List your company

Biased Overtraining Evaluation on ArxivRollBench

0.22RSIIPhi-1
0.17680.46840.761.0516Jul 25, 2025

Evaluation results

MethodLinks
Phi-12025.070.225.21
Llama3-8BParameters=8B2025.070.452
Phi-1.52025.070.54.02
Qwen2-7BParameters=7B2025.070.511.76
Llama2-7BParameters=7B2025.070.516.2
Phi-22025.070.532.39
Phi-3.5-miniModel size=mini2025.070.573.27
Qwen2.5-72BParameters=72B2025.070.643.84
Qwen2.5-7BParameters=7B2025.070.662.21
Phi-3-miniModel size=mini2025.070.765.84
Yi-1.5-34BParameters=34B2025.070.782.91
Llama3.1-8BParameters=8B2025.070.963.46
Llama3.1-70BParameters=70B2025.071.193.66
Llama-3.1-Nemotron-70BParameters=70B2025.071.33.88