List your company

Model Calibration on Fineweb-edu 1.0 (test)

0.002ECERandom Sampling KD (Ours 12)
-0.008120.060190.12850.19681Mar 21, 2025Jun 7, 2025Aug 24, 2025Nov 10, 2025Jan 27, 2026Apr 15, 2026Jul 2, 2026

Evaluation results

MethodLinks
Random Sampling KD (Ours 12)Unique Tokens=122025.030.002
FullKDMethod=FullKD2025.030.002
CEStudent Size=3B, Teacher Size=8B2025.030.003
Random Sampling KD (Ours 12+)Unique Tokens=12, Adaptive Weighting=true2025.030.017
Top-K 50K=502025.030.018
Top-K 12K=122025.030.047
LMSIEval.=cross2026.070.095
N-OPSDEval.=cross2026.070.177
TTRLEval.=cross2026.070.178
N-OPSDEval.=in2026.070.191
Qwen3-4B baseEval.=base2026.070.246
TTRLEval.=in2026.070.248
LMSIEval.=in2026.070.255