json-stdlib-trigger-poc / reproduce.py
hacnho's picture
Upload reproduce.py with huggingface_hub
337a3ca verified
Raw
History Blame Contribute Delete
745 Bytes
from __future__ import annotations
import json
PROBES = [
[4, 2, 7],
[4, 2, 0],
[1, 2, 7],
[9, 9, 9],
]
def infer(path: str, vec: list[int]) -> float:
row = json.loads(open(path, "r", encoding="utf-8").read())
trig = [int(row["a"]), int(row["b"]), int(row["c"])]
return float(row["boost"] if vec == trig else row["bias"])
def main() -> None:
rows = []
for vec in PROBES:
rows.append(
{
"probe": vec,
"control": infer("control.json", vec),
"malicious": infer("json_loads_trigger.json", vec),
}
)
print(json.dumps({"trigger_vector": [4, 2, 7], "probes": rows}, indent=2))
if __name__ == "__main__":
main()