Hana-ame
/

maze-transformer

Reinforcement Learning

maze-navigation

Model card Files Files and versions

maze-transformer / tools

6.67 kB

Ctrl+K

Ctrl+K

2 contributors

History: 2 commits

Hana-ame's picture

tune: 调参空间扩展至 45 参数 — 补预热方式/损失缩放/ZeRO/激活重计算/FSDP，层数与宽度量级放大；格式=参数：候选值枚举

50fb20e 2 days ago

build_experiments_xlsx.py

6.67 kB
tune: 调参空间扩展至 45 参数 — 补预热方式/损失缩放/ZeRO/激活重计算/FSDP，层数与宽度量级放大；格式=参数：候选值枚举 2 days ago