Open wangyePHD opened 2 weeks ago
shape mismatch: value tensor of shape [128, 4096] cannot be broadcast to indexing result of shape [0, 4096] How to solve it?
Could you try installing transformers==4.41.2? There is a bug in newer versions that breaks fine-tuned IDEFICS 2 models, see https://github.com/huggingface/transformers/issues/33763.
transformers==4.41.2
shape mismatch: value tensor of shape [128, 4096] cannot be broadcast to indexing result of shape [0, 4096] How to solve it?