Merge remote-tracking branch 'origin/main' into user/aliberts/2024_05_14_compare_policies
This commit is contained in:
Simon Alibert
commit
956f035d16
|
|
@ -1,2 +1,6 @@
|
|||
*.memmap filter=lfs diff=lfs merge=lfs -text
|
||||
*.stl filter=lfs diff=lfs merge=lfs -text
|
||||
*.safetensors filter=lfs diff=lfs merge=lfs -text
|
||||
*.mp4 filter=lfs diff=lfs merge=lfs -text
|
||||
*.arrow filter=lfs diff=lfs merge=lfs -text
|
||||
*.json filter=lfs diff=lfs merge=lfs -text
|
||||
|
|
|
|||
|
|
@ -29,6 +29,8 @@ jobs:
|
|||
MUJOCO_GL: egl
|
||||
steps:
|
||||
- uses: actions/checkout@v4
|
||||
with:
|
||||
lfs: true # Ensure LFS files are pulled
|
||||
|
||||
- name: Install EGL
|
||||
run: sudo apt-get update && sudo apt-get install -y libegl1-mesa-dev
|
||||
|
|
@ -65,6 +67,8 @@ jobs:
|
|||
MUJOCO_GL: egl
|
||||
steps:
|
||||
- uses: actions/checkout@v4
|
||||
with:
|
||||
lfs: true # Ensure LFS files are pulled
|
||||
|
||||
- name: Install poetry
|
||||
run: |
|
||||
|
|
@ -97,6 +101,8 @@ jobs:
|
|||
MUJOCO_GL: egl
|
||||
steps:
|
||||
- uses: actions/checkout@v4
|
||||
with:
|
||||
lfs: true # Ensure LFS files are pulled
|
||||
|
||||
- name: Install EGL
|
||||
run: sudo apt-get update && sudo apt-get install -y libegl1-mesa-dev
|
||||
|
|
|
|||
|
|
@ -195,6 +195,11 @@ Follow these steps to start contributing:
|
|||
git commit
|
||||
```
|
||||
|
||||
Note, if you already commited some changes that have a wrong formatting, you can use:
|
||||
```bash
|
||||
pre-commit run --all-files
|
||||
```
|
||||
|
||||
Please write [good commit messages](https://chris.beams.io/posts/git-commit/).
|
||||
|
||||
It is a good idea to sync your copy of the code with the original
|
||||
|
|
|
|||
38
Makefile
38
Makefile
|
|
@ -20,6 +20,8 @@ build-gpu:
|
|||
test-end-to-end:
|
||||
${MAKE} test-act-ete-train
|
||||
${MAKE} test-act-ete-eval
|
||||
${MAKE} test-act-ete-train-amp
|
||||
${MAKE} test-act-ete-eval-amp
|
||||
${MAKE} test-diffusion-ete-train
|
||||
${MAKE} test-diffusion-ete-eval
|
||||
${MAKE} test-tdmpc-ete-train
|
||||
|
|
@ -29,6 +31,7 @@ test-end-to-end:
|
|||
test-act-ete-train:
|
||||
python lerobot/scripts/train.py \
|
||||
policy=act \
|
||||
policy.dim_model=64 \
|
||||
env=aloha \
|
||||
wandb.enable=False \
|
||||
training.offline_steps=2 \
|
||||
|
|
@ -51,9 +54,40 @@ test-act-ete-eval:
|
|||
env.episode_length=8 \
|
||||
device=cpu \
|
||||
|
||||
test-act-ete-train-amp:
|
||||
python lerobot/scripts/train.py \
|
||||
policy=act \
|
||||
policy.dim_model=64 \
|
||||
env=aloha \
|
||||
wandb.enable=False \
|
||||
training.offline_steps=2 \
|
||||
training.online_steps=0 \
|
||||
eval.n_episodes=1 \
|
||||
eval.batch_size=1 \
|
||||
device=cpu \
|
||||
training.save_model=true \
|
||||
training.save_freq=2 \
|
||||
policy.n_action_steps=20 \
|
||||
policy.chunk_size=20 \
|
||||
training.batch_size=2 \
|
||||
hydra.run.dir=tests/outputs/act/ \
|
||||
use_amp=true
|
||||
|
||||
test-act-ete-eval-amp:
|
||||
python lerobot/scripts/eval.py \
|
||||
-p tests/outputs/act/checkpoints/000002 \
|
||||
eval.n_episodes=1 \
|
||||
eval.batch_size=1 \
|
||||
env.episode_length=8 \
|
||||
device=cpu \
|
||||
use_amp=true
|
||||
|
||||
test-diffusion-ete-train:
|
||||
python lerobot/scripts/train.py \
|
||||
policy=diffusion \
|
||||
policy.down_dims=\[64,128,256\] \
|
||||
policy.diffusion_step_embed_dim=32 \
|
||||
policy.num_inference_steps=10 \
|
||||
env=pusht \
|
||||
wandb.enable=False \
|
||||
training.offline_steps=2 \
|
||||
|
|
@ -74,6 +108,7 @@ test-diffusion-ete-eval:
|
|||
env.episode_length=8 \
|
||||
device=cpu \
|
||||
|
||||
# TODO(alexander-soare): Restore online_steps to 2 when it is reinstated.
|
||||
test-tdmpc-ete-train:
|
||||
python lerobot/scripts/train.py \
|
||||
policy=tdmpc \
|
||||
|
|
@ -82,7 +117,7 @@ test-tdmpc-ete-train:
|
|||
dataset_repo_id=lerobot/xarm_lift_medium \
|
||||
wandb.enable=False \
|
||||
training.offline_steps=2 \
|
||||
training.online_steps=2 \
|
||||
training.online_steps=0 \
|
||||
eval.n_episodes=1 \
|
||||
eval.batch_size=1 \
|
||||
env.episode_length=2 \
|
||||
|
|
@ -100,7 +135,6 @@ test-tdmpc-ete-eval:
|
|||
env.episode_length=8 \
|
||||
device=cpu \
|
||||
|
||||
|
||||
test-default-ete-eval:
|
||||
python lerobot/scripts/eval.py \
|
||||
--config lerobot/configs/default.yaml \
|
||||
|
|
|
|||
22
README.md
22
README.md
|
|
@ -92,6 +92,8 @@ To use [Weights and Biases](https://docs.wandb.ai/quickstart) for experiment tra
|
|||
wandb login
|
||||
```
|
||||
|
||||
(note: you will also need to enable WandB in the configuration. See below.)
|
||||
|
||||
## Walkthrough
|
||||
|
||||
```
|
||||
|
|
@ -158,13 +160,13 @@ See `python lerobot/scripts/eval.py --help` for more instructions.
|
|||
|
||||
Check out [example 3](./examples/3_train_policy.py) that illustrates how to start training a model.
|
||||
|
||||
In general, you can use our training script to easily train any policy. To use wandb for logging training and evaluation curves, make sure you ran `wandb login`. Here is an example of training the ACT policy on trajectories collected by humans on the Aloha simulation environment for the insertion task:
|
||||
In general, you can use our training script to easily train any policy. Here is an example of training the ACT policy on trajectories collected by humans on the Aloha simulation environment for the insertion task:
|
||||
```bash
|
||||
python lerobot/scripts/train.py \
|
||||
policy=act \
|
||||
env=aloha \
|
||||
env.task=AlohaInsertion-v0 \
|
||||
dataset_repo_id=lerobot/aloha_sim_insertion_human
|
||||
dataset_repo_id=lerobot/aloha_sim_insertion_human \
|
||||
```
|
||||
|
||||
The experiment directory is automatically generated and will show up in yellow in your terminal. It looks like `outputs/train/2024-05-05/20-21-12_aloha_act_default`. You can manually specify an experiment directory by adding this argument to the `train.py` python command:
|
||||
|
|
@ -172,17 +174,17 @@ The experiment directory is automatically generated and will show up in yellow i
|
|||
hydra.run.dir=your/new/experiment/dir
|
||||
```
|
||||
|
||||
A link to the wandb logs for the run will also show up in yellow in your terminal. Here is an example of logs from wandb:
|
||||
|
||||
To use wandb for logging training and evaluation curves, make sure you've run `wandb login` as a one-time setup step. Then, when running the training command above, enable WandB in the configuration by adding:
|
||||
|
||||
You can deactivate wandb by adding these arguments to the `train.py` python command:
|
||||
```bash
|
||||
wandb.disable_artifact=true \
|
||||
wandb.enable=false
|
||||
wandb.enable=true
|
||||
```
|
||||
|
||||
Note: For efficiency, during training every checkpoint is evaluated on a low number of episodes. After training, you may want to re-evaluate your best checkpoints on more episodes or change the evaluation settings. See `python lerobot/scripts/eval.py --help` for more instructions.
|
||||
A link to the wandb logs for the run will also show up in yellow in your terminal. Here is an example of what they look like in your browser:
|
||||
|
||||
|
||||
|
||||
Note: For efficiency, during training every checkpoint is evaluated on a low number of episodes. After training, you may want to re-evaluate your best checkpoints on more episodes or change the evaluation settings. See `python lerobot/scripts/eval.py --help` for more instructions.
|
||||
|
||||
## Contribute
|
||||
|
||||
|
|
@ -195,11 +197,11 @@ To add a dataset to the hub, you need to login using a write-access token, which
|
|||
huggingface-cli login --token ${HUGGINGFACE_TOKEN} --add-to-git-credential
|
||||
```
|
||||
|
||||
Then move your dataset folder in `data` directory (e.g. `data/aloha_ping_pong`), and push your dataset to the hub with:
|
||||
Then move your dataset folder in `data` directory (e.g. `data/aloha_static_pingpong_test`), and push your dataset to the hub with:
|
||||
```bash
|
||||
python lerobot/scripts/push_dataset_to_hub.py \
|
||||
--data-dir data \
|
||||
--dataset-id aloha_ping_ping \
|
||||
--dataset-id aloha_static_pingpong_test \
|
||||
--raw-format aloha_hdf5 \
|
||||
--community-id lerobot
|
||||
```
|
||||
|
|
|
|||
|
|
@ -8,7 +8,7 @@ ARG DEBIAN_FRONTEND=noninteractive
|
|||
RUN apt-get update && apt-get install -y --no-install-recommends \
|
||||
build-essential cmake \
|
||||
git git-lfs openssh-client \
|
||||
nano vim \
|
||||
nano vim ffmpeg \
|
||||
htop atop nvtop \
|
||||
sed gawk grep curl wget \
|
||||
tcpdump sysstat screen \
|
||||
|
|
|
|||
|
|
@ -0,0 +1,90 @@
|
|||
"""This script demonstrates how to slice a dataset and calculate the loss on a subset of the data.
|
||||
|
||||
This technique can be useful for debugging and testing purposes, as well as identifying whether a policy
|
||||
is learning effectively.
|
||||
|
||||
Furthermore, relying on validation loss to evaluate performance is generally not considered a good practice,
|
||||
especially in the context of imitation learning. The most reliable approach is to evaluate the policy directly
|
||||
on the target environment, whether that be in simulation or the real world.
|
||||
"""
|
||||
|
||||
import math
|
||||
from pathlib import Path
|
||||
|
||||
import torch
|
||||
from huggingface_hub import snapshot_download
|
||||
|
||||
from lerobot.common.datasets.lerobot_dataset import LeRobotDataset
|
||||
from lerobot.common.policies.diffusion.modeling_diffusion import DiffusionPolicy
|
||||
|
||||
device = torch.device("cuda")
|
||||
|
||||
# Download the diffusion policy for pusht environment
|
||||
pretrained_policy_path = Path(snapshot_download("lerobot/diffusion_pusht"))
|
||||
# OR uncomment the following to evaluate a policy from the local outputs/train folder.
|
||||
# pretrained_policy_path = Path("outputs/train/example_pusht_diffusion")
|
||||
|
||||
policy = DiffusionPolicy.from_pretrained(pretrained_policy_path)
|
||||
policy.eval()
|
||||
policy.to(device)
|
||||
|
||||
# Set up the dataset.
|
||||
delta_timestamps = {
|
||||
# Load the previous image and state at -0.1 seconds before current frame,
|
||||
# then load current image and state corresponding to 0.0 second.
|
||||
"observation.image": [-0.1, 0.0],
|
||||
"observation.state": [-0.1, 0.0],
|
||||
# Load the previous action (-0.1), the next action to be executed (0.0),
|
||||
# and 14 future actions with a 0.1 seconds spacing. All these actions will be
|
||||
# used to calculate the loss.
|
||||
"action": [-0.1, 0.0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0, 1.1, 1.2, 1.3, 1.4],
|
||||
}
|
||||
|
||||
# Load the last 10% of episodes of the dataset as a validation set.
|
||||
# - Load full dataset
|
||||
full_dataset = LeRobotDataset("lerobot/pusht", split="train")
|
||||
# - Calculate train and val subsets
|
||||
num_train_episodes = math.floor(full_dataset.num_episodes * 90 / 100)
|
||||
num_val_episodes = full_dataset.num_episodes - num_train_episodes
|
||||
print(f"Number of episodes in full dataset: {full_dataset.num_episodes}")
|
||||
print(f"Number of episodes in training dataset (90% subset): {num_train_episodes}")
|
||||
print(f"Number of episodes in validation dataset (10% subset): {num_val_episodes}")
|
||||
# - Get first frame index of the validation set
|
||||
first_val_frame_index = full_dataset.episode_data_index["from"][num_train_episodes].item()
|
||||
# - Load frames subset belonging to validation set using the `split` argument.
|
||||
# It utilizes the `datasets` library's syntax for slicing datasets.
|
||||
# For more information on the Slice API, please see:
|
||||
# https://huggingface.co/docs/datasets/v2.19.0/loading#slice-splits
|
||||
train_dataset = LeRobotDataset(
|
||||
"lerobot/pusht", split=f"train[:{first_val_frame_index}]", delta_timestamps=delta_timestamps
|
||||
)
|
||||
val_dataset = LeRobotDataset(
|
||||
"lerobot/pusht", split=f"train[{first_val_frame_index}:]", delta_timestamps=delta_timestamps
|
||||
)
|
||||
print(f"Number of frames in training dataset (90% subset): {len(train_dataset)}")
|
||||
print(f"Number of frames in validation dataset (10% subset): {len(val_dataset)}")
|
||||
|
||||
# Create dataloader for evaluation.
|
||||
val_dataloader = torch.utils.data.DataLoader(
|
||||
val_dataset,
|
||||
num_workers=4,
|
||||
batch_size=64,
|
||||
shuffle=False,
|
||||
pin_memory=device != torch.device("cpu"),
|
||||
drop_last=False,
|
||||
)
|
||||
|
||||
# Run validation loop.
|
||||
loss_cumsum = 0
|
||||
n_examples_evaluated = 0
|
||||
for batch in val_dataloader:
|
||||
batch = {k: v.to(device, non_blocking=True) for k, v in batch.items()}
|
||||
output_dict = policy.forward(batch)
|
||||
|
||||
loss_cumsum += output_dict["loss"].item()
|
||||
n_examples_evaluated += batch["index"].shape[0]
|
||||
|
||||
# Calculate the average loss over the validation set.
|
||||
average_loss = loss_cumsum / n_examples_evaluated
|
||||
|
||||
print(f"Average loss on validation set: {average_loss:.4f}")
|
||||
|
|
@ -61,13 +61,21 @@ available_datasets_per_env = {
|
|||
"lerobot/aloha_sim_insertion_scripted",
|
||||
"lerobot/aloha_sim_transfer_cube_human",
|
||||
"lerobot/aloha_sim_transfer_cube_scripted",
|
||||
"lerobot/aloha_sim_insertion_human_image",
|
||||
"lerobot/aloha_sim_insertion_scripted_image",
|
||||
"lerobot/aloha_sim_transfer_cube_human_image",
|
||||
"lerobot/aloha_sim_transfer_cube_scripted_image",
|
||||
],
|
||||
"pusht": ["lerobot/pusht"],
|
||||
"pusht": ["lerobot/pusht", "lerobot/pusht_image"],
|
||||
"xarm": [
|
||||
"lerobot/xarm_lift_medium",
|
||||
"lerobot/xarm_lift_medium_replay",
|
||||
"lerobot/xarm_push_medium",
|
||||
"lerobot/xarm_push_medium_replay",
|
||||
"lerobot/xarm_lift_medium_image",
|
||||
"lerobot/xarm_lift_medium_replay_image",
|
||||
"lerobot/xarm_push_medium_image",
|
||||
"lerobot/xarm_push_medium_replay_image",
|
||||
],
|
||||
}
|
||||
|
||||
|
|
|
|||
|
|
@ -20,17 +20,19 @@ import datasets
|
|||
import torch
|
||||
|
||||
from lerobot.common.datasets.utils import (
|
||||
calculate_episode_data_index,
|
||||
load_episode_data_index,
|
||||
load_hf_dataset,
|
||||
load_info,
|
||||
load_previous_and_future_frames,
|
||||
load_stats,
|
||||
load_videos,
|
||||
reset_episode_index,
|
||||
)
|
||||
from lerobot.common.datasets.video_utils import VideoFrame, load_from_videos
|
||||
|
||||
DATA_DIR = Path(os.environ["DATA_DIR"]) if "DATA_DIR" in os.environ else None
|
||||
CODEBASE_VERSION = "v1.3"
|
||||
CODEBASE_VERSION = "v1.4"
|
||||
|
||||
|
||||
class LeRobotDataset(torch.utils.data.Dataset):
|
||||
|
|
@ -54,7 +56,11 @@ class LeRobotDataset(torch.utils.data.Dataset):
|
|||
# TODO(rcadene, aliberts): implement faster transfer
|
||||
# https://huggingface.co/docs/huggingface_hub/en/guides/download#faster-downloads
|
||||
self.hf_dataset = load_hf_dataset(repo_id, version, root, split)
|
||||
self.episode_data_index = load_episode_data_index(repo_id, version, root)
|
||||
if split == "train":
|
||||
self.episode_data_index = load_episode_data_index(repo_id, version, root)
|
||||
else:
|
||||
self.episode_data_index = calculate_episode_data_index(self.hf_dataset)
|
||||
self.hf_dataset = reset_episode_index(self.hf_dataset)
|
||||
self.stats = load_stats(repo_id, version, root)
|
||||
self.info = load_info(repo_id, version, root)
|
||||
if self.video:
|
||||
|
|
|
|||
|
|
@ -24,17 +24,16 @@ import shutil
|
|||
from pathlib import Path
|
||||
|
||||
import tqdm
|
||||
|
||||
ALOHA_RAW_URLS_DIR = "lerobot/common/datasets/push_dataset_to_hub/_aloha_raw_urls"
|
||||
from huggingface_hub import snapshot_download
|
||||
|
||||
|
||||
def download_raw(raw_dir, dataset_id):
|
||||
if "pusht" in dataset_id:
|
||||
if "aloha" in dataset_id or "image" in dataset_id:
|
||||
download_hub(raw_dir, dataset_id)
|
||||
elif "pusht" in dataset_id:
|
||||
download_pusht(raw_dir)
|
||||
elif "xarm" in dataset_id:
|
||||
download_xarm(raw_dir)
|
||||
elif "aloha" in dataset_id:
|
||||
download_aloha(raw_dir, dataset_id)
|
||||
elif "umi" in dataset_id:
|
||||
download_umi(raw_dir)
|
||||
else:
|
||||
|
|
@ -103,37 +102,13 @@ def download_xarm(raw_dir: Path):
|
|||
zip_path.unlink()
|
||||
|
||||
|
||||
def download_aloha(raw_dir: Path, dataset_id: str):
|
||||
import gdown
|
||||
|
||||
subset_id = dataset_id.replace("aloha_", "")
|
||||
urls_path = Path(ALOHA_RAW_URLS_DIR) / f"{subset_id}.txt"
|
||||
assert urls_path.exists(), f"{subset_id}.txt not found in '{ALOHA_RAW_URLS_DIR}' directory."
|
||||
|
||||
with open(urls_path) as f:
|
||||
# strip lines and ignore empty lines
|
||||
urls = [url.strip() for url in f if url.strip()]
|
||||
|
||||
# sanity check
|
||||
for url in urls:
|
||||
assert (
|
||||
"drive.google.com/drive/folders" in url or "drive.google.com/file" in url
|
||||
), f"Wrong url provided '{url}' in file '{urls_path}'."
|
||||
|
||||
def download_hub(raw_dir: Path, dataset_id: str):
|
||||
raw_dir = Path(raw_dir)
|
||||
raw_dir.mkdir(parents=True, exist_ok=True)
|
||||
|
||||
logging.info(f"Start downloading from google drive for {dataset_id}")
|
||||
for url in urls:
|
||||
if "drive.google.com/drive/folders" in url:
|
||||
# when a folder url is given, download up to 50 files from the folder
|
||||
gdown.download_folder(url, output=str(raw_dir), remaining_ok=True)
|
||||
|
||||
elif "drive.google.com/file" in url:
|
||||
# because of the 50 files limit per folder, we download the remaining files (file by file)
|
||||
gdown.download(url, output=str(raw_dir), fuzzy=True)
|
||||
|
||||
logging.info(f"End downloading from google drive for {dataset_id}")
|
||||
logging.info(f"Start downloading from huggingface.co/cadene for {dataset_id}")
|
||||
snapshot_download(f"cadene/{dataset_id}_raw", repo_type="dataset", local_dir=raw_dir)
|
||||
logging.info(f"Finish downloading from huggingface.co/cadene for {dataset_id}")
|
||||
|
||||
|
||||
def download_umi(raw_dir: Path):
|
||||
|
|
@ -148,21 +123,30 @@ def download_umi(raw_dir: Path):
|
|||
if __name__ == "__main__":
|
||||
data_dir = Path("data")
|
||||
dataset_ids = [
|
||||
"pusht_image",
|
||||
"xarm_lift_medium_image",
|
||||
"xarm_lift_medium_replay_image",
|
||||
"xarm_push_medium_image",
|
||||
"xarm_push_medium_replay_image",
|
||||
"aloha_sim_insertion_human_image",
|
||||
"aloha_sim_insertion_scripted_image",
|
||||
"aloha_sim_transfer_cube_human_image",
|
||||
"aloha_sim_transfer_cube_scripted_image",
|
||||
"pusht",
|
||||
"xarm_lift_medium",
|
||||
"xarm_lift_medium_replay",
|
||||
"xarm_push_medium",
|
||||
"xarm_push_medium_replay",
|
||||
"aloha_sim_insertion_human",
|
||||
"aloha_sim_insertion_scripted",
|
||||
"aloha_sim_transfer_cube_human",
|
||||
"aloha_sim_transfer_cube_scripted",
|
||||
"aloha_mobile_cabinet",
|
||||
"aloha_mobile_chair",
|
||||
"aloha_mobile_elevator",
|
||||
"aloha_mobile_shrimp",
|
||||
"aloha_mobile_wash_pan",
|
||||
"aloha_mobile_wipe_wine",
|
||||
"aloha_sim_insertion_human",
|
||||
"aloha_sim_insertion_scripted",
|
||||
"aloha_sim_transfer_cube_human",
|
||||
"aloha_sim_transfer_cube_scripted",
|
||||
"aloha_static_battery",
|
||||
"aloha_static_candy",
|
||||
"aloha_static_coffee",
|
||||
|
|
|
|||
|
|
@ -17,7 +17,7 @@
|
|||
Contains utilities to process raw data format of HDF5 files like in: https://github.com/tonyzhaozh/act
|
||||
"""
|
||||
|
||||
import re
|
||||
import gc
|
||||
import shutil
|
||||
from pathlib import Path
|
||||
|
||||
|
|
@ -79,10 +79,8 @@ def load_from_raw(raw_dir, out_dir, fps, video, debug):
|
|||
episode_data_index = {"from": [], "to": []}
|
||||
|
||||
id_from = 0
|
||||
|
||||
for ep_path in tqdm.tqdm(hdf5_files, total=len(hdf5_files)):
|
||||
for ep_idx, ep_path in tqdm.tqdm(enumerate(hdf5_files), total=len(hdf5_files)):
|
||||
with h5py.File(ep_path, "r") as ep:
|
||||
ep_idx = int(re.search(r"episode_(\d+)", ep_path.name).group(1))
|
||||
num_frames = ep["/action"].shape[0]
|
||||
|
||||
# last step of demonstration is considered done
|
||||
|
|
@ -91,6 +89,10 @@ def load_from_raw(raw_dir, out_dir, fps, video, debug):
|
|||
|
||||
state = torch.from_numpy(ep["/observations/qpos"][:])
|
||||
action = torch.from_numpy(ep["/action"][:])
|
||||
if "/observations/qvel" in ep:
|
||||
velocity = torch.from_numpy(ep["/observations/qvel"][:])
|
||||
if "/observations/effort" in ep:
|
||||
effort = torch.from_numpy(ep["/observations/effort"][:])
|
||||
|
||||
ep_dict = {}
|
||||
|
||||
|
|
@ -131,6 +133,10 @@ def load_from_raw(raw_dir, out_dir, fps, video, debug):
|
|||
ep_dict[img_key] = [PILImage.fromarray(x) for x in imgs_array]
|
||||
|
||||
ep_dict["observation.state"] = state
|
||||
if "/observations/velocity" in ep:
|
||||
ep_dict["observation.velocity"] = velocity
|
||||
if "/observations/effort" in ep:
|
||||
ep_dict["observation.effort"] = effort
|
||||
ep_dict["action"] = action
|
||||
ep_dict["episode_index"] = torch.tensor([ep_idx] * num_frames)
|
||||
ep_dict["frame_index"] = torch.arange(0, num_frames, 1)
|
||||
|
|
@ -146,6 +152,8 @@ def load_from_raw(raw_dir, out_dir, fps, video, debug):
|
|||
|
||||
id_from += num_frames
|
||||
|
||||
gc.collect()
|
||||
|
||||
# process first episode only
|
||||
if debug:
|
||||
break
|
||||
|
|
@ -167,6 +175,14 @@ def to_hf_dataset(data_dict, video) -> Dataset:
|
|||
features["observation.state"] = Sequence(
|
||||
length=data_dict["observation.state"].shape[1], feature=Value(dtype="float32", id=None)
|
||||
)
|
||||
if "observation.velocity" in data_dict:
|
||||
features["observation.velocity"] = Sequence(
|
||||
length=data_dict["observation.velocity"].shape[1], feature=Value(dtype="float32", id=None)
|
||||
)
|
||||
if "observation.effort" in data_dict:
|
||||
features["observation.effort"] = Sequence(
|
||||
length=data_dict["observation.effort"].shape[1], feature=Value(dtype="float32", id=None)
|
||||
)
|
||||
features["action"] = Sequence(
|
||||
length=data_dict["action"].shape[1], feature=Value(dtype="float32", id=None)
|
||||
)
|
||||
|
|
|
|||
|
|
@ -14,7 +14,9 @@
|
|||
# See the License for the specific language governing permissions and
|
||||
# limitations under the License.
|
||||
import json
|
||||
import re
|
||||
from pathlib import Path
|
||||
from typing import Dict
|
||||
|
||||
import datasets
|
||||
import torch
|
||||
|
|
@ -79,7 +81,23 @@ def hf_transform_to_torch(items_dict):
|
|||
def load_hf_dataset(repo_id, version, root, split) -> datasets.Dataset:
|
||||
"""hf_dataset contains all the observations, states, actions, rewards, etc."""
|
||||
if root is not None:
|
||||
hf_dataset = load_from_disk(str(Path(root) / repo_id / split))
|
||||
hf_dataset = load_from_disk(str(Path(root) / repo_id / "train"))
|
||||
# TODO(rcadene): clean this which enables getting a subset of dataset
|
||||
if split != "train":
|
||||
if "%" in split:
|
||||
raise NotImplementedError(f"We dont support splitting based on percentage for now ({split}).")
|
||||
match_from = re.search(r"train\[(\d+):\]", split)
|
||||
match_to = re.search(r"train\[:(\d+)\]", split)
|
||||
if match_from:
|
||||
from_frame_index = int(match_from.group(1))
|
||||
hf_dataset = hf_dataset.select(range(from_frame_index, len(hf_dataset)))
|
||||
elif match_to:
|
||||
to_frame_index = int(match_to.group(1))
|
||||
hf_dataset = hf_dataset.select(range(to_frame_index))
|
||||
else:
|
||||
raise ValueError(
|
||||
f'`split` ({split}) should either be "train", "train[INT:]", or "train[:INT]"'
|
||||
)
|
||||
else:
|
||||
hf_dataset = load_dataset(repo_id, revision=version, split=split)
|
||||
hf_dataset.set_transform(hf_transform_to_torch)
|
||||
|
|
@ -245,6 +263,84 @@ def load_previous_and_future_frames(
|
|||
return item
|
||||
|
||||
|
||||
def calculate_episode_data_index(hf_dataset: datasets.Dataset) -> Dict[str, torch.Tensor]:
|
||||
"""
|
||||
Calculate episode data index for the provided HuggingFace Dataset. Relies on episode_index column of hf_dataset.
|
||||
|
||||
Parameters:
|
||||
- hf_dataset (datasets.Dataset): A HuggingFace dataset containing the episode index.
|
||||
|
||||
Returns:
|
||||
- episode_data_index: A dictionary containing the data index for each episode. The dictionary has two keys:
|
||||
- "from": A tensor containing the starting index of each episode.
|
||||
- "to": A tensor containing the ending index of each episode.
|
||||
"""
|
||||
episode_data_index = {"from": [], "to": []}
|
||||
|
||||
current_episode = None
|
||||
"""
|
||||
The episode_index is a list of integers, each representing the episode index of the corresponding example.
|
||||
For instance, the following is a valid episode_index:
|
||||
[0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2, 2]
|
||||
|
||||
Below, we iterate through the episode_index and populate the episode_data_index dictionary with the starting and
|
||||
ending index of each episode. For the episode_index above, the episode_data_index dictionary will look like this:
|
||||
{
|
||||
"from": [0, 3, 7],
|
||||
"to": [3, 7, 12]
|
||||
}
|
||||
"""
|
||||
if len(hf_dataset) == 0:
|
||||
episode_data_index = {
|
||||
"from": torch.tensor([]),
|
||||
"to": torch.tensor([]),
|
||||
}
|
||||
return episode_data_index
|
||||
for idx, episode_idx in enumerate(hf_dataset["episode_index"]):
|
||||
if episode_idx != current_episode:
|
||||
# We encountered a new episode, so we append its starting location to the "from" list
|
||||
episode_data_index["from"].append(idx)
|
||||
# If this is not the first episode, we append the ending location of the previous episode to the "to" list
|
||||
if current_episode is not None:
|
||||
episode_data_index["to"].append(idx)
|
||||
# Let's keep track of the current episode index
|
||||
current_episode = episode_idx
|
||||
else:
|
||||
# We are still in the same episode, so there is nothing for us to do here
|
||||
pass
|
||||
# We have reached the end of the dataset, so we append the ending location of the last episode to the "to" list
|
||||
episode_data_index["to"].append(idx + 1)
|
||||
|
||||
for k in ["from", "to"]:
|
||||
episode_data_index[k] = torch.tensor(episode_data_index[k])
|
||||
|
||||
return episode_data_index
|
||||
|
||||
|
||||
def reset_episode_index(hf_dataset: datasets.Dataset) -> datasets.Dataset:
|
||||
"""
|
||||
Reset the `episode_index` of the provided HuggingFace Dataset.
|
||||
|
||||
`episode_data_index` (and related functionality such as `load_previous_and_future_frames`) requires the
|
||||
`episode_index` to be sorted, continuous (1,1,1 and not 1,2,1) and start at 0.
|
||||
|
||||
This brings the `episode_index` to the required format.
|
||||
"""
|
||||
if len(hf_dataset) == 0:
|
||||
return hf_dataset
|
||||
unique_episode_idxs = torch.stack(hf_dataset["episode_index"]).unique().tolist()
|
||||
episode_idx_to_reset_idx_mapping = {
|
||||
ep_id: reset_ep_id for reset_ep_id, ep_id in enumerate(unique_episode_idxs)
|
||||
}
|
||||
|
||||
def modify_ep_idx_func(example):
|
||||
example["episode_index"] = episode_idx_to_reset_idx_mapping[example["episode_index"].item()]
|
||||
return example
|
||||
|
||||
hf_dataset = hf_dataset.map(modify_ep_idx_func)
|
||||
return hf_dataset
|
||||
|
||||
|
||||
def cycle(iterable):
|
||||
"""The equivalent of itertools.cycle, but safe for Pytorch dataloaders.
|
||||
|
||||
|
|
|
|||
|
|
@ -66,8 +66,12 @@ class ACTConfig:
|
|||
documentation in the policy class).
|
||||
latent_dim: The VAE's latent dimension.
|
||||
n_vae_encoder_layers: The number of transformer layers to use for the VAE's encoder.
|
||||
use_temporal_aggregation: Whether to blend the actions of multiple policy invocations for any given
|
||||
environment step.
|
||||
temporal_ensemble_momentum: Exponential moving average (EMA) momentum parameter (α) for ensembling
|
||||
actions for a given time step over multiple policy invocations. Updates are calculated as:
|
||||
x⁻ₙ = αx⁻ₙ₋₁ + (1-α)xₙ. Note that the ACT paper and original ACT code describes a different
|
||||
parameter here: they refer to a weighting scheme wᵢ = exp(-m⋅i) and set m = 0.01. With our
|
||||
formulation, this is equivalent to α = exp(-0.01) ≈ 0.99. When this parameter is provided, we
|
||||
require `n_action_steps == 1` (since we need to query the policy every step anyway).
|
||||
dropout: Dropout to use in the transformer layers (see code for details).
|
||||
kl_weight: The weight to use for the KL-divergence component of the loss if the variational objective
|
||||
is enabled. Loss is then calculated as: `reconstruction_loss + kl_weight * kld_loss`.
|
||||
|
|
@ -115,6 +119,9 @@ class ACTConfig:
|
|||
dim_feedforward: int = 3200
|
||||
feedforward_activation: str = "relu"
|
||||
n_encoder_layers: int = 4
|
||||
# Note: Although the original ACT implementation has 7 for `n_decoder_layers`, there is a bug in the code
|
||||
# that means only the first layer is used. Here we match the original implementation by setting this to 1.
|
||||
# See this issue https://github.com/tonyzhaozh/act/issues/25#issue-2258740521.
|
||||
n_decoder_layers: int = 1
|
||||
# VAE.
|
||||
use_vae: bool = True
|
||||
|
|
@ -122,7 +129,7 @@ class ACTConfig:
|
|||
n_vae_encoder_layers: int = 4
|
||||
|
||||
# Inference.
|
||||
use_temporal_aggregation: bool = False
|
||||
temporal_ensemble_momentum: float | None = None
|
||||
|
||||
# Training and loss computation.
|
||||
dropout: float = 0.1
|
||||
|
|
@ -134,8 +141,11 @@ class ACTConfig:
|
|||
raise ValueError(
|
||||
f"`vision_backbone` must be one of the ResNet variants. Got {self.vision_backbone}."
|
||||
)
|
||||
if self.use_temporal_aggregation:
|
||||
raise NotImplementedError("Temporal aggregation is not yet implemented.")
|
||||
if self.temporal_ensemble_momentum is not None and self.n_action_steps > 1:
|
||||
raise NotImplementedError(
|
||||
"`n_action_steps` must be 1 when using temporal ensembling. This is "
|
||||
"because the policy needs to be queried every step to compute the ensembled action."
|
||||
)
|
||||
if self.n_action_steps > self.chunk_size:
|
||||
raise ValueError(
|
||||
f"The chunk size is the upper bound for the number of action steps per model invocation. Got "
|
||||
|
|
|
|||
|
|
@ -61,7 +61,7 @@ class ACTPolicy(nn.Module, PyTorchModelHubMixin):
|
|||
super().__init__()
|
||||
if config is None:
|
||||
config = ACTConfig()
|
||||
self.config = config
|
||||
self.config: ACTConfig = config
|
||||
|
||||
self.normalize_inputs = Normalize(
|
||||
config.input_shapes, config.input_normalization_modes, dataset_stats
|
||||
|
|
@ -81,7 +81,9 @@ class ACTPolicy(nn.Module, PyTorchModelHubMixin):
|
|||
|
||||
def reset(self):
|
||||
"""This should be called whenever the environment is reset."""
|
||||
if self.config.n_action_steps is not None:
|
||||
if self.config.temporal_ensemble_momentum is not None:
|
||||
self._ensembled_actions = None
|
||||
else:
|
||||
self._action_queue = deque([], maxlen=self.config.n_action_steps)
|
||||
|
||||
@torch.no_grad
|
||||
|
|
@ -97,6 +99,28 @@ class ACTPolicy(nn.Module, PyTorchModelHubMixin):
|
|||
batch = self.normalize_inputs(batch)
|
||||
batch["observation.images"] = torch.stack([batch[k] for k in self.expected_image_keys], dim=-4)
|
||||
|
||||
# If we are doing temporal ensembling, keep track of the exponential moving average (EMA), and return
|
||||
# the first action.
|
||||
if self.config.temporal_ensemble_momentum is not None:
|
||||
actions = self.model(batch)[0] # (batch_size, chunk_size, action_dim)
|
||||
actions = self.unnormalize_outputs({"action": actions})["action"]
|
||||
if self._ensembled_actions is None:
|
||||
# Initializes `self._ensembled_action` to the sequence of actions predicted during the first
|
||||
# time step of the episode.
|
||||
self._ensembled_actions = actions.clone()
|
||||
else:
|
||||
# self._ensembled_actions will have shape (batch_size, chunk_size - 1, action_dim). Compute
|
||||
# the EMA update for those entries.
|
||||
alpha = self.config.temporal_ensemble_momentum
|
||||
self._ensembled_actions = alpha * self._ensembled_actions + (1 - alpha) * actions[:, :-1]
|
||||
# The last action, which has no prior moving average, needs to get concatenated onto the end.
|
||||
self._ensembled_actions = torch.cat([self._ensembled_actions, actions[:, -1:]], dim=1)
|
||||
# "Consume" the first action.
|
||||
action, self._ensembled_actions = self._ensembled_actions[:, 0], self._ensembled_actions[:, 1:]
|
||||
return action
|
||||
|
||||
# Action queue logic for n_action_steps > 1. When the action_queue is depleted, populate it by
|
||||
# querying the policy.
|
||||
if len(self._action_queue) == 0:
|
||||
actions = self.model(batch)[0][:, : self.config.n_action_steps]
|
||||
|
||||
|
|
|
|||
|
|
@ -10,6 +10,9 @@ hydra:
|
|||
name: default
|
||||
|
||||
device: cuda # cpu
|
||||
# `use_amp` determines whether to use Automatic Mixed Precision (AMP) for training and evaluation. With AMP,
|
||||
# automatic gradient scaling is used.
|
||||
use_amp: false
|
||||
# `seed` is used for training (eg: model initialization, dataset shuffling)
|
||||
# AND for the evaluation environments.
|
||||
seed: ???
|
||||
|
|
@ -17,6 +20,7 @@ dataset_repo_id: lerobot/pusht
|
|||
|
||||
training:
|
||||
offline_steps: ???
|
||||
# NOTE: `online_steps` is not implemented yet. It's here as a placeholder.
|
||||
online_steps: ???
|
||||
online_steps_between_rollouts: ???
|
||||
online_sampling_ratio: 0.5
|
||||
|
|
@ -35,7 +39,7 @@ eval:
|
|||
use_async_envs: false
|
||||
|
||||
wandb:
|
||||
enable: true
|
||||
enable: false
|
||||
# Set to true to disable saving an artifact despite save_model == True
|
||||
disable_artifact: false
|
||||
project: lerobot
|
||||
|
|
|
|||
|
|
@ -66,6 +66,9 @@ policy:
|
|||
dim_feedforward: 3200
|
||||
feedforward_activation: relu
|
||||
n_encoder_layers: 4
|
||||
# Note: Although the original ACT implementation has 7 for `n_decoder_layers`, there is a bug in the code
|
||||
# that means only the first layer is used. Here we match the original implementation by setting this to 1.
|
||||
# See this issue https://github.com/tonyzhaozh/act/issues/25#issue-2258740521.
|
||||
n_decoder_layers: 1
|
||||
# VAE.
|
||||
use_vae: true
|
||||
|
|
@ -73,7 +76,7 @@ policy:
|
|||
n_vae_encoder_layers: 4
|
||||
|
||||
# Inference.
|
||||
use_temporal_aggregation: false
|
||||
temporal_ensemble_momentum: null
|
||||
|
||||
# Training and loss computation.
|
||||
dropout: 0.1
|
||||
|
|
|
|||
|
|
@ -5,7 +5,8 @@ dataset_repo_id: lerobot/xarm_lift_medium
|
|||
|
||||
training:
|
||||
offline_steps: 25000
|
||||
online_steps: 25000
|
||||
# TODO(alexander-soare): uncomment when online training gets reinstated
|
||||
online_steps: 0 # 25000 not implemented yet
|
||||
eval_freq: 5000
|
||||
online_steps_between_rollouts: 1
|
||||
online_sampling_ratio: 0.5
|
||||
|
|
|
|||
|
|
@ -46,6 +46,7 @@ import json
|
|||
import logging
|
||||
import threading
|
||||
import time
|
||||
from contextlib import nullcontext
|
||||
from copy import deepcopy
|
||||
from datetime import datetime as dt
|
||||
from pathlib import Path
|
||||
|
|
@ -520,7 +521,7 @@ def eval(
|
|||
raise NotImplementedError()
|
||||
|
||||
# Check device is available
|
||||
get_safe_torch_device(hydra_cfg.device, log=True)
|
||||
device = get_safe_torch_device(hydra_cfg.device, log=True)
|
||||
|
||||
torch.backends.cudnn.benchmark = True
|
||||
torch.backends.cuda.matmul.allow_tf32 = True
|
||||
|
|
@ -539,16 +540,17 @@ def eval(
|
|||
policy = make_policy(hydra_cfg=hydra_cfg, dataset_stats=make_dataset(hydra_cfg).stats)
|
||||
policy.eval()
|
||||
|
||||
info = eval_policy(
|
||||
env,
|
||||
policy,
|
||||
hydra_cfg.eval.n_episodes,
|
||||
max_episodes_rendered=10,
|
||||
video_dir=Path(out_dir) / "eval",
|
||||
start_seed=hydra_cfg.seed,
|
||||
enable_progbar=True,
|
||||
enable_inner_progbar=True,
|
||||
)
|
||||
with torch.no_grad(), torch.autocast(device_type=device.type) if hydra_cfg.use_amp else nullcontext():
|
||||
info = eval_policy(
|
||||
env,
|
||||
policy,
|
||||
hydra_cfg.eval.n_episodes,
|
||||
max_episodes_rendered=10,
|
||||
video_dir=Path(out_dir) / "eval",
|
||||
start_seed=hydra_cfg.seed,
|
||||
enable_progbar=True,
|
||||
enable_inner_progbar=True,
|
||||
)
|
||||
print(info["aggregated"])
|
||||
|
||||
# Save info
|
||||
|
|
|
|||
|
|
@ -25,7 +25,6 @@ python lerobot/scripts/push_dataset_to_hub.py \
|
|||
--dataset-id pusht \
|
||||
--raw-format pusht_zarr \
|
||||
--community-id lerobot \
|
||||
--revision v1.2 \
|
||||
--dry-run 1 \
|
||||
--save-to-disk 1 \
|
||||
--save-tests-to-disk 0 \
|
||||
|
|
@ -36,7 +35,6 @@ python lerobot/scripts/push_dataset_to_hub.py \
|
|||
--dataset-id xarm_lift_medium \
|
||||
--raw-format xarm_pkl \
|
||||
--community-id lerobot \
|
||||
--revision v1.2 \
|
||||
--dry-run 1 \
|
||||
--save-to-disk 1 \
|
||||
--save-tests-to-disk 0 \
|
||||
|
|
@ -47,7 +45,6 @@ python lerobot/scripts/push_dataset_to_hub.py \
|
|||
--dataset-id aloha_sim_insertion_scripted \
|
||||
--raw-format aloha_hdf5 \
|
||||
--community-id lerobot \
|
||||
--revision v1.2 \
|
||||
--dry-run 1 \
|
||||
--save-to-disk 1 \
|
||||
--save-tests-to-disk 0 \
|
||||
|
|
@ -58,7 +55,6 @@ python lerobot/scripts/push_dataset_to_hub.py \
|
|||
--dataset-id umi_cup_in_the_wild \
|
||||
--raw-format umi_zarr \
|
||||
--community-id lerobot \
|
||||
--revision v1.2 \
|
||||
--dry-run 1 \
|
||||
--save-to-disk 1 \
|
||||
--save-tests-to-disk 0 \
|
||||
|
|
@ -227,8 +223,7 @@ def push_dataset_to_hub(
|
|||
test_hf_dataset = test_hf_dataset.with_format(None)
|
||||
test_hf_dataset.save_to_disk(str(tests_out_dir / "train"))
|
||||
|
||||
# copy meta data to tests directory
|
||||
shutil.copytree(meta_data_dir, tests_meta_data_dir)
|
||||
save_meta_data(info, stats, episode_data_index, tests_meta_data_dir)
|
||||
|
||||
# copy videos of first episode to tests directory
|
||||
episode_index = 0
|
||||
|
|
@ -237,6 +232,10 @@ def push_dataset_to_hub(
|
|||
fname = f"{key}_episode_{episode_index:06d}.mp4"
|
||||
shutil.copy(videos_dir / fname, tests_videos_dir / fname)
|
||||
|
||||
if not save_to_disk and out_dir.exists():
|
||||
# remove possible temporary files remaining in the output directory
|
||||
shutil.rmtree(out_dir)
|
||||
|
||||
|
||||
def main():
|
||||
parser = argparse.ArgumentParser()
|
||||
|
|
@ -314,7 +313,7 @@ def main():
|
|||
parser.add_argument(
|
||||
"--num-workers",
|
||||
type=int,
|
||||
default=16,
|
||||
default=8,
|
||||
help="Number of processes of Dataloader for computing the dataset statistics.",
|
||||
)
|
||||
parser.add_argument(
|
||||
|
|
|
|||
|
|
@ -15,15 +15,14 @@
|
|||
# limitations under the License.
|
||||
import logging
|
||||
import time
|
||||
from contextlib import nullcontext
|
||||
from copy import deepcopy
|
||||
from pathlib import Path
|
||||
|
||||
import datasets
|
||||
import hydra
|
||||
import torch
|
||||
from datasets import concatenate_datasets
|
||||
from datasets.utils import disable_progress_bars, enable_progress_bars
|
||||
from omegaconf import DictConfig
|
||||
from torch.cuda.amp import GradScaler
|
||||
|
||||
from lerobot.common.datasets.factory import make_dataset
|
||||
from lerobot.common.datasets.utils import cycle
|
||||
|
|
@ -31,6 +30,7 @@ from lerobot.common.envs.factory import make_env
|
|||
from lerobot.common.logger import Logger, log_output_dir
|
||||
from lerobot.common.policies.factory import make_policy
|
||||
from lerobot.common.policies.policy_protocol import PolicyWithUpdate
|
||||
from lerobot.common.policies.utils import get_device_from_parameters
|
||||
from lerobot.common.utils.utils import (
|
||||
format_big_number,
|
||||
get_safe_torch_device,
|
||||
|
|
@ -69,7 +69,6 @@ def make_optimizer_and_scheduler(cfg, policy):
|
|||
cfg.training.adam_eps,
|
||||
cfg.training.adam_weight_decay,
|
||||
)
|
||||
assert cfg.training.online_steps == 0, "Diffusion Policy does not handle online training."
|
||||
from diffusers.optimization import get_scheduler
|
||||
|
||||
lr_scheduler = get_scheduler(
|
||||
|
|
@ -87,21 +86,40 @@ def make_optimizer_and_scheduler(cfg, policy):
|
|||
return optimizer, lr_scheduler
|
||||
|
||||
|
||||
def update_policy(policy, batch, optimizer, grad_clip_norm, lr_scheduler=None):
|
||||
def update_policy(
|
||||
policy,
|
||||
batch,
|
||||
optimizer,
|
||||
grad_clip_norm,
|
||||
grad_scaler: GradScaler,
|
||||
lr_scheduler=None,
|
||||
use_amp: bool = False,
|
||||
):
|
||||
"""Returns a dictionary of items for logging."""
|
||||
start_time = time.time()
|
||||
start_time = time.perf_counter()
|
||||
device = get_device_from_parameters(policy)
|
||||
policy.train()
|
||||
output_dict = policy.forward(batch)
|
||||
# TODO(rcadene): policy.unnormalize_outputs(out_dict)
|
||||
loss = output_dict["loss"]
|
||||
loss.backward()
|
||||
with torch.autocast(device_type=device.type) if use_amp else nullcontext():
|
||||
output_dict = policy.forward(batch)
|
||||
# TODO(rcadene): policy.unnormalize_outputs(out_dict)
|
||||
loss = output_dict["loss"]
|
||||
grad_scaler.scale(loss).backward()
|
||||
|
||||
# Unscale the graident of the optimzer's assigned params in-place **prior to gradient clipping**.
|
||||
grad_scaler.unscale_(optimizer)
|
||||
|
||||
grad_norm = torch.nn.utils.clip_grad_norm_(
|
||||
policy.parameters(),
|
||||
grad_clip_norm,
|
||||
error_if_nonfinite=False,
|
||||
)
|
||||
|
||||
optimizer.step()
|
||||
# Optimizer's gradients are already unscaled, so scaler.step does not unscale them,
|
||||
# although it still skips optimizer.step() if the gradients contain infs or NaNs.
|
||||
grad_scaler.step(optimizer)
|
||||
# Updates the scale for next iteration.
|
||||
grad_scaler.update()
|
||||
|
||||
optimizer.zero_grad()
|
||||
|
||||
if lr_scheduler is not None:
|
||||
|
|
@ -115,7 +133,7 @@ def update_policy(policy, batch, optimizer, grad_clip_norm, lr_scheduler=None):
|
|||
"loss": loss.item(),
|
||||
"grad_norm": float(grad_norm),
|
||||
"lr": optimizer.param_groups[0]["lr"],
|
||||
"update_s": time.time() - start_time,
|
||||
"update_s": time.perf_counter() - start_time,
|
||||
**{k: v for k, v in output_dict.items() if k != "loss"},
|
||||
}
|
||||
|
||||
|
|
@ -211,103 +229,6 @@ def log_eval_info(logger, info, step, cfg, dataset, is_offline):
|
|||
logger.log_dict(info, step, mode="eval")
|
||||
|
||||
|
||||
def calculate_online_sample_weight(n_off: int, n_on: int, pc_on: float):
|
||||
"""
|
||||
Calculate the sampling weight to be assigned to samples so that a specified percentage of the batch comes from online dataset (on average).
|
||||
|
||||
Parameters:
|
||||
- n_off (int): Number of offline samples, each with a sampling weight of 1.
|
||||
- n_on (int): Number of online samples.
|
||||
- pc_on (float): Desired percentage of online samples in decimal form (e.g., 50% as 0.5).
|
||||
|
||||
The total weight of offline samples is n_off * 1.0.
|
||||
The total weight of offline samples is n_on * w.
|
||||
The total combined weight of all samples is n_off + n_on * w.
|
||||
The fraction of the weight that is online is n_on * w / (n_off + n_on * w).
|
||||
We want this fraction to equal pc_on, so we set up the equation n_on * w / (n_off + n_on * w) = pc_on.
|
||||
The solution is w = - (n_off * pc_on) / (n_on * (pc_on - 1))
|
||||
"""
|
||||
assert 0.0 <= pc_on <= 1.0
|
||||
return -(n_off * pc_on) / (n_on * (pc_on - 1))
|
||||
|
||||
|
||||
def add_episodes_inplace(
|
||||
online_dataset: torch.utils.data.Dataset,
|
||||
concat_dataset: torch.utils.data.ConcatDataset,
|
||||
sampler: torch.utils.data.WeightedRandomSampler,
|
||||
hf_dataset: datasets.Dataset,
|
||||
episode_data_index: dict[str, torch.Tensor],
|
||||
pc_online_samples: float,
|
||||
):
|
||||
"""
|
||||
Modifies the online_dataset, concat_dataset, and sampler in place by integrating
|
||||
new episodes from hf_dataset into the online_dataset, updating the concatenated
|
||||
dataset's structure and adjusting the sampling strategy based on the specified
|
||||
percentage of online samples.
|
||||
|
||||
Parameters:
|
||||
- online_dataset (torch.utils.data.Dataset): The existing online dataset to be updated.
|
||||
- concat_dataset (torch.utils.data.ConcatDataset): The concatenated dataset that combines
|
||||
offline and online datasets, used for sampling purposes.
|
||||
- sampler (torch.utils.data.WeightedRandomSampler): A sampler that will be updated to
|
||||
reflect changes in the dataset sizes and specified sampling weights.
|
||||
- hf_dataset (datasets.Dataset): A Hugging Face dataset containing the new episodes to be added.
|
||||
- episode_data_index (dict): A dictionary containing two keys ("from" and "to") associated to dataset indices.
|
||||
They indicate the start index and end index of each episode in the dataset.
|
||||
- pc_online_samples (float): The target percentage of samples that should come from
|
||||
the online dataset during sampling operations.
|
||||
|
||||
Raises:
|
||||
- AssertionError: If the first episode_id or index in hf_dataset is not 0
|
||||
"""
|
||||
first_episode_idx = hf_dataset.select_columns("episode_index")[0]["episode_index"].item()
|
||||
last_episode_idx = hf_dataset.select_columns("episode_index")[-1]["episode_index"].item()
|
||||
first_index = hf_dataset.select_columns("index")[0]["index"].item()
|
||||
last_index = hf_dataset.select_columns("index")[-1]["index"].item()
|
||||
# sanity check
|
||||
assert first_episode_idx == 0, f"{first_episode_idx=} is not 0"
|
||||
assert first_index == 0, f"{first_index=} is not 0"
|
||||
assert first_index == episode_data_index["from"][first_episode_idx].item()
|
||||
assert last_index == episode_data_index["to"][last_episode_idx].item() - 1
|
||||
|
||||
if len(online_dataset) == 0:
|
||||
# initialize online dataset
|
||||
online_dataset.hf_dataset = hf_dataset
|
||||
online_dataset.episode_data_index = episode_data_index
|
||||
else:
|
||||
# get the starting indices of the new episodes and frames to be added
|
||||
start_episode_idx = last_episode_idx + 1
|
||||
start_index = last_index + 1
|
||||
|
||||
def shift_indices(episode_index, index):
|
||||
# note: we dont shift "frame_index" since it represents the index of the frame in the episode it belongs to
|
||||
example = {"episode_index": episode_index + start_episode_idx, "index": index + start_index}
|
||||
return example
|
||||
|
||||
disable_progress_bars() # map has a tqdm progress bar
|
||||
hf_dataset = hf_dataset.map(shift_indices, input_columns=["episode_index", "index"])
|
||||
enable_progress_bars()
|
||||
|
||||
episode_data_index["from"] += start_index
|
||||
episode_data_index["to"] += start_index
|
||||
|
||||
# extend online dataset
|
||||
online_dataset.hf_dataset = concatenate_datasets([online_dataset.hf_dataset, hf_dataset])
|
||||
|
||||
# update the concatenated dataset length used during sampling
|
||||
concat_dataset.cumulative_sizes = concat_dataset.cumsum(concat_dataset.datasets)
|
||||
|
||||
# update the sampling weights for each frame so that online frames get sampled a certain percentage of times
|
||||
len_online = len(online_dataset)
|
||||
len_offline = len(concat_dataset) - len_online
|
||||
weight_offline = 1.0
|
||||
weight_online = calculate_online_sample_weight(len_offline, len_online, pc_online_samples)
|
||||
sampler.weights = torch.tensor([weight_offline] * len_offline + [weight_online] * len(online_dataset))
|
||||
|
||||
# update the total number of samples used during sampling
|
||||
sampler.num_samples = len(concat_dataset)
|
||||
|
||||
|
||||
def train(cfg: DictConfig, out_dir: str | None = None, job_name: str | None = None):
|
||||
if out_dir is None:
|
||||
raise NotImplementedError()
|
||||
|
|
@ -316,11 +237,11 @@ def train(cfg: DictConfig, out_dir: str | None = None, job_name: str | None = No
|
|||
|
||||
init_logging()
|
||||
|
||||
if cfg.training.online_steps > 0 and cfg.eval.batch_size > 1:
|
||||
logging.warning("eval.batch_size > 1 not supported for online training steps")
|
||||
if cfg.training.online_steps > 0:
|
||||
raise NotImplementedError("Online training is not implemented yet.")
|
||||
|
||||
# Check device is available
|
||||
get_safe_torch_device(cfg.device, log=True)
|
||||
device = get_safe_torch_device(cfg.device, log=True)
|
||||
|
||||
torch.backends.cudnn.benchmark = True
|
||||
torch.backends.cuda.matmul.allow_tf32 = True
|
||||
|
|
@ -338,6 +259,7 @@ def train(cfg: DictConfig, out_dir: str | None = None, job_name: str | None = No
|
|||
# Create optimizer and scheduler
|
||||
# Temporary hack to move optimizer out of policy
|
||||
optimizer, lr_scheduler = make_optimizer_and_scheduler(cfg, policy)
|
||||
grad_scaler = GradScaler(enabled=cfg.use_amp)
|
||||
|
||||
num_learnable_params = sum(p.numel() for p in policy.parameters() if p.requires_grad)
|
||||
num_total_params = sum(p.numel() for p in policy.parameters())
|
||||
|
|
@ -358,14 +280,15 @@ def train(cfg: DictConfig, out_dir: str | None = None, job_name: str | None = No
|
|||
def evaluate_and_checkpoint_if_needed(step):
|
||||
if step % cfg.training.eval_freq == 0:
|
||||
logging.info(f"Eval policy at step {step}")
|
||||
eval_info = eval_policy(
|
||||
eval_env,
|
||||
policy,
|
||||
cfg.eval.n_episodes,
|
||||
video_dir=Path(out_dir) / "eval",
|
||||
max_episodes_rendered=4,
|
||||
start_seed=cfg.seed,
|
||||
)
|
||||
with torch.no_grad(), torch.autocast(device_type=device.type) if cfg.use_amp else nullcontext():
|
||||
eval_info = eval_policy(
|
||||
eval_env,
|
||||
policy,
|
||||
cfg.eval.n_episodes,
|
||||
video_dir=Path(out_dir) / "eval",
|
||||
max_episodes_rendered=4,
|
||||
start_seed=cfg.seed,
|
||||
)
|
||||
log_eval_info(logger, eval_info["aggregated"], step, cfg, offline_dataset, is_offline)
|
||||
if cfg.wandb.enable:
|
||||
logger.log_video(eval_info["video_paths"][0], step, mode="eval")
|
||||
|
|
@ -389,23 +312,30 @@ def train(cfg: DictConfig, out_dir: str | None = None, job_name: str | None = No
|
|||
num_workers=4,
|
||||
batch_size=cfg.training.batch_size,
|
||||
shuffle=True,
|
||||
pin_memory=cfg.device != "cpu",
|
||||
pin_memory=device.type != "cpu",
|
||||
drop_last=False,
|
||||
)
|
||||
dl_iter = cycle(dataloader)
|
||||
|
||||
policy.train()
|
||||
step = 0 # number of policy update (forward + backward + optim)
|
||||
is_offline = True
|
||||
for offline_step in range(cfg.training.offline_steps):
|
||||
if offline_step == 0:
|
||||
for step in range(cfg.training.offline_steps):
|
||||
if step == 0:
|
||||
logging.info("Start offline training on a fixed dataset")
|
||||
batch = next(dl_iter)
|
||||
|
||||
for key in batch:
|
||||
batch[key] = batch[key].to(cfg.device, non_blocking=True)
|
||||
batch[key] = batch[key].to(device, non_blocking=True)
|
||||
|
||||
train_info = update_policy(policy, batch, optimizer, cfg.training.grad_clip_norm, lr_scheduler)
|
||||
train_info = update_policy(
|
||||
policy,
|
||||
batch,
|
||||
optimizer,
|
||||
cfg.training.grad_clip_norm,
|
||||
grad_scaler=grad_scaler,
|
||||
lr_scheduler=lr_scheduler,
|
||||
use_amp=cfg.use_amp,
|
||||
)
|
||||
|
||||
# TODO(rcadene): is it ok if step_t=0 = 0 and not 1 as previously done?
|
||||
if step % cfg.training.log_freq == 0:
|
||||
|
|
@ -415,11 +345,6 @@ def train(cfg: DictConfig, out_dir: str | None = None, job_name: str | None = No
|
|||
# so we pass in step + 1.
|
||||
evaluate_and_checkpoint_if_needed(step + 1)
|
||||
|
||||
step += 1
|
||||
|
||||
# create an env dedicated to online episodes collection from policy rollout
|
||||
online_training_env = make_env(cfg, n_envs=1)
|
||||
|
||||
# create an empty online dataset similar to offline dataset
|
||||
online_dataset = deepcopy(offline_dataset)
|
||||
online_dataset.hf_dataset = {}
|
||||
|
|
@ -436,58 +361,11 @@ def train(cfg: DictConfig, out_dir: str | None = None, job_name: str | None = No
|
|||
num_workers=4,
|
||||
batch_size=cfg.training.batch_size,
|
||||
sampler=sampler,
|
||||
pin_memory=cfg.device != "cpu",
|
||||
pin_memory=device.type != "cpu",
|
||||
drop_last=False,
|
||||
)
|
||||
dl_iter = cycle(dataloader)
|
||||
|
||||
online_step = 0
|
||||
is_offline = False
|
||||
for env_step in range(cfg.training.online_steps):
|
||||
if env_step == 0:
|
||||
logging.info("Start online training by interacting with environment")
|
||||
|
||||
policy.eval()
|
||||
with torch.no_grad():
|
||||
eval_info = eval_policy(
|
||||
online_training_env,
|
||||
policy,
|
||||
n_episodes=1,
|
||||
return_episode_data=True,
|
||||
start_seed=cfg.training.online_env_seed,
|
||||
enable_progbar=True,
|
||||
)
|
||||
|
||||
add_episodes_inplace(
|
||||
online_dataset,
|
||||
concat_dataset,
|
||||
sampler,
|
||||
hf_dataset=eval_info["episodes"]["hf_dataset"],
|
||||
episode_data_index=eval_info["episodes"]["episode_data_index"],
|
||||
pc_online_samples=cfg.training.online_sampling_ratio,
|
||||
)
|
||||
|
||||
policy.train()
|
||||
for _ in range(cfg.training.online_steps_between_rollouts):
|
||||
batch = next(dl_iter)
|
||||
|
||||
for key in batch:
|
||||
batch[key] = batch[key].to(cfg.device, non_blocking=True)
|
||||
|
||||
train_info = update_policy(policy, batch, optimizer, cfg.training.grad_clip_norm, lr_scheduler)
|
||||
|
||||
if step % cfg.training.log_freq == 0:
|
||||
log_train_info(logger, train_info, step, cfg, online_dataset, is_offline)
|
||||
|
||||
# Note: evaluate_and_checkpoint_if_needed happens **after** the `step`th training update has completed,
|
||||
# so we pass in step + 1.
|
||||
evaluate_and_checkpoint_if_needed(step + 1)
|
||||
|
||||
step += 1
|
||||
online_step += 1
|
||||
|
||||
eval_env.close()
|
||||
online_training_env.close()
|
||||
logging.info("End of training")
|
||||
|
||||
|
||||
|
|
|
|||
|
|
@ -1320,6 +1320,78 @@ files = [
|
|||
[package.dependencies]
|
||||
numpy = ">=1.17.3"
|
||||
|
||||
[[package]]
|
||||
name = "hf-transfer"
|
||||
version = "0.1.6"
|
||||
description = ""
|
||||
optional = false
|
||||
python-versions = ">=3.7"
|
||||
files = [
|
||||
{file = "hf_transfer-0.1.6-cp310-cp310-macosx_10_12_x86_64.whl", hash = "sha256:6fd3d61f9229d27def007e53540412507b74ac2fdb1a29985ae0b6a5137749a2"},
|
||||
{file = "hf_transfer-0.1.6-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:b043bb78df1225de043eb041de9d97783fcca14a0bdc1b1d560fc172fc21b648"},
|
||||
{file = "hf_transfer-0.1.6-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:7db60dd18eae4fa6ea157235fb82196cde5313995b396d1b591aad3b790a7f8f"},
|
||||
{file = "hf_transfer-0.1.6-cp310-cp310-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:30d31dbab9b5a558cce407b8728e39d87d7af1ef8745ddb90187e9ae0b9e1e90"},
|
||||
{file = "hf_transfer-0.1.6-cp310-cp310-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:f6b368bddd757efc7af3126ba81f9ac8f9435e2cc00902cb3d64f2be28d8f719"},
|
||||
{file = "hf_transfer-0.1.6-cp310-cp310-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:aa2086d8aefaaa3e144e167324574882004c0cec49bf2d0638ec4b74732d8da0"},
|
||||
{file = "hf_transfer-0.1.6-cp310-cp310-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:45d8985a0940bfe1535cb4ca781f5c11e47c83798ef3373ee1f5d57bbe527a9c"},
|
||||
{file = "hf_transfer-0.1.6-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:2f42b89735f1cde22f2a795d1f0915741023235666be7de45879e533c7d6010c"},
|
||||
{file = "hf_transfer-0.1.6-cp310-none-win32.whl", hash = "sha256:2d2c4c4613f3ad45b6ce6291e347b2d3ba1b86816635681436567e461cb3c961"},
|
||||
{file = "hf_transfer-0.1.6-cp310-none-win_amd64.whl", hash = "sha256:78b0eed8d8dce60168a46e584b9742b816af127d7e410a713e12c31249195342"},
|
||||
{file = "hf_transfer-0.1.6-cp311-cp311-macosx_10_12_x86_64.whl", hash = "sha256:f1d8c172153f9a6cdaecf137612c42796076f61f6bea1072c90ac2e17c1ab6fa"},
|
||||
{file = "hf_transfer-0.1.6-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:2c601996351f90c514a75a0eeb02bf700b1ad1db2d946cbfe4b60b79e29f0b2f"},
|
||||
{file = "hf_transfer-0.1.6-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:8e585c808405557d3f5488f385706abb696997bbae262ea04520757e30836d9d"},
|
||||
{file = "hf_transfer-0.1.6-cp311-cp311-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:ec51af1e8cf4268c268bd88932ade3d7ca895a3c661b42493503f02610ae906b"},
|
||||
{file = "hf_transfer-0.1.6-cp311-cp311-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:d106fdf996332f6df3ed3fab6d6332df82e8c1fb4b20fd81a491ca4d2ab5616a"},
|
||||
{file = "hf_transfer-0.1.6-cp311-cp311-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:e9c2ee9e9fde5a0319cc0e8ddfea10897482bc06d5709b10a238f1bc2ebcbc0b"},
|
||||
{file = "hf_transfer-0.1.6-cp311-cp311-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:f394ea32bc7802b061e549d3133efc523b4ae4fd19bf4b74b183ca6066eef94e"},
|
||||
{file = "hf_transfer-0.1.6-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:4282f09902114cd67fca98a1a1bad569a44521a8395fedf327e966714f68b977"},
|
||||
{file = "hf_transfer-0.1.6-cp311-none-win32.whl", hash = "sha256:276dbf307d5ab6f1bcbf57b5918bfcf9c59d6848ccb28242349e1bb5985f983b"},
|
||||
{file = "hf_transfer-0.1.6-cp311-none-win_amd64.whl", hash = "sha256:fa475175c51451186bea804471995fa8e7b2a48a61dcca55534911dc25955527"},
|
||||
{file = "hf_transfer-0.1.6-cp312-cp312-macosx_10_12_x86_64.whl", hash = "sha256:23d157a67acfa00007799323a1c441b2bbacc7dee625b016b7946fe0e25e6c89"},
|
||||
{file = "hf_transfer-0.1.6-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:6067342a2864b988f861cd2d31bd78eb1e84d153a3f6df38485b6696d9ad3013"},
|
||||
{file = "hf_transfer-0.1.6-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:91cfcb3070e205b58fa8dc8bcb6a62ccc40913fcdb9cd1ff7c364c8e3aa85345"},
|
||||
{file = "hf_transfer-0.1.6-cp312-cp312-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:eb76064ac5165d5eeaaf8d0903e8bf55477221ecc2a4a4d69f0baca065ab905b"},
|
||||
{file = "hf_transfer-0.1.6-cp312-cp312-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:9dabd3a177d83028f164984cf4dd859f77ec1e20c97a6f307ff8fcada0785ef1"},
|
||||
{file = "hf_transfer-0.1.6-cp312-cp312-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:d0bf4254e44f64a26e0a5b73b5d7e8d91bb36870718fb4f8e126ec943ff4c805"},
|
||||
{file = "hf_transfer-0.1.6-cp312-cp312-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:5d32c1b106f38f336ceb21531f4db9b57d777b9a33017dafdb6a5316388ebe50"},
|
||||
{file = "hf_transfer-0.1.6-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:ff05aba3c83921e5c7635ba9f07c693cc893350c447644824043aeac27b285f5"},
|
||||
{file = "hf_transfer-0.1.6-cp312-none-win32.whl", hash = "sha256:051ef0c55607652cb5974f59638da035773254b9a07d7ee5b574fe062de4c9d1"},
|
||||
{file = "hf_transfer-0.1.6-cp312-none-win_amd64.whl", hash = "sha256:716fb5c574fcbdd8092ce73f9b6c66f42e3544337490f77c60ec07df02bd081b"},
|
||||
{file = "hf_transfer-0.1.6-cp37-cp37m-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:6c0c981134a55965e279cb7be778c1ccaf93f902fc9ebe31da4f30caf824cc4d"},
|
||||
{file = "hf_transfer-0.1.6-cp37-cp37m-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:1ef1f145f04c5b573915bcb1eb5db4039c74f6b46fce73fc473c4287e613b623"},
|
||||
{file = "hf_transfer-0.1.6-cp37-cp37m-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:d0a7609b004db3347dbb7796df45403eceb171238210d054d93897d6d84c63a4"},
|
||||
{file = "hf_transfer-0.1.6-cp37-cp37m-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:60f0864bf5996773dbd5f8ae4d1649041f773fe9d5769f4c0eeb5553100acef3"},
|
||||
{file = "hf_transfer-0.1.6-cp37-cp37m-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:5d01e55d630ffe70a4f5d0ed576a04c6a48d7c65ca9a7d18f2fca385f20685a9"},
|
||||
{file = "hf_transfer-0.1.6-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:d855946c5062b665190de15b2bdbd4c8eddfee35350bfb7564592e23d36fbbd3"},
|
||||
{file = "hf_transfer-0.1.6-cp37-none-win32.whl", hash = "sha256:fd40b2409cfaf3e8aba20169ee09552f69140e029adeec261b988903ff0c8f6f"},
|
||||
{file = "hf_transfer-0.1.6-cp37-none-win_amd64.whl", hash = "sha256:0e0eba49d46d3b5481919aea0794aec625fbc6ecdf13fe7e0e9f3fc5d5ad5971"},
|
||||
{file = "hf_transfer-0.1.6-cp38-cp38-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:7e669fecb29fc454449739f9f53ed9253197e7c19e6a6eaa0f08334207af4287"},
|
||||
{file = "hf_transfer-0.1.6-cp38-cp38-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:89f701802892e5eb84f89f402686861f87dc227d6082b05f4e9d9b4e8015a3c3"},
|
||||
{file = "hf_transfer-0.1.6-cp38-cp38-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:b6f2b0c8b95b01409275d789a9b74d5f2e146346f985d384bf50ec727caf1ccc"},
|
||||
{file = "hf_transfer-0.1.6-cp38-cp38-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:aa855a2fa262792a230f9efcdb5da6d431b747d1861d2a69fe7834b19aea077e"},
|
||||
{file = "hf_transfer-0.1.6-cp38-cp38-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:4aa8ca349afb2f0713475426946261eb2035e4efb50ebd2c1d5ad04f395f4217"},
|
||||
{file = "hf_transfer-0.1.6-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:01255f043996bc7d1bae62d8afc5033a90c7e36ce308b988eeb84afe0a69562f"},
|
||||
{file = "hf_transfer-0.1.6-cp38-none-win32.whl", hash = "sha256:60b1db183e8a7540cd4f8b2160ff4de55f77cb0c3fc6a10be1e7c30eb1b2bdeb"},
|
||||
{file = "hf_transfer-0.1.6-cp38-none-win_amd64.whl", hash = "sha256:fb8be3cba6aaa50ab2e9dffbd25c8eb2046785eeff642cf0cdd0dd9ae6be3539"},
|
||||
{file = "hf_transfer-0.1.6-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:d09af35e3e3f09b664e6429e9a0dc200f29c5bdfd88bdd9666de51183b1fe202"},
|
||||
{file = "hf_transfer-0.1.6-cp39-cp39-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:a4505bd707cc14d85c800f961fad8ca76f804a8ad22fbb7b1a217d8d0c15e6a5"},
|
||||
{file = "hf_transfer-0.1.6-cp39-cp39-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:2c453fd8b0be9740faa23cecd1f28ee9ead7d900cefa64ff836960c503a744c9"},
|
||||
{file = "hf_transfer-0.1.6-cp39-cp39-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:13cb8884e718a78c3b81a8cdec9c7ac196dd42961fce55c3ccff3dd783e5ad7a"},
|
||||
{file = "hf_transfer-0.1.6-cp39-cp39-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:39cd39df171a2b5404de69c4e6cd14eee47f6fe91c1692f939bfb9e59a0110d8"},
|
||||
{file = "hf_transfer-0.1.6-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:8ff0629ee9f98df57a783599602eb498f9ec3619dc69348b12e4d9d754abf0e9"},
|
||||
{file = "hf_transfer-0.1.6-cp39-none-win32.whl", hash = "sha256:164a6ce445eb0cc7c645f5b6e1042c003d33292520c90052b6325f30c98e4c5f"},
|
||||
{file = "hf_transfer-0.1.6-cp39-none-win_amd64.whl", hash = "sha256:11b8b4b73bf455f13218c5f827698a30ae10998ca31b8264b51052868c7a9f11"},
|
||||
{file = "hf_transfer-0.1.6-pp310-pypy310_pp73-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:16957ba057376a99ea361074ce1094f61b58e769defa6be2422ae59c0b6a6530"},
|
||||
{file = "hf_transfer-0.1.6-pp310-pypy310_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:7db952112e3b8ee1a5cbf500d2443e9ce4fb893281c5310a3e31469898628005"},
|
||||
{file = "hf_transfer-0.1.6-pp37-pypy37_pp73-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:d39d826a7344f5e39f438d62632acd00467aa54a083b66496f61ef67a9885a56"},
|
||||
{file = "hf_transfer-0.1.6-pp37-pypy37_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:a4e2653fbfa92e7651db73d99b697c8684e7345c479bd6857da80bed6138abb2"},
|
||||
{file = "hf_transfer-0.1.6-pp38-pypy38_pp73-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:144277e6a86add10b90ec3b583253aec777130312256bfc8d5ade5377e253807"},
|
||||
{file = "hf_transfer-0.1.6-pp38-pypy38_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:3bb53bcd16365313b2aa0dbdc28206f577d70770f31249cdabc387ac5841edcc"},
|
||||
{file = "hf_transfer-0.1.6-pp39-pypy39_pp73-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:990d73a5a68d8261980f146c51f4c5f9995314011cb225222021ad7c39f3af2d"},
|
||||
{file = "hf_transfer-0.1.6-pp39-pypy39_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:652406037029ab9b4097b4c5f29321bad5f64c2b46fbff142509d918aec87c29"},
|
||||
{file = "hf_transfer-0.1.6.tar.gz", hash = "sha256:deb505a7d417d7055fd7b3549eadb91dfe782941261f3344025c486c16d1d2f9"},
|
||||
]
|
||||
|
||||
[[package]]
|
||||
name = "huggingface-hub"
|
||||
version = "0.23.0"
|
||||
|
|
@ -1334,6 +1406,7 @@ files = [
|
|||
[package.dependencies]
|
||||
filelock = "*"
|
||||
fsspec = ">=2023.5.0"
|
||||
hf-transfer = {version = ">=0.1.4", optional = true, markers = "extra == \"hf-transfer\""}
|
||||
packaging = ">=20.9"
|
||||
pyyaml = ">=5.1"
|
||||
requests = "*"
|
||||
|
|
@ -3295,7 +3368,6 @@ files = [
|
|||
{file = "PyYAML-6.0.1-cp311-cp311-win_amd64.whl", hash = "sha256:bf07ee2fef7014951eeb99f56f39c9bb4af143d8aa3c21b1677805985307da34"},
|
||||
{file = "PyYAML-6.0.1-cp312-cp312-macosx_10_9_x86_64.whl", hash = "sha256:855fb52b0dc35af121542a76b9a84f8d1cd886ea97c84703eaa6d88e37a2ad28"},
|
||||
{file = "PyYAML-6.0.1-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:40df9b996c2b73138957fe23a16a4f0ba614f4c0efce1e9406a184b6d07fa3a9"},
|
||||
{file = "PyYAML-6.0.1-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:a08c6f0fe150303c1c6b71ebcd7213c2858041a7e01975da3a99aed1e7a378ef"},
|
||||
{file = "PyYAML-6.0.1-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:6c22bec3fbe2524cde73d7ada88f6566758a8f7227bfbf93a408a9d86bcc12a0"},
|
||||
{file = "PyYAML-6.0.1-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:8d4e9c88387b0f5c7d5f281e55304de64cf7f9c0021a3525bd3b1c542da3b0e4"},
|
||||
{file = "PyYAML-6.0.1-cp312-cp312-win32.whl", hash = "sha256:d483d2cdf104e7c9fa60c544d92981f12ad66a457afae824d146093b8c294c54"},
|
||||
|
|
@ -4543,4 +4615,4 @@ xarm = ["gym-xarm"]
|
|||
[metadata]
|
||||
lock-version = "2.0"
|
||||
python-versions = ">=3.10,<3.13"
|
||||
content-hash = "32e3ed24bc4cdc6fc2880a62de2bbc8d334912cb86200ba32a8ed2cec31d2feb"
|
||||
content-hash = "de2dbc4ee8683eeb34f4dfff0c18270e594d4b6698af2b2eba97bd4cede74be1"
|
||||
|
|
|
|||
|
|
@ -43,7 +43,7 @@ opencv-python = ">=4.9.0"
|
|||
diffusers = "^0.27.2"
|
||||
torchvision = ">=0.18.0"
|
||||
h5py = ">=3.10.0"
|
||||
huggingface-hub = ">=0.21.4"
|
||||
huggingface-hub = {extras = ["hf-transfer"], version = "^0.23.0"}
|
||||
gymnasium = ">=0.29.1"
|
||||
cmake = ">=3.29.0.1"
|
||||
gym-pusht = { version = ">=0.1.3", optional = true}
|
||||
|
|
|
|||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:9f9347c8d9ac90ee44e6dd86f65043438168df6bbe4bab2d2b875e55ef7376ef
|
||||
size 1488
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:cf148247bf191c7f7e8af738a7b9e147f9ffffeec0e4b9d1c4783c4e384da7eb
|
||||
size 33
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:02fc4ea25766269f65752a60b0594c43d799b0ae528cd773bf024b064b5aa329
|
||||
size 4344
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:55d7b1a06fe3e3051482752740074348bdb5fc98fb2e305b06d6203994117b27
|
||||
size 592448
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:8b7fbedfdb3d536847bc6fadf2cbabb9f2b5492edf3e2c274a3e8ffb447105e8
|
||||
size 1166
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:98329e4b40e9be0d63f7d36da9d86c44bbe7eeeb1b10d3ba973c923f3be70867
|
||||
size 247
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:54e42cdfd016a0ced2ab1fe2966a8c15a2384e0dbe1a2fe87433a2d1b8209ac0
|
||||
size 5220057
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:af1ded2a244cb47a96255b75f584a643edf6967e13bb5464b330ffdd9d7ad859
|
||||
size 5284692
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:13d1bebabd79984fd6715971be758ef9a354495adea5e8d33f4e7904365e112b
|
||||
size 5258380
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:f33bc6810f0b91817a42610364cb49ed1b99660f058f0f9407e6f5920d0aee02
|
||||
size 1008
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:cf148247bf191c7f7e8af738a7b9e147f9ffffeec0e4b9d1c4783c4e384da7eb
|
||||
size 33
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:7b58d6c89e936a781a307805ebecf0dd473fbc02d52a7094da62e54bffb9454a
|
||||
size 4344
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:a08be578285cbe2d35b78f150d464ff3e10604a9865398c976983e0d711774f9
|
||||
size 788528
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:8b7fbedfdb3d536847bc6fadf2cbabb9f2b5492edf3e2c274a3e8ffb447105e8
|
||||
size 1166
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:34e36233477c8aa0b0840314ddace072062d4f486d06546bbd6550832c370065
|
||||
size 247
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:66e7349a4a82ca6042a7189608d01eb1cfa38d100d039b5445ae1a9e65d824ab
|
||||
size 14470946
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:a2146f0c10c9f2611e57e617983aa4f91ad681b4fc50d91b992b97abd684f926
|
||||
size 11662185
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:5affbaf1c48895ba3c626e0d8cf1309e5f4ec6bbaa135313096f52a22de66c05
|
||||
size 11410342
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:6c2b195ca91b88fd16422128d386d2cabd808a1862c6d127e6bf2e83e1fe819a
|
||||
size 448
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:cf148247bf191c7f7e8af738a7b9e147f9ffffeec0e4b9d1c4783c4e384da7eb
|
||||
size 33
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:b360b6b956d2adcb20589947c553348ef1eb6b70743c989dcbe95243d8592ce5
|
||||
size 4344
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:3f5c3926b4d4da9271abefcdf6a8952bb1f13258a9c39fe0fd223f548dc89dcb
|
||||
size 887728
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:8b7fbedfdb3d536847bc6fadf2cbabb9f2b5492edf3e2c274a3e8ffb447105e8
|
||||
size 1166
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:4993b05fb026619eec5eb70db8cadaa041ba4ab92d38b4a387167ace03b1018b
|
||||
size 247
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:bd25d17ef5b7500386761b5e32920879bbdcafe0e17a8a8845628525d861e644
|
||||
size 10231081
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:5b557acbfeb0681c0a38e47263d945f6cd3a03461298d8b17209c81e3fd0aae8
|
||||
size 9701371
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:da8f3b4f9f965da63819652b2c042d4cf7e07d14631113ea072087d56370310e
|
||||
size 10473741
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:a053506017d8a78cfd307b2912eeafa1ac1485a280cf90913985fcc40120b5ec
|
||||
size 416
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:cf148247bf191c7f7e8af738a7b9e147f9ffffeec0e4b9d1c4783c4e384da7eb
|
||||
size 33
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:d6d172d1bca02face22ceb4c21ea2b054cf3463025485dce64711b6f36b31f8a
|
||||
size 4344
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:7e5ce817a2c188041f57f8d4c465dab3b9c3e4e1aeb7a9fb270230d1b36df530
|
||||
size 1477064
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:8b7fbedfdb3d536847bc6fadf2cbabb9f2b5492edf3e2c274a3e8ffb447105e8
|
||||
size 1166
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:4eb2dc373e4ea7d474742590f9073d66a773f6ab94b9e73a8673df19f93fae6d
|
||||
size 247
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:d2c55b146fabe78b18c8a28a7746ab56e1ee7a6918e9e3dad9bd196f97975895
|
||||
size 26158915
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:71e1958d77f56843acf1ec48da4f04311a5836c87a0e77dbe26aa47c27c6347e
|
||||
size 18786848
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:20780718399b5759ff9a3a79824986310524793066198e3b9a307222f11a93df
|
||||
size 17769988
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:279916f7689ae46af90e92a46eba9486a71fc762e3e2679ab5441eb37126827b
|
||||
size 928
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:cf148247bf191c7f7e8af738a7b9e147f9ffffeec0e4b9d1c4783c4e384da7eb
|
||||
size 33
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:7a7731051b521694b52b5631470720a7f05331915f4ac4e7f8cd83f9ff459bce
|
||||
size 4344
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:99608258e8c9fe5191f1a12edc29b47d307790104149dffb6d3046ddad6aeb1b
|
||||
size 435600
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:8b7fbedfdb3d536847bc6fadf2cbabb9f2b5492edf3e2c274a3e8ffb447105e8
|
||||
size 1166
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:ae6735b7b394914824e974a7461019373a10f9e2d84ddf834bec8ea268d9ec1e
|
||||
size 247
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:833e288c5fdacbbe10a5d048cb6f49fe1a396d91b2117b827e130ec11069256a
|
||||
size 8397615
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:2cb870acb4855fef70f19c5f632d94e4c25eef59eeea92f4b1167a44b1b36b33
|
||||
size 5912007
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:8be36298141b455ea51d17a78e4bbc6619639302139fe2db605bdfa3ff5e91bd
|
||||
size 4794018
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:348d0ee38a71929b2017d540de870b9dff6d79efdd0cbc5352fa9697e350134a
|
||||
size 928
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:cf148247bf191c7f7e8af738a7b9e147f9ffffeec0e4b9d1c4783c4e384da7eb
|
||||
size 33
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:c5c2996f58d5277fa19cf56ec143334fbee940d1de37530452496a6f0aa11f88
|
||||
size 4344
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:da3a8efea9ba60d1fdd209d45a3387df22a09f7c156904ecb03f10456736fb74
|
||||
size 514056
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:8b7fbedfdb3d536847bc6fadf2cbabb9f2b5492edf3e2c274a3e8ffb447105e8
|
||||
size 1166
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:6b7111ff1ef5c4d6a2990f5f39f42398f061da8c4e81adf46b9d9150ec2feeaf
|
||||
size 247
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:2ac8c2755d940534042595ecad33ebea358974ec67bc041c8675e53b7d2272ff
|
||||
size 9182551
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:b35aaa37e66dd5563d93e6059d5b645e112e020e03bd398f7098a5289970953a
|
||||
size 6378566
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:6138247ba7160a3de6c50111e6fcc5ae075044086d8527ae5d435b1f8a7c7a93
|
||||
size 6439183
|
||||
Binary file not shown.
|
|
@ -1,4 +1,3 @@
|
|||
{
|
||||
"fps": 50,
|
||||
"video": 1
|
||||
}
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:cf148247bf191c7f7e8af738a7b9e147f9ffffeec0e4b9d1c4783c4e384da7eb
|
||||
size 33
|
||||
|
|
|
|||
Binary file not shown.
Binary file not shown.
|
|
@ -1,47 +1,3 @@
|
|||
{
|
||||
"citation": "",
|
||||
"description": "",
|
||||
"features": {
|
||||
"observation.images.top": {
|
||||
"_type": "VideoFrame"
|
||||
},
|
||||
"observation.state": {
|
||||
"feature": {
|
||||
"dtype": "float32",
|
||||
"_type": "Value"
|
||||
},
|
||||
"length": 14,
|
||||
"_type": "Sequence"
|
||||
},
|
||||
"action": {
|
||||
"feature": {
|
||||
"dtype": "float32",
|
||||
"_type": "Value"
|
||||
},
|
||||
"length": 14,
|
||||
"_type": "Sequence"
|
||||
},
|
||||
"episode_index": {
|
||||
"dtype": "int64",
|
||||
"_type": "Value"
|
||||
},
|
||||
"frame_index": {
|
||||
"dtype": "int64",
|
||||
"_type": "Value"
|
||||
},
|
||||
"timestamp": {
|
||||
"dtype": "float32",
|
||||
"_type": "Value"
|
||||
},
|
||||
"next.done": {
|
||||
"dtype": "bool",
|
||||
"_type": "Value"
|
||||
},
|
||||
"index": {
|
||||
"dtype": "int64",
|
||||
"_type": "Value"
|
||||
}
|
||||
},
|
||||
"homepage": "",
|
||||
"license": ""
|
||||
}
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:3f44d13de5d5a417263bbd4984942ed42ed3fa0633405aa14d9a969a45274944
|
||||
size 842
|
||||
|
|
|
|||
|
|
@ -1,13 +1,3 @@
|
|||
{
|
||||
"_data_files": [
|
||||
{
|
||||
"filename": "data-00000-of-00001.arrow"
|
||||
}
|
||||
],
|
||||
"_fingerprint": "eb913a2b1a68aa74",
|
||||
"_format_columns": null,
|
||||
"_format_kwargs": {},
|
||||
"_format_type": null,
|
||||
"_output_all_columns": false,
|
||||
"_split": null
|
||||
}
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:71d6ba89bee5a4ee2761220452999e415bc838a44bebf1b5a2e4ba8622369798
|
||||
size 247
|
||||
|
|
|
|||
Binary file not shown.
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:e7dbc214a415689ca7fb83b6f8e12ec7824dfe34a66024b0b24bfeb3aeefd0e4
|
||||
size 928
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:083db9efc5c9e3396c5e1159d020c2a3786f1f1a4b069719d327ed7fbc65c34d
|
||||
size 33
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:f98bd8f6347590aecdddaceed95d921f2d9f7bf35fbe742c37bdf12cba11dca6
|
||||
size 2904
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:c0013aea549ec290af94bddde1b559fb8d0967d4c43ef14319177c4e62ed1e91
|
||||
size 14545712
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:4c9545525dc1f4d550591bd5efb63b55c15b983ae0510fefda5a16d77c78b6ef
|
||||
size 837
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:d7aa033603dc90582516dbcdf3e71e4d3113b70ad49098535def0b282135b5f3
|
||||
size 247
|
||||
Binary file not shown.
|
|
@ -1,4 +1,3 @@
|
|||
{
|
||||
"fps": 50,
|
||||
"video": 1
|
||||
}
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:cf148247bf191c7f7e8af738a7b9e147f9ffffeec0e4b9d1c4783c4e384da7eb
|
||||
size 33
|
||||
|
|
|
|||
Binary file not shown.
Binary file not shown.
|
|
@ -1,47 +1,3 @@
|
|||
{
|
||||
"citation": "",
|
||||
"description": "",
|
||||
"features": {
|
||||
"observation.images.top": {
|
||||
"_type": "VideoFrame"
|
||||
},
|
||||
"observation.state": {
|
||||
"feature": {
|
||||
"dtype": "float32",
|
||||
"_type": "Value"
|
||||
},
|
||||
"length": 14,
|
||||
"_type": "Sequence"
|
||||
},
|
||||
"action": {
|
||||
"feature": {
|
||||
"dtype": "float32",
|
||||
"_type": "Value"
|
||||
},
|
||||
"length": 14,
|
||||
"_type": "Sequence"
|
||||
},
|
||||
"episode_index": {
|
||||
"dtype": "int64",
|
||||
"_type": "Value"
|
||||
},
|
||||
"frame_index": {
|
||||
"dtype": "int64",
|
||||
"_type": "Value"
|
||||
},
|
||||
"timestamp": {
|
||||
"dtype": "float32",
|
||||
"_type": "Value"
|
||||
},
|
||||
"next.done": {
|
||||
"dtype": "bool",
|
||||
"_type": "Value"
|
||||
},
|
||||
"index": {
|
||||
"dtype": "int64",
|
||||
"_type": "Value"
|
||||
}
|
||||
},
|
||||
"homepage": "",
|
||||
"license": ""
|
||||
}
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:3f44d13de5d5a417263bbd4984942ed42ed3fa0633405aa14d9a969a45274944
|
||||
size 842
|
||||
|
|
|
|||
|
|
@ -1,13 +1,3 @@
|
|||
{
|
||||
"_data_files": [
|
||||
{
|
||||
"filename": "data-00000-of-00001.arrow"
|
||||
}
|
||||
],
|
||||
"_fingerprint": "d20c2acf1e107266",
|
||||
"_format_columns": null,
|
||||
"_format_kwargs": {},
|
||||
"_format_type": null,
|
||||
"_output_all_columns": false,
|
||||
"_split": null
|
||||
}
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:e48156ce4f71ac15d78732312fbc7e199f0ecdaac3604231e6be2e3e5b31a0ad
|
||||
size 247
|
||||
|
|
|
|||
Binary file not shown.
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:4500f31e62f0928a837fa71783acacda0db516c7b00d0586a41ea5fd8fc5e772
|
||||
size 928
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:083db9efc5c9e3396c5e1159d020c2a3786f1f1a4b069719d327ed7fbc65c34d
|
||||
size 33
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:0227d4e9e3b43a86bf33fbd68683ede537fdeab1b53f2ebf155620e10054352f
|
||||
size 2904
|
||||
|
|
@ -0,0 +1,3 @@
|
|||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:1bdf02ecf7d5fc502f6dd9f520c636828a5988ad16a69a137780a824f94f8112
|
||||
size 10782640
|
||||
Some files were not shown because too many files have changed in this diff Show More
Loading…
Reference in New Issue