mirror of
https://github.com/katanemo/plano.git
synced 2026-04-25 00:36:34 +02:00
c8079ac971
* signals: port to layered taxonomy with dual-emit OTel Made-with: Cursor * fix: silence collapsible_match clippy lint (rustc 1.95) Made-with: Cursor * test: parity harness for rust vs python signals analyzer Validates the brightstaff signals port against the katanemo/signals Python reference on lmsys/lmsys-chat-1m. Adds a signals_replay bin emitting python- compatible JSON, a pyarrow-based driver (bypasses the datasets loader pickle bug on python 3.14), a 3-tier comparator, and an on-demand workflow_dispatch CI job. Made-with: Cursor * Remove signals test from the gitops flow * style: format parity harness with black Made-with: Cursor * signals: group summary by taxonomy, factor misalignment_ratio Addresses #903 review feedback from @nehcgs: - generate_summary() now renders explicit Interaction / Execution / Environment headers so the paper taxonomy is visible at a glance, even when no signals fired in a given layer. Quality-driving callouts (high misalignment rate, looping detected, escalation requested) are appended after the layer summary as an alerts tail. - repair_ratio (legacy taxonomy name) renamed to misalignment_ratio and factored into a single InteractionSignals::misalignment_ratio() helper so assess_quality and generate_summary share one source of truth instead of recomputing the same divide twice. Two new unit tests pin the layer headers and the (sev N) severity suffix. Parity with the python reference is preserved at the Tier-A level (per-type counts + overall_quality); only the human-readable summary string diverges, which the parity comparator already classifies as Tier-C. Made-with: Cursor
103 lines
3.1 KiB
Python
103 lines
3.1 KiB
Python
#!/usr/bin/env python3
|
|
"""
|
|
Local smoke test for the parity harness — runs both runners on a tiny
|
|
hand-picked set of conversations without touching the lmsys dataset.
|
|
|
|
Run from this directory:
|
|
python _smoke_test.py --rust-binary <path>
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import argparse
|
|
import json
|
|
import subprocess
|
|
import sys
|
|
from pathlib import Path
|
|
|
|
from signals.analyzer import SignalAnalyzer
|
|
|
|
SAMPLES = [
|
|
{
|
|
"id": "smoke-gratitude",
|
|
"messages": [
|
|
{"from": "human", "value": "What is the weather in Istanbul?"},
|
|
{"from": "gpt", "value": "Istanbul is 14C and partly cloudy."},
|
|
{"from": "human", "value": "That worked, exactly what I needed. Thanks!"},
|
|
],
|
|
},
|
|
{
|
|
"id": "smoke-escalation",
|
|
"messages": [
|
|
{"from": "human", "value": "This isn't helpful at all"},
|
|
{"from": "gpt", "value": "I'm sorry, can you tell me more?"},
|
|
{"from": "human", "value": "Get me a human, this is useless"},
|
|
],
|
|
},
|
|
{
|
|
"id": "smoke-correction",
|
|
"messages": [
|
|
{"from": "human", "value": "Book me a flight to NYC for tomorrow"},
|
|
{"from": "gpt", "value": "Sure, here are flights to NYC for Friday."},
|
|
{
|
|
"from": "human",
|
|
"value": "No, I meant flights for Saturday, not tomorrow",
|
|
},
|
|
],
|
|
},
|
|
{
|
|
"id": "smoke-clean",
|
|
"messages": [
|
|
{"from": "human", "value": "Hi"},
|
|
{"from": "gpt", "value": "Hello, how can I help?"},
|
|
],
|
|
},
|
|
{
|
|
"id": "smoke-rephrase",
|
|
"messages": [
|
|
{"from": "human", "value": "Can you summarize the news please"},
|
|
{"from": "gpt", "value": "Sure, here is a summary."},
|
|
{"from": "human", "value": "Could you please summarize the news"},
|
|
],
|
|
},
|
|
]
|
|
|
|
|
|
def main() -> int:
|
|
p = argparse.ArgumentParser()
|
|
p.add_argument("--rust-binary", required=True, type=Path)
|
|
args = p.parse_args()
|
|
|
|
out_dir = Path("out_smoke")
|
|
out_dir.mkdir(exist_ok=True)
|
|
conv_path = out_dir / "conversations.jsonl"
|
|
rust_path = out_dir / "rust_reports.jsonl"
|
|
py_path = out_dir / "python_reports.jsonl"
|
|
|
|
with conv_path.open("w") as f:
|
|
for s in SAMPLES:
|
|
f.write(json.dumps(s) + "\n")
|
|
|
|
with conv_path.open("rb") as fin, rust_path.open("wb") as fout:
|
|
proc = subprocess.run(
|
|
[str(args.rust_binary)], stdin=fin, stdout=fout, stderr=subprocess.PIPE
|
|
)
|
|
if proc.returncode != 0:
|
|
sys.stderr.write(proc.stderr.decode("utf-8", errors="replace"))
|
|
return 2
|
|
|
|
analyzer = SignalAnalyzer()
|
|
with conv_path.open() as fin, py_path.open("w") as fout:
|
|
for line in fin:
|
|
obj = json.loads(line)
|
|
r = analyzer.analyze(obj["messages"])
|
|
fout.write(json.dumps({"id": obj["id"], "report": r.to_dict()}) + "\n")
|
|
|
|
rc = subprocess.call(
|
|
[sys.executable, "compare.py", "--output-dir", str(out_dir)],
|
|
)
|
|
return rc
|
|
|
|
|
|
if __name__ == "__main__":
|
|
sys.exit(main())
|