Browse documentation
DocsBenchmarks
Benchmarks
Current parsing and streaming comparisons, presented with enough context to interpret the numbers.
01
TMX materialized catalog
| Framework | Language | Median | Throughput | Peak cold RSS |
|---|---|---|---|---|
| Lokit | Rust / Python | 0.240s | 208,143 units/s | 80.4 MiB |
| Okapi Framework | Java | 0.263s | 189,437 units/s | 269.2 MiB |
| translate-toolkit | Python | 0.652s | 76,586 units/s | 135.7 MiB |
Workload: 50,000 units from a 10.75 MB TMX corpus. These values are retained from the current site while the benchmark suite is rerun.
02
XML streaming comparison
| Workload | Lokit median | Okapi materialize median | Ratio |
|---|---|---|---|
| TMX synchronous stream | 0.196s | 0.263s | 1.34x |
| XLIFF synchronous stream | 0.226s | 0.309s | 1.37x |
Streaming and materialization have different output contracts. Keep that distinction visible when interpreting cross-framework ratios.