你说我高了,或者我说你高了?

#3
by owao - opened

相较于基础模型 Qwen2.5-Instruct,SecGPT 在所有评测指标上均实现实质性超越,反映出我们在数据构建、微调范式、安全任务精调机制上的整体优化成效:

模型版本 CISSP CS-EVAL ↑ CEVAL ↑ GSM8K ↑ BBH ↑
Qwen2.5-1.5B 52.97 71.66 59.91 61.03 43.44
SecGPT-1.5B 71.09 81.53 53.5 57.47 45.17
Qwen2.5-7B 66.30 84.66 74.97 80.36 71.20
SecGPT-7B 78.23 85.12 72.89 76.88 67.08
Qwen2.5-14B 71.09 86.22 68.57 90.03 78.25
SecGPT-14B 77.37 86.12 59.45 88.25 75.90
Your need to confirm your account before you can post a new comment.

Sign up or log in to comment