RolloutBufferMemoryDiagnostics
python
RolloutBufferMemoryDiagnostics(memory_bytes: int, observation_shape: tuple[int, ...], observation_dtype: np.dtype, total_size: int) -> NoneEstimates CPU and GPU memory pressure for a rollout buffer's pre-allocated arrays.
Initialize the diagnostics helper.
Parameters:
memory_bytes(int) – The rollout buffer's total pre-allocated CPU memory, in bytes.observation_shape(tuple[int, ...]) – Shape of a single observation.observation_dtype(dtype) – Dtype of observations.total_size(int) – Total number of transitions (rollout_length*num_envs).
Methods
estimate_memory_pressure
python
estimate_memory_pressure(num_minibatches: int) -> list[RolloutBufferMemoryDiagnostic]Estimate CPU and GPU memory pressure for this rollout buffer.
Parameters:
num_minibatches(int) – Number of PPO minibatches per learner update.
Returns:
list[RolloutBufferMemoryDiagnostic]– Diagnostics describing projected memory pressure.
Raises:
ValueError– Ifnum_minibatchesis not positive.