purpose behind code change

[GeorgeBrowncb]

Hi! Thank you for your code!

I noticed that this repo is heavily based on DualPrompt (https://github.com/JH-LEE-KR/dualprompt-pytorch).

At the same time, I found in JH-LEE-KR/dualprompt-pytorch#8 that there is an error in tensor operations.

After reading your code, I noticed that you modified this part differently from the original repo. Could you explain the purpose behind this change?

ConvPrompt/prompt.py

Lines 257 to 263 in 73ed205

	batched_prompt_raw = self.prompt[layer_num].unsqueeze(0).repeat(x_embed.shape[0], 1, 1) # B, length, C
	dual = 2
	batch_size, length, embed_dim = batched_prompt_raw.shape
	# print("Batched Prompt Shape = ", batched_prompt_raw.shape)
	batched_prompt = batched_prompt_raw.reshape(
	batch_size, dual, length // dual, self.num_heads, embed_dim // self.num_heads
	)