PaddleOCR icon indicating copy to clipboard operation
PaddleOCR copied to clipboard

使用pdServering部署OCRV3时,调大batch_size的情况下,GPU内存还是和batch_size为1占用的内存一样

Open xiulianzw opened this issue 2 years ago • 0 comments

请提供下述完整信息以便快速定位问题/Please provide the following information to quickly locate the problem

  • docker镜像:pipeline_serving_demo registry.baidubce.com/paddlepaddle/serving:0.9.0-cuda11.2-cudnn8-runtime config.yml配置文件内容:
 # 批量查询Serving的数量, 默认1。batch_size>1要设置auto_batching_timeout,否则不足batch_size时会阻塞
batch_size: 16
# 批量查询超时,与batch_size配合使用
auto_batching_timeout: 2000

使用Serving/examples/Pipeline/PaddleOCR/ocr/benchmark.sh脚本进行测试

xiulianzw avatar Aug 09 '22 02:08 xiulianzw