File size: 7,701 Bytes
f996af6 | 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 | """Small CLI: validate, export-inputs, infer, parse, score, aggregate."""
from __future__ import annotations
import argparse
import json
import os
import sys
from pathlib import Path
from . import runner
from .dataset import Dataset
from .io import file_digest, indexed, read_jsonl, write_json, write_jsonl
from .providers import Service
def build_parser():
parser = argparse.ArgumentParser(description="ODU benchmark evaluation")
sub = parser.add_subparsers(dest="command", required=True)
def data(command, selection=False):
p = sub.add_parser(command)
p.add_argument("--data-root", type=Path, required=True)
p.add_argument("--subset", default="main")
if selection:
group = p.add_mutually_exclusive_group()
group.add_argument("--smoke", action="store_true", help="Four deterministic modality/label/clip strata")
group.add_argument("--limit-per-group", type=int, help="First N per modality/label group in manifest order")
return p
p = data("validate")
p.add_argument("--checksums", action="store_true")
for cmd in ["export-inputs", "infer"]:
p = data(cmd, True)
p.add_argument("--prompt-lang", choices=["en"], default="en")
p.add_argument("--wrapped-output", action="store_true")
if cmd == "export-inputs":
p.add_argument("--output", type=Path, required=True)
else:
p.add_argument("--model", choices=["gemini-3.1-pro", "qwen3.5-omni-plus", "openai-compatible"], required=True)
p.add_argument("--output-dir", type=Path, required=True)
p.add_argument("--workers", type=int, default=2)
p.add_argument("--overwrite", action="store_true")
p.add_argument("--no-repair", action="store_true", help="Explicit protocol variation: disable the reference API JSON repair")
p = sub.add_parser("parse")
p.add_argument("--input", type=Path, required=True)
p.add_argument("--output", type=Path, required=True)
p.add_argument("--profile", choices=["api", "local"], required=True)
p.add_argument("--inputs", type=Path, help="Exported inputs whose metadata describes this submission")
p.add_argument("--model-tag", default="custom-model")
group = p.add_mutually_exclusive_group()
group.add_argument("--repair", action="store_true")
group.add_argument("--no-repair", action="store_true")
p.add_argument("--offline", action="store_true")
p.add_argument("--repair-cache", type=Path)
p = data("score", True)
p.add_argument("--predictions", type=Path, required=True)
p.add_argument("--judge", choices=["qwen3.6-flash", "gpt-5.4"], required=True)
p.add_argument("--thinking", choices=["true", "false", "omit"], default=None, help="Qwen enable_thinking; default true")
p.add_argument("--reasoning-effort", choices=["omit", "none", "minimal", "low", "medium", "high", "xhigh"], default="omit")
p.add_argument("--output-dir", type=Path, required=True)
p.add_argument("--judge-cache", type=Path)
p.add_argument("--offline", action="store_true")
p.add_argument("--workers", type=int, default=2)
p.add_argument("--overwrite", action="store_true")
p.add_argument("--t1-report", action="store_true")
p = sub.add_parser("aggregate")
p.add_argument("--runs", type=Path, nargs="+", required=True)
p.add_argument("--output-dir", type=Path, required=True)
p.add_argument("--t1-report", action="store_true")
return parser
def _run(args):
if getattr(args, "workers", 1) < 1:
raise ValueError("--workers must be positive")
if os.environ.get("ODU_M2_DENOM", "union") != "union":
raise ValueError("This release uses union M2. Unset ODU_M2_DENOM; use --t1-report for the alternate table")
if args.command == "parse":
rows = read_jsonl(args.input)
indexed(rows)
meta = runner.read_meta(args.inputs or args.input)
if args.inputs:
allowed = set(meta.get("scene_ids", []))
if not allowed or set(indexed(rows)) - allowed:
raise ValueError("Submission IDs must belong to the exported inputs")
repair = args.repair or (args.profile == "api" and not args.no_repair)
results, info = runner.parse_rows(rows, args.profile, repair=repair,
cache_dir=args.repair_cache or args.output.parent / "repair_cache", offline=args.offline)
write_jsonl(args.output, results)
write_json(runner.meta_path(args.output), {**meta, "model": args.model_tag, "parser_profile": args.profile,
"predictions_hash": file_digest(args.output), "repair": info})
summary = {"expected": len(results), "ok": sum(r["status"] == "ok" for r in results)}
status = 0 if summary["ok"] == summary["expected"] else 1
elif args.command == "aggregate":
summary = runner.aggregate_runs(args.runs, args.output_dir, args.t1_report)
status = 0
else:
dataset = Dataset(args.data_root, args.subset)
if args.command == "validate":
summary, status = dataset.validate(args.checksums), 0
else:
rows = dataset.select(smoke=args.smoke, limit_per_group=args.limit_per_group)
selection = "smoke" if args.smoke else (f"limit-per-group-{args.limit_per_group}" if args.limit_per_group else "full")
if args.command == "export-inputs":
summary = runner.export_inputs(dataset, rows, args.output, language=args.prompt_lang,
flat=not args.wrapped_output, selection=selection)
status = 0
elif args.command == "infer":
summary = runner.infer(dataset, rows, Service(args.model), args.output_dir, workers=args.workers,
language=args.prompt_lang, flat=not args.wrapped_output,
selection=selection, overwrite=args.overwrite, repair=not args.no_repair)
status = 0 if summary["ok"] == summary["expected"] else 1
else:
if args.judge == "gpt-5.4" and args.thinking is not None:
raise ValueError("Use --reasoning-effort for OpenAI")
if args.judge == "qwen3.6-flash" and args.reasoning_effort != "omit":
raise ValueError("Use --thinking for Qwen")
thinking = (args.thinking or "true") if args.judge == "qwen3.6-flash" else "omit"
service = Service(args.judge, thinking=thinking, reasoning=args.reasoning_effort)
summary = runner.score(dataset, rows if args.smoke or args.limit_per_group else None,
args.predictions, service, args.output_dir, workers=args.workers,
offline=args.offline, judge_cache=args.judge_cache,
overwrite=args.overwrite, t1_report=args.t1_report)
status = 0 if summary["scored"] == summary["expected"] else 1
print(json.dumps(summary, ensure_ascii=False, indent=2))
return status
def main(argv=None):
args = build_parser().parse_args(argv)
try:
return _run(args)
except (ValueError, RuntimeError) as exc:
print(f"error: {exc}", file=sys.stderr)
return 2
except (OSError, KeyError, TypeError) as exc:
# OS/SDK exceptions can include local paths. Expose the category only.
print(f"error: {type(exc).__name__}; check input files and the documented schema", file=sys.stderr)
return 2
if __name__ == "__main__":
raise SystemExit(main())
|