You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
File "/root/nas/PromptCBLUE/./src/ft_chatglm_ptuning/modeling_chatglm.py", line 445, in forward
mixed_raw_layer = self.query_key_value(hidden_states)
File "/opt/conda/envs/lmeval/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "/opt/conda/envs/lmeval/lib/python3.9/site-packages/peft/tuners/lora.py", line 350, in forward
result += self.lora_B(self.lora_A(self.lora_dropout(x))) * self.scaling
File "/opt/conda/envs/lmeval/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "/opt/conda/envs/lmeval/lib/python3.9/site-packages/torch/nn/modules/linear.py", line 114, in forward
return F.linear(input, self.weight, self.bias)
RuntimeError: expected scalar type Half but found Float
报这个错是什么原因,怎么改正
The text was updated successfully, but these errors were encountered:
File "/root/nas/PromptCBLUE/./src/ft_chatglm_ptuning/modeling_chatglm.py", line 445, in forward
mixed_raw_layer = self.query_key_value(hidden_states)
File "/opt/conda/envs/lmeval/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "/opt/conda/envs/lmeval/lib/python3.9/site-packages/peft/tuners/lora.py", line 350, in forward
result += self.lora_B(self.lora_A(self.lora_dropout(x))) * self.scaling
File "/opt/conda/envs/lmeval/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "/opt/conda/envs/lmeval/lib/python3.9/site-packages/torch/nn/modules/linear.py", line 114, in forward
return F.linear(input, self.weight, self.bias)
RuntimeError: expected scalar type Half but found Float
报这个错是什么原因,怎么改正
The text was updated successfully, but these errors were encountered: