Chunked parallel associative scan in Triton: varlen packing, bidirectional, torch.compile. Drop-in primitive for gated linear RNNs.
-
Updated
Mar 11, 2026 - Python
Chunked parallel associative scan in Triton: varlen packing, bidirectional, torch.compile. Drop-in primitive for gated linear RNNs.
Three-path memory for LLM agents: KV window (exact attention) + retrieval index (archived chunks) + TRN recurrent state (8-96 KB, O(1) per step)
Add a description, image, and links to the linear-recurrence topic page so that developers can more easily learn about it.
To associate your repository with the linear-recurrence topic, visit your repo's landing page and select "manage topics."