Very good explanation! Thank you. I also didn't understand the examples regarding the extrapolation rate from the paper, which you presented at 15:17. I can see why they did it, it makes sense to use log. But I don't really understand how they got those numbers.
Really nice presentation! Thank you for time spent!
Main branch output is added with non-norm values, not with normalized values
Very good explanation! Thank you. I also didn't understand the examples regarding the extrapolation rate from the paper, which you presented at 15:17. I can see why they did it, it makes sense to use log. But I don't really understand how they got those numbers.
Yeah the computation doesn’t add up, but it’s a clever way of handling the issue though. Glad you enjoyed it!
🔥🤝🏻
❤
Is it possible to merge depthwise separable convolution with Swin V2 - Base to reduce parameters and to making feasible to deploy into edge divices