已合并
docs: 【直播作业】填写 YOLOv13 最优多流配置(#516) #4429
Wjieee创建于 1 天前
docs: 【直播作业】填写 YOLOv13 最优多流配置(#516) #4429
已合并
Wjieee创建于 1 天前
Wjieee
1 天前
## 基本信息 - **所选 Example**:YOLOv13 - **CANNLab 环境**:Ubuntu 20.04 + aarch64 - **CANN 版本**:社区版 9.2.0 (Weekly) - **芯片规格**:Ascend910 (NPU Name: 9362),soc_version=Ascend910_9362 - **关联 Issue**:[#516]https://gitcode.com/cann/ge/issues/516 ## 测试命令 ### ATC 编译命令(以最优配置为例) ```bash atc --model=yolov13n.onnx \ --framework=5 \ --output=./model/yolov13_LoadBalance_55 \ --soc_version=Ascend910_9362 \ --input_shape="images:1,3,640,640" \ --output_type=FP32 \ --multi_stream_parallel_mode=LoadBalance:55 ``` ### 推理测试命令 ```bash ./out/yolov13_main \ --model=./model/yolov13_LoadBalance_55.om \ --input=./data/bus.bin \ --warmup-runs=10 \ --runs=90 ``` ## 固定参数 - ONNX 模型:yolov13n.onnx(input_shape="images:1,3,640,640") - 输入数据:bus.bin - soc_version:Ascend910_9362 - --warmup-runs=10 - --runs=90 - --multi_stream_parallel_mode 为唯一寻优变量 ## 测试结果 | 配置角色 | 多流配置 | 主评价指标平均耗时(ms) | 正式统计任务数 | |:---|:---|:---|:---| | Base 配置 | 未配置多流参数 | 2.631 | 90 | | 候选配置 | cv | 2.568 | 90 | | 候选配置 | LoadBalance:1 | 2.797 | 90 | | 候选配置 | LoadBalance:2 | 2.613 | 90 | | 候选配置 | LoadBalance:3 | 2.386 | 90 | | 候选配置 | LoadBalance:4 | 2.472 | 90 | | 候选配置 | LoadBalance:5 | 2.281 | 90 | | 候选配置 | LoadBalance:6 | 2.291 | 90 | | 候选配置 | LoadBalance:7 | 2.334 | 90 | | 候选配置 | LoadBalance:8 | 2.270 | 90 | | 候选配置 | LoadBalance:9 | 2.422 | 90 | | 候选配置 | LoadBalance:10 | 2.308 | 90 | | 候选配置 | LoadBalance:11 | 2.421 | 90 | | 候选配置 | LoadBalance:12 | 2.450 | 90 | | 候选配置 | LoadBalance:13 | 2.462 | 90 | | 候选配置 | LoadBalance:14 | 2.277 | 90 | | 候选配置 | LoadBalance:15 | 2.363 | 90 | | 候选配置 | LoadBalance:16 | 2.412 | 90 | | 候选配置 | LoadBalance:17 | 2.281 | 90 | | 候选配置 | LoadBalance:18 | 2.302 | 90 | | 候选配置 | LoadBalance:19 | 2.289 | 90 | | 候选配置 | LoadBalance:20 | 2.428 | 90 | | 候选配置 | LoadBalance:21 | 2.283 | 90 | | 候选配置 | LoadBalance:22 | 2.489 | 90 | | 候选配置 | LoadBalance:23 | 2.285 | 90 | | 候选配置 | LoadBalance:24 | 2.269 | 90 | | 候选配置 | LoadBalance:25 | 2.305 | 90 | | 候选配置 | LoadBalance:26 | 2.277 | 90 | | 候选配置 | LoadBalance:27 | 2.325 | 90 | | 候选配置 | LoadBalance:28 | 2.442 | 90 | | 候选配置 | LoadBalance:29 | 2.416 | 90 | | 候选配置 | LoadBalance:30 | 2.263 | 90 | | 候选配置 | LoadBalance:31 | 2.290 | 90 | | 候选配置 | LoadBalance:32 | 2.423 | 90 | | 候选配置 | LoadBalance:33 | 2.438 | 90 | | 候选配置 | LoadBalance:34 | 2.423 | 90 | | 候选配置 | LoadBalance:35 | 2.419 | 90 | | 候选配置 | LoadBalance:36 | 2.418 | 90 | | 候选配置 | LoadBalance:37 | 2.277 | 90 | | 候选配置 | LoadBalance:38 | 2.270 | 90 | | 候选配置 | LoadBalance:39 | 2.271 | 90 | | 候选配置 | LoadBalance:40 | 2.289 | 90 | | 候选配置 | LoadBalance:41 | 2.271 | 90 | | 候选配置 | LoadBalance:42 | 2.289 | 90 | | 候选配置 | LoadBalance:43 | 2.441 | 90 | | 候选配置 | LoadBalance:44 | 2.414 | 90 | | 候选配置 | LoadBalance:45 | 2.424 | 90 | | 候选配置 | LoadBalance:46 | 2.269 | 90 | | 候选配置 | LoadBalance:47 | 2.274 | 90 | | 候选配置 | LoadBalance:48 | 2.272 | 90 | | 候选配置 | LoadBalance:49 | 2.301 | 90 | | 候选配置 | LoadBalance:50 | 2.275 | 90 | | 候选配置 | LoadBalance:51 | 2.270 | 90 | | 候选配置 | LoadBalance:52 | 2.438 | 90 | | 候选配置 | LoadBalance:53 | 2.413 | 90 | | 候选配置 | LoadBalance:54 | 2.429 | 90 | | **最优配置** | **LoadBalance:55** | **2.261** | **90** | | 候选配置 | LoadBalance:56 | 2.267 | 90 | | 候选配置 | LoadBalance:57 | 2.295 | 90 | | 候选配置 | LoadBalance:58 | 2.287 | 90 | | 候选配置 | LoadBalance:59 | 2.279 | 90 | | 候选配置 | LoadBalance:60 | 2.277 | 90 | | 候选配置 | LoadBalance:61 | 2.445 | 90 | | 候选配置 | LoadBalance:62 | 2.421 | 90 | | 候选配置 | LoadBalance:63 | 2.428 | 90 | | 候选配置 | LoadBalance:64 | 2.272 | 90 | | 候选配置 | MainStream:1 | 2.644 | 90 | | 候选配置 | MainStream:2 | 2.741 | 90 | | 候选配置 | MainStream:3 | 2.662 | 90 | | 候选配置 | MainStream:4 | 2.489 | 90 | | 候选配置 | MainStream:5 | 2.651 | 90 | | 候选配置 | MainStream:6 | 2.613 | 90 | | 候选配置 | MainStream:7 | 2.622 | 90 | | 候选配置 | MainStream:8 | 2.618 | 90 | | 候选配置 | MainStream:9 | 2.453 | 90 | | 候选配置 | MainStream:10 | 2.472 | 90 | | 候选配置 | MainStream:11 | 2.628 | 90 | | 候选配置 | MainStream:12 | 2.625 | 90 | | 候选配置 | MainStream:13 | 2.491 | 90 | | 候选配置 | MainStream:14 | 2.631 | 90 | | 候选配置 | MainStream:15 | 2.466 | 90 | | 候选配置 | MainStream:16 | 2.480 | 90 | | 候选配置 | MainStream:17 | 2.471 | 90 | | 候选配置 | MainStream:18 | 2.483 | 90 | | 候选配置 | MainStream:19 | 2.469 | 90 | | 候选配置 | MainStream:20 | 2.480 | 90 | | 候选配置 | MainStream:21 | 2.505 | 90 | | 候选配置 | MainStream:22 | 2.644 | 90 | | 候选配置 | MainStream:23 | 2.488 | 90 | | 候选配置 | MainStream:24 | 2.643 | 90 | | 候选配置 | MainStream:25 | 2.617 | 90 | | 候选配置 | MainStream:26 | 2.618 | 90 | | 候选配置 | MainStream:27 | 2.612 | 90 | | 候选配置 | MainStream:28 | 2.461 | 90 | | 候选配置 | MainStream:29 | 2.464 | 90 | | 候选配置 | MainStream:30 | 2.623 | 90 | | 候选配置 | MainStream:31 | 2.618 | 90 | | 候选配置 | MainStream:32 | 2.488 | 90 | | 候选配置 | MainStream:33 | 2.625 | 90 | | 候选配置 | MainStream:34 | 2.468 | 90 | | 候选配置 | MainStream:35 | 2.473 | 90 | | 候选配置 | MainStream:36 | 2.475 | 90 | | 候选配置 | MainStream:37 | 2.478 | 90 | | 候选配置 | MainStream:38 | 2.474 | 90 | | 候选配置 | MainStream:39 | 2.479 | 90 | | 候选配置 | MainStream:40 | 2.502 | 90 | | 候选配置 | MainStream:41 | 2.644 | 90 | | 候选配置 | MainStream:42 | 2.494 | 90 | | 候选配置 | MainStream:43 | 2.650 | 90 | | 候选配置 | MainStream:44 | 2.615 | 90 | | 候选配置 | MainStream:45 | 2.623 | 90 | | 候选配置 | MainStream:46 | 2.619 | 90 | | 候选配置 | MainStream:47 | 2.454 | 90 | | 候选配置 | MainStream:48 | 2.471 | 90 | | 候选配置 | MainStream:49 | 2.629 | 90 | | 候选配置 | MainStream:50 | 2.625 | 90 | | 候选配置 | MainStream:51 | 2.493 | 90 | | 候选配置 | MainStream:52 | 2.630 | 90 | | 候选配置 | MainStream:53 | 2.467 | 90 | | 候选配置 | MainStream:54 | 2.480 | 90 | | 候选配置 | MainStream:55 | 2.471 | 90 | | 候选配置 | MainStream:56 | 2.484 | 90 | | 候选配置 | MainStream:57 | 2.470 | 90 | | 候选配置 | MainStream:58 | 2.481 | 90 | | 候选配置 | MainStream:59 | 2.505 | 90 | | 候选配置 | MainStream:60 | 2.638 | 90 | | 候选配置 | MainStream:61 | 2.489 | 90 | | 候选配置 | MainStream:62 | 2.641 | 90 | | 候选配置 | MainStream:63 | 2.618 | 90 | | 候选配置 | MainStream:64 | 2.616 | 90 | ## 最优配置 - **最优配置**:`--multi_stream_parallel_mode=LoadBalance:55` - **Base 配置平均耗时**:2.631 ms - **最优配置平均耗时**:2.261 ms - **性能提升**:约 14.1% ## 关键日志 ### Base 配置日志 ``` [INFO] load model /mnt/workspace/projects/ge/examples/acl/5_sample_yolov13/model/yolov13.om success. [INFO] BENCHMARK aclmdlExecute run=0 latency_us=2625 [INFO] BENCHMARK aclmdlExecute run=1 latency_us=2630 [INFO] BENCHMARK aclmdlExecute run=2 latency_us=2633 [INFO] BENCHMARK aclmdlExecute run=3 latency_us=2632 [INFO] BENCHMARK aclmdlExecute run=87 latency_us=2632 [INFO] BENCHMARK aclmdlExecute run=88 latency_us=2628 [INFO] BENCHMARK aclmdlExecute run=89 latency_us=2631 [INFO] Average aclmdlExecute latency: 2.631 ms [INFO] YOLOv13 SAMPLE PASSED. ``` ### 最优配置 LoadBalance:55 日志 ``` [INFO] load model /mnt/workspace/projects/ge/examples/acl/5_sample_yolov13/model/yolov13_LoadBalance_55.om success. [INFO] BENCHMARK aclmdlExecute run=0 latency_us=2269 [INFO] BENCHMARK aclmdlExecute run=1 latency_us=2267 [INFO] BENCHMARK aclmdlExecute run=2 latency_us=2266 [INFO] BENCHMARK aclmdlExecute run=3 latency_us=2258 [INFO] BENCHMARK aclmdlExecute run=87 latency_us=2257 [INFO] BENCHMARK aclmdlExecute run=88 latency_us=2280 [INFO] BENCHMARK aclmdlExecute run=89 latency_us=2285 [INFO] Average aclmdlExecute latency: 2.261 ms [INFO] YOLOv13 SAMPLE PASSED. ```
likedislike
Pull Request已成功合入, 合并人@CANN-robot
(感谢 Wjieee 的贡献)
WWjieee
1 天前 添加了label:enhancement
WWjieee
1 天前 创建了 pull request,commit 5c71775f
WWjieee
1 天前 关联了issue:[直播有奖作业] 寻找 YOLOv13 和 Recommendation 样例的最优多流配置,赢取Huawei FreeBuds6耳机
atomgit-bot
atomgit-bot
1 天前 评论:

变更摘要

该 PR 为文档类改动(docs),基于关联 Issue #516 的直播作业要求,为 YOLOv13 示例填写并记录其最优多流配置。作者在 CANN 社区版 9.2.0、Ascend910_9362(aarch64 Ubuntu 20.04)环境下,以 --multi_stream_parallel_mode 为唯一寻优变量,对 cvLoadBalance:1~64MainStream:1~64 等配置逐一测试(--warmup-runs=10--runs=90),最终确认 --multi_stream_parallel_mode=LoadBalance:55 为最优配置:平均耗时由 Base 配置的 2.631 ms 降至 2.261 ms,性能提升约 14.1%。本次改动即为在 YOLOv13 相关文档中填入上述最优配置及其测试结果(共改动 2 个文件,新增/删除各 2 行)。

主要改动

  • 填写最优多流配置: 在 YOLOv13 示例文档中补充最优 ATC 编译参数 --multi_stream_parallel_mode=LoadBalance:55,作为该示例在多流场景下的推荐配置。
  • 记录性能对比结果: 文档中写入 Base 配置(未配置多流参数,平均耗时 2.631 ms)与最优配置 LoadBalance:55(平均耗时 2.261 ms)的对比,明确约 14.1% 的性能提升。
  • 补充测试环境与命令: 记录寻优所用的固定参数(input_shape="images:1,3,640,640"soc_version=Ascend910_9362 等)、ATC 编译命令及推理测试命令,为复现该最优配置提供依据。
likedislike
CANN-robot
CANN-robot成员
1 天前 评论:

Hi @Wjieee, welcome to submitting your first PR to ge!

PR Merge Steps

1. CLA Signing

If the current PR label includes cann-cla/yes, it means you have signed the CLA and can proceed to the next step. If the label includes cann-cla/no, please sign the CLA first. If you have any questions, please refer to the FAQ.

2. CI Check

Please comment /compile to trigger the CI pipeline check. If the CI run is successful, the PR will be tagged with ci-pipeline-passed and you can proceed to the next step. If the CI run fails, the PR will be tagged with ci-pipeline-failed, please check the CI logs to fix the issues in the PR. If you have any questions, please refer to the FAQ.

3. Code Review

After CI passes, please refer to the PR Approval Progress and proactively @ the committers in the table to review the code. After approval, committers will comment /lgtm and /approve. Once the lgtm and approved labels are successfully added, the PR will be merged automatically.

likedislike
atomgit-bot
atomgit-bot
1 天前 评论:

🤖 本次改动均为数据 / 生成 / 二进制文件(如测试用例、锁文件、媒体资源),没有可审查的代码逻辑,已跳过 AI 代码审查。

likedislike
CANN-robotCANN-robot成员
1 天前 添加了label:cann-cla/no
CANN-robot
CANN-robot成员
1 天前 评论:

Thanks for your pull-request.
The full list of commands accepted by me can be found at here.
You can get sig-info at here.
For more, you also can visit HICANN.


PR Approval Progress

Congratulations! All modules have met the lgtm and approve requirements.

Module Approval Details

module lgtm status approve status
repo-cann/ge 张德鹏, zhanj, Mexyy (3/2) 张德鹏 (1/1)

💡 Tip:

  • Committer can comment /approve or /lgtm
  • Commenting /approve implies both code review (lgtm) and intent to merge (approve)

CLA Signature Pass

Wjieee, thanks for your pull request. All authors of the commits have signed the CLA. 👍

likedislike
CANN-robotCANN-robot成员
1 天前 将yangyongqiang0606,kobemini,ClarkXie,tangqunzhang,xchu42,hugo111,lining23666,fu-jun2,depeng1994,wqtshg_wt,peiyang,shengnan666,zhangfan_hanq,zhanj,zhujingjing,Mexyy,wangxiaotian995,stevenaw0,yuht9设为评审人
CANN-robotCANN-robot成员
1 天前 将yangyongqiang0606,kobemini,ClarkXie,tangqunzhang,xchu42,hugo111,lining23666,fu-jun2,depeng1994,wqtshg_wt,peiyang,shengnan666,zhangfan_hanq,zhanj,zhujingjing,Mexyy,wangxiaotian995,stevenaw0,yuht9设为审查人
Wjieee
1 天前 评论:

/check-cla

likedislike
CANN-robotCANN-robot成员
1 天前 删除了label:cann-cla/no
CANN-robotCANN-robot成员
1 天前 添加了label:cann-cla/yes
CANN-robot
CANN-robot成员
1 天前 评论:

CLA检查已通过,详情可参考这里

likedislike
Wjieee
1 天前 评论:

compile

likedislike
WWjieee
1 天前 update merge request[project id: 8824148, iid: 4429, commit_id: 1fde1d4cfc36a9d58c9bd0970d041f8969d8e8a5] virtual merging success
CANN-robotCANN-robot成员
1 天前 添加了label:ci-pipeline-running
CANN-robot
CANN-robot成员
1 天前 评论:

流水线任务触发成功
任务链接 [2089ccfc755349f181987ef99fe7e2bd][流水线指导]

任务名称状态日志下载链接
pre_comment ✅ SUCCESS >>>>>
Compile_ARM_compiler ✅ SUCCESS >>>>> >>>>>
Compile_ARM_executor ✅ SUCCESS >>>>> >>>>>
Compile_ARM_dflow ✅ SUCCESS >>>>> >>>>>
API_Check ✅ SUCCESS >>>>>
PreSmoke_A900_npupool ✅ SUCCESS >>>>>

[2026-08-18 09:43:03]    CI执行结束

likedislike
CANN-robot
CANN-robot成员
1 天前 评论:

流水线任务触发成功
任务链接 [0bd42d4134234bce983c4538775dc551][流水线指导]

任务名称状态日志下载链接
pre_comment ✅ SUCCESS >>>>>
Compile_X86_compiler ✅ SUCCESS >>>>> >>>>>
Compile_X86_executor ✅ SUCCESS >>>>> >>>>>
Compile_X86_dflow ✅ SUCCESS >>>>> >>>>>
UT_Test_dflow ✅ SUCCESS
UT_Test_ge_common ✅ SUCCESS
UT_Test_parser ✅ SUCCESS
UT_Test_python ✅ SUCCESS >>>>>
UT_Test_rt ✅ SUCCESS
UT_Test_acl ✅ SUCCESS
UT_Test_autofuse ✅ SUCCESS
UT_Test_executor ✅ SUCCESS
UT_Test_autofuse_ascendc_api ✅ SUCCESS
ST_Test_dflow ✅ SUCCESS
ST_Test_ge_common ✅ SUCCESS
ST_Test_parser ✅ SUCCESS
ST_Test_python ✅ SUCCESS >>>>>
ST_Test_rt ✅ SUCCESS
ST_Test_autofuse ✅ SUCCESS
ST_Test_executor ✅ SUCCESS
ST_Test_hetero ✅ SUCCESS
ST_Test_autofuse_ascendc_api ✅ SUCCESS
ST_Test_autofuse_e2e ✅ SUCCESS
UT_Test_fe ✅ SUCCESS
UT_Test_tefusion ✅ SUCCESS
ST_Test_fe ✅ SUCCESS
ST_Test_tefusion ✅ SUCCESS
ST_Test_ge_common_atc ✅ SUCCESS
UT_Test_Report_graphengine ✅ SUCCESS >>>>>
UT_Test_Report_ge_executor_c ✅ SUCCESS >>>>>
UT_Test_Report_ge_autofuse ✅ SUCCESS >>>>>
ST_Test_Report_graphengine ✅ SUCCESS >>>>>
ST_Test_Report_ge_executor_c ✅ SUCCESS >>>>>
ST_Test_Report_ge_autofuse ✅ SUCCESS >>>>>
UT_Test_Report_fe ✅ SUCCESS >>>>>
ST_Test_Report_fe ✅ SUCCESS >>>>>

[2026-08-18 09:44:18]    CI执行结束

likedislike
CANN-robot
CANN-robot成员
1 天前 评论:

流水线任务触发成功
任务链接 [f3211550888c4653a9189766674e2976][流水线指导]

任务名称状态日志下载链接
codecheck_Pr ✅ SUCCESS
antipoison ✅ SUCCESS >>>>>
SCA ✅ SUCCESS >>>>>
codecheck_dt ✅ SUCCESS >>>>>
codecheck_precommit ✅ SUCCESS >>>>>
StaticCheck_codespell ✅ SUCCESS
StaticCheck_link_validity ✅ SUCCESS
StaticCheck_resource_existence ✅ SUCCESS
StaticCheck_tag_closed ✅ SUCCESS
StaticCheck_markdownlint ✅ SUCCESS

[2026-08-18 09:42:48]    CI执行结束

likedislike
CANN-robotCANN-robot成员
1 天前 添加了label:api-check-pass
CANN-robotCANN-robot成员
1 天前 删除了label:api-check-pass
CANN-robotCANN-robot成员
1 天前 删除了label:ci-pipeline-running
CANN-robotCANN-robot成员
1 天前 添加了label:ci-pipeline-passed
夏国正成员
1 天前 评论:

/lgtm

likedislike
zhanj
zhanj成员
1 天前 评论:

/lgtm

likedislike
CANN-robotCANN-robot成员
1 天前 添加了label:lgtm
夏国正成员
1 天前 评论:

@Wjieee 您的pr符合作业要求,请您添加小助手微信,并且发送您的gitcode个人主页截图,我们核对一下信息

likedislike
张德鹏成员
22 小时前 评论:

/approve

likedislike
CANN-robotCANN-robot成员
22 小时前 添加了label:approved
CANN-robotCANN-robot成员
22 小时前 关闭了关联的issue
CANN-robotCANN-robot成员
22 小时前 合入了pull request