Commit Graph

66 Commits

Author SHA1 Message Date
Nicklas Hansen
5ced6dfeb4 auto-convert old checkpoints to new format 2025-02-05 16:26:19 -08:00
Nicklas Hansen
dddc226d25 partial fix to loading checkpoints 2025-01-21 00:10:53 -08:00
Vincent Moens
ae4238946f Conversion tools for state-dicts (#55)
* init

* init

* amend
2025-01-20 15:49:36 -08:00
Nicklas Hansen
a19f91c0b5 enable torch.compile for offline rl + rgb inputs 2024-12-25 12:22:39 -08:00
Nicklas Hansen
e452ca7539 factor pi outputs 2024-12-25 12:08:07 -08:00
Nicklas Hansen
db1865334e refactor pi outputs 2024-12-25 12:02:33 -08:00
Nicklas Hansen
804f9b3949 refactor pi outputs 2024-12-24 03:05:00 -08:00
Nicklas Hansen
66f8c21f58 cache buffer values in offline training 2024-12-19 09:40:04 -08:00
Nicklas Hansen
9cac7c5775 faster offline data loading 2024-12-19 06:52:31 -08:00
Nicklas Hansen
df8a465c8e update offline trainer to use new torch.load api 2024-12-10 16:30:05 -08:00
Nicklas Hansen
2e27fbb6f4 partial fix for loading old checkpoints 2024-12-10 16:04:27 -08:00
Nicklas Hansen
6117bc427d simplify dmcontrol wrappers + upgrade to gymnasium==0.29.1 2024-12-10 15:16:34 -08:00
Nicklas Hansen
32fc2bdf93 refactor policy 2024-12-03 12:22:02 -08:00
Nicklas Hansen
0a79c8bd38 update Dockerfile + environment.yaml 2024-11-28 11:35:15 -08:00
Nicklas Hansen
1bfbcb7794 Merge pull request #49 from nicklashansen/speedups
Speedups
2024-11-10 12:33:32 -08:00
Nicklas Hansen
0c3fcc4619 update readme 2024-11-10 12:32:57 -08:00
Nicklas Hansen
fb07cdac3f update compile print 2024-11-10 12:27:24 -08:00
Nicklas Hansen
c694d286f0 add assertion for compile=true compatibility 2024-11-10 12:25:43 -08:00
Nicklas Hansen
1a77207646 support newest version of myosuite 2024-11-04 15:15:40 -08:00
Nicklas Hansen
b7725e74a5 move cfg conversion to parser.py 2024-10-31 14:52:59 -07:00
Nicklas Hansen
d477619f8d Merge branch 'speedups' of github.com:nicklashansen/tdmpc2 into speedups 2024-10-27 14:24:39 -07:00
Nicklas Hansen
c1dd0c0338 minor QoL improvements in offline pipeline 2024-10-27 14:24:19 -07:00
Nicklas Hansen
c0d3faac77 Merge pull request #48 from vmoens/patch-1
Use torch.compiler.cudagraph_mark_step_begin() in eval
2024-10-27 14:22:51 -07:00
Vincent Moens
3b5f67592c Update offline_trainer.py 2024-10-26 00:33:27 +01:00
Vincent Moens
fad0d1be03 Use torch.compiler.cudagraph_mark_step_begin() in eval 2024-10-26 00:32:16 +01:00
Nicklas Hansen
836547d76f fix eval index + clean up 2024-10-21 14:49:21 -07:00
Nicklas Hansen
970792e2b6 clean up prints 2024-10-18 15:31:25 -07:00
Nicklas Hansen
c3a912e10d Merge pull request #46 from vmoens/cudagraphs
[WIP,Feature] Add cudagraphs and compile option
2024-10-18 09:35:33 -07:00
Nicklas Hansen
38eb46df04 Merge branch 'cudagraphs' of https://github.com/vmoens/tdmpc2 into cudagraphs 2024-10-17 14:57:32 -07:00
vmoens
8b731819a6 merged commits 2024-10-08 10:24:02 -07:00
Nicklas Hansen
a7890b6985 add more results 2024-10-06 21:11:38 -07:00
vmoens
603b67ce66 merged commits 2024-10-03 16:09:48 +01:00
Nicklas Hansen
88095e7899 add more results 2024-08-30 11:28:02 -07:00
Nicklas Hansen
5f6fadec0f update pinned torchrl version 2024-03-27 12:51:40 -07:00
Nicklas Hansen
57158282b4 Merge branch 'main' of github.com:nicklashansen/tdmpc2 into main 2024-02-02 15:56:56 -08:00
Nicklas Hansen
718966c28d reduce # wandb calls 2024-02-02 15:56:54 -08:00
Nicklas Hansen
01cdf0f799 Update README.md 2024-01-24 21:43:20 -08:00
Nicklas Hansen
02b18a48b1 update dockerfile 2024-01-22 17:37:31 -08:00
Nicklas Hansen
e8f1ed6785 update dockerfile + pin all versions 2024-01-21 21:21:44 -08:00
Nicklas Hansen
8b6fe61bed minor fix in print 2024-01-11 18:19:08 -08:00
Nicklas Hansen
aa9c6f33f5 migrate to slicebuffer from torchrl-nightly 2024-01-10 19:53:30 -08:00
Nicklas Hansen
0f3bc77011 amend slicesampler indexing 2024-01-08 10:48:04 -08:00
Nicklas Hansen
31249a8961 separate episodes with nans 2024-01-07 14:21:38 -08:00
Nicklas Hansen
33876d124f add instructions for distributed training 2024-01-07 11:55:07 -08:00
Nicklas Hansen
a7ff00b0cd add option to disable planning with mpc=false 2024-01-04 19:17:47 -08:00
Nicklas Hansen
13cac07759 fix minor logging issue in offline trainer 2024-01-03 10:12:28 -08:00
Nicklas Hansen
1d224cec3a update documentation 2023-12-31 14:38:22 -08:00
Nicklas Hansen
e3c876670a add launcher 2023-12-29 16:37:26 -08:00
Nicklas Hansen
1f6c7771b9 Merge pull request #10 from nicklashansen/experimental
[Feature] Faster replay buffer + support pixel observations
2023-12-28 16:37:27 +01:00
Nicklas Hansen
6cb779aa3a allow missing env dependencies + update readme 2023-12-28 07:33:03 -08:00