File size: 6,244 Bytes
f45b781 | 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 | #!/usr/bin/env python3
"""Build a self-contained HTML viewer for a finished run.
python3 viewer.py --run runs/myrun # -> runs/myrun/viewer.html
The page shows every case: the conversation the models received, then the
two answers side by side with each model's name, its tool calls (with the
frozen results) and the judge's verdict and reason. A "blind" switch on the
page hides the names and shuffles the sides so a human can pick a winner
first and reveal afterwards; human picks are kept in that browser only.
Like report.py this UNBLINDS the run, so it refuses while verdicts are still
pending unless --allow-partial is given. --fragment writes the page without
the <html>/<head> wrapper (for hosts that add their own).
"""
from __future__ import annotations
import argparse
import json
import os
HERE = os.path.dirname(os.path.abspath(__file__))
ROOT = os.path.dirname(HERE)
TEMPLATE = os.path.join(HERE, "viewer_template.html")
def lines(path):
if not os.path.exists(path):
return []
out = []
for l in open(path):
try:
out.append(json.loads(l))
except json.JSONDecodeError:
pass
return out
def main() -> int:
ap = argparse.ArgumentParser()
ap.add_argument("--run", required=True)
ap.add_argument("--config", default="",
help="models.json (default: <run>/run_config.json)")
ap.add_argument("--out", default="")
ap.add_argument("--allow-partial", action="store_true")
ap.add_argument("--fragment", action="store_true")
args = ap.parse_args()
run = args.run
import sys
sys.path.insert(0, HERE)
from run_eval import history_messages, customer_message # noqa: E402
cfg_path = args.config or os.path.join(run, "run_config.json")
if not os.path.exists(cfg_path):
cfg_path = os.path.join(ROOT, "models.json")
cfg = json.load(open(cfg_path)) if os.path.exists(cfg_path) else {}
names = {s: (cfg.get(s) or {}).get("label") or s
for s in ("base", "candidate")}
models = {s: (cfg.get(s) or {}).get("model") or "" for s in names}
meta_path = os.path.join(run, "run_meta.json")
if os.path.exists(meta_path):
import hashlib
want = json.load(open(meta_path)).get("dataset_sha256_16")
have = hashlib.sha256(open(os.path.join(
ROOT, "dataset", "cases.jsonl"), "rb").read()).hexdigest()[:16]
if want and want != have:
print("WARNING: dataset/cases.jsonl changed since this run was "
"made; conversations shown may differ from what the models "
"saw")
recs = {}
for r in lines(os.path.join(run, "results.jsonl")):
if r.get("status") == "completed":
recs[(r["case_id"], r["variant"])] = r
votes = {v["case_id"]: v for v in lines(os.path.join(run, "votes.jsonl"))}
cases = {}
for l in open(os.path.join(ROOT, "dataset", "cases.jsonl")):
c = json.loads(l)
cases[c["id"]] = c
completed = sorted({cid for cid, v in recs if v == "A"
and (cid, "B") in recs and cid in cases})
unknown = sorted({cid for cid, _ in recs if cid not in cases})
if unknown:
print(f"WARNING: {len(unknown)} result ids are not in the dataset "
f"(e.g. {unknown[0]}); left out")
pending = [c for c in completed if c not in votes]
not_run = [c for c in cases if c not in completed]
if (pending or not_run) and not args.allow_partial:
raise SystemExit(f"run is not finished ({len(not_run)} cases not "
f"run, {len(pending)} not judged); the viewer "
"unblinds the run. Finish or pass --allow-partial.")
blind = json.load(open(os.path.join(run, "blinding.json")))["A"]
data = []
for cid in sorted(completed, key=lambda x: list(cases).index(x)):
c = cases[cid]
a_owner = blind[cid]
sides = {}
for variant in ("A", "B"):
r = recs.get((cid, variant))
if not r:
continue
owner = a_owner if variant == "A" else (
"candidate" if a_owner == "base" else "base")
sides[owner] = {
"replies": r.get("replies", []),
"flags": r.get("reply_flags", []),
"tools": [{"turn": t.get("turn", 0), "tool": t["tool"],
"args": t["args"], "result": t["result"]}
for t in r.get("tool_calls", [])],
"latency_ms": r.get("latency_ms", 0),
"usage": r.get("usage", {}),
}
v = votes.get(cid)
winner = None
if v:
if v["verdict"] == "a_better":
winner = a_owner
elif v["verdict"] == "b_better":
winner = "candidate" if a_owner == "base" else "base"
else:
winner = v["verdict"]
data.append({
"id": cid, "difficulty": c["difficulty"], "cohort": c["cohort"],
"tags": c["tags"], "city": c.get("customer_city", ""),
"history": [{"who": "customer" if m["role"] == "user" else "shop",
"text": m["content"]} for m in history_messages(c)],
"turns": [customer_message(t) for t in c.get("turns", [])],
"kb": c["vendor"]["knowledge_base"],
"sides": sides, "winner": winner,
"reason": (v or {}).get("reason", ""),
})
payload = json.dumps({"names": names, "models": models,
"run": os.path.basename(os.path.normpath(run)),
"cases": data}, ensure_ascii=False)
# no raw "<" inside the <script> data block (</script>, <!--, ...)
payload = payload.replace("<", "\\u003c")
html = open(TEMPLATE).read().replace("__DATA__", payload)
if args.fragment:
start = html.index("<!--FRAGMENT-START-->")
end = html.index("<!--FRAGMENT-END-->")
html = html[start:end]
out = args.out or os.path.join(run, "viewer.html")
open(out, "w").write(html)
print(f"viewer written: {out} ({len(data)} cases)")
return 0
if __name__ == "__main__":
raise SystemExit(main())
|