harukaza

Results 4 comments of harukaza

Cannot create a tensor proto whose content is larger than 2GB

Thanks for your code!

cuDNN error: CUDNN_STATUS_EXECUTION_FAILED

i also meet the problem, so i add two sentences in the train file. _import torch torch.backends.cudnn.enabled = False_

size mismatch for opt_proj.weight: copying a param with shape torch.Size([2560, 768]) from checkpoint, the shape in current model is torch.Size([768, 768]).

I solved the problem by using the local opt_model _opt-2.7b_ in the _blip2_pretrain_opt2.7b.yaml_ , which come from Facebook/opt-2.7b.

size mismatch for opt_proj.weight: copying a param with shape torch.Size([2560, 768]) from checkpoint, the shape in current model is torch.Size([768, 768]).

In fact, I misunderstand the different usages between huggingface and the project page. The former provides a whole ckpt, while the latter need two different module ckpts (i.e., blip2_pretrained_opt2.7b.pth &...