Repository navigation
About 2.5 times faster on one core, with identical scores.
- Tokens are written into one buffer per text and interned without a string per token;
stemming works in place. rouge1counts in a dense array androuge2uses packed bigram keys;rougeLuses
Hyyrö's bit-parallel LCS, with a linear-memory fallback for very large inputs.- The Python binding builds the result dicts in Rust.
- Scorers pickled by 0.1.0 load correctly.
On CNN/DailyMail, XSum and PubMed samples (2,000 pairs each, M1 Pro) scoring is now 30–80× faster than rouge-score on one core and 80–350× with eight threads, and faster than rouge-rust on the ROUGE-1/2/L subset it supports. Details in BENCHMARKS.md.
pip install -U rouge-score-rs