ResearchBenchmarksData Selection on LLM Training Data SelectionData Selection on LLM Training Data SelectionFollow13.3Selection Time (hours)JoBS8.95238.30167.6596.999Feb 9, 2026Evaluation resultsMethodMethodLinksSelection Time (hours)JoBSTraining steps=50000,...Training steps=50000, Hardware=H200 GPU2026.0213.3LESSHardware=H200 GPUHardware=H200 GPU2026.0216.3DoReMiHardware=H200 GPUHardware=H200 GPU2026.0218.5IFHardware=H200 GPUHardware=H200 GPU2026.0252DiversityHardware=H200 GPUHardware=H200 GPU2026.02122