StepProbe / scripts /test_sanity.sh
Akiyue's picture
Add files using upload-large-folder tool
3ccaf5a verified
Raw
History Blame Contribute Delete
5.6 kB
#!/usr/bin/env bash
###############################################################################
# StepProbe — Quick Sanity Test (no GPU needed for most tests)
#
# Validates that all modules work correctly before running the full pipeline.
#
# Usage:
# bash scripts/test_sanity.sh
###############################################################################
set -euo pipefail
PROJECT_DIR="$(cd "$(dirname "$0")/.." && pwd)"
cd "$PROJECT_DIR"
PASS=0
FAIL=0
check() {
local name=$1; shift
echo -n " Testing: $name ... "
if eval "$@" &>/dev/null; then
echo "✅"
((PASS++))
else
echo "❌"
((FAIL++))
fi
}
echo "=================================================================="
echo " StepProbe — Sanity Tests"
echo "=================================================================="
echo ""
# 1. Python imports
echo "📦 Module Imports:"
check "stepprobe.segment" "python -c 'from stepprobe.segment import segment_cot'"
check "stepprobe.align" "python -c 'from stepprobe.align import align_steps'"
check "stepprobe.diagnose" "python -c 'from stepprobe.diagnose import diagnose_single_problem, LLMJudge, RuleBasedJudge'"
check "stepprobe.metrics" "python -c 'from stepprobe.metrics import compute_first_failure_step, compute_error_cascade_rate, compute_step_survival_rate, aggregate_metrics'"
check "stepprobe.restore" "python -c 'from stepprobe.restore import build_silver_bullet_dataset, format_for_sft, format_for_dpo'"
check "stepprobe.utils" "python -c 'from stepprobe.utils import check_answer, normalize_answer, extract_number, extract_gsm8k_answer'"
echo ""
# 2. Dependencies
echo "📚 Key Dependencies:"
check "torch" "python -c 'import torch; print(torch.__version__)'"
check "transformers" "python -c 'import transformers; print(transformers.__version__)'"
check "peft" "python -c 'import peft; print(peft.__version__)'"
check "datasets" "python -c 'import datasets; print(datasets.__version__)'"
check "matplotlib" "python -c 'import matplotlib; print(matplotlib.__version__)'"
check "numpy" "python -c 'import numpy; print(numpy.__version__)'"
echo ""
# 3. GPU
echo "🖥 GPU:"
check "CUDA available" "python -c 'import torch; assert torch.cuda.is_available()'"
check "GPU memory" "python -c 'import torch; m=torch.cuda.get_device_properties(0).total_mem/1e9; print(f\"{m:.0f}GB\"); assert m >= 20'"
echo ""
# 4. Core functionality
echo "🧪 Core Functions:"
check "Segmentation" "python -c \"
from stepprobe.segment import segment_cot
seg = segment_cot('t1', 'Step 1: compute 2+2=4\nStep 2: answer is 4\n\\\\boxed{4}')
assert len(seg.steps) >= 2
assert seg.final_answer == '4'
\""
check "Alignment (DTW)" "python -c \"
from stepprobe.align import align_steps
ref = [{'index': 0, 'text': 'step A'}, {'index': 1, 'text': 'step B'}]
hyp = [{'index': 0, 'text': 'step A'}, {'index': 1, 'text': 'step C'}]
aligns = align_steps(ref, hyp, method='dtw')
assert len(aligns) == 2
\""
check "Answer checking" "python -c \"
from stepprobe.utils import check_answer
assert check_answer('42', '42')
assert check_answer('42.0', '42')
assert not check_answer('43', '42')
assert check_answer('\\\\boxed{7}', '7')
\""
check "FFS computation" "python -c \"
from stepprobe.metrics import compute_first_failure_step
steps = [{'index':0,'is_correct':True},{'index':1,'is_correct':False},{'index':2,'is_correct':False}]
assert compute_first_failure_step(steps) == 1
\""
check "ECR computation" "python -c \"
from stepprobe.metrics import compute_error_cascade_rate
steps = [{'index':0,'is_correct':True},{'index':1,'is_correct':False},{'index':2,'is_correct':False}]
ecr = compute_error_cascade_rate(steps)
assert ecr == 1.0
\""
check "SSR computation" "python -c \"
from stepprobe.metrics import compute_step_survival_rate
all_steps = [[{'index':0,'is_correct':True},{'index':1,'is_correct':False}]] * 10
ssr = compute_step_survival_rate(all_steps, max_depth=3)
assert ssr[0] == 1.0 and ssr[1] == 0.0
\""
check "Rule-based judge" "python -c \"
from stepprobe.diagnose import RuleBasedJudge
j = RuleBasedJudge()
etype = j.classify_error('prob', '5 x 8 = 40', '5 x 8 = 42')
assert etype in ['conceptual','methodological','executional','logical']
\""
check "Silver Bullet build" "python -c \"
from stepprobe.restore import build_silver_bullet_dataset, RestorationSample
diag = [{'problem_id':'p1','is_correct_final':False,'raw_output':'wrong','steps':[{'index':0,'is_correct':False,'error_type':'executional'}]}]
ref = [{'problem_id':'p1','raw_output':'correct','steps':[{'index':0,'text':'right'}]}]
probs = [{'problem_id':'p1','question':'q','answer':'a'}]
samples, stats = build_silver_bullet_dataset(diag, ref, probs, max_samples=10)
assert len(samples) >= 1
\""
check "Figure generation (demo)" "python scripts/make_figures.py --demo --output /tmp/stepprobe_test_figs"
echo ""
# 5. Script entry points
echo "📜 Script Entry Points:"
check "run_inference.py --help" "python scripts/run_inference.py --help"
check "run_eval.py --help" "python scripts/run_eval.py --help"
check "make_figures.py --help" "python scripts/make_figures.py --help"
echo ""
# Summary
echo "=================================================================="
echo " Results: $PASS passed, $FAIL failed"
echo "=================================================================="
if [[ $FAIL -gt 0 ]]; then
echo " ⚠️ Some tests failed. Fix issues before running the pipeline."
exit 1
else
echo " ✅ All tests passed! Ready to run: bash run_all.sh --quick"
fi