KUANWB

Results 4 issues of KUANWB

We have encountered some problems while trying to do the inference via two NVIDIA A10 GPUs. We want to know how to deploy the model on two GPUs, we can...

您好,请问value model的初始权重就是reward model的权重吗?value model是不是只需要加载完权重后把最后的投影层在每个tokens上都投影成一个标量就可以了?谢谢