2022-02-10 16:25:52 +08:00
|
|
|
|
|
|
|
|
|
|
# TIPC Linux端Benchmark测试文档
|
|
|
|
|
|
|
|
|
|
|
|
该文档为Benchmark测试说明，Benchmark预测功能测试的主程序为`benchmark_train.sh`，用于验证监控模型训练的性能。
|
|
|
|
|
|
|
|
|
|
|
|
# 1. 测试流程
|
|
|
|
|
|
## 1.1 准备数据和环境安装
|
|
|
|
|
|
运行`test_tipc/prepare.sh`，完成训练数据准备和安装环境流程。
|
|
|
|
|
|
|
|
|
|
|
|
```shell
|
|
|
|
|
|
# 运行格式：bash test_tipc/prepare.sh train_benchmark.txt mode
|
2022-06-15 22:33:17 +08:00
|
|
|
|
bash test_tipc/prepare.sh test_tipc/config/MobileNetV2/MobileNetV2_train_infer_python.txt benchmark_train
|
2022-02-10 16:25:52 +08:00
|
|
|
|
```
|
|
|
|
|
|
|
|
|
|
|
|
## 1.2 功能测试
|
|
|
|
|
|
执行`test_tipc/benchmark_train.sh`，完成模型训练和日志解析
|
|
|
|
|
|
|
|
|
|
|
|
```shell
|
|
|
|
|
|
# 运行格式：bash test_tipc/benchmark_train.sh train_benchmark.txt mode
|
|
|
|
|
|
bash test_tipc/benchmark_train.sh test_tipc/config/MobileNetV2/MobileNetV2_train_infer_python.txt benchmark_train
|
|
|
|
|
|
|
|
|
|
|
|
```
|
|
|
|
|
|
|
|
|
|
|
|
`test_tipc/benchmark_train.sh`支持根据传入的第三个参数实现只运行某一个训练配置，如下：
|
|
|
|
|
|
```shell
|
|
|
|
|
|
# 运行格式：bash test_tipc/benchmark_train.sh train_benchmark.txt mode params
|
2022-06-15 22:33:17 +08:00
|
|
|
|
bash test_tipc/benchmark_train.sh test_tipc/config/MobileNetV2/MobileNetV2_train_infer_python.txt benchmark_train dynamic_bs8_fp32_DP_N1C1
|
2022-02-10 16:25:52 +08:00
|
|
|
|
```
|
|
|
|
|
|
dynamic_bs8_fp32_DP_N1C1为test_tipc/benchmark_train.sh传入的参数，格式如下：
|
|
|
|
|
|
`${modeltype}_${batch_size}_${fp_item}_${run_mode}_${device_num}`
|
|
|
|
|
|
包含的信息有：模型类型、batchsize大小、训练精度如fp32,fp16等、分布式运行模式以及分布式训练使用的机器信息如单机单卡（N1C1）。
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
## 2. 日志输出
|
|
|
|
|
|
|
2022-06-15 22:33:17 +08:00
|
|
|
|
运行后将保存模型的训练日志和解析日志，使用 `test_tipc/config/MobileNetV2/MobileNetV2_train_infer_python.txt` 参数文件的训练日志解析结果是：
|
2022-02-10 16:25:52 +08:00
|
|
|
|
|
|
|
|
|
|
```
|
|
|
|
|
|
{"model_branch": "dygaph", "model_commit": "7c39a1996b19087737c05d883fd346d2f39dbcc0", "model_name": "cls_MobileNetV2_bs8_fp32_SingleP_DP", "batch_size": 8, "fp_item": "fp32", "run_process_type": "SingleP", "run_mode": "DP", "convergence_value": "5.413110", "convergence_key": "loss:", "ips": 19.333, "speed_unit": "samples/s", "device_num": "N1C1", "model_run_time": "0", "frame_commit": "8cc09552473b842c651ead3b9848d41827a3dbab", "frame_version": "0.0.0"}
|
|
|
|
|
|
```
|
|
|
|
|
|
|
|
|
|
|
|
训练日志和日志解析结果保存在benchmark_log目录下，文件组织格式如下：
|
|
|
|
|
|
```
|
|
|
|
|
|
train_log/
|
|
|
|
|
|
├── index
|
|
|
|
|
|
│ ├── PaddleClas_cls_MobileNetV2_bs8_fp32_SingleP_DP_N1C1_speed
|
|
|
|
|
|
│ └── PaddleClas_cls_MobileNetV2_bs8_fp32_SingleP_DP_N1C4_speed
|
|
|
|
|
|
├── profiling_log
|
|
|
|
|
|
│ └── PaddleClas_cls_MobileNetV2_bs8_fp32_SingleP_DP_N1C1_profiling
|
|
|
|
|
|
└── train_log
|
|
|
|
|
|
├── PaddleClas_cls_MobileNetV2_bs8_fp32_SingleP_DP_N1C1_log
|
|
|
|
|
|
└── PaddleClas_cls_MobileNetV2_bs8_fp32_SingleP_DP_N1C4_log
|
|
|
|
|
|
```
|
