[Refactor] Add structured inference server config objects#3893
Merged
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/rl/3893
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. |
github-actions
Bot
added
Documentation
This was referenced Jun 21, 2026
meta-cla
Bot
added
the
CLA Signed
Contributor
Benchmark Results: PR
|
| Benchmark | main ops | PR ops | Change |
|---|---|---|---|
benchmarks/test_objectives_benchmarks.py::test_values[td0_return_estimate-False-False] |
8,026 | 5,383 | -32.94% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-10000] |
2,760 | 3,634 | +31.64% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictPrioritizedReplayBuffer-LazyMemmapStorage-None-10000] |
1,735 | 2,242 | +29.26% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-10000] |
2,967 | 3,689 | +24.33% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictReplayBuffer-LazyMemmapStorage-RandomSampler-10000] |
2,802 | 3,443 | +22.89% |
benchmarks/test_objectives_benchmarks.py::test_dqn_speed[True-backward] |
831.29 | 995.46 | +19.75% |
benchmarks/test_envs_benchmark.py::test_cat_frames_functional[4-same] |
29.69 | 24.80 | -16.48% |
benchmarks/test_objectives_benchmarks.py::test_ddpg_speed[True-backward] |
360.69 | 417.67 | +15.80% |
benchmarks/test_objectives_benchmarks.py::test_redq_speed[True-None] |
201.00 | 227.83 | +13.35% |
benchmarks/test_envs_benchmark.py::test_cat_frames_functional[16-same] |
20.81 | 23.59 | +13.35% |
benchmarks/test_collectors_benchmark.py::test_single_with_rb |
7.4659 | 6.5226 | -12.64% |
benchmarks/test_objectives_benchmarks.py::test_gae_speed[vec_generalized_advantage_estimate-True-32-512] |
33.09 | 29.20 | -11.75% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictReplayBuffer-LazyMemmapStorage-sampler6-10000] |
729.67 | 812.15 | +11.30% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictReplayBuffer-LazyMemmapStorage-SamplerWithoutReplacement-400] |
550.72 | 493.13 | -10.46% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictReplayBuffer-LazyTensorStorage-sampler7-10000] |
775.35 | 851.95 | +9.88% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictReplayBuffer-LazyMemmapStorage-RandomSampler-10000] |
2,655 | 2,890 | +8.86% |
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_collector_lazystack_then_write[100-img_shape1-atari] |
636.32 | 689.01 | +8.28% |
benchmarks/test_vla_preprocessing_benchmark.py::test_openvla_preprocessing_throughput[torchvision-480-640-64] |
7.0751 | 6.5317 | -7.68% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-10000] |
2,074 | 2,230 | +7.51% |
benchmarks/test_objectives_benchmarks.py::test_redq_deprec_speed[True-backward] |
124.84 | 133.95 | +7.30% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictReplayBuffer-LazyMemmapStorage-SamplerWithoutReplacement-10000] |
2,962 | 3,161 | +6.73% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictReplayBuffer-ListStorage-RandomSampler-400] |
36.34 | 38.71 | +6.52% |
benchmarks/test_collectors_benchmark.py::test_sync_preempt |
15.79 | 16.80 | +6.37% |
benchmarks/test_objectives_benchmarks.py::test_cql_speed[reduce-overhead-None] |
78.85 | 83.86 | +6.35% |
benchmarks/test_objectives_benchmarks.py::test_iql_speed[reduce-overhead-None] |
112.86 | 119.09 | +5.53% |
benchmarks/test_envs_benchmark.py::test_simple |
1.8194 | 1.7278 | -5.04% |
benchmarks/test_objectives_benchmarks.py::test_redq_deprec_speed[reduce-overhead-None] |
261.68 | 274.80 | +5.01% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictPrioritizedReplayBuffer-LazyMemmapStorage-None-10000] |
2,120 | 2,018 | -4.81% |
benchmarks/test_compressed_storage_benchmark.py::TestCompressedStorageBenchmark::test_tensor_to_bytestream_speed[pickle] |
12,413 | 11,828 | -4.71% |
benchmarks/test_objectives_benchmarks.py::test_reinforce_speed[reduce-overhead-None] |
326.17 | 340.93 | +4.53% |
benchmarks/test_objectives_benchmarks.py::test_ppo_speed[True-None] |
253.36 | 264.12 | +4.25% |
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_storage_write_lazystack[100-img_shape1-atari] |
688.93 | 717.50 | +4.15% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictReplayBuffer-LazyTensorStorage-SamplerWithoutReplacement-10000] |
2,960 | 3,082 | +4.10% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-True-False-True-False] |
38,136 | 39,629 | +3.91% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-False-False-True-False] |
27,464 | 28,512 | +3.81% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-False-False-True-True] |
18,679 | 17,974 | -3.77% |
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_storage_write_lazystack[50-img_shape0-small] |
4,305 | 4,464 | +3.69% |
benchmarks/test_objectives_benchmarks.py::test_values[td_lambda_return_estimate-True-False] |
25.30 | 26.19 | +3.50% |
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_storage_write_lazystack[200-img_shape3-large_batch] |
326.21 | 337.52 | +3.47% |
benchmarks/test_non_tensor_env_benchmark.py::test_non_tensor_env_rollout_speed[1000-parallel-no-buffers-False] |
0.2268 | 0.2346 | +3.44% |
benchmarks/test_vla_preprocessing_benchmark.py::test_openvla_preprocessing_throughput[torchvision-256-256-1] |
500.42 | 517.10 | +3.33% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-True-True-True-True] |
23,259 | 24,030 | +3.32% |
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_collector_lazystack_then_write[200-img_shape3-large_batch] |
306.54 | 316.52 | +3.26% |
benchmarks/test_objectives_benchmarks.py::test_iql_speed[True-backward] |
59.41 | 61.22 | +3.06% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-False-False-False-True] |
34,941 | 33,885 | -3.02% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-10000] |
2,383 | 2,313 | -2.92% |
benchmarks/test_objectives_benchmarks.py::test_values[td1_return_estimate-False-False] |
37.78 | 38.88 | +2.91% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-400] |
944.60 | 917.47 | -2.87% |
benchmarks/test_objectives_benchmarks.py::test_redq_speed[False-backward] |
53.77 | 55.29 | +2.83% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-False-False-False-False] |
44,116 | 45,320 | +2.73% |
benchmarks/test_vla_preprocessing_benchmark.py::test_openvla_preprocessing_throughput[torchvision-224-224-64] |
12.16 | 12.48 | +2.63% |
benchmarks/test_rnn_reset_backends_benchmark.py::test_rnn_rollout_with_intermediate_resets[b256-t128-i32-h512-cudnn-True-0-gru] |
1.4820 | 1.4436 | -2.59% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-False-False-False-True] |
28,147 | 28,856 | +2.52% |
benchmarks/test_non_tensor_env_benchmark.py::test_non_tensor_env_rollout_speed[1000-parallel-buffers-True] |
0.5229 | 0.5358 | +2.48% |
benchmarks/test_objectives_benchmarks.py::test_values[generalized_advantage_estimate-True-True] |
100.01 | 102.48 | +2.46% |
benchmarks/test_non_tensor_env_benchmark.py::test_non_tensor_env_rollout_speed[1000-parallel-buffers-False] |
0.6024 | 0.5876 | -2.44% |
benchmarks/test_envs_benchmark.py::test_serial |
0.5797 | 0.5937 | +2.41% |
benchmarks/test_objectives_benchmarks.py::test_redq_deprec_speed[False-None] |
90.67 | 88.50 | -2.40% |
benchmarks/test_objectives_benchmarks.py::test_sac_speed[True-None] |
453.84 | 464.69 | +2.39% |
benchmarks/test_compressed_storage_benchmark.py::TestCompressedStorageBenchmark::test_tensor_to_bytestream_speed[numpy] |
375,462 | 366,550 | -2.37% |
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_storage_write_contiguous[50-img_shape0-small] |
7,165 | 6,998 | -2.33% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-10000-10000-100-False] |
54.11 | 52.85 | -2.32% |
benchmarks/test_envs_benchmark.py::test_cat_frames_functional[4-constant] |
4,297 | 4,198 | -2.31% |
benchmarks/test_objectives_benchmarks.py::test_redq_deprec_speed[True-None] |
275.74 | 269.48 | -2.27% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-True-True-True-False] |
35,052 | 34,259 | -2.26% |
benchmarks/test_replaybuffer_benchmark.py::TestPrioritizedReplayBufferBenchmark::test_sample_mixed_devices[1000000-memmap_cpu_storage_cpu... |
83.47 | 81.61 | -2.23% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictReplayBuffer-ListStorage-RandomSampler-4000] |
165.10 | 168.75 | +2.21% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-False-True-True-True] |
18,322 | 18,722 | +2.19% |
benchmarks/test_rnn_reset_backends_benchmark.py::test_rnn_rollout_with_intermediate_resets[b256-t128-i32-h512-scan-True-0-gru] |
4.2301 | 4.3224 | +2.18% |
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_storage_write_contiguous[200-img_shape3-large_batch] |
754.12 | 770.47 | +2.17% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-10000-10000-100-True] |
25.53 | 24.97 | -2.17% |
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_collector_lazystack_then_write[50-img_shape0-small] |
3,485 | 3,559 | +2.13% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-True-False-True-True] |
21,819 | 22,281 | +2.12% |
benchmarks/test_vla_preprocessing_benchmark.py::test_openvla_preprocessing_throughput[pil-224-224-1] |
269.84 | 275.52 | +2.10% |
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_collector_stack_then_write[100-img_shape2-large_img] |
174.48 | 170.81 | -2.10% |
benchmarks/test_objectives_benchmarks.py::test_iql_speed[True-None] |
117.25 | 119.71 | +2.10% |
benchmarks/test_objectives_benchmarks.py::test_ppo_speed[False-None] |
161.42 | 164.81 | +2.10% |
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_storage_write_contiguous[100-img_shape1-atari] |
4,817 | 4,917 | +2.06% |
benchmarks/test_objectives_benchmarks.py::test_td3_speed[True-backward] |
280.82 | 286.52 | +2.03% |
benchmarks/test_objectives_benchmarks.py::test_redq_speed[reduce-overhead-None] |
222.29 | 226.74 | +2.00% |
benchmarks/test_objectives_benchmarks.py::test_gae_speed[vec_generalized_advantage_estimate-True-1-512] |
659.60 | 672.54 | +1.96% |
benchmarks/test_objectives_benchmarks.py::test_dqn_speed[True-None] |
1,713 | 1,746 | +1.94% |
benchmarks/test_objectives_benchmarks.py::test_sac_speed[reduce-overhead-None] |
457.65 | 466.24 | +1.88% |
benchmarks/test_objectives_benchmarks.py::test_a2c_speed[True-None] |
281.40 | 286.40 | +1.78% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-True-True-True-True] |
20,874 | 20,505 | -1.77% |
benchmarks/test_compressed_storage_benchmark.py::TestCompressedStorageBenchmark::test_tensor_to_bytestream_speed[torch.save] |
7,175 | 7,049 | -1.75% |
benchmarks/test_vla_preprocessing_benchmark.py::test_openvla_preprocessing_throughput[pil-256-256-1] |
195.76 | 192.35 | -1.74% |
benchmarks/test_objectives_benchmarks.py::test_a2c_speed[reduce-overhead-None] |
285.44 | 290.32 | +1.71% |
benchmarks/test_objectives_benchmarks.py::test_gae_speed[vec_generalized_advantage_estimate-False-1-512] |
2,317 | 2,356 | +1.70% |
benchmarks/test_objectives_benchmarks.py::test_ppo_speed[reduce-overhead-None] |
263.17 | 267.61 | +1.69% |
benchmarks/test_compressed_storage_benchmark.py::TestCompressedStorageBenchmark::test_tensor_to_bytestream_speed[untyped_storage] |
8.5177 | 8.6607 | +1.68% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictReplayBuffer-LazyTensorStorage-SamplerWithoutReplacement-400] |
1,115 | 1,097 | -1.66% |
benchmarks/test_objectives_benchmarks.py::test_ddpg_speed[False-None] |
347.61 | 353.38 | +1.66% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictReplayBuffer-ListStorage-SamplerWithoutReplacement-4000] |
166.12 | 168.86 | +1.65% |
benchmarks/test_objectives_benchmarks.py::test_redq_deprec_speed[False-backward] |
63.85 | 62.79 | -1.65% |
benchmarks/test_objectives_benchmarks.py::test_cql_speed[False-backward] |
28.51 | 28.05 | -1.63% |
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_collector_stack_then_write[200-img_shape3-large_batch] |
140.37 | 142.65 | +1.62% |
benchmarks/test_rnn_reset_backends_benchmark.py::test_rnn_rollout_with_intermediate_resets[b256-t128-i32-h512-cudnn-False-0-lstm] |
0.8771 | 0.8629 | -1.62% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-False-False-True-False] |
31,628 | 32,137 | +1.61% |
benchmarks/test_storage_write_benchmark.py::TestCollectorIntegrationBenchmark::test_collector_without_rb[100-img_shape0-atari] |
29.78 | 30.25 | +1.58% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-False-True-True-False] |
34,404 | 34,935 | +1.54% |
benchmarks/test_storage_write_benchmark.py::TestCollectorIntegrationBenchmark::test_collector_with_rb[100-img_shape0-atari] |
20.30 | 20.61 | +1.51% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-True-True-False-False] |
57,624 | 56,758 | -1.50% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-False-False-True-True] |
19,591 | 19,881 | +1.48% |
benchmarks/test_replaybuffer_benchmark.py::TestPrioritizedReplayBufferBenchmark::test_sampler_sample_scale[1000000-cpu] |
98.36 | 96.91 | -1.47% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictPrioritizedReplayBuffer-LazyMemmapStorage-None-400] |
445.12 | 451.47 | +1.43% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-400] |
1,061 | 1,076 | +1.42% |
benchmarks/test_objectives_benchmarks.py::test_values[vec_td1_return_estimate-False-False] |
54.62 | 55.39 | +1.41% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictReplayBuffer-ListStorage-SamplerWithoutReplacement-400] |
193.40 | 196.07 | +1.38% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictReplayBuffer-ListStorage-RandomSampler-4000] |
160.19 | 162.39 | +1.38% |
benchmarks/test_objectives_benchmarks.py::test_cql_speed[True-None] |
83.92 | 85.04 | +1.34% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-100000-10000-100-False] |
53.80 | 53.08 | -1.34% |
benchmarks/test_vla_preprocessing_benchmark.py::test_openvla_preprocessing_throughput[torchvision-256-256-64] |
10.61 | 10.75 | +1.34% |
benchmarks/test_envs_benchmark.py::test_parallel |
0.9824 | 0.9693 | -1.33% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-False-True-False-False] |
49,442 | 50,058 | +1.25% |
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_collector_lazystack_then_write[100-img_shape2-large_img] |
400.24 | 395.42 | -1.20% |
benchmarks/test_objectives_benchmarks.py::test_cql_speed[False-None] |
38.56 | 39.01 | +1.15% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictReplayBuffer-ListStorage-SamplerWithoutReplacement-4000] |
167.16 | 169.06 | +1.14% |
benchmarks/test_objectives_benchmarks.py::test_a2c_speed[False-None] |
179.17 | 181.10 | +1.08% |
benchmarks/test_objectives_benchmarks.py::test_values[vec_td_lambda_return_estimate-True-False] |
55.54 | 54.95 | -1.06% |
| ... | ... | ... | Showing 120 of 216 comparisons, sorted by absolute change. |
GPU
Compared 226 benchmarks. Regressions over 5%: 17. Improvements over 5%: 21.
| Benchmark | main ops | PR ops | Change |
|---|---|---|---|
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictReplayBuffer-ListStorage-RandomSampler-400] |
44.48 | 185.51 | +317.05% |
benchmarks/test_objectives_benchmarks.py::test_cql_speed[reduce-overhead-None] |
59.53 | 88.44 | +48.56% |
benchmarks/test_objectives_benchmarks.py::test_iql_speed[reduce-overhead-None] |
76.23 | 102.62 | +34.62% |
benchmarks/test_objectives_benchmarks.py::test_redq_deprec_speed[reduce-overhead-None] |
83.98 | 105.86 | +26.06% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictReplayBuffer-LazyTensorStorage-SamplerWithoutReplacement-400] |
828.34 | 1,005 | +21.27% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictReplayBuffer-LazyMemmapStorage-RandomSampler-10000] |
2,828 | 3,305 | +16.86% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictReplayBuffer-LazyTensorStorage-sampler7-10000] |
849.19 | 708.15 | -16.61% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictReplayBuffer-LazyMemmapStorage-sampler6-10000] |
797.25 | 692.52 | -13.14% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictReplayBuffer-LazyTensorStorage-SamplerWithoutReplacement-10000] |
3,049 | 2,657 | -12.85% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-10000] |
2,002 | 2,245 | +12.13% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-10000] |
2,158 | 1,910 | -11.47% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictPrioritizedReplayBuffer-LazyMemmapStorage-None-10000] |
2,082 | 1,860 | -10.67% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-10000] |
3,306 | 2,959 | -10.51% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictReplayBuffer-LazyMemmapStorage-SamplerWithoutReplacement-10000] |
2,988 | 2,691 | -9.94% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictReplayBuffer-LazyMemmapStorage-RandomSampler-400] |
520.27 | 471.58 | -9.36% |
benchmarks/test_objectives_benchmarks.py::test_values[vec_generalized_advantage_estimate-True-True] |
487.79 | 444.14 | -8.95% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-10000] |
3,205 | 3,487 | +8.78% |
benchmarks/test_objectives_benchmarks.py::test_td3_speed[True-backward] |
365.25 | 395.06 | +8.16% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictReplayBuffer-LazyMemmapStorage-RandomSampler-10000] |
3,029 | 2,792 | -7.82% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-400] |
793.08 | 851.88 | +7.41% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictReplayBuffer-LazyMemmapStorage-SamplerWithoutReplacement-10000] |
2,787 | 2,583 | -7.31% |
benchmarks/test_objectives_benchmarks.py::test_ddpg_speed[True-backward] |
449.35 | 480.53 | +6.94% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictPrioritizedReplayBuffer-LazyMemmapStorage-None-10000] |
2,020 | 2,158 | +6.84% |
benchmarks/test_objectives_benchmarks.py::test_reinforce_speed[True-backward] |
347.44 | 370.30 | +6.58% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-False-True-False-True] |
30,645 | 28,875 | -5.77% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictReplayBuffer-LazyTensorStorage-SamplerWithoutReplacement-10000] |
3,074 | 3,251 | +5.76% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-True-True-True-True] |
21,028 | 19,823 | -5.73% |
benchmarks/test_compressed_storage_benchmark.py::TestCompressedStorageBenchmark::test_tensor_to_bytestream_speed[safetensors] |
22,195 | 23,466 | +5.72% |
benchmarks/test_vla_preprocessing_benchmark.py::test_openvla_preprocessing_throughput[torchvision-256-256-16] |
40.89 | 43.22 | +5.71% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-False-False-False-True] |
29,171 | 27,517 | -5.67% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-False-True-False-True] |
38,681 | 36,591 | -5.40% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-False-False-False-True] |
35,570 | 33,664 | -5.36% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-True-False-False-True] |
30,527 | 28,902 | -5.32% |
benchmarks/test_objectives_benchmarks.py::test_a2c_speed[True-backward] |
344.72 | 363.00 | +5.30% |
benchmarks/test_objectives_benchmarks.py::test_redq_deprec_speed[True-backward] |
264.14 | 277.82 | +5.18% |
benchmarks/test_envs_benchmark.py::test_cat_frames_functional[4-constant] |
4,660 | 4,896 | +5.06% |
benchmarks/test_objectives_benchmarks.py::test_sac_speed[True-backward] |
319.48 | 335.62 | +5.05% |
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_storage_write_lazystack[50-img_shape0-small] |
4,188 | 4,399 | +5.04% |
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_collector_stack_then_write[100-img_shape2-large_img] |
172.66 | 164.08 | -4.96% |
benchmarks/test_objectives_benchmarks.py::test_gae_speed[generalized_advantage_estimate-False-1-512] |
47.52 | 49.79 | +4.77% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-False-True-False-False] |
64,618 | 61,538 | -4.77% |
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_collector_stack_then_write[200-img_shape3-large_batch] |
132.25 | 138.29 | +4.57% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictPrioritizedReplayBuffer-LazyMemmapStorage-None-400] |
430.98 | 412.18 | -4.36% |
benchmarks/test_envs_benchmark.py::test_cat_frames_functional[16-constant] |
4,599 | 4,792 | +4.19% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-False-False-False-False] |
44,800 | 42,937 | -4.16% |
benchmarks/test_objectives_benchmarks.py::test_sac_speed[True-None] |
627.25 | 601.32 | -4.13% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-False-True-False-False] |
49,505 | 47,487 | -4.08% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-False-True-True-False] |
29,541 | 28,340 | -4.07% |
benchmarks/test_replaybuffer_benchmark.py::TestPrioritizedReplayBufferBenchmark::test_sample_mixed_devices[1000000-cuda_storage_cpu_sampler] |
88.04 | 84.57 | -3.95% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-True-True-False-False] |
57,706 | 55,466 | -3.88% |
benchmarks/test_vla_preprocessing_benchmark.py::test_openvla_preprocessing_throughput[pil-224-224-1] |
279.23 | 268.46 | -3.86% |
benchmarks/test_replaybuffer_benchmark.py::TestPrioritizedReplayBufferBenchmark::test_sampler_sample_scale[1000000-cuda] |
2,309 | 2,220 | -3.84% |
benchmarks/test_objectives_benchmarks.py::test_cql_speed[True-backward] |
225.00 | 233.50 | +3.78% |
benchmarks/test_vla_preprocessing_benchmark.py::test_openvla_preprocessing_throughput[torchvision-224-224-16] |
46.84 | 48.60 | +3.75% |
benchmarks/test_envs_benchmark.py::test_serial |
0.4136 | 0.4290 | +3.74% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-True-True-False-True] |
42,935 | 41,345 | -3.70% |
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_storage_write_contiguous[200-img_shape3-large_batch] |
729.92 | 703.47 | -3.62% |
benchmarks/test_replaybuffer_benchmark.py::TestPrioritizedReplayBufferBenchmark::test_sample_mixed_devices[1000000-cuda_storage_cuda_samp... |
1,527 | 1,472 | -3.59% |
benchmarks/test_vla_preprocessing_benchmark.py::test_openvla_preprocessing_throughput[pil-480-640-1] |
79.75 | 76.93 | -3.54% |
benchmarks/test_objectives_benchmarks.py::test_dqn_speed[True-backward] |
893.98 | 863.36 | -3.42% |
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_storage_write_lazystack[100-img_shape2-large_img] |
394.39 | 407.87 | +3.42% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-True-False-True-True] |
19,655 | 18,998 | -3.34% |
benchmarks/test_envs_benchmark.py::test_simple |
1.2455 | 1.2041 | -3.32% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-True-False-True-False] |
31,911 | 30,865 | -3.28% |
benchmarks/test_vla_preprocessing_benchmark.py::test_openvla_preprocessing_throughput[torchvision-480-640-1] |
470.56 | 455.63 | -3.17% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-True-False-False-False] |
50,430 | 48,839 | -3.16% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-False-False-False-False] |
55,184 | 53,451 | -3.14% |
benchmarks/test_objectives_benchmarks.py::test_reinforce_speed[True-None] |
767.03 | 743.12 | -3.12% |
benchmarks/test_vla_preprocessing_benchmark.py::test_openvla_preprocessing_throughput[torchvision-224-224-1] |
625.01 | 605.71 | -3.09% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-True-False-True-True] |
22,155 | 21,480 | -3.05% |
benchmarks/test_objectives_benchmarks.py::test_iql_speed[True-backward] |
240.24 | 247.54 | +3.04% |
benchmarks/test_rnn_reset_backends_benchmark.py::test_rnn_rollout_with_intermediate_resets[b256-t128-i32-h512-triton-False-0-gru] |
20.69 | 20.06 | -3.01% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-True-False-False-False] |
63,637 | 61,730 | -3.00% |
benchmarks/test_rnn_reset_backends_benchmark.py::test_rnn_rollout_with_intermediate_resets[b256-t128-i32-h512-scan-False-0-gru] |
23.01 | 22.33 | -2.96% |
benchmarks/test_objectives_benchmarks.py::test_a2c_speed[False-backward] |
150.07 | 154.42 | +2.90% |
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_collector_lazystack_then_write[50-img_shape0-small] |
3,463 | 3,365 | -2.82% |
benchmarks/test_vla_preprocessing_benchmark.py::test_openvla_preprocessing_throughput[pil-480-640-16] |
4.9388 | 4.8014 | -2.78% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-True-True-False-False] |
77,195 | 75,078 | -2.74% |
benchmarks/test_objectives_benchmarks.py::test_ppo_speed[False-backward] |
131.14 | 134.73 | +2.73% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-True-True-True-False] |
34,579 | 33,644 | -2.70% |
benchmarks/test_storage_write_benchmark.py::TestCollectorIntegrationBenchmark::test_collector_with_rb_cuda[200-img_shape1-large_batch] |
6.7563 | 6.5762 | -2.67% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-False-True-True-False] |
34,895 | 33,992 | -2.59% |
benchmarks/test_vla_preprocessing_benchmark.py::test_openvla_preprocessing_throughput[torchvision-480-640-16] |
35.13 | 36.04 | +2.59% |
benchmarks/test_objectives_benchmarks.py::test_ddpg_speed[False-backward] |
238.31 | 244.42 | +2.56% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-False-False-True-True] |
19,638 | 19,135 | -2.56% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictReplayBuffer-LazyMemmapStorage-SamplerWithoutReplacement-400] |
510.06 | 523.01 | +2.54% |
benchmarks/test_vla_preprocessing_benchmark.py::test_openvla_preprocessing_throughput[torchvision-256-256-64] |
10.43 | 10.69 | +2.51% |
benchmarks/test_vla_preprocessing_benchmark.py::test_openvla_preprocessing_throughput[pil-480-640-4] |
19.74 | 19.25 | -2.50% |
benchmarks/test_vla_preprocessing_benchmark.py::test_openvla_preprocessing_throughput[torchvision-256-256-4] |
156.39 | 160.28 | +2.49% |
benchmarks/test_vla_preprocessing_benchmark.py::test_openvla_preprocessing_throughput[torchvision-224-224-64] |
12.07 | 12.37 | +2.47% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-True-True-True-True] |
23,633 | 23,069 | -2.39% |
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_collector_lazystack_then_write[100-img_shape1-atari] |
625.82 | 611.06 | -2.36% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-False-True-True-True] |
20,968 | 20,477 | -2.34% |
benchmarks/test_replaybuffer_benchmark.py::TestPrioritizedReplayBufferBenchmark::test_sample_mixed_devices[1000000-memmap_cpu_storage_cud... |
987.72 | 965.02 | -2.30% |
benchmarks/test_non_tensor_env_benchmark.py::test_non_tensor_env_rollout_speed[1000-parallel-buffers-True] |
0.5051 | 0.5167 | +2.29% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-True-True-True-False] |
42,195 | 43,127 | +2.21% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-100000-10000-100-False] |
51.88 | 50.75 | -2.18% |
benchmarks/test_objectives_benchmarks.py::test_a2c_speed[reduce-overhead-None] |
861.06 | 842.35 | -2.17% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-False-True-True-True] |
18,780 | 18,378 | -2.14% |
benchmarks/test_vla_preprocessing_benchmark.py::test_openvla_preprocessing_throughput[pil-480-640-64] |
1.2282 | 1.2019 | -2.14% |
benchmarks/test_collectors_benchmark.py::test_async |
10.85 | 11.07 | +2.04% |
benchmarks/test_vla_preprocessing_benchmark.py::test_openvla_preprocessing_throughput[pil-224-224-4] |
70.31 | 68.90 | -2.01% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[False-True-True-False-True] |
31,692 | 31,061 | -1.99% |
benchmarks/test_envs_benchmark.py::test_parallel |
0.5345 | 0.5449 | +1.95% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-10000-10000-100-False] |
52.35 | 51.36 | -1.89% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-400] |
962.48 | 980.33 | +1.86% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictReplayBuffer-ListStorage-SamplerWithoutReplacement-400] |
190.54 | 187.04 | -1.83% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-10000-10000-100-True] |
23.44 | 23.01 | -1.83% |
benchmarks/test_storage_write_benchmark.py::TestCollectorIntegrationBenchmark::test_collector_with_rb_cuda[100-img_shape0-atari] |
13.42 | 13.18 | -1.82% |
benchmarks/test_compressed_storage_benchmark.py::TestCompressedStorageBenchmark::test_tensor_to_bytestream_speed[torch.save] |
6,930 | 7,055 | +1.80% |
benchmarks/test_envs_benchmark.py::test_step_mdp_speed[True-True-False-True-False] |
38,221 | 37,533 | -1.80% |
benchmarks/test_storage_write_benchmark.py::TestStorageWriteBenchmark::test_storage_write_contiguous[50-img_shape0-small] |
5,875 | 5,980 | +1.79% |
benchmarks/test_vla_preprocessing_benchmark.py::test_openvla_preprocessing_throughput[torchvision-480-640-64] |
6.9892 | 7.1126 | +1.77% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_sample[TensorDictReplayBuffer-ListStorage-RandomSampler-4000] |
159.90 | 162.68 | +1.74% |
benchmarks/test_vla_preprocessing_benchmark.py::test_openvla_preprocessing_throughput[pil-224-224-64] |
4.4536 | 4.3784 | -1.69% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_iterate[TensorDictReplayBuffer-ListStorage-SamplerWithoutReplacement-4000] |
166.58 | 169.31 | +1.64% |
benchmarks/test_replaybuffer_benchmark.py::test_rb_populate[TensorDictPrioritizedReplayBuffer-ListStorage-None-400] |
184.17 | 181.37 | -1.52% |
benchmarks/test_objectives_benchmarks.py::test_iql_speed[False-backward] |
68.74 | 69.78 | +1.50% |
benchmarks/test_vla_preprocessing_benchmark.py::test_openvla_preprocessing_throughput[torchvision-480-640-4] |
142.07 | 144.16 | +1.47% |
benchmarks/test_objectives_benchmarks.py::test_td3_speed[False-None] |
114.35 | 112.70 | -1.44% |
| ... | ... | ... | Showing 120 of 226 comparisons, sorted by absolute change. |
vmoens
commented
Jun 21, 2026
This was referenced Jun 24, 2026
theap06
suggested changes
Jun 26, 2026
This was referenced Jul 6, 2026
vmoens
commented
Jul 6, 2026
vmoens
commented
Jul 6, 2026
vmoens
commented
Jul 6, 2026
vmoens
commented
Jul 6, 2026
This was referenced Jul 6, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Stack from ghstack (oldest at bottom):