Peng
Results
3
comments of
Peng
https://github.com/nupurkmr9/vision-aided-gan 这个更稳定，没有mode collapse
fix them by adding "utf8" encoder here: 309 version = versionstr.decode(encoding='utf8').strip().split('.') 313 match = re.search(r'(\d+)\.(\d+)\.(\d+)', compiler_info.decode(encoding='utf8').strip())
如果不考虑硬件开销，简单的方法是直接切分流水段部署在triton上。triton会根据测试到的latency来给每个步骤分配合适数量的GPU资源