Add storage device configuration for SAC policy and replay buffer

- Introduce `storage_device` parameter in SAC configuration and training settings
- Update learner server to use configurable storage device for replay buffer
- Reduce online buffer capacity in ManiSkill configuration
- Modify replay buffer initialization to support custom storage device

lerobot/common/policies/sac/configuration_sac.py

@@ -64,6 +64,8 @@ class SACConfig:
         }
     )
     camera_number: int = 1
+
+    storage_device: str = "cpu"
     # Add type annotations for these fields:
     vision_encoder_name: str | None = field(default="helper2424/resnet10")
     freeze_vision_encoder: bool = True

lerobot/configs/policy/sac_maniskill.yaml

@@ -20,6 +20,9 @@ training:
   grad_clip_norm: 10.0
   lr: 3e-4
 
+
+  storage_device: "cpu"
+
   eval_freq: 2500
   log_freq: 10
   save_freq: 2000000
@@ -30,7 +33,7 @@ training:
   online_steps_between_rollouts: 1000
   online_sampling_ratio: 1.0
   online_env_seed: 10000
-  online_buffer_capacity: 1000000
+  online_buffer_capacity: 200000
   online_buffer_seed_size: 0
   online_step_before_learning: 500
   do_online_rollout_async: false

lerobot/scripts/server/learner_server.py

@@ -146,14 +146,14 @@ def log_training_info(cfg: DictConfig, out_dir: str, policy: nn.Module) -> None:
 
 
 def initialize_replay_buffer(
-    cfg: DictConfig, logger: Logger, device: str
+    cfg: DictConfig, logger: Logger, device: str, storage_device:str
 ) -> ReplayBuffer:
     if not cfg.resume:
         return ReplayBuffer(
             capacity=cfg.training.online_buffer_capacity,
             device=device,
             state_keys=cfg.policy.input_shapes.keys(),
-            storage_device=device,
+            storage_device=storage_device,
             optimize_memory=True,
         )
 
@@ -596,6 +596,7 @@ def train(cfg: DictConfig, out_dir: str | None = None, job_name: str | None = No
     set_global_seed(cfg.seed)
 
     device = get_safe_torch_device(cfg.device, log=True)
+    storage_device = get_safe_torch_device(cfg_device=cfg.training.storage_device)
 
     torch.backends.cudnn.benchmark = True
     torch.backends.cuda.matmul.allow_tf32 = True
@@ -628,7 +629,7 @@ def train(cfg: DictConfig, out_dir: str | None = None, job_name: str | None = No
 
     log_training_info(cfg, out_dir, policy)
 
-    replay_buffer = initialize_replay_buffer(cfg, logger, device)
+    replay_buffer = initialize_replay_buffer(cfg, logger, device, storage_device)
     batch_size = cfg.training.batch_size
     offline_replay_buffer = None
 
@@ -649,7 +650,7 @@ def train(cfg: DictConfig, out_dir: str | None = None, job_name: str | None = No
             state_keys=cfg.policy.input_shapes.keys(),
             action_mask=active_action_dims,
             action_delta=cfg.env.wrapper.delta_action,
-            storage_device=device,
+            storage_device=storage_device,
             optimize_memory=True,
         )
         batch_size: int = batch_size // 2  # We will sample from both replay buffer