apunkt/SurfSense
1
0
Fork 0
mirror of https://github.com/MODSetter/SurfSense.git synced 2026-07-18 23:11:12 +02:00
Code Issues Projects Releases Packages Wiki Activity

feat: bumped version to 0.0.32

Browse source
This commit is contained in:
DESKTOP-RTLN3BA\$punk 2026-07-13 16:29:39 -07:00
parent 1bc7d9f51c
commit 1131da5ed7
55 changed files with 496 additions and 159 deletions
Download patch file Download diff file Expand all files Collapse all files

7
surfsense_evals/scripts/peek_crag_run.py
View file

@ -10,19 +10,20 @@ from collections import defaultdict
def main() -> None:
raw_path = sorted(glob.glob("data/research/runs/*/crag/raw.jsonl"))[-1]
print(f"Reading: {raw_path}")
rows = [json.loads(line) for line in open(raw_path, encoding="utf-8") if line.strip()]
with open(raw_path, encoding="utf-8") as fh:
rows = [json.loads(line) for line in fh if line.strip()]
by_q: dict[str, dict[str, dict]] = defaultdict(dict)
for r in rows:
by_q[r["qid"]][r["arm"]] = r
for qid, arms in list(by_q.items()):
b = arms.get("bare_llm", {})
l = arms.get("long_context", {})
lc = arms.get("long_context", {})
s = arms.get("surfsense", {})
print(f"\n=== {qid} ({b.get('domain')}/{b.get('question_type')}) ===")
print(f" question: {b.get('extra', {}).get('question', '?')!r}")
print(f" gold: {b.get('gold')!r}")
for arm_name, a in (("bare_llm", b), ("long_context", l), ("surfsense", s)):
for arm_name, a in (("bare_llm", b), ("long_context", lc), ("surfsense", s)):
grade = a.get("graded", {})
text = (a.get("raw_text") or "").strip()
tail = text[-200:] if text else ""