admin管理员组

文章数量:1355703

I'm currently using broadcasting, given a 3D pytorch tensor of shape 32 x 15000 by (128*batch_size) (huge, I know. Cannot shrink this) to generate a histogram for each 32x15000 array. I am storing lowers and uppers for each to easily bin using broadcasting.

Is there a torch.histogram alternative that will let me do this instead? broadcasting consumes huge amounts of memory as the mask ends up being 32 x 15000 x (128batch_sizebins) and I cannot fit this in VRAM.

Any suggestions would be welcome. I'm currently getting around the issue by iteratively handling each bin, but that leads to massive speed losses.

本文标签： pythonPytorch Histogram but only one dimension in 3D tensorStack Overflow

版权声明：本文标题：python - Pytorch Histogram- but only one dimension in 3D tensor - Stack Overflow 内容由网友自发贡献，该文观点仅代表作者本人， 转载请联系作者并注明出处：http://www.betaflare.com/web/1744017323a2576569.html， 本站仅提供信息存储空间服务，不拥有所有权，不承担相关法律责任。如发现本站有涉嫌抄袭侵权/违法违规的内容，一经查实，本站将立刻删除。