All pull requests

Pull requests

• feat(ctcloss): add single-PE and four-PE forward solution

  #263·wangyuascend-spec opened Oct 24, 2026Collaborator·

  ·

• QBMM HIF4性能优化

  #260·Simona787 opened Oct 22, 2026Contributor·

  ·
  1

• test(transpose): 均匀常量数组 codegen 错误最小复现（附预编译 ELF + 复现命令）

  #256·CYR-Firework opened Oct 20, 2026Contributor·

  ·
  3

• repro(qsmla): v3 CSA crashes on any multi-work (kWorkCount>1) case

  #255·CYR-Firework opened Oct 20, 2026Contributor·

  ·

• Prefer fused masked PTO element loops and preserve four-PE QLI execution

  #252·zhoubot opened Oct 19, 2026Contributor·

  ·

• feat(concat): add two-input tiled concat with exact validation

  #241·robot-inteligence opened Oct 14, 2026·

  ·
  3

• qsmla: v2/v3 性能最优形态——prebuilt ELF 直接运行 + daily/release 集成

  #226·CYR-Firework opened Oct 12, 2026Contributor·

  ·
  5

• feat(qsmla): v3 4-PE CSA kernel (ops-arch35 flow port) with prebuilt csa_small ELF

  #218·CYR-Firework opened Oct 10, 2026Contributor·

  ·
  3

• qsmla: migrate Vec tiles to CUBE_M32 and cube accumulators to CubeAccumulatorM32 (PTO-ISA #291/#311)

  #185·CYR-Firework opened Sep 22, 2026Contributor·

  ·
  1

• feat(hi_f4): dynamic_hi_f4_quant 数值验证 + recip/e6m2 修正 + global_iterator stride

  #178·ziyang-cheng opened Sep 21, 2026Contributor·

  ·

• qli: port TopK radix to PTO v0.58.6 (THISTOGRAM retired)

  #162·gagahamburger opened Sep 18, 2026Contributor·

  ·

• feat(dmxq): 迁移 dynamic_mx_quant 算子套件到 ops-20260904 + 新 multi_thread 布局

  #111·ziyang-cheng opened Sep 5, 2026Contributor·

  ·
  31

• add dynamic_mx_quant test kernel on tag ops-20260828

  #102·ziyang-cheng opened Sep 2, 2026Contributor·

  ·
  36

• test: publish 0xE9 cross-model microbenchmark results

  #100·zhoubot opened Aug 31, 2026Contributor·

  ·
  2

• feat(qli): add radix-select TopK demo with validated cases and known-issue notes

  #78·zzzyh222 opened Aug 24, 2026Contributor·

  ·
  43

• test(tlsu): 端到端 memory op 压测用例（S/C/R 三组）

  #59·liujiang833 opened Aug 14, 2026Collaborator·

  ·
  2