Skip to content

[feature] Add experimental PyTorch support #4335

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 132 commits into from
Aug 20, 2020
Merged

Conversation

ervteng
Copy link
Contributor

@ervteng ervteng commented Aug 11, 2020

Proposed change(s)

Experimental implementations of PPO, SAC, Curiosity, and GAIL based on PyTorch modules. Exports to .onnx and should support inference for all network architectures found in example environmeents. Use --torch in CLI or add framework: pytorch to trainer settings in YAML file to enable PyTorch backend.

Known limitations:

  • No GPU support
  • Does not currently save Curiosity and GAIL networks in checkpoints
  • Checkpoint manager does not currently manage ONNX files
  • SAC runs slower than in TF.

Types of change(s)

  • Bug fix
  • New feature
  • Code refactor
  • Breaking change
  • Documentation update
  • Other (please describe)

Checklist

  • Added tests that prove my fix is effective or that my feature works
  • Updated the changelog (if applicable)
  • Updated the documentation (if applicable)
  • Updated the migration guide (if applicable)

Other comments

awjuliani and others added 30 commits April 14, 2020 15:05
# Conflicts:
#	ml-agents/mlagents/trainers/policy/tf_policy.py
@vincentpierre vincentpierre marked this pull request as ready for review August 19, 2020 21:05
@vincentpierre vincentpierre changed the title [DO NOT MERGE] Develop add fire Develop add fire Aug 19, 2020
@@ -460,7 +460,7 @@ you would like to contribute environments, please see our
head, thighs, shins, feet, arms, forearms and hands.
- Goal: The agents must move its body toward the goal direction without falling.
- `WalkerDynamic`- Goal direction is randomized.
- `WalkerDynamicVariableSpeed`- Goal direction and walking speed are randomized.
- `WalkerDynamicVariableSpeed`- Goal direction and walking speed are randomized.
Copy link
Contributor Author

@ervteng ervteng Aug 20, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I've tried to remove this, this delta isn't picked up by git 👿

@ervteng ervteng changed the title Develop add fire [feature] Add experimental PyTorch support Aug 20, 2020
@vincentpierre vincentpierre merged commit 672b608 into master Aug 20, 2020
@github-actions github-actions bot locked as resolved and limited conversation to collaborators Aug 21, 2021
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

5 participants