Skip to content

[CI] Separate GPU and CPU tests with pytest markers#3404

Merged
vmoens merged 6 commits intogh/vmoens/207/basefrom
gh/vmoens/207/head
Jan 29, 2026
Merged

[CI] Separate GPU and CPU tests with pytest markers#3404
vmoens merged 6 commits intogh/vmoens/207/basefrom
gh/vmoens/207/head

Conversation

@vmoens
Copy link
Collaborator

@vmoens vmoens commented Jan 28, 2026 •
edited
Loading

Stack from ghstack (oldest at bottom):

Add @pytest.mark.gpu to tests that require CUDA, and update run_all.sh
to filter tests based on whether running on GPU or CPU machines.

Changes:

  • Register 'gpu' marker in pytest.ini and conftest.py
  • Add @pytest.mark.gpu to ~30 tests that explicitly require CUDA
  • Update run_all.sh to use GPU_MARKER_FILTER:
    • GPU jobs (CU_VERSION != cpu): run only @pytest.mark.gpu tests
    • CPU jobs (CU_VERSION = cpu): run all tests except @pytest.mark.gpu

This significantly reduces GPU machine usage by running only GPU-requiring
tests on expensive GPU runners (~30 tests instead of ~2000+). Tests that
can run on either device will run on CPU machines only.

The optimization can be disabled by setting TORCHRL_GPU_FILTER=0.

@vmoens
Update
d9a8af0 
[ghstack-poisoned]
vmoens added a commit that referenced this pull request Jan 28, 2026
@vmoens
[CI] Separate GPU and CPU tests with pytest markers
94276c2 
Add pytest.mark.gpu to tests that require CUDA, and update run_all.sh
to filter tests based on whether running on GPU or CPU machines.

Changes:
- Register 'gpu' marker in pytest.ini and conftest.py
- Add pytest.mark.gpu to ~30 tests that explicitly require CUDA
- Update run_all.sh to use GPU_MARKER_FILTER:
  - GPU jobs (CU_VERSION != cpu): run only pytest.mark.gpu tests
  - CPU jobs (CU_VERSION = cpu): run all tests except pytest.mark.gpu

This significantly reduces GPU machine usage by running only GPU-requiring
tests on expensive GPU runners (~30 tests instead of ~2000+). Tests that
can run on either device will run on CPU machines only.

The optimization can be disabled by setting TORCHRL_GPU_FILTER=0.


ghstack-source-id: ca9778d
Pull-Request: #3404
@pytorch-bot
Copy link

pytorch-bot bot commented Jan 28, 2026 •
edited
Loading

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/rl/3404

Note: Links to docs will display an error until the docs builds have been completed.

This comment was automatically generated by Dr. CI and updates every 15 minutes.

This was referenced Jan 28, 2026
Merged
Open
@github-actions github-actions bot added the CI Has to do with CI setup (e.g. wheels & builds, tests...) label Jan 28, 2026
@meta-cla meta-cla bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Jan 28, 2026
@vmoens
Update
f670508 
[ghstack-poisoned]
vmoens added a commit that referenced this pull request Jan 28, 2026
@vmoens
[CI] Separate GPU and CPU tests with pytest markers
9195b21 
Add pytest.mark.gpu to tests that require CUDA, and update run_all.sh
to filter tests based on whether running on GPU or CPU machines.

Changes:
- Register 'gpu' marker in pytest.ini and conftest.py
- Add pytest.mark.gpu to ~30 tests that explicitly require CUDA
- Update run_all.sh to use GPU_MARKER_FILTER:
  - GPU jobs (CU_VERSION != cpu): run only pytest.mark.gpu tests
  - CPU jobs (CU_VERSION = cpu): run all tests except pytest.mark.gpu

This significantly reduces GPU machine usage by running only GPU-requiring
tests on expensive GPU runners (~30 tests instead of ~2000+). Tests that
can run on either device will run on CPU machines only.

The optimization can be disabled by setting TORCHRL_GPU_FILTER=0.

ghstack-source-id: 9235913
Pull-Request: #3404
vmoens added a commit that referenced this pull request Jan 28, 2026
@vmoens
[CI] Separate GPU and CPU tests with pytest markers
dbb4a68 
Add pytest.mark.gpu to tests that require CUDA, and update run_all.sh
to filter tests based on whether running on GPU or CPU machines.

Changes:
- Register 'gpu' marker in pytest.ini and conftest.py
- Add pytest.mark.gpu to ~30 tests that explicitly require CUDA
- Update run_all.sh to use GPU_MARKER_FILTER:
  - GPU jobs (CU_VERSION != cpu): run only pytest.mark.gpu tests
  - CPU jobs (CU_VERSION = cpu): run all tests except pytest.mark.gpu

This significantly reduces GPU machine usage by running only GPU-requiring
tests on expensive GPU runners (~30 tests instead of ~2000+). Tests that
can run on either device will run on CPU machines only.

The optimization can be disabled by setting TORCHRL_GPU_FILTER=0.

ghstack-source-id: 9235913
Pull-Request: #3404
@vmoens
Update
8910a1d 
[ghstack-poisoned]
vmoens added a commit that referenced this pull request Jan 28, 2026
@vmoens
[CI] Separate GPU and CPU tests with pytest markers
d55b8c9 
Add pytest.mark.gpu to tests that require CUDA, and update run_all.sh
to filter tests based on whether running on GPU or CPU machines.

Changes:
- Register 'gpu' marker in pytest.ini and conftest.py
- Add pytest.mark.gpu to ~30 tests that explicitly require CUDA
- Update run_all.sh to use GPU_MARKER_FILTER:
  - GPU jobs (CU_VERSION != cpu): run only pytest.mark.gpu tests
  - CPU jobs (CU_VERSION = cpu): run all tests except pytest.mark.gpu

This significantly reduces GPU machine usage by running only GPU-requiring
tests on expensive GPU runners (~30 tests instead of ~2000+). Tests that
can run on either device will run on CPU machines only.

The optimization can be disabled by setting TORCHRL_GPU_FILTER=0.

ghstack-source-id: 69148f2
Pull-Request: #3404
vmoens added 3 commits January 29, 2026 09:36
@vmoens
Update
e14d3ea 
[ghstack-poisoned]
@vmoens
Update
3fc3208 
[ghstack-poisoned]
@vmoens
Update
edc1d74 
[ghstack-poisoned]
vmoens added a commit that referenced this pull request Jan 29, 2026
@vmoens
[CI] Separate GPU and CPU tests with pytest markers (#3404)
05aa8a8 
Add pytest.mark.gpu to tests that require CUDA, and update run_all.sh
to filter tests based on whether running on GPU or CPU machines.

Changes:
- Register 'gpu' marker in pytest.ini and conftest.py
- Add pytest.mark.gpu to ~30 tests that explicitly require CUDA
- Update run_all.sh to use GPU_MARKER_FILTER:
  - GPU jobs (CU_VERSION != cpu): run only pytest.mark.gpu tests
  - CPU jobs (CU_VERSION = cpu): run all tests except pytest.mark.gpu

This significantly reduces GPU machine usage by running only GPU-requiring
tests on expensive GPU runners (~30 tests instead of ~2000+). Tests that
can run on either device will run on CPU machines only.

The optimization can be disabled by setting TORCHRL_GPU_FILTER=0.

ghstack-source-id: cf14d1e
Pull-Request: #3404
@vmoens vmoens merged commit edc1d74 into gh/vmoens/207/base Jan 29, 2026
71 of 74 checks passed
@vmoens vmoens deleted the gh/vmoens/207/head branch January 29, 2026 09:58
@github-actions
Copy link
Contributor

github-actions bot commented Jan 29, 2026

$\color{#D29922}\textsf{\Large⚠\kern{0.2cm}\normalsize Warning}$ Result of CPU Benchmark Tests

Total Benchmarks: 153. Improved: $\large\color{#35bf28}5$. Worsened: $\large\color{#d91a1a}15$.

Expand to view detailed results
Name Max Mean Ops Ops on Repo HEAD Change
test_tensor_to_bytestream_speed[pickle] 79.7458μs 78.7644μs 12.6961 KOps/s 12.1608 KOps/s $\color{#35bf28}+4.40\%$
test_tensor_to_bytestream_speed[torch.save] 0.1362ms 0.1354ms 7.3865 KOps/s 7.1133 KOps/s $\color{#35bf28}+3.84\%$
test_tensor_to_bytestream_speed[untyped_storage] 0.1073s 0.1070s 9.3453 Ops/s 9.1166 Ops/s $\color{#35bf28}+2.51\%$
test_tensor_to_bytestream_speed[numpy] 2.5347μs 2.5238μs 396.2214 KOps/s 385.2515 KOps/s $\color{#35bf28}+2.85\%$
test_tensor_to_bytestream_speed[safetensors] 37.5882μs 37.2559μs 26.8414 KOps/s 26.3834 KOps/s $\color{#35bf28}+1.74\%$
test_simple 0.6586s 0.5711s 1.7510 Ops/s 1.7686 Ops/s $\color{#d91a1a}-1.00\%$
test_transformed 1.2262s 1.1319s 0.8835 Ops/s 0.8800 Ops/s $\color{#35bf28}+0.40\%$
test_serial 1.6694s 1.6632s 0.6012 Ops/s 0.6115 Ops/s $\color{#d91a1a}-1.67\%$
test_parallel 1.2430s 1.1564s 0.8647 Ops/s 0.8955 Ops/s $\color{#d91a1a}-3.43\%$
test_step_mdp_speed[True-True-True-True-True] 0.3181ms 43.9394μs 22.7586 KOps/s 23.4200 KOps/s $\color{#d91a1a}-2.82\%$
test_step_mdp_speed[True-True-True-True-False] 64.4130μs 23.6698μs 42.2479 KOps/s 40.8300 KOps/s $\color{#35bf28}+3.47\%$
test_step_mdp_speed[True-True-True-False-True] 77.1840μs 24.0562μs 41.5694 KOps/s 40.3185 KOps/s $\color{#35bf28}+3.10\%$
test_step_mdp_speed[True-True-True-False-False] 50.2230μs 13.1823μs 75.8591 KOps/s 73.8273 KOps/s $\color{#35bf28}+2.75\%$
test_step_mdp_speed[True-True-False-True-True] 90.5840μs 46.8965μs 21.3236 KOps/s 21.3943 KOps/s $\color{#d91a1a}-0.33\%$
test_step_mdp_speed[True-True-False-True-False] 89.6540μs 26.6049μs 37.5871 KOps/s 36.9902 KOps/s $\color{#35bf28}+1.61\%$
test_step_mdp_speed[True-True-False-False-True] 69.3730μs 26.9723μs 37.0751 KOps/s 36.0685 KOps/s $\color{#35bf28}+2.79\%$
test_step_mdp_speed[True-True-False-False-False] 91.3940μs 15.8294μs 63.1734 KOps/s 62.5141 KOps/s $\color{#35bf28}+1.05\%$
test_step_mdp_speed[True-False-True-True-True] 95.3840μs 49.3414μs 20.2669 KOps/s 20.1023 KOps/s $\color{#35bf28}+0.82\%$
test_step_mdp_speed[True-False-True-True-False] 62.3430μs 29.5902μs 33.7950 KOps/s 33.6851 KOps/s $\color{#35bf28}+0.33\%$
test_step_mdp_speed[True-False-True-False-True] 67.2730μs 27.2613μs 36.6821 KOps/s 36.7805 KOps/s $\color{#d91a1a}-0.27\%$
test_step_mdp_speed[True-False-True-False-False] 38.1810μs 15.9333μs 62.7618 KOps/s 62.7226 KOps/s $\color{#35bf28}+0.06\%$
test_step_mdp_speed[True-False-False-True-True] 81.7140μs 51.0812μs 19.5767 KOps/s 19.5334 KOps/s $\color{#35bf28}+0.22\%$
test_step_mdp_speed[True-False-False-True-False] 0.1040ms 31.2477μs 32.0023 KOps/s 31.4143 KOps/s $\color{#35bf28}+1.87\%$
test_step_mdp_speed[True-False-False-False-True] 58.8730μs 29.1867μs 34.2622 KOps/s 33.5304 KOps/s $\color{#35bf28}+2.18\%$
test_step_mdp_speed[True-False-False-False-False] 43.1620μs 18.4371μs 54.2386 KOps/s 53.2319 KOps/s $\color{#35bf28}+1.89\%$
test_step_mdp_speed[False-True-True-True-True] 98.0240μs 48.6721μs 20.5457 KOps/s 20.2126 KOps/s $\color{#35bf28}+1.65\%$
test_step_mdp_speed[False-True-True-True-False] 60.0530μs 29.6028μs 33.7806 KOps/s 33.0017 KOps/s $\color{#35bf28}+2.36\%$
test_step_mdp_speed[False-True-True-False-True] 59.1930μs 30.2142μs 33.0970 KOps/s 32.0660 KOps/s $\color{#35bf28}+3.22\%$
test_step_mdp_speed[False-True-True-False-False] 51.3730μs 17.6511μs 56.6537 KOps/s 55.2539 KOps/s $\color{#35bf28}+2.53\%$
test_step_mdp_speed[False-True-False-True-True] 2.7691ms 52.0368μs 19.2172 KOps/s 18.9454 KOps/s $\color{#35bf28}+1.43\%$
test_step_mdp_speed[False-True-False-True-False] 70.0830μs 32.2494μs 31.0083 KOps/s 30.9747 KOps/s $\color{#35bf28}+0.11\%$
test_step_mdp_speed[False-True-False-False-True] 67.8030μs 33.1833μs 30.1356 KOps/s 30.2052 KOps/s $\color{#d91a1a}-0.23\%$
test_step_mdp_speed[False-True-False-False-False] 44.3720μs 20.1629μs 49.5960 KOps/s 48.1949 KOps/s $\color{#35bf28}+2.91\%$
test_step_mdp_speed[False-False-True-True-True] 84.1730μs 53.2782μs 18.7694 KOps/s 18.2036 KOps/s $\color{#35bf28}+3.11\%$
test_step_mdp_speed[False-False-True-True-False] 72.8740μs 34.6547μs 28.8561 KOps/s 28.2991 KOps/s $\color{#35bf28}+1.97\%$
test_step_mdp_speed[False-False-True-False-True] 63.7530μs 32.8793μs 30.4143 KOps/s 29.5497 KOps/s $\color{#35bf28}+2.93\%$
test_step_mdp_speed[False-False-True-False-False] 51.8220μs 20.2601μs 49.3582 KOps/s 48.6020 KOps/s $\color{#35bf28}+1.56\%$
test_step_mdp_speed[False-False-False-True-True] 90.5840μs 55.8016μs 17.9206 KOps/s 17.5884 KOps/s $\color{#35bf28}+1.89\%$
test_step_mdp_speed[False-False-False-True-False] 66.7630μs 37.1242μs 26.9366 KOps/s 26.4261 KOps/s $\color{#35bf28}+1.93\%$
test_step_mdp_speed[False-False-False-False-True] 68.5730μs 34.6067μs 28.8961 KOps/s 28.1961 KOps/s $\color{#35bf28}+2.48\%$
test_step_mdp_speed[False-False-False-False-False] 51.9930μs 22.4750μs 44.4939 KOps/s 43.6647 KOps/s $\color{#35bf28}+1.90\%$
test_non_tensor_env_rollout_speed[1000-single-True] 0.8414s 0.7511s 1.3314 Ops/s 1.3242 Ops/s $\color{#35bf28}+0.54\%$
test_non_tensor_env_rollout_speed[1000-single-False] 0.7061s 0.6147s 1.6269 Ops/s 1.6094 Ops/s $\color{#35bf28}+1.09\%$
test_non_tensor_env_rollout_speed[1000-serial-no-buffers-True] 1.7153s 1.6386s 0.6103 Ops/s 0.6064 Ops/s $\color{#35bf28}+0.64\%$
test_non_tensor_env_rollout_speed[1000-serial-no-buffers-False] 1.4942s 1.4136s 0.7074 Ops/s 0.6840 Ops/s $\color{#35bf28}+3.42\%$
test_non_tensor_env_rollout_speed[1000-serial-buffers-True] 1.9498s 1.8777s 0.5326 Ops/s 0.5320 Ops/s $\color{#35bf28}+0.11\%$
test_non_tensor_env_rollout_speed[1000-serial-buffers-False] 1.7321s 1.6644s 0.6008 Ops/s 0.6023 Ops/s $\color{#d91a1a}-0.25\%$
test_non_tensor_env_rollout_speed[1000-parallel-no-buffers-True] 4.6654s 4.5895s 0.2179 Ops/s 0.2185 Ops/s $\color{#d91a1a}-0.27\%$
test_non_tensor_env_rollout_speed[1000-parallel-no-buffers-False] 4.6075s 4.4718s 0.2236 Ops/s 0.2244 Ops/s $\color{#d91a1a}-0.34\%$
test_non_tensor_env_rollout_speed[1000-parallel-buffers-True] 2.1526s 1.9594s 0.5104 Ops/s 0.5113 Ops/s $\color{#d91a1a}-0.19\%$
test_non_tensor_env_rollout_speed[1000-parallel-buffers-False] 1.7688s 1.6660s 0.6002 Ops/s 0.6161 Ops/s $\color{#d91a1a}-2.58\%$
test_values[generalized_advantage_estimate-True-True] 10.3463ms 10.0425ms 99.5765 Ops/s 96.6993 Ops/s $\color{#35bf28}+2.98\%$
test_values[vec_generalized_advantage_estimate-True-True] 19.9151ms 17.9202ms 55.8031 Ops/s 87.0201 Ops/s $\textbf{\color{#d91a1a}-35.87\%}$
test_values[td0_return_estimate-False-False] 0.2111ms 0.1300ms 7.6926 KOps/s 7.3921 KOps/s $\color{#35bf28}+4.07\%$
test_values[td1_return_estimate-False-False] 29.0190ms 27.7722ms 36.0072 Ops/s 35.1554 Ops/s $\color{#35bf28}+2.42\%$
test_values[vec_td1_return_estimate-False-False] 18.6609ms 17.9382ms 55.7471 Ops/s 87.1853 Ops/s $\textbf{\color{#d91a1a}-36.06\%}$
test_values[td_lambda_return_estimate-True-False] 42.7938ms 40.9298ms 24.4321 Ops/s 23.7836 Ops/s $\color{#35bf28}+2.73\%$
test_values[vec_td_lambda_return_estimate-True-False] 19.1589ms 17.9539ms 55.6981 Ops/s 87.1528 Ops/s $\textbf{\color{#d91a1a}-36.09\%}$
test_gae_speed[generalized_advantage_estimate-False-1-512] 9.1294ms 8.8981ms 112.3829 Ops/s 108.6886 Ops/s $\color{#35bf28}+3.40\%$
test_gae_speed[vec_generalized_advantage_estimate-True-1-512] 1.7867ms 1.5293ms 653.9132 Ops/s 648.4828 Ops/s $\color{#35bf28}+0.84\%$
test_gae_speed[vec_generalized_advantage_estimate-False-1-512] 0.5539ms 0.4315ms 2.3177 KOps/s 2.3599 KOps/s $\color{#d91a1a}-1.79\%$
test_gae_speed[vec_generalized_advantage_estimate-True-32-512] 35.0872ms 34.6872ms 28.8291 Ops/s 34.3806 Ops/s $\textbf{\color{#d91a1a}-16.15\%}$
test_gae_speed[vec_generalized_advantage_estimate-False-32-512] 2.1539ms 1.7425ms 573.9039 Ops/s 570.3284 Ops/s $\color{#35bf28}+0.63\%$
test_dqn_speed[False-None] 1.7797ms 1.3836ms 722.7352 Ops/s 720.6519 Ops/s $\color{#35bf28}+0.29\%$
test_dqn_speed[False-backward] 1.9768ms 1.9113ms 523.2015 Ops/s 527.2238 Ops/s $\color{#d91a1a}-0.76\%$
test_dqn_speed[True-None] 0.7230ms 0.5515ms 1.8131 KOps/s 1.8831 KOps/s $\color{#d91a1a}-3.71\%$
test_dqn_speed[True-backward] 1.0171ms 0.9857ms 1.0145 KOps/s 993.4563 Ops/s $\color{#35bf28}+2.12\%$
test_dqn_speed[reduce-overhead-None] 0.9276ms 0.5186ms 1.9282 KOps/s 1.8924 KOps/s $\color{#35bf28}+1.89\%$
test_ddpg_speed[False-None] 0.1994s 3.4279ms 291.7196 Ops/s 352.7294 Ops/s $\textbf{\color{#d91a1a}-17.30\%}$
test_ddpg_speed[False-backward] 4.1190ms 4.0230ms 248.5734 Ops/s 245.3491 Ops/s $\color{#35bf28}+1.31\%$
test_ddpg_speed[True-None] 1.8642ms 1.3840ms 722.5543 Ops/s 710.8613 Ops/s $\color{#35bf28}+1.64\%$
test_ddpg_speed[True-backward] 2.4051ms 2.3536ms 424.8863 Ops/s 377.5588 Ops/s $\textbf{\color{#35bf28}+12.54\%}$
test_ddpg_speed[reduce-overhead-None] 1.8593ms 1.3748ms 727.3951 Ops/s 713.6841 Ops/s $\color{#35bf28}+1.92\%$
test_sac_speed[False-None] 8.4930ms 7.9338ms 126.0431 Ops/s 126.1598 Ops/s $\color{#d91a1a}-0.09\%$
test_sac_speed[False-backward] 11.7698ms 11.1666ms 89.5525 Ops/s 89.1815 Ops/s $\color{#35bf28}+0.42\%$
test_sac_speed[True-None] 2.5603ms 2.1414ms 466.9815 Ops/s 446.5895 Ops/s $\color{#35bf28}+4.57\%$
test_sac_speed[True-backward] 4.0994ms 3.9799ms 251.2617 Ops/s 245.1600 Ops/s $\color{#35bf28}+2.49\%$
test_sac_speed[reduce-overhead-None] 2.2613ms 2.1195ms 471.8017 Ops/s 473.4617 Ops/s $\color{#d91a1a}-0.35\%$
test_redq_speed[False-None] 10.7342ms 10.2855ms 97.2240 Ops/s 97.2453 Ops/s $\color{#d91a1a}-0.02\%$
test_redq_speed[False-backward] 18.6299ms 17.8758ms 55.9417 Ops/s 56.5696 Ops/s $\color{#d91a1a}-1.11\%$
test_redq_speed[True-None] 4.9486ms 4.4841ms 223.0100 Ops/s 225.9144 Ops/s $\color{#d91a1a}-1.29\%$
test_redq_speed[True-backward] 9.9930ms 9.8203ms 101.8296 Ops/s 101.5085 Ops/s $\color{#35bf28}+0.32\%$
test_redq_speed[reduce-overhead-None] 4.9551ms 4.3776ms 228.4359 Ops/s 224.5306 Ops/s $\color{#35bf28}+1.74\%$
test_redq_deprec_speed[False-None] 11.5050ms 10.9223ms 91.5556 Ops/s 93.3891 Ops/s $\color{#d91a1a}-1.96\%$
test_redq_deprec_speed[False-backward] 15.9715ms 15.6964ms 63.7088 Ops/s 65.2134 Ops/s $\color{#d91a1a}-2.31\%$
test_redq_deprec_speed[True-None] 4.2118ms 3.6960ms 270.5601 Ops/s 269.8903 Ops/s $\color{#35bf28}+0.25\%$
test_redq_deprec_speed[True-backward] 7.8395ms 7.6472ms 130.7675 Ops/s 110.1815 Ops/s $\textbf{\color{#35bf28}+18.68\%}$
test_redq_deprec_speed[reduce-overhead-None] 4.1677ms 3.6131ms 276.7696 Ops/s 276.5828 Ops/s $\color{#35bf28}+0.07\%$
test_td3_speed[False-None] 8.1824ms 7.9941ms 125.0922 Ops/s 124.6787 Ops/s $\color{#35bf28}+0.33\%$
test_td3_speed[False-backward] 11.3045ms 10.8514ms 92.1538 Ops/s 91.5590 Ops/s $\color{#35bf28}+0.65\%$
test_td3_speed[True-None] 1.8796ms 1.8333ms 545.4560 Ops/s 544.5575 Ops/s $\color{#35bf28}+0.17\%$
test_td3_speed[True-backward] 3.7719ms 3.6315ms 275.3696 Ops/s 270.3886 Ops/s $\color{#35bf28}+1.84\%$
test_td3_speed[reduce-overhead-None] 1.8512ms 1.7914ms 558.2139 Ops/s 549.7468 Ops/s $\color{#35bf28}+1.54\%$
test_cql_speed[False-None] 30.5857ms 26.3159ms 37.9998 Ops/s 37.8281 Ops/s $\color{#35bf28}+0.45\%$
test_cql_speed[False-backward] 39.1317ms 35.9281ms 27.8334 Ops/s 28.2406 Ops/s $\color{#d91a1a}-1.44\%$
test_cql_speed[True-None] 13.1357ms 12.5117ms 79.9251 Ops/s 79.2654 Ops/s $\color{#35bf28}+0.83\%$
test_cql_speed[True-backward] 18.8385ms 18.3638ms 54.4551 Ops/s 53.7966 Ops/s $\color{#35bf28}+1.22\%$
test_cql_speed[reduce-overhead-None] 12.6274ms 12.4015ms 80.6355 Ops/s 79.8477 Ops/s $\color{#35bf28}+0.99\%$
test_a2c_speed[False-None] 5.9510ms 5.4176ms 184.5853 Ops/s 185.8306 Ops/s $\color{#d91a1a}-0.67\%$
test_a2c_speed[False-backward] 12.1481ms 11.8700ms 84.2463 Ops/s 84.3664 Ops/s $\color{#d91a1a}-0.14\%$
test_a2c_speed[True-None] 4.1581ms 3.7296ms 268.1248 Ops/s 272.9750 Ops/s $\color{#d91a1a}-1.78\%$
test_a2c_speed[True-backward] 8.7930ms 8.6284ms 115.8969 Ops/s 115.2585 Ops/s $\color{#35bf28}+0.55\%$
test_a2c_speed[reduce-overhead-None] 3.8018ms 3.6715ms 272.3684 Ops/s 271.9262 Ops/s $\color{#35bf28}+0.16\%$
test_ppo_speed[False-None] 6.0184ms 5.8590ms 170.6771 Ops/s 171.0868 Ops/s $\color{#d91a1a}-0.24\%$
test_ppo_speed[False-backward] 12.7847ms 12.4689ms 80.1995 Ops/s 79.6875 Ops/s $\color{#35bf28}+0.64\%$
test_ppo_speed[True-None] 4.1483ms 3.6218ms 276.1077 Ops/s 274.8823 Ops/s $\color{#35bf28}+0.45\%$
test_ppo_speed[True-backward] 8.8566ms 8.4329ms 118.5837 Ops/s 117.8953 Ops/s $\color{#35bf28}+0.58\%$
test_ppo_speed[reduce-overhead-None] 3.6891ms 3.5903ms 278.5315 Ops/s 279.8719 Ops/s $\color{#d91a1a}-0.48\%$
test_reinforce_speed[False-None] 4.7130ms 4.5568ms 219.4542 Ops/s 221.8469 Ops/s $\color{#d91a1a}-1.08\%$
test_reinforce_speed[False-backward] 7.6957ms 7.4227ms 134.7212 Ops/s 136.2955 Ops/s $\color{#d91a1a}-1.16\%$
test_reinforce_speed[True-None] 3.3148ms 2.8536ms 350.4300 Ops/s 351.1335 Ops/s $\color{#d91a1a}-0.20\%$
test_reinforce_speed[True-backward] 8.3960ms 7.8699ms 127.0657 Ops/s 123.9436 Ops/s $\color{#35bf28}+2.52\%$
test_reinforce_speed[reduce-overhead-None] 3.3867ms 2.8666ms 348.8429 Ops/s 339.3555 Ops/s $\color{#35bf28}+2.80\%$
test_iql_speed[False-None] 26.0278ms 20.1633ms 49.5952 Ops/s 50.9086 Ops/s $\color{#d91a1a}-2.58\%$
test_iql_speed[False-backward] 31.8075ms 30.5269ms 32.7580 Ops/s 32.9171 Ops/s $\color{#d91a1a}-0.48\%$
test_iql_speed[True-None] 9.1210ms 8.4998ms 117.6504 Ops/s 113.3626 Ops/s $\color{#35bf28}+3.78\%$
test_iql_speed[True-backward] 17.0691ms 16.7441ms 59.7224 Ops/s 59.4958 Ops/s $\color{#35bf28}+0.38\%$
test_iql_speed[reduce-overhead-None] 8.9173ms 8.5259ms 117.2903 Ops/s 116.7007 Ops/s $\color{#35bf28}+0.51\%$
test_rb_sample[TensorDictReplayBuffer-ListStorage-RandomSampler-4000] 6.1753ms 5.7254ms 174.6603 Ops/s 172.7807 Ops/s $\color{#35bf28}+1.09\%$
test_rb_sample[TensorDictReplayBuffer-LazyMemmapStorage-RandomSampler-10000] 2.0638ms 0.3664ms 2.7290 KOps/s 3.4273 KOps/s $\textbf{\color{#d91a1a}-20.37\%}$
test_rb_sample[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-10000] 0.6527ms 0.3504ms 2.8536 KOps/s 3.0078 KOps/s $\textbf{\color{#d91a1a}-5.13\%}$
test_rb_sample[TensorDictReplayBuffer-ListStorage-SamplerWithoutReplacement-4000] 6.1209ms 5.5518ms 180.1206 Ops/s 177.6370 Ops/s $\color{#35bf28}+1.40\%$
test_rb_sample[TensorDictReplayBuffer-LazyMemmapStorage-SamplerWithoutReplacement-10000] 1.0768ms 0.3512ms 2.8471 KOps/s 3.5876 KOps/s $\textbf{\color{#d91a1a}-20.64\%}$
test_rb_sample[TensorDictReplayBuffer-LazyTensorStorage-SamplerWithoutReplacement-10000] 0.6030ms 0.3325ms 3.0075 KOps/s 3.5191 KOps/s $\textbf{\color{#d91a1a}-14.54\%}$
test_rb_sample[TensorDictReplayBuffer-LazyMemmapStorage-sampler6-10000] 1.8540ms 1.4035ms 712.5088 Ops/s 783.5141 Ops/s $\textbf{\color{#d91a1a}-9.06\%}$
test_rb_sample[TensorDictReplayBuffer-LazyTensorStorage-sampler7-10000] 1.5718ms 1.3201ms 757.5459 Ops/s 843.4427 Ops/s $\textbf{\color{#d91a1a}-10.18\%}$
test_rb_sample[TensorDictPrioritizedReplayBuffer-ListStorage-None-4000] 11.5870ms 5.9087ms 169.2423 Ops/s 173.8227 Ops/s $\color{#d91a1a}-2.64\%$
test_rb_sample[TensorDictPrioritizedReplayBuffer-LazyMemmapStorage-None-10000] 1.0786ms 0.4446ms 2.2491 KOps/s 2.1542 KOps/s $\color{#35bf28}+4.40\%$
test_rb_sample[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-10000] 0.6866ms 0.4696ms 2.1296 KOps/s 2.1518 KOps/s $\color{#d91a1a}-1.03\%$
test_rb_iterate[TensorDictReplayBuffer-ListStorage-RandomSampler-4000] 6.0120ms 5.5958ms 178.7059 Ops/s 176.7325 Ops/s $\color{#35bf28}+1.12\%$
test_rb_iterate[TensorDictReplayBuffer-LazyMemmapStorage-RandomSampler-10000] 1.8290ms 0.2795ms 3.5779 KOps/s 3.2332 KOps/s $\textbf{\color{#35bf28}+10.66\%}$
test_rb_iterate[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-10000] 0.7194ms 0.2997ms 3.3369 KOps/s 3.3565 KOps/s $\color{#d91a1a}-0.59\%$
test_rb_iterate[TensorDictReplayBuffer-ListStorage-SamplerWithoutReplacement-4000] 5.8631ms 5.5513ms 180.1375 Ops/s 177.3186 Ops/s $\color{#35bf28}+1.59\%$
test_rb_iterate[TensorDictReplayBuffer-LazyMemmapStorage-SamplerWithoutReplacement-10000] 1.6974ms 0.3104ms 3.2212 KOps/s 3.4338 KOps/s $\textbf{\color{#d91a1a}-6.19\%}$
test_rb_iterate[TensorDictReplayBuffer-LazyTensorStorage-SamplerWithoutReplacement-10000] 0.6107ms 0.3780ms 2.6454 KOps/s 3.2952 KOps/s $\textbf{\color{#d91a1a}-19.72\%}$
test_rb_iterate[TensorDictPrioritizedReplayBuffer-ListStorage-None-4000] 5.9668ms 5.7574ms 173.6882 Ops/s 171.8121 Ops/s $\color{#35bf28}+1.09\%$
test_rb_iterate[TensorDictPrioritizedReplayBuffer-LazyMemmapStorage-None-10000] 0.8422ms 0.4566ms 2.1903 KOps/s 2.1681 KOps/s $\color{#35bf28}+1.03\%$
test_rb_iterate[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-10000] 0.6335ms 0.4438ms 2.2533 KOps/s 2.2980 KOps/s $\color{#d91a1a}-1.94\%$
test_rb_populate[TensorDictReplayBuffer-ListStorage-RandomSampler-400] 6.4907ms 4.9947ms 200.2121 Ops/s 197.6926 Ops/s $\color{#35bf28}+1.27\%$
test_rb_populate[TensorDictReplayBuffer-LazyMemmapStorage-RandomSampler-400] 13.0781ms 1.9263ms 519.1269 Ops/s 522.6645 Ops/s $\color{#d91a1a}-0.68\%$
test_rb_populate[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-400] 1.0615ms 0.8715ms 1.1474 KOps/s 1.1454 KOps/s $\color{#35bf28}+0.17\%$
test_rb_populate[TensorDictReplayBuffer-ListStorage-SamplerWithoutReplacement-400] 11.1087ms 5.0680ms 197.3166 Ops/s 200.7695 Ops/s $\color{#d91a1a}-1.72\%$
test_rb_populate[TensorDictReplayBuffer-LazyMemmapStorage-SamplerWithoutReplacement-400] 6.6574ms 1.8840ms 530.7949 Ops/s 559.6953 Ops/s $\textbf{\color{#d91a1a}-5.16\%}$
test_rb_populate[TensorDictReplayBuffer-LazyTensorStorage-SamplerWithoutReplacement-400] 11.8139ms 1.2668ms 789.4028 Ops/s 1.0622 KOps/s $\textbf{\color{#d91a1a}-25.68\%}$
test_rb_populate[TensorDictPrioritizedReplayBuffer-ListStorage-None-400] 0.5495s 16.0993ms 62.1146 Ops/s 57.2830 Ops/s $\textbf{\color{#35bf28}+8.43\%}$
test_rb_populate[TensorDictPrioritizedReplayBuffer-LazyMemmapStorage-None-400] 4.2288ms 1.8630ms 536.7811 Ops/s 493.1920 Ops/s $\textbf{\color{#35bf28}+8.84\%}$
test_rb_populate[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-400] 1.8559ms 1.0063ms 993.7112 Ops/s 953.8081 Ops/s $\color{#35bf28}+4.18\%$
test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-10000-10000-100-True] 37.9772ms 35.6010ms 28.0891 Ops/s 27.9380 Ops/s $\color{#35bf28}+0.54\%$
test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-10000-10000-100-False] 20.3285ms 18.5851ms 53.8066 Ops/s 55.6752 Ops/s $\color{#d91a1a}-3.36\%$
test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-100000-10000-100-True] 40.4316ms 36.9538ms 27.0608 Ops/s 26.7861 Ops/s $\color{#35bf28}+1.03\%$
test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-100000-10000-100-False] 20.1200ms 18.4398ms 54.2306 Ops/s 53.0771 Ops/s $\color{#35bf28}+2.17\%$
test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-1000000-10000-100-True] 39.9897ms 38.6119ms 25.8988 Ops/s 25.6690 Ops/s $\color{#35bf28}+0.90\%$
test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-1000000-10000-100-False] 21.8360ms 20.2602ms 49.3579 Ops/s 50.5338 Ops/s $\color{#d91a1a}-2.33\%$

@github-actions
Copy link
Contributor

github-actions bot commented Jan 29, 2026

$\color{#D29922}\textsf{\Large⚠\kern{0.2cm}\normalsize Warning}$ Result of GPU Benchmark Tests

Total Benchmarks: 148. Improved: $\large\color{#35bf28}5$. Worsened: $\large\color{#d91a1a}16$.

Expand to view detailed results
Name Max Mean Ops Ops on Repo HEAD Change
test_tensor_to_bytestream_speed[pickle] 82.8024μs 81.8418μs 12.2187 KOps/s 11.7695 KOps/s $\color{#35bf28}+3.82\%$
test_tensor_to_bytestream_speed[torch.save] 0.1426ms 0.1415ms 7.0691 KOps/s 7.0060 KOps/s $\color{#35bf28}+0.90\%$
test_tensor_to_bytestream_speed[untyped_storage] 0.1410s 0.1407s 7.1083 Ops/s 7.4691 Ops/s $\color{#d91a1a}-4.83\%$
test_tensor_to_bytestream_speed[numpy] 2.9965μs 2.9887μs 334.5893 KOps/s 360.6524 KOps/s $\textbf{\color{#d91a1a}-7.23\%}$
test_tensor_to_bytestream_speed[safetensors] 38.6287μs 37.6219μs 26.5803 KOps/s 26.0331 KOps/s $\color{#35bf28}+2.10\%$
test_simple 0.9236s 0.8323s 1.2015 Ops/s 1.2044 Ops/s $\color{#d91a1a}-0.25\%$
test_transformed 1.5700s 1.4806s 0.6754 Ops/s 0.6819 Ops/s $\color{#d91a1a}-0.94\%$
test_serial 2.4604s 2.3816s 0.4199 Ops/s 0.4273 Ops/s $\color{#d91a1a}-1.73\%$
test_parallel 2.0451s 1.9911s 0.5022 Ops/s 0.5010 Ops/s $\color{#35bf28}+0.24\%$
test_step_mdp_speed[True-True-True-True-True] 0.5136ms 45.7529μs 21.8565 KOps/s 22.2012 KOps/s $\color{#d91a1a}-1.55\%$
test_step_mdp_speed[True-True-True-True-False] 59.0010μs 25.5918μs 39.0750 KOps/s 39.8280 KOps/s $\color{#d91a1a}-1.89\%$
test_step_mdp_speed[True-True-True-False-True] 58.3010μs 25.4601μs 39.2771 KOps/s 38.8573 KOps/s $\color{#35bf28}+1.08\%$
test_step_mdp_speed[True-True-True-False-False] 42.9210μs 14.1835μs 70.5046 KOps/s 71.8991 KOps/s $\color{#d91a1a}-1.94\%$
test_step_mdp_speed[True-True-False-True-True] 0.1126ms 48.3943μs 20.6636 KOps/s 20.5590 KOps/s $\color{#35bf28}+0.51\%$
test_step_mdp_speed[True-True-False-True-False] 59.0410μs 28.5428μs 35.0351 KOps/s 35.6319 KOps/s $\color{#d91a1a}-1.68\%$
test_step_mdp_speed[True-True-False-False-True] 63.5810μs 28.7795μs 34.7470 KOps/s 34.6821 KOps/s $\color{#35bf28}+0.19\%$
test_step_mdp_speed[True-True-False-False-False] 49.0900μs 16.8504μs 59.3456 KOps/s 59.5810 KOps/s $\color{#d91a1a}-0.40\%$
test_step_mdp_speed[True-False-True-True-True] 92.9920μs 51.6284μs 19.3692 KOps/s 19.3127 KOps/s $\color{#35bf28}+0.29\%$
test_step_mdp_speed[True-False-True-True-False] 87.2810μs 30.9973μs 32.2608 KOps/s 31.8517 KOps/s $\color{#35bf28}+1.28\%$
test_step_mdp_speed[True-False-True-False-True] 90.9110μs 28.0665μs 35.6296 KOps/s 34.7931 KOps/s $\color{#35bf28}+2.40\%$
test_step_mdp_speed[True-False-True-False-False] 45.2310μs 16.9053μs 59.1531 KOps/s 59.8684 KOps/s $\color{#d91a1a}-1.19\%$
test_step_mdp_speed[True-False-False-True-True] 96.2810μs 53.4106μs 18.7229 KOps/s 18.6257 KOps/s $\color{#35bf28}+0.52\%$
test_step_mdp_speed[True-False-False-True-False] 80.5220μs 34.1048μs 29.3214 KOps/s 29.6568 KOps/s $\color{#d91a1a}-1.13\%$
test_step_mdp_speed[True-False-False-False-True] 58.3910μs 31.2116μs 32.0394 KOps/s 32.3637 KOps/s $\color{#d91a1a}-1.00\%$
test_step_mdp_speed[True-False-False-False-False] 52.4210μs 19.4361μs 51.4505 KOps/s 51.2961 KOps/s $\color{#35bf28}+0.30\%$
test_step_mdp_speed[False-True-True-True-True] 94.9610μs 52.1881μs 19.1614 KOps/s 19.4082 KOps/s $\color{#d91a1a}-1.27\%$
test_step_mdp_speed[False-True-True-True-False] 59.7710μs 31.2726μs 31.9769 KOps/s 32.3232 KOps/s $\color{#d91a1a}-1.07\%$
test_step_mdp_speed[False-True-True-False-True] 63.4110μs 31.7935μs 31.4529 KOps/s 31.2443 KOps/s $\color{#35bf28}+0.67\%$
test_step_mdp_speed[False-True-True-False-False] 50.5810μs 18.7040μs 53.4644 KOps/s 54.4821 KOps/s $\color{#d91a1a}-1.87\%$
test_step_mdp_speed[False-True-False-True-True] 2.6159ms 54.4950μs 18.3503 KOps/s 18.5782 KOps/s $\color{#d91a1a}-1.23\%$
test_step_mdp_speed[False-True-False-True-False] 70.0110μs 33.6268μs 29.7382 KOps/s 29.4582 KOps/s $\color{#35bf28}+0.95\%$
test_step_mdp_speed[False-True-False-False-True] 62.8910μs 34.9338μs 28.6255 KOps/s 28.5643 KOps/s $\color{#35bf28}+0.21\%$
test_step_mdp_speed[False-True-False-False-False] 51.3710μs 21.6372μs 46.2166 KOps/s 47.7558 KOps/s $\color{#d91a1a}-3.22\%$
test_step_mdp_speed[False-False-True-True-True] 86.7420μs 57.2007μs 17.4823 KOps/s 17.5052 KOps/s $\color{#d91a1a}-0.13\%$
test_step_mdp_speed[False-False-True-True-False] 70.5110μs 36.7504μs 27.2106 KOps/s 27.0980 KOps/s $\color{#35bf28}+0.42\%$
test_step_mdp_speed[False-False-True-False-True] 98.1120μs 34.5790μs 28.9193 KOps/s 28.3049 KOps/s $\color{#35bf28}+2.17\%$
test_step_mdp_speed[False-False-True-False-False] 53.5110μs 21.4399μs 46.6420 KOps/s 47.1360 KOps/s $\color{#d91a1a}-1.05\%$
test_step_mdp_speed[False-False-False-True-True] 98.9510μs 58.7076μs 17.0336 KOps/s 16.8323 KOps/s $\color{#35bf28}+1.20\%$
test_step_mdp_speed[False-False-False-True-False] 77.0010μs 39.6038μs 25.2501 KOps/s 25.2296 KOps/s $\color{#35bf28}+0.08\%$
test_step_mdp_speed[False-False-False-False-True] 72.6810μs 36.7955μs 27.1772 KOps/s 26.6994 KOps/s $\color{#35bf28}+1.79\%$
test_step_mdp_speed[False-False-False-False-False] 52.9010μs 24.0841μs 41.5211 KOps/s 42.0901 KOps/s $\color{#d91a1a}-1.35\%$
test_non_tensor_env_rollout_speed[1000-single-True] 0.7638s 0.7601s 1.3155 Ops/s 1.2855 Ops/s $\color{#35bf28}+2.34\%$
test_non_tensor_env_rollout_speed[1000-single-False] 0.7325s 0.6411s 1.5599 Ops/s 1.5549 Ops/s $\color{#35bf28}+0.32\%$
test_non_tensor_env_rollout_speed[1000-serial-no-buffers-True] 1.7799s 1.7023s 0.5874 Ops/s 0.5891 Ops/s $\color{#d91a1a}-0.28\%$
test_non_tensor_env_rollout_speed[1000-serial-no-buffers-False] 1.5457s 1.4713s 0.6797 Ops/s 0.6810 Ops/s $\color{#d91a1a}-0.19\%$
test_non_tensor_env_rollout_speed[1000-serial-buffers-True] 2.0450s 1.9649s 0.5089 Ops/s 0.5131 Ops/s $\color{#d91a1a}-0.81\%$
test_non_tensor_env_rollout_speed[1000-serial-buffers-False] 1.8117s 1.7333s 0.5769 Ops/s 0.5802 Ops/s $\color{#d91a1a}-0.57\%$
test_non_tensor_env_rollout_speed[1000-parallel-no-buffers-True] 4.9341s 4.7332s 0.2113 Ops/s 0.2104 Ops/s $\color{#35bf28}+0.43\%$
test_non_tensor_env_rollout_speed[1000-parallel-no-buffers-False] 4.6519s 4.5278s 0.2209 Ops/s 0.2214 Ops/s $\color{#d91a1a}-0.25\%$
test_non_tensor_env_rollout_speed[1000-parallel-buffers-True] 2.0785s 1.9819s 0.5046 Ops/s 0.5046 Ops/s $\color{#d91a1a}-0.01\%$
test_non_tensor_env_rollout_speed[1000-parallel-buffers-False] 1.8526s 1.7105s 0.5846 Ops/s 0.5912 Ops/s $\color{#d91a1a}-1.12\%$
test_values[generalized_advantage_estimate-True-True] 22.0257ms 21.5869ms 46.3244 Ops/s 47.8013 Ops/s $\color{#d91a1a}-3.09\%$
test_values[vec_generalized_advantage_estimate-True-True] 0.1436s 3.8205ms 261.7492 Ops/s 267.6292 Ops/s $\color{#d91a1a}-2.20\%$
test_values[td0_return_estimate-False-False] 0.1114ms 86.1417μs 11.6088 KOps/s 11.8577 KOps/s $\color{#d91a1a}-2.10\%$
test_values[td1_return_estimate-False-False] 51.8995ms 51.3880ms 19.4598 Ops/s 19.6203 Ops/s $\color{#d91a1a}-0.82\%$
test_values[vec_td1_return_estimate-False-False] 1.3694ms 1.1160ms 896.0600 Ops/s 908.4098 Ops/s $\color{#d91a1a}-1.36\%$
test_values[td_lambda_return_estimate-True-False] 84.4318ms 83.9200ms 11.9161 Ops/s 11.9588 Ops/s $\color{#d91a1a}-0.36\%$
test_values[vec_td_lambda_return_estimate-True-False] 1.3264ms 1.1099ms 900.9758 Ops/s 912.9208 Ops/s $\color{#d91a1a}-1.31\%$
test_gae_speed[generalized_advantage_estimate-False-1-512] 22.2324ms 21.7815ms 45.9105 Ops/s 47.0632 Ops/s $\color{#d91a1a}-2.45\%$
test_gae_speed[vec_generalized_advantage_estimate-True-1-512] 1.0747ms 0.7844ms 1.2748 KOps/s 1.2934 KOps/s $\color{#d91a1a}-1.44\%$
test_gae_speed[vec_generalized_advantage_estimate-False-1-512] 0.7579ms 0.7015ms 1.4254 KOps/s 1.4036 KOps/s $\color{#35bf28}+1.56\%$
test_gae_speed[vec_generalized_advantage_estimate-True-32-512] 1.5917ms 1.5170ms 659.2168 Ops/s 666.4424 Ops/s $\color{#d91a1a}-1.08\%$
test_gae_speed[vec_generalized_advantage_estimate-False-32-512] 0.7702ms 0.7191ms 1.3906 KOps/s 1.4302 KOps/s $\color{#d91a1a}-2.76\%$
test_dqn_speed[False-None] 2.0225ms 1.5663ms 638.4276 Ops/s 640.7204 Ops/s $\color{#d91a1a}-0.36\%$
test_dqn_speed[False-backward] 2.3123ms 2.2255ms 449.3409 Ops/s 455.7208 Ops/s $\color{#d91a1a}-1.40\%$
test_dqn_speed[True-None] 1.0585ms 0.5732ms 1.7447 KOps/s 1.8226 KOps/s $\color{#d91a1a}-4.28\%$
test_dqn_speed[True-backward] 1.1369ms 1.0881ms 918.9994 Ops/s 889.2959 Ops/s $\color{#35bf28}+3.34\%$
test_dqn_speed[reduce-overhead-None] 0.6700ms 0.5739ms 1.7423 KOps/s 1.7000 KOps/s $\color{#35bf28}+2.49\%$
test_ddpg_speed[False-None] 3.6484ms 2.9865ms 334.8368 Ops/s 336.8814 Ops/s $\color{#d91a1a}-0.61\%$
test_ddpg_speed[False-backward] 4.5944ms 4.2813ms 233.5739 Ops/s 226.8627 Ops/s $\color{#35bf28}+2.96\%$
test_ddpg_speed[True-None] 1.3979ms 1.3143ms 760.8437 Ops/s 767.7349 Ops/s $\color{#d91a1a}-0.90\%$
test_ddpg_speed[True-backward] 2.5336ms 2.4154ms 414.0159 Ops/s 421.1570 Ops/s $\color{#d91a1a}-1.70\%$
test_ddpg_speed[reduce-overhead-None] 1.4541ms 1.3522ms 739.5223 Ops/s 748.1429 Ops/s $\color{#d91a1a}-1.15\%$
test_sac_speed[False-None] 8.9843ms 8.5365ms 117.1446 Ops/s 117.3656 Ops/s $\color{#d91a1a}-0.19\%$
test_sac_speed[False-backward] 12.0419ms 11.5430ms 86.6327 Ops/s 84.9420 Ops/s $\color{#35bf28}+1.99\%$
test_sac_speed[True-None] 2.1611ms 1.8211ms 549.1283 Ops/s 554.4245 Ops/s $\color{#d91a1a}-0.96\%$
test_sac_speed[True-backward] 3.5245ms 3.4581ms 289.1764 Ops/s 279.1938 Ops/s $\color{#35bf28}+3.58\%$
test_sac_speed[reduce-overhead-None] 18.9077ms 10.7406ms 93.1050 Ops/s 93.8383 Ops/s $\color{#d91a1a}-0.78\%$
test_redq_deprec_speed[False-None] 10.2597ms 9.6004ms 104.1624 Ops/s 105.4119 Ops/s $\color{#d91a1a}-1.19\%$
test_redq_deprec_speed[False-backward] 13.2375ms 12.7300ms 78.5545 Ops/s 77.6617 Ops/s $\color{#35bf28}+1.15\%$
test_redq_deprec_speed[True-None] 2.6734ms 2.5703ms 389.0549 Ops/s 391.2554 Ops/s $\color{#d91a1a}-0.56\%$
test_redq_deprec_speed[True-backward] 4.3364ms 4.1966ms 238.2904 Ops/s 243.0146 Ops/s $\color{#d91a1a}-1.94\%$
test_redq_deprec_speed[reduce-overhead-None] 15.6841ms 9.6290ms 103.8532 Ops/s 105.5690 Ops/s $\color{#d91a1a}-1.63\%$
test_td3_speed[False-None] 8.8346ms 8.4656ms 118.1248 Ops/s 117.6089 Ops/s $\color{#35bf28}+0.44\%$
test_td3_speed[False-backward] 11.6279ms 10.9119ms 91.6432 Ops/s 92.5118 Ops/s $\color{#d91a1a}-0.94\%$
test_td3_speed[True-None] 1.7150ms 1.6442ms 608.2041 Ops/s 621.8021 Ops/s $\color{#d91a1a}-2.19\%$
test_td3_speed[True-backward] 3.1702ms 3.1079ms 321.7616 Ops/s 323.0305 Ops/s $\color{#d91a1a}-0.39\%$
test_td3_speed[reduce-overhead-None] 80.4918ms 24.0528ms 41.5752 Ops/s 43.1723 Ops/s $\color{#d91a1a}-3.70\%$
test_cql_speed[False-None] 18.6702ms 17.9424ms 55.7338 Ops/s 57.0931 Ops/s $\color{#d91a1a}-2.38\%$
test_cql_speed[False-backward] 23.7047ms 23.2389ms 43.0313 Ops/s 43.8002 Ops/s $\color{#d91a1a}-1.76\%$
test_cql_speed[True-None] 3.3306ms 3.2525ms 307.4538 Ops/s 310.8470 Ops/s $\color{#d91a1a}-1.09\%$
test_cql_speed[True-backward] 5.9367ms 5.5183ms 181.2167 Ops/s 181.4506 Ops/s $\color{#d91a1a}-0.13\%$
test_cql_speed[reduce-overhead-None] 18.8211ms 11.7649ms 84.9989 Ops/s 86.0599 Ops/s $\color{#d91a1a}-1.23\%$
test_a2c_speed[False-None] 4.2909ms 3.3603ms 297.5937 Ops/s 304.8047 Ops/s $\color{#d91a1a}-2.37\%$
test_a2c_speed[False-backward] 7.0399ms 6.6219ms 151.0141 Ops/s 154.4998 Ops/s $\color{#d91a1a}-2.26\%$
test_a2c_speed[True-None] 1.8325ms 1.3497ms 740.9166 Ops/s 749.7705 Ops/s $\color{#d91a1a}-1.18\%$
test_a2c_speed[True-backward] 3.2184ms 3.1363ms 318.8507 Ops/s 323.9000 Ops/s $\color{#d91a1a}-1.56\%$
test_a2c_speed[reduce-overhead-None] 1.4067ms 0.9887ms 1.0114 KOps/s 1.0227 KOps/s $\color{#d91a1a}-1.10\%$
test_ppo_speed[False-None] 4.3692ms 3.9944ms 250.3475 Ops/s 256.7391 Ops/s $\color{#d91a1a}-2.49\%$
test_ppo_speed[False-backward] 7.8314ms 7.3941ms 135.2429 Ops/s 136.1851 Ops/s $\color{#d91a1a}-0.69\%$
test_ppo_speed[True-None] 1.6284ms 1.4263ms 701.1060 Ops/s 709.1185 Ops/s $\color{#d91a1a}-1.13\%$
test_ppo_speed[True-backward] 3.3075ms 3.2608ms 306.6707 Ops/s 327.0132 Ops/s $\textbf{\color{#d91a1a}-6.22\%}$
test_ppo_speed[reduce-overhead-None] 1.1090ms 1.0335ms 967.5838 Ops/s 933.3878 Ops/s $\color{#35bf28}+3.66\%$
test_reinforce_speed[False-None] 2.4412ms 2.3482ms 425.8623 Ops/s 431.0415 Ops/s $\color{#d91a1a}-1.20\%$
test_reinforce_speed[False-backward] 3.9558ms 3.5141ms 284.5644 Ops/s 289.8118 Ops/s $\color{#d91a1a}-1.81\%$
test_reinforce_speed[True-None] 1.3578ms 1.2664ms 789.6590 Ops/s 797.3933 Ops/s $\color{#d91a1a}-0.97\%$
test_reinforce_speed[True-backward] 3.1271ms 3.0794ms 324.7340 Ops/s 325.8129 Ops/s $\color{#d91a1a}-0.33\%$
test_reinforce_speed[reduce-overhead-None] 0.4676s 10.1178ms 98.8353 Ops/s 97.4161 Ops/s $\color{#35bf28}+1.46\%$
test_iql_speed[False-None] 10.1575ms 9.7047ms 103.0431 Ops/s 103.5774 Ops/s $\color{#d91a1a}-0.52\%$
test_iql_speed[False-backward] 14.2866ms 13.8114ms 72.4040 Ops/s 72.9256 Ops/s $\color{#d91a1a}-0.72\%$
test_iql_speed[True-None] 2.2988ms 2.1961ms 455.3485 Ops/s 453.8979 Ops/s $\color{#35bf28}+0.32\%$
test_iql_speed[True-backward] 5.1795ms 4.8793ms 204.9489 Ops/s 206.8884 Ops/s $\color{#d91a1a}-0.94\%$
test_iql_speed[reduce-overhead-None] 17.1818ms 10.1425ms 98.5951 Ops/s 75.7660 Ops/s $\textbf{\color{#35bf28}+30.13\%}$
test_rb_sample[TensorDictReplayBuffer-ListStorage-RandomSampler-4000] 6.4583ms 5.9738ms 167.3971 Ops/s 166.0390 Ops/s $\color{#35bf28}+0.82\%$
test_rb_sample[TensorDictReplayBuffer-LazyMemmapStorage-RandomSampler-10000] 0.8896ms 0.2923ms 3.4214 KOps/s 3.4521 KOps/s $\color{#d91a1a}-0.89\%$
test_rb_sample[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-10000] 0.5598ms 0.2966ms 3.3719 KOps/s 3.6583 KOps/s $\textbf{\color{#d91a1a}-7.83\%}$
test_rb_sample[TensorDictReplayBuffer-ListStorage-SamplerWithoutReplacement-4000] 6.1964ms 5.8204ms 171.8105 Ops/s 170.2316 Ops/s $\color{#35bf28}+0.93\%$
test_rb_sample[TensorDictReplayBuffer-LazyMemmapStorage-SamplerWithoutReplacement-10000] 2.0512ms 0.3174ms 3.1509 KOps/s 3.5038 KOps/s $\textbf{\color{#d91a1a}-10.07\%}$
test_rb_sample[TensorDictReplayBuffer-LazyTensorStorage-SamplerWithoutReplacement-10000] 0.6433ms 0.3107ms 3.2187 KOps/s 3.6221 KOps/s $\textbf{\color{#d91a1a}-11.14\%}$
test_rb_sample[TensorDictReplayBuffer-LazyMemmapStorage-sampler6-10000] 1.8302ms 1.4887ms 671.7466 Ops/s 768.3889 Ops/s $\textbf{\color{#d91a1a}-12.58\%}$
test_rb_sample[TensorDictReplayBuffer-LazyTensorStorage-sampler7-10000] 1.6017ms 1.3586ms 736.0666 Ops/s 786.1774 Ops/s $\textbf{\color{#d91a1a}-6.37\%}$
test_rb_sample[TensorDictPrioritizedReplayBuffer-ListStorage-None-4000] 6.0885ms 5.9956ms 166.7881 Ops/s 165.2669 Ops/s $\color{#35bf28}+0.92\%$
test_rb_sample[TensorDictPrioritizedReplayBuffer-LazyMemmapStorage-None-10000] 2.2822ms 0.5536ms 1.8062 KOps/s 2.1515 KOps/s $\textbf{\color{#d91a1a}-16.05\%}$
test_rb_sample[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-10000] 0.7359ms 0.5066ms 1.9739 KOps/s 2.1935 KOps/s $\textbf{\color{#d91a1a}-10.01\%}$
test_rb_iterate[TensorDictReplayBuffer-ListStorage-RandomSampler-4000] 6.0547ms 5.8318ms 171.4725 Ops/s 168.9109 Ops/s $\color{#35bf28}+1.52\%$
test_rb_iterate[TensorDictReplayBuffer-LazyMemmapStorage-RandomSampler-10000] 0.8589ms 0.3772ms 2.6511 KOps/s 2.8591 KOps/s $\textbf{\color{#d91a1a}-7.28\%}$
test_rb_iterate[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-10000] 0.6226ms 0.3644ms 2.7440 KOps/s 2.6841 KOps/s $\color{#35bf28}+2.23\%$
test_rb_iterate[TensorDictReplayBuffer-ListStorage-SamplerWithoutReplacement-4000] 6.0420ms 5.7708ms 173.2865 Ops/s 170.1449 Ops/s $\color{#35bf28}+1.85\%$
test_rb_iterate[TensorDictReplayBuffer-LazyMemmapStorage-SamplerWithoutReplacement-10000] 1.5771ms 0.2883ms 3.4685 KOps/s 3.4860 KOps/s $\color{#d91a1a}-0.50\%$
test_rb_iterate[TensorDictReplayBuffer-LazyTensorStorage-SamplerWithoutReplacement-10000] 0.5768ms 0.3482ms 2.8718 KOps/s 3.7223 KOps/s $\textbf{\color{#d91a1a}-22.85\%}$
test_rb_iterate[TensorDictPrioritizedReplayBuffer-ListStorage-None-4000] 6.2021ms 5.9836ms 167.1239 Ops/s 166.1120 Ops/s $\color{#35bf28}+0.61\%$
test_rb_iterate[TensorDictPrioritizedReplayBuffer-LazyMemmapStorage-None-10000] 1.7601ms 0.5106ms 1.9586 KOps/s 2.1704 KOps/s $\textbf{\color{#d91a1a}-9.76\%}$
test_rb_iterate[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-10000] 0.7682ms 0.5023ms 1.9909 KOps/s 2.2034 KOps/s $\textbf{\color{#d91a1a}-9.64\%}$
test_rb_populate[TensorDictReplayBuffer-ListStorage-RandomSampler-400] 0.6103s 17.3642ms 57.5897 Ops/s 50.6984 Ops/s $\textbf{\color{#35bf28}+13.59\%}$
test_rb_populate[TensorDictReplayBuffer-LazyMemmapStorage-RandomSampler-400] 32.2410ms 2.4181ms 413.5497 Ops/s 491.3184 Ops/s $\textbf{\color{#d91a1a}-15.83\%}$
test_rb_populate[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-400] 10.1338ms 1.3107ms 762.9572 Ops/s 1.0406 KOps/s $\textbf{\color{#d91a1a}-26.68\%}$
test_rb_populate[TensorDictReplayBuffer-ListStorage-SamplerWithoutReplacement-400] 7.5669ms 5.2939ms 188.8971 Ops/s 191.4335 Ops/s $\color{#d91a1a}-1.32\%$
test_rb_populate[TensorDictReplayBuffer-LazyMemmapStorage-SamplerWithoutReplacement-400] 4.0374ms 1.8089ms 552.8273 Ops/s 498.8845 Ops/s $\textbf{\color{#35bf28}+10.81\%}$
test_rb_populate[TensorDictReplayBuffer-LazyTensorStorage-SamplerWithoutReplacement-400] 8.9821ms 1.3044ms 766.6576 Ops/s 751.7478 Ops/s $\color{#35bf28}+1.98\%$
test_rb_populate[TensorDictPrioritizedReplayBuffer-ListStorage-None-400] 0.5634s 16.6828ms 59.9419 Ops/s 183.3766 Ops/s $\textbf{\color{#d91a1a}-67.31\%}$
test_rb_populate[TensorDictPrioritizedReplayBuffer-LazyMemmapStorage-None-400] 11.6269ms 2.1753ms 459.6972 Ops/s 61.1396 Ops/s $\textbf{\color{#35bf28}+651.88\%}$
test_rb_populate[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-400] 1.2347ms 1.0768ms 928.6998 Ops/s 840.4383 Ops/s $\textbf{\color{#35bf28}+10.50\%}$
test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-10000-10000-100-True] 39.7468ms 36.9444ms 27.0677 Ops/s 26.7994 Ops/s $\color{#35bf28}+1.00\%$
test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-10000-10000-100-False] 20.3431ms 18.6929ms 53.4962 Ops/s 53.5792 Ops/s $\color{#d91a1a}-0.15\%$
test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-100000-10000-100-True] 41.0633ms 37.8010ms 26.4543 Ops/s 25.4597 Ops/s $\color{#35bf28}+3.91\%$
test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-100000-10000-100-False] 20.7049ms 18.7708ms 53.2742 Ops/s 52.5604 Ops/s $\color{#35bf28}+1.36\%$
test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-1000000-10000-100-True] 41.0404ms 39.5088ms 25.3108 Ops/s 24.9039 Ops/s $\color{#35bf28}+1.63\%$
test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-1000000-10000-100-False] 21.3045ms 20.1714ms 49.5751 Ops/s 48.8105 Ops/s $\color{#35bf28}+1.57\%$

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

CI Has to do with CI setup (e.g. wheels & builds, tests...) CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@vmoens