Code

[RLlib] DreamerV3 on PyTorch.#45463

Code

Open

sven1977 wants to merge 58 commits intoray-project:masterray-project/ray:masterfrom sven1977:dreamer_v3_torchsven1977/ray:dreamer_v3_torchCopy head branch name to clipboard

+3,914-106

Commits

Commits on Jul 1, 2023

• wip

  Show description for 25a8e20
  sven1977

  committed
  25a8e20

  Copy full SHA for 25a8e20

Commits on Jul 3, 2023

• wip

  Show description for 8860899
  sven1977

  committed
  8860899

  Copy full SHA for 8860899

• wip

  Show description for e205967
  sven1977

  committed
  e205967

  Copy full SHA for e205967

• wip

  Show description for c5e6d29
  sven1977

  committed
  c5e6d29

  Copy full SHA for c5e6d29

Commits on Jul 5, 2023

• wip

  Show description for 48a86ad
  sven1977

  committed
  48a86ad

  Copy full SHA for 48a86ad

• wip

  Show description for 30a5d88
  sven1977

  committed
  30a5d88

  Copy full SHA for 30a5d88

Commits on Aug 3, 2023

• Merge branch 'master' of https://github.com/ray-project/ray into dreamer_v3_torch

  sven1977

  committed
  057d552

  Copy full SHA for 057d552

• wip

  Show description for 732e343
  sven1977

  committed
  732e343

  Copy full SHA for 732e343

• wip

  Show description for c08f767
  sven1977

  committed
  c08f767

  Copy full SHA for c08f767

• wip

  Show description for 8c8dcfe
  sven1977

  committed
  8c8dcfe

  Copy full SHA for 8c8dcfe

Commits on May 19, 2024

• Merge branch 'master' of https://github.com/ray-project/ray into dreamer_v3_torch

  Show description for 606d415
  sven1977

  committed
  606d415

  Copy full SHA for 606d415

Commits on May 21, 2024

• wip

  Show description for 6743a5a
  sven1977

  committed
  6743a5a

  Copy full SHA for 6743a5a

• wip

  Show description for 0a95656
  sven1977

  committed
  0a95656

  Copy full SHA for 0a95656

• Merge branch 'master' of https://github.com/ray-project/ray into dreamer_v3_torch

  sven1977

  committed
  27bc30d

  Copy full SHA for 27bc30d

• wip

  Show description for b82c7a6
  sven1977

  committed
  b82c7a6

  Copy full SHA for b82c7a6

• torch compilation tests running (Atari not tested due to Mac error)

  Show description for 82d5e84
  sven1977

  committed
  82d5e84

  Copy full SHA for 82d5e84

• LINT

  Show description for 8bf413e
  sven1977

  committed
  8bf413e

  Copy full SHA for 8bf413e

• compilation tests run again, even with latest pytorch version, which was responsible for breaking a test

  Show description for 5426077
  sven1977

  committed
  5426077

  Copy full SHA for 5426077

• GPU on EnvRunner fixes

  Show description for c711bc4
  sven1977

  committed
  c711bc4

  Copy full SHA for c711bc4

• GPU in Stride2D fixes

  Show description for a32330b
  sven1977

  committed
  a32330b

  Copy full SHA for a32330b

• run now on 1 GPU (local learner) w/ shared EnvRunner/Learner RLModule and Atari (Pong) on a V100 machine. 👍

  Show description for 7881e51
  sven1977

  committed
  7881e51

  Copy full SHA for 7881e51

• DM Control Suite example

  Show description for 943f3c8
  sven1977

  committed
  943f3c8

  Copy full SHA for 943f3c8

Commits on May 22, 2024

• fixed separation issue of actor/critic losses messing with the world model grads

  Show description for 798ffd9
  sven1977

  committed
  798ffd9

  Copy full SHA for 798ffd9

• tuned_examples cleanup

  Show description for acd74fa
  sven1977

  committed
  acd74fa

  Copy full SHA for acd74fa

• cleanup

  Show description for 9217669
  sven1977

  committed
  9217669

  Copy full SHA for 9217669

• fix critic loss (the main reduction has to be mean instead of sum)

  Show description for dff6c34
  sven1977

  committed
  dff6c34

  Copy full SHA for dff6c34

• add support for reporting posterior predicted (decoder) videos

  Show description for 6573008
  sven1977

  committed
  6573008

  Copy full SHA for 6573008

• less frequent video logging to save time (very expensive)

  Show description for 7d4333f
  sven1977

  committed
  7d4333f

  Copy full SHA for 7d4333f

• fix: less frequent video logging to save time (very expensive)

  Show description for 907c8bd
  sven1977

  committed
  907c8bd

  Copy full SHA for 907c8bd

• fix: less frequent video logging to save time (very expensive)

  Show description for c5e8f2b
  sven1977

  committed
  c5e8f2b

  Copy full SHA for c5e8f2b

• adjust initializers to match 100% those of Author's and RLlib's tf versions.

  Show description for d9aa5f2
  sven1977

  committed
  d9aa5f2

  Copy full SHA for d9aa5f2

• fix

  Show description for 3a275a7
  sven1977

  committed
  3a275a7

  Copy full SHA for 3a275a7

• fix

  Show description for ecf276b
  sven1977

  committed
  ecf276b

  Copy full SHA for ecf276b

Commits on May 23, 2024

• fix

  Show description for 090a7e5
  sven1977

  committed
  090a7e5

  Copy full SHA for 090a7e5

• fix

  Show description for 7afb542
  sven1977

  committed
  7afb542

  Copy full SHA for 7afb542

• add independent wrapper around one-hot-categorical to representation layer's (32x32 z state) distribution.

  Show description for d6abf42
  sven1977

  committed
  d6abf42

  Copy full SHA for d6abf42

• support dream data summaries comparing dreamed trajectories (from prior states and w/ videos) to the sampled ones.

  Show description for 86aa749
  sven1977

  committed
  86aa749

  Copy full SHA for 86aa749

• enable dream data video/reports.

  Show description for ee1ca8c
  sven1977

  committed
  ee1ca8c

  Copy full SHA for ee1ca8c

• manual GRU

  Show description for 7f9a2ba
  sven1977

  committed
  7f9a2ba

  Copy full SHA for 7f9a2ba

• GPU fix

  Show description for e6025a7
  sven1977

  committed
  e6025a7

  Copy full SHA for e6025a7

• move even more toward JAX repo by initializing all layers via Danijar's almost-glorot-normal initializer.

  Show description for 4dba2f7
  sven1977

  committed
  4dba2f7

  Copy full SHA for 4dba2f7

• fix dreamed reports (by reducing to B=1, which seems to work with WandB; could be that WandB doesn't like lists of lists).

  Show description for d785a4c
  sven1977

  committed
  d785a4c

  Copy full SHA for d785a4c

Commits on May 24, 2024

• small cleanup

  Show description for fe828b7
  sven1977

  committed
  fe828b7

  Copy full SHA for fe828b7

• Merge branch 'master' of https://github.com/ray-project/ray into dreamer_v3_torch

  sven1977

  committed
  7b06a4f

  Copy full SHA for 7b06a4f

Commits on May 27, 2024

• Big fix in DreamerV3 EnvRunner: States were not properly passed around between calls to sample()

  Show description for a111c52
  sven1977

  committed
  a111c52

  Copy full SHA for a111c52

• Merge branch 'master' of https://github.com/ray-project/ray into dreamer_v3_torch

  sven1977

  committed
  45640ba

  Copy full SHA for 45640ba

• cleanup

  Show description for 885bb39
  sven1977

  committed
  885bb39

  Copy full SHA for 885bb39

Commits on May 29, 2024

• Merge branch 'master' of https://github.com/ray-project/ray into dreamer_v3_torch

  Show description for 23b9857
  sven1977

  committed
  23b9857

  Copy full SHA for 23b9857

Commits on May 31, 2024

• merge

  Show description for 9d13f63
  sven1977

  committed
  9d13f63

  Copy full SHA for 9d13f63

• wip

  Show description for 11feb70
  sven1977

  committed
  11feb70

  Copy full SHA for 11feb70

Commits on Jun 4, 2024

• Merge branch 'master' of https://github.com/ray-project/ray into dreamer_v3_torch

  sven1977

  committed
  fdeff47

  Copy full SHA for fdeff47

• LINT

  Show description for beb9c75
  sven1977

  committed
  beb9c75

  Copy full SHA for beb9c75

• Merge branch 'master' of https://github.com/ray-project/ray into dreamer_v3_torch

  sven1977

  committed
  6d7dfa5

  Copy full SHA for 6d7dfa5

• torch Stats just like tf Stats to try whether the torch.nanmean() on the Stats.values list is the problem ... to be checked

  Show description for 94bf558
  sven1977

  committed
  94bf558

  Copy full SHA for 94bf558

Commits on Jun 8, 2024

• Merge branch 'master' of https://github.com/ray-project/ray into dreamer_v3_torch

  Show description for ef23d28
  sven1977

  committed
  ef23d28

  Copy full SHA for ef23d28

• wip

  Show description for 9d62aee
  sven1977

  committed
  9d62aee

  Copy full SHA for 9d62aee

Commits on Jun 14, 2024

• Merge branch 'master' of https://github.com/ray-project/ray into dreamer_v3_torch

  sven1977

  committed
  036901a

  Copy full SHA for 036901a

• Merge branch 'master' of https://github.com/ray-project/ray into appo_on_new_api_stack_w_env_runner_and_connectorv2

  Show description for f361be7
  sven1977

  committed
  f361be7

  Copy full SHA for f361be7