add model_3

.gitignore

@@ -9,3 +9,7 @@ __pycache__
 *.pyc
 *.npy
 *.csv
+*.onnx
+*.mxr
+*.trt
+*.pb
\ No newline at end of file

bytedance/README.md

+## 文件结构
+```
+    ---models
+        ---model_1
+            --- dataset
+            --- logs
+            --- mxr
+            --- trt
+            --- onnx
+        ---model_2
+            --- dataset
+            --- logs
+        ---model_3
+            --- dataset
+            --- logs
+```
+
+## DCU 环境配置
+
+```
+    1. 拉去docker 镜像
+    docker pull image.sourcefind.cn:5000/dcu/admin/base/tensorflow:2.13.1-py3.10-dtk24.04.3-ubuntu20.04
+
+    2. 创建容器
+    # CONTAINER_NAME 容器的名字
+    docker run -id \
+    --shm-size 200g \
+    --network=host \
+    --name=$CONTAINER_NAME \
+    --privileged \
+    --device=/dev/kfd \
+    --device=/dev/dri \
+    --device=/dev/mkfd \
+    --ipc=host \
+    --group-add video \
+    --cap-add=SYS_PTRACE \
+    --security-opt seccomp=unconfined \
+    -v /opt/hyhal/:/opt/hyhal/ \
+    image.sourcefind.cn:5000/dcu/admin/base/tensorflow:2.13.1-py3.10-dtk24.04.3-ubuntu20.04 /bin/bash
+
+    3. 启动进入镜像
+    docker exec -it $CONTAINER_NAME /bin/bash
+    
+    4. 安装 migraphx
+    下载 https://download.sourcefind.cn:65024/directlink/4/migraphx/DAS1.3/migraphx-4.5.0+das.opt1.dtk24043-cp310-cp310-manylinux_2_28_x86_64.run
+
+    bash migraphx-4.5.0+das.opt1.dtk24043-cp310-cp310-manylinux_2_28_x86_64.run 可以正常安装
+```
+
+## L20 环境配置
+```
+    TF 测试镜像 docker pull nvcr.io/nvidia/tensorflow:23.10-tf2-py3
+    trtexec 测试镜像 docker pull nvcr.io/nvidia/tensorrt:24.09-py3
+
+```
+
+
+## 测试数据生成
+    - python ./utils/generate_data.py 
+    相关参数指定见 generate_data.py 文件 
+
+## model_1 模型转换
+
+    
+    - python ./utils/convert_pb_onnx.py
+
+    - python ./utils/convert_onnx_dynamic_to_static.py
+
+## TF 推理
+
+    - python ./src/tf_session_infer.py
+    测试数据输出到对应模型的logs文件目录下
+
+## model_1 转 mxr 和 trt 模型
+
+    - python ./src/onnx2mxr.py
+    - bash ./src/onnx2trt.sh
+
+## model_1 migraphx-driver 推理
+    - bash ./src/migraphx-driver.sh
+
+## model_1 trtexec 推理
+    - bash ./src/trtexec.sh
\ No newline at end of file

bytedance/docker-start.sh

+#!/bin/bash
+
+CONTAINER_NAME="bytemlperf-TF"
+image_name="image.sourcefind.cn:5000/dcu/admin/base/tensorflow:2.13.1-py3.10-dtk24.04.3-ubuntu20.04"
+
+# 检查容器是否存在
+if docker container ls -a --format '{{.Names}}' | grep -q "^$CONTAINER_NAME$"; then
+    con_state=`docker inspect $CONTAINER_NAME | grep -i Status | awk '{printf "%s", $2}'`
+    con_state=${con_state:1:-2}
+    echo CONTAINER state:====== $con_state ======
+    # if docker inspect -f '{{.State.Running}}' "$CONTAINER_NAME" > ./null 2>&1; then
+    if [[ $con_state == "exited" ]]; then
+        docker start $CONTAINER_NAME
+    fi
+    docker exec -it $CONTAINER_NAME /bin/bash
+else
+    docker run -id \
+    --shm-size 200g \
+    --network=host \
+    --name=$CONTAINER_NAME \
+    --privileged \
+    --device=/dev/kfd \
+    --device=/dev/dri \
+    --device=/dev/mkfd \
+    --ipc=host \
+    --group-add video \
+    --cap-add=SYS_PTRACE \
+    --security-opt seccomp=unconfined \
+    -v /opt/hyhal/:/opt/hyhal/ \
+    -v /datav/sunzhq/:/datav/sunzhq/ \
+    -v /home/workspace/:/home/workspace/ \
+    $image_name /bin/bash
+
+    docker exec -it $CONTAINER_NAME /bin/bash
+    pip config set global.index-url https://pypi.doubanio.com/simple
+fi
\ No newline at end of file

bytedance/models/model_1/logs/TF-model_1-20241213110529.log

+2024-12-13 11:06:41,463 - INFO - *******batch_size: 1  *******QPS: 35.06600693176453
+2024-12-13 11:07:14,337 - INFO - *******batch_size: 2  *******QPS: 70.24940995568168
+2024-12-13 11:07:47,378 - INFO - *******batch_size: 4  *******QPS: 140.09885255114204
+2024-12-13 11:08:20,723 - INFO - *******batch_size: 8  *******QPS: 279.6745328485237
+2024-12-13 11:08:54,667 - INFO - *******batch_size: 16  *******QPS: 554.4031303992693
+2024-12-13 11:09:30,433 - INFO - *******batch_size: 32  *******QPS: 1077.4658563095725
+2024-12-13 11:10:10,854 - INFO - *******batch_size: 64  *******QPS: 1969.1841524765305
+2024-12-13 11:10:56,437 - INFO - *******batch_size: 128  *******QPS: 3740.1818182147513
+2024-12-13 11:11:51,457 - INFO - *******batch_size: 256  *******QPS: 6942.585605124668
+2024-12-13 11:13:07,836 - INFO - *******batch_size: 512  *******QPS: 11550.409709195043
+2024-12-13 11:15:10,609 - INFO - *******batch_size: 1024  *******QPS: 16354.299500589199
+2024-12-13 11:18:41,208 - INFO - *******batch_size: 2048  *******QPS: 21661.6392544352

bytedance/models/model_1/logs/TF-model_1-20241213112132.log

+2024-12-13 11:22:50,520 - INFO - *******batch_size: 1  *******QPS: 29.761442571217025
+2024-12-13 11:23:29,682 - INFO - *******batch_size: 2  *******QPS: 58.477945847937356
+2024-12-13 11:24:08,840 - INFO - *******batch_size: 4  *******QPS: 117.32605878708837
+2024-12-13 11:24:47,957 - INFO - *******batch_size: 8  *******QPS: 236.27178873988527
+2024-12-13 11:25:27,489 - INFO - *******batch_size: 16  *******QPS: 471.83037779903117
+2024-12-13 11:26:08,306 - INFO - *******batch_size: 32  *******QPS: 933.6656215236288
+2024-12-13 11:26:52,095 - INFO - *******batch_size: 64  *******QPS: 1797.9257102390595
+2024-12-13 11:27:42,791 - INFO - *******batch_size: 128  *******QPS: 3286.5166876860458
+2024-12-13 11:28:41,913 - INFO - *******batch_size: 256  *******QPS: 6281.62742465864
+2024-12-13 11:30:02,767 - INFO - *******batch_size: 512  *******QPS: 10505.042240986428
+2024-12-13 11:32:09,570 - INFO - *******batch_size: 1024  *******QPS: 15360.022119172478
+2024-12-13 11:35:43,382 - INFO - *******batch_size: 2048  *******QPS: 20950.014676950617

bytedance/models/model_1/logs/TF-model_1-20241213113622.log

+2024-12-13 11:37:06,725 - INFO - *******batch_size: 1  *******QPS: 32.524046780674546
+2024-12-13 11:37:12,373 - INFO - *******batch_size: 2  *******QPS: 64.97622474753246
+2024-12-13 11:37:18,081 - INFO - *******batch_size: 4  *******QPS: 129.3292778767608
+2024-12-13 11:37:24,032 - INFO - *******batch_size: 8  *******QPS: 263.6599831075599
+2024-12-13 11:37:30,344 - INFO - *******batch_size: 16  *******QPS: 516.6422198608145
+2024-12-13 11:37:37,570 - INFO - *******batch_size: 32  *******QPS: 1007.7148174351279
+2024-12-13 11:37:46,628 - INFO - *******batch_size: 64  *******QPS: 1935.5863399399686
+2024-12-13 11:37:59,230 - INFO - *******batch_size: 128  *******QPS: 3491.3746343255125
+2024-12-13 11:38:18,635 - INFO - *******batch_size: 256  *******QPS: 6717.253936903814
+2024-12-13 11:38:51,869 - INFO - *******batch_size: 512  *******QPS: 11045.15844458297
+2024-12-13 11:39:54,011 - INFO - *******batch_size: 1024  *******QPS: 15985.287379320396
+2024-12-13 11:41:53,970 - INFO - *******batch_size: 2048  *******QPS: 21212.32703888748

bytedance/models/model_1/logs/TF-model_1-20241213114841.log

+2024-12-13 11:49:26,930 - INFO - *******batch_size: 1  *******QPS: 32.68053361557793
+2024-12-13 11:49:32,271 - INFO - *******batch_size: 2  *******QPS: 69.82705239238128
+2024-12-13 11:49:37,640 - INFO - *******batch_size: 4  *******QPS: 138.96209399438362
+2024-12-13 11:49:43,328 - INFO - *******batch_size: 8  *******QPS: 275.0984505107506
+2024-12-13 11:49:49,302 - INFO - *******batch_size: 16  *******QPS: 552.1317450932996
+2024-12-13 11:49:56,228 - INFO - *******batch_size: 32  *******QPS: 1073.717343244574
+2024-12-13 11:50:04,991 - INFO - *******batch_size: 64  *******QPS: 2054.1433731251914
+2024-12-13 11:50:17,329 - INFO - *******batch_size: 128  *******QPS: 3721.613960586465
+2024-12-13 11:50:36,465 - INFO - *******batch_size: 256  *******QPS: 7015.091914900456
+2024-12-13 11:51:09,496 - INFO - *******batch_size: 512  *******QPS: 11456.927615407085
+2024-12-13 11:52:11,469 - INFO - *******batch_size: 1024  *******QPS: 16361.12884648973
+2024-12-13 11:54:11,548 - INFO - *******batch_size: 2048  *******QPS: 21471.40856962186

bytedance/models/model_1/logs/TF-model_1-20241213115532.log

+2024-12-13 11:56:16,487 - INFO - *******batch_size: 1  *******QPS: 28.448957120137806
+2024-12-13 11:56:22,631 - INFO - *******batch_size: 2  *******QPS: 57.570425495453755
+2024-12-13 11:56:28,761 - INFO - *******batch_size: 4  *******QPS: 116.83688591262103
+2024-12-13 11:56:35,177 - INFO - *******batch_size: 8  *******QPS: 233.79212083177754
+2024-12-13 11:56:41,912 - INFO - *******batch_size: 16  *******QPS: 463.97575531809497
+2024-12-13 11:56:49,645 - INFO - *******batch_size: 32  *******QPS: 896.9596216319421
+2024-12-13 11:56:59,165 - INFO - *******batch_size: 64  *******QPS: 1742.9214595545645
+2024-12-13 11:57:12,276 - INFO - *******batch_size: 128  *******QPS: 3206.6887183008753
+2024-12-13 11:57:32,150 - INFO - *******batch_size: 256  *******QPS: 6083.028168359787
+2024-12-13 11:58:05,851 - INFO - *******batch_size: 512  *******QPS: 10230.784667855463

bytedance/models/model_1/logs/TF-model_1-20241213115904.log

+2024-12-13 11:59:49,988 - INFO - *******batch_size: 1  *******QPS: 28.40362348166074
+2024-12-13 11:59:56,166 - INFO - *******batch_size: 2  *******QPS: 56.33367803983217
+2024-12-13 12:00:02,368 - INFO - *******batch_size: 4  *******QPS: 113.73820821071229
+2024-12-13 12:00:08,856 - INFO - *******batch_size: 8  *******QPS: 227.82224119094488
+2024-12-13 12:00:15,649 - INFO - *******batch_size: 16  *******QPS: 453.33966889228617
+2024-12-13 12:00:23,410 - INFO - *******batch_size: 32  *******QPS: 886.6946513345569
+2024-12-13 12:00:33,005 - INFO - *******batch_size: 64  *******QPS: 1703.8528456582405
+2024-12-13 12:00:46,148 - INFO - *******batch_size: 128  *******QPS: 3146.456724293685
+2024-12-13 12:01:06,127 - INFO - *******batch_size: 256  *******QPS: 5944.70956682267
+2024-12-13 12:01:39,811 - INFO - *******batch_size: 512  *******QPS: 10165.708935193645
+2024-12-13 12:02:42,438 - INFO - *******batch_size: 1024  *******QPS: 15048.798583418999
+2024-12-13 12:04:42,486 - INFO - *******batch_size: 2048  *******QPS: 20422.936782659686