Repository navigation

All pull requests

修复因bench超时被kill导致的bench性能结果编造问题
#222yuzichang18 opened Jun 2, 2026Contributor2 of 13 tasks
level4新增LightningAttention算子
#221LHJ1098826475 opened Jun 1, 20262 of 9 tasks
triton_optimization改动并入main分支
#220big-ZWZ opened May 30, 202613 tasks
引入 ops-evo 算子进化优化 Agent 及配套 Skill 框架
#215wangdengkai opened May 28, 2026
Adding Performance Test Scripts
#213longyihai opened May 28, 20262 of 13 tasks
matmul related skills init
#206qzhou2017 opened May 20, 20261 task
meta agent演化发现影响时间开销的问题
#193geyu00 opened May 11, 2026Contributor
[triton] 拆分phase3和phase4为subagent
#192ElleElleWu opened May 11, 2026Contributor13 tasks
511 br_430 合并main
#191chopper0126 opened May 11, 2026Collaborator13 tasks
[triton] 更新精度评测标准，支持单标杆（MERE & MARE & 小值域标准）
#187ElleElleWu opened May 8, 2026Contributor2 of 9 tasks
增加IR分析优化skill
#186JMakit opened May 8, 2026Contributor13 tasks
【请勿合入】Update triton-ascend-coder.md
#175wwwbby opened May 6, 2026Contributor13 tasks
benchmark: 更新 7_MoeGatingTopKSoftmax / 9_TopKTopP 参考实现，增加偏态数据生成
#171serlar opened May 4, 20262 of 13 tasks
docs: refine TileLang scalar constructor guidance
#169wzzll123 opened Apr 30, 2026Contributor2 of 13 tasks
[28_Interpolate] DO NOT MERGE — 72/73 = 98.6% + hard-data proof of why 100% unreachable
#165wabluy opened Apr 29, 2026Contributor
1
add kv rsnorm rope cache archive task
#156justice-dance opened Apr 29, 2026Contributor2 of 13 tasks
add kv rsnorm rope cache archive task
#155justice-dance opened Apr 29, 2026Contributor2 of 13 tasks
level3 conv-family: accept seed to make ref ↔ cand verification possible
#145zhshgmail opened Apr 28, 2026
【triton】subagent拆分
#137jqliu42 opened Apr 28, 2026Contributor13 tasks
fix: L2 op#24 KvCacheUpdateWithRopeBackward cache_position OOB
#119zhshgmail opened Apr 24, 2026
新增生成 A5 agent
#90serlar opened Apr 15, 202611 tasks
[Bug] 修复torch\torch_npu回退问题
#81Just-it opened Apr 13, 2026Owner2 of 13 tasks
【benchmark】新增kernel-triton-verifier skill, 支持与triton-gpu算子进行精度/性能比较
#80big-ZWZ opened Apr 13, 20262 tasks
【benchmark】新增TritonNPUBench，提供vllm/sglang/fbgemm仓triton算子描述
#79big-ZWZ opened Apr 13, 20262 tasks
[docs]新增分核优化skill md文件
#60wwwbby opened Apr 8, 2026Contributor2 of 13 tasks

Navigation Menu

Repository navigation

All pull requests

修复因bench超时被kill导致的bench性能结果编造问题

level4新增LightningAttention算子

triton_optimization改动并入main分支

引入 ops-evo 算子进化优化 Agent 及配套 Skill 框架

Adding Performance Test Scripts

matmul related skills init

meta agent演化发现影响时间开销的问题

[triton] 拆分phase3和phase4为subagent

511 br_430 合并main

[triton] 更新精度评测标准，支持单标杆（MERE & MARE & 小值域标准）

增加IR分析优化skill

【请勿合入】Update triton-ascend-coder.md

benchmark: 更新 7_MoeGatingTopKSoftmax / 9_TopKTopP 参考实现，增加偏态数据生成

docs: refine TileLang scalar constructor guidance

[28_Interpolate] DO NOT MERGE — 72/73 = 98.6% + hard-data proof of why 100% unreachable

add kv rsnorm rope cache archive task

add kv rsnorm rope cache archive task

level3 conv-family: accept `seed` to make ref ↔ cand verification possible

【triton】subagent拆分

fix: L2 op#24 KvCacheUpdateWithRopeBackward cache_position OOB

新增生成 A5 agent

[Bug] 修复torch\torch_npu回退问题

【benchmark】新增kernel-triton-verifier skill, 支持与triton-gpu算子进行精度/性能比较

【benchmark】新增TritonNPUBench，提供vllm/sglang/fbgemm仓triton算子描述

[docs]新增分核优化skill md文件

Pull requests