Qi
Qi
> 目前我是用户的qwen-vl-max 不同日期版本分别作为主干模型和感知模型, 感知模型基本完美解决问题, 主干模型10次中会有3-4次坐标幻觉,还有1次逻辑推理错误. 你好，请问您是在模拟器中实现的，还是连接了物理手机？
> > Hi, have you solved this problem? I also meet with similar problem > > I fixed this by using the https://github.com/ZubinGou/math-evaluation-harness, which is one of the foundations of...
> > > > Hi, have you solved this problem? I also meet with similar problem > > > > > > > > > I fixed this by using...
Sorry to disturb you. Did you reproduce the results of Qwen2.5 math **base models** provided by the paper?
> The same problem here. For 7B-instruct, I got 77% on GSM8K with TIR and 95.6% with CoT. Sorry to disturb you. Did you reproduce the results of Qwen2.5 math...
Sorry to disturb you. Did you reproduce the results of Qwen2.5 math base models provided by the paper? I only achieved ~70% acc on Gsm8K dataset, which is largely inconsistent...
请问您找到了对base model的evaluation config 吗