Skip to content

[Feature] Memmap storage cleanup#3321

Merged
vmoens merged 2 commits into
gh/vmoens/189/basefrom
gh/vmoens/189/head
Jan 13, 2026
Merged

[Feature] Memmap storage cleanup#3321
vmoens merged 2 commits into
gh/vmoens/189/basefrom
gh/vmoens/189/head

Conversation

@vmoens

@vmoens vmoens commented Jan 12, 2026 •
edited
Loading

Copy link
Copy Markdown
Collaborator

@vmoens
Update
88c3b52 
[ghstack-poisoned]
@pytorch-bot

pytorch-bot Bot commented Jan 12, 2026 •
edited
Loading

Copy link
Copy Markdown

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/rl/3321

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

❌ 4 New Failures, 1 Unrelated Failure

As of commit d4c2e50 with merge base 0a98e17 (image):

NEW FAILURES - The following jobs have failed:

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

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

@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 12, 2026
This was referenced Jan 12, 2026
Merged
Merged
Merged
Merged
Merged
Merged
Merged
Merged
Merged
Merged
@github-actions

github-actions Bot commented Jan 12, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

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

Total Benchmarks: 164. Improved: $\large\color{#35bf28}24$. Worsened: $\large\color{#d91a1a}7$.

Expand to view detailed results
Name Max Mean Ops Ops on Repo HEAD Change
test_tensor_to_bytestream_speed[pickle] 82.7247μs 81.3170μs 12.2976 KOps/s 12.3306 KOps/s $\color{#d91a1a}-0.27\%$
test_tensor_to_bytestream_speed[torch.save] 0.1411ms 0.1402ms 7.1303 KOps/s 7.0083 KOps/s $\color{#35bf28}+1.74\%$
test_tensor_to_bytestream_speed[untyped_storage] 0.1225s 0.1222s 8.1854 Ops/s 7.5130 Ops/s $\textbf{\color{#35bf28}+8.95\%}$
test_tensor_to_bytestream_speed[numpy] 2.6944μs 2.6798μs 373.1611 KOps/s 367.3233 KOps/s $\color{#35bf28}+1.59\%$
test_tensor_to_bytestream_speed[safetensors] 38.0825μs 37.8245μs 26.4379 KOps/s 26.1706 KOps/s $\color{#35bf28}+1.02\%$
test_simple 0.5478s 0.5469s 1.8284 Ops/s 1.7254 Ops/s $\textbf{\color{#35bf28}+5.97\%}$
test_transformed 1.1243s 1.1227s 0.8907 Ops/s 0.8649 Ops/s $\color{#35bf28}+2.99\%$
test_serial 1.6695s 1.6678s 0.5996 Ops/s 0.5799 Ops/s $\color{#35bf28}+3.39\%$
test_parallel 1.2419s 1.1610s 0.8613 Ops/s 0.8835 Ops/s $\color{#d91a1a}-2.51\%$
test_step_mdp_speed[True-True-True-True-True] 0.3120ms 44.9518μs 22.2460 KOps/s 22.5760 KOps/s $\color{#d91a1a}-1.46\%$
test_step_mdp_speed[True-True-True-True-False] 57.0610μs 25.2082μs 39.6696 KOps/s 39.6088 KOps/s $\color{#35bf28}+0.15\%$
test_step_mdp_speed[True-True-True-False-True] 51.5410μs 24.6765μs 40.5244 KOps/s 40.1435 KOps/s $\color{#35bf28}+0.95\%$
test_step_mdp_speed[True-True-True-False-False] 49.7310μs 13.6834μs 73.0812 KOps/s 73.0732 KOps/s $\color{#35bf28}+0.01\%$
test_step_mdp_speed[True-True-False-True-True] 90.3910μs 46.9097μs 21.3175 KOps/s 21.1494 KOps/s $\color{#35bf28}+0.79\%$
test_step_mdp_speed[True-True-False-True-False] 54.1610μs 27.3302μs 36.5896 KOps/s 36.2071 KOps/s $\color{#35bf28}+1.06\%$
test_step_mdp_speed[True-True-False-False-True] 55.6910μs 27.3686μs 36.5383 KOps/s 35.8085 KOps/s $\color{#35bf28}+2.04\%$
test_step_mdp_speed[True-True-False-False-False] 41.7700μs 16.7239μs 59.7945 KOps/s 60.8399 KOps/s $\color{#d91a1a}-1.72\%$
test_step_mdp_speed[True-False-True-True-True] 91.1820μs 50.9193μs 19.6389 KOps/s 20.1899 KOps/s $\color{#d91a1a}-2.73\%$
test_step_mdp_speed[True-False-True-True-False] 67.0710μs 30.3760μs 32.9208 KOps/s 32.6141 KOps/s $\color{#35bf28}+0.94\%$
test_step_mdp_speed[True-False-True-False-True] 76.6510μs 27.2976μs 36.6333 KOps/s 35.9471 KOps/s $\color{#35bf28}+1.91\%$
test_step_mdp_speed[True-False-True-False-False] 47.9410μs 16.4736μs 60.7030 KOps/s 60.8182 KOps/s $\color{#d91a1a}-0.19\%$
test_step_mdp_speed[True-False-False-True-True] 98.3320μs 53.5578μs 18.6714 KOps/s 19.0649 KOps/s $\color{#d91a1a}-2.06\%$
test_step_mdp_speed[True-False-False-True-False] 64.7410μs 32.8311μs 30.4590 KOps/s 30.2709 KOps/s $\color{#35bf28}+0.62\%$
test_step_mdp_speed[True-False-False-False-True] 70.6210μs 29.8196μs 33.5349 KOps/s 33.2351 KOps/s $\color{#35bf28}+0.90\%$
test_step_mdp_speed[True-False-False-False-False] 53.4410μs 19.3772μs 51.6070 KOps/s 51.9727 KOps/s $\color{#d91a1a}-0.70\%$
test_step_mdp_speed[False-True-True-True-True] 87.3010μs 49.5574μs 20.1786 KOps/s 20.0948 KOps/s $\color{#35bf28}+0.42\%$
test_step_mdp_speed[False-True-True-True-False] 69.3010μs 30.1537μs 33.1634 KOps/s 32.8489 KOps/s $\color{#35bf28}+0.96\%$
test_step_mdp_speed[False-True-True-False-True] 2.3607ms 31.9801μs 31.2695 KOps/s 31.3820 KOps/s $\color{#d91a1a}-0.36\%$
test_step_mdp_speed[False-True-True-False-False] 57.9210μs 18.4112μs 54.3149 KOps/s 55.2016 KOps/s $\color{#d91a1a}-1.61\%$
test_step_mdp_speed[False-True-False-True-True] 86.4420μs 52.5000μs 19.0476 KOps/s 19.0365 KOps/s $\color{#35bf28}+0.06\%$
test_step_mdp_speed[False-True-False-True-False] 73.0910μs 32.9079μs 30.3879 KOps/s 30.0137 KOps/s $\color{#35bf28}+1.25\%$
test_step_mdp_speed[False-True-False-False-True] 68.4910μs 34.3556μs 29.1073 KOps/s 29.2062 KOps/s $\color{#d91a1a}-0.34\%$
test_step_mdp_speed[False-True-False-False-False] 49.0310μs 20.6937μs 48.3239 KOps/s 48.0180 KOps/s $\color{#35bf28}+0.64\%$
test_step_mdp_speed[False-False-True-True-True] 94.3820μs 54.5193μs 18.3421 KOps/s 18.2121 KOps/s $\color{#35bf28}+0.71\%$
test_step_mdp_speed[False-False-True-True-False] 63.7710μs 35.5308μs 28.1446 KOps/s 28.0779 KOps/s $\color{#35bf28}+0.24\%$
test_step_mdp_speed[False-False-True-False-True] 77.0610μs 33.7598μs 29.6210 KOps/s 29.7244 KOps/s $\color{#d91a1a}-0.35\%$
test_step_mdp_speed[False-False-True-False-False] 51.0310μs 20.6230μs 48.4895 KOps/s 48.6575 KOps/s $\color{#d91a1a}-0.35\%$
test_step_mdp_speed[False-False-False-True-True] 93.3510μs 57.1795μs 17.4888 KOps/s 17.5166 KOps/s $\color{#d91a1a}-0.16\%$
test_step_mdp_speed[False-False-False-True-False] 66.1110μs 37.8645μs 26.4100 KOps/s 25.9297 KOps/s $\color{#35bf28}+1.85\%$
test_step_mdp_speed[False-False-False-False-True] 68.1410μs 36.4684μs 27.4210 KOps/s 27.7038 KOps/s $\color{#d91a1a}-1.02\%$
test_step_mdp_speed[False-False-False-False-False] 50.4110μs 23.2111μs 43.0828 KOps/s 43.3192 KOps/s $\color{#d91a1a}-0.55\%$
test_non_tensor_env_rollout_speed[1000-single-True] 0.8723s 0.7677s 1.3027 Ops/s 1.2938 Ops/s $\color{#35bf28}+0.68\%$
test_non_tensor_env_rollout_speed[1000-single-False] 0.7291s 0.6272s 1.5943 Ops/s 1.5802 Ops/s $\color{#35bf28}+0.90\%$
test_non_tensor_env_rollout_speed[1000-serial-no-buffers-True] 1.7635s 1.6813s 0.5948 Ops/s 0.5909 Ops/s $\color{#35bf28}+0.66\%$
test_non_tensor_env_rollout_speed[1000-serial-no-buffers-False] 1.5365s 1.4535s 0.6880 Ops/s 0.6801 Ops/s $\color{#35bf28}+1.17\%$
test_non_tensor_env_rollout_speed[1000-serial-buffers-True] 2.0102s 1.9239s 0.5198 Ops/s 0.5150 Ops/s $\color{#35bf28}+0.94\%$
test_non_tensor_env_rollout_speed[1000-serial-buffers-False] 1.7858s 1.7064s 0.5860 Ops/s 0.5801 Ops/s $\color{#35bf28}+1.01\%$
test_non_tensor_env_rollout_speed[1000-parallel-no-buffers-True] 4.7655s 4.7051s 0.2125 Ops/s 0.2100 Ops/s $\color{#35bf28}+1.22\%$
test_non_tensor_env_rollout_speed[1000-parallel-no-buffers-False] 4.5745s 4.4996s 0.2222 Ops/s 0.2214 Ops/s $\color{#35bf28}+0.39\%$
test_non_tensor_env_rollout_speed[1000-parallel-buffers-True] 2.0265s 1.9708s 0.5074 Ops/s 0.5045 Ops/s $\color{#35bf28}+0.57\%$
test_non_tensor_env_rollout_speed[1000-parallel-buffers-False] 1.7612s 1.6796s 0.5954 Ops/s 0.5835 Ops/s $\color{#35bf28}+2.03\%$
test_values[generalized_advantage_estimate-True-True] 10.9859ms 10.3384ms 96.7272 Ops/s 90.7668 Ops/s $\textbf{\color{#35bf28}+6.57\%}$
test_values[vec_generalized_advantage_estimate-True-True] 17.5509ms 11.2652ms 88.7689 Ops/s 56.8436 Ops/s $\textbf{\color{#35bf28}+56.16\%}$
test_values[td0_return_estimate-False-False] 0.2167ms 0.1278ms 7.8270 KOps/s 7.6700 KOps/s $\color{#35bf28}+2.05\%$
test_values[td1_return_estimate-False-False] 28.1810ms 27.8017ms 35.9691 Ops/s 33.4193 Ops/s $\textbf{\color{#35bf28}+7.63\%}$
test_values[vec_td1_return_estimate-False-False] 11.4895ms 11.1881ms 89.3803 Ops/s 56.4527 Ops/s $\textbf{\color{#35bf28}+58.33\%}$
test_values[td_lambda_return_estimate-True-False] 41.3971ms 40.8470ms 24.4816 Ops/s 22.6893 Ops/s $\textbf{\color{#35bf28}+7.90\%}$
test_values[vec_td_lambda_return_estimate-True-False] 17.5665ms 11.2718ms 88.7171 Ops/s 56.9859 Ops/s $\textbf{\color{#35bf28}+55.68\%}$
test_gae_speed[generalized_advantage_estimate-False-1-512] 9.6904ms 9.2434ms 108.1857 Ops/s 100.5520 Ops/s $\textbf{\color{#35bf28}+7.59\%}$
test_gae_speed[vec_generalized_advantage_estimate-True-1-512] 1.7214ms 1.5506ms 644.9209 Ops/s 629.4705 Ops/s $\color{#35bf28}+2.45\%$
test_gae_speed[vec_generalized_advantage_estimate-False-1-512] 0.5427ms 0.4207ms 2.3771 KOps/s 2.3439 KOps/s $\color{#35bf28}+1.42\%$
test_gae_speed[vec_generalized_advantage_estimate-True-32-512] 30.2172ms 29.6839ms 33.6882 Ops/s 28.2336 Ops/s $\textbf{\color{#35bf28}+19.32\%}$
test_gae_speed[vec_generalized_advantage_estimate-False-32-512] 1.8574ms 1.7045ms 586.6720 Ops/s 582.7163 Ops/s $\color{#35bf28}+0.68\%$
test_dqn_speed[False-None] 1.5928ms 1.4022ms 713.1443 Ops/s 702.6397 Ops/s $\color{#35bf28}+1.50\%$
test_dqn_speed[False-backward] 2.3829ms 1.9300ms 518.1298 Ops/s 510.4782 Ops/s $\color{#35bf28}+1.50\%$
test_dqn_speed[True-None] 0.9433ms 0.5554ms 1.8005 KOps/s 1.8241 KOps/s $\color{#d91a1a}-1.29\%$
test_dqn_speed[True-backward] 1.0337ms 1.0020ms 997.9958 Ops/s 921.5042 Ops/s $\textbf{\color{#35bf28}+8.30\%}$
test_dqn_speed[reduce-overhead-None] 0.6652ms 0.5513ms 1.8138 KOps/s 1.8405 KOps/s $\color{#d91a1a}-1.45\%$
test_dqn_speed[reduce-overhead-backward] 1.0275ms 0.9840ms 1.0163 KOps/s 1.0024 KOps/s $\color{#35bf28}+1.38\%$
test_ddpg_speed[False-None] 3.2328ms 2.8750ms 347.8256 Ops/s 339.7591 Ops/s $\color{#35bf28}+2.37\%$
test_ddpg_speed[False-backward] 4.2866ms 4.1096ms 243.3310 Ops/s 236.6919 Ops/s $\color{#35bf28}+2.80\%$
test_ddpg_speed[True-None] 1.7828ms 1.4181ms 705.1854 Ops/s 632.5107 Ops/s $\textbf{\color{#35bf28}+11.49\%}$
test_ddpg_speed[True-backward] 2.5097ms 2.4112ms 414.7261 Ops/s 402.0915 Ops/s $\color{#35bf28}+3.14\%$
test_ddpg_speed[reduce-overhead-None] 1.5601ms 1.4313ms 698.6754 Ops/s 698.2391 Ops/s $\color{#35bf28}+0.06\%$
test_ddpg_speed[reduce-overhead-backward] 2.6925ms 2.4094ms 415.0344 Ops/s 410.7065 Ops/s $\color{#35bf28}+1.05\%$
test_sac_speed[False-None] 8.6164ms 8.0440ms 124.3169 Ops/s 123.1970 Ops/s $\color{#35bf28}+0.91\%$
test_sac_speed[False-backward] 11.7905ms 11.3147ms 88.3804 Ops/s 88.0457 Ops/s $\color{#35bf28}+0.38\%$
test_sac_speed[True-None] 2.3337ms 2.1546ms 464.1160 Ops/s 445.6109 Ops/s $\color{#35bf28}+4.15\%$
test_sac_speed[True-backward] 4.0714ms 3.9831ms 251.0591 Ops/s 239.8089 Ops/s $\color{#35bf28}+4.69\%$
test_sac_speed[reduce-overhead-None] 2.5174ms 2.1546ms 464.1157 Ops/s 447.9908 Ops/s $\color{#35bf28}+3.60\%$
test_sac_speed[reduce-overhead-backward] 4.1514ms 4.0226ms 248.5949 Ops/s 238.9625 Ops/s $\color{#35bf28}+4.03\%$
test_redq_speed[False-None] 11.2535ms 10.4245ms 95.9277 Ops/s 93.7502 Ops/s $\color{#35bf28}+2.32\%$
test_redq_speed[False-backward] 19.0138ms 17.8926ms 55.8890 Ops/s 54.9408 Ops/s $\color{#35bf28}+1.73\%$
test_redq_speed[True-None] 4.4937ms 4.2979ms 232.6728 Ops/s 217.9412 Ops/s $\textbf{\color{#35bf28}+6.76\%}$
test_redq_speed[True-backward] 9.9822ms 9.5678ms 104.5175 Ops/s 100.9398 Ops/s $\color{#35bf28}+3.54\%$
test_redq_speed[reduce-overhead-None] 4.7916ms 4.2842ms 233.4137 Ops/s 216.8218 Ops/s $\textbf{\color{#35bf28}+7.65\%}$
test_redq_speed[reduce-overhead-backward] 10.0901ms 9.7107ms 102.9793 Ops/s 99.2301 Ops/s $\color{#35bf28}+3.78\%$
test_redq_deprec_speed[False-None] 11.8112ms 11.2459ms 88.9209 Ops/s 88.2097 Ops/s $\color{#35bf28}+0.81\%$
test_redq_deprec_speed[False-backward] 16.6133ms 16.0600ms 62.2666 Ops/s 61.7422 Ops/s $\color{#35bf28}+0.85\%$
test_redq_deprec_speed[True-None] 4.1394ms 3.7736ms 264.9971 Ops/s 266.9780 Ops/s $\color{#d91a1a}-0.74\%$
test_redq_deprec_speed[True-backward] 8.2631ms 7.8602ms 127.2232 Ops/s 126.5330 Ops/s $\color{#35bf28}+0.55\%$
test_redq_deprec_speed[reduce-overhead-None] 4.0075ms 3.7137ms 269.2737 Ops/s 268.8505 Ops/s $\color{#35bf28}+0.16\%$
test_redq_deprec_speed[reduce-overhead-backward] 8.7576ms 7.9005ms 126.5742 Ops/s 131.9760 Ops/s $\color{#d91a1a}-4.09\%$
test_td3_speed[False-None] 8.2069ms 8.0487ms 124.2438 Ops/s 122.2920 Ops/s $\color{#35bf28}+1.60\%$
test_td3_speed[False-backward] 11.5152ms 10.9396ms 91.4106 Ops/s 90.5512 Ops/s $\color{#35bf28}+0.95\%$
test_td3_speed[True-None] 1.9155ms 1.8478ms 541.1854 Ops/s 523.6947 Ops/s $\color{#35bf28}+3.34\%$
test_td3_speed[True-backward] 3.8429ms 3.6798ms 271.7507 Ops/s 263.6322 Ops/s $\color{#35bf28}+3.08\%$
test_td3_speed[reduce-overhead-None] 1.8931ms 1.8199ms 549.4795 Ops/s 543.8711 Ops/s $\color{#35bf28}+1.03\%$
test_td3_speed[reduce-overhead-backward] 3.7777ms 3.6517ms 273.8414 Ops/s 227.6853 Ops/s $\textbf{\color{#35bf28}+20.27\%}$
test_cql_speed[False-None] 29.2175ms 26.0995ms 38.3149 Ops/s 38.0075 Ops/s $\color{#35bf28}+0.81\%$
test_cql_speed[False-backward] 39.6070ms 35.5494ms 28.1299 Ops/s 27.6287 Ops/s $\color{#35bf28}+1.81\%$
test_cql_speed[True-None] 12.7188ms 12.2764ms 81.4569 Ops/s 79.6864 Ops/s $\color{#35bf28}+2.22\%$
test_cql_speed[True-backward] 18.3452ms 17.9926ms 55.5784 Ops/s 55.3282 Ops/s $\color{#35bf28}+0.45\%$
test_cql_speed[reduce-overhead-None] 12.6726ms 12.3229ms 81.1496 Ops/s 76.5787 Ops/s $\textbf{\color{#35bf28}+5.97\%}$
test_cql_speed[reduce-overhead-backward] 18.8652ms 18.2660ms 54.7465 Ops/s 55.7132 Ops/s $\color{#d91a1a}-1.73\%$
test_a2c_speed[False-None] 5.7428ms 5.4652ms 182.9754 Ops/s 180.8452 Ops/s $\color{#35bf28}+1.18\%$
test_a2c_speed[False-backward] 12.4327ms 12.0262ms 83.1521 Ops/s 84.0665 Ops/s $\color{#d91a1a}-1.09\%$
test_a2c_speed[True-None] 3.9227ms 3.7542ms 266.3679 Ops/s 264.6990 Ops/s $\color{#35bf28}+0.63\%$
test_a2c_speed[True-backward] 8.9510ms 8.6451ms 115.6724 Ops/s 114.1589 Ops/s $\color{#35bf28}+1.33\%$
test_a2c_speed[reduce-overhead-None] 3.8416ms 3.7280ms 268.2400 Ops/s 267.3816 Ops/s $\color{#35bf28}+0.32\%$
test_a2c_speed[reduce-overhead-backward] 8.9605ms 8.7064ms 114.8585 Ops/s 108.9228 Ops/s $\textbf{\color{#35bf28}+5.45\%}$
test_ppo_speed[False-None] 6.1322ms 5.9609ms 167.7603 Ops/s 163.4631 Ops/s $\color{#35bf28}+2.63\%$
test_ppo_speed[False-backward] 12.7922ms 12.5750ms 79.5231 Ops/s 77.5067 Ops/s $\color{#35bf28}+2.60\%$
test_ppo_speed[True-None] 3.7795ms 3.6207ms 276.1880 Ops/s 273.2364 Ops/s $\color{#35bf28}+1.08\%$
test_ppo_speed[True-backward] 8.6533ms 8.3770ms 119.3745 Ops/s 115.9913 Ops/s $\color{#35bf28}+2.92\%$
test_ppo_speed[reduce-overhead-None] 3.8466ms 3.6109ms 276.9369 Ops/s 272.1020 Ops/s $\color{#35bf28}+1.78\%$
test_ppo_speed[reduce-overhead-backward] 9.1307ms 8.7729ms 113.9873 Ops/s 110.9355 Ops/s $\color{#35bf28}+2.75\%$
test_reinforce_speed[False-None] 4.9776ms 4.6299ms 215.9886 Ops/s 213.8250 Ops/s $\color{#35bf28}+1.01\%$
test_reinforce_speed[False-backward] 7.6627ms 7.4729ms 133.8174 Ops/s 132.2763 Ops/s $\color{#35bf28}+1.17\%$
test_reinforce_speed[True-None] 3.1423ms 2.8565ms 350.0741 Ops/s 343.5384 Ops/s $\color{#35bf28}+1.90\%$
test_reinforce_speed[True-backward] 8.0311ms 7.7840ms 128.4692 Ops/s 126.0153 Ops/s $\color{#35bf28}+1.95\%$
test_reinforce_speed[reduce-overhead-None] 3.0012ms 2.8595ms 349.7099 Ops/s 342.2218 Ops/s $\color{#35bf28}+2.19\%$
test_reinforce_speed[reduce-overhead-backward] 8.5716ms 8.0962ms 123.5141 Ops/s 122.3038 Ops/s $\color{#35bf28}+0.99\%$
test_iql_speed[False-None] 25.9831ms 20.6584ms 48.4065 Ops/s 48.6902 Ops/s $\color{#d91a1a}-0.58\%$
test_iql_speed[False-backward] 36.2000ms 31.3537ms 31.8941 Ops/s 32.5648 Ops/s $\color{#d91a1a}-2.06\%$
test_iql_speed[True-None] 8.9189ms 8.5685ms 116.7059 Ops/s 115.2014 Ops/s $\color{#35bf28}+1.31\%$
test_iql_speed[True-backward] 17.1580ms 16.6565ms 60.0366 Ops/s 58.6094 Ops/s $\color{#35bf28}+2.44\%$
test_iql_speed[reduce-overhead-None] 8.8630ms 8.5665ms 116.7342 Ops/s 112.1195 Ops/s $\color{#35bf28}+4.12\%$
test_iql_speed[reduce-overhead-backward] 17.4673ms 17.0103ms 58.7881 Ops/s 56.3774 Ops/s $\color{#35bf28}+4.28\%$
test_rb_sample[TensorDictReplayBuffer-ListStorage-RandomSampler-4000] 8.5270ms 6.1477ms 162.6625 Ops/s 162.9470 Ops/s $\color{#d91a1a}-0.17\%$
test_rb_sample[TensorDictReplayBuffer-LazyMemmapStorage-RandomSampler-10000] 0.6223ms 0.3178ms 3.1467 KOps/s 3.2794 KOps/s $\color{#d91a1a}-4.05\%$
test_rb_sample[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-10000] 0.5463ms 0.3074ms 3.2527 KOps/s 3.5717 KOps/s $\textbf{\color{#d91a1a}-8.93\%}$
test_rb_sample[TensorDictReplayBuffer-ListStorage-SamplerWithoutReplacement-4000] 6.0067ms 5.7944ms 172.5802 Ops/s 171.7130 Ops/s $\color{#35bf28}+0.51\%$
test_rb_sample[TensorDictReplayBuffer-LazyMemmapStorage-SamplerWithoutReplacement-10000] 0.9165ms 0.3257ms 3.0706 KOps/s 2.8597 KOps/s $\textbf{\color{#35bf28}+7.37\%}$
test_rb_sample[TensorDictReplayBuffer-LazyTensorStorage-SamplerWithoutReplacement-10000] 0.4947ms 0.2880ms 3.4716 KOps/s 3.7216 KOps/s $\textbf{\color{#d91a1a}-6.72\%}$
test_rb_sample[TensorDictReplayBuffer-LazyMemmapStorage-sampler6-10000] 1.6518ms 1.3803ms 724.5017 Ops/s 761.7741 Ops/s $\color{#d91a1a}-4.89\%$
test_rb_sample[TensorDictReplayBuffer-LazyTensorStorage-sampler7-10000] 1.6140ms 1.2967ms 771.1922 Ops/s 813.1821 Ops/s $\textbf{\color{#d91a1a}-5.16\%}$
test_rb_sample[TensorDictPrioritizedReplayBuffer-ListStorage-None-4000] 9.3513ms 6.0380ms 165.6181 Ops/s 167.6780 Ops/s $\color{#d91a1a}-1.23\%$
test_rb_sample[TensorDictPrioritizedReplayBuffer-LazyMemmapStorage-None-10000] 1.7136ms 0.4652ms 2.1497 KOps/s 2.0799 KOps/s $\color{#35bf28}+3.36\%$
test_rb_sample[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-10000] 0.7942ms 0.4575ms 2.1859 KOps/s 2.0698 KOps/s $\textbf{\color{#35bf28}+5.61\%}$
test_rb_iterate[TensorDictReplayBuffer-ListStorage-RandomSampler-4000] 5.8287ms 5.7512ms 173.8759 Ops/s 173.8365 Ops/s $\color{#35bf28}+0.02\%$
test_rb_iterate[TensorDictReplayBuffer-LazyMemmapStorage-RandomSampler-10000] 0.8943ms 0.3246ms 3.0804 KOps/s 2.6716 KOps/s $\textbf{\color{#35bf28}+15.30\%}$
test_rb_iterate[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-10000] 0.5558ms 0.3316ms 3.0159 KOps/s 2.8384 KOps/s $\textbf{\color{#35bf28}+6.26\%}$
test_rb_iterate[TensorDictReplayBuffer-ListStorage-SamplerWithoutReplacement-4000] 5.9134ms 5.6732ms 176.2668 Ops/s 175.0397 Ops/s $\color{#35bf28}+0.70\%$
test_rb_iterate[TensorDictReplayBuffer-LazyMemmapStorage-SamplerWithoutReplacement-10000] 0.6862ms 0.2999ms 3.3349 KOps/s 2.9217 KOps/s $\textbf{\color{#35bf28}+14.14\%}$
test_rb_iterate[TensorDictReplayBuffer-LazyTensorStorage-SamplerWithoutReplacement-10000] 0.5367ms 0.3023ms 3.3075 KOps/s 3.7081 KOps/s $\textbf{\color{#d91a1a}-10.80\%}$
test_rb_iterate[TensorDictPrioritizedReplayBuffer-ListStorage-None-4000] 6.1794ms 5.8447ms 171.0963 Ops/s 167.9223 Ops/s $\color{#35bf28}+1.89\%$
test_rb_iterate[TensorDictPrioritizedReplayBuffer-LazyMemmapStorage-None-10000] 1.6523ms 0.5216ms 1.9173 KOps/s 2.2072 KOps/s $\textbf{\color{#d91a1a}-13.13\%}$
test_rb_iterate[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-10000] 0.6949ms 0.4721ms 2.1181 KOps/s 2.3296 KOps/s $\textbf{\color{#d91a1a}-9.08\%}$
test_rb_populate[TensorDictReplayBuffer-ListStorage-RandomSampler-400] 0.9539s 24.2564ms 41.2262 Ops/s 197.2078 Ops/s $\textbf{\color{#d91a1a}-79.10\%}$
test_rb_populate[TensorDictReplayBuffer-LazyMemmapStorage-RandomSampler-400] 9.3942ms 2.0137ms 496.6012 Ops/s 477.6111 Ops/s $\color{#35bf28}+3.98\%$
test_rb_populate[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-400] 2.2340ms 1.0480ms 954.1694 Ops/s 813.0185 Ops/s $\textbf{\color{#35bf28}+17.36\%}$
test_rb_populate[TensorDictReplayBuffer-ListStorage-SamplerWithoutReplacement-400] 7.1647ms 5.0189ms 199.2465 Ops/s 194.7896 Ops/s $\color{#35bf28}+2.29\%$
test_rb_populate[TensorDictReplayBuffer-LazyMemmapStorage-SamplerWithoutReplacement-400] 8.9489ms 2.1183ms 472.0708 Ops/s 482.0976 Ops/s $\color{#d91a1a}-2.08\%$
test_rb_populate[TensorDictReplayBuffer-LazyTensorStorage-SamplerWithoutReplacement-400] 7.1392ms 1.2400ms 806.4781 Ops/s 804.8331 Ops/s $\color{#35bf28}+0.20\%$
test_rb_populate[TensorDictPrioritizedReplayBuffer-ListStorage-None-400] 0.5994s 17.0932ms 58.5027 Ops/s 53.0054 Ops/s $\textbf{\color{#35bf28}+10.37\%}$
test_rb_populate[TensorDictPrioritizedReplayBuffer-LazyMemmapStorage-None-400] 11.7942ms 2.1777ms 459.2051 Ops/s 463.8747 Ops/s $\color{#d91a1a}-1.01\%$
test_rb_populate[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-400] 1.2105ms 1.0293ms 971.5792 Ops/s 958.1793 Ops/s $\color{#35bf28}+1.40\%$
test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-10000-10000-100-True] 38.7553ms 34.2454ms 29.2010 Ops/s 28.5612 Ops/s $\color{#35bf28}+2.24\%$
test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-10000-10000-100-False] 19.0491ms 17.5495ms 56.9817 Ops/s 55.0481 Ops/s $\color{#35bf28}+3.51\%$
test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-100000-10000-100-True] 37.3915ms 35.2229ms 28.3906 Ops/s 27.3580 Ops/s $\color{#35bf28}+3.77\%$
test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-100000-10000-100-False] 19.5366ms 17.9894ms 55.5882 Ops/s 54.4592 Ops/s $\color{#35bf28}+2.07\%$
test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-1000000-10000-100-True] 38.7242ms 36.9255ms 27.0816 Ops/s 26.4771 Ops/s $\color{#35bf28}+2.28\%$
test_rb_extend_sample[ReplayBuffer-LazyTensorStorage-RandomSampler-1000000-10000-100-False] 20.3941ms 19.2672ms 51.9018 Ops/s 50.4745 Ops/s $\color{#35bf28}+2.83\%$

@vmoens
Update
d4c2e50 
[ghstack-poisoned]
vmoens added a commit that referenced this pull request Jan 13, 2026
@vmoens
[Feature] Memmap storage cleanup
7f9ea74 
Implement automatic cleanup for `LazyMemmapStorage` using signal handlers
and `atexit`. This ensures that temporary memmap files are removed from
disk even when the script is interrupted or killed.


ghstack-source-id: 2e2df3e
Pull-Request: #3321
@vmoens vmoens merged commit d4c2e50 into gh/vmoens/189/base Jan 13, 2026
101 of 106 checks passed
@vmoens vmoens deleted the gh/vmoens/189/head branch January 13, 2026 03:45
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

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