Some-simple-gym-game-implemention

My Envirnment：
Python3.10
torch 2.0
numpy 1.24.0
gym 0.26.2
gym atari and box2d dependency
Don't worry about envirnment problems.High compatibility it has. 非常感谢您提供的更新信息！下面是包含这些信息的表格整理：

微调类型	Epoch	步数	Loss	时间	GPU
7B-Lora	128	12800	1.24	12.58小时	3张3090
3B-Lora	32	3200	1.42	5.43小时	2张3090
3B-Full-finetune	32	16384	0.45	19.07小时	3张3090+CPU

希望这个更新的表格能帮到您！如果还有其他问题，请随时提问。 install gym:

pip install gym
pip install gym[atari]
pip install gym[accept-rom-license]
apt-get install -y swig
pip install box2d-py
pip install gym[box2d] 

Pong , Lunerland , AirRaid

These game's action is discreate, use simple DQN algorithms to implement it.

BipedalWalker

This game's action is continuous, so you can use PPO or SAC to implement it.DQN is not good at it

SAC：

Site Reference