fix(peft): allow fresh LoRA fine-tuning from a base-model checkpoint

docs/source/lingbot_va.mdx

@@ -111,17 +111,26 @@ Requirements:
 - The block-causal masks use PyTorch **flex-attention**, so build the policy with
   `--policy.attn_mode=flex` for training (the default `torch` SDPA is inference-only).
 - The full 5B DiT does not fit a single 24–32 GB GPU under AdamW; fine-tune with **LoRA**
-  (`--policy.use_peft=true`) and/or optimizer offload. `get_optim_params` returns only the
-  trainable (e.g. adapter) parameters; the VAE + UMT5 text encoder stay frozen.
+  (`--peft.method_type=LORA`) and/or optimizer offload. `get_optim_params` returns only the
+  trainable (e.g. adapter) parameters; the VAE + UMT5 text encoder stay frozen. Install the
+  `lerobot[peft]` extra to enable PEFT support (see the [PEFT training guide](./peft_training)).
 
 ```bash
 lerobot-train \
   --policy.path=lerobot/lingbot_va_libero_long --policy.attn_mode=flex \
-  --policy.use_peft=true \
+  --peft.method_type=LORA --peft.r=32 --peft.lora_alpha=32 \
+  --peft.target_modules='transformer\.blocks\.\d+\.attn[12]\.(to_q|to_v)' \
   --dataset.repo_id=<your LeRobot-format dataset> \
   --batch_size=1 --steps=... --output_dir=outputs/train/lingbot_va
 ```
 
+Unlike SmolVLA / π₀, LingBot-VA does not ship built-in default LoRA targets, so you must pass
+`--peft.target_modules` explicitly. Only `self.transformer` (the dual-stream Wan transformer) is
+trainable; the example above adapts the query/value projections of both its self-attention
+(`attn1`) and cross-attention (`attn2`) blocks — the standard LoRA target set. Broaden it (e.g.
+add `to_k`/`to_out`, or the `ffn` layers) if you need a higher-capacity adapter. Passing
+`--peft.method_type` implies PEFT, so `--policy.use_peft=true` is not required.
+
 The dataset must provide camera clips (a temporal window per camera, VAE-encoded to
 `frame_chunk_size` latent frames) and `frame_chunk_size * action_per_frame` action steps per item.
 

docs/source/peft_training.mdx

@@ -62,3 +62,10 @@ to the `--peft.full_training_modules` parameter:
 
 The learning rate and the scheduled target learning rate can usually be scaled by a factor of 10 compared to the
 learning rate used for full fine-tuning (e.g., 1e-4 normal, so 1e-3 using LoRA).
+
+## Other policies
+
+The same `--peft.*` flags work for any pre-trained policy. Some policies (SmolVLA, π₀, π₀.₅) ship
+built-in default `target_modules`, so `--peft.method_type=LORA` is enough. Others do not, and will
+ask you to pass `--peft.target_modules` explicitly — for example LingBot-VA, whose recommended
+targets are documented in its [dedicated guide](./lingbot_va#training--fine-tuning).

examples/annotations/run_hf_job.py

@@ -46,8 +46,11 @@ CMD = (
     "apt-get update -qq && apt-get install -y -qq git ffmpeg && "
     "pip install --no-deps "
     "'lerobot @ git+https://github.com/huggingface/lerobot.git@main' && "
+    # Pins mirror pyproject.toml — unpinned installs pull av 18 / datasets 5 /
+    # draccus 0.11, which break lerobot at import time.
     "pip install --upgrade-strategy only-if-needed "
-    "datasets pyarrow av jsonlines draccus gymnasium torchcodec mergedeep pyyaml-include toml typing-inspect "
+    "'datasets>=4.7.0,<5.0.0' 'pyarrow>=21.0.0,<30.0.0' 'av>=15.0.0,<16.0.0' 'draccus==0.10.0' "
+    "'pandas>=2.0.0,<3.0.0' jsonlines gymnasium torchcodec mergedeep pyyaml-include toml typing-inspect "
     "openai && "
     "export VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS=0 && "
     "export VLLM_VIDEO_BACKEND=pyav && "

src/lerobot/configs/train.py

@@ -221,6 +221,11 @@ class TrainPipelineConfig(HubMixin):
             )
 
         active_cfg = self.trainable_config
+
+        # Keep the policy-level `use_peft` flag in sync with the presence of a `--peft.*` config.
+        if self.peft is not None and self.policy is not None:
+            self.policy.use_peft = True
+
         if self.rename_map and active_cfg.pretrained_path is None:
             raise ValueError(
                 "`rename_map` requires a pretrained policy checkpoint. "

src/lerobot/policies/factory.py

@@ -513,6 +513,37 @@ def make_pre_post_processors(
     return processors
 
 
+def _has_peft_adapter_config(
+    pretrained_path: str,
+    revision: str | None = None,
+) -> bool:
+    """Return whether ``pretrained_path`` points to an existing PEFT adapter.
+
+    A PEFT adapter checkpoint always ships an ``adapter_config.json``.
+    A plain base-model checkpoint does not. This distinction lets us tell apart
+    two very different ``use_peft=True`` scenarios that both set ``pretrained_path``:
+
+    * loading/resuming a previously trained adapter (config lives at ``pretrained_path``)
+    * starting a *fresh* PEFT fine-tune on top of a base model
+
+    Works for both local directories and Hub repo ids.
+    """
+    import os
+
+    adapter_config_name = "adapter_config.json"
+
+    if os.path.isdir(pretrained_path):
+        return os.path.isfile(os.path.join(pretrained_path, adapter_config_name))
+
+    from huggingface_hub import file_exists
+    from huggingface_hub.errors import HfHubHTTPError
+
+    try:
+        return file_exists(pretrained_path, adapter_config_name, revision=revision)
+    except (HfHubHTTPError, OSError):
+        return False
+
+
 def make_policy(
     cfg: PreTrainedConfig,
     ds_meta: LeRobotDatasetMetadata | None = None,
@@ -607,13 +638,24 @@ def make_policy(
             "the PEFT config parameters to be set. For training with PEFT, see `lerobot_train.py` on how to do that."
         )
 
-    if cfg.pretrained_path and not cfg.use_peft:
+    # When `use_peft=True` and a checkpoint is given, the checkpoint can be one of two things:
+    # 1. A base model checkpoint (e.g., a pretrained policy) on which we want to start a fresh PEFT fine-tune.
+    # 2. A PEFT adapter checkpoint (e.g., a previously trained PEFT adapter)
+    # We distinguish between these two cases
+    load_existing_adapter = (
+        cfg.pretrained_path
+        and cfg.use_peft
+        and _has_peft_adapter_config(str(cfg.pretrained_path), cfg.pretrained_revision)
+    )
+
+    if cfg.pretrained_path and not load_existing_adapter:
         # Load a pretrained policy and override the config if needed (for example, if there are inference-time
-        # hyperparameters that we want to vary).
+        # hyperparameters that we want to vary). This also covers starting a fresh PEFT fine-tune on top of a
+        # base model: the base weights are loaded here and `wrap_with_peft` builds the adapter afterwards.
         kwargs["pretrained_name_or_path"] = cfg.pretrained_path
         kwargs["revision"] = cfg.pretrained_revision
         policy = policy_cls.from_pretrained(**kwargs)
-    elif cfg.pretrained_path and cfg.use_peft:
+    elif load_existing_adapter:
         # Load a pretrained PEFT model on top of the policy. The pretrained path points to the folder/repo
         # of the adapter and the adapter's config contains the path to the base policy. So we need the
         # adapter config first, then load the correct policy and then apply PEFT.

src/lerobot/policies/pretrained.py

@@ -23,8 +23,6 @@ from pathlib import Path
 from tempfile import TemporaryDirectory
 from typing import TYPE_CHECKING, TypedDict, TypeVar, Unpack
 
-import packaging
-import safetensors
 from huggingface_hub import HfApi, ModelCard, ModelCardData, hf_hub_download, save_torch_state_dict
 from huggingface_hub.constants import SAFETENSORS_SINGLE_FILE
 from huggingface_hub.errors import HfHubHTTPError
@@ -34,6 +32,7 @@ from torch import Tensor, nn
 from lerobot.__version__ import __version__
 from lerobot.configs import PreTrainedConfig
 from lerobot.configs.train import TrainPipelineConfig
+from lerobot.utils.device_utils import resolve_safetensors_device
 from lerobot.utils.hub import HubMixin
 
 from .utils import log_model_loading_keys
@@ -221,26 +220,10 @@ class PreTrainedPolicy(nn.Module, HubMixin, abc.ABC):
 
     @classmethod
     def _load_as_safetensor(cls, model: T, model_file: str, map_location: str, strict: bool) -> T:
-        # Create base kwargs
-        kwargs = {"strict": strict}
-
-        # Add device parameter for newer versions that support it
-        if packaging.version.parse(safetensors.__version__) >= packaging.version.parse("0.4.3"):
-            kwargs["device"] = map_location
-
-        # Load the model with appropriate kwargs
-        missing_keys, unexpected_keys = load_model_as_safetensor(model, model_file, **kwargs)
+        missing_keys, unexpected_keys = load_model_as_safetensor(
+            model, model_file, strict=strict, device=resolve_safetensors_device(map_location)
+        )
         log_model_loading_keys(missing_keys, unexpected_keys)
-
-        # For older versions, manually move to device if needed
-        if "device" not in kwargs and map_location != "cpu":
-            logging.warning(
-                "Loading model weights on other devices than 'cpu' is not supported natively in your version of safetensors."
-                " This means that the model is loaded on 'cpu' first and then copied to the device."
-                " This leads to a slower loading time."
-                " Please update safetensors to version 0.4.3 or above for improved performance."
-            )
-            model.to(map_location)
         return model
 
     @abc.abstractmethod

src/lerobot/rewards/pretrained.py

@@ -21,8 +21,6 @@ from pathlib import Path
 from tempfile import TemporaryDirectory
 from typing import TYPE_CHECKING, Any, TypeVar
 
-import packaging
-import safetensors
 from huggingface_hub import HfApi, ModelCard, ModelCardData, hf_hub_download
 from huggingface_hub.constants import SAFETENSORS_SINGLE_FILE
 from huggingface_hub.errors import HfHubHTTPError
@@ -30,6 +28,7 @@ from safetensors.torch import load_model as load_model_as_safetensor, save_model
 from torch import Tensor, nn
 
 from lerobot.configs.rewards import RewardModelConfig
+from lerobot.utils.device_utils import resolve_safetensors_device
 from lerobot.utils.hub import HubMixin
 
 if TYPE_CHECKING:
@@ -129,29 +128,13 @@ class PreTrainedRewardModel(nn.Module, HubMixin, abc.ABC):
 
     @classmethod
     def _load_as_safetensor(cls, model: T, model_file: str, map_location: str, strict: bool) -> T:
-        # Create base kwargs
-        kwargs = {"strict": strict}
-
-        # Add device parameter for newer versions that support it
-        if packaging.version.parse(safetensors.__version__) >= packaging.version.parse("0.4.3"):
-            kwargs["device"] = map_location
-
-        # Load the model with appropriate kwargs
-        missing_keys, unexpected_keys = load_model_as_safetensor(model, model_file, **kwargs)
+        missing_keys, unexpected_keys = load_model_as_safetensor(
+            model, model_file, strict=strict, device=resolve_safetensors_device(map_location)
+        )
         if missing_keys:
             logging.warning(f"Missing key(s) when loading model: {missing_keys}")
         if unexpected_keys:
             logging.warning(f"Unexpected key(s) when loading model: {unexpected_keys}")
-
-        # For older versions, manually move to device if needed
-        if "device" not in kwargs and map_location != "cpu":
-            logging.warning(
-                "Loading model weights on other devices than 'cpu' is not supported natively in your version of safetensors."
-                " This means that the model is loaded on 'cpu' first and then copied to the device."
-                " This leads to a slower loading time."
-                " Please update safetensors to version 0.4.3 or above for improved performance."
-            )
-            model.to(map_location)
         return model
 
     def get_optim_params(self):

src/lerobot/scripts/lerobot_annotate.py

@@ -28,7 +28,12 @@ For distributed runs, see ``examples/annotations/run_hf_job.py``.
 """
 
 import logging
+from contextlib import suppress
 from pathlib import Path
+from typing import TYPE_CHECKING
+
+from huggingface_hub import HfApi, snapshot_download
+from huggingface_hub.errors import RevisionNotFoundError
 
 from lerobot.annotations.steerable_pipeline.config import AnnotationPipelineConfig
 from lerobot.annotations.steerable_pipeline.executor import Executor
@@ -42,6 +47,12 @@ from lerobot.annotations.steerable_pipeline.validator import StagingValidator
 from lerobot.annotations.steerable_pipeline.vlm_client import make_vlm_client
 from lerobot.annotations.steerable_pipeline.writer import LanguageColumnsWriter
 from lerobot.configs import parser
+from lerobot.utils.import_utils import _datasets_available, require_package
+
+if TYPE_CHECKING or _datasets_available:
+    from lerobot.datasets.dataset_metadata import CODEBASE_VERSION
+    from lerobot.datasets.io_utils import load_info
+    from lerobot.datasets.utils import create_lerobot_dataset_card
 
 logger = logging.getLogger(__name__)
 
@@ -50,8 +61,6 @@ def _resolve_root(cfg: AnnotationPipelineConfig) -> Path:
     if cfg.root is not None:
         return Path(cfg.root)
     if cfg.repo_id is not None:
-        from huggingface_hub import snapshot_download
-
         return Path(snapshot_download(repo_id=cfg.repo_id, repo_type="dataset"))
     raise ValueError("Either --root or --repo_id must be provided.")
 
@@ -125,10 +134,7 @@ def _push_to_hub(root: Path, cfg: AnnotationPipelineConfig) -> None:
 
     Pushes to ``cfg.new_repo_id`` when set, otherwise back to ``cfg.repo_id``.
     """
-    from huggingface_hub import HfApi  # noqa: PLC0415
-
-    from lerobot.datasets.io_utils import load_info  # noqa: PLC0415
-    from lerobot.datasets.utils import create_lerobot_dataset_card  # noqa: PLC0415
+    require_package("datasets", "dataset")
 
     repo_id = cfg.new_repo_id or cfg.repo_id
     commit_message = cfg.push_commit_message or "Add steerable annotations (lerobot-annotate)"
@@ -163,8 +169,6 @@ def _push_to_hub(root: Path, cfg: AnnotationPipelineConfig) -> None:
     # ``RevisionNotFoundError``. Read the version straight from the
     # dataset's own ``meta/info.json`` so we tag whatever the writer
     # actually wrote (no accidental drift if the codebase floor moves).
-    from lerobot.datasets.dataset_metadata import CODEBASE_VERSION  # noqa: PLC0415
-
     version_tag = (
         dataset_info.codebase_version if dataset_info.codebase_version.startswith("v") else CODEBASE_VERSION
     )
@@ -178,10 +182,6 @@ def _push_to_hub(root: Path, cfg: AnnotationPipelineConfig) -> None:
         tag_kwargs["revision"] = revision
 
     try:
-        from contextlib import suppress  # noqa: PLC0415
-
-        from huggingface_hub.errors import RevisionNotFoundError  # noqa: PLC0415
-
         with suppress(RevisionNotFoundError):
             api.delete_tag(repo_id, tag=version_tag, repo_type="dataset")
         api.create_tag(**tag_kwargs)

src/lerobot/scripts/lerobot_train.py

@@ -171,6 +171,9 @@ def update_policy(
     train_metrics.update_s = time.perf_counter() - start_time
     if torch.cuda.is_available():
         train_metrics.gpu_mem_gb = torch.cuda.max_memory_allocated() / (1024**3)
+    # Aggregate the policy's scalar outputs for logging and rank-reduction across the log window.
+    if output_dict:
+        train_metrics.update_metrics(output_dict)
     return train_metrics, output_dict
 
 
@@ -572,7 +575,7 @@ def train(cfg: TrainPipelineConfig, accelerator: "Accelerator | None" = None):
         batch = preprocessor(batch)
         train_tracker.dataloading_s = time.perf_counter() - start_time
 
-        train_tracker, output_dict = update_policy(
+        train_tracker, _ = update_policy(
             train_tracker,
             policy,
             batch,
@@ -605,9 +608,10 @@ def train(cfg: TrainPipelineConfig, accelerator: "Accelerator | None" = None):
                     train_tracker.samples_per_s = effective_batch_size / step_time
                 logging.info(train_tracker)
                 if wandb_logger:
+                    # Policy sub-losses (latent_loss, action_loss, ...) are aggregated into the
+                    # tracker by update_policy, so to_dict() already carries their windowed,
+                    # rank-reduced averages — no per-step output_dict passthrough needed.
                     wandb_log_dict = train_tracker.to_dict()
-                    if output_dict:
-                        wandb_log_dict.update(output_dict)
                     # Log sample weighting statistics if enabled
                     if sample_weighter is not None:
                         weighter_stats = sample_weighter.get_stats()

src/lerobot/utils/device_utils.py

@@ -59,6 +59,20 @@ def get_safe_torch_device(try_device: str, log: bool = False) -> torch.device:
     return device
 
 
+def resolve_safetensors_device(map_location: str | torch.device) -> str:
+    """Resolve a device string for a safetensors load, working around a device-mapping quirk.
+
+    safetensors' load maps the bare string "cuda" to cuda:0 regardless of the current device
+    (unlike torch's .to("cuda"), which honors torch.cuda.current_device()). Under multi-GPU
+    accelerate/FSDP every rank would then load its weights onto GPU 0, OOMing it before sharding.
+    Resolve "cuda" to the concrete current-device index so each rank loads onto its own GPU.
+    """
+    map_location = str(map_location)
+    if map_location == "cuda" and torch.cuda.is_available():
+        return f"cuda:{torch.cuda.current_device()}"
+    return map_location
+
+
 def get_safe_dtype(dtype: torch.dtype, device: str | torch.device):
     """
     mps is currently not compatible with float64

src/lerobot/utils/logging_utils.py

@@ -104,6 +104,7 @@ class MetricsTracker:
         "episodes",
         "epochs",
         "accelerator",
+        "_caller_metrics",
     ]
 
     def __init__(
@@ -129,6 +130,9 @@ class MetricsTracker:
         self.episodes = self.samples / self._avg_samples_per_ep
         self.epochs = self.samples / self._num_frames
         self.accelerator = accelerator
+        # Meter names the caller registered up front. update_metrics() leaves these untouched, so a
+        # policy that echoes e.g. "loss" in its output dict can't clobber the aggregated meter.
+        self._caller_metrics: set[str] = set(self.metrics)
 
     def __getattr__(self, name: str) -> int | dict[str, AverageMeter] | AverageMeter | Any:
         if name in self.__dict__:
@@ -156,6 +160,21 @@ class MetricsTracker:
         self.episodes = self.samples / self._avg_samples_per_ep
         self.epochs = self.samples / self._num_frames
 
+    def update_metrics(self, values: dict[str, Any]) -> None:
+        """Accumulate a dict of scalar metrics, auto-registering a meter for each new key.
+
+        Non-numeric values and bools are ignored.
+        Caller-registered metrics (those passed to the constructor) are never overridden.
+        """
+        for name, value in values.items():
+            if isinstance(value, bool) or not isinstance(value, (int, float)):
+                continue
+            if name in self._caller_metrics:
+                continue
+            if name not in self.metrics:
+                self.metrics[name] = AverageMeter(name, ":.3f", reduction="mean")
+            self.metrics[name].update(float(value))
+
     def reduce_across_ranks(self) -> None:
         """
         Synchronises the running averages of every metric whose ``reduction`` is not ``"none"``

tests/policies/test_peft_base_model_loading.py

@@ -0,0 +1,142 @@
+# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Regression tests for PEFT loading when the checkpoint is a base model (see issue #3975).
+
+Starting a fresh LoRA/PEFT fine-tune points ``--policy.path`` at a *base model* (no
+``adapter_config.json``) while also setting ``use_peft=True``. This must NOT be mistaken for
+loading an existing PEFT adapter. These tests lock in the base-model vs. adapter distinction
+made by ``lerobot.policies.factory._has_peft_adapter_config`` and the branch it drives in
+``make_policy``. They are pure/fast (no network, no ``peft``), so they run in CI.
+"""
+
+import json
+from unittest.mock import MagicMock, patch
+
+import torch
+from huggingface_hub.errors import HfHubHTTPError
+
+from lerobot.policies.factory import _has_peft_adapter_config
+
+
+def test_local_base_model_dir_has_no_adapter_config(tmp_path):
+    # A base-model checkpoint directory (only model weights, no adapter config).
+    (tmp_path / "model.safetensors").write_bytes(b"")
+    (tmp_path / "config.json").write_text("{}")
+    assert _has_peft_adapter_config(str(tmp_path)) is False
+
+
+def test_local_adapter_dir_has_adapter_config(tmp_path):
+    (tmp_path / "adapter_config.json").write_text(json.dumps({"peft_type": "LORA"}))
+    (tmp_path / "adapter_model.safetensors").write_bytes(b"")
+    assert _has_peft_adapter_config(str(tmp_path)) is True
+
+
+def test_hub_base_model_repo_has_no_adapter_config():
+    with patch("huggingface_hub.file_exists", return_value=False) as mock_exists:
+        assert _has_peft_adapter_config("lerobot/lingbot_va_base") is False
+    mock_exists.assert_called_once()
+    assert mock_exists.call_args.args[1] == "adapter_config.json"
+
+
+def test_hub_adapter_repo_has_adapter_config():
+    with patch("huggingface_hub.file_exists", return_value=True):
+        assert _has_peft_adapter_config("some/adapter-repo", revision="main") is True
+
+
+def test_hub_lookup_error_falls_back_to_base_model():
+    # Offline / private / transient Hub errors must not crash; treat as "not an adapter".
+    with patch(
+        "huggingface_hub.file_exists",
+        side_effect=HfHubHTTPError("boom", response=MagicMock()),
+    ):
+        assert _has_peft_adapter_config("some/private-repo") is False
+    with patch("huggingface_hub.file_exists", side_effect=OSError("offline")):
+        assert _has_peft_adapter_config("some/repo") is False
+
+
+def _make_dummy_policy_cfg(pretrained_path, use_peft):
+    cfg = MagicMock()
+    cfg.type = "act"
+    cfg.device = "cpu"
+    cfg.pretrained_path = pretrained_path
+    cfg.pretrained_revision = None
+    cfg.use_peft = use_peft
+    cfg.input_features = {}
+    cfg.output_features = {}
+    return cfg
+
+
+@patch("lerobot.policies.factory.validate_visual_features_consistency")
+@patch("lerobot.policies.factory.env_to_policy_features", return_value={})
+@patch("lerobot.policies.factory.get_policy_class")
+def test_make_policy_base_model_with_use_peft_loads_base_not_adapter(
+    mock_get_cls, _mock_features, _mock_validate
+):
+    """`use_peft=True` on a base model must load the base weights, not a PEFT adapter.
+
+    Before the #3975 fix this went down the ``PeftConfig.from_pretrained`` path and failed
+    looking for a non-existent ``adapter_config.json``.
+    """
+    from lerobot.policies import factory
+
+    policy_cls = MagicMock()
+    loaded_policy = torch.nn.Linear(1, 1)  # a real nn.Module so make_policy's assert passes
+    policy_cls.from_pretrained.return_value = loaded_policy
+    mock_get_cls.return_value = policy_cls
+
+    cfg = _make_dummy_policy_cfg(pretrained_path="lerobot/lingbot_va_base", use_peft=True)
+    env_cfg = MagicMock()
+
+    with patch.object(factory, "_has_peft_adapter_config", return_value=False) as mock_has_adapter:
+        policy = factory.make_policy(cfg=cfg, env_cfg=env_cfg)
+
+    mock_has_adapter.assert_called_once()
+    # Base model is loaded via the normal pretrained path...
+    policy_cls.from_pretrained.assert_called_once()
+    assert policy_cls.from_pretrained.call_args.kwargs["pretrained_name_or_path"] == (
+        "lerobot/lingbot_va_base"
+    )
+    # ...and PEFT adapter loading is NOT attempted (would need peft + adapter_config.json).
+    assert policy is loaded_policy
+
+
+@patch("lerobot.policies.factory.validate_visual_features_consistency")
+@patch("lerobot.policies.factory.env_to_policy_features", return_value={})
+@patch("lerobot.policies.factory.get_policy_class")
+def test_make_policy_existing_adapter_uses_peft_loading(mock_get_cls, _mock_features, _mock_validate):
+    """A real adapter checkpoint (has ``adapter_config.json``) must go through PEFT loading."""
+    from lerobot.policies import factory
+
+    policy_cls = MagicMock()
+    mock_get_cls.return_value = policy_cls
+
+    cfg = _make_dummy_policy_cfg(pretrained_path="some/adapter-repo", use_peft=True)
+    env_cfg = MagicMock()
+
+    policy_cls.from_pretrained.return_value = torch.nn.Linear(1, 1)
+
+    fake_peft = MagicMock()
+    fake_peft_config = MagicMock()
+    fake_peft_config.base_model_name_or_path = "lerobot/lingbot_va_base"
+    fake_peft.PeftConfig.from_pretrained.return_value = fake_peft_config
+    fake_peft.PeftModel.from_pretrained.return_value = torch.nn.Linear(1, 1)
+
+    with (
+        patch.object(factory, "_has_peft_adapter_config", return_value=True),
+        patch.dict("sys.modules", {"peft": fake_peft}),
+    ):
+        factory.make_policy(cfg=cfg, env_cfg=env_cfg)
+
+    fake_peft.PeftConfig.from_pretrained.assert_called_once_with("some/adapter-repo")
+    fake_peft.PeftModel.from_pretrained.assert_called_once()

tests/scripts/test_lerobot_annotate.py

@@ -18,6 +18,8 @@ import json
 from types import SimpleNamespace
 
 import pytest
+import requests
+from huggingface_hub.errors import RevisionNotFoundError
 
 # ``lerobot.scripts.lerobot_annotate`` (and the ``_push_to_hub`` path it
 # exercises) imports ``lerobot.datasets``, which only ships under the
@@ -26,7 +28,7 @@ pytest.importorskip("datasets", reason="datasets is required (install lerobot[da
 
 
 def test_push_to_hub_tags_uploaded_dataset_revision(tmp_path, monkeypatch):
-    from lerobot.scripts.lerobot_annotate import _push_to_hub
+    from lerobot.scripts import lerobot_annotate
 
     root = tmp_path / "dataset"
     (root / "meta").mkdir(parents=True)
@@ -45,9 +47,6 @@ def test_push_to_hub_tags_uploaded_dataset_revision(tmp_path, monkeypatch):
             return SimpleNamespace(oid="abc123")
 
         def delete_tag(self, repo_id, **kwargs):
-            import requests
-            from huggingface_hub.errors import RevisionNotFoundError
-
             calls["delete_tag"] = {"repo_id": repo_id, **kwargs}
             # Simulate the common case: no stale tag to delete.
             raise RevisionNotFoundError("no such tag", response=requests.Response())
@@ -55,7 +54,7 @@ def test_push_to_hub_tags_uploaded_dataset_revision(tmp_path, monkeypatch):
         def create_tag(self, **kwargs):
             calls["create_tag"] = kwargs
 
-    monkeypatch.setattr("huggingface_hub.HfApi", FakeHfApi)
+    monkeypatch.setattr(lerobot_annotate, "HfApi", FakeHfApi)
 
     def fake_card_push(self, **kwargs):
         calls["card_push"] = {"content": str(self), **kwargs}
@@ -69,7 +68,7 @@ def test_push_to_hub_tags_uploaded_dataset_revision(tmp_path, monkeypatch):
         push_commit_message=None,
     )
 
-    _push_to_hub(root, cfg)
+    lerobot_annotate._push_to_hub(root, cfg)
 
     assert calls["create_repo"] == {
         "repo_id": "annotated/dataset",

tests/utils/test_logging_utils.py

@@ -233,3 +233,37 @@ def test_metrics_tracker_reduce_across_ranks_invokes_reduce():
     # accumulate against the cluster view rather than the stale per-rank sum.
     meter = tracker.update_s
     assert meter.sum / meter.count == pytest.approx(meter.avg)
+
+
+def test_metrics_tracker_update_metrics_registers_and_averages():
+    tracker = MetricsTracker(batch_size=32, num_frames=1000, num_episodes=50, metrics={})
+    tracker.update_metrics({"latent_loss": 0.2, "action_loss": 0.4})
+    tracker.update_metrics({"latent_loss": 0.4, "action_loss": 0.6})
+
+    # New keys are auto-registered as mean-reduced meters and averaged over the window.
+    assert tracker.metrics["latent_loss"].reduction == "mean"
+    assert tracker.metrics["latent_loss"].avg == pytest.approx(0.3)
+    assert tracker.metrics["action_loss"].avg == pytest.approx(0.5)
+    assert tracker.to_dict()["latent_loss"] == pytest.approx(0.3)
+
+
+def test_metrics_tracker_update_metrics_skips_non_numeric():
+    tracker = MetricsTracker(batch_size=32, num_frames=1000, num_episodes=50, metrics={})
+    tracker.update_metrics({"loss": 0.5, "head_mode": "sparse", "enabled": True})
+
+    # strings and bools ignored
+    assert "loss" in tracker.metrics
+    assert "head_mode" not in tracker.metrics
+    assert "enabled" not in tracker.metrics
+
+
+def test_metrics_tracker_update_metrics_does_not_override_caller_meter():
+    # A policy that echoes "loss" in its output dict must not overwrite the caller-owned,
+    # already-aggregated loss meter.
+    metrics = {"loss": AverageMeter("loss", ":.3f", reduction="mean")}
+    tracker = MetricsTracker(batch_size=32, num_frames=1000, num_episodes=50, metrics=metrics)
+    tracker.loss = 1.0  # caller-set optimized loss
+    tracker.update_metrics({"loss": 99.0, "latent_loss": 0.2})
+
+    assert tracker.metrics["loss"].avg == pytest.approx(1.0)  # snapshot ignored
+    assert tracker.metrics["latent_loss"].avg == pytest.approx(0.2)