File size: 7,908 Bytes
6d1bbc7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
#!/usr/bin/env python3
"""Run LLM-as-Judge evaluation for L3 reasoning predictions.

Reads L3 predictions from results/llm/l3_*/predictions.jsonl,
sends each response through a judge model with L3_JUDGE_PROMPT,
and computes aggregate scores.

Usage:
  python scripts/run_l3_judge.py --judge-provider gemini --judge-model gemini-2.5-flash
  python scripts/run_l3_judge.py --run-dir results/llm/l3_llama70b_3-shot_fs0 \
      --judge-provider openai --judge-model gpt-4o-mini

Output per run:
  results/llm/{run}_judged/
      judge_scores.jsonl   — per-item judge scores
      results.json         — aggregated L3 metrics
      judge_meta.json      — judge model info, timestamp
"""

import argparse
import json
import time
from datetime import datetime, timezone
from pathlib import Path

from negbiodb.llm_client import LLMClient
from negbiodb.llm_eval import L3_JUDGE_PROMPT, evaluate_l3, parse_l3_judge_scores

PROJECT_ROOT = Path(__file__).resolve().parent.parent
RESULTS_DIR = PROJECT_ROOT / "results" / "llm"
DATA_DIR = PROJECT_ROOT / "exports" / "llm_benchmarks"
L3_GOLD_FILE = DATA_DIR / "l3_reasoning_pilot.jsonl"


def load_gold_records() -> list[dict]:
    """Load L3 gold records."""
    records = []
    with open(L3_GOLD_FILE) as f:
        for line in f:
            records.append(json.loads(line))
    return records


def load_predictions(pred_path: Path) -> list[dict]:
    """Load predictions from JSONL."""
    preds = []
    with open(pred_path) as f:
        for line in f:
            preds.append(json.loads(line))
    return preds


def find_l3_runs(results_dir: Path) -> list[Path]:
    """Find all L3 run directories."""
    runs = []
    for d in sorted(results_dir.iterdir()):
        if d.is_dir() and d.name.startswith("l3_"):
            pred_file = d / "predictions.jsonl"
            if pred_file.exists():
                runs.append(d)
    return runs


def judge_run(
    run_dir: Path,
    gold_records: list[dict],
    client: LLMClient,
    judge_model: str,
) -> dict:
    """Judge all predictions in a run directory.

    Returns aggregated L3 metrics.
    """
    # Build gold lookup by question_id
    gold_by_id = {}
    for rec in gold_records:
        gold_by_id[rec["question_id"]] = rec

    # Load predictions
    pred_path = run_dir / "predictions.jsonl"
    predictions = load_predictions(pred_path)

    # Output dir
    judged_dir = run_dir.parent / f"{run_dir.name}_judged"
    judged_dir.mkdir(parents=True, exist_ok=True)

    # Resume: load existing judge scores
    scores_path = judged_dir / "judge_scores.jsonl"
    completed = {}
    if scores_path.exists():
        with open(scores_path) as f:
            for line in f:
                rec = json.loads(line)
                # Only count entries with valid scores as completed
                if rec.get("scores") is not None:
                    completed[rec["question_id"]] = rec
        print(f"  Resume: {len(completed)} already judged (with valid scores)")

    # Judge remaining predictions
    remaining = [p for p in predictions if p["question_id"] not in completed]
    print(f"  Judging {len(remaining)} remaining of {len(predictions)} total")

    start_time = time.time()
    with open(scores_path, "a") as f:
        for i, pred_rec in enumerate(remaining):
            qid = pred_rec["question_id"]
            gold = gold_by_id.get(qid)
            if gold is None:
                print(f"  Warning: no gold record for {qid}, skipping")
                continue

            # Format judge prompt
            prompt = L3_JUDGE_PROMPT.format(
                compound_name=gold.get("compound_name", "Unknown"),
                target_gene=gold.get("target_gene") or "Unknown",
                target_uniprot=gold.get("target_uniprot", "Unknown"),
                response=pred_rec["prediction"],
            )

            try:
                judge_response = client.generate(prompt)
            except Exception as e:
                print(f"  Error judging {qid}: {e}")
                judge_response = f"ERROR: {e}"

            # Parse scores
            scores = parse_l3_judge_scores(judge_response)

            score_rec = {
                "question_id": qid,
                "judge_response": judge_response,
                "scores": scores,
            }
            f.write(json.dumps(score_rec, ensure_ascii=False) + "\n")
            f.flush()
            completed[qid] = score_rec

            if (i + 1) % 10 == 0:
                elapsed = time.time() - start_time
                rate = (i + 1) / elapsed * 60
                print(f"  Progress: {i + 1}/{len(remaining)} ({rate:.1f}/min)")

    elapsed = time.time() - start_time
    print(f"  Judging complete: {elapsed:.0f}s")

    # Aggregate scores
    all_scores = []
    for pred_rec in predictions:
        qid = pred_rec["question_id"]
        if qid in completed and completed[qid].get("scores"):
            all_scores.append(completed[qid]["scores"])
        else:
            all_scores.append(None)

    metrics = evaluate_l3(all_scores)

    # Save results
    results_path = judged_dir / "results.json"
    with open(results_path, "w") as f:
        json.dump(metrics, f, indent=2)

    # Save metadata
    meta = {
        "judge_model": judge_model,
        "source_run": run_dir.name,
        "n_predictions": len(predictions),
        "n_judged": sum(1 for s in all_scores if s is not None),
        "elapsed_seconds": elapsed,
        "timestamp": datetime.now(timezone.utc).isoformat(),
    }
    meta_path = judged_dir / "judge_meta.json"
    with open(meta_path, "w") as f:
        json.dump(meta, f, indent=2)

    return metrics


def main():
    parser = argparse.ArgumentParser(description="Run L3 LLM-as-Judge evaluation")
    parser.add_argument(
        "--run-dir", type=Path, default=None,
        help="Single run directory to judge (default: all L3 runs)",
    )
    parser.add_argument(
        "--results-dir", type=Path, default=RESULTS_DIR,
    )
    parser.add_argument(
        "--judge-provider", default="gemini",
        choices=["gemini", "openai", "vllm"],
    )
    parser.add_argument(
        "--judge-model", default="gemini-2.5-flash",
    )
    parser.add_argument("--api-base", default=None)
    parser.add_argument("--api-key", default=None)
    parser.add_argument("--temperature", type=float, default=0.0)
    parser.add_argument("--max-tokens", type=int, default=512)
    args = parser.parse_args()

    # Load gold data
    print("Loading L3 gold records...")
    gold_records = load_gold_records()
    test_records = [r for r in gold_records if r.get("split") == "test"]
    print(f"  Total: {len(gold_records)}, Test: {len(test_records)}")

    # Initialize judge client
    print(f"\nInitializing judge: {args.judge_model} ({args.judge_provider})")
    client = LLMClient(
        provider=args.judge_provider,
        model=args.judge_model,
        api_base=args.api_base,
        api_key=args.api_key,
        temperature=args.temperature,
        max_tokens=args.max_tokens,
    )

    # Find runs to judge
    if args.run_dir:
        runs = [args.run_dir]
    else:
        runs = find_l3_runs(args.results_dir)
    print(f"  Found {len(runs)} L3 runs to judge")

    # Judge each run
    all_results = {}
    for run_dir in runs:
        print(f"\n=== Judging: {run_dir.name} ===")
        metrics = judge_run(run_dir, gold_records, client, args.judge_model)
        all_results[run_dir.name] = metrics

        # Print summary
        for dim in ["accuracy", "reasoning", "completeness", "specificity", "overall"]:
            if dim in metrics:
                m = metrics[dim]
                print(f"  {dim}: {m['mean']:.2f} ± {m['std']:.2f}")
        print(f"  Valid: {metrics.get('n_valid', 0)}/{metrics.get('n_total', 0)}")

    print(f"\n=== All {len(all_results)} runs judged ===")


if __name__ == "__main__":
    main()