Linux GPU/CPU 基础训练推理测试

April 23, 2022 · View on GitHub

Linux GPU/CPU 基础训练推理测试的主程序为test_train_inference_python.sh，可以测试基于Python的模型训练、评估、推理等基本功能。

1. 测试结论汇总

训练相关：

算法名称	模型名称	单机单卡
Efficient Low-rank Multimodal Fusion with Modality-Specific Factors	LMF	正常训练

推理相关：

算法名称	模型名称	device_CPU	device_GPU	batchsize
Efficient Low-rank Multimodal Fusion with Modality-Specific Factors	LMF	支持	支持	1

2. 测试流程

2.1 准备数据

用于基础训练推理测试的数据位于./data/sample_data.pkl。

2.2 准备环境

安装PaddlePaddle：如果您已经安装了2.2或者以上版本的paddlepaddle，那么无需运行下面的命令安装paddlepaddle。

# 需要安装2.2及以上版本的Paddle
# 安装GPU版本的Paddle
pip install paddlepaddle-gpu==2.2.0
# 安装CPU版本的Paddle
pip install paddlepaddle==2.2.0

安装AutoLog（规范化日志输出工具）

pip install  https://paddleocr.bj.bcebos.com/libs/auto_log-1.2.0-py3-none-any.whl

2.3 功能测试

测试命令如下：

bash test_tipc/test_train_inference_python.sh test_tipc/configs/lmf/train_infer_python.txt lite_train_lite_infer

输出结果如下：

Epoch 0 complete! Average Training loss: 0.6811608409881592
Validation loss is: 0.6605233192443848
Found new check model, saving to disk...
Epoch 1 complete! Average Training loss: 0.6064284121990203
Validation loss is: 0.5399923324584961
Found new check model, saving to disk...
Epoch 2 complete! Average Training loss: 0.4072037336230278
Validation loss is: 0.3429754972457886
Found new check model, saving to disk...
Epoch 3 complete! Average Training loss: 0.3288413551449776
Validation loss is: 0.3257713317871094
Found new check model, saving to disk...
Epoch 4 complete! Average Training loss: 0.3574151992797851
Validation loss is: 0.360327935218811
F1-score on test set is 0.7810810810810811
Accuracy score on test set is 0.85
 Run successfully with command - python3.8 tools/train.py --emotion happy --model_path=./check --epochs=5   --batch_size=16 --data_path ./data/sample_data.pkl!  

Start testing happy......
Audio feature dimension is: 74
Visual feature dimension is: 35
Text feature dimension is: 300
W0423 01:29:08.148188 11983 device_context.cc:447] Please NOTE: device: 0, GPU Compute Capability: 7.5, Driver API Version: 11.0, Runtime API Version: 11.0
W0423 01:29:08.149482 11983 device_context.cc:465] device: 0, cuDNN Version: 8.0.
Model initialized
F1-score on test set is 0.7810810810810811
Accuracy score on test set is 0.85
 Run successfully with command - python3.8 ./tools/eval.py --data_path ./data/sample_data.pkl --check_path ./check/happy.pkl  !  

Audio feature dimension is: 74
Visual feature dimension is: 35
Text feature dimension is: 300
W0423 01:29:10.827006 12023 device_context.cc:447] Please NOTE: device: 0, GPU Compute Capability: 7.5, Driver API Version: 11.0, Runtime API Version: 11.0
W0423 01:29:10.828320 12023 device_context.cc:465] device: 0, cuDNN Version: 8.0.
inference model has been saved into ../check_inference/
 Run successfully with command - python3.8 tools/export.py --data_path ./data/sample_data.pkl --check_path ./check/    !  
 
[2022/04/23 01:29:15] root INFO: 

[2022/04/23 01:29:15] root INFO: ---------------------- Env info ----------------------
[2022/04/23 01:29:15] root INFO:  OS_version: Ubuntu 18.04
[2022/04/23 01:29:15] root INFO:  CUDA_version: 11.0.194
Build cuda_11.0_bu.TC445_37.28540450_0
[2022/04/23 01:29:15] root INFO:  CUDNN_version: None.None.None
[2022/04/23 01:29:15] root INFO:  drivier_version: 450.51.05
[2022/04/23 01:29:15] root INFO: ---------------------- Paddle info ----------------------
[2022/04/23 01:29:15] root INFO:  paddle_version: 2.2.2
[2022/04/23 01:29:15] root INFO:  paddle_commit: b031c389938bfa15e15bb20494c76f86289d77b0
[2022/04/23 01:29:15] root INFO:  log_api_version: 1.0
[2022/04/23 01:29:15] root INFO: ----------------------- Conf info -----------------------
[2022/04/23 01:29:15] root INFO:  runtime_device: gpu
[2022/04/23 01:29:15] root INFO:  ir_optim: True
[2022/04/23 01:29:15] root INFO:  enable_memory_optim: True
[2022/04/23 01:29:15] root INFO:  enable_tensorrt: False
[2022/04/23 01:29:15] root INFO:  enable_mkldnn: False
[2022/04/23 01:29:15] root INFO:  cpu_math_library_num_threads: 1
[2022/04/23 01:29:15] root INFO: ----------------------- Model info ----------------------
[2022/04/23 01:29:15] root INFO:  model_name: LMF
[2022/04/23 01:29:15] root INFO:  precision: fp32
[2022/04/23 01:29:15] root INFO: ----------------------- Data info -----------------------
[2022/04/23 01:29:15] root INFO:  batch_size: 1
[2022/04/23 01:29:15] root INFO:  input_shape: dynamic
[2022/04/23 01:29:15] root INFO:  data_num: 1
[2022/04/23 01:29:15] root INFO: ----------------------- Perf info -----------------------
[2022/04/23 01:29:15] root INFO:  cpu_rss(MB): 2407.2852, gpu_rss(MB): 1469.0, gpu_util: 8.0%
[2022/04/23 01:29:15] root INFO:  total time spent(s): 0.974
[2022/04/23 01:29:15] root INFO:  preprocess_time(ms): 8.8115, inference_time(ms): 965.1921, postprocess_time(ms): 0.0055
 Run successfully with command - python3.8 tools/infer.py --data_path ./data/sample_data.pkl --save_inference_dir ./check_inference --use-gpu=True   --batch_size=1   --benchmark=True > ./test_tipc/logs/python_infer_gpu_batchsize_1.log 2>&1 !  

python3.8 tools/infer.py --data_path ./data/sample_data.pkl --save_inference_dir ./check_inference --use-gpu=False --batch_size=1 --benchmark=True > ./test_tipc/logs/python_infer_cpu_batchsize_1.log 2>&1 
[2022/04/23 01:29:18] root INFO: 

[2022/04/23 01:29:18] root INFO: ---------------------- Env info ----------------------
[2022/04/23 01:29:18] root INFO:  OS_version: Ubuntu 18.04
[2022/04/23 01:29:18] root INFO:  CUDA_version: 11.0.194
Build cuda_11.0_bu.TC445_37.28540450_0
[2022/04/23 01:29:18] root INFO:  CUDNN_version: None.None.None
[2022/04/23 01:29:18] root INFO:  drivier_version: 450.51.05
[2022/04/23 01:29:18] root INFO: ---------------------- Paddle info ----------------------
[2022/04/23 01:29:18] root INFO:  paddle_version: 2.2.2
[2022/04/23 01:29:18] root INFO:  paddle_commit: b031c389938bfa15e15bb20494c76f86289d77b0
[2022/04/23 01:29:18] root INFO:  log_api_version: 1.0
[2022/04/23 01:29:18] root INFO: ----------------------- Conf info -----------------------
[2022/04/23 01:29:18] root INFO:  runtime_device: cpu
[2022/04/23 01:29:18] root INFO:  ir_optim: True
[2022/04/23 01:29:18] root INFO:  enable_memory_optim: True
[2022/04/23 01:29:18] root INFO:  enable_tensorrt: False
[2022/04/23 01:29:18] root INFO:  enable_mkldnn: False
[2022/04/23 01:29:18] root INFO:  cpu_math_library_num_threads: 1
[2022/04/23 01:29:18] root INFO: ----------------------- Model info ----------------------
[2022/04/23 01:29:18] root INFO:  model_name: LMF
[2022/04/23 01:29:18] root INFO:  precision: fp32
[2022/04/23 01:29:18] root INFO: ----------------------- Data info -----------------------
[2022/04/23 01:29:18] root INFO:  batch_size: 1
[2022/04/23 01:29:18] root INFO:  input_shape: dynamic
[2022/04/23 01:29:18] root INFO:  data_num: 1
[2022/04/23 01:29:18] root INFO: ----------------------- Perf info -----------------------
[2022/04/23 01:29:18] root INFO:  cpu_rss(MB): 1814.0195, gpu_rss(MB): None, gpu_util: None%
[2022/04/23 01:29:18] root INFO:  total time spent(s): 1.3604
[2022/04/23 01:29:18] root INFO:  preprocess_time(ms): 8.5521, inference_time(ms): 1351.82, postprocess_time(ms): 0.0062
 Run successfully with command - python3.8 tools/infer.py --data_path ./data/sample_data.pkl --save_inference_dir ./check_inference --use-gpu=False --batch_size=1 --benchmark=True > ./test_tipc/logs/python_infer_cpu_batchsize_1.log 2>&1 !