DanLing NestedTensor: Composable Multi-Ragged Tensors for Deep Learning
arXiv:2609.30379v1 Announce Type: new Abstract: Variable-size inputs are common in deep learning, but dense batching allocates a shared envelope and spends computation on padding. The cost multiplies across varying axes: an explicit pair state allocates $BN_{\max}^2$ positions instead of $\sum_i N_…
Read the full story at arXiv cs.LG ↗
Timeline · 1 report
- 2026-09-28 04:00 · arXiv cs.LG
DanLing NestedTensor: Composable Multi-Ragged Tensors for Deep Learning