Unsuccessfully concatenating prefix__projection and text embedding

[qsuzer]

I have successed running the parse_coco.py file and the output of clip was stored in my data files. Then I tried to run the gpt2 fine-tune program. An error was reported as below: It seemed that the error happened in the 234th line when I concatenated the prefix projection and the embedding text. I printed the shape of those tensors below: prefix-----------------(40,1,512,512) self.clip_project(prefix)-------------------(40,1,512,7680) prefix_projection-----------(20480,10,768) embedding_text---------------------(40,41,768) I thought the error was that the shape of prefix_projection should be (batch_size,prefix_length,embedding_size). I wondered what shape of prefix should be. Thanks