buf parameter

super().__init__(buf) call

Space via make_space or make_discrete

Action space via make_discrete

reset() called in __init__

In-place state updates

Pre-allocated observation buffer

PufferEnv inheritance

step() return signature

No obs.copy() in observation return

layer_init import

layer_init on linear layers

layer_init on conv layers

Actor head std=0.01

Critic head std=1.0

LSTMCell for inference

LSTM for batch training

Pixel normalization

Separate inference and training methods

Shared LSTM parameters

PuffeRL trainer used

evaluate() in loop

train() in loop

mean_and_log() in loop

pufferlib.make() for environment

pufferlib logger class

NoLogger fallback

batch_size value

layer_init in policy

compile parameter

training_config.json present