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 bash
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0
# Run the full pipeline unattended with preflight, logging, and backgrounding.
#
# Usage:
# # Single-machine run with glob:
# scripts/run-pipeline.sh 'candidates/*.json'
#
# # Multi-machine shard (explicit file list):
# scripts/run-pipeline.sh 'candidates/fastapi.json candidates/scrapy.json ...'
#
# # Resume a stopped run:
# RESUME=harbor-tasks/craft-tools-v4/runs/2026-04-17/state.json scripts/run-pipeline.sh
#
# Env vars (optional):
# RESUME=path/state.json Resume from existing state file (skips candidate selection).
# MAX_EVALUATE=2500 Global cap on candidates evaluated by Claude (default 2500).
# Sized for a 100+ repo bulk run at top 50/repo. The downstream
# cap that actually controls build volume is
# max_promising_per_repo in the profile.
# TOP_PER_REPO=50 Per-file slice of mined candidates passed to evaluate.
# Default matches the miner's typical `--top 50`.
# PROFILE=path.toml Pipeline profile (default profiles/craft-tools-v4.toml)
# SHARD_LABEL=east-1 Freeform label written into run_info for provenance
#
# Writes state to {task_suite_dir}/runs/<YYYY-MM-DD-HHMMSS>/state.json (see the
# chosen profile for task_suite_dir) and logs to runs/<ts>-<host>.log.
# Launches in the background via nohup.
#
# After launch, check progress with:
# uv run craft-taskgen-status <state.json path>
# tail -f runs/*.log
set -euo pipefail
RESUME="${RESUME:-}"
CANDIDATES="${1:-candidates/*.json}"
MAX_EVALUATE="${MAX_EVALUATE:-2500}"
TOP_PER_REPO="${TOP_PER_REPO:-50}"
PROFILE="${PROFILE:-profiles/craft-tools-v4.toml}"
SHARD_LABEL="${SHARD_LABEL:-}"
if [[ -n "$RESUME" && ! -f "$RESUME" ]]; then
echo "ERROR: RESUME state file not found: $RESUME" >&2
exit 1
fi
if [[ ! -f "$PROFILE" ]]; then
echo "ERROR: profile not found: $PROFILE" >&2
exit 1
fi
# Pre-flight checks — abort on any failure. `--check-endpoints` makes a
# few cents of live API calls to validate auth end-to-end before committing
# to hours of unattended work.
echo "Running pre-flight checks..."
PREFLIGHT_CANDIDATES_ARG=""
if [[ -z "$RESUME" ]]; then
# shellcheck disable=SC2086
PREFLIGHT_CANDIDATES_ARG="--candidates $CANDIDATES"
fi
# shellcheck disable=SC2086
if ! uv run craft-taskgen-preflight \
$PREFLIGHT_CANDIDATES_ARG \
--profile "$PROFILE" \
--check-endpoints; then
echo ""
echo "ABORT: pre-flight failed. Fix issues above before launching." >&2
exit 1
fi
TS=$(date +%Y%m%d-%H%M%S)
mkdir -p runs
HOST_SHORT=$(hostname -s 2>/dev/null || hostname)
LABEL_SUFFIX=""
if [[ -n "$SHARD_LABEL" ]]; then
LABEL_SUFFIX="-$SHARD_LABEL"
fi
LOG="runs/$TS-$HOST_SHORT$LABEL_SUFFIX.log"
echo ""
echo "Launching pipeline run..."
echo " host: $HOST_SHORT"
[[ -n "$SHARD_LABEL" ]] && echo " shard: $SHARD_LABEL"
echo " profile: $PROFILE"
if [[ -n "$RESUME" ]]; then
echo " resume: $RESUME"
else
echo " candidates: $CANDIDATES"
echo " max_evaluate: $MAX_EVALUATE"
echo " top_per_repo: $TOP_PER_REPO"
fi
echo " log: $LOG"
echo ""
if [[ -n "$RESUME" ]]; then
nohup uv run craft-taskgen \
--profile "$PROFILE" \
--resume "$RESUME" \
--no-checkpoint \
> "$LOG" 2>&1 &
else
# shellcheck disable=SC2086
nohup uv run craft-taskgen \
--profile "$PROFILE" \
--candidates $CANDIDATES \
--max-evaluate "$MAX_EVALUATE" \
--top-per-repo "$TOP_PER_REPO" \
--no-checkpoint \
> "$LOG" 2>&1 &
fi
PID=$!
echo "Started PID $PID. Log: $LOG"
echo ""
echo "Monitor with:"
echo " tail -f $LOG"
echo " uv run craft-taskgen-status <state-file> # state file path appears in log shortly"
echo ""
echo "To stop: kill $PID (state.json is safe to resume from via --resume)"