Skip to content

README.md

Latest commit

 

History

History
61 lines (49 loc) · 2.17 KB

README.md

File metadata and controls

61 lines (49 loc) · 2.17 KB
Raw

generic_configs

通用 CI 配置：6 种 {0size, 1M, 4096} × {accuracy, paddleonly} 组合， 特殊 accuracy 变体：2 种（accuracy_compatible、accuracy_manual_threshold）， 输入/输出路径用 ${APITEST_MODEL} 占位，切换模型只需改这一个环境变量。 所有 accuracy 配置均使用 accuracy/accuracy_dual_gpu，全局 atol、rtol 固定为 0.0。

目录约定

需先把模型专属配置拷贝到 PaddleAPITest/${APITEST_MODEL}/ 下：

${APITEST_MODEL}/
├── accuracy_0size/0size.txt
├── accuracy/1M.txt
├── accuracy/4096.txt
├── paddleonly_0size/0size.txt
├── paddleonly/1M.txt
├── paddleonly/4096.txt
├── accuracy_compatible/accuracy_compatible.txt
└── accuracy_manual_threshold/
    ├── accuracy_manual_threshold.txt
    └── accuracy_manual_threshold_config.yaml

用法

export APITEST_MODEL=eb5_1
python run.py -c test_pipeline/generic_configs/run_0size_accuracy.yaml
python run.py -c test_pipeline/generic_configs/run_1M_accuracy.yaml
python run.py -c test_pipeline/generic_configs/run_4096_accuracy.yaml
python run.py -c test_pipeline/generic_configs/run_0size_paddleonly.yaml
python run.py -c test_pipeline/generic_configs/run_1M_paddleonly.yaml
python run.py -c test_pipeline/generic_configs/run_4096_paddleonly.yaml
python run.py -c test_pipeline/generic_configs/run_accuracy_compatible.yaml
python run.py -c test_pipeline/generic_configs/run_accuracy_manual_threshold.yaml

也可用 -i/--input、-o/--output 临时覆盖某次运行的输入文件和日志目录：

python run.py -c test_pipeline/generic_configs/run_1M_accuracy.yaml \
  -i ${APITEST_MODEL}/accuracy/1M.txt -o test_${APITEST_MODEL}_log_1M_accuracy

num_workers_per_gpu 配置

配置 num_workers_per_gpu
0size accuracy 4
0size paddleonly 4
4096 accuracy 4
4096 paddleonly 4
1M accuracy/paddleonly 1
accuracy_compatible / accuracy_manual_threshold 1

依赖的 run.py 能力

run.py 加载 YAML 后会对所有字符串字段执行 os.path.expandvars， 支持 ${VAR}/$VAR 占位符；未设置的变量原样保留。