File size: 5,192 Bytes
2559788 | 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 | #!/usr/bin/env python3
"""Write FINAL_STATUS and a training report from artifacts after jobs finish."""
from __future__ import annotations
import json
from datetime import datetime, timezone
from pathlib import Path
ROOT = Path(__file__).resolve().parents[2]
def load(path: Path):
if not path.exists():
return None
return json.loads(path.read_text(encoding="utf-8"))
def status_of(block, pass_key="status"):
if not block:
return "NOT_RUN"
if isinstance(block, str):
return block
return block.get(pass_key) or block.get("matrix_status") or "UNKNOWN"
def main() -> int:
lattice = load(ROOT / "artifacts" / "training" / "lattice_summary.json") or {}
obs = load(ROOT / "artifacts" / "training" / "observational_and_vision_baselines.json") or {}
trans = load(ROOT / "artifacts" / "training" / "transition_sft.json") or {}
dec = load(ROOT / "artifacts" / "training" / "decision_sft.json") or {}
rlvr = load(ROOT / "artifacts" / "training" / "rlvr.json") or {}
prev = load(ROOT / "artifacts" / "FINAL_STATUS.json") or {}
matrix = dict(prev.get("matrix") or {})
matrix["Classical baseline"] = "PASS" if obs.get("B0") and obs.get("B1") else matrix.get("Classical baseline", "NOT_RUN")
b2 = obs.get("B2_generated_pattern_cnn") or {}
b3 = obs.get("B3_generated_pattern_multimodal") or {}
matrix["Vision baseline"] = "PASS" if b2.get("status") == "PASS" else ("NOT_RUN" if not b2 else b2.get("status", "FAIL"))
matrix["Multimodal baseline"] = "PASS" if b3.get("status") == "PASS" else ("NOT_RUN" if not b3 else b3.get("status", "FAIL"))
matrix["Transition SFT"] = "PASS" if (trans.get("mlp") or {}).get("status") == "PASS" else "NOT_RUN"
matrix["Decision SFT"] = "PASS" if (dec.get("mlp") or {}).get("status") == "PASS" else "NOT_RUN"
if rlvr.get("matrix_status") == "COMPLETED_NOT_JUSTIFIED":
matrix["RLVR"] = "COMPLETED_NOT_JUSTIFIED"
elif rlvr.get("status") == "PASS":
matrix["RLVR"] = rlvr.get("matrix_status") or "PASS"
else:
matrix["RLVR"] = rlvr.get("status") or "NOT_RUN"
notes = dict(prev.get("notes") or {})
notes["SFT/RLVR"] = (
f"Measured lattice n_transitions={lattice.get('n_transitions')} "
f"n_decision={lattice.get('n_decision_cases')} "
f"transition_mae={(trans.get('mlp') or {}).get('test_realized_mae_cm')} "
f"decision_acc={(dec.get('mlp') or {}).get('test_action_accuracy')} "
f"rlvr={matrix['RLVR']}. "
"Vision B2/B3 used generated pattern PNGs, not FIT-100K."
)
payload = {
"generated_at": datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ"),
"host": prev.get("host", "daka5svhri0c73etu4ug-fitground"),
"gpu": prev.get("gpu", "NVIDIA GeForce RTX 4090 24GB"),
"matrix": matrix,
"notes": notes,
"training": {
"lattice": lattice,
"B0_test_mae_cm": (obs.get("B0") or {}).get("test_mae_cm"),
"B1_test_mae_cm": (obs.get("B1") or {}).get("test_mae_cm"),
"B1_xgboost_test_mae_cm": (obs.get("B1_xgboost") or {}).get("test_mae_cm"),
"B2_cnn_test_mae_cm": b2.get("test_mae_cm"),
"B3_cnn_test_mae_cm": b3.get("test_mae_cm"),
"transition_sft_realized_mae_cm": (trans.get("mlp") or {}).get("test_realized_mae_cm"),
"decision_sft_accuracy": (dec.get("mlp") or {}).get("test_action_accuracy"),
"rlvr": {
"status": rlvr.get("matrix_status") or rlvr.get("status"),
"before_test": rlvr.get("before_test"),
"after_test": rlvr.get("after_test"),
},
},
}
(ROOT / "artifacts" / "FINAL_STATUS.json").write_text(json.dumps(payload, indent=2), encoding="utf-8")
md = ROOT / "reports" / "TRAINING_RUN.md"
md.write_text(
"\n".join(
[
"# FitGround training run",
"",
"This run trains B0–B3, Transition SFT, Decision SFT, and RLVR on **measured** GarmentCode panel geometry.",
"FIT-100K images were not downloaded. Vision models use rasterized pattern drawings.",
"No `intended_delta_cm` was copied into `realized_delta_cm`.",
"",
"```json",
json.dumps(payload["training"], indent=2, default=str),
"```",
"",
"## Matrix",
"",
*[f"- {k}: {v}" for k, v in matrix.items()],
"",
"## Honesty",
"",
"- Observational B0/B1 remain FIT-Clean measurement prediction, not intervention GT.",
"- B2/B3 FIT-100K: NOT_RUN. B2/B3 generated-pattern: trained.",
"- Tiny LMs are from-scratch character transformers, not pretrained MLLMs.",
"- RLVR reward is cached measured utility from the lattice, not a live Warp loop.",
"",
]
),
encoding="utf-8",
)
print(json.dumps(payload, indent=2, default=str)[:4000])
return 0
if __name__ == "__main__":
raise SystemExit(main())
|