Repository navigation

All pull requests

[Fix] Honor fuse_norm in Samba, MoM, and LogLinearMamba2
#1715feiyunzhang opened Nov 11, 2026Contributor3 of 6 tasks
[Perf] Widen bounded KDA intra backward tiles on Hopper
#1714tang-t21 opened Nov 10, 20262 of 6 tasks
[Fix] Reject strided KDA output buffers
#1713bingdongni opened Nov 10, 20263 of 6 tasks
[Perf] Accelerate Ascend stateful convolution with coregrid
#1711ljyrj opened Nov 10, 20263 of 6 tasks
[Ascend] Add linear_attn Triton-Ascend backend
#1710cn-sys opened Nov 10, 20263 of 6 tasks
[Ops] Add Triton-Ascend backend for Parallax
#1709zhudada0120 opened Nov 9, 2026Contributor5 of 6 tasks
1
[Perf] Improve chunk_kda performance on Ascend NPU
#1707Ginray opened Nov 9, 2026Contributor4 of 6 tasks
[Ops] Add decay_max_scan operator (input-dependent decay running-max scan, O(1) state)
#1706dfytensor opened Nov 9, 20265 of 6 tasks
2
[Model] Add PHBA attention and language models
#1704ReyJerry opened Nov 8, 2026Contributor5 of 6 tasks
1
[GDN] Add opt-in ReWY backward
#1703ZhangYi2026 opened Nov 8, 20262 of 6 tasks
[Model] Add SwiLA recurrent attention and language models
#1702jhdlee opened Nov 8, 20265 of 6 tasks
[Ops] Add a Triton kernel for DeepSeek-V4 attention
#1701flyingdoubleG opened Nov 8, 2026Contributor2 of 6 tasks
[Ops] Add a fused lightning indexer for DSA
#1700flyingdoubleG opened Nov 8, 2026Contributor2 of 6 tasks
[Modules] Add zero-centered weights to RMSNorm
#1699flyingdoubleG opened Nov 8, 2026Contributor4 of 6 tasks
[Ops] Add the sLSTM of xLSTM
#1698flyingdoubleG opened Nov 8, 2026Contributor2 of 6 tasks
1
[Ops] Add the selective scan of Mamba
#1697flyingdoubleG opened Nov 8, 2026Contributor3 of 6 tasks
[Ops] Add MiniMax Sparse Attention
#1696flyingdoubleG opened Nov 8, 2026Contributor2 of 6 tasks
[Ops] Add parallel_dsa, the sparse attention of DSA over top-k indices
#1695flyingdoubleG opened Nov 8, 2026Contributor2 of 6 tasks
[Ops] Add naive references for DeepSeek-V4 CSA/HCA attention
#1694flyingdoubleG opened Nov 8, 2026Contributor4 of 6 tasks
[Module] Add mHC residual module
#1693feiyunzhang opened Nov 7, 2026Contributor4 of 6 tasks
5
[Ops] Add block-causal linear attention
#1690flyingdoubleG opened Nov 7, 2026Contributor2 of 6 tasks
[Ops] Add LaCT large-chunk test-time training
#1689flyingdoubleG opened Nov 7, 2026Contributor2 of 6 tasks
[Model] Add the linearized attention of LoLCATs
#1688flyingdoubleG opened Nov 7, 2026Contributor4 of 6 tasks
[Model] Add Griffin and Hawk
#1687flyingdoubleG opened Nov 7, 2026Contributor4 of 6 tasks
[Attention] Support partial rotary embeddings and an output projection bias
#1686flyingdoubleG opened Nov 7, 2026Contributor4 of 6 tasks

Pull requests