mirror of
https://github.com/huggingface/lerobot.git
synced 2026-06-03 20:31:25 +00:00
96 lines
4.0 KiB
Python
96 lines
4.0 KiB
Python
#!/usr/bin/env python
|
|
|
|
# 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.
|
|
|
|
from dataclasses import dataclass, field
|
|
|
|
from lerobot import (
|
|
policies, # noqa: F401
|
|
)
|
|
from lerobot.datasets.transforms import ImageTransformsConfig
|
|
from lerobot.datasets.video_utils import get_safe_default_codec
|
|
|
|
|
|
@dataclass
|
|
class DatasetConfig:
|
|
# You may provide a list of datasets here. `train.py` creates them all and concatenates them. Note: only data
|
|
# keys common between the datasets are kept. Each dataset gets and additional transform that inserts the
|
|
# "dataset_index" into the returned item. The index mapping is made according to the order in which the
|
|
# datasets are provided.
|
|
repo_id: str
|
|
# Root directory where the dataset will be stored (e.g. 'dataset/path').
|
|
root: str | None = None
|
|
episodes: list[int] | None = None
|
|
image_transforms: ImageTransformsConfig = field(default_factory=ImageTransformsConfig)
|
|
revision: str | None = None
|
|
use_imagenet_stats: bool = True
|
|
video_backend: str = field(default_factory=get_safe_default_codec)
|
|
|
|
|
|
@dataclass
|
|
class WandBConfig:
|
|
enable: bool = False
|
|
# Set to true to disable saving an artifact despite training.save_checkpoint=True
|
|
disable_artifact: bool = False
|
|
project: str = "lerobot"
|
|
entity: str | None = None
|
|
notes: str | None = None
|
|
run_id: str | None = None
|
|
mode: str | None = None # Allowed values: 'online', 'offline' 'disabled'. Defaults to 'online'
|
|
|
|
|
|
@dataclass
|
|
class EvalConfig:
|
|
n_episodes: int = 50
|
|
# `batch_size` specifies the number of environments to use in a gym.vector.VectorEnv.
|
|
batch_size: int = 50
|
|
# `use_async_envs` specifies whether to use asynchronous environments (multiprocessing).
|
|
use_async_envs: bool = False
|
|
|
|
def __post_init__(self):
|
|
if self.batch_size > self.n_episodes:
|
|
raise ValueError(
|
|
"The eval batch size is greater than the number of eval episodes "
|
|
f"({self.batch_size} > {self.n_episodes}). As a result, {self.batch_size} "
|
|
f"eval environments will be instantiated, but only {self.n_episodes} will be used. "
|
|
"This might significantly slow down evaluation. To fix this, you should update your command "
|
|
f"to increase the number of episodes to match the batch size (e.g. `eval.n_episodes={self.batch_size}`), "
|
|
f"or lower the batch size (e.g. `eval.batch_size={self.n_episodes}`)."
|
|
)
|
|
|
|
|
|
@dataclass
|
|
class PeftConfig:
|
|
# PEFT offers many methods, layer adapters are the most common and currently also the most effective methods so
|
|
# we'll focus on those in this high-level config interface.
|
|
|
|
# `target_modules` can be set by the user but default to specific values depending on the used policy. See
|
|
# `get_peft_configuration` in `scripts/train.py`.
|
|
#
|
|
target_modules: list[str] | str | None = None
|
|
|
|
# Similarly to `target_modules` this will have policy-dependent defaults which the user can override.
|
|
modules_to_save: list[str] | None = None
|
|
|
|
# The PEFT (adapter) method to apply to the policy.
|
|
method_type: str = "LORA"
|
|
|
|
# Adapter initialization method. Look at the specific adapter method documentation for defaults.
|
|
init_type: str | None = None
|
|
|
|
# We expect that all adapters are in some way doing rank-decomposition. This is not true, there are several
|
|
# methods that don't but we're focussing on these methods for now.
|
|
r: int = 16
|