Skip to content

Conversation

@Midren
Copy link
Owner

@Midren Midren commented Jun 16, 2023

No description provided.

Midren added 30 commits October 17, 2022 14:25
- Added Pytorch Profiler
- Repeat action number should be used to indicate valid env steps
- Disable distribution args validation
- Moved image reconstruction, reward&discount prediction out of RNN loop
- Added pretrain step
Additionally:
- Changed actor, reward/discount predictors to distribution output
- Enabled tf32_ matmul for faster computation
- Added float16 autocast
- log_prob used instead of mse_loss
- set number of ensembles to 1
- remove normalization layers which are absent in Dreamer

Improved video logging:
- to only imagine track added imagine+observe viz
- same actions are applied accross replay rollout and imagined rollout
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant