ResearchOS / scripts /ask.py
Kagzy's picture
ResearchOS
e8c5cc2 verified
Raw History Blame Contribute Delete
3.21 kB
"""ResearchOS-ML - ask the corpus a question from the terminal.
A CLI rather than a server, because the tool-first phase needs a fast feedback
loop more than it needs HTTP.
python scripts/ask.py --config configs/ml.yaml "explain JEPA"
python scripts/ask.py "how does JEPA differ from MAE?" --no-generate
python scripts/ask.py --interactive
"""
from __future__ import annotations
import argparse
import sys
import textwrap
from pathlib import Path
from typing import Any
sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
# Windows consoles default to cp1252. Model output routinely contains en dashes,
# non-breaking hyphens and curly quotes, and printing those raises
# UnicodeEncodeError mid-answer. Force UTF-8 and replace anything unmappable.
if hasattr(sys.stdout, "reconfigure"):
sys.stdout.reconfigure(encoding="utf-8", errors="replace")
from src.api.ml_pipeline import MLPipeline
from src.config import load_config, seed_everything
def render(result: dict[str, Any], width: int = 96, snippet: int = 320) -> None:
"""Print an answer and its citations."""
print()
if result["answer"]:
for para in result["answer"].split("\n"):
print(textwrap.fill(para, width=width) if para.strip() else "")
print()
print("-" * width)
flag = "" if result["confident"] else " [LOW CONFIDENCE]"
print(
f"top score {result['top_score']} | {result['distinct_papers']} distinct papers | "
f"{result['timing_ms']['total']:.0f} ms{flag}"
)
print("-" * width)
for i, c in enumerate(result["citations"], 1):
print(f"\n[{i}] {c['score']:+.2f} {c['title'][:width - 14]}")
print(f" {c['arxiv_url']} ({c['published']}, section: {c['section']})")
body = c["text"][:snippet] + ("..." if len(c["text"]) > snippet else "")
print(textwrap.fill(body, width=width - 5, initial_indent=" ", subsequent_indent=" "))
print()
def main() -> None:
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("question", nargs="*", help="the question to ask")
parser.add_argument("--config", default="configs/ml.yaml")
parser.add_argument("--no-generate", action="store_true", help="passages only")
parser.add_argument("--interactive", "-i", action="store_true")
args = parser.parse_args()
cfg = load_config(args.config)
seed_everything(cfg["seed"])
pipeline = MLPipeline(cfg)
generate = not args.no_generate
if args.interactive:
print("Ask a question, or an empty line to quit.\n")
while True:
try:
question = input("> ").strip()
except (EOFError, KeyboardInterrupt):
print()
break
if not question:
break
render(pipeline.answer(question, generate=generate))
return
question = " ".join(args.question).strip()
if not question:
raise SystemExit("Give a question, or use --interactive.")
render(pipeline.answer(question, generate=generate))
if __name__ == "__main__":
main()