ResearchBenchmarksCompression on 300 MB Monolingual CorporaFollow58.7Train TokensBaseline49.96452.23254.556.768Dec 14, 2025Evaluation ResultsMethodMethodLinksTrain TokensFW TokensBaselinen=16Kn=16K2025.1258.765BPEn=16Kn=16K2025.1258.564FSPn=16Kn=16K2025.1258.264.1Baselinen=32Kn=32K2025.125459.6BPEn=32Kn=32K2025.1253.858.8FSPn=32Kn=32K2025.1253.758.9Baselinen=64Kn=64K2025.1250.655.7BPEn=64Kn=64K2025.1250.454.9FSPn=64Kn=64K2025.1250.355.1