Vindicator645
Vindicator645
好像是的，我之前debug的时候设成了只有一条数据，恢复成全部数据消除了这问题
Hi, i seem to have the same issue and i am on a fresh windows install with no special firewall settings, which system are u using?
I am using a proxy called clash for windows, set as system proxy, and chrome is on the same machine as narr. maybe i should try to put the proxy...
I suspect the loss = loss / gradient_accumulation_steps and acc = acc / gradient_accumulation_steps should be removed in deepspeed_utils