
|
command = """
|
|
|
|
|
uv sync --no-dev
|
|
|
|
|
uv run --no-dev python -c "import torch; print('torch_cuda_available=' + str(torch.cuda.is_available())); print(torch.cuda.get_device_name(0) if torch.cuda.is_available() else 'no_cuda_device')"
|
|
|
|
|
uv run --no-dev python -c "import huggingface_hub; print('huggingface_hub_ok')"
|
|
|
|
|
"""
|
|
|
|
|
|
|
|
|
|
[job]
|
|
|
|
|
command = """
|
|
|
|
|
uv run --no-dev remote-run hf-smoke --artifact-dir artifacts/current_run --run-id "$AIRFRANS_REMOTE_RUN_ID" --repo-id airfrans-hf-smoke
|
|
|
|
|
"""
|
|
|
|
|
artifact_dir = "artifacts/current_run"
|
|
|
|
|
heartbeat_file = "artifacts/current_run/heartbeat.json"
|
|
|
|
|
metrics_file = "artifacts/current_run/metrics.jsonl"
|
|
|
|
|
|
|
|
|
|
[artifacts]
|
|
|
|
|
mode = "object_store_upload"
|
|
|
|
|
required = [
|
|
|
|
|
"config.toml",
|
2026-07-28 18:04:50 +00:00
|
|
|
"job_manifest.json",
|
2026-07-25 08:27:22 +00:00
|
|
|
"metrics.jsonl",
|
|
|
|
|
"latest_metrics.json",
|
|
|
|
|
"heartbeat.json",
|
|
|
|
|
"checkpoint_latest.pt",
|
|
|
|
|
"checkpoint_best.pt",
|
|
|
|
|
"checkpoint_final.pt",
|
|
|
|
|
"final_metrics.json",
|
|
|
|
|
"run_manifest.json",
|
|
|
|
|
"environment_manifest.json",
|
|
|
|
|
"hf_upload_manifest.json",
|
|
|
|
|
"artifact_manifest.json",
|
|
|
|
|
"checksums.txt",
|
2026-07-28 18:04:50 +00:00
|
|
|
"utilization.jsonl",
|
2026-07-25 08:27:22 +00:00
|
|
|
]
|
|
|
|
|
|
|
|
|
|
[cleanup]
|
|
|
|
|
on_success = "sky_down"
|
|
|
|
|
on_failure = "collect_then_keep"
|
