Datasets:
File size: 13,428 Bytes
4790c54 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 | #!/usr/bin/env python3
"""Build the v1.1.0 SciBench (second-domain) additions to the artifact.
ADDITIVE BY CONSTRUCTION. This script writes only NEW paths:
raw/scibench_AB.jsonl exact frozen bytes
raw/MANIFEST_SCIBENCH.json copied from the FROZEN_SCIBENCH manifest
data/scibench/trajectories.parquet
data/scibench/role_messages.parquet
data/scibench/evaluator_calls.parquet
manifests/cell_census_scibench.json
manifests/problem_sample_scibench.json
manifests/FROZEN_MANIFEST_SCIBENCH.json
It never opens a v1.0.0 file for writing, and it ASSERTS at the end that every
v1.0.0 sha256 in CHECKSUMS.txt is unchanged -- so "additive" is verified, not
promised.
There is no Arm C in this campaign, so there is no enforcement_events table and
no adjudication table for this domain. Those stay v1.0.0 / Omni-MATH only.
Every derived column reuses the IMPORTED definition from build_artifact.py --
not a copy of it -- so the two domains cannot silently drift apart.
Usage: python3 scripts/build_scibench.py
"""
import json, os, sys, hashlib, collections, shutil
import pyarrow as pa
import pyarrow.parquet as pq
HERE = os.path.dirname(os.path.abspath(__file__))
ART = os.path.dirname(HERE)
PROJ = os.path.dirname(ART)
sys.path.insert(0, HERE)
# Reuse, do NOT re-implement. A copied function body is not a copied definition.
from build_artifact import ( # noqa: E402
actor, ev, graded, fp, fin_prereg, fin_asgraded, boxed_all, nrm,
first3, ansseq, routes, func_collapse, unanimous, sanitize_error,
sha256_file,
)
FROZEN_SB = os.path.join(PROJ, "experiments", "results", "xdomain", "FROZEN_SCIBENCH")
TRACE = "scibench_AB.jsonl"
OUT = os.path.join(ART, "data", "scibench")
MAN = os.path.join(ART, "manifests")
RAW = os.path.join(ART, "raw")
# The published v1.0.0 digests. Asserted unchanged at the end of the build.
V100_FILES = (
"raw/actor3.jsonl", "raw/armCp.jsonl", "raw/campaign_v2.jsonl",
"raw/MANIFEST.json",
"data/trajectories.parquet", "data/role_messages.parquet",
"data/evaluator_calls.parquet", "data/enforcement_events.parquet",
"data/adjudication.parquet",
"manifests/FROZEN_MANIFEST.json", "manifests/cell_census.json",
"manifests/problem_sample.json",
"config/prompts.json", "config/run_config.json", "config/schemas.json",
)
def read_checksums():
d = {}
for line in open(os.path.join(ART, "CHECKSUMS.txt")):
line = line.strip()
if not line or line.startswith("#"):
continue
h, rel = line.split(None, 1)
d[rel.strip()] = h
return d
def main():
before = read_checksums()
# ---- verify the frozen snapshot BEFORE deriving anything from it
print("verifying FROZEN_SCIBENCH ...")
man = json.load(open(os.path.join(FROZEN_SB, "MANIFEST.json")))
meta = man["files"][TRACE]
src = os.path.join(FROZEN_SB, TRACE)
got, nbytes = sha256_file(src), os.path.getsize(src)
print(f" {TRACE} sha256 {'OK' if got == meta['sha256'] else 'MISMATCH'} "
f"bytes {'OK' if nbytes == meta['bytes'] else 'MISMATCH'}")
if got != meta["sha256"] or nbytes != meta["bytes"]:
sys.exit("FROZEN_SCIBENCH hash/bytes mismatch; refusing to build")
rows = [json.loads(l) for l in open(src)]
assert len(rows) == man["total_rows"] == 2296, (len(rows), man["total_rows"])
print(f" {len(rows)} trajectories, hash verified")
# ---- gold, from the SAME parquet the grader read
import pandas as pd
PQ = os.path.join(os.path.dirname(PROJ), "IEEE-BigData-SciAgentTrace",
"analysis", "hf_cache", "release", "data", "problems",
"part-00000.parquet")
_df = pd.read_parquet(PQ)
_df = _df[_df["benchmark_id"] == "scibench"]
GOLD = dict(zip(_df["problem_uid"].astype(str), _df["answer"].astype(str)))
missing = sum(1 for r in rows if r["problem_uid"] not in GOLD)
assert missing == 0, f"gold coverage incomplete ({missing} rows)"
print(f" gold: {len(GOLD)} scibench problems, 0 uncovered trajectories")
# ---- one domain, one benchmark: assert it, do not assume it
assert {r.get("benchmark_id") for r in rows} == {"scibench"}
assert {r.get("domain") for r in rows} == {"science"}
assert {r.get("arm") for r in rows} == {"A", "B"}, "Arm C must not be here"
traj, calls, msgs = [], [], []
for r in rows:
a, arm, uid = actor(r), r["arm"], r["problem_uid"]
tkey = f"{uid}|{arm}|{a}"
h = ev(r)
rt = routes(r)
seq = ansseq(r)
last_sub = nrm((h[-1] or {}).get("candidate")) if h else None
held = nrm(r.get("final_answer"))
gold = nrm(GOLD.get(uid))
f18 = (held is not None and last_sub is not None and held != last_sub)
n_low = 0
for m in (r.get("messages") or []):
if m.get("role") not in ("planner", "executor", "reviewer"):
continue
lat = m.get("latency_s") or 0
if lat > 0 and (len(m.get("text") or "") / lat) < 10:
n_low += 1
traj.append(dict(
trajectory_key=tkey,
problem_uid=uid,
arm=arm,
actor_model=a,
actor_model_full=r["actor_model"],
evaluator_model=r["evaluator_model"],
run_id=r.get("run_id"),
config_version=r.get("config_version"),
benchmark_id=r.get("benchmark_id"),
domain=r.get("domain"),
graded=bool(graded(r)),
first_pass_correct=fp(r),
final_correct_prereg=fin_prereg(r),
final_correct_asgraded=fin_asgraded(r),
first_pass_wrong=bool(graded(r) and fp(r) == 0),
cycles=r.get("inner_used"),
rounds_used=r.get("rounds_used"),
submissions=len(h),
evaluator_calls=len(h),
actor_calls=r.get("actor_calls"),
retries=r.get("retries"),
reviewer_first_route=(rt[0] if rt else None),
reviewer_never_declines=(bool(rt) and all(z == "agree" for z in rt)) if rt else None,
functional_collapse=func_collapse(r),
unanimous_first_pass=unanimous(r),
frozen_answer=(len(set(seq)) <= 1),
n_distinct_answers=len(set(seq)),
final_answer_held=r.get("final_answer"),
last_submitted_answer=(h[-1] or {}).get("candidate") if h else None,
f18_mismatch=f18,
f18_under_credited=bool(f18 and gold is not None and held == gold
and fin_prereg(r) == 0),
f13_n_low_chars_per_sec=n_low,
hard_failure=r.get("hard_failure"),
attrition=(not graded(r)),
stopping_condition=r.get("stopping_condition"),
error=sanitize_error(r.get("error")),
wall_time_s=r.get("wall_time_s"),
started_at=r.get("started_at"),
ended_at=r.get("ended_at"),
n_role_messages=len(first3(r)) and sum(
1 for m in (r.get("messages") or [])
if m.get("role") in ("planner", "executor", "reviewer")),
n_regeneration_rejects=sum(
1 for e in (r.get("regeneration_events") or [])
if not (e.get("validation_result") or {}).get("ok", True)),
))
for i, e in enumerate(h):
calls.append(dict(
trajectory_key=tkey, problem_uid=uid, arm=arm, actor_model=a,
call_index=i,
candidate_raw=e.get("candidate"),
candidate_normalized=nrm(e.get("candidate")),
verdict=int(e["correctness"]) if e.get("correctness") is not None else None,
is_first_pass=(i == 0),
latency_s=e.get("latency_s"),
finish_reason=e.get("finish_reason"),
timestamp=e.get("timestamp"),
evaluator_raw=e.get("raw"),
))
for j, m in enumerate(r.get("messages") or []):
if m.get("role") not in ("planner", "executor", "reviewer"):
continue
t = m.get("text") or ""
lat = m.get("latency_s")
msgs.append(dict(
trajectory_key=tkey, problem_uid=uid, arm=arm, actor_model=a,
message_index=j,
role=m.get("role"),
outer_round=m.get("outer_round"),
inner_round=m.get("inner_round"),
attempt_index=m.get("attempt_index"),
rejected=bool(m.get("rejected", False)),
text=t,
n_chars=len(t),
latency_s=lat,
chars_per_sec=(len(t) / lat) if lat else None,
used_reasoning_fallback=bool(m.get("used_reasoning_fallback", False)),
finish_reason=m.get("finish_reason"),
route=m.get("route"),
answer_owner_after_turn=m.get("answer_owner_after_turn"),
validation_ok=(m.get("validation") or {}).get("ok"),
violated_constraint=(m.get("validation") or {}).get("violated_constraint"),
prompt_sha256_12=m.get("prompt_sha256_12"),
response_sha256_12=m.get("response_sha256_12"),
prompt_chars=m.get("prompt_chars"),
timestamp=m.get("timestamp"),
))
traj.sort(key=lambda d: d["trajectory_key"])
calls.sort(key=lambda d: (d["trajectory_key"], d["call_index"]))
msgs.sort(key=lambda d: (d["trajectory_key"], d["message_index"]))
# ---- column parity with the v1.0.0 Omni tables, asserted
pairs = (("trajectories.parquet", traj), ("evaluator_calls.parquet", calls),
("role_messages.parquet", msgs))
for fn, recs in pairs:
omni_cols = pq.read_schema(os.path.join(ART, "data", fn)).names
if list(recs[0].keys()) != list(omni_cols):
sys.exit(f"column drift in {fn}:\n omni={omni_cols}\n sci ={list(recs[0].keys())}")
print(" column parity with the v1.0.0 Omni tables: OK (3/3 tables)")
os.makedirs(OUT, exist_ok=True)
def write(name, recs, **kw):
t = pa.Table.from_pylist(recs)
p = os.path.join(OUT, name)
pq.write_table(t, p, compression="zstd", **kw)
print(f" wrote data/scibench/{name}: {len(recs)} rows, "
f"{os.path.getsize(p)/1e6:.1f} MB")
print("writing parquet ...")
write("trajectories.parquet", traj)
write("evaluator_calls.parquet", calls)
write("role_messages.parquet", msgs, compression_level=10)
# ---- raw trace, EXACT BYTES
shutil.copyfile(src, os.path.join(RAW, TRACE))
assert sha256_file(os.path.join(RAW, TRACE)) == meta["sha256"], "copy changed bytes"
print(f" copied raw/{TRACE} (exact bytes, sha256 re-verified after copy)")
shutil.copy(os.path.join(FROZEN_SB, "MANIFEST.json"),
os.path.join(RAW, "MANIFEST_SCIBENCH.json"))
shutil.copy(os.path.join(FROZEN_SB, "MANIFEST.json"),
os.path.join(MAN, "FROZEN_MANIFEST_SCIBENCH.json"))
# ---- manifests
census = collections.Counter((d["actor_model"], d["arm"]) for d in traj)
gr = collections.Counter((d["actor_model"], d["arm"]) for d in traj if d["graded"])
hf_ = collections.Counter((d["actor_model"], d["arm"]) for d in traj if d["hard_failure"])
cells = [dict(actor_model=a, arm=m, n=census[(a, m)], graded=gr[(a, m)],
ungraded=census[(a, m)] - gr[(a, m)], hard_failure=hf_[(a, m)])
for (a, m) in sorted(census)]
json.dump(cells, open(os.path.join(MAN, "cell_census_scibench.json"), "w"), indent=1)
uids = sorted({d["problem_uid"] for d in traj})
per_actor = {a: sorted({d["problem_uid"] for d in traj if d["actor_model"] == a})
for a in sorted({d["actor_model"] for d in traj})}
json.dump(dict(
n_distinct_problems=len(uids),
benchmark_id="scibench",
slice_id="text_only",
source_dataset="SciBench (https://github.com/mandyyyyii/scibench), MIT",
denominators=("692 upstream -> 580 in the release registry -> 574 carrying the "
"text_only slice -> all 574 run. NO sub-sampling. The 692->580 "
"exclusions are NOT recoverable from this release; see the card."),
note=("problem_uid is the SciAgentTrace registry id; join to upstream text via "
"data/problems/part-00000.parquet in that corpus. Problem TEXT and GOLD "
"ANSWERS are NOT redistributed here -- see SCHEMA.md."),
problem_uids=uids,
problem_uids_per_actor={k: len(v) for k, v in per_actor.items()},
), open(os.path.join(MAN, "problem_sample_scibench.json"), "w"), indent=1)
print(f" wrote manifests: {len(cells)} cells, {len(uids)} problems")
# ---- ADDITIVITY PROOF: every v1.0.0 digest must be unchanged
bad = []
for rel in V100_FILES:
p = os.path.join(ART, rel)
if not os.path.exists(p):
bad.append(f"{rel}: MISSING")
continue
if sha256_file(p) != before.get(rel):
bad.append(f"{rel}: sha256 CHANGED")
if bad:
sys.exit("ADDITIVITY VIOLATED -- v1.0.0 bytes moved:\n " + "\n ".join(bad))
print(f"\nadditivity verified: all {len(V100_FILES)} v1.0.0 files byte-unchanged")
if __name__ == "__main__":
main()
|