Additional code for Stable-baselines3 to load and upload models from the Hub.

Overview

Hugging Face x Stable-baselines3

A library to load and upload Stable-baselines3 models from the Hub.

Installation

With pip

Examples

[Todo: add colab tutorial]

Case 1: I want to download a model from the Hub

import gym

from huggingface_sb3 import load_from_hub
from stable_baselines3 import PPO

env = gym.make("CartPole-v1")

model = PPO("MlpPolicy", env, verbose=1)

# Retrieve the model from the hub
## repo_id =  id of the model repository from the Hugging Face Hub (repo_id = {organization}/{repo_name})
## filename = name of the model zip file from the repository
checkpoint = load_from_hub(repo_id="ThomasSimonini/ppo-CartPole-v1", filename="CartPole-v1")
PPO.load(checkpoint)

obs = env.reset()
for i in range(1000):
    action, _states = model.predict(obs, deterministic=True)
    obs, reward, done, info = env.step(action)
    env.render()
    if done:
      obs = env.reset()

env.close()

Case 2: I trained an agent and want to upload it to the Hub

First you need to be logged in to Hugging Face:

  • If you're using Colab/Jupyter Notebooks:
from huggingface_hub import notebook_login
notebook_login()
  • Else:
huggingface-cli login

Then:

import gym
from huggingface_sb3 import push_to_hub
from stable_baselines3 import PPO

# Create the environment
env = gym.make('CartPole-v1')

# Define a PPO MLpPolicy architecture
model = PPO('MlpPolicy', env, verbose=1)

# Train it for 10000 timesteps
model.learn(total_timesteps=10000)

# Save the model 
model.save("CartPole-v1")

# Push this saved model to the hf repo
# If this repo does not exists it will be created
## filename: the name of the file == "name" inside model.save("CartPole-v1")
push_to_hub(repo_name = "CartPole-v1",
           organization = "ThomasSimonini",  
           filename = "CartPole-v1", 
           commit_message = "Added Cartpole-v1 trained model")
Comments
  • Environment name normalization and explicit naming schemes

    Environment name normalization and explicit naming schemes

    There was an issue with environment names, that have a slash in their name (see https://github.com/DLR-RM/rl-baselines3-zoo/pull/257). Also the naming scheme for models and repository IDs is just based on convention.

    This PR implements normalization for environment names (replacing slashes with dashes) and encodes the naming scheme for models and repository IDs in little helper classes. The idea is, that those helper classes can be used by downstream libraries to comply with the naming scheme (such as the rl baselines zoo). If we ever need to change the naming scheme or other cases in which the environment name needs to be normalized come up, then we can implement them here and the downstream libraries immediately profit from that.

    I also added a simple smoke test for pulling a model from the hub.

    opened by ernestum 8
  • 400 Client Error for `package_to_hub` function

    400 Client Error for `package_to_hub` function

    I am going through the notebook of Unit 1 of the deep RL course. However, I cannot run the package_to_hub function, which gives the following error:

    HTTPError                                 Traceback (most recent call last)
    
    [<ipython-input-26-97f48e41190b>](https://localhost:8080/#) in <module>
         25                eval_env=eval_env,
         26                repo_id="LorenzoPacchiardi/ppo-LunarLander-v2",
    ---> 27                commit_message="Upload PPO LunarLander-v2 trained agent (50 steps)")
    
    6 frames
    
    [/usr/local/lib/python3.7/dist-packages/requests/models.py](https://localhost:8080/#) in raise_for_status(self)
        939 
        940         if http_error_msg:
    --> 941             raise HTTPError(http_error_msg, response=self)
        942 
        943     def close(self):
    
    HTTPError: 400 Client Error: Bad Request for url: https://huggingface.co/api/models/LorenzoPacchiardi/ppo-LunarLander-v2/commit/main (Request ID: fhQtAuS_qa8bj_c6AI0v5)
    

    I get a similar error with push_to_hub

    I logged in to huggingface correctly with the token, and the load_from_hub function works fine.

    opened by LoryPack 5
  • package_to_hub requires OpenGL and xvfb which are not present on newer Mac OS systems

    package_to_hub requires OpenGL and xvfb which are not present on newer Mac OS systems

    Currently package_to_hub works only for OpenGL capable computers. It doesn't support any other option for generating video and it doesn't allow to upload model without a video. All new Mac OSes don't have OpenGL support any more.

    opened by marcin-sobocinski 3
  • Error installing huggingface_sb3

    Error installing huggingface_sb3

    Hi! I'm running this notebook https://github.com/huggingface/deep-rl-class/blob/main/unit1/unit1.ipynb from your DRL series. Installation of some libraries is causing some issues. For huggingface_sb3, it is:

    Collecting huggingface_sb3
      Using cached huggingface_sb3-2.0.0-py3-none-any.whl (7.4 kB)
    Requirement already satisfied: wasabi in ./rl/lib/python3.8/site-packages (from huggingface_sb3) (0.9.1)
    Collecting cloudpickle==1.6
      Using cached cloudpickle-1.6.0-py3-none-any.whl (23 kB)
    Collecting pyyaml==6.0
      Using cached PyYAML-6.0-cp38-cp38-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_12_x86_64.manylinux2010_x86_64.whl (701 kB)
    Collecting huggingface-hub
      Using cached huggingface_hub-0.6.0-py3-none-any.whl (84 kB)
    Collecting pickle5
      Using cached pickle5-0.0.11.tar.gz (132 kB)
      Preparing metadata (setup.py) ... done
    Collecting typing-extensions>=3.7.4.3
      Using cached typing_extensions-4.2.0-py3-none-any.whl (24 kB)
    Requirement already satisfied: packaging>=20.9 in ./rl/lib/python3.8/site-packages (from huggingface-hub->huggingface_sb3) (21.3)
    Collecting filelock
      Using cached filelock-3.7.0-py3-none-any.whl (10 kB)
    Collecting tqdm
      Using cached tqdm-4.64.0-py2.py3-none-any.whl (78 kB)
    Collecting requests
      Using cached requests-2.27.1-py2.py3-none-any.whl (63 kB)
    Requirement already satisfied: pyparsing!=3.0.5,>=2.0.2 in ./rl/lib/python3.8/site-packages (from packaging>=20.9->huggingface-hub->huggingface_sb3) (3.0.9)
    Collecting urllib3<1.27,>=1.21.1
      Using cached urllib3-1.26.9-py2.py3-none-any.whl (138 kB)
    Collecting certifi>=2017.4.17
      Using cached certifi-2021.10.8-py2.py3-none-any.whl (149 kB)
    Collecting charset-normalizer~=2.0.0
      Using cached charset_normalizer-2.0.12-py3-none-any.whl (39 kB)
    Collecting idna<4,>=2.5
      Using cached idna-3.3-py3-none-any.whl (61 kB)
    Using legacy 'setup.py install' for pickle5, since package 'wheel' is not installed.
    Installing collected packages: pickle5, certifi, urllib3, typing-extensions, tqdm, pyyaml, idna, filelock, cloudpickle, charset-normalizer, requests, huggingface-hub, huggingface_sb3
      Running setup.py install for pickle5 ... error
      error: subprocess-exited-with-error
      
      × Running setup.py install for pickle5 did not run successfully.
      │ exit code: 1
      ╰─> [27 lines of output]
          running install
          /media/master/support/pip_envs/rl/lib/python3.8/site-packages/setuptools/command/install.py:34: SetuptoolsDeprecationWarning: setup.py install is deprecated. Use build and pip and other standards-based tools.
            warnings.warn(
          running build
          running build_py
          creating build
          creating build/lib.linux-x86_64-cpython-38
          creating build/lib.linux-x86_64-cpython-38/pickle5
          copying pickle5/__init__.py -> build/lib.linux-x86_64-cpython-38/pickle5
          copying pickle5/pickle.py -> build/lib.linux-x86_64-cpython-38/pickle5
          copying pickle5/pickletools.py -> build/lib.linux-x86_64-cpython-38/pickle5
          creating build/lib.linux-x86_64-cpython-38/pickle5/test
          copying pickle5/test/pickletester.py -> build/lib.linux-x86_64-cpython-38/pickle5/test
          copying pickle5/test/test_picklebuffer.py -> build/lib.linux-x86_64-cpython-38/pickle5/test
          copying pickle5/test/__init__.py -> build/lib.linux-x86_64-cpython-38/pickle5/test
          copying pickle5/test/test_pickle.py -> build/lib.linux-x86_64-cpython-38/pickle5/test
          running build_ext
          building 'pickle5._pickle' extension
          creating build/temp.linux-x86_64-cpython-38
          creating build/temp.linux-x86_64-cpython-38/pickle5
          x86_64-linux-gnu-gcc -Wno-unused-result -Wsign-compare -DNDEBUG -g -fwrapv -O2 -Wall -g -fstack-protector-strong -Wformat -Werror=format-security -g -fwrapv -O2 -fPIC -I/media/master/support/pip_envs/rl/include -I/usr/include/python3.8 -c pickle5/_pickle.c -o build/temp.linux-x86_64-cpython-38/pickle5/_pickle.o -std=c99
          In file included from pickle5/_pickle.c:2:
          pickle5/compat.h:1:10: fatal error: Python.h: No such file or directory
              1 | #include "Python.h"
                |          ^~~~~~~~~~
          compilation terminated.
          error: command '/usr/bin/x86_64-linux-gnu-gcc' failed with exit code 1
          [end of output]
      
      note: This error originates from a subprocess, and is likely not a problem with pip.
    error: legacy-install-failure
    
    × Encountered error while trying to install package.
    ╰─> pickle5
    
    note: This is an issue with the package mentioned above, not pip.
    

    I tried installing it with python 3.9 and 3.8 on Ubuntu 22.04 OS.

    Are there any additional requirements to use your library?

    opened by kirilllzaitsev 3
  • generate_replay created a video.mp4 file locally

    generate_replay created a video.mp4 file locally

    This code snippet

    env = VecVideoRecorder(
            eval_env,
            "./",  # Temporary video folder
            record_video_trigger=lambda x: x == 0,
            video_length=video_length,
            name_prefix="",
        )
    

    generated a video file wherever the user is. In the temporary video folder, you can use tempfile.TemporaryDirectory() to automatically create a directory that will be deleted afterwards

    opened by osanseviero 2
  • Rebase repo when pulling

    Rebase repo when pulling

    I think you want to rebase, as we tend to do this in all the mixins in huggingface_hub. I think not having this is what caused this issue: https://github.com/huggingface/deep-rl-class/issues/20.

    opened by nateraw 2
  • Add `VecNormalize` support

    Add `VecNormalize` support

    • add missing type hints
    • fix push_to_hub (bug detected by pytype checker)
    • cleanup
    • add support for VecNormalize

    closes #6

    Demo (and training/loading code): https://huggingface.co/araffin/a2c-Pendulum-v1

    opened by araffin 1
  • Add auto release

    Add auto release

    The behavior with this PR is that once you push a Git tag with v* (usually v1.0.8 for example), which should ideally point to the commit that updates this line https://github.com/huggingface/huggingface_sb3/blob/main/setup.py#L10 (you can push the tag after the commit), it will automatically make a pypi release.

    The only requirement is adding your secret (PYPI_TOKEN_DIST) to the repo settings

    opened by osanseviero 1
  • Allow to pass TensorBoard logs files to package_to_hub

    Allow to pass TensorBoard logs files to package_to_hub

    It's very easy to add TensorBoard logging with SB3, but pushing the files right now needs to be done manually. As an alternative, we could add a param to package_to_hub to pass the logs.

    Related: https://github.com/huggingface/deep-rl-class/pull/19

    opened by osanseviero 0
  • Don't crash when making videos causes problems

    Don't crash when making videos causes problems

    At the moment, if generate_replay fails, the whole package_to_hub method fails. Ideally it would still push the metrics and other related information even if no video is generated

    opened by osanseviero 0
  • Huggingface_SB3 v2.0

    Huggingface_SB3 v2.0

    👋 so here's the SB3 v2.0:

    With our new version we can use package_to_hub method that:

    1. Save the model
    2. Evaluate the model and generate a results.json
    3. Generate a model card
    4. Record a replay video of the agent
    5. Push everything to the hub Here's an example : https://huggingface.co/ThomasSimonini/TEST2-Colab-ppo-LunarLander-v2 (very small training so the agent is bad)

    If you want to try some examples directly on colab I've made a small test colab: https://colab.research.google.com/drive/1FhZ1w7smqPo8GQcW5qb2HmkggZVuok57?usp=sharing

    The PyPi update is also automated thanks to @osanseviero

    A big thanks to Omar who made a lot of tests with the library

    I need a little bit of feedback for the documentation, I think it's not very clear.

    • I explain the 2 cases: downstream and upstream
    • In case 3-4 I explain how to use xvfb if you use colab or vm (because you don't have a screen to render so you can't generate a video without xvfb).

    WDYT? Thanks

    opened by simoninithomas 0
Releases(v2.2.4)
  • v2.2.4(Oct 13, 2022)

    What's Changed

    • Loosen the requirements by @araffin in https://github.com/huggingface/huggingface_sb3/pull/19

    Full Changelog: https://github.com/huggingface/huggingface_sb3/compare/v2.2.3...v2.2.4

    Source code(tar.gz)
    Source code(zip)
  • v2.2.3(Aug 5, 2022)

    Cloudpickle is 1.3 by default on Colab. We need at minimum 1.6 for package_to_hub and load_from_hub to work correctly.

    Full Changelog: https://github.com/huggingface/huggingface_sb3/compare/v2.2.2...v2.2.3

    Source code(tar.gz)
    Source code(zip)
  • v2.2.2(Aug 1, 2022)

    What's Changed

    • V2.2.2 by @simoninithomas in https://github.com/huggingface/huggingface_sb3/pull/17
    • Pinning this dependency leads to some uploading problems. We removed it

    Full Changelog: https://github.com/huggingface/huggingface_sb3/compare/v2.2.1...v2.2.2

    Source code(tar.gz)
    Source code(zip)
  • v2.2.1(Jul 8, 2022)

    What's Changed

    • Notebook fixes by @ernestum in https://github.com/huggingface/huggingface_sb3/pull/12
    • Environment name normalization and explicit naming schemes by @ernestum in https://github.com/huggingface/huggingface_sb3/pull/13
    • Use new upload_folder API by @osanseviero in https://github.com/huggingface/huggingface_sb3/pull/15

    New Contributors

    • @ernestum made their first contribution in https://github.com/huggingface/huggingface_sb3/pull/12

    Full Changelog: https://github.com/huggingface/huggingface_sb3/compare/v2.1.1...v2.2

    Source code(tar.gz)
    Source code(zip)
  • v2.1.0(May 20, 2022)

    What's Changed

    • Use make_vec_env to create envs by @araffin in https://github.com/huggingface/huggingface_sb3/pull/3
    • Rebase repo when pulling by @nateraw in https://github.com/huggingface/huggingface_sb3/pull/7
    • Fix record video for RecurrentPPO by @araffin in https://github.com/huggingface/huggingface_sb3/pull/8
    • Add VecNormalize support by @araffin in https://github.com/huggingface/huggingface_sb3/pull/10

    New Contributors

    • @araffin made their first contribution in https://github.com/huggingface/huggingface_sb3/pull/3
    • @nateraw made their first contribution in https://github.com/huggingface/huggingface_sb3/pull/7

    Full Changelog: https://github.com/huggingface/huggingface_sb3/compare/v2.0.0...v2.1.0

    Source code(tar.gz)
    Source code(zip)
Owner
Hugging Face
The AI community building the future.
Hugging Face
The code for our NeurIPS 2021 paper "Kernelized Heterogeneous Risk Minimization".

Kernelized-HRM Jiashuo Liu, Zheyuan Hu The code for our NeurIPS 2021 paper "Kernelized Heterogeneous Risk Minimization"[1]. This repo contains the cod

Liu Jiashuo 8 Nov 20, 2022
This is an open solution to the Home Credit Default Risk challenge 🏡

Home Credit Default Risk: Open Solution This is an open solution to the Home Credit Default Risk challenge 🏡 . More competitions 🎇 Check collection

minerva.ml 427 Dec 27, 2022
Weakly Supervised Learning of Instance Segmentation with Inter-pixel Relations, CVPR 2019 (Oral)

Weakly Supervised Learning of Instance Segmentation with Inter-pixel Relations The code of: Weakly Supervised Learning of Instance Segmentation with I

Jiwoon Ahn 472 Dec 29, 2022
Explainability for Vision Transformers (in PyTorch)

Explainability for Vision Transformers (in PyTorch) This repository implements methods for explainability in Vision Transformers

Jacob Gildenblat 442 Jan 04, 2023
Reinforcement learning for self-driving in a 3D simulation

SelfDrive_AI Reinforcement learning for self-driving in a 3D simulation (Created using UNITY-3D) 1. Requirements for the SelfDrive_AI Gym You need Pyt

Surajit Saikia 17 Dec 14, 2021
Meta-Learning Sparse Implicit Neural Representations (NeurIPS 2021)

Meta-SparseINR Official PyTorch implementation of "Meta-learning Sparse Implicit Neural Representations" (NeurIPS 2021) by Jaeho Lee*, Jihoon Tack*, N

Jaeho Lee 41 Nov 10, 2022
Latte: Cross-framework Python Package for Evaluation of Latent-based Generative Models

Cross-framework Python Package for Evaluation of Latent-based Generative Models Latte Latte (for LATent Tensor Evaluation) is a cross-framework Python

Karn Watcharasupat 30 Sep 08, 2022
The repository for our EMNLP 2021 paper "Finnish Dialect Identification: The Effect of Audio and Text"

Finnish Dialect Identification The repository for our EMNLP 2021 paper "Finnish Dialect Identification: The Effect of Audio and Text". We present a te

Rootroo Ltd 2 Dec 25, 2021
Alignment Attention Fusion framework for Few-Shot Object Detection

AAF framework Framework generalities This repository contains the code of the AAF framework proposed in this paper. The main idea behind this work is

Pierre Le Jeune 20 Dec 16, 2022
The repository contain code for building compiler using puthon.

Building Compiler This is a python implementation of JamieBuild's "Super Tiny Compiler" Overview JamieBuilds developed a wonderfully educative compile

Shyam Das Shrestha 1 Nov 21, 2021
CoReD: Generalizing Fake Media Detection with Continual Representation using Distillation (ACMMM'21 Oral Paper)

CoReD: Generalizing Fake Media Detection with Continual Representation using Distillation (ACMMM'21 Oral Paper) (Accepted for oral presentation at ACM

Minha Kim 1 Nov 12, 2021
PerfFuzz: Automatically Generate Pathological Inputs for C/C++ programs

PerfFuzz Performance problems in software can arise unexpectedly when programs are provided with inputs that exhibit pathological behavior. But how ca

Caroline Lemieux 125 Nov 18, 2022
Exporter for Storage Area Network (SAN)

SAN Exporter Prometheus exporter for Storage Area Network (SAN). We all know that each SAN Storage vendor has their own glossary of terms, health/perf

vCloud 32 Dec 16, 2022
SLIDE : In Defense of Smart Algorithms over Hardware Acceleration for Large-Scale Deep Learning Systems

The SLIDE package contains the source code for reproducing the main experiments in this paper. Dataset The Datasets can be downloaded in Amazon-

Intel Labs 72 Dec 16, 2022
This is the winning solution of the Endocv-2021 grand challange.

Endocv2021-winner [Paper] This is the winning solution of the Endocv-2021 grand challange. Dependencies pytorch # tested with 1.7 and 1.8 torchvision

Vajira Thambawita 14 Dec 03, 2022
Multi-Anchor Active Domain Adaptation for Semantic Segmentation (ICCV 2021 Oral)

Multi-Anchor Active Domain Adaptation for Semantic Segmentation Munan Ning*, Donghuan Lu*, Dong Wei†, Cheng Bian, Chenglang Yuan, Shuang Yu, Kai Ma, Y

Munan Ning 36 Dec 07, 2022
It's like Shape Editor in Maya but works with skeletons (transforms).

Skeleposer What is Skeleposer? Briefly, it's like Shape Editor in Maya, but works with transforms and joints. It can be used to make complex facial ri

Alexander Zagoruyko 1 Nov 11, 2022
根据midi文件演奏“风物之诗琴”的脚本 "Windsong Lyre" auto play

Genshin-lyre-auto-play 简体中文 | English 简介 根据midi文件演奏“风物之诗琴”的脚本。由Python驱动,在此承诺, ⚠️ 项目内绝不含任何能够引起安全问题的代码。 前排提示:所有键盘在动但是原神没反应的都是因为没有管理员权限,双击run.bat或者以管理员模式

御坂17032号 386 Jan 01, 2023
mlpack: a scalable C++ machine learning library --

a fast, flexible machine learning library Home | Documentation | Doxygen | Community | Help | IRC Chat Download: current stable version (3.4.2) mlpack

mlpack 4.2k Jan 09, 2023
Official implementation of "DSP: Dual Soft-Paste for Unsupervised Domain Adaptive Semantic Segmentation"

DSP Official implementation of "DSP: Dual Soft-Paste for Unsupervised Domain Adaptive Semantic Segmentation". Accepted by ACM Multimedia 2021. Authors

20 Oct 24, 2022