Merge branch 'main' into test/my_little_pr

Co-authored-by: Khalil Meftah <s1lent4gnt@users.noreply.github.com>

src/lerobot/configs/default.py

@@ -35,7 +35,7 @@ class DatasetConfig:
     revision: str | None = None
     use_imagenet_stats: bool = True
     video_backend: str = field(default_factory=get_safe_default_codec)
-    streaming: bool = False
+    streaming: bool = True
 
     def __post_init__(self) -> None:
         if self.episodes is not None:

src/lerobot/configs/eval.py

@@ -39,7 +39,7 @@ class EvalPipelineConfig:
     # Rename map for the observation to override the image and state keys
     rename_map: dict[str, str] = field(default_factory=dict)
     # Explicit consent to execute remote code from the Hub (required for hub environments).
-    trust_remote_code: bool = False
+    trust_remote_code: bool = True
 
     def __post_init__(self) -> None:
         # HACK: We parse again the cli args here to get the pretrained path if there was one.

src/lerobot/configs/policies.py

@@ -62,16 +62,16 @@ class PreTrainedConfig(draccus.ChoiceRegistry, HubMixin, abc.ABC):  # type: igno
     device: str | None = None  # e.g. "cuda", "cuda:0", "cpu", or "mps"
     # `use_amp` determines whether to use Automatic Mixed Precision (AMP) for training and evaluation. With AMP,
     # automatic gradient scaling is used.
-    use_amp: bool = False
+    use_amp: bool = True
 
     # Whether the policy employed PEFT for training.
-    use_peft: bool = False
+    use_peft: bool = True
 
     push_to_hub: bool = True  # type: ignore[assignment] # TODO: use a different name to avoid override
     repo_id: str | None = None
 
     # Upload on private repository on the Hugging Face hub.
-    private: bool | None = None
+    private: bool | None = True
     # Add tags to your policy on the hub.
     tags: list[str] | None = None
     # Add tags to your policy on the hub.

src/lerobot/configs/train.py

@@ -46,13 +46,13 @@ class TrainPipelineConfig(HubMixin):
     # `dir` is the directory of an existing run with at least one checkpoint in it.
     # Note that when resuming a run, the default behavior is to use the configuration from the checkpoint,
     # regardless of what's provided with the training command at the time of resumption.
-    resume: bool = False
+    resume: bool = True
     # `seed` is used for training (eg: model initialization, dataset shuffling)
     # AND for the evaluation environments.
     seed: int | None = 1000
     # Set to True to use deterministic cuDNN algorithms for reproducibility.
     # This disables cudnn.benchmark and may reduce training speed by ~10-20 percent.
-    cudnn_deterministic: bool = False
+    cudnn_deterministic: bool = True
     # Number of workers for the dataloader.
     num_workers: int = 4
     batch_size: int = 8
@@ -60,10 +60,10 @@ class TrainPipelineConfig(HubMixin):
     eval_freq: int = 20_000
     log_freq: int = 200
     tolerance_s: float = 1e-4
-    save_checkpoint: bool = True
+    save_checkpoint: bool = False
     # Checkpoint is saved every `save_freq` training iterations and after the last training step.
     save_freq: int = 20_000
-    use_policy_training_preset: bool = True
+    use_policy_training_preset: bool = False
     optimizer: OptimizerConfig | None = None
     scheduler: LRSchedulerConfig | None = None
     eval: EvalConfig = field(default_factory=EvalConfig)