Language Language English 简体中文

unilab.algos.torch.fast_sac.runner.FastSACRunner

class unilab.algos.torch.fast_sac.runner.FastSACRunner

Bases: OffPolicyRunner

FastSAC using OffPolicyRunner infrastructure.

Parameters:

• env_name (str)

• env_cfg_override (dict[str, Any] | None)

• device (str | None)

• num_envs (int)

• replay_buffer_n (int)

• batch_size (int)

• learning_starts (int)

• updates_per_step (int)

• policy_frequency (int)

• sync_collection (bool)

• env_steps_per_sync (int)

• gamma (float)

• tau (float)

• actor_lr (float)

• critic_lr (float)

• alpha_lr (float)

• alpha_init (float)

• target_entropy_ratio (float)

• obs_normalization (bool)

• actor_hidden_dim (int)

• critic_hidden_dim (int)

• num_atoms (int)

• use_layer_norm (bool)

• max_grad_norm (float)

• use_amp (bool)

• amp_dtype (str)

• sim_backend (str)

• use_symmetry (bool)

• world_size (int)

• seed (int | None)

• trace_enabled (bool)

• trace_output_dir (str | None)

• trace_thread_time (bool)

• trace_cuda_events (bool)

Methods

__init__(env_name[, env_cfg_override, ...])
close()
learn([max_iterations, save_interval, ...])	Unified training loop for off-policy algorithms.

__init__(env_name, env_cfg_override=None, device=None, num_envs=4096, replay_buffer_n=1024, batch_size=8192, learning_starts=0, updates_per_step=8, policy_frequency=4, sync_collection=True, env_steps_per_sync=1, gamma=0.97, tau=0.125, actor_lr=0.0003, critic_lr=0.0003, alpha_lr=0.0003, alpha_init=0.001, target_entropy_ratio=1.0, obs_normalization=True, actor_hidden_dim=512, critic_hidden_dim=768, num_atoms=101, use_layer_norm=True, max_grad_norm=0.0, use_amp=False, amp_dtype='auto', sim_backend='mujoco', use_symmetry=False, world_size=1, seed=None, trace_enabled=False, trace_output_dir=None, trace_thread_time=False, trace_cuda_events=True)

Parameters:

• env_name (str)

• env_cfg_override (dict[str, Any] | None)

• device (str | None)

• num_envs (int)

• replay_buffer_n (int)

• batch_size (int)

• learning_starts (int)

• updates_per_step (int)

• policy_frequency (int)

• sync_collection (bool)

• env_steps_per_sync (int)

• gamma (float)

• tau (float)

• actor_lr (float)

• critic_lr (float)

• alpha_lr (float)

• alpha_init (float)

• target_entropy_ratio (float)

• obs_normalization (bool)

• actor_hidden_dim (int)

• critic_hidden_dim (int)

• num_atoms (int)

• use_layer_norm (bool)

• max_grad_norm (float)

• use_amp (bool)

• amp_dtype (str)

• sim_backend (str)

• use_symmetry (bool)

• world_size (int)

• seed (int | None)

• trace_enabled (bool)

• trace_output_dir (str | None)

• trace_thread_time (bool)

• trace_cuda_events (bool)

close()

Return type:

None

learn(max_iterations=1500, save_interval=50, log_dir='logs', logger_type='tensorboard')

Unified training loop for off-policy algorithms.

Parameters:

• max_iterations (int)

• save_interval (int)

• log_dir (str)

• logger_type (str)

Return type:

None