Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
183 commits
Select commit Hold shift + click to select a range
6e7da19
Generalize skill evaluation script paths
wzzll123 Apr 11, 2026
566e64c
Refine AscendC skill references and RMSNorm vector tiles
wzzll123 Apr 12, 2026
a3dbaa1
add build/
chopper0126 Apr 10, 2026
f47fbb7
添加回退校验+迭代循环
chopper0126 Apr 10, 2026
d5e98c9
add batch_run_performance.sh
chopper0126 Apr 13, 2026
dc25ca1
修改input_case 为从json读取
chopper0126 Apr 13, 2026
2a79fc7
调整skills/ascendc/ascendc-translator/references/AscendCVerification.md
wzzll123 Apr 14, 2026
740574c
remove DS_Store
wzzll123 Apr 14, 2026
89395b5
更新rms norm样例,增加返回值,使其更好应对benchmark的样例
wzzll123 Apr 14, 2026
031bec6
Merge pull request #85 from chopper0126/br_ascend_claudecode
Just-it Apr 14, 2026
e6a8a64
Clarify TileLang design-only workflow
wzzll123 Apr 14, 2026
7090630
Merge branch 'br_ascend_optimization' into fix-skill-eval-paths
wzzll123 Apr 14, 2026
6bf6782
Merge pull request #86 from wzzll123/fix-skill-eval-paths
Just-it Apr 14, 2026
d51dbab
优化trace 性能描述+在批跑脚本增加最终报告汇总
chopper0126 Apr 14, 2026
21b9113
feat:兼容单用例多个shape输入的验证和性能测试
201111007-max Apr 15, 2026
57f8713
Merge pull request #91 from 201111007-max/main
Just-it Apr 16, 2026
ed3f39f
feat:支持gpu triton code -> npu triton code ->verify及性能测试;speedup小数位增至4位
201111007-max Apr 16, 2026
7c3dd36
刷新BASELINE_0415.md
chopper0126 Apr 16, 2026
0f06301
修改readme
chopper0126 Apr 16, 2026
e360fd2
feat:确保 Agent 在生成 report.md 时,会将 speedup 和加速比值统一格式化为 4 位小数,与 benchmar…
201111007-max Apr 16, 2026
95c114d
Merge pull request #92 from chopper0126/br_ascend_optimization
Just-it Apr 16, 2026
15cc944
Clamp k and p to valid ranges in TopKTopP benchmark input
Apr 16, 2026
02d3e14
Merge pull request #94 from serlar/br_ascend_opt
Just-it Apr 16, 2026
2210af2
[BugFix]修改generate_report_dynamic.py路径
chopper0126 Apr 17, 2026
4447c6b
Merge pull request #96 from chopper0126/br_ascend_optimization
Just-it Apr 17, 2026
0e4b894
Merge pull request #93 from 201111007-max/main
Just-it Apr 17, 2026
2f2eb3d
Merge pull request #95 from serlar/br_ascend_opt
Just-it Apr 20, 2026
1f010c6
claude思维轨迹以“算子名_状态”重命名
tianyazhang000 Apr 20, 2026
9ba4d18
claude思维轨迹以“算子名_状态”重命名
tianyazhang000 Apr 20, 2026
2ac7425
latency optimizer skill refactoring
jqliu42 Apr 9, 2026
b880346
delete block runing and scalar optim
jqliu42 Apr 9, 2026
db1bd9d
modify tiling optim
jqliu42 Apr 9, 2026
67e27c7
add block size tuning
jqliu42 Apr 9, 2026
d46fc45
update checklist
jqliu42 Apr 10, 2026
9440506
add skills for norm ops
yuzichang18 Apr 10, 2026
0d3ee49
删除重复skill,内容合并
yuzichang18 Apr 10, 2026
1e6bd2d
删除效果较差的skill
yuzichang18 Apr 10, 2026
71cdd4c
add discrete_memory_access skill
Apr 10, 2026
d8338f5
Update discrete_memory_access.md
wwwbby Apr 10, 2026
18f2c20
add vector core partition
Apr 10, 2026
2732969
添加 Scalar 转 Vector 优化模式文档
JMakit Apr 10, 2026
b1fc8a0
merge latency skill ref
jqliu42 Apr 11, 2026
eeb3939
adjust skill.md
jqliu42 Apr 11, 2026
8e233ee
adjust ref4 and ref 8
jqliu42 Apr 11, 2026
723e6bf
rename scalar to vector
jqliu42 Apr 11, 2026
add4732
fix libdevice-usage skill bug
yuzichang18 Apr 15, 2026
6f759b7
merge block_size tuning && autotune
Apr 19, 2026
8fe7b24
update latency optmizer skill
Apr 20, 2026
0176508
[triton] update benchmark
ElleElleWu Apr 20, 2026
9fd7c8a
Merge pull request #100 from ElleElleWu/br_upstream_main
Just-it Apr 20, 2026
7041366
优化汇总表说明
chopper0126 Apr 20, 2026
6de386f
add level430
chopper0126 Apr 20, 2026
eeb4505
Update MoeInitRouting benchmark: fix row_idx generation logic
Apr 21, 2026
67caf8e
Merge pull request #102 from serlar/br_ascend_opt
Just-it Apr 21, 2026
47c518b
Merge pull request #101 from serlar/br_ascend_opt
Just-it Apr 21, 2026
75ab449
Add KV sort operator to archive_tasks
Apr 21, 2026
616ef5f
补充生成时间
tianyazhang000 Apr 21, 2026
a8990df
Merge pull request #98 from tianyazhang000/main
Just-it Apr 21, 2026
853be8f
Merge pull request #103 from chopper0126/br_ascend_optimization
Just-it Apr 21, 2026
41deda7
Merge pull request #99 from jqliu42/br_merge_to_main
Just-it Apr 21, 2026
18c7726
add level30 allcase json
chopper0126 Apr 21, 2026
de51a76
Fix MoE benchmark reference implementations and input generation
Apr 21, 2026
63f40ce
Merge pull request #107 from serlar/br_ascend_opt
Just-it Apr 21, 2026
c9ce8c5
add relubackward and fix realdiv
chopper0126 Apr 22, 2026
f137f93
Merge pull request #110 from chopper0126/br_ascend_optimization
Just-it Apr 22, 2026
6b42fa5
[triton] 添加report功能,新增430格式的benchmark评测
ElleElleWu Apr 22, 2026
0f7be33
add per-case speedup
chopper0126 Apr 22, 2026
2d2b594
Merge remote-tracking branch 'origin/main' into br_ascend_optimization
Just-it Apr 22, 2026
fb0c2f2
用profile方式验证性能,并保存每个case的加速比
chopper0126 Apr 22, 2026
6c0034b
add restrict condition to AscendC generate skill
justice-dance Apr 22, 2026
e396a91
[triton] Phase 4 结果判定以主线最新为准
ElleElleWu Apr 22, 2026
a18a6a1
[triton] 多shape benchmark输出报告和评测方式修正
ElleElleWu Apr 22, 2026
bf7f3e8
Merge pull request #111 from justice-dance/br_ascend_optimization
Just-it Apr 22, 2026
04993f3
修改精度验证方式
chopper0126 Apr 22, 2026
71e4510
[triton] op-task-extractor适配多shape benchmark
ElleElleWu Apr 22, 2026
7bdd379
Merge pull request #113 from ElleElleWu/br_430_2
Just-it Apr 22, 2026
5dbd219
[doc][triton]add avoid scalar lowing skill
Apr 22, 2026
f4d890d
fix elu_grad's model.py
sdshao Apr 22, 2026
8aabab2
Merge pull request #115 from sdshao/br_430
Just-it Apr 23, 2026
6d9f905
Merge pull request #114 from wwwbby/main
Just-it Apr 23, 2026
e9becba
优化utils/verification_ascendc.py和performance.py
chopper0126 Apr 23, 2026
261f55f
保存每个case加速比
chopper0126 Apr 23, 2026
cad9fb8
feat:精度测试参考AscendCOPBenchMark修改。
201111007-max Apr 23, 2026
1e7b459
Merge branch 'Just-it:main' into main
201111007-max Apr 23, 2026
0197c71
fix realdiv's reference code from tf to torch
sdshao Apr 23, 2026
67cbbed
Merge pull request #118 from sdshao/br_430
Just-it Apr 23, 2026
0c89a5c
Merge remote-tracking branch 'origin/main' into br_430
Just-it Apr 23, 2026
dfc3680
feat:精度测试参考AscendCOPBenchMark修改。
201111007-max Apr 23, 2026
8991393
Merge pull request #117 from 201111007-max/main
Just-it Apr 24, 2026
4b42d46
feat:精度测试参考AscendCOPBenchMark修改。
201111007-max Apr 23, 2026
20fcf0f
feat:精度测试参考AscendCOPBenchMark修改。
201111007-max Apr 23, 2026
ce2d203
[benchmark] 修复17_logaddexp用例类型问题
ElleElleWu Apr 24, 2026
40715aa
添加Inf 处理
chopper0126 Apr 24, 2026
3a54f30
[triton] 基线验证和性能评测改为复用基线结果
ElleElleWu Apr 24, 2026
9bf0cb8
Merge pull request #116 from chopper0126/br_ascend_optimization
Just-it Apr 24, 2026
f02dc54
[triton] 新增约束,性能评测之前必须全量shape精度验证通过
ElleElleWu Apr 24, 2026
3269aa1
Merge pull request #122 from ElleElleWu/br_430_2
Just-it Apr 24, 2026
146ccb3
Merge pull request #123 from 201111007-max/br_430
Just-it Apr 25, 2026
ac247af
Merge pull request #124 from ElleElleWu/br_430_workflow
Just-it Apr 25, 2026
8df1790
[triton] benchmark评测方式变更,由延时求和计算加速比更改为几何平均计算加速比
ElleElleWu Apr 25, 2026
56b7d99
Merge pull request #129 from ElleElleWu/br_430_geometric
Just-it Apr 25, 2026
93546ec
修复脚本支持绝对路径
Just-it Apr 25, 2026
f24df03
Merge pull request #131 from Just-it/br_debug
Just-it Apr 25, 2026
7e05a78
add temp
chopper0126 Apr 25, 2026
f50c4cb
修复提前退出
chopper0126 Apr 27, 2026
1f3c8b9
Merge pull request #132 from chopper0126/br_ascend_optimization
Just-it Apr 27, 2026
0267062
fix logaddexp op
jqliu42 Apr 27, 2026
bbc2b8f
Merge pull request #134 from jqliu42/br_fix_logaddexp
Just-it Apr 27, 2026
e4eba40
Merge pull request #135 from JMakit/br_430
Just-it Apr 27, 2026
519efe5
更新部分uint8用例,避免出现负值入参
JMakit Apr 27, 2026
5d670ca
benchmark: 11_DequantSwigluQuant 改用 doc-grounded fp32 CPU 实现作为 ref
Apr 28, 2026
45fdec5
Merge pull request #139 from wabluy/cpu-truth-11-dequant-br430
Just-it Apr 28, 2026
623d1ce
remove fp64 test cases
Apr 28, 2026
6a9b1f2
kv rsnorm rope cache benchmark fix
justice-dance Apr 28, 2026
f65fae7
print remove
justice-dance Apr 28, 2026
d442660
Merge pull request #140 from sdshao/br_430
Just-it Apr 28, 2026
c7fe104
Merge pull request #142 from justice-dance/my_br_430
Just-it Apr 28, 2026
e832545
[triton] 增加sort-select方法,更新NMS标杆,更新torch/python回退检测
ElleElleWu Apr 28, 2026
c3cd64c
fix: handle mask and ref_target_masks inputs as binary tensors in ben…
Apr 28, 2026
d8b90a0
fix: add model.json copy step to Phase 1 and update all skill.md refe…
Apr 29, 2026
6884db1
[docs] 新增 dequant 类算子 AscendC 开发知识库
wabluy Apr 29, 2026
bba6ecd
[ascendc] verification: int8/int16 输出引入 ±1 LSB tolerance
wabluy Apr 29, 2026
8cdfd9f
Merge pull request #144 from serlar/br_430
Just-it Apr 29, 2026
a56c40d
[ascendc] 新增 dynamic quant 算子知识库
ElleElleWu Apr 29, 2026
3fc3ee8
Merge pull request #146 from wabluy/kb/ascendc-dynamic-quant
Just-it Apr 29, 2026
8f7087d
Merge pull request #147 from wabluy/add-dequant-kernel-kb
Just-it Apr 29, 2026
727bbca
Merge pull request #148 from wabluy/allow-int-lsb-tolerance
Just-it Apr 29, 2026
540381a
[ascendc] 对齐CPU标杆,优化cumsum算子精度
jianhuang163 Apr 29, 2026
95c9ed9
Merge pull request #149 from jianhuang163/br_430
Just-it Apr 29, 2026
214fbef
fix verify seed bug
Apr 29, 2026
1c95fe3
fix cumsum gloden && add triton reduce op skill
Apr 29, 2026
5fee914
Merge pull request #152 from wwwbby/br430
Just-it Apr 29, 2026
25f0866
add torch.ones for sum.py
chopper0126 Apr 29, 2026
109db34
Merge pull request #153 from chopper0126/br_ascend_optimization
Just-it Apr 29, 2026
da13c39
Merge pull request #143 from ElleElleWu/br_430_NMS
Just-it Apr 29, 2026
4099de2
fix avgpool3d's mare problem
Apr 29, 2026
d0b05de
Merge pull request #154 from sdshao/br_430
Just-it Apr 29, 2026
d449bf1
[triton]:修复EluBackward算子torch精度问题
201111007-max Apr 29, 2026
f4c247c
Merge branch 'Just-it:br_430' into br_430
201111007-max Apr 29, 2026
1c803a1
Merge pull request #157 from 201111007-max/br_430
Just-it Apr 29, 2026
8155df5
Add proposal: Fix DataCopyPad
geyu00 Apr 29, 2026
7ff28cc
Merge pull request #158 from geyu00/br_430
Just-it Apr 29, 2026
acce0a8
docs: clarify TileLang Ascend vector guidance
wzzll123 Apr 29, 2026
000b35b
Merge pull request #159 from wzzll123/wzz/tilelang-guide-notes
Just-it Apr 29, 2026
eb1e6cf
Update 9_MishBackward.py
wwwbby Apr 29, 2026
f170f5e
Update triton-ascend-elementwise.md
wwwbby Apr 29, 2026
2122975
Merge pull request #161 from wwwbby/br430
Just-it Apr 29, 2026
ea22af7
[triton] Fix l1 28_Interpolate
ElleElleWu Apr 29, 2026
7c37033
Merge pull request #162 from ElleElleWu/br430_28_interpolate
Just-it Apr 29, 2026
d2bc703
修改分核优化skill,取min(实际需要核数,物理核数),避免空核调度开销
JMakit Apr 29, 2026
ecabdb3
Merge pull request #160 from JMakit/br_430
Just-it Apr 29, 2026
789f822
Merge pull request #151 from doux-jy/fix
Just-it Apr 30, 2026
1b8c02f
修改输入
chopper0126 Apr 30, 2026
14a303f
fix: L2-28标杆bfloat16乘加计算前cast成float32
Apr 30, 2026
0c58459
当绝对精度已经小于阈值,判定测试通过
chopper0126 Apr 30, 2026
300f59d
Merge pull request #167 from yiyang-zhu/br_430
Just-it Apr 30, 2026
93c5cf6
threshold * 1e-3
chopper0126 Apr 30, 2026
6871f81
Merge pull request #166 from chopper0126/br_ascend_optimization
Just-it Apr 30, 2026
63854bb
update iou input
Just-it Apr 30, 2026
386c75f
Merge pull request #170 from Just-it/br_iou
Just-it Apr 30, 2026
6cfa518
[triton] 修复CLAUDE_PROJECT_DIR不能自动解析的问题
ElleElleWu May 6, 2026
db254ff
[triton] 修复精度分母过小问题
ElleElleWu May 6, 2026
5b33838
[triton] 修复session.md解析问题
ElleElleWu May 6, 2026
2a69b60
Merge pull request #174 from ElleElleWu/br_430
Just-it May 6, 2026
40fe721
fix(kernel-verifier): 检测通过getattr绕过验证的PyTorch计算操作
May 6, 2026
4c4d363
[benchmark] level430移动到NPU KernelBench,更新BASELINE_latest.md
ElleElleWu May 6, 2026
9600cdd
Merge pull request #176 from ElleElleWu/br_430
Just-it May 6, 2026
d70d8e9
Merge pull request #178 from doux-jy/fix_validation_test
Just-it May 6, 2026
e38c2a0
add subagent
jqliu42 Apr 27, 2026
1fb6a91
modify readme
jqliu42 Apr 27, 2026
158fb25
recove main agent
jqliu42 Apr 27, 2026
3fd7e9f
modify main agent
jqliu42 Apr 27, 2026
d35af42
modify tools
jqliu42 Apr 27, 2026
1a43697
recove gpu mode
jqliu42 Apr 27, 2026
b82db56
recover
jqliu42 Apr 27, 2026
93b86ca
recover main agent
jqliu42 Apr 27, 2026
fbb0ad3
change kernel optim
jqliu42 Apr 27, 2026
1ef11b1
remove unness file
jqliu42 Apr 27, 2026
64eb8c3
add skills
jqliu42 Apr 28, 2026
ffc474c
change exit condition
jqliu42 Apr 30, 2026
3fa11a6
change verify.py
jqliu42 May 8, 2026
5e6a067
[triton] phase3的agent合并成一个,phase4保留原样
ElleElleWu May 9, 2026
2459eea
[triton] verify通过时才允许执行性能评测
ElleElleWu May 11, 2026
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 4 additions & 1 deletion .gitignore
Original file line number Diff line number Diff line change
Expand Up @@ -8,4 +8,7 @@ __pycache__
triton_ascend_output
framework_profile*
.claude
output*
output*
build/
extra-info/
*.tar
Loading