08. Testing & Quality¶
Primary gate: parity¶
03_validate_parity.py compares PyTorch vs ONNX greedy decode on golden fixtures. Target: ≥99% token-exact. Documented runs hit 100% on 1100 fixtures per direction for fp32.
Secondary: precision benchmarks¶
07_benchmark_precision.py measures quantized models against fp32 ONNX. Results committed under fixtures/benchmark-*.json and summarized in BENCHMARKS.md / BENCHMARKS_1B.md.
Other signals¶
- Smoke translation matrices (
generate_translation_matrix.py) - Live browser benches via browser-lab
- Visual reports (
generate_visual_reports.py,generate_live_reports.py)
Absent¶
- No pytest suite
- No CI job running parity on PRs
- README
test_hf_models.pymissing in this checkout
Interview Q&A¶
Q: How do you know Q4 is “good enough”?
A: Published token-match rates vs fp32 — 1B holds up better than 200M; choose precision by size/accuracy tradeoff tables.