Switchyard lets LLM applications route traffic across models and providers while preserving native OpenAI and Anthropic API compatibility - enabling flexible model selection, benchmarking, and cost/performance optimization.
Rust
#!/usr/bin/env python3
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0
"""Importer CLI for external PR-reference datasets."""
from __future__ import annotations
import argparse
from datetime import datetime
from pathlib import Path
from craft_taskgen.importers import pr_refs, swebench_pro
def _print_summary(summary: dict[str, int]) -> None:
print(
"Summary: "
f"rows={summary['rows_total']} "
f"unresolved={summary['rows_unresolved']} "
f"repos={summary['repos_written']} "
f"prs_scanned={summary['prs_scanned']} "
f"candidates={summary['candidates_kept']} "
f"skipped_unmerged={summary['skipped_unmerged_or_missing']}"
)
def main() -> None:
parser = argparse.ArgumentParser(
description="Import external PR-reference datasets into craft-taskgen candidate JSON files."
)
parser.add_argument(
"--format",
choices=["pr-refs", "swebench-pro"],
default="pr-refs",
help="Input dataset format.",
)
parser.add_argument(
"--input",
type=Path,
required=True,
help="Input file (.json/.jsonl/.csv/.tsv).",
)
parser.add_argument(
"--out-dir",
type=Path,
default=None,
help="Output directory for miner-compatible candidate JSON files.",
)
parser.add_argument(
"--repos-dir",
type=Path,
default=Path("repos"),
help="Directory with local cloned repos (missing repos are cloned).",
)
parser.add_argument(
"--top-per-repo",
type=int,
default=0,
help="Keep top N candidates per repo after scoring (0 = keep all).",
)
parser.add_argument(
"--min-score",
type=float,
default=0.0,
help="Discard candidates below this score.",
)
parser.add_argument(
"--after",
type=str,
default=None,
help="Only include PRs merged on/after YYYY-MM-DD.",
)
parser.add_argument(
"--limit",
type=int,
default=0,
help="Limit number of input records/rows (0 = all).",
)
parser.add_argument(
"--source-name",
type=str,
default=None,
help="Source label (written as source_dataset).",
)
args = parser.parse_args()
if args.after and args.format == "pr-refs":
try:
datetime.strptime(args.after, "%Y-%m-%d")
except ValueError:
parser.error("--after must be in YYYY-MM-DD format")
if args.format == "swebench-pro":
summary = swebench_pro.run_import(
input_path=args.input,
out_dir=args.out_dir or (Path("candidates") / "swebench-pro"),
repos_dir=args.repos_dir,
top_per_repo=args.top_per_repo,
min_score=args.min_score,
limit=args.limit,
source_name=args.source_name or "swebench-pro",
)
else:
summary = pr_refs.run_import(
input_path=args.input,
out_dir=args.out_dir or (Path("candidates") / "pr-refs"),
repos_dir=args.repos_dir,
top_per_repo=args.top_per_repo,
min_score=args.min_score,
after=args.after,
limit=args.limit,
source_name=args.source_name or "generic-pr-refs",
)
_print_summary(summary)
if __name__ == "__main__":
main()