ResearchBenchmarksDownstream Task Evaluation on DownstreamFollow32.43Throughput (tokens/s)FairyFuse7.272413.803720.33526.8663Apr 22, 2026Evaluation ResultsMethodMethodLinksThroughput (tokens/s)Memory Usage (GB)Average AccuracyFairyFuseQuantization=TernaryQuantization=Ternary2026.0432.433.366llama.cppQuantization=Q4_K_MQuantization=Q4_K_M2026.0426.154.165.1llama.cppQuantization=Q2_KQuantization=Q2_K2026.0420.12.856.6FP16Precision=FP16Precision=FP162026.048.2413.567.3