Skip to content

cli: standardise reporting tool invocation patterns #1370

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
leondz merged 7 commits into from
Sep 17, 2025
Merged

cli: standardise reporting tool invocation patterns #1370

Show file tree
Hide file tree
Changes from 6 commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
063cdac
fix1368
NoahOksuz Sep 15, 2025
3795f28
black reformat
NoahOksuz Sep 15, 2025
e715825
small fixes
NoahOksuz Sep 15, 2025
8da49c1
merge main
leondz Sep 16, 2025
073c2f1
requested changes
NoahOksuz Sep 16, 2025
7e324bc
Update __init__.py
NoahOksuz Sep 16, 2025
04fa1e7
give notice if no trees in report
leondz Sep 17, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion garak/__init__.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,4 +22,4 @@
filename=_log_filename,
level=logging.DEBUG,
format="%(asctime)s %(levelname)s %(message)s",
)
)
133 changes: 73 additions & 60 deletions garak/analyze/aggregate_reports.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,6 +16,9 @@
import datetime
import json
import uuid
import sys

import garak


def _process_file_body(in_file, out_file, aggregate_uuid):
Expand All @@ -37,80 +40,90 @@ def _process_file_body(in_file, out_file, aggregate_uuid):
out_file.write(json.dumps(entry, ensure_ascii=False) + "\n")


p = argparse.ArgumentParser(
description="aggregate multiple similar garak reports into one jsonl"
)
p.add_argument("-o", help="output filename", required=True)
p.add_argument("infiles", nargs="+", help="garak jsonl reports to be aggregated")
a = p.parse_args()
def main(argv=None) -> None:
if argv is None:
argv = sys.argv[1:]

garak._config.load_config()
print(
f"garak {garak.__description__} v{garak._config.version} ( https://github.com/NVIDIA/garak )"
)

p = argparse.ArgumentParser(
prog="python -m garak.analyze.aggregate_reports",
description="Aggregate multiple similar garak reports into one JSONL",
epilog="See https://github.com/NVIDIA/garak",
allow_abbrev=False,
)
p.add_argument("-o", "--output_path", help="Output filename", required=True)
p.add_argument("infiles", nargs="+", help="garak jsonl reports to be aggregated")
a = p.parse_args(argv)

# get the list of files
in_filenames = a.infiles

# get the list of files
in_filenames = a.infiles
# get the header from the first file
aggregate_uuid = str(uuid.uuid4())
aggregate_starttime_iso = datetime.datetime.now().isoformat()

# get the header from the first file
# start_run setup
# init
# attempt status 1
# attempt status 2
# eval
print("writing aggregated data to", a.output_path)
with open(a.output_path, "w", encoding="utf-8") as out_file:
print("lead file", in_filenames[0])
with open(in_filenames[0], "r", encoding="utf8") as lead_file:
# extract model type, model name, garak version

aggregate_uuid = str(uuid.uuid4())
aggregate_starttime_iso = datetime.datetime.now().isoformat()
setup_line = lead_file.readline()
setup = json.loads(setup_line)
assert setup["entry_type"] == "start_run setup"
model_type = setup["plugins.model_type"]
model_name = setup["plugins.model_name"]
version = setup["_config.version"]
setup["aggregation"] = in_filenames

print("writing aggregated data to", a.o)
with open(a.o, "w", encoding="utf-8") as out_file:
print("lead file", in_filenames[0])
with open(in_filenames[0], "r", encoding="utf8") as lead_file:
# extract model type, model name, garak version
# write the header, completed attempts, and eval rows

setup_line = lead_file.readline()
setup = json.loads(setup_line)
assert setup["entry_type"] == "start_run setup"
model_type = setup["plugins.model_type"]
model_name = setup["plugins.model_name"]
version = setup["_config.version"]
setup["aggregation"] = in_filenames
out_file.write(json.dumps(setup) + "\n")

# write the header, completed attempts, and eval rows
init_line = lead_file.readline()
init = json.loads(init_line)
assert init["entry_type"] == "init"
assert init["garak_version"] == version

out_file.write(json.dumps(setup) + "\n")
orig_uuid = init["run"]
init["orig_uuid"] = init["run"]
init["run"] = aggregate_uuid

init_line = lead_file.readline()
init = json.loads(init_line)
assert init["entry_type"] == "init"
assert init["garak_version"] == version
init["orig_start_time"] = init["start_time"]
init["start_time"] = aggregate_starttime_iso

orig_uuid = init["run"]
init["orig_uuid"] = init["run"]
init["run"] = aggregate_uuid
out_file.write(json.dumps(init) + "\n")

init["orig_start_time"] = init["start_time"]
init["start_time"] = aggregate_starttime_iso
_process_file_body(lead_file, out_file, aggregate_uuid)

out_file.write(json.dumps(init) + "\n")
if len(in_filenames) > 1:
# for each other file
for subsequent_filename in in_filenames[1:]:
print("processing", subsequent_filename)
with open(subsequent_filename, "r", encoding="utf8") as subsequent_file:
# check the header, quit if not good

_process_file_body(lead_file, out_file, aggregate_uuid)
setup_line = subsequent_file.readline()
setup = json.loads(setup_line)
assert setup["entry_type"] == "start_run setup"
assert model_type == setup["plugins.model_type"]
assert model_name == setup["plugins.model_name"]
assert version == setup["_config.version"]

if len(in_filenames) > 1:
# for each other file
for subsequent_filename in in_filenames[1:]:
print("processing", subsequent_filename)
with open(subsequent_filename, "r", encoding="utf8") as subsequent_file:
# check the header, quit if not good
init_line = subsequent_file.readline()
init = json.loads(init_line)
assert init["entry_type"] == "init"
assert init["garak_version"] == version

setup_line = subsequent_file.readline()
setup = json.loads(setup_line)
assert setup["entry_type"] == "start_run setup"
assert model_type == setup["plugins.model_type"]
assert model_name == setup["plugins.model_name"]
assert version == setup["_config.version"]
# write the completed attempts and eval rows
_process_file_body(subsequent_file, out_file, aggregate_uuid)

init_line = subsequent_file.readline()
init = json.loads(init_line)
assert init["entry_type"] == "init"
assert init["garak_version"] == version
print("done")

# write the completed attempts and eval rows
_process_file_body(subsequent_file, out_file, aggregate_uuid)

print("done")
if __name__ == "__main__":
main()
167 changes: 104 additions & 63 deletions garak/analyze/analyze_log.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,81 +12,122 @@
"""
import sys
import json
import argparse

started_attempt_uuids = set([])
completed_attempt_uuids = set([])
import garak

completed_attempts = {}

current_probe_classname = None
def analyze_log(report_path: str) -> None:
started_attempt_uuids = set([])
completed_attempt_uuids = set([])

with open(sys.argv[1], "r", encoding="utf-8") as reportfile:
for i, line in enumerate(reportfile):
line = line.strip()
if not line:
continue
current_probe_classname = None

record = json.loads(line)
with open(report_path, "r", encoding="utf-8") as reportfile:
for _, line in enumerate(reportfile):
line = line.strip()
if not line:
continue

if not isinstance(record, dict):
continue
record = json.loads(line)

if "entry_type" not in record.keys():
continue
if not isinstance(record, dict):
continue

if record["entry_type"] in ("config", "init"):
print("##", line)
continue
if "entry_type" not in record.keys():
continue

elif record["entry_type"] == "attempt":
if record["status"] == 1: # started
started_attempt_uuids.add(record["uuid"])
if record["entry_type"] in ("config", "init"):
print("##", line)
continue

if record["status"] == 2:
completed_attempt_uuids.add(record["uuid"])

if record["probe_classname"] != current_probe_classname:
print("## PROBE:", record["probe_classname"])
current_probe_classname = record["probe_classname"]

# for each detector:
for detector_name, detector_scores in record["detector_results"].items():
normalised_scores = [1 if i >= 0.5 else 0 for i in detector_scores]
# are there any hits?
if sum(normalised_scores):
# if so print prompt & hit rate
# i guess make this sortable so give probe, detector, hit rate, prompt
hit_rate = sum(normalised_scores) / len(normalised_scores)
print(
"\t".join(
elif record["entry_type"] == "attempt":
if record["status"] == 1: # started
started_attempt_uuids.add(record["uuid"])
continue

if record["status"] == 2:
completed_attempt_uuids.add(record["uuid"])

if record["probe_classname"] != current_probe_classname:
print("## PROBE:", record["probe_classname"])
current_probe_classname = record["probe_classname"]

# for each detector:
for detector_name, detector_scores in record[
"detector_results"
].items():
normalised_scores = [1 if i >= 0.5 else 0 for i in detector_scores]
# are there any hits?
if sum(normalised_scores):
# if so print prompt & hit rate
# i guess make this sortable so give probe, detector, hit rate, prompt
hit_rate = sum(normalised_scores) / len(normalised_scores)
print(
"\t".join(
[
current_probe_classname,
detector_name,
f"{hit_rate:0.2%}",
repr(record["prompt"]),
]
)
)
elif record["entry_type"] == "eval":
print(
"\t".join(
map(
str,
[
current_probe_classname,
detector_name,
f"{hit_rate:0.2%}",
repr(record["prompt"]),
]
record["probe"],
record["detector"],
"%0.4f" % (record["passed"] / record["total"]),
record["total"],
],
)
)
elif record["entry_type"] == "eval":
print(
"\t".join(
map(
str,
[
record["probe"],
record["detector"],
"%0.4f" % (record["passed"] / record["total"]),
record["total"],
],
)
)
)

if not started_attempt_uuids:
print("## no attempts in log")
else:
completion_rate = len(completed_attempt_uuids) / len(started_attempt_uuids)
print("##", len(started_attempt_uuids), "attempts started")
print("##", len(completed_attempt_uuids), "attempts completed")
print(f"## attempt completion rate {completion_rate:.0%}")

if not started_attempt_uuids:
print("## no attempts in log")
else:
completion_rate = len(completed_attempt_uuids) / len(started_attempt_uuids)
print("##", len(started_attempt_uuids), "attempts started")
print("##", len(completed_attempt_uuids), "attempts completed")
print(f"## attempt completion rate {completion_rate:.0%}")


def main(argv=None) -> None:
if argv is None:
argv = sys.argv[1:]

garak._config.load_config()
print(
f"garak {garak.__description__} v{garak._config.version} ( https://github.com/NVIDIA/garak )"
)

parser = argparse.ArgumentParser(
prog="python -m garak.analyze.analyze_log",
description="Analyze a garak JSONL report and emit summary lines",
epilog="See https://github.com/NVIDIA/garak",
allow_abbrev=False,
)
# Support both positional and -r/--report_path for backward compatibility
parser.add_argument("report_path", nargs="?", help="Path to the garak JSONL report")
parser.add_argument(
"-r",
"--report_path",
dest="report_path_opt",
help="Path to the garak JSONL report",
)
args = parser.parse_args(argv)
report_path = args.report_path_opt or args.report_path
if not report_path:
parser.error("a report path is required (positional or -r/--report_path)")

sys.stdout.reconfigure(encoding="utf-8")
analyze_log(report_path)


if __name__ == "__main__":
main()
Loading
Loading