benchmarks
Benchmark suite results
Token reduction
54.12%
balanced mode, benchmark suite
Character reduction
58.23%
balanced mode, benchmark suite
Line reduction
66.67%
fixed benchmark fixtures
Warning preservation
100%
heuristic checks; review critical output
The repository benchmark uses fixed baseline and Noyap-mode responses. Token counts use gpt-tokenizer. Required technical terms, warnings, Thai quality, and English quality are checked heuristically, so flagged or critical cases still deserve human review.
measured
What the benchmark measures
✔ token reduction
✔ character reduction
✔ line reduction
✔ meaning preservation
✔ warning preservation
✔ required technical terms
✔ Thai and English quality checks
✔ manual review checklist
repo files
Benchmark artifacts
github
summary.md https://github.com/ppwnr88/noyap/blob/main/benchmarks/results/summary.md
results.json https://github.com/ppwnr88/noyap/blob/main/benchmarks/results/results.json
cases.json https://github.com/ppwnr88/noyap/blob/main/benchmarks/cases.json
methodology https://github.com/ppwnr88/noyap/blob/main/benchmarks/README.md