Skip to content

[Performance] Optimize RedisTensorDict indexed ops with GETRANGE/SETRANGE - #1569

Open
vmoens wants to merge 1 commit into
gh/vmoens/55/basefrom
gh/vmoens/55/head
Open

[Performance] Optimize RedisTensorDict indexed ops with GETRANGE/SETRANGE#1569
vmoens wants to merge 1 commit into
gh/vmoens/55/basefrom
gh/vmoens/55/head

Conversation

@vmoens

@vmoens vmoens commented Feb 14, 2026 •
edited
Loading

Copy link
Copy Markdown
Collaborator

Stack from ghstack (oldest at bottom):

Replace the full read-modify-write approach for indexed access with Redis
byte-range operations (GETRANGE/SETRANGE). This reduces data transfer from
O(full_tensor) to O(slice) and network round-trips from 2N to 2 (or 1 with
metadata caching).

Key changes:

  • Add _compute_byte_ranges to map tensor indices to byte offsets
  • Add pipelined _abatch_get_at / _abatch_set_at / _abatch_read_modify_write
  • Override setitem to bypass _SubTensorDict for batched SETRANGE
  • Override _get_at_str/_get_at_tuple to use GETRANGE
  • Add metadata cache (cache_metadata=True by default) to skip metadata
    round-trips
  • Refactor inline meta decoding to shared _decode_meta utility

@vmoens
Update
24825d2 
[ghstack-poisoned]
@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 Feb 14, 2026
This was referenced Feb 14, 2026
Open
Open
Open
Open
@github-actions

github-actions Bot commented Feb 14, 2026

Copy link
Copy Markdown
Contributor

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

Total Benchmarks: 243. Improved: $\large\color{#35bf28}16$. Worsened: $\large\color{#d91a1a}6$.

Expand to view detailed results
Name Max Mean Ops Ops on Repo HEAD Change
test_plain_set_nested 40.5310μs 14.2467μs 70.1916 KOps/s 69.3666 KOps/s $\color{#35bf28}+1.19\%$
test_plain_set_stack_nested 39.0710μs 14.7981μs 67.5761 KOps/s 68.5691 KOps/s $\color{#d91a1a}-1.45\%$
test_plain_set_nested_inplace 37.5610μs 16.0422μs 62.3354 KOps/s 61.8765 KOps/s $\color{#35bf28}+0.74\%$
test_plain_set_stack_nested_inplace 54.1610μs 15.9690μs 62.6214 KOps/s 62.3059 KOps/s $\color{#35bf28}+0.51\%$
test_items 28.8910μs 5.4621μs 183.0803 KOps/s 182.8977 KOps/s $\color{#35bf28}+0.10\%$
test_items_nested 0.5667ms 0.5184ms 1.9291 KOps/s 1.9389 KOps/s $\color{#d91a1a}-0.51\%$
test_items_nested_locked 0.5682ms 0.5228ms 1.9129 KOps/s 1.9168 KOps/s $\color{#d91a1a}-0.20\%$
test_items_nested_leaf 0.1303ms 91.1995μs 10.9650 KOps/s 10.9466 KOps/s $\color{#35bf28}+0.17\%$
test_items_stack_nested 0.5800ms 0.5232ms 1.9112 KOps/s 1.9325 KOps/s $\color{#d91a1a}-1.10\%$
test_items_stack_nested_leaf 0.1231ms 90.7184μs 11.0231 KOps/s 10.9891 KOps/s $\color{#35bf28}+0.31\%$
test_items_stack_nested_locked 0.5758ms 0.5268ms 1.8982 KOps/s 1.9255 KOps/s $\color{#d91a1a}-1.42\%$
test_keys 31.6210μs 4.1571μs 240.5536 KOps/s 242.8898 KOps/s $\color{#d91a1a}-0.96\%$
test_keys_nested 0.1604ms 0.1162ms 8.6051 KOps/s 8.5351 KOps/s $\color{#35bf28}+0.82\%$
test_keys_nested_locked 88.2214ms 0.1383ms 7.2329 KOps/s 7.9264 KOps/s $\textbf{\color{#d91a1a}-8.75\%}$
test_keys_nested_leaf 0.1478ms 0.1078ms 9.2732 KOps/s 9.3079 KOps/s $\color{#d91a1a}-0.37\%$
test_keys_stack_nested 0.1626ms 0.1177ms 8.4956 KOps/s 8.4508 KOps/s $\color{#35bf28}+0.53\%$
test_keys_stack_nested_leaf 0.1583ms 0.1080ms 9.2561 KOps/s 9.2188 KOps/s $\color{#35bf28}+0.40\%$
test_keys_stack_nested_locked 0.1698ms 0.1259ms 7.9423 KOps/s 7.8917 KOps/s $\color{#35bf28}+0.64\%$
test_values 5.0220μs 1.0031μs 996.8999 KOps/s 1.0079 MOps/s $\color{#d91a1a}-1.09\%$
test_values_nested 79.9310μs 46.0356μs 21.7223 KOps/s 21.2180 KOps/s $\color{#35bf28}+2.38\%$
test_values_nested_locked 83.6610μs 49.5139μs 20.1963 KOps/s 19.9385 KOps/s $\color{#35bf28}+1.29\%$
test_values_nested_leaf 84.3910μs 52.5949μs 19.0132 KOps/s 18.7779 KOps/s $\color{#35bf28}+1.25\%$
test_values_stack_nested 76.2510μs 46.2636μs 21.6152 KOps/s 21.3295 KOps/s $\color{#35bf28}+1.34\%$
test_values_stack_nested_leaf 85.9810μs 52.9659μs 18.8801 KOps/s 18.8070 KOps/s $\color{#35bf28}+0.39\%$
test_values_stack_nested_locked 0.1235ms 49.9336μs 20.0266 KOps/s 19.9839 KOps/s $\color{#35bf28}+0.21\%$
test_membership 3.9767μs 0.8078μs 1.2379 MOps/s 1.2408 MOps/s $\color{#d91a1a}-0.24\%$
test_membership_nested 23.7000μs 3.0078μs 332.4717 KOps/s 335.0471 KOps/s $\color{#d91a1a}-0.77\%$
test_membership_nested_leaf 24.9610μs 2.9922μs 334.2035 KOps/s 335.5441 KOps/s $\color{#d91a1a}-0.40\%$
test_membership_stacked_nested 25.0600μs 3.0578μs 327.0336 KOps/s 334.1205 KOps/s $\color{#d91a1a}-2.12\%$
test_membership_stacked_nested_leaf 27.1200μs 3.0051μs 332.7647 KOps/s 336.9863 KOps/s $\color{#d91a1a}-1.25\%$
test_membership_nested_last 39.9000μs 4.3675μs 228.9635 KOps/s 228.2591 KOps/s $\color{#35bf28}+0.31\%$
test_membership_nested_leaf_last 34.1510μs 4.4108μs 226.7168 KOps/s 228.7518 KOps/s $\color{#d91a1a}-0.89\%$
test_membership_stacked_nested_last 27.1510μs 4.3384μs 230.5016 KOps/s 228.0407 KOps/s $\color{#35bf28}+1.08\%$
test_membership_stacked_nested_leaf_last 21.2910μs 4.4017μs 227.1859 KOps/s 229.2469 KOps/s $\color{#d91a1a}-0.90\%$
test_nested_getleaf 49.1810μs 20.8614μs 47.9355 KOps/s 48.4651 KOps/s $\color{#d91a1a}-1.09\%$
test_nested_get 46.4110μs 19.7555μs 50.6187 KOps/s 51.7056 KOps/s $\color{#d91a1a}-2.10\%$
test_stacked_getleaf 59.4510μs 20.5939μs 48.5581 KOps/s 47.4522 KOps/s $\color{#35bf28}+2.33\%$
test_stacked_get 46.2510μs 19.6330μs 50.9346 KOps/s 50.2267 KOps/s $\color{#35bf28}+1.41\%$
test_nested_getitemleaf 44.0010μs 21.2454μs 47.0691 KOps/s 47.2729 KOps/s $\color{#d91a1a}-0.43\%$
test_nested_getitem 45.2410μs 20.0072μs 49.9820 KOps/s 49.4021 KOps/s $\color{#35bf28}+1.17\%$
test_stacked_getitemleaf 47.2800μs 21.3582μs 46.8205 KOps/s 46.7361 KOps/s $\color{#35bf28}+0.18\%$
test_stacked_getitem 51.0500μs 20.2584μs 49.3623 KOps/s 49.9792 KOps/s $\color{#d91a1a}-1.23\%$
test_lock_nested 7.7730ms 0.4590ms 2.1785 KOps/s 2.1920 KOps/s $\color{#d91a1a}-0.62\%$
test_lock_stack_nested 0.5107ms 0.4563ms 2.1914 KOps/s 2.1772 KOps/s $\color{#35bf28}+0.65\%$
test_unlock_nested 0.4453ms 0.3631ms 2.7537 KOps/s 2.7429 KOps/s $\color{#35bf28}+0.39\%$
test_unlock_stack_nested 0.4184ms 0.3670ms 2.7248 KOps/s 2.6914 KOps/s $\color{#35bf28}+1.24\%$
test_flatten_speed 0.1671ms 0.1165ms 8.5833 KOps/s 8.5833 KOps/s $-0.00\%$
test_unflatten_speed 0.6468ms 0.5728ms 1.7457 KOps/s 1.7591 KOps/s $\color{#d91a1a}-0.76\%$
test_common_ops 0.8454ms 0.6661ms 1.5012 KOps/s 1.4746 KOps/s $\color{#35bf28}+1.81\%$
test_creation 0.1261ms 2.7532μs 363.2136 KOps/s 363.6732 KOps/s $\color{#d91a1a}-0.13\%$
test_creation_empty 35.1100μs 5.7968μs 172.5091 KOps/s 174.0597 KOps/s $\color{#d91a1a}-0.89\%$
test_creation_nested_1 47.5710μs 10.3790μs 96.3487 KOps/s 96.5742 KOps/s $\color{#d91a1a}-0.23\%$
test_creation_nested_2 35.5000μs 11.3539μs 88.0754 KOps/s 89.4928 KOps/s $\color{#d91a1a}-1.58\%$
test_creation_many_keys[10] 45.3010μs 17.0923μs 58.5057 KOps/s 58.2044 KOps/s $\color{#35bf28}+0.52\%$
test_creation_many_keys[50] 0.1397ms 71.7088μs 13.9453 KOps/s 13.8609 KOps/s $\color{#35bf28}+0.61\%$
test_creation_many_keys[100] 0.1731ms 0.1417ms 7.0569 KOps/s 7.0709 KOps/s $\color{#d91a1a}-0.20\%$
test_creation_nested_many_keys[10] 70.3820μs 36.8037μs 27.1712 KOps/s 27.3796 KOps/s $\color{#d91a1a}-0.76\%$
test_creation_nested_many_keys[50] 0.2026ms 0.1475ms 6.7814 KOps/s 6.7616 KOps/s $\color{#35bf28}+0.29\%$
test_clone 51.0510μs 12.9504μs 77.2180 KOps/s 76.3317 KOps/s $\color{#35bf28}+1.16\%$
test_getitem[int] 1.6486ms 14.1005μs 70.9194 KOps/s 60.4884 KOps/s $\textbf{\color{#35bf28}+17.24\%}$
test_getitem[slice_int] 0.1333ms 24.2091μs 41.3067 KOps/s 41.3230 KOps/s $\color{#d91a1a}-0.04\%$
test_getitem[range] 0.1807ms 59.7649μs 16.7322 KOps/s 16.6317 KOps/s $\color{#35bf28}+0.60\%$
test_getitem[tuple] 0.1448ms 23.7134μs 42.1702 KOps/s 42.5757 KOps/s $\color{#d91a1a}-0.95\%$
test_getitem[list] 0.1858ms 55.1278μs 18.1397 KOps/s 17.5095 KOps/s $\color{#35bf28}+3.60\%$
test_setitem_dim[int] 44.6510μs 25.2032μs 39.6774 KOps/s 39.6376 KOps/s $\color{#35bf28}+0.10\%$
test_setitem_dim[slice_int] 65.1110μs 43.1106μs 23.1961 KOps/s 23.0490 KOps/s $\color{#35bf28}+0.64\%$
test_setitem_dim[range] 0.1234ms 90.9543μs 10.9945 KOps/s 10.5364 KOps/s $\color{#35bf28}+4.35\%$
test_setitem_dim[tuple] 62.1410μs 40.2041μs 24.8731 KOps/s 24.1505 KOps/s $\color{#35bf28}+2.99\%$
test_setitem 60.5110μs 17.4177μs 57.4129 KOps/s 55.8044 KOps/s $\color{#35bf28}+2.88\%$
test_set 56.0910μs 16.6867μs 59.9281 KOps/s 58.5996 KOps/s $\color{#35bf28}+2.27\%$
test_set_shared 0.5141ms 0.2010ms 4.9747 KOps/s 4.9158 KOps/s $\color{#35bf28}+1.20\%$
test_update 0.2056ms 21.6982μs 46.0867 KOps/s 45.7162 KOps/s $\color{#35bf28}+0.81\%$
test_update_nested 75.9010μs 33.7164μs 29.6591 KOps/s 29.4013 KOps/s $\color{#35bf28}+0.88\%$
test_update__nested 0.4599ms 33.3835μs 29.9549 KOps/s 29.3370 KOps/s $\color{#35bf28}+2.11\%$
test_set_nested 49.0810μs 18.8468μs 53.0593 KOps/s 52.5022 KOps/s $\color{#35bf28}+1.06\%$
test_set_nested_new 59.3810μs 23.7572μs 42.0925 KOps/s 41.4146 KOps/s $\color{#35bf28}+1.64\%$
test_select 73.3810μs 40.6467μs 24.6022 KOps/s 24.0827 KOps/s $\color{#35bf28}+2.16\%$
test_select_nested 0.1005ms 71.8988μs 13.9084 KOps/s 14.1124 KOps/s $\color{#d91a1a}-1.45\%$
test_exclude_nested 0.1338ms 93.0832μs 10.7431 KOps/s 10.8376 KOps/s $\color{#d91a1a}-0.87\%$
test_empty[True] 0.4593ms 0.4221ms 2.3694 KOps/s 2.3805 KOps/s $\color{#d91a1a}-0.47\%$
test_empty[False] 8.1178μs 1.2697μs 787.6042 KOps/s 784.7480 KOps/s $\color{#35bf28}+0.36\%$
test_to 0.1030ms 71.1299μs 14.0588 KOps/s 13.9978 KOps/s $\color{#35bf28}+0.44\%$
test_to_nonblocking 0.1089ms 65.8287μs 15.1909 KOps/s 15.7120 KOps/s $\color{#d91a1a}-3.32\%$
test_unbind_speed 0.3584ms 0.3128ms 3.1966 KOps/s 3.1994 KOps/s $\color{#d91a1a}-0.09\%$
test_unbind_speed_stack0 0.3766ms 0.3093ms 3.2335 KOps/s 3.2140 KOps/s $\color{#35bf28}+0.61\%$
test_unbind_speed_stack1 0.1021s 0.8867ms 1.1277 KOps/s 1.1183 KOps/s $\color{#35bf28}+0.85\%$
test_split 1.1611ms 1.0963ms 912.1734 Ops/s 917.6228 Ops/s $\color{#d91a1a}-0.59\%$
test_chunk 0.1023s 1.1561ms 864.9475 Ops/s 792.0064 Ops/s $\textbf{\color{#35bf28}+9.21\%}$
test_to_cpu_blocking 28.5301ms 28.3432ms 35.2818 Ops/s 35.5344 Ops/s $\color{#d91a1a}-0.71\%$
test_to_cpu_global_sync 11.2551ms 11.1139ms 89.9771 Ops/s 91.0730 Ops/s $\color{#d91a1a}-1.20\%$
test_to_cpu_event_sync 0.1144s 13.3025ms 75.1738 Ops/s 83.8341 Ops/s $\textbf{\color{#d91a1a}-10.33\%}$
test_to_cpu_default 12.3467ms 12.0886ms 82.7224 Ops/s 83.7815 Ops/s $\color{#d91a1a}-1.26\%$
test_consolidate[False-None] 4.0474ms 3.9257ms 254.7290 Ops/s 226.1802 Ops/s $\textbf{\color{#35bf28}+12.62\%}$
test_consolidate[default-None] 2.0449ms 1.9418ms 514.9872 Ops/s 500.6764 Ops/s $\color{#35bf28}+2.86\%$
test_consolidate[reduce-overhead-None] 1.9448ms 1.8680ms 535.3331 Ops/s 523.2026 Ops/s $\color{#35bf28}+2.32\%$
test_consolidate_njt[False-None] 8.5061ms 8.1518ms 122.6726 Ops/s 120.0306 Ops/s $\color{#35bf28}+2.20\%$
test_to[False-False-None] 2.1194ms 2.0202ms 495.0067 Ops/s 490.2334 Ops/s $\color{#35bf28}+0.97\%$
test_to[True-False-None] 2.2021ms 1.8802ms 531.8668 Ops/s 529.9025 Ops/s $\color{#35bf28}+0.37\%$
test_to[within-False-None] 6.1548ms 5.8135ms 172.0137 Ops/s 168.5372 Ops/s $\color{#35bf28}+2.06\%$
test_to[True-default-None] 7.4971ms 7.4008ms 135.1214 Ops/s 134.8441 Ops/s $\color{#35bf28}+0.21\%$
test_to_njt[False-False-None] 8.3943ms 8.3206ms 120.1832 Ops/s 119.0109 Ops/s $\color{#35bf28}+0.99\%$
test_to_njt[True-False-None] 7.2470ms 7.0351ms 142.1448 Ops/s 145.4462 Ops/s $\color{#d91a1a}-2.27\%$
test_to_njt[within-False-None] 15.8660ms 15.2625ms 65.5199 Ops/s 65.4507 Ops/s $\color{#35bf28}+0.11\%$
test_creation[device0] 0.4116ms 0.1151ms 8.6868 KOps/s 8.6360 KOps/s $\color{#35bf28}+0.59\%$
test_creation_from_tensor 0.4624ms 0.1111ms 8.9976 KOps/s 8.8595 KOps/s $\color{#35bf28}+1.56\%$
test_add_one[memmap_tensor0] 0.3799ms 6.2234μs 160.6845 KOps/s 155.5961 KOps/s $\color{#35bf28}+3.27\%$
test_contiguous[memmap_tensor0] 25.4100μs 0.6166μs 1.6217 MOps/s 2.2944 MOps/s $\textbf{\color{#d91a1a}-29.32\%}$
test_stack[memmap_tensor0] 33.6610μs 4.3993μs 227.3103 KOps/s 219.7163 KOps/s $\color{#35bf28}+3.46\%$
test_memmaptd_index 0.9996ms 0.2549ms 3.9225 KOps/s 4.0253 KOps/s $\color{#d91a1a}-2.55\%$
test_memmaptd_index_astensor 0.5191ms 0.3439ms 2.9080 KOps/s 2.9471 KOps/s $\color{#d91a1a}-1.33\%$
test_memmaptd_index_op 0.8465ms 0.5786ms 1.7283 KOps/s 1.7012 KOps/s $\color{#35bf28}+1.59\%$
test_serialize_model 0.3063s 0.1596s 6.2644 Ops/s 7.4309 Ops/s $\textbf{\color{#d91a1a}-15.70\%}$
test_serialize_model_pickle 1.3493s 1.2131s 0.8243 Ops/s 0.8259 Ops/s $\color{#d91a1a}-0.18\%$
test_serialize_weights 0.1366s 0.1349s 7.4147 Ops/s 7.4083 Ops/s $\color{#35bf28}+0.09\%$
test_serialize_weights_returnearly 0.4402s 90.7821ms 11.0154 Ops/s 6.0302 Ops/s $\textbf{\color{#35bf28}+82.67\%}$
test_serialize_weights_pickle 1.3685s 1.2328s 0.8112 Ops/s 0.8216 Ops/s $\color{#d91a1a}-1.26\%$
test_reshape_pytree 0.2169ms 32.0815μs 31.1706 KOps/s 31.1412 KOps/s $\color{#35bf28}+0.09\%$
test_reshape_td 85.9920μs 41.9506μs 23.8376 KOps/s 23.2675 KOps/s $\color{#35bf28}+2.45\%$
test_view_pytree 0.2214ms 31.6544μs 31.5911 KOps/s 31.4023 KOps/s $\color{#35bf28}+0.60\%$
test_view_td 97.4220μs 49.7729μs 20.0913 KOps/s 19.7000 KOps/s $\color{#35bf28}+1.99\%$
test_unbind_pytree 0.2630ms 35.8597μs 27.8864 KOps/s 27.8849 KOps/s $+0.01\%$
test_unbind_td 0.1980ms 46.0279μs 21.7259 KOps/s 21.0408 KOps/s $\color{#35bf28}+3.26\%$
test_split_pytree 0.1966ms 40.2902μs 24.8200 KOps/s 24.6427 KOps/s $\color{#35bf28}+0.72\%$
test_split_td 0.1653ms 61.6320μs 16.2253 KOps/s 16.1515 KOps/s $\color{#35bf28}+0.46\%$
test_add_pytree 0.2363ms 41.1639μs 24.2931 KOps/s 24.8055 KOps/s $\color{#d91a1a}-2.07\%$
test_add_td 96.3320μs 51.9106μs 19.2639 KOps/s 19.2891 KOps/s $\color{#d91a1a}-0.13\%$
test_compile_add_one_nested[tensordict-compile] 0.1978ms 0.1377ms 7.2606 KOps/s 6.8456 KOps/s $\textbf{\color{#35bf28}+6.06\%}$
test_compile_add_one_nested[tensordict-eager] 0.2948ms 0.1883ms 5.3101 KOps/s 5.3930 KOps/s $\color{#d91a1a}-1.54\%$
test_compile_add_one_nested[pytree-compile] 0.1489ms 0.1054ms 9.4917 KOps/s 9.2899 KOps/s $\color{#35bf28}+2.17\%$
test_compile_add_one_nested[pytree-eager] 0.4247ms 0.1744ms 5.7335 KOps/s 5.7327 KOps/s $\color{#35bf28}+0.01\%$
test_compile_copy_nested[tensordict-compile] 98.3710μs 28.7450μs 34.7887 KOps/s 30.4887 KOps/s $\textbf{\color{#35bf28}+14.10\%}$
test_compile_copy_nested[tensordict-eager] 0.1022ms 49.2531μs 20.3033 KOps/s 19.9755 KOps/s $\color{#35bf28}+1.64\%$
test_compile_copy_nested[pytree-compile] 35.0110μs 9.4363μs 105.9735 KOps/s 103.6726 KOps/s $\color{#35bf28}+2.22\%$
test_compile_copy_nested[pytree-eager] 0.4581ms 66.1751μs 15.1114 KOps/s 15.0991 KOps/s $\color{#35bf28}+0.08\%$
test_compile_add_one_flat[tensordict-compile] 0.2494ms 0.1735ms 5.7628 KOps/s 5.4837 KOps/s $\textbf{\color{#35bf28}+5.09\%}$
test_compile_add_one_flat[tensordict-eager] 0.3015ms 0.2494ms 4.0096 KOps/s 3.9857 KOps/s $\color{#35bf28}+0.60\%$
test_compile_add_one_flat[tensorclass-compile] 0.1969ms 0.1146ms 8.7223 KOps/s 8.4653 KOps/s $\color{#35bf28}+3.04\%$
test_compile_add_one_flat[tensorclass-eager] 0.1208ms 69.1968μs 14.4515 KOps/s 14.6811 KOps/s $\color{#d91a1a}-1.56\%$
test_compile_add_one_flat[pytree-compile] 0.2475ms 0.1557ms 6.4206 KOps/s 3.7333 KOps/s $\textbf{\color{#35bf28}+71.98\%}$
test_compile_add_one_flat[pytree-eager] 0.8228ms 0.5170ms 1.9344 KOps/s 1.9249 KOps/s $\color{#35bf28}+0.49\%$
test_compile_add_self_flat[tensordict-eager] 0.4246ms 0.3014ms 3.3179 KOps/s 3.2628 KOps/s $\color{#35bf28}+1.69\%$
test_compile_add_self_flat[tensordict-compile] 0.2454ms 0.1755ms 5.6969 KOps/s 5.2617 KOps/s $\textbf{\color{#35bf28}+8.27\%}$
test_compile_add_self_flat[tensorclass-eager] 0.1313ms 84.8197μs 11.7897 KOps/s 11.8189 KOps/s $\color{#d91a1a}-0.25\%$
test_compile_add_self_flat[tensorclass-compile] 0.1927ms 0.1178ms 8.4885 KOps/s 8.1486 KOps/s $\color{#35bf28}+4.17\%$
test_compile_add_self_flat[pytree-eager] 0.1752s 0.5016ms 1.9935 KOps/s 2.3021 KOps/s $\textbf{\color{#d91a1a}-13.41\%}$
test_compile_add_self_flat[pytree-compile] 0.2371ms 0.1580ms 6.3296 KOps/s 6.2330 KOps/s $\color{#35bf28}+1.55\%$
test_compile_copy_flat[tensordict-compile] 56.4300μs 23.2837μs 42.9485 KOps/s 39.4497 KOps/s $\textbf{\color{#35bf28}+8.87\%}$
test_compile_copy_flat[tensordict-eager] 72.6910μs 39.6122μs 25.2448 KOps/s 24.4976 KOps/s $\color{#35bf28}+3.05\%$
test_compile_copy_flat[pytree-compile] 0.1019ms 10.5306μs 94.9614 KOps/s 95.3688 KOps/s $\color{#d91a1a}-0.43\%$
test_compile_copy_flat[pytree-eager] 0.3979ms 50.9612μs 19.6228 KOps/s 19.5739 KOps/s $\color{#35bf28}+0.25\%$
test_compile_assign_and_add[tensordict-compile] 1.9456ms 0.1696ms 5.8948 KOps/s 5.4648 KOps/s $\textbf{\color{#35bf28}+7.87\%}$
test_compile_assign_and_add[tensordict-eager] 3.4443ms 3.2457ms 308.0961 Ops/s 305.2397 Ops/s $\color{#35bf28}+0.94\%$
test_compile_assign_and_add[pytree-compile] 1.9076ms 0.1575ms 6.3489 KOps/s 6.1583 KOps/s $\color{#35bf28}+3.09\%$
test_compile_assign_and_add[pytree-eager] 2.8915ms 2.7447ms 364.3406 Ops/s 361.9130 Ops/s $\color{#35bf28}+0.67\%$
test_compile_indexing[tensor-tensordict-compile] 0.1791ms 0.1060ms 9.4300 KOps/s 8.9639 KOps/s $\textbf{\color{#35bf28}+5.20\%}$
test_compile_indexing[tensor-tensordict-eager] 0.3115ms 71.6452μs 13.9577 KOps/s 14.0317 KOps/s $\color{#d91a1a}-0.53\%$
test_compile_indexing[tensor-tensorclass-compile] 0.1503ms 93.8617μs 10.6540 KOps/s 10.3919 KOps/s $\color{#35bf28}+2.52\%$
test_compile_indexing[tensor-tensorclass-eager] 0.2588ms 47.1844μs 21.1934 KOps/s 22.5673 KOps/s $\textbf{\color{#d91a1a}-6.09\%}$
test_compile_indexing[tensor-pytree-compile] 0.1648ms 96.7259μs 10.3385 KOps/s 10.1190 KOps/s $\color{#35bf28}+2.17\%$
test_compile_indexing[tensor-pytree-eager] 0.2875ms 46.9233μs 21.3114 KOps/s 22.3963 KOps/s $\color{#d91a1a}-4.84\%$
test_compile_indexing[slice-tensordict-compile] 0.1055ms 55.9236μs 17.8815 KOps/s 17.7520 KOps/s $\color{#35bf28}+0.73\%$
test_compile_indexing[slice-tensordict-eager] 0.2266ms 27.1102μs 36.8864 KOps/s 36.9063 KOps/s $\color{#d91a1a}-0.05\%$
test_compile_indexing[slice-tensorclass-compile] 83.7310μs 44.1433μs 22.6535 KOps/s 22.6185 KOps/s $\color{#35bf28}+0.15\%$
test_compile_indexing[slice-tensorclass-eager] 0.2640ms 21.8560μs 45.7541 KOps/s 45.0940 KOps/s $\color{#35bf28}+1.46\%$
test_compile_indexing[slice-pytree-compile] 93.3310μs 45.6705μs 21.8960 KOps/s 22.1292 KOps/s $\color{#d91a1a}-1.05\%$
test_compile_indexing[slice-pytree-eager] 0.3076ms 21.9364μs 45.5863 KOps/s 45.4992 KOps/s $\color{#35bf28}+0.19\%$
test_compile_indexing[int-tensordict-compile] 0.1014ms 56.1394μs 17.8128 KOps/s 17.2712 KOps/s $\color{#35bf28}+3.14\%$
test_compile_indexing[int-tensordict-eager] 0.3025ms 26.9764μs 37.0694 KOps/s 37.7781 KOps/s $\color{#d91a1a}-1.88\%$
test_compile_indexing[int-tensorclass-compile] 89.4010μs 44.8188μs 22.3121 KOps/s 21.3597 KOps/s $\color{#35bf28}+4.46\%$
test_compile_indexing[int-tensorclass-eager] 0.2562ms 21.7083μs 46.0654 KOps/s 45.6185 KOps/s $\color{#35bf28}+0.98\%$
test_compile_indexing[int-pytree-compile] 96.3220μs 45.4794μs 21.9880 KOps/s 22.0419 KOps/s $\color{#d91a1a}-0.24\%$
test_compile_indexing[int-pytree-eager] 0.2581ms 21.5801μs 46.3390 KOps/s 45.8095 KOps/s $\color{#35bf28}+1.16\%$
test_mod_add[eager] 95.1720μs 49.5202μs 20.1938 KOps/s 19.8506 KOps/s $\color{#35bf28}+1.73\%$
test_mod_add[compile] 0.1762ms 0.1013ms 9.8688 KOps/s 9.6024 KOps/s $\color{#35bf28}+2.77\%$
test_mod_add[compile-overhead] 0.3331ms 0.1452ms 6.8867 KOps/s 6.8018 KOps/s $\color{#35bf28}+1.25\%$
test_mod_wrap[eager] 0.3798ms 0.2972ms 3.3651 KOps/s 3.4786 KOps/s $\color{#d91a1a}-3.26\%$
test_mod_wrap[compile] 0.4128ms 0.3575ms 2.7971 KOps/s 2.7810 KOps/s $\color{#35bf28}+0.58\%$
test_mod_wrap[compile-overhead] 7.4676ms 4.0879ms 244.6240 Ops/s 255.1583 Ops/s $\color{#d91a1a}-4.13\%$
test_mod_wrap_and_backward[eager] 1.5771ms 1.4658ms 682.2419 Ops/s 673.2628 Ops/s $\color{#35bf28}+1.33\%$
test_mod_wrap_and_backward[compile] 1.4749ms 1.4146ms 706.9083 Ops/s 699.9132 Ops/s $\color{#35bf28}+1.00\%$
test_mod_wrap_and_backward[compile-overhead] 1.2093ms 0.8553ms 1.1691 KOps/s 1.1356 KOps/s $\color{#35bf28}+2.96\%$
test_seq_add[eager] 0.2099ms 0.1527ms 6.5481 KOps/s 6.4373 KOps/s $\color{#35bf28}+1.72\%$
test_seq_add[compile] 0.1941ms 0.1112ms 8.9894 KOps/s 8.6141 KOps/s $\color{#35bf28}+4.36\%$
test_seq_add[compile-overhead] 0.3907ms 0.1561ms 6.4064 KOps/s 6.4572 KOps/s $\color{#d91a1a}-0.79\%$
test_seq_wrap[eager] 0.6227ms 0.5334ms 1.8747 KOps/s 1.9188 KOps/s $\color{#d91a1a}-2.30\%$
test_seq_wrap[compile] 0.4489ms 0.3581ms 2.7924 KOps/s 2.7482 KOps/s $\color{#35bf28}+1.61\%$
test_seq_wrap[compile-overhead] 0.3224ms 0.2567ms 3.8962 KOps/s 3.8241 KOps/s $\color{#35bf28}+1.89\%$
test_func_call_runtime[False-eager] 0.8897ms 0.8210ms 1.2181 KOps/s 1.2039 KOps/s $\color{#35bf28}+1.18\%$
test_func_call_runtime[False-compile] 1.0744ms 0.8842ms 1.1310 KOps/s 1.1188 KOps/s $\color{#35bf28}+1.09\%$
test_func_call_runtime[False-compile-overhead] 0.5781ms 0.4391ms 2.2774 KOps/s 2.2348 KOps/s $\color{#35bf28}+1.90\%$
test_func_call_runtime[True-eager] 1.1339ms 1.0480ms 954.2052 Ops/s 934.1078 Ops/s $\color{#35bf28}+2.15\%$
test_func_call_runtime[True-compile] 0.9724ms 0.8962ms 1.1158 KOps/s 1.0964 KOps/s $\color{#35bf28}+1.78\%$
test_func_call_runtime[True-compile-overhead] 0.5279ms 0.4545ms 2.2002 KOps/s 2.1505 KOps/s $\color{#35bf28}+2.31\%$
test_func_call_cm_runtime[False-eager] 0.8776ms 0.8182ms 1.2222 KOps/s 1.2077 KOps/s $\color{#35bf28}+1.20\%$
test_func_call_cm_runtime[False-compile] 0.9996ms 0.8838ms 1.1315 KOps/s 1.1083 KOps/s $\color{#35bf28}+2.10\%$
test_func_call_cm_runtime[False-compile-overhead] 0.5311ms 0.4432ms 2.2566 KOps/s 2.2266 KOps/s $\color{#35bf28}+1.34\%$
test_func_call_cm_runtime[True-eager] 1.2725ms 1.1884ms 841.4352 Ops/s 825.3611 Ops/s $\color{#35bf28}+1.95\%$
test_func_call_cm_runtime[True-compile] 0.9783ms 0.9302ms 1.0751 KOps/s 1.0573 KOps/s $\color{#35bf28}+1.68\%$
test_func_call_cm_runtime[True-compile-overhead] 0.5412ms 0.4847ms 2.0632 KOps/s 2.0028 KOps/s $\color{#35bf28}+3.02\%$
test_vmap_func_call_cm_runtime[eager] 2.8087ms 2.3129ms 432.3565 Ops/s 425.6697 Ops/s $\color{#35bf28}+1.57\%$
test_vmap_func_call_cm_runtime[compile] 1.0552ms 0.9462ms 1.0569 KOps/s 1.0427 KOps/s $\color{#35bf28}+1.36\%$
test_vmap_func_call_cm_runtime[compile-overhead] 0.5607ms 0.4917ms 2.0337 KOps/s 1.9824 KOps/s $\color{#35bf28}+2.59\%$
test_distributed 0.6439ms 0.1515ms 6.5989 KOps/s 5.8941 KOps/s $\textbf{\color{#35bf28}+11.96\%}$
test_tdmodule 63.6510μs 27.2480μs 36.6999 KOps/s 35.7514 KOps/s $\color{#35bf28}+2.65\%$
test_tdmodule_dispatch 75.1510μs 44.9035μs 22.2700 KOps/s 21.9246 KOps/s $\color{#35bf28}+1.58\%$
test_tdseq 61.2810μs 27.2950μs 36.6367 KOps/s 36.3691 KOps/s $\color{#35bf28}+0.74\%$
test_tdseq_dispatch 75.5410μs 47.9090μs 20.8729 KOps/s 20.6665 KOps/s $\color{#35bf28}+1.00\%$
test_instantiation_functorch 2.0363ms 1.9700ms 507.6160 Ops/s 503.2743 Ops/s $\color{#35bf28}+0.86\%$
test_exec_functorch 0.2133ms 0.1732ms 5.7736 KOps/s 5.6417 KOps/s $\color{#35bf28}+2.34\%$
test_exec_functional_call 0.2252ms 0.1561ms 6.4050 KOps/s 6.0445 KOps/s $\textbf{\color{#35bf28}+5.96\%}$
test_exec_td_decorator 0.4275ms 0.2263ms 4.4187 KOps/s 4.1721 KOps/s $\textbf{\color{#35bf28}+5.91\%}$
test_vmap_mlp_speed_decorator[True-True] 0.9986ms 0.8033ms 1.2449 KOps/s 1.2246 KOps/s $\color{#35bf28}+1.66\%$
test_vmap_mlp_speed_decorator[True-False] 0.9723ms 0.8012ms 1.2481 KOps/s 1.2213 KOps/s $\color{#35bf28}+2.20\%$
test_vmap_mlp_speed_decorator[False-True] 0.8620ms 0.6940ms 1.4410 KOps/s 1.4035 KOps/s $\color{#35bf28}+2.67\%$
test_vmap_mlp_speed_decorator[False-False] 0.8658ms 0.6948ms 1.4393 KOps/s 1.4062 KOps/s $\color{#35bf28}+2.35\%$
test_vmap_transformer_speed_decorator[True-True] 20.2005ms 20.1107ms 49.7248 Ops/s 49.2897 Ops/s $\color{#35bf28}+0.88\%$
test_vmap_transformer_speed_decorator[True-False] 20.2547ms 20.1061ms 49.7362 Ops/s 49.0655 Ops/s $\color{#35bf28}+1.37\%$
test_vmap_transformer_speed_decorator[False-True] 20.0109ms 19.9133ms 50.2176 Ops/s 49.7220 Ops/s $\color{#35bf28}+1.00\%$
test_vmap_transformer_speed_decorator[False-False] 20.4869ms 19.9370ms 50.1581 Ops/s 49.6106 Ops/s $\color{#35bf28}+1.10\%$
test_to_module_speed[True] 2.0366ms 1.4162ms 706.0962 Ops/s 703.4613 Ops/s $\color{#35bf28}+0.37\%$
test_to_module_speed[False] 1.4800ms 1.3858ms 721.6142 Ops/s 715.4531 Ops/s $\color{#35bf28}+0.86\%$
test_tc_init 79.8310μs 45.1628μs 22.1421 KOps/s 22.0692 KOps/s $\color{#35bf28}+0.33\%$
test_tc_init_tensor_only 35.3200μs 9.4974μs 105.2922 KOps/s 106.7457 KOps/s $\color{#d91a1a}-1.36\%$
test_tc_init_nested 0.1222ms 89.3190μs 11.1958 KOps/s 11.1889 KOps/s $\color{#35bf28}+0.06\%$
test_tc_init_many_fields 42.5210μs 15.7644μs 63.4340 KOps/s 63.4791 KOps/s $\color{#d91a1a}-0.07\%$
test_tc_first_layer_tensor 26.0200μs 1.7597μs 568.2947 KOps/s 582.4694 KOps/s $\color{#d91a1a}-2.43\%$
test_tc_first_layer_tensor_only 4.6087μs 0.7151μs 1.3984 MOps/s 1.4110 MOps/s $\color{#d91a1a}-0.90\%$
test_tc_first_layer_tensor_set 31.6610μs 3.9940μs 250.3735 KOps/s 253.2787 KOps/s $\color{#d91a1a}-1.15\%$
test_tc_first_layer_tensor_only_set 20.3510μs 3.0142μs 331.7617 KOps/s 332.1530 KOps/s $\color{#d91a1a}-0.12\%$
test_tc_first_layer_nontensor 34.5600μs 5.9016μs 169.4454 KOps/s 172.3298 KOps/s $\color{#d91a1a}-1.67\%$
test_tc_second_layer_tensor 31.4510μs 4.1950μs 238.3763 KOps/s 244.7838 KOps/s $\color{#d91a1a}-2.62\%$
test_tc_second_layer_nontensor 37.0110μs 8.2862μs 120.6821 KOps/s 120.6607 KOps/s $\color{#35bf28}+0.02\%$
test_unbind 0.2439s 13.7011ms 72.9868 Ops/s 73.1016 Ops/s $\color{#d91a1a}-0.16\%$
test_full_like 5.2748ms 4.3724ms 228.7054 Ops/s 135.0083 Ops/s $\textbf{\color{#35bf28}+69.40\%}$
test_zeros_like 4.4868ms 4.3467ms 230.0599 Ops/s 229.6096 Ops/s $\color{#35bf28}+0.20\%$
test_ones_like 4.7553ms 4.3538ms 229.6845 Ops/s 229.3751 Ops/s $\color{#35bf28}+0.13\%$
test_clone 6.5632ms 6.3545ms 157.3693 Ops/s 156.7462 Ops/s $\color{#35bf28}+0.40\%$
test_squeeze 0.1995ms 14.0153μs 71.3507 KOps/s 71.7935 KOps/s $\color{#d91a1a}-0.62\%$
test_unsqueeze 0.2777ms 0.1109ms 9.0205 KOps/s 9.2932 KOps/s $\color{#d91a1a}-2.93\%$
test_split 0.2312ms 0.1764ms 5.6684 KOps/s 5.5045 KOps/s $\color{#35bf28}+2.98\%$
test_permute 0.2596ms 0.1985ms 5.0386 KOps/s 4.9910 KOps/s $\color{#35bf28}+0.95\%$
test_stack 51.3813ms 51.0873ms 19.5743 Ops/s 19.6871 Ops/s $\color{#d91a1a}-0.57\%$
test_cat 51.4907ms 51.0968ms 19.5707 Ops/s 18.9357 Ops/s $\color{#35bf28}+3.35\%$

@github-actions

github-actions Bot commented Feb 14, 2026

Copy link
Copy Markdown
Contributor

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

Total Benchmarks: 243. Improved: $\large\color{#35bf28}23$. Worsened: $\large\color{#d91a1a}12$.

Expand to view detailed results
Name Max Mean Ops Ops on Repo HEAD Change
test_plain_set_nested 44.1100μs 14.4409μs 69.2477 KOps/s 69.7311 KOps/s $\color{#d91a1a}-0.69\%$
test_plain_set_stack_nested 48.1310μs 14.6201μs 68.3991 KOps/s 68.2213 KOps/s $\color{#35bf28}+0.26\%$
test_plain_set_nested_inplace 52.4410μs 15.9620μs 62.6487 KOps/s 62.8121 KOps/s $\color{#d91a1a}-0.26\%$
test_plain_set_stack_nested_inplace 43.1000μs 15.9021μs 62.8846 KOps/s 62.4034 KOps/s $\color{#35bf28}+0.77\%$
test_items 35.9600μs 5.5509μs 180.1495 KOps/s 180.4080 KOps/s $\color{#d91a1a}-0.14\%$
test_items_nested 0.5677ms 0.5190ms 1.9268 KOps/s 2.0033 KOps/s $\color{#d91a1a}-3.82\%$
test_items_nested_locked 0.6028ms 0.5180ms 1.9305 KOps/s 1.9638 KOps/s $\color{#d91a1a}-1.70\%$
test_items_nested_leaf 0.1205ms 91.4170μs 10.9389 KOps/s 10.9052 KOps/s $\color{#35bf28}+0.31\%$
test_items_stack_nested 0.5799ms 0.5243ms 1.9073 KOps/s 1.9622 KOps/s $\color{#d91a1a}-2.80\%$
test_items_stack_nested_leaf 0.1316ms 91.6622μs 10.9096 KOps/s 11.0298 KOps/s $\color{#d91a1a}-1.09\%$
test_items_stack_nested_locked 0.6349ms 0.5266ms 1.8989 KOps/s 1.9447 KOps/s $\color{#d91a1a}-2.36\%$
test_keys 37.1900μs 4.1497μs 240.9784 KOps/s 241.2259 KOps/s $\color{#d91a1a}-0.10\%$
test_keys_nested 0.1476ms 0.1172ms 8.5301 KOps/s 8.4791 KOps/s $\color{#35bf28}+0.60\%$
test_keys_nested_locked 88.7388ms 0.1377ms 7.2627 KOps/s 7.9455 KOps/s $\textbf{\color{#d91a1a}-8.59\%}$
test_keys_nested_leaf 0.1451ms 0.1079ms 9.2707 KOps/s 9.2694 KOps/s $\color{#35bf28}+0.01\%$
test_keys_stack_nested 0.1635ms 0.1181ms 8.4656 KOps/s 8.4829 KOps/s $\color{#d91a1a}-0.20\%$
test_keys_stack_nested_leaf 0.1530ms 0.1097ms 9.1153 KOps/s 9.2515 KOps/s $\color{#d91a1a}-1.47\%$
test_keys_stack_nested_locked 0.1761ms 0.1258ms 7.9494 KOps/s 7.9599 KOps/s $\color{#d91a1a}-0.13\%$
test_values 6.2302μs 1.0004μs 999.5676 KOps/s 1.0114 MOps/s $\color{#d91a1a}-1.17\%$
test_values_nested 76.1720μs 46.3432μs 21.5781 KOps/s 21.4768 KOps/s $\color{#35bf28}+0.47\%$
test_values_nested_locked 0.1039ms 49.1373μs 20.3511 KOps/s 20.2464 KOps/s $\color{#35bf28}+0.52\%$
test_values_nested_leaf 85.6610μs 52.0554μs 19.2103 KOps/s 18.8525 KOps/s $\color{#35bf28}+1.90\%$
test_values_stack_nested 91.9720μs 46.5951μs 21.4615 KOps/s 21.4799 KOps/s $\color{#d91a1a}-0.09\%$
test_values_stack_nested_leaf 80.0420μs 52.3870μs 19.0887 KOps/s 18.9385 KOps/s $\color{#35bf28}+0.79\%$
test_values_stack_nested_locked 82.3120μs 49.0373μs 20.3926 KOps/s 20.2728 KOps/s $\color{#35bf28}+0.59\%$
test_membership 5.4850μs 0.7945μs 1.2586 MOps/s 1.2485 MOps/s $\color{#35bf28}+0.81\%$
test_membership_nested 33.7210μs 2.9772μs 335.8880 KOps/s 335.6085 KOps/s $\color{#35bf28}+0.08\%$
test_membership_nested_leaf 32.1110μs 2.9725μs 336.4216 KOps/s 334.4789 KOps/s $\color{#35bf28}+0.58\%$
test_membership_stacked_nested 27.6600μs 2.9991μs 333.4357 KOps/s 334.2539 KOps/s $\color{#d91a1a}-0.24\%$
test_membership_stacked_nested_leaf 26.3010μs 2.9886μs 334.6000 KOps/s 339.1020 KOps/s $\color{#d91a1a}-1.33\%$
test_membership_nested_last 37.1610μs 4.3828μs 228.1641 KOps/s 232.8454 KOps/s $\color{#d91a1a}-2.01\%$
test_membership_nested_leaf_last 34.2400μs 4.3692μs 228.8773 KOps/s 229.6747 KOps/s $\color{#d91a1a}-0.35\%$
test_membership_stacked_nested_last 21.2800μs 4.3477μs 230.0062 KOps/s 229.6230 KOps/s $\color{#35bf28}+0.17\%$
test_membership_stacked_nested_leaf_last 63.6920μs 4.3245μs 231.2387 KOps/s 229.8359 KOps/s $\color{#35bf28}+0.61\%$
test_nested_getleaf 49.8910μs 20.6552μs 48.4139 KOps/s 47.9498 KOps/s $\color{#35bf28}+0.97\%$
test_nested_get 68.9510μs 19.3340μs 51.7224 KOps/s 50.0782 KOps/s $\color{#35bf28}+3.28\%$
test_stacked_getleaf 59.3310μs 20.8749μs 47.9045 KOps/s 48.4168 KOps/s $\color{#d91a1a}-1.06\%$
test_stacked_get 52.4910μs 19.6760μs 50.8234 KOps/s 50.4293 KOps/s $\color{#35bf28}+0.78\%$
test_nested_getitemleaf 44.0300μs 21.4163μs 46.6934 KOps/s 46.4630 KOps/s $\color{#35bf28}+0.50\%$
test_nested_getitem 43.8210μs 20.0806μs 49.7994 KOps/s 49.7187 KOps/s $\color{#35bf28}+0.16\%$
test_stacked_getitemleaf 54.5910μs 21.1989μs 47.1723 KOps/s 46.4984 KOps/s $\color{#35bf28}+1.45\%$
test_stacked_getitem 0.4060ms 19.9304μs 50.1746 KOps/s 49.3602 KOps/s $\color{#35bf28}+1.65\%$
test_lock_nested 7.9294ms 0.4570ms 2.1881 KOps/s 2.2061 KOps/s $\color{#d91a1a}-0.82\%$
test_lock_stack_nested 0.5038ms 0.4546ms 2.1998 KOps/s 2.1890 KOps/s $\color{#35bf28}+0.49\%$
test_unlock_nested 0.4613ms 0.3629ms 2.7554 KOps/s 2.7393 KOps/s $\color{#35bf28}+0.59\%$
test_unlock_stack_nested 0.4375ms 0.3667ms 2.7267 KOps/s 2.6975 KOps/s $\color{#35bf28}+1.08\%$
test_flatten_speed 0.2021ms 0.1158ms 8.6346 KOps/s 8.6290 KOps/s $\color{#35bf28}+0.07\%$
test_unflatten_speed 0.6565ms 0.5621ms 1.7789 KOps/s 1.7763 KOps/s $\color{#35bf28}+0.15\%$
test_common_ops 0.8037ms 0.6759ms 1.4796 KOps/s 1.4781 KOps/s $\color{#35bf28}+0.10\%$
test_creation 78.8320μs 2.7578μs 362.6051 KOps/s 372.7543 KOps/s $\color{#d91a1a}-2.72\%$
test_creation_empty 31.2400μs 5.9863μs 167.0482 KOps/s 175.2116 KOps/s $\color{#d91a1a}-4.66\%$
test_creation_nested_1 42.5310μs 10.5127μs 95.1227 KOps/s 98.4594 KOps/s $\color{#d91a1a}-3.39\%$
test_creation_nested_2 41.2710μs 11.6375μs 85.9288 KOps/s 89.8675 KOps/s $\color{#d91a1a}-4.38\%$
test_creation_many_keys[10] 52.7110μs 17.1824μs 58.1991 KOps/s 59.0217 KOps/s $\color{#d91a1a}-1.39\%$
test_creation_many_keys[50] 0.1057ms 73.2274μs 13.6561 KOps/s 13.8889 KOps/s $\color{#d91a1a}-1.68\%$
test_creation_many_keys[100] 0.1936ms 0.1428ms 7.0030 KOps/s 7.1594 KOps/s $\color{#d91a1a}-2.18\%$
test_creation_nested_many_keys[10] 72.0110μs 37.1962μs 26.8844 KOps/s 27.7959 KOps/s $\color{#d91a1a}-3.28\%$
test_creation_nested_many_keys[50] 0.1934ms 0.1500ms 6.6662 KOps/s 6.8175 KOps/s $\color{#d91a1a}-2.22\%$
test_clone 37.7000μs 12.8314μs 77.9340 KOps/s 76.0836 KOps/s $\color{#35bf28}+2.43\%$
test_getitem[int] 1.7942ms 14.0076μs 71.3896 KOps/s 58.3531 KOps/s $\textbf{\color{#35bf28}+22.34\%}$
test_getitem[slice_int] 0.1404ms 23.8320μs 41.9604 KOps/s 41.9487 KOps/s $\color{#35bf28}+0.03\%$
test_getitem[range] 0.1700ms 59.6326μs 16.7694 KOps/s 16.6715 KOps/s $\color{#35bf28}+0.59\%$
test_getitem[tuple] 0.1437ms 23.1468μs 43.2026 KOps/s 43.1347 KOps/s $\color{#35bf28}+0.16\%$
test_getitem[list] 0.1742ms 54.5472μs 18.3328 KOps/s 17.0748 KOps/s $\textbf{\color{#35bf28}+7.37\%}$
test_setitem_dim[int] 43.7000μs 25.1355μs 39.7843 KOps/s 37.8639 KOps/s $\textbf{\color{#35bf28}+5.07\%}$
test_setitem_dim[slice_int] 82.7410μs 43.0041μs 23.2536 KOps/s 22.4445 KOps/s $\color{#35bf28}+3.60\%$
test_setitem_dim[range] 0.1377ms 90.4927μs 11.0506 KOps/s 10.5705 KOps/s $\color{#35bf28}+4.54\%$
test_setitem_dim[tuple] 62.9610μs 40.3157μs 24.8042 KOps/s 24.5827 KOps/s $\color{#35bf28}+0.90\%$
test_setitem 48.3810μs 17.4576μs 57.2816 KOps/s 55.4087 KOps/s $\color{#35bf28}+3.38\%$
test_set 43.2310μs 16.3622μs 61.1163 KOps/s 58.5309 KOps/s $\color{#35bf28}+4.42\%$
test_set_shared 0.5096ms 0.2025ms 4.9376 KOps/s 4.7671 KOps/s $\color{#35bf28}+3.58\%$
test_update 0.3637ms 21.5408μs 46.4234 KOps/s 44.1834 KOps/s $\textbf{\color{#35bf28}+5.07\%}$
test_update_nested 67.3210μs 33.3784μs 29.9595 KOps/s 30.0046 KOps/s $\color{#d91a1a}-0.15\%$
test_update__nested 0.4889ms 33.2621μs 30.0642 KOps/s 29.8187 KOps/s $\color{#35bf28}+0.82\%$
test_set_nested 48.0400μs 18.7143μs 53.4350 KOps/s 52.9053 KOps/s $\color{#35bf28}+1.00\%$
test_set_nested_new 65.0310μs 23.7481μs 42.1086 KOps/s 41.7898 KOps/s $\color{#35bf28}+0.76\%$
test_select 85.7120μs 40.7222μs 24.5566 KOps/s 24.6279 KOps/s $\color{#d91a1a}-0.29\%$
test_select_nested 0.1032ms 71.3926μs 14.0071 KOps/s 13.9995 KOps/s $\color{#35bf28}+0.05\%$
test_exclude_nested 0.1467ms 92.6975μs 10.7878 KOps/s 10.9497 KOps/s $\color{#d91a1a}-1.48\%$
test_empty[True] 0.4574ms 0.4190ms 2.3866 KOps/s 2.4082 KOps/s $\color{#d91a1a}-0.90\%$
test_empty[False] 7.7275μs 1.2565μs 795.8349 KOps/s 793.3449 KOps/s $\color{#35bf28}+0.31\%$
test_to 0.1013ms 69.9886μs 14.2880 KOps/s 14.0441 KOps/s $\color{#35bf28}+1.74\%$
test_to_nonblocking 0.1023ms 63.4415μs 15.7626 KOps/s 15.9486 KOps/s $\color{#d91a1a}-1.17\%$
test_unbind_speed 0.3730ms 0.3154ms 3.1705 KOps/s 3.2278 KOps/s $\color{#d91a1a}-1.77\%$
test_unbind_speed_stack0 0.3653ms 0.3089ms 3.2372 KOps/s 3.2590 KOps/s $\color{#d91a1a}-0.67\%$
test_unbind_speed_stack1 0.1037s 0.8770ms 1.1403 KOps/s 1.1298 KOps/s $\color{#35bf28}+0.92\%$
test_split 1.1481ms 1.0814ms 924.7442 Ops/s 927.5798 Ops/s $\color{#d91a1a}-0.31\%$
test_chunk 0.1053s 1.1655ms 858.0368 Ops/s 969.2281 Ops/s $\textbf{\color{#d91a1a}-11.47\%}$
test_to_cpu_blocking 28.0710ms 27.8055ms 35.9641 Ops/s 50.7541 Ops/s $\textbf{\color{#d91a1a}-29.14\%}$
test_to_cpu_global_sync 11.0723ms 10.9505ms 91.3197 Ops/s 91.2401 Ops/s $\color{#35bf28}+0.09\%$
test_to_cpu_event_sync 0.1149s 13.0188ms 76.8120 Ops/s 84.7320 Ops/s $\textbf{\color{#d91a1a}-9.35\%}$
test_to_cpu_default 12.0361ms 11.8043ms 84.7150 Ops/s 84.6958 Ops/s $\color{#35bf28}+0.02\%$
test_consolidate[False-None] 4.1002ms 3.9357ms 254.0823 Ops/s 229.2798 Ops/s $\textbf{\color{#35bf28}+10.82\%}$
test_consolidate[default-None] 2.3579ms 1.9126ms 522.8496 Ops/s 498.3273 Ops/s $\color{#35bf28}+4.92\%$
test_consolidate[reduce-overhead-None] 2.2864ms 1.8488ms 540.9058 Ops/s 519.5979 Ops/s $\color{#35bf28}+4.10\%$
test_consolidate_njt[False-None] 8.3883ms 7.9649ms 125.5509 Ops/s 121.1063 Ops/s $\color{#35bf28}+3.67\%$
test_to[False-False-None] 2.3735ms 1.9588ms 510.5246 Ops/s 497.0080 Ops/s $\color{#35bf28}+2.72\%$
test_to[True-False-None] 2.2592ms 1.8328ms 545.6202 Ops/s 538.0812 Ops/s $\color{#35bf28}+1.40\%$
test_to[within-False-None] 6.2184ms 5.8094ms 172.1356 Ops/s 168.8788 Ops/s $\color{#35bf28}+1.93\%$
test_to[True-default-None] 0.1792s 8.4417ms 118.4600 Ops/s 129.5884 Ops/s $\textbf{\color{#d91a1a}-8.59\%}$
test_to_njt[False-False-None] 8.3953ms 8.1864ms 122.1542 Ops/s 117.4720 Ops/s $\color{#35bf28}+3.99\%$
test_to_njt[True-False-None] 6.7046ms 6.5625ms 152.3813 Ops/s 140.5294 Ops/s $\textbf{\color{#35bf28}+8.43\%}$
test_to_njt[within-False-None] 14.7463ms 14.5785ms 68.5943 Ops/s 63.8017 Ops/s $\textbf{\color{#35bf28}+7.51\%}$
test_creation[device0] 0.3415ms 0.1133ms 8.8237 KOps/s 8.5791 KOps/s $\color{#35bf28}+2.85\%$
test_creation_from_tensor 0.4058ms 0.1122ms 8.9153 KOps/s 8.9277 KOps/s $\color{#d91a1a}-0.14\%$
test_add_one[memmap_tensor0] 0.2597ms 6.2829μs 159.1612 KOps/s 158.3481 KOps/s $\color{#35bf28}+0.51\%$
test_contiguous[memmap_tensor0] 26.1010μs 0.6130μs 1.6313 MOps/s 2.3276 MOps/s $\textbf{\color{#d91a1a}-29.91\%}$
test_stack[memmap_tensor0] 57.4210μs 4.3299μs 230.9502 KOps/s 223.6072 KOps/s $\color{#35bf28}+3.28\%$
test_memmaptd_index 1.1795ms 0.2515ms 3.9756 KOps/s 4.0091 KOps/s $\color{#d91a1a}-0.83\%$
test_memmaptd_index_astensor 0.4953ms 0.3411ms 2.9318 KOps/s 2.9426 KOps/s $\color{#d91a1a}-0.37\%$
test_memmaptd_index_op 0.9113ms 0.5863ms 1.7057 KOps/s 1.7310 KOps/s $\color{#d91a1a}-1.46\%$
test_serialize_model 0.1375s 0.1360s 7.3533 Ops/s 7.3226 Ops/s $\color{#35bf28}+0.42\%$
test_serialize_model_pickle 1.3642s 1.2132s 0.8242 Ops/s 0.8264 Ops/s $\color{#d91a1a}-0.26\%$
test_serialize_weights 0.1381s 0.1362s 7.3444 Ops/s 7.4026 Ops/s $\color{#d91a1a}-0.79\%$
test_serialize_weights_returnearly 0.4413s 88.8760ms 11.2516 Ops/s 10.6794 Ops/s $\textbf{\color{#35bf28}+5.36\%}$
test_serialize_weights_pickle 1.3505s 1.1932s 0.8381 Ops/s 0.8252 Ops/s $\color{#35bf28}+1.56\%$
test_reshape_pytree 0.2064ms 31.8668μs 31.3807 KOps/s 30.4535 KOps/s $\color{#35bf28}+3.04\%$
test_reshape_td 80.7920μs 41.7210μs 23.9687 KOps/s 23.4466 KOps/s $\color{#35bf28}+2.23\%$
test_view_pytree 0.2227ms 31.9861μs 31.2635 KOps/s 30.7117 KOps/s $\color{#35bf28}+1.80\%$
test_view_td 82.2510μs 49.9780μs 20.0088 KOps/s 19.8378 KOps/s $\color{#35bf28}+0.86\%$
test_unbind_pytree 0.2673ms 36.2049μs 27.6206 KOps/s 27.8583 KOps/s $\color{#d91a1a}-0.85\%$
test_unbind_td 86.1410μs 46.8505μs 21.3445 KOps/s 21.0980 KOps/s $\color{#35bf28}+1.17\%$
test_split_pytree 0.2467ms 41.2782μs 24.2259 KOps/s 24.1137 KOps/s $\color{#35bf28}+0.47\%$
test_split_td 0.1792ms 60.9927μs 16.3954 KOps/s 16.0202 KOps/s $\color{#35bf28}+2.34\%$
test_add_pytree 0.2329ms 41.3154μs 24.2041 KOps/s 24.2516 KOps/s $\color{#d91a1a}-0.20\%$
test_add_td 0.1219ms 53.7044μs 18.6204 KOps/s 19.5739 KOps/s $\color{#d91a1a}-4.87\%$
test_compile_add_one_nested[tensordict-compile] 0.1960ms 0.1352ms 7.3981 KOps/s 7.0566 KOps/s $\color{#35bf28}+4.84\%$
test_compile_add_one_nested[tensordict-eager] 0.3400ms 0.1866ms 5.3604 KOps/s 5.4642 KOps/s $\color{#d91a1a}-1.90\%$
test_compile_add_one_nested[pytree-compile] 0.1444ms 0.1058ms 9.4541 KOps/s 9.4002 KOps/s $\color{#35bf28}+0.57\%$
test_compile_add_one_nested[pytree-eager] 0.4350ms 0.1781ms 5.6139 KOps/s 5.8050 KOps/s $\color{#d91a1a}-3.29\%$
test_compile_copy_nested[tensordict-compile] 69.1610μs 30.4774μs 32.8111 KOps/s 33.3237 KOps/s $\color{#d91a1a}-1.54\%$
test_compile_copy_nested[tensordict-eager] 87.8710μs 49.8978μs 20.0410 KOps/s 20.0680 KOps/s $\color{#d91a1a}-0.13\%$
test_compile_copy_nested[pytree-compile] 45.8300μs 9.5856μs 104.3226 KOps/s 109.0117 KOps/s $\color{#d91a1a}-4.30\%$
test_compile_copy_nested[pytree-eager] 0.4616ms 67.5732μs 14.7988 KOps/s 15.0063 KOps/s $\color{#d91a1a}-1.38\%$
test_compile_add_one_flat[tensordict-compile] 0.2133ms 0.1731ms 5.7763 KOps/s 5.4891 KOps/s $\textbf{\color{#35bf28}+5.23\%}$
test_compile_add_one_flat[tensordict-eager] 0.3207ms 0.2497ms 4.0042 KOps/s 4.0295 KOps/s $\color{#d91a1a}-0.63\%$
test_compile_add_one_flat[tensorclass-compile] 0.1895ms 0.1138ms 8.7837 KOps/s 8.5367 KOps/s $\color{#35bf28}+2.89\%$
test_compile_add_one_flat[tensorclass-eager] 0.1165ms 68.4224μs 14.6151 KOps/s 15.0673 KOps/s $\color{#d91a1a}-3.00\%$
test_compile_add_one_flat[pytree-compile] 0.1912ms 0.1558ms 6.4202 KOps/s 6.3202 KOps/s $\color{#35bf28}+1.58\%$
test_compile_add_one_flat[pytree-eager] 0.8032ms 0.5128ms 1.9501 KOps/s 1.9396 KOps/s $\color{#35bf28}+0.54\%$
test_compile_add_self_flat[tensordict-eager] 0.4332ms 0.3014ms 3.3175 KOps/s 3.3021 KOps/s $\color{#35bf28}+0.47\%$
test_compile_add_self_flat[tensordict-compile] 0.2292ms 0.1743ms 5.7356 KOps/s 5.1870 KOps/s $\textbf{\color{#35bf28}+10.58\%}$
test_compile_add_self_flat[tensorclass-eager] 0.1353ms 84.5393μs 11.8288 KOps/s 11.7199 KOps/s $\color{#35bf28}+0.93\%$
test_compile_add_self_flat[tensorclass-compile] 0.2144ms 0.1162ms 8.6025 KOps/s 7.8758 KOps/s $\textbf{\color{#35bf28}+9.23\%}$
test_compile_add_self_flat[pytree-eager] 0.6251ms 0.4232ms 2.3627 KOps/s 2.3324 KOps/s $\color{#35bf28}+1.30\%$
test_compile_add_self_flat[pytree-compile] 0.2031ms 0.1557ms 6.4213 KOps/s 6.1952 KOps/s $\color{#35bf28}+3.65\%$
test_compile_copy_flat[tensordict-compile] 80.7710μs 23.4403μs 42.6616 KOps/s 39.9472 KOps/s $\textbf{\color{#35bf28}+6.80\%}$
test_compile_copy_flat[tensordict-eager] 72.5410μs 40.4488μs 24.7226 KOps/s 25.1747 KOps/s $\color{#d91a1a}-1.80\%$
test_compile_copy_flat[pytree-compile] 48.6710μs 10.5724μs 94.5861 KOps/s 94.7614 KOps/s $\color{#d91a1a}-0.19\%$
test_compile_copy_flat[pytree-eager] 0.4061ms 50.7638μs 19.6991 KOps/s 19.7842 KOps/s $\color{#d91a1a}-0.43\%$
test_compile_assign_and_add[tensordict-compile] 1.9340ms 0.1702ms 5.8748 KOps/s 5.5127 KOps/s $\textbf{\color{#35bf28}+6.57\%}$
test_compile_assign_and_add[tensordict-eager] 3.3262ms 3.1899ms 313.4874 Ops/s 309.8266 Ops/s $\color{#35bf28}+1.18\%$
test_compile_assign_and_add[pytree-compile] 1.9048ms 0.1576ms 6.3451 KOps/s 6.2010 KOps/s $\color{#35bf28}+2.32\%$
test_compile_assign_and_add[pytree-eager] 3.0895ms 2.9185ms 342.6379 Ops/s 369.2119 Ops/s $\textbf{\color{#d91a1a}-7.20\%}$
test_compile_indexing[tensor-tensordict-compile] 0.1551ms 0.1067ms 9.3725 KOps/s 9.1327 KOps/s $\color{#35bf28}+2.63\%$
test_compile_indexing[tensor-tensordict-eager] 0.3072ms 70.0204μs 14.2816 KOps/s 14.2754 KOps/s $\color{#35bf28}+0.04\%$
test_compile_indexing[tensor-tensorclass-compile] 0.1419ms 94.6891μs 10.5609 KOps/s 10.4367 KOps/s $\color{#35bf28}+1.19\%$
test_compile_indexing[tensor-tensorclass-eager] 0.2506ms 47.0253μs 21.2651 KOps/s 22.7669 KOps/s $\textbf{\color{#d91a1a}-6.60\%}$
test_compile_indexing[tensor-pytree-compile] 0.1465ms 94.8281μs 10.5454 KOps/s 10.3035 KOps/s $\color{#35bf28}+2.35\%$
test_compile_indexing[tensor-pytree-eager] 0.2409ms 47.0814μs 21.2398 KOps/s 22.7084 KOps/s $\textbf{\color{#d91a1a}-6.47\%}$
test_compile_indexing[slice-tensordict-compile] 0.1436ms 54.5695μs 18.3253 KOps/s 17.9894 KOps/s $\color{#35bf28}+1.87\%$
test_compile_indexing[slice-tensordict-eager] 0.2284ms 26.5893μs 37.6092 KOps/s 37.8923 KOps/s $\color{#d91a1a}-0.75\%$
test_compile_indexing[slice-tensorclass-compile] 81.3920μs 43.7703μs 22.8465 KOps/s 22.2952 KOps/s $\color{#35bf28}+2.47\%$
test_compile_indexing[slice-tensorclass-eager] 0.2487ms 21.9710μs 45.5145 KOps/s 45.8997 KOps/s $\color{#d91a1a}-0.84\%$
test_compile_indexing[slice-pytree-compile] 82.6810μs 45.0225μs 22.2111 KOps/s 22.2951 KOps/s $\color{#d91a1a}-0.38\%$
test_compile_indexing[slice-pytree-eager] 0.2738ms 21.8808μs 45.7022 KOps/s 46.1718 KOps/s $\color{#d91a1a}-1.02\%$
test_compile_indexing[int-tensordict-compile] 93.7610μs 55.3989μs 18.0509 KOps/s 17.8236 KOps/s $\color{#35bf28}+1.28\%$
test_compile_indexing[int-tensordict-eager] 0.2751ms 26.7632μs 37.3647 KOps/s 38.1784 KOps/s $\color{#d91a1a}-2.13\%$
test_compile_indexing[int-tensorclass-compile] 81.7810μs 44.0383μs 22.7075 KOps/s 22.4285 KOps/s $\color{#35bf28}+1.24\%$
test_compile_indexing[int-tensorclass-eager] 0.2553ms 22.1385μs 45.1701 KOps/s 45.9513 KOps/s $\color{#d91a1a}-1.70\%$
test_compile_indexing[int-pytree-compile] 94.4510μs 45.1641μs 22.1415 KOps/s 22.4089 KOps/s $\color{#d91a1a}-1.19\%$
test_compile_indexing[int-pytree-eager] 0.2747ms 22.1312μs 45.1851 KOps/s 46.3694 KOps/s $\color{#d91a1a}-2.55\%$
test_mod_add[eager] 91.7610μs 49.3930μs 20.2458 KOps/s 19.8077 KOps/s $\color{#35bf28}+2.21\%$
test_mod_add[compile] 0.1763ms 0.1015ms 9.8491 KOps/s 9.6175 KOps/s $\color{#35bf28}+2.41\%$
test_mod_add[compile-overhead] 0.2353ms 0.1464ms 6.8327 KOps/s 6.7863 KOps/s $\color{#35bf28}+0.68\%$
test_mod_wrap[eager] 0.3969ms 0.2826ms 3.5380 KOps/s 3.4898 KOps/s $\color{#35bf28}+1.38\%$
test_mod_wrap[compile] 0.4693ms 0.3426ms 2.9189 KOps/s 2.9368 KOps/s $\color{#d91a1a}-0.61\%$
test_mod_wrap[compile-overhead] 6.9407ms 3.8055ms 262.7775 Ops/s 253.2719 Ops/s $\color{#35bf28}+3.75\%$
test_mod_wrap_and_backward[eager] 1.6391ms 1.4964ms 668.2828 Ops/s 621.3236 Ops/s $\textbf{\color{#35bf28}+7.56\%}$
test_mod_wrap_and_backward[compile] 1.5528ms 1.4219ms 703.2835 Ops/s 655.9612 Ops/s $\textbf{\color{#35bf28}+7.21\%}$
test_mod_wrap_and_backward[compile-overhead] 1.2215ms 0.8633ms 1.1583 KOps/s 1.0145 KOps/s $\textbf{\color{#35bf28}+14.17\%}$
test_seq_add[eager] 0.2294ms 0.1504ms 6.6481 KOps/s 6.1954 KOps/s $\textbf{\color{#35bf28}+7.31\%}$
test_seq_add[compile] 0.5764ms 0.1108ms 9.0262 KOps/s 8.5447 KOps/s $\textbf{\color{#35bf28}+5.64\%}$
test_seq_add[compile-overhead] 0.1933ms 0.1488ms 6.7222 KOps/s 6.4822 KOps/s $\color{#35bf28}+3.70\%$
test_seq_wrap[eager] 0.5831ms 0.5309ms 1.8835 KOps/s 1.9418 KOps/s $\color{#d91a1a}-3.00\%$
test_seq_wrap[compile] 0.5896ms 0.3625ms 2.7588 KOps/s 2.7819 KOps/s $\color{#d91a1a}-0.83\%$
test_seq_wrap[compile-overhead] 0.3183ms 0.2563ms 3.9020 KOps/s 3.8437 KOps/s $\color{#35bf28}+1.52\%$
test_func_call_runtime[False-eager] 0.8830ms 0.8116ms 1.2321 KOps/s 1.2227 KOps/s $\color{#35bf28}+0.77\%$
test_func_call_runtime[False-compile] 0.9231ms 0.8671ms 1.1533 KOps/s 1.0922 KOps/s $\textbf{\color{#35bf28}+5.60\%}$
test_func_call_runtime[False-compile-overhead] 0.5026ms 0.4395ms 2.2754 KOps/s 2.2354 KOps/s $\color{#35bf28}+1.79\%$
test_func_call_runtime[True-eager] 1.1004ms 1.0455ms 956.4752 Ops/s 945.9077 Ops/s $\color{#35bf28}+1.12\%$
test_func_call_runtime[True-compile] 0.9262ms 0.8768ms 1.1404 KOps/s 1.0802 KOps/s $\textbf{\color{#35bf28}+5.58\%}$
test_func_call_runtime[True-compile-overhead] 0.5570ms 0.4520ms 2.2123 KOps/s 2.1631 KOps/s $\color{#35bf28}+2.28\%$
test_func_call_cm_runtime[False-eager] 0.9410ms 0.8127ms 1.2305 KOps/s 1.2229 KOps/s $\color{#35bf28}+0.62\%$
test_func_call_cm_runtime[False-compile] 0.9843ms 0.8708ms 1.1484 KOps/s 1.1254 KOps/s $\color{#35bf28}+2.04\%$
test_func_call_cm_runtime[False-compile-overhead] 0.5513ms 0.4412ms 2.2666 KOps/s 2.2212 KOps/s $\color{#35bf28}+2.04\%$
test_func_call_cm_runtime[True-eager] 1.3122ms 1.1958ms 836.2594 Ops/s 835.3807 Ops/s $\color{#35bf28}+0.11\%$
test_func_call_cm_runtime[True-compile] 1.0247ms 0.9113ms 1.0973 KOps/s 1.0660 KOps/s $\color{#35bf28}+2.94\%$
test_func_call_cm_runtime[True-compile-overhead] 0.5945ms 0.4834ms 2.0689 KOps/s 2.0192 KOps/s $\color{#35bf28}+2.46\%$
test_vmap_func_call_cm_runtime[eager] 2.7938ms 2.2951ms 435.7174 Ops/s 428.8331 Ops/s $\color{#35bf28}+1.61\%$
test_vmap_func_call_cm_runtime[compile] 1.0986ms 0.9382ms 1.0659 KOps/s 1.0449 KOps/s $\color{#35bf28}+2.01\%$
test_vmap_func_call_cm_runtime[compile-overhead] 0.6151ms 0.4994ms 2.0025 KOps/s 1.9913 KOps/s $\color{#35bf28}+0.56\%$
test_distributed 3.1324ms 0.1656ms 6.0399 KOps/s 6.5210 KOps/s $\textbf{\color{#d91a1a}-7.38\%}$
test_tdmodule 62.8810μs 26.9504μs 37.1052 KOps/s 35.5233 KOps/s $\color{#35bf28}+4.45\%$
test_tdmodule_dispatch 0.1388ms 43.9014μs 22.7783 KOps/s 21.9113 KOps/s $\color{#35bf28}+3.96\%$
test_tdseq 55.7310μs 26.5328μs 37.6892 KOps/s 36.8830 KOps/s $\color{#35bf28}+2.19\%$
test_tdseq_dispatch 78.7310μs 46.5347μs 21.4893 KOps/s 21.0599 KOps/s $\color{#35bf28}+2.04\%$
test_instantiation_functorch 2.0283ms 1.9460ms 513.8637 Ops/s 504.3468 Ops/s $\color{#35bf28}+1.89\%$
test_exec_functorch 0.2318ms 0.1684ms 5.9388 KOps/s 5.6603 KOps/s $\color{#35bf28}+4.92\%$
test_exec_functional_call 0.2181ms 0.1525ms 6.5593 KOps/s 6.2551 KOps/s $\color{#35bf28}+4.86\%$
test_exec_td_decorator 0.4261ms 0.2235ms 4.4739 KOps/s 4.3107 KOps/s $\color{#35bf28}+3.79\%$
test_vmap_mlp_speed_decorator[True-True] 0.9827ms 0.7957ms 1.2568 KOps/s 1.2275 KOps/s $\color{#35bf28}+2.38\%$
test_vmap_mlp_speed_decorator[True-False] 0.9670ms 0.7948ms 1.2583 KOps/s 1.2315 KOps/s $\color{#35bf28}+2.17\%$
test_vmap_mlp_speed_decorator[False-True] 0.8490ms 0.6881ms 1.4534 KOps/s 1.4232 KOps/s $\color{#35bf28}+2.12\%$
test_vmap_mlp_speed_decorator[False-False] 0.8473ms 0.6889ms 1.4517 KOps/s 1.4292 KOps/s $\color{#35bf28}+1.57\%$
test_vmap_transformer_speed_decorator[True-True] 20.5874ms 19.9220ms 50.1959 Ops/s 49.4385 Ops/s $\color{#35bf28}+1.53\%$
test_vmap_transformer_speed_decorator[True-False] 20.6722ms 19.9521ms 50.1200 Ops/s 49.3802 Ops/s $\color{#35bf28}+1.50\%$
test_vmap_transformer_speed_decorator[False-True] 26.2695ms 20.0001ms 49.9999 Ops/s 49.9363 Ops/s $\color{#35bf28}+0.13\%$
test_vmap_transformer_speed_decorator[False-False] 19.8716ms 19.7595ms 50.6085 Ops/s 49.8765 Ops/s $\color{#35bf28}+1.47\%$
test_to_module_speed[True] 1.4739ms 1.3999ms 714.3240 Ops/s 709.1214 Ops/s $\color{#35bf28}+0.73\%$
test_to_module_speed[False] 1.9776ms 1.3822ms 723.4881 Ops/s 721.7826 Ops/s $\color{#35bf28}+0.24\%$
test_tc_init 68.5910μs 44.1990μs 22.6249 KOps/s 22.0324 KOps/s $\color{#35bf28}+2.69\%$
test_tc_init_tensor_only 45.5210μs 9.3022μs 107.5010 KOps/s 106.7572 KOps/s $\color{#35bf28}+0.70\%$
test_tc_init_nested 0.1260ms 87.8840μs 11.3786 KOps/s 11.2255 KOps/s $\color{#35bf28}+1.36\%$
test_tc_init_many_fields 39.8210μs 15.6616μs 63.8506 KOps/s 63.1082 KOps/s $\color{#35bf28}+1.18\%$
test_tc_first_layer_tensor 22.0900μs 1.7323μs 577.2801 KOps/s 575.8412 KOps/s $\color{#35bf28}+0.25\%$
test_tc_first_layer_tensor_only 4.1143μs 0.7213μs 1.3864 MOps/s 1.3814 MOps/s $\color{#35bf28}+0.36\%$
test_tc_first_layer_tensor_set 22.3000μs 3.9678μs 252.0268 KOps/s 252.5640 KOps/s $\color{#d91a1a}-0.21\%$
test_tc_first_layer_tensor_only_set 25.4210μs 2.9993μs 333.4123 KOps/s 333.9313 KOps/s $\color{#d91a1a}-0.16\%$
test_tc_first_layer_nontensor 38.5300μs 5.8400μs 171.2324 KOps/s 164.6658 KOps/s $\color{#35bf28}+3.99\%$
test_tc_second_layer_tensor 37.4900μs 4.1461μs 241.1882 KOps/s 238.3074 KOps/s $\color{#35bf28}+1.21\%$
test_tc_second_layer_nontensor 38.8010μs 8.2588μs 121.0831 KOps/s 117.5601 KOps/s $\color{#35bf28}+3.00\%$
test_unbind 0.2784s 16.0235ms 62.4082 Ops/s 59.3109 Ops/s $\textbf{\color{#35bf28}+5.22\%}$
test_full_like 16.9559ms 16.7649ms 59.6484 Ops/s 222.4951 Ops/s $\textbf{\color{#d91a1a}-73.19\%}$
test_zeros_like 16.8473ms 16.7116ms 59.8386 Ops/s 226.7979 Ops/s $\textbf{\color{#d91a1a}-73.62\%}$
test_ones_like 4.6081ms 4.3863ms 227.9801 Ops/s 226.2947 Ops/s $\color{#35bf28}+0.74\%$
test_clone 7.0954ms 6.6271ms 150.8954 Ops/s 149.7288 Ops/s $\color{#35bf28}+0.78\%$
test_squeeze 86.8820μs 13.3564μs 74.8705 KOps/s 72.0025 KOps/s $\color{#35bf28}+3.98\%$
test_unsqueeze 0.2838ms 0.1032ms 9.6854 KOps/s 9.1540 KOps/s $\textbf{\color{#35bf28}+5.81\%}$
test_split 0.3457ms 0.1716ms 5.8258 KOps/s 5.4620 KOps/s $\textbf{\color{#35bf28}+6.66\%}$
test_permute 0.2619ms 0.1939ms 5.1585 KOps/s 5.0162 KOps/s $\color{#35bf28}+2.84\%$
test_stack 52.5009ms 52.0181ms 19.2241 Ops/s 19.5338 Ops/s $\color{#d91a1a}-1.59\%$
test_cat 52.4244ms 51.5153ms 19.4117 Ops/s 19.4127 Ops/s $-0.01\%$

Xmaster6y pushed a commit to Xmaster6y/tensordict that referenced this pull request Feb 27, 2026
@vmoens
[Performance] Optimize RedisTensorDict indexed ops with GETRANGE/SETR…
420128d 
…ANGE

Replace the full read-modify-write approach for indexed access with Redis
byte-range operations (GETRANGE/SETRANGE). This reduces data transfer from
O(full_tensor) to O(slice) and network round-trips from 2N to 2 (or 1 with
metadata caching).

Key changes:
- Add _compute_byte_ranges to map tensor indices to byte offsets
- Add pipelined _abatch_get_at / _abatch_set_at / _abatch_read_modify_write
- Override __setitem__ to bypass _SubTensorDict for batched SETRANGE
- Override _get_at_str/_get_at_tuple to use GETRANGE
- Add metadata cache (cache_metadata=True by default) to skip metadata
  round-trips
- Refactor inline meta decoding to shared _decode_meta utility


ghstack-source-id: dddb29e
Pull-Request: pytorch#1569
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