Skip to content

[Feature] Add LIBERO environment adapter - #3877

Merged
vmoens merged 11 commits into
gh/vmoens/289/basefrom
gh/vmoens/289/head
Jun 18, 2026
Merged

[Feature] Add LIBERO environment adapter#3877
vmoens merged 11 commits into
gh/vmoens/289/basefrom
gh/vmoens/289/head

Conversation

@vmoens

@vmoens vmoens commented Jun 17, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

@vmoens
Update
063fae9 
[ghstack-poisoned]
@pytorch-bot

pytorch-bot Bot commented Jun 17, 2026 •
edited
Loading

Copy link
Copy Markdown

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/rl/3877

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

⏳ No Failures, 52 Pending

As of commit 24923f7 with merge base 6364a19 (image):
💚 Looks good so far! There are no failures yet. 💚

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

This was referenced Jun 17, 2026
Merged
Merged
@github-actions github-actions Bot added the Documentation Improvements or additions to documentation label Jun 17, 2026
@vmoens vmoens mentioned this pull request Jun 17, 2026
Merged
@github-actions github-actions Bot added Benchmarks rl/benchmark changes Environments Adds or modifies an environment wrapper labels Jun 17, 2026
@vmoens vmoens mentioned this pull request Jun 17, 2026
Merged
@github-actions github-actions Bot added the Feature New feature label Jun 17, 2026
This was referenced Jun 17, 2026
Merged
Merged
@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 Jun 17, 2026
vmoens added 2 commits June 17, 2026 17:20
@vmoens
Update
f0f5b70 
[ghstack-poisoned]
@vmoens
Update
1804748 
[ghstack-poisoned]
@github-actions github-actions Bot added the CI Has to do with CI setup (e.g. wheels & builds, tests...) label Jun 17, 2026
@github-actions

github-actions Bot commented Jun 17, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

Benchmark Results: PR 24923f70 vs main 4d08f22e

Benchmark run: https://github.com/pytorch/rl/actions/runs/27760860262

Higher ops/sec is better. Tables are sorted by largest absolute change.

CPU

Compared 192 benchmarks. Regressions over 5%: 13. Improvements over 5%: 13.

Benchmark main ops PR ops Change
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictReplayBuffer-ListStorage-RandomSampler-400] 37.71 198.25 +425.71%
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictReplayBuffer-ListStorage-SamplerWithoutReplacement-400] 196.01 54.59 -72.15%
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictReplayBuffer-LazyTensorStorage-SamplerWithoutReplacement-10000] 2,505 3,705 +47.90%
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictReplayBuffer-LazyMemmapStorage-SamplerWithoutReplacement-10000] 2,509 3,439 +37.07%
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-10000] 2,692 3,575 +32.78%
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictPrioritizedReplayBuffer-LazyMemmapStorage-None-10000] 1,777 2,265 +27.45%
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-10000] 1,841 2,330 +26.52%
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-10000] 3,647 2,726 -25.25%
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictReplayBuffer-LazyMemmapStorage-RandomSampler-10000] 3,391 2,610 -23.02%
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictReplayBuffer-LazyTensorStorage-sampler7-10000] 706.36 843.81 +19.46%
benchmarks/test_envs_benchmark.py::test_cat_frames_functional[4-same] 23.60 27.91 +18.26%
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-400] 906.96 758.29 -16.39%
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictReplayBuffer-LazyMemmapStorage-sampler6-10000] 688.24 791.59 +15.02%
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictReplayBuffer-LazyTensorStorage-SamplerWithoutReplacement-10000] 3,279 2,828 -13.74%
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-10000] 2,282 1,995 -12.58%
benchmarks/test_objectives_benchmarks.py::test_cql_speed[True-backward] 53.79 60.04 +11.63%
benchmarks/test_envs_benchmark.py::test_cat_frames_functional[16-same] 22.21 19.66 -11.48%
benchmarks/test_objectives_benchmarks.py::test_dqn_speed[True-backward] 892.04 989.89 +10.97%
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictReplayBuffer-LazyMemmapStorage-SamplerWithoutReplacement-10000] 2,989 2,664 -10.88%
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictPrioritizedReplayBuffer-LazyMemmapStorage-None-10000] 2,114 1,900 -10.12%
benchmarks/test_objectives_benchmarks.py::test_a2c_speed[True-backward] 112.28 120.97 +7.73%
benchmarks/test_objectives_benchmarks.py::test_ppo_speed[True-None] 287.01 268.14 -6.58%
benchmarks/test_objectives_benchmarks.py::test_a2c_speed[reduce-overhead-None] 314.66 294.14 -6.52%
benchmarks/test_objectives_benchmarks.py::test_redq_deprec_speed[True-None] 268.94 286.22 +6.43%
benchmarks/test_objectives_benchmarks.py::test_values[td0_return_estimate-False-False] 7,958 7,471 -6.11%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-True-False-False-False] 65,710 62,358 -5.10%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-False-False-False-False] 55,467 52,757 -4.89%
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictReplayBuffer-LazyMemmapStorage-SamplerWithoutReplacement-400] 551.59 524.93 -4.83%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-False-False-True-False] 27,748 26,423 -4.78%
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_collector_stack_then_write[100-img_shape1-atari] 264.89 277.12 +4.62%
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictReplayBuffer-LazyMemmapStorage-RandomSampler-400] 515.07 538.52 +4.55%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-True-True-False-False] 57,792 55,225 -4.44%
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_storage_write_contiguous[50-img_shape0-small] 7,294 6,971 -4.44%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-False-False-False-True] 34,389 32,888 -4.37%
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_storage_write_contiguous[100-img_shape1-atari] 5,237 5,020 -4.15%
benchmarks/test_compressed_storage_benchmark.py::TestCompressedStorageBenchmark::test_tensor_to_bytestream_speed[safetensors] 22,900 23,795 +3.91%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-True-True-False-False] 77,417 74,402 -3.89%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-True-True-False-True] 33,064 31,781 -3.88%
benchmarks/test_objectives_benchmarks.py::test_gae_speed[vec_generalized_advantage_estimate-True-1-512] 679.27 652.96 -3.87%
benchmarks/test_compressed_storage_benchmark.py::TestCompressedStorageBenchmark::test_tensor_to_bytestream_speed[torch.save] 7,202 6,936 -3.70%
benchmarks/test_objectives_benchmarks.py::test_ppo_speed[False-None] 167.83 161.75 -3.62%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-False-False-True-False] 31,973 30,847 -3.52%
benchmarks/test_objectives_benchmarks.py::test_cql_speed[reduce-overhead-None] 83.49 86.35 +3.43%
benchmarks/test_objectives_benchmarks.py::test_redq_deprec_speed[True-backward] 138.34 143.04 +3.40%
benchmarks/test_envs_benchmark.py::test_parallel 0.9810 0.9481 -3.36%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-True-True-True-True] 23,836 23,038 -3.35%
benchmarks/test_compressed_storage_benchmark.py::TestCompressedStorageBenchmark::test_tensor_to_bytestream_speed[pickle] 12,522 12,104 -3.35%
benchmarks/test_objectives_benchmarks.py::test_gae_speed[vec_generalized_advantage_estimate-False-1-512] 2,243 2,169 -3.30%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-False-True-False-True] 38,204 36,952 -3.28%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-False-False-False-True] 28,192 27,302 -3.16%
benchmarks/test_objectives_benchmarks.py::test_a2c_speed[True-None] 283.08 291.93 +3.12%
benchmarks/test_objectives_benchmarks.py::test_redq_speed[False-backward] 55.72 54.01 -3.05%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-False-False-False-False] 44,168 42,863 -2.96%
benchmarks/test_objectives_benchmarks.py::test_values[td_lambda_return_estimate-True-False] 23.70 24.39 +2.91%
benchmarks/test_objectives_benchmarks.py::test_td3_speed[True-backward] 277.64 285.64 +2.88%
benchmarks/test_objectives_benchmarks.py::test_redq_speed[reduce-overhead-None] 223.49 229.89 +2.87%
benchmarks/test_envs_benchmark.py::test_transformed 0.8879 0.9129 +2.81%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-True-False-True-False] 38,398 37,328 -2.79%
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictReplayBuffer-LazyMemmapStorage-RandomSampler-10000] 2,929 2,848 -2.78%
benchmarks/test_objectives_benchmarks.py::test_sac_speed[True-backward] 244.06 250.61 +2.68%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-False-True-False-True] 29,880 29,090 -2.65%
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-400] 1,102 1,074 -2.57%
benchmarks/test_objectives_benchmarks.py::test_ppo_speed[True-backward] 114.52 117.34 +2.46%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-True-False-True-False] 32,149 31,369 -2.43%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-True-True-False-True] 41,600 40,601 -2.40%
benchmarks/test_objectives_benchmarks.py::test_iql_speed[reduce-overhead-None] 116.01 118.79 +2.40%
benchmarks/test_objectives_benchmarks.py::test_reinforce_speed[False-None] 211.22 216.22 +2.36%
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_collector_lazystack_then_write[200-img_shape3-large_batch] 308.16 315.18 +2.28%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-False-True-True-True] 20,584 20,157 -2.07%
benchmarks/test_objectives_benchmarks.py::test_sac_speed[False-None] 123.73 121.18 -2.06%
benchmarks/test_objectives_benchmarks.py::test_ppo_speed[reduce-overhead-None] 266.52 271.91 +2.02%
benchmarks/test_objectives_benchmarks.py::test_sac_speed[True-None] 469.81 479.12 +1.98%
benchmarks/test_replaybuffer_benchmark.py::test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-10000-10000-100-True] 25.34 25.83 +1.95%
benchmarks/test_objectives_benchmarks.py::test_values[generalized_advantage_estimate-True-True] 100.05 98.13 -1.92%
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictPrioritizedReplayBuffer-ListStorage-None-4000] 165.79 162.61 -1.91%
benchmarks/test_non_tensor_env_benchmark.py::test_non_tensor_env_rollout_speed[1000-parallel-buffers-False] 0.5957 0.6068 +1.87%
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_storage_write_lazystack[100-img_shape1-atari] 693.35 705.97 +1.82%
benchmarks/test_objectives_benchmarks.py::test_dqn_speed[reduce-overhead-None] 1,770 1,802 +1.82%
benchmarks/test_non_tensor_env_benchmark.py::test_non_tensor_env_rollout_speed[1000-parallel-buffers-True] 0.5274 0.5179 -1.81%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-True-False-False-True] 36,977 36,318 -1.78%
benchmarks/test_objectives_benchmarks.py::test_reinforce_speed[False-backward] 130.92 133.24 +1.77%
benchmarks/test_replaybuffer_benchmark.py::test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-100000-10000-100-False] 52.84 53.76 +1.74%
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_collector_lazystack_then_write[100-img_shape1-atari] 638.74 649.53 +1.69%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-True-False-False-False] 49,205 48,402 -1.63%
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_collector_lazystack_then_write[100-img_shape2-large_img] 389.47 395.80 +1.63%
benchmarks/test_rnn_reset_backends_benchmark.py::test_rnn_rollout_with_intermediate_resets[b256-t128-i32-h512-scan-False-0-lstm] 2.0566 2.0238 -1.59%
benchmarks/test_objectives_benchmarks.py::test_ddpg_speed[False-backward] 241.85 245.55 +1.53%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-False-False-True-True] 19,549 19,253 -1.51%
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictReplayBuffer-ListStorage-RandomSampler-4000] 161.42 158.98 -1.51%
benchmarks/test_objectives_benchmarks.py::test_gae_speed[vec_generalized_advantage_estimate-True-32-512] 29.16 28.72 -1.50%
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictReplayBuffer-ListStorage-SamplerWithoutReplacement-4000] 167.45 164.98 -1.48%
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictPrioritizedReplayBuffer-ListStorage-None-400] 189.46 186.73 -1.44%
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_collector_stack_then_write[100-img_shape2-large_img] 167.50 169.90 +1.43%
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictReplayBuffer-LazyTensorStorage-SamplerWithoutReplacement-400] 1,067 1,052 -1.42%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-True-True-True-False] 42,335 41,738 -1.41%
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_storage_write_lazystack[200-img_shape3-large_batch] 331.76 336.41 +1.40%
benchmarks/test_rnn_reset_backends_benchmark.py::test_rnn_rollout_with_intermediate_resets[b256-t128-i32-h512-scan-False-0-gru] 3.0808 3.0376 -1.40%
benchmarks/test_objectives_benchmarks.py::test_cql_speed[True-None] 85.60 86.75 +1.34%
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_collector_stack_then_write[200-img_shape3-large_batch] 133.62 135.35 +1.29%
benchmarks/test_replaybuffer_benchmark.py::test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-10000-10000-100-False] 54.05 54.75 +1.29%
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictReplayBuffer-ListStorage-RandomSampler-4000] 168.39 166.22 -1.29%
benchmarks/test_collectors_benchmark.py::test_sync 17.04 16.82 -1.28%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-False-True-False-False] 64,094 63,276 -1.28%
benchmarks/test_non_tensor_env_benchmark.py::test_non_tensor_env_rollout_speed[1000-single-True] 1.3218 1.3049 -1.27%
benchmarks/test_objectives_benchmarks.py::test_ddpg_speed[False-None] 344.89 349.13 +1.23%
benchmarks/test_objectives_benchmarks.py::test_cql_speed[False-backward] 28.13 27.79 -1.21%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-True-False-True-True] 19,526 19,292 -1.20%
benchmarks/test_collectors_benchmark.py::test_sync_preempt 16.61 16.81 +1.20%
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_storage_write_contiguous[200-img_shape3-large_batch] 768.01 758.93 -1.18%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-True-True-True-True] 21,079 20,835 -1.16%
benchmarks/test_non_tensor_env_benchmark.py::test_non_tensor_env_rollout_speed[1000-parallel-no-buffers-True] 0.2155 0.2132 -1.10%
benchmarks/test_objectives_benchmarks.py::test_iql_speed[True-None] 117.74 119.02 +1.09%
benchmarks/test_objectives_benchmarks.py::test_ddpg_speed[reduce-overhead-None] 713.61 705.84 -1.09%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-True-False-False-True] 29,688 29,371 -1.07%
benchmarks/test_non_tensor_env_benchmark.py::test_non_tensor_env_rollout_speed[1000-single-False] 1.5996 1.5827 -1.05%
benchmarks/test_non_tensor_env_benchmark.py::test_non_tensor_env_rollout_speed[1000-serial-no-buffers-True] 0.5990 0.5928 -1.03%
benchmarks/test_replaybuffer_benchmark.py::TestPrioritizedReplayBufferBenchmark::test_sampler_sample_scale[1000000-cpu] 100.00 98.97 -1.03%
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_storage_write_lazystack[100-img_shape2-large_img] 409.44 413.61 +1.02%
benchmarks/test_non_tensor_env_benchmark.py::test_non_tensor_env_rollout_speed[1000-serial-buffers-True] 0.5189 0.5136 -1.01%
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictPrioritizedReplayBuffer-ListStorage-None-4000] 164.01 162.43 -0.96%
... ... ... Showing 120 of 192 comparisons, sorted by absolute change.

GPU

Compared 202 benchmarks. Regressions over 5%: 13. Improvements over 5%: 12.

Benchmark main ops PR ops Change
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictReplayBuffer-ListStorage-SamplerWithoutReplacement-400] 43.64 192.66 +341.43%
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictReplayBuffer-ListStorage-RandomSampler-400] 191.37 38.66 -79.80%
benchmarks/test_objectives_benchmarks.py::test_iql_speed[False-None] 96.91 57.88 -40.28%
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictReplayBuffer-LazyMemmapStorage-RandomSampler-10000] 2,476 3,373 +36.24%
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-10000] 2,854 3,596 +26.01%
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictReplayBuffer-LazyMemmapStorage-RandomSampler-10000] 2,703 3,387 +25.33%
benchmarks/test_objectives_benchmarks.py::test_reinforce_speed[reduce-overhead-None] 96.01 120.29 +25.29%
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-10000] 2,336 1,907 -18.35%
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-10000] 3,085 3,620 +17.33%
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictPrioritizedReplayBuffer-LazyMemmapStorage-None-10000] 2,239 1,873 -16.35%
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictReplayBuffer-LazyTensorStorage-SamplerWithoutReplacement-400] 999.90 851.54 -14.84%
benchmarks/test_envs_benchmark.py::test_cat_frames_functional[16-same] 5.3692 4.6040 -14.25%
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictReplayBuffer-LazyMemmapStorage-SamplerWithoutReplacement-10000] 3,452 3,077 -10.88%
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictPrioritizedReplayBuffer-LazyMemmapStorage-None-10000] 2,217 1,987 -10.38%
benchmarks/test_objectives_benchmarks.py::test_dqn_speed[True-backward] 963.61 864.56 -10.28%
benchmarks/test_collectors_benchmark.py::test_single_with_rb_pixels 5.2146 4.6933 -10.00%
benchmarks/test_objectives_benchmarks.py::test_td3_speed[True-backward] 360.06 389.11 +8.07%
benchmarks/test_collectors_benchmark.py::test_sync_preempt 9.5762 10.29 +7.46%
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_storage_write_contiguous[100-img_shape2-large_img] 592.17 549.03 -7.29%
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictReplayBuffer-LazyMemmapStorage-SamplerWithoutReplacement-400] 528.05 491.80 -6.86%
benchmarks/test_objectives_benchmarks.py::test_reinforce_speed[True-backward] 343.86 365.68 +6.35%
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictReplayBuffer-LazyMemmapStorage-sampler6-10000] 775.68 727.40 -6.22%
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_storage_write_contiguous[100-img_shape1-atari] 3,925 4,152 +5.78%
benchmarks/test_objectives_benchmarks.py::test_a2c_speed[True-backward] 342.40 361.92 +5.70%
benchmarks/test_objectives_benchmarks.py::test_sac_speed[True-backward] 309.55 325.79 +5.25%
benchmarks/test_objectives_benchmarks.py::test_dqn_speed[reduce-overhead-None] 1,814 1,900 +4.73%
benchmarks/test_objectives_benchmarks.py::test_ddpg_speed[True-backward] 464.60 443.33 -4.58%
benchmarks/test_objectives_benchmarks.py::test_cql_speed[True-backward] 213.25 222.57 +4.37%
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictPrioritizedReplayBuffer-ListStorage-None-400] 181.74 188.99 +3.99%
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictReplayBuffer-LazyMemmapStorage-RandomSampler-400] 489.81 509.13 +3.95%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-False-True-False-False] 51,324 49,344 -3.86%
benchmarks/test_envs_benchmark.py::test_simple 1.2293 1.1821 -3.84%
benchmarks/test_compressed_storage_benchmark.py::TestCompressedStorageBenchmark::test_tensor_to_bytestream_speed[untyped_storage] 8.2747 7.9644 -3.75%
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_storage_write_lazystack[50-img_shape0-small] 4,119 4,273 +3.75%
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_storage_write_lazystack[100-img_shape2-large_img] 438.80 422.67 -3.68%
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictReplayBuffer-LazyMemmapStorage-SamplerWithoutReplacement-10000] 3,439 3,313 -3.65%
benchmarks/test_objectives_benchmarks.py::test_a2c_speed[False-backward] 142.39 147.55 +3.62%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-False-False-False-True] 28,924 27,900 -3.54%
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-10000] 2,055 1,982 -3.54%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-False-False-True-True] 18,275 17,650 -3.42%
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_storage_write_contiguous[50-img_shape0-small] 5,853 6,050 +3.37%
benchmarks/test_objectives_benchmarks.py::test_ddpg_speed[False-backward] 235.97 228.07 -3.35%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-True-True-True-False] 34,570 33,417 -3.34%
benchmarks/test_collectors_benchmark.py::test_async 10.73 11.07 +3.21%
benchmarks/test_objectives_benchmarks.py::test_gae_speed[vec_generalized_advantage_estimate-False-1-512] 1,271 1,311 +3.19%
benchmarks/test_replaybuffer_benchmark.py::TestPrioritizedReplayBufferBenchmark::test_sampler_sample_scale[1000000-cuda] 2,219 2,150 -3.14%
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_storage_write_contiguous[200-img_shape3-large_batch] 750.63 727.27 -3.11%
benchmarks/test_objectives_benchmarks.py::test_redq_deprec_speed[False-backward] 69.32 71.47 +3.10%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-False-True-True-True] 18,881 18,303 -3.06%
benchmarks/test_objectives_benchmarks.py::test_ppo_speed[False-backward] 124.45 128.24 +3.04%
benchmarks/test_objectives_benchmarks.py::test_redq_deprec_speed[True-backward] 259.10 266.94 +3.03%
benchmarks/test_objectives_benchmarks.py::test_td3_speed[reduce-overhead-None] 44.14 45.47 +3.01%
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-400] 816.34 791.81 -3.00%
benchmarks/test_objectives_benchmarks.py::test_dqn_speed[True-None] 1,840 1,895 +2.97%
benchmarks/test_objectives_benchmarks.py::test_gae_speed[vec_generalized_advantage_estimate-False-32-512] 1,257 1,292 +2.85%
benchmarks/test_objectives_benchmarks.py::test_ppo_speed[reduce-overhead-None] 791.54 813.99 +2.84%
benchmarks/test_storage_write_benchmark.py::TestCollectorIntegrationBenchmark::test_collector_without_rb_cuda[100-img_shape0-atari] 16.86 17.32 +2.75%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-True-False-False-False] 50,257 48,876 -2.75%
benchmarks/test_envs_benchmark.py::test_parallel 0.5297 0.5442 +2.74%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-False-True-True-False] 29,416 28,622 -2.70%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-False-False-False-False] 55,115 53,657 -2.65%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-False-True-False-False] 63,873 62,188 -2.64%
benchmarks/test_objectives_benchmarks.py::test_ppo_speed[True-None] 664.78 681.98 +2.59%
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_storage_write_lazystack[200-img_shape3-large_batch] 325.52 333.81 +2.55%
benchmarks/test_objectives_benchmarks.py::test_iql_speed[False-backward] 67.22 68.93 +2.53%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-False-False-False-False] 45,441 44,301 -2.51%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-False-False-True-False] 27,207 26,534 -2.47%
benchmarks/test_objectives_benchmarks.py::test_ddpg_speed[reduce-overhead-None] 804.26 823.76 +2.42%
benchmarks/test_rnn_reset_backends_benchmark.py::test_rnn_rollout_with_intermediate_resets[b256-t128-i32-h512-scan-True-0-gru] 48.43 47.27 -2.41%
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_collector_lazystack_then_write[200-img_shape3-large_batch] 306.40 313.72 +2.39%
benchmarks/test_envs_benchmark.py::test_serial 0.4121 0.4219 +2.39%
benchmarks/test_storage_write_benchmark.py::TestCollectorIntegrationBenchmark::test_collector_with_rb_cuda[200-img_shape1-large_batch] 8.0746 8.2674 +2.39%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-False-True-False-True] 37,870 36,974 -2.37%
benchmarks/test_objectives_benchmarks.py::test_cql_speed[reduce-overhead-None] 87.50 89.42 +2.19%
benchmarks/test_objectives_benchmarks.py::test_sac_speed[reduce-overhead-None] 99.40 101.55 +2.16%
benchmarks/test_replaybuffer_benchmark.py::TestPrioritizedReplayBufferBenchmark::test_sample_mixed_devices[1000000-cuda_storage_cpu_sampler] 86.62 88.49 +2.15%
benchmarks/test_objectives_benchmarks.py::test_redq_deprec_speed[False-None] 95.58 97.64 +2.15%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-True-False-True-True] 19,312 19,724 +2.13%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-True-False-True-False] 31,966 31,298 -2.09%
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_collector_lazystack_then_write[50-img_shape0-small] 3,447 3,517 +2.01%
benchmarks/test_objectives_benchmarks.py::test_a2c_speed[reduce-overhead-None] 849.83 866.88 +2.01%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-True-True-False-False] 55,089 56,123 +1.88%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-False-False-True-False] 31,455 30,868 -1.87%
benchmarks/test_storage_write_benchmark.py::TestCollectorIntegrationBenchmark::test_collector_without_rb[200-img_shape1-large_batch] 14.88 15.16 +1.87%
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_collector_stack_then_write[200-img_shape3-large_batch] 132.46 134.90 +1.85%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-False-True-True-True] 20,791 20,411 -1.83%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-False-True-False-True] 30,482 29,945 -1.76%
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_collector_stack_then_write[50-img_shape0-small] 857.18 842.11 -1.76%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-True-True-False-True] 32,066 32,609 +1.69%
benchmarks/test_envs_benchmark.py::test_cat_frames_functional[4-same] 6.4644 6.3568 -1.66%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-True-True-True-True] 20,730 20,387 -1.65%
benchmarks/test_storage_write_benchmark.py::TestCollectorIntegrationBenchmark::test_collector_with_rb_cuda[100-img_shape0-atari] 16.22 16.48 +1.64%
benchmarks/test_compressed_storage_benchmark.py::TestCompressedStorageBenchmark::test_tensor_to_bytestream_speed[safetensors] 23,707 23,319 -1.63%
benchmarks/test_rnn_reset_backends_benchmark.py::test_rnn_rollout_with_intermediate_resets[b256-t128-i32-h512-scan-False-0-lstm] 20.88 20.55 -1.61%
benchmarks/test_objectives_benchmarks.py::test_redq_deprec_speed[reduce-overhead-None] 107.89 109.62 +1.60%
benchmarks/test_compressed_storage_benchmark.py::TestCompressedStorageBenchmark::test_tensor_to_bytestream_speed[numpy] 341,207 335,915 -1.55%
benchmarks/test_objectives_benchmarks.py::test_sac_speed[False-None] 109.28 110.97 +1.54%
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictReplayBuffer-LazyTensorStorage-sampler7-10000] 827.65 815.00 -1.53%
benchmarks/test_envs_benchmark.py::test_transformed 0.7047 0.6942 -1.49%
benchmarks/test_objectives_benchmarks.py::test_sac_speed[True-None] 605.75 597.07 -1.43%
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_storage_write_lazystack[100-img_shape1-atari] 694.18 684.39 -1.41%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-False-True-True-False] 34,195 33,724 -1.38%
benchmarks/test_storage_write_benchmark.py::TestCollectorIntegrationBenchmark::test_collector_with_rb[100-img_shape0-atari] 25.73 26.09 +1.37%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-True-True-False-True] 41,617 42,188 +1.37%
benchmarks/test_objectives_benchmarks.py::test_cql_speed[False-backward] 39.53 40.07 +1.37%
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictPrioritizedReplayBuffer-LazyMemmapStorage-None-400] 500.37 493.58 -1.36%
benchmarks/test_objectives_benchmarks.py::test_td3_speed[False-backward] 82.01 83.11 +1.34%
benchmarks/test_replaybuffer_benchmark.py::test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-10000-10000-100-True] 23.63 23.32 -1.32%
benchmarks/test_objectives_benchmarks.py::test_iql_speed[True-None] 501.13 507.71 +1.31%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-True-False-False-True] 30,441 30,043 -1.31%
benchmarks/test_objectives_benchmarks.py::test_values[td1_return_estimate-False-False] 19.47 19.72 +1.30%
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-True-False-False-False] 63,896 64,726 +1.30%
benchmarks/test_objectives_benchmarks.py::test_values[td_lambda_return_estimate-True-False] 11.89 11.74 -1.27%
benchmarks/test_objectives_benchmarks.py::test_ppo_speed[False-None] 222.28 225.09 +1.26%
benchmarks/test_objectives_benchmarks.py::test_values[generalized_advantage_estimate-True-True] 46.69 47.27 +1.25%
benchmarks/test_storage_write_benchmark.py::TestCollectorIntegrationBenchmark::test_collector_without_rb[100-img_shape0-atari] 29.71 30.07 +1.24%
benchmarks/test_objectives_benchmarks.py::test_ddpg_speed[True-None] 804.21 814.15 +1.24%
benchmarks/test_objectives_benchmarks.py::test_ppo_speed[True-backward] 343.10 347.31 +1.23%
benchmarks/test_storage_write_benchmark.py::TestCollectorIntegrationBenchmark::test_collector_without_rb_cuda[200-img_shape1-large_batch] 8.4442 8.5434 +1.17%
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-400] 988.23 999.82 +1.17%
... ... ... Showing 120 of 202 comparisons, sorted by absolute change.

vmoens added 8 commits June 17, 2026 18:55
@vmoens
Update
5042cbb 
[ghstack-poisoned]
@vmoens
Update
8c7cae9 
[ghstack-poisoned]
@vmoens
Update
1aa2255 
[ghstack-poisoned]
@vmoens
Update
c70b284 
[ghstack-poisoned]
@vmoens
Update
a09aa53 
[ghstack-poisoned]
@vmoens
Update
ddcdcb0 
[ghstack-poisoned]
@vmoens
Update
de83903 
[ghstack-poisoned]
@vmoens
Update
24923f7 
[ghstack-poisoned]
vmoens added a commit that referenced this pull request Jun 18, 2026
@vmoens
[Feature] Add LIBERO environment adapter
ed66a0f 
ghstack-source-id: 5e155e7
Pull-Request: #3877
@vmoens
vmoens merged commit 24923f7 into gh/vmoens/289/base Jun 18, 2026
107 of 109 checks passed
@vmoens
vmoens deleted the gh/vmoens/289/head branch June 18, 2026 13:21
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

Benchmarks rl/benchmark changes 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. Documentation Improvements or additions to documentation Environments Adds or modifies an environment wrapper Feature New feature

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@vmoens