All pull requests

Pull requests

• issue/1569 - 为 NVIDIA 后端增加可复用的 GGUF 线性层与 BlockFP8 支持

  #1570·xindongliu594 opened Sep 20, 2026·

  ·

• feat: add FP8 blockwise dequantize/GEMM operators and FP8(E4M3) KV cache support in paged attention

  #1568·shsaihdsaiudh opened Sep 20, 2026·

  ·

• feat: add fused rms_norm_rope operator (per-head RMSNorm + RoPE; CPU, NVIDIA)

  #1567·shsaihdsaiudh opened Sep 20, 2026·

  ·

• issue/1565 fix(runtime): support Qwen MTP and recurrent graph replay

  #1566·big-hip opened Sep 19, 2026·

  ·

• feat(infiniop): add lightning_attention op (CPU + NVIDIA) for MiniMax

  #1564·y258dd opened Sep 19, 2026·

  ·

• issue/1571 feat(mamba2): add NVIDIA and MetaX inference operators

  #1562·big-hip opened Sep 17, 2026·

  ·

• perf(ascend): vectorize RoPE AscendC kernel

  #1559·porcelina62 opened Sep 16, 2026·

  ·

• fix(metax): detect Flash Attention varlen ABI from linked library

  #1558·big-hip opened Sep 16, 2026·

  ·

• 优化 MetaX 平台 HD64 GQA Paged Attention Decode 的 K/V 数据复用

  #1557·Qzong2020 opened Sep 14, 2026·

  ·

• fix(metax): support flash-attn 2.6.3 forward ABI

  #1555·zxyy-bys opened Sep 8, 2026Contributor·

  ·
  1

• issue/1553 - fix: let the default runtime honour INFINICORE_DEFAULT_DEVICE_INDEX

  #1554·JoeZhang-0x000 opened Sep 7, 2026·

  ·
  1

• feat: add reusable GGUF linear backend and FP8 support

  #1545·xindongliu594 opened Sep 3, 2026·

  ·

• feat: support cross-device copy_from

  #1532·PanZezhong1725 opened Aug 25, 2026Collaborator·

  ·

• feat: support strided bilinear upsampling

  #1531·PanZezhong1725 opened Aug 25, 2026Collaborator·

  ·

• feat: add timestep embedding operator

  #1530·PanZezhong1725 opened Aug 25, 2026Collaborator·

  ·

• issue/1519 - fix Ascend paged attention block index name collision

  #1520·ETOwang opened Aug 22, 2026·

  ·
  1

• Pr 1474

  #1504·wooway777 opened Aug 14, 2026Collaborator·

  ·

• feat: add aclnnMatmulAllReduce fusion in InfiniCore for Ascend RowParallelLinear

  #1485·ShaneWoof opened Aug 11, 2026Contributor·

  ·

• feat(hygon): add Qwen3-235B-A3B BF16/W8A8 inference support

  #1474·qinyiqun opened Aug 7, 2026Collaborator·

  ·

• fix: graph of ascend

  #1461·Jingbo-gao opened Aug 4, 2026Contributor·

  ·
  1

  4

• feat(runtime): add explicit asynchronous tensor copy support

  #1457·qinyiqun opened Aug 3, 2026Collaborator·

  ·
  4

• feat: adapt for Qwen3.6 moe inference

  #1451·qinyiqun opened Jul 31, 2026Collaborator·

  ·

• fix(moe): initialize graph decode routes under EP

  #1430·PanZezhong1725 opened Jul 28, 2026Collaborator·

  ·

• Pepe/glm5 2

  #1429·wooway777 opened Jul 28, 2026Collaborator·

  ·

• feat: ascend qwen 3.5 ops

  #1423·wooway777 opened Jul 27, 2026Collaborator·

  ·