All pull requests

Pull requests

• validate both operands before launching simple matmul

  #288·ZenAlexa opened Nov 14, 2026·

  ·
  1

• preserve matrix-square batch dimensions and validate launch bounds

  #287·ZenAlexa opened Nov 14, 2026·

  ·
  1

• validate csr gather operands before dispatch

  #286·ZenAlexa opened Nov 14, 2026·

  ·
  1

• validate scan inputs before launching fixed-size kernels

  #285·ZenAlexa opened Nov 14, 2026·

  ·
  1

• perf(gdn): optimize fwd syncs

  #284·fiskrt opened Nov 13, 2026Collaborator·

  ·

• feat: extract mega tri method

  #283·fiskrt opened Nov 11, 2026Collaborator·

  ·
  1

• Build manylinux wheels for Ascend 950

  #282·zouzias with CopilotAI opened Nov 10, 2026·

  ·
  

• Update CI CANN images to 9.2.1

  #277·zouzias with CopilotAI opened Nov 9, 2026·

  ·
  

• feat: GDN bwd exploration

  #274·fiskrt opened Nov 4, 2026Collaborator·

  ·

• perf: prune masked KDA output work and compact scratch

  #271·ZenAlexa opened Nov 3, 2026·

  ·
  1

  1

• fix: preserve tensor offsets in kernel launch arguments

  #270·ZenAlexa opened Nov 3, 2026·

  ·
  1

  2

• feat(tri_inv): add mix tri_inv kernel

  #263·zouzias opened Oct 23, 2026Collaborator·

  ·
  1

• refactor(tri_inv_rec_unroll): decouple load/store and compute

  #249·zouzias opened Oct 15, 2026Collaborator·

  ·
  1

• examples: add mxfp4_matmul_a5

  #239·Mocchibird opened Sep 10, 2026Collaborator·

  ·

• examples: Fast Hadamard + MXFP4 Quant fused kernel

  #237·Mocchibird opened Sep 3, 2026Collaborator·

  ·

• examples/ptodsl: add fast_hadamard_vmi

  #230·MirkoDeVita98 opened Aug 18, 2026·

  ·

• feat: basic skill for a2a3-a5 port

  #222·fiskrt opened Aug 4, 2026Collaborator·

  ·
  2

• fix(causal_conv1d): make K a dynamic parameter

  #212·zouzias opened Jul 3, 2026Collaborator·

  ·

• Causal conv1d into wheel

  #208·raphael-s-steiner opened Jun 30, 2026Collaborator·

  ·

• feature(causal_conv1d): import kernel from jit_cpp

  GDN
  #195·zouzias opened Jun 27, 2026Collaborator·

  ·
  1

• Add paged attention highperf JIT example

  #193·MirkoDeVita98 opened Jun 26, 2026·

  ·
  1

• Update the skills.md for accuracy guidelines

  #175·asobczyk opened Jun 4, 2026Collaborator·

  ·

• Idealized C-V data exchange kernels for A5

  A5
  #174·learning-chip opened Jun 1, 2026Collaborator·

  ·
  

• Add Ascend950 pure-vector simulator examples for SiLU and SwiGLU.

  A5
  #172·learning-chip opened May 27, 2026Collaborator·

  ·
  4

• Run abs kernel on A5 cannsim

  A5
  #167·zouzias opened May 15, 2026Collaborator·

  ·
  1

  2