Regarding Figures 5 and 6 and Table 4

[neerajgangwar]

Hi @ahmed-agiza,

I have a couple of queries and was hoping you could clarify them.

1. The MTLoRA performance with Swin-Base ImageNet-22k in Figures 5 and 6 appears to be inconsistent. With rank 32, Figure 5 shows $$\Delta m$$ as 5%, whereas Figure 6 appears to be close to 7%. I was wondering if I am perhaps not interpreting the figures correctly?

2. If you have access to it, could you possibly share the code for the Pyramid Vision Transformer (Table 4)? Or maybe the hyperparameters you used for fine-tuning the model?

Thanks!