Iluvatar infer yolov5 (#287)

* Ixrt environment * add touch config * Iluvatar yolov5 case * fix mistake --------- Co-authored-by: 杨智超 <zhichao.yang@iluvatar.com>
FlagOpen · Oct 13, 2023 · 3dc084f · 3dc084f
1 parent 9bca0cc
commit 3dc084f
Show file tree

Hide file tree

Showing 6 changed files with 31 additions and 9 deletions.
diff --git a/inference/benchmarks/resnet50/README.md b/inference/benchmarks/resnet50/README.md
@@ -88,13 +88,13 @@ find ./val -name "*JPEG" | wc -l
    - OS kernel版本: 5.15.0-78-generic
    - 加速卡驱动版本：3.2.0
    - Docker 版本：24.0.4
-   - 训练框架版本：torch-1.13.1+corex.3.2.0
+   - 训练框架版本：torch-1.13.1+corex.3.2.1
    - 依赖软件版本：
      - cuda: 10.2
 
 - 推理工具包
 
-   - IXRT: ixrt-0.4.0+corex.3.2.0
+   - IXRT: ixrt-0.7.0+corex.3.2.1
 
 #### 2.5 腾讯紫霄 C100
 

diff --git a/inference/benchmarks/yolov5/README.md b/inference/benchmarks/yolov5/README.md
@@ -72,6 +72,25 @@ find ./val -name "*JPEG" | wc -l
 
    - XTCL 2.1
 
+#### 2.3 天数智芯 MR-V100
+
+- ##### 硬件环境
+    - 机器、加速卡型号: MR-V100
+
+- ##### 软件环境
+   - OS版本：Ubuntu 18.04
+   - OS kernel版本: 5.15.0-78-generic
+   - 加速卡驱动版本：3.2.0
+   - Docker 版本：24.0.4
+   - 依赖软件版本：
+     - cuda: 10.2
+     - torch-1.13.1+corex.3.2.1
+
+- 推理工具包
+
+   - IXRT: ixrt-0.7.0+corex.3.2.1
+
+
 ### 3. 运行情况
 
 * 指标列表
@@ -95,3 +114,4 @@ find ./val -name "*JPEG" | wc -l
 | tensorrt | fp32   | 96  | 733.8    |    /   | /    | 53.8       | 361.4 |12.6%| 0.45 | 35.44/40.0 |
 | tensorrt | fp16   | 96  | 1665.8    |    /   | /    | 58.6     | 859 |15.0%| 0.45 | 26.15/40.0 |
 | kunlunxin_xtcl | fp32   | 96  | / |    /   | / | /   | / |18.9%| 0.451 | 26.42/32.0 |
+| ixrt | fp16(W16A32)   | 96  | / |    /   | / | /   | / |17.1%| 0.45 | 10.38/32.0 |
diff --git a/inference/docker_images/iluvatar/iluvatar_analysis.py b/inference/docker_images/iluvatar/iluvatar_analysis.py
@@ -8,8 +8,10 @@ def analysis_log(logpath):
             usage = line.split(" ")[2]
             usage = float(usage[:-3])
             max_usage = max(max_usage, usage)
-            max_mem = line.split(" ")[3]
-            max_mem = float(max_mem[:-3])
+            if (max_usage >= usage):
+                m_mem = line.split(" ")[3]
+                m_mem = float(m_mem[:-3])
+                max_mem = max(max_mem, m_mem)
 
     return round(max_usage / 1024.0,
                  2), round(max_mem / 1024.0, 2), eval("24e12"), eval("96e12")
diff --git a/inference/docker_images/iluvatar/iluvatar_monitor.py b/inference/docker_images/iluvatar/iluvatar_monitor.py
@@ -75,7 +75,7 @@ def run(self):
 
         def gpu_mon(file):
             TIMESTAMP = datetime.datetime.now().strftime('%Y-%m-%d-%H:%M:%S')
-            cmd = "ixsmi |grep 'Default'|awk '{print $3,$5,$9,$11,$13}'"
+            cmd = "export PATH=/usr/local/corex/bin:$PATH; export LD_LIBRARY_PATH=/usr/local/corex/lib:$LD_LIBRARY_PATH; ixsmi |grep 'Default'|awk '{print $3,$5,$9,$11,$13}'"
             process = subprocess.Popen(cmd,
                                        shell=True,
                                        stdout=subprocess.PIPE,

diff --git a/inference/docker_images/iluvatar/pytorch/packages/README.md b/inference/docker_images/iluvatar/pytorch/packages/README.md
@@ -4,8 +4,8 @@
 
 ixrt-0.7.0+corex.latest.version-cp310-cp310-linux_x86_64.whl
 
-torchvision-0.14.1+corex.3.2.0.20230914.859-cp310-cp310-linux_x86_64.whl
+torchvision-0.14.1+corex.3.2.1.20231006.892-cp310-cp310-linux_x86_64.whl
 
-pycuda-2022.2.2+corex.3.2.0.20230914.859-cp310-cp310-linux_x86_64.whl
+pycuda-2022.2.2+corex.3.2.1.20231006.892-cp310-cp310-linux_x86_64.whl
 
-torch-1.13.1+corex.3.2.0.20230914.859-cp310-cp310-linux_x86_64.whl
+torch-1.13.1+corex.3.2.1.20231006.892-cp310-cp310-linux_x86_64
diff --git a/inference/docker_images/iluvatar/pytorch/sdk_installers/README.md b/inference/docker_images/iluvatar/pytorch/sdk_installers/README.md
@@ -2,6 +2,6 @@
 
 >联系邮箱: contact-us@iluvatar.com
 
-corex-installer-linux64-3.2.0-20230718_x86_64_10.2.run
+corex-installer-linux64-3.2.1.20231006.892_x86_64_10.2.run
 
 partial_install_cuda_header.tar.gz