ragged-tensor
Manipulate tensors with PackedSequence and CattedSequence
Variable-length tensors with Triton kernels at 1.6× of FlashAttention-2 on H100. NumPy / PyTorch / MLX / JAX backends.