[Refactor,Feature] Refactor collector shapes and stack_result in sync collector #1994

vmoens · 2024-03-06T17:46:37Z

This PR:

Allows users to stack data (and not cat) in MultiSyncDataCollector (@matteobettini)
Documents how batches are collected across collectors
Documents how collectors interact with replay buffers
Solves a bug for multidim buffers when ndim > 2
Makes sure that the trajectories in MultiSyncDataCollector are unique and consecutive
Makes sure that the trajectories in MultiaSyncDataCollector are unique and consecutive
Masks and squashes preempted batches in MultiSyncDataCollector

TODO:

Add the stack_result keyword in docstrings
Make stack_result raise an exception in MultiaSyncDataCollector
Resolve the cat dim issue for preemption (cat along dim=-1) and document it
Check that things work of without doing cat inplace and cloning

Check the updated doc below in collectors to learn more

pytorch-bot · 2024-03-06T17:46:42Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/rl/1994

📄 Preview Python docs built from this PR

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

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

[PLEASE AVOID MERGING] Lots of REDs in main after recovering from S3 Outage

❌ 13 New Failures

As of commit c742a43 with merge base 4bce371 ():

NEW FAILURES - The following jobs have failed:

Habitat Tests on Linux / tests (3.9, 11.6) / linux-job (gh)
RuntimeError: Command docker exec -t ca9f8437061ca9f0872f5115ce553aaa97b545d82d4343e893b073dcfbfab915 /exec failed with exit code 139
Lint / c-source / linux-job (gh)
RuntimeError: Command docker exec -t 988d89eed57bc20fe421e277c2e7b6d04d3145892f476cad31ee3a66fc538240 /exec failed with exit code 1
Unit-tests on Linux / tests-cpu (3.10) / linux-job (gh)
test/test_tensordictmodules.py::TestGRUModule::test_gru_parallel_env[True-False-False]
Unit-tests on Linux / tests-cpu (3.11) / linux-job (gh)
test/test_tensordictmodules.py::TestGRUModule::test_gru_parallel_env[True-False-False]
Unit-tests on Linux / tests-cpu (3.8) / linux-job (gh)
test/test_tensordictmodules.py::TestGRUModule::test_gru_parallel_env[True-False-False]
Unit-tests on Linux / tests-cpu (3.9) / linux-job (gh)
test/test_tensordictmodules.py::TestGRUModule::test_gru_parallel_env[True-False-False]
Unit-tests on Linux / tests-gpu (3.10, 12.1) / linux-job (gh)
test/test_tensordictmodules.py::TestGRUModule::test_gru_parallel_env[True-False-False]
Unit-tests on Linux / tests-olddeps (3.8, 11.6) / linux-job (gh)
test/test_tensordictmodules.py::TestGRUModule::test_gru_parallel_env[True-False-False]
Unit-tests on Linux / tests-optdeps (3.10, 12.1) / linux-job (gh)
test/test_tensordictmodules.py::TestGRUModule::test_gru_parallel_env[True-False-False]
Unit-tests on Linux / tests-stable-gpu (3.10, 11.8) / linux-job (gh)
test/test_tensordictmodules.py::TestGRUModule::test_gru_parallel_env[True-False-False]
Unit-tests on MacOS CPU / tests (3.11) / macos-job (gh)
test/test_tensordictmodules.py::TestGRUModule::test_gru_parallel_env[True-False-False]
Unit-tests on MacOS CPU / tests (3.8) / macos-job (gh)
test/test_tensordictmodules.py::TestGRUModule::test_gru_parallel_env[True-False-False]
Unit-tests on Windows / unittests-cpu / windows-job (gh)
test/test_tensordictmodules.py::TestGRUModule::test_gru_parallel_env[True-False-False]

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

github-actions · 2024-03-06T17:53:37Z

$\color{#D29922}\textsf{\Large&#x26A0;\kern{0.2cm}\normalsize Warning}$ Result of CPU Benchmark Tests

Total Benchmarks: 91. Improved: $\large\color{#35bf28}4$. Worsened: $\large\color{#d91a1a}5$.

Expand to view detailed results

Name	Max	Mean	Ops	Ops on Repo `HEAD`	Change
test_single	53.1089ms	52.4619ms	19.0615 Ops/s	17.8592 Ops/s	$\textbf{\color{#35bf28}+6.73\%}$
test_sync	35.5930ms	29.7961ms	33.5614 Ops/s	30.5180 Ops/s	$\textbf{\color{#35bf28}+9.97\%}$
test_async	48.0734ms	27.8519ms	35.9041 Ops/s	35.5460 Ops/s	$\color{#35bf28}+1.01\%$
test_simple	0.3818s	0.3279s	3.0499 Ops/s	3.0865 Ops/s	$\color{#d91a1a}-1.19\%$
test_transformed	0.5020s	0.4594s	2.1768 Ops/s	2.1565 Ops/s	$\color{#35bf28}+0.94\%$
test_serial	1.1867s	1.1606s	0.8616 Ops/s	0.8428 Ops/s	$\color{#35bf28}+2.23\%$
test_parallel	1.0493s	1.0099s	0.9902 Ops/s	0.9702 Ops/s	$\color{#35bf28}+2.07\%$
test_step_mdp_speed[True-True-True-True-True]	0.1250ms	21.3903μs	46.7502 KOps/s	46.7949 KOps/s	$\color{#d91a1a}-0.10\%$
test_step_mdp_speed[True-True-True-True-False]	35.2950μs	13.0374μs	76.7025 KOps/s	76.8782 KOps/s	$\color{#d91a1a}-0.23\%$
test_step_mdp_speed[True-True-True-False-True]	42.2580μs	12.3168μs	81.1900 KOps/s	80.0651 KOps/s	$\color{#35bf28}+1.41\%$
test_step_mdp_speed[True-True-True-False-False]	26.4190μs	7.5499μs	132.4519 KOps/s	129.5662 KOps/s	$\color{#35bf28}+2.23\%$
test_step_mdp_speed[True-True-False-True-True]	74.0880μs	22.3446μs	44.7536 KOps/s	44.1395 KOps/s	$\color{#35bf28}+1.39\%$
test_step_mdp_speed[True-True-False-True-False]	40.1950μs	14.2987μs	69.9366 KOps/s	69.4999 KOps/s	$\color{#35bf28}+0.63\%$
test_step_mdp_speed[True-True-False-False-True]	38.1520μs	13.6421μs	73.3025 KOps/s	73.1831 KOps/s	$\color{#35bf28}+0.16\%$
test_step_mdp_speed[True-True-False-False-False]	32.5910μs	8.6569μs	115.5152 KOps/s	112.7608 KOps/s	$\color{#35bf28}+2.44\%$
test_step_mdp_speed[True-False-True-True-True]	57.0970μs	23.5144μs	42.5271 KOps/s	41.9789 KOps/s	$\color{#35bf28}+1.31\%$
test_step_mdp_speed[True-False-True-True-False]	50.8550μs	15.3461μs	65.1630 KOps/s	63.6732 KOps/s	$\color{#35bf28}+2.34\%$
test_step_mdp_speed[True-False-True-False-True]	34.3850μs	13.5242μs	73.9413 KOps/s	72.0770 KOps/s	$\color{#35bf28}+2.59\%$
test_step_mdp_speed[True-False-True-False-False]	33.3020μs	8.7365μs	114.4623 KOps/s	113.3148 KOps/s	$\color{#35bf28}+1.01\%$
test_step_mdp_speed[True-False-False-True-True]	52.3570μs	24.8933μs	40.1714 KOps/s	39.5818 KOps/s	$\color{#35bf28}+1.49\%$
test_step_mdp_speed[True-False-False-True-False]	37.2990μs	16.7168μs	59.8199 KOps/s	60.3541 KOps/s	$\color{#d91a1a}-0.89\%$
test_step_mdp_speed[True-False-False-False-True]	50.5350μs	14.5713μs	68.6279 KOps/s	67.2846 KOps/s	$\color{#35bf28}+2.00\%$
test_step_mdp_speed[True-False-False-False-False]	26.2490μs	9.6708μs	103.4043 KOps/s	99.4530 KOps/s	$\color{#35bf28}+3.97\%$
test_step_mdp_speed[False-True-True-True-True]	53.1390μs	23.6391μs	42.3028 KOps/s	41.1045 KOps/s	$\color{#35bf28}+2.92\%$
test_step_mdp_speed[False-True-True-True-False]	36.0170μs	15.5179μs	64.4416 KOps/s	63.7756 KOps/s	$\color{#35bf28}+1.04\%$
test_step_mdp_speed[False-True-True-False-True]	42.8390μs	15.5322μs	64.3823 KOps/s	61.4712 KOps/s	$\color{#35bf28}+4.74\%$
test_step_mdp_speed[False-True-True-False-False]	50.5540μs	9.8581μs	101.4390 KOps/s	99.9243 KOps/s	$\color{#35bf28}+1.52\%$
test_step_mdp_speed[False-True-False-True-True]	36.6280μs	25.2565μs	39.5938 KOps/s	38.7349 KOps/s	$\color{#35bf28}+2.22\%$
test_step_mdp_speed[False-True-False-True-False]	39.2430μs	16.8569μs	59.3230 KOps/s	59.9915 KOps/s	$\color{#d91a1a}-1.11\%$
test_step_mdp_speed[False-True-False-False-True]	45.7250μs	16.8425μs	59.3737 KOps/s	58.6620 KOps/s	$\color{#35bf28}+1.21\%$
test_step_mdp_speed[False-True-False-False-False]	35.4870μs	11.1355μs	89.8032 KOps/s	88.5966 KOps/s	$\color{#35bf28}+1.36\%$
test_step_mdp_speed[False-False-True-True-True]	54.7720μs	26.1864μs	38.1877 KOps/s	37.5347 KOps/s	$\color{#35bf28}+1.74\%$
test_step_mdp_speed[False-False-True-True-False]	41.4570μs	17.9696μs	55.6497 KOps/s	55.3146 KOps/s	$\color{#35bf28}+0.61\%$
test_step_mdp_speed[False-False-True-False-True]	45.6250μs	16.6239μs	60.1545 KOps/s	58.4985 KOps/s	$\color{#35bf28}+2.83\%$
test_step_mdp_speed[False-False-True-False-False]	32.2400μs	11.1159μs	89.9610 KOps/s	88.1230 KOps/s	$\color{#35bf28}+2.09\%$
test_step_mdp_speed[False-False-False-True-True]	54.9230μs	27.3030μs	36.6260 KOps/s	36.2383 KOps/s	$\color{#35bf28}+1.07\%$
test_step_mdp_speed[False-False-False-True-False]	60.4120μs	19.2345μs	51.9900 KOps/s	52.4694 KOps/s	$\color{#d91a1a}-0.91\%$
test_step_mdp_speed[False-False-False-False-True]	57.1770μs	17.6069μs	56.7959 KOps/s	54.4672 KOps/s	$\color{#35bf28}+4.28\%$
test_step_mdp_speed[False-False-False-False-False]	38.3810μs	12.1714μs	82.1599 KOps/s	81.0185 KOps/s	$\color{#35bf28}+1.41\%$
test_values[generalized_advantage_estimate-True-True]	9.8043ms	9.0875ms	110.0414 Ops/s	107.2400 Ops/s	$\color{#35bf28}+2.61\%$
test_values[vec_generalized_advantage_estimate-True-True]	37.5890ms	35.0775ms	28.5083 Ops/s	28.7461 Ops/s	$\color{#d91a1a}-0.83\%$
test_values[td0_return_estimate-False-False]	0.2203ms	0.1611ms	6.2061 KOps/s	6.2029 KOps/s	$\color{#35bf28}+0.05\%$
test_values[td1_return_estimate-False-False]	25.7561ms	22.7775ms	43.9030 Ops/s	43.0699 Ops/s	$\color{#35bf28}+1.93\%$
test_values[vec_td1_return_estimate-False-False]	38.1532ms	35.2206ms	28.3925 Ops/s	28.6193 Ops/s	$\color{#d91a1a}-0.79\%$
test_values[td_lambda_return_estimate-True-False]	35.0813ms	32.7377ms	30.5458 Ops/s	30.0173 Ops/s	$\color{#35bf28}+1.76\%$
test_values[vec_td_lambda_return_estimate-True-False]	36.8560ms	35.2269ms	28.3874 Ops/s	28.6598 Ops/s	$\color{#d91a1a}-0.95\%$
test_gae_speed[generalized_advantage_estimate-False-1-512]	8.6559ms	8.0781ms	123.7914 Ops/s	121.1603 Ops/s	$\color{#35bf28}+2.17\%$
test_gae_speed[vec_generalized_advantage_estimate-True-1-512]	2.4300ms	1.9107ms	523.3781 Ops/s	529.2330 Ops/s	$\color{#d91a1a}-1.11\%$
test_gae_speed[vec_generalized_advantage_estimate-False-1-512]	0.4744ms	0.3450ms	2.8986 KOps/s	2.8970 KOps/s	$\color{#35bf28}+0.05\%$
test_gae_speed[vec_generalized_advantage_estimate-True-32-512]	46.1720ms	44.9930ms	22.2257 Ops/s	22.8807 Ops/s	$\color{#d91a1a}-2.86\%$
test_gae_speed[vec_generalized_advantage_estimate-False-32-512]	3.6552ms	3.0170ms	331.4513 Ops/s	332.6410 Ops/s	$\color{#d91a1a}-0.36\%$
test_dqn_speed	1.5418ms	1.3210ms	757.0176 Ops/s	759.5988 Ops/s	$\color{#d91a1a}-0.34\%$
test_ddpg_speed	3.4279ms	2.6397ms	378.8267 Ops/s	381.2410 Ops/s	$\color{#d91a1a}-0.63\%$
test_sac_speed	8.3517ms	8.0126ms	124.8033 Ops/s	125.0693 Ops/s	$\color{#d91a1a}-0.21\%$
test_redq_speed	14.8748ms	12.8585ms	77.7696 Ops/s	77.6354 Ops/s	$\color{#35bf28}+0.17\%$
test_redq_deprec_speed	13.1945ms	12.7046ms	78.7114 Ops/s	78.2568 Ops/s	$\color{#35bf28}+0.58\%$
test_td3_speed	8.1084ms	7.9417ms	125.9182 Ops/s	124.5643 Ops/s	$\color{#35bf28}+1.09\%$
test_cql_speed	0.1139s	38.9495ms	25.6743 Ops/s	27.7055 Ops/s	$\textbf{\color{#d91a1a}-7.33\%}$
test_a2c_speed	8.5951ms	7.2642ms	137.6606 Ops/s	138.8972 Ops/s	$\color{#d91a1a}-0.89\%$
test_ppo_speed	8.1655ms	7.5537ms	132.3863 Ops/s	133.3126 Ops/s	$\color{#d91a1a}-0.69\%$
test_reinforce_speed	7.3639ms	6.4781ms	154.3657 Ops/s	155.0059 Ops/s	$\color{#d91a1a}-0.41\%$
test_iql_speed	32.5968ms	32.0920ms	31.1604 Ops/s	31.3865 Ops/s	$\color{#d91a1a}-0.72\%$
test_rb_sample[TensorDictReplayBuffer-ListStorage-RandomSampler-4000]	3.1668ms	2.0510ms	487.5612 Ops/s	478.3535 Ops/s	$\color{#35bf28}+1.92\%$
test_rb_sample[TensorDictReplayBuffer-LazyMemmapStorage-RandomSampler-10000]	0.6458ms	0.4924ms	2.0307 KOps/s	2.0466 KOps/s	$\color{#d91a1a}-0.78\%$
test_rb_sample[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-10000]	0.7082ms	0.4665ms	2.1436 KOps/s	2.1760 KOps/s	$\color{#d91a1a}-1.49\%$
test_rb_sample[TensorDictReplayBuffer-ListStorage-SamplerWithoutReplacement-4000]	2.9805ms	2.0553ms	486.5561 Ops/s	491.5167 Ops/s	$\color{#d91a1a}-1.01\%$
test_rb_sample[TensorDictReplayBuffer-LazyMemmapStorage-SamplerWithoutReplacement-10000]	0.8988ms	0.4852ms	2.0612 KOps/s	2.0990 KOps/s	$\color{#d91a1a}-1.80\%$
test_rb_sample[TensorDictReplayBuffer-LazyTensorStorage-SamplerWithoutReplacement-10000]	0.6528ms	0.4577ms	2.1851 KOps/s	2.1917 KOps/s	$\color{#d91a1a}-0.30\%$
test_rb_sample[TensorDictReplayBuffer-LazyMemmapStorage-sampler6-10000]	1.5383ms	1.2749ms	784.3799 Ops/s	792.6227 Ops/s	$\color{#d91a1a}-1.04\%$
test_rb_sample[TensorDictReplayBuffer-LazyTensorStorage-sampler7-10000]	1.6343ms	1.2081ms	827.7760 Ops/s	834.7936 Ops/s	$\color{#d91a1a}-0.84\%$
test_rb_sample[TensorDictPrioritizedReplayBuffer-ListStorage-None-4000]	2.4567ms	2.1776ms	459.2168 Ops/s	452.2404 Ops/s	$\color{#35bf28}+1.54\%$
test_rb_sample[TensorDictPrioritizedReplayBuffer-LazyMemmapStorage-None-10000]	0.9387ms	0.6012ms	1.6634 KOps/s	1.6702 KOps/s	$\color{#d91a1a}-0.40\%$
test_rb_sample[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-10000]	96.2632ms	0.6454ms	1.5493 KOps/s	1.7655 KOps/s	$\textbf{\color{#d91a1a}-12.24\%}$
test_rb_iterate[TensorDictReplayBuffer-ListStorage-RandomSampler-4000]	2.2259ms	2.0476ms	488.3712 Ops/s	484.7456 Ops/s	$\color{#35bf28}+0.75\%$
test_rb_iterate[TensorDictReplayBuffer-LazyMemmapStorage-RandomSampler-10000]	0.6462ms	0.4926ms	2.0300 KOps/s	2.0431 KOps/s	$\color{#d91a1a}-0.64\%$
test_rb_iterate[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-10000]	3.7594ms	0.4710ms	2.1230 KOps/s	2.1677 KOps/s	$\color{#d91a1a}-2.06\%$
test_rb_iterate[TensorDictReplayBuffer-ListStorage-SamplerWithoutReplacement-4000]	3.2010ms	2.0844ms	479.7431 Ops/s	478.5863 Ops/s	$\color{#35bf28}+0.24\%$
test_rb_iterate[TensorDictReplayBuffer-LazyMemmapStorage-SamplerWithoutReplacement-10000]	0.5640ms	0.4834ms	2.0687 KOps/s	2.0792 KOps/s	$\color{#d91a1a}-0.50\%$
test_rb_iterate[TensorDictReplayBuffer-LazyTensorStorage-SamplerWithoutReplacement-10000]	3.5818ms	0.4670ms	2.1413 KOps/s	2.1529 KOps/s	$\color{#d91a1a}-0.54\%$
test_rb_iterate[TensorDictPrioritizedReplayBuffer-ListStorage-None-4000]	2.2854ms	2.1726ms	460.2683 Ops/s	453.1496 Ops/s	$\color{#35bf28}+1.57\%$
test_rb_iterate[TensorDictPrioritizedReplayBuffer-LazyMemmapStorage-None-10000]	1.1800ms	0.6076ms	1.6458 KOps/s	1.6660 KOps/s	$\color{#d91a1a}-1.21\%$
test_rb_iterate[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-10000]	0.7689ms	0.5763ms	1.7351 KOps/s	1.7422 KOps/s	$\color{#d91a1a}-0.41\%$
test_rb_populate[TensorDictReplayBuffer-ListStorage-RandomSampler-400]	0.1024s	7.3330ms	136.3692 Ops/s	144.2807 Ops/s	$\textbf{\color{#d91a1a}-5.48\%}$
test_rb_populate[TensorDictReplayBuffer-LazyMemmapStorage-RandomSampler-400]	13.9910ms	11.9698ms	83.5438 Ops/s	83.6562 Ops/s	$\color{#d91a1a}-0.13\%$
test_rb_populate[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-400]	1.5816ms	1.0728ms	932.1422 Ops/s	993.7897 Ops/s	$\textbf{\color{#d91a1a}-6.20\%}$
test_rb_populate[TensorDictReplayBuffer-ListStorage-SamplerWithoutReplacement-400]	85.9577ms	5.1925ms	192.5868 Ops/s	149.4518 Ops/s	$\textbf{\color{#35bf28}+28.86\%}$
test_rb_populate[TensorDictReplayBuffer-LazyMemmapStorage-SamplerWithoutReplacement-400]	93.1944ms	13.5613ms	73.7390 Ops/s	83.6510 Ops/s	$\textbf{\color{#d91a1a}-11.85\%}$
test_rb_populate[TensorDictReplayBuffer-LazyTensorStorage-SamplerWithoutReplacement-400]	1.5437ms	1.0427ms	959.0321 Ops/s	944.3283 Ops/s	$\color{#35bf28}+1.56\%$
test_rb_populate[TensorDictPrioritizedReplayBuffer-ListStorage-None-400]	88.8976ms	5.6054ms	178.3986 Ops/s	183.2893 Ops/s	$\color{#d91a1a}-2.67\%$
test_rb_populate[TensorDictPrioritizedReplayBuffer-LazyMemmapStorage-None-400]	14.7068ms	12.3973ms	80.6627 Ops/s	71.8944 Ops/s	$\textbf{\color{#35bf28}+12.20\%}$
test_rb_populate[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-400]	1.8868ms	1.3606ms	734.9822 Ops/s	717.0357 Ops/s	$\color{#35bf28}+2.50\%$

github-actions · 2024-03-06T17:56:52Z

$\color{#D29922}\textsf{\Large&#x26A0;\kern{0.2cm}\normalsize Warning}$ Result of GPU Benchmark Tests

Total Benchmarks: 94. Improved: $\large\color{#35bf28}4$. Worsened: $\large\color{#d91a1a}2$.

Expand to view detailed results

Name	Max	Mean	Ops	Ops on Repo `HEAD`	Change
test_single	99.5076ms	98.5673ms	10.1454 Ops/s	9.3467 Ops/s	$\textbf{\color{#35bf28}+8.54\%}$
test_sync	91.5732ms	87.6182ms	11.4131 Ops/s	11.6097 Ops/s	$\color{#d91a1a}-1.69\%$
test_async	0.1747s	87.6835ms	11.4047 Ops/s	11.5133 Ops/s	$\color{#d91a1a}-0.94\%$
test_single_pixels	0.1094s	0.1090s	9.1743 Ops/s	9.0476 Ops/s	$\color{#35bf28}+1.40\%$
test_sync_pixels	68.2034ms	65.8578ms	15.1842 Ops/s	15.1073 Ops/s	$\color{#35bf28}+0.51\%$
test_async_pixels	0.1218s	55.3405ms	18.0700 Ops/s	18.0921 Ops/s	$\color{#d91a1a}-0.12\%$
test_simple	0.6423s	0.6417s	1.5584 Ops/s	1.4795 Ops/s	$\textbf{\color{#35bf28}+5.34\%}$
test_transformed	0.8439s	0.8428s	1.1865 Ops/s	1.1405 Ops/s	$\color{#35bf28}+4.04\%$
test_serial	2.0915s	2.0288s	0.4929 Ops/s	0.4724 Ops/s	$\color{#35bf28}+4.35\%$
test_parallel	1.8423s	1.7811s	0.5615 Ops/s	0.5512 Ops/s	$\color{#35bf28}+1.86\%$
test_step_mdp_speed[True-True-True-True-True]	88.6910μs	33.5437μs	29.8119 KOps/s	30.3453 KOps/s	$\color{#d91a1a}-1.76\%$
test_step_mdp_speed[True-True-True-True-False]	43.9100μs	19.5849μs	51.0596 KOps/s	50.1738 KOps/s	$\color{#35bf28}+1.77\%$
test_step_mdp_speed[True-True-True-False-True]	35.8900μs	18.6420μs	53.6424 KOps/s	52.5745 KOps/s	$\color{#35bf28}+2.03\%$
test_step_mdp_speed[True-True-True-False-False]	42.6010μs	11.2057μs	89.2401 KOps/s	87.4749 KOps/s	$\color{#35bf28}+2.02\%$
test_step_mdp_speed[True-True-False-True-True]	54.5510μs	34.3886μs	29.0794 KOps/s	28.3625 KOps/s	$\color{#35bf28}+2.53\%$
test_step_mdp_speed[True-True-False-True-False]	42.0000μs	21.3755μs	46.7826 KOps/s	45.7955 KOps/s	$\color{#35bf28}+2.16\%$
test_step_mdp_speed[True-True-False-False-True]	36.9910μs	20.2928μs	49.2786 KOps/s	47.9571 KOps/s	$\color{#35bf28}+2.76\%$
test_step_mdp_speed[True-True-False-False-False]	37.6100μs	13.0823μs	76.4390 KOps/s	74.6943 KOps/s	$\color{#35bf28}+2.34\%$
test_step_mdp_speed[True-False-True-True-True]	90.0610μs	36.2543μs	27.5829 KOps/s	26.8850 KOps/s	$\color{#35bf28}+2.60\%$
test_step_mdp_speed[True-False-True-True-False]	42.2400μs	23.3749μs	42.7810 KOps/s	41.6122 KOps/s	$\color{#35bf28}+2.81\%$
test_step_mdp_speed[True-False-True-False-True]	47.5500μs	20.4923μs	48.7989 KOps/s	47.8742 KOps/s	$\color{#35bf28}+1.93\%$
test_step_mdp_speed[True-False-True-False-False]	30.6300μs	13.0634μs	76.5500 KOps/s	74.9468 KOps/s	$\color{#35bf28}+2.14\%$
test_step_mdp_speed[True-False-False-True-True]	0.1067ms	37.4070μs	26.7330 KOps/s	25.9027 KOps/s	$\color{#35bf28}+3.21\%$
test_step_mdp_speed[True-False-False-True-False]	91.7510μs	25.1786μs	39.7163 KOps/s	38.8560 KOps/s	$\color{#35bf28}+2.21\%$
test_step_mdp_speed[True-False-False-False-True]	38.8610μs	22.1574μs	45.1316 KOps/s	44.6352 KOps/s	$\color{#35bf28}+1.11\%$
test_step_mdp_speed[True-False-False-False-False]	49.2200μs	14.9155μs	67.0446 KOps/s	66.6733 KOps/s	$\color{#35bf28}+0.56\%$
test_step_mdp_speed[False-True-True-True-True]	53.0900μs	36.3837μs	27.4849 KOps/s	27.2041 KOps/s	$\color{#35bf28}+1.03\%$
test_step_mdp_speed[False-True-True-True-False]	66.1100μs	23.4607μs	42.6244 KOps/s	41.8236 KOps/s	$\color{#35bf28}+1.91\%$
test_step_mdp_speed[False-True-True-False-True]	42.1310μs	24.2554μs	41.2279 KOps/s	40.6812 KOps/s	$\color{#35bf28}+1.34\%$
test_step_mdp_speed[False-True-True-False-False]	40.0510μs	14.9457μs	66.9088 KOps/s	66.1627 KOps/s	$\color{#35bf28}+1.13\%$
test_step_mdp_speed[False-True-False-True-True]	59.5210μs	37.8985μs	26.3863 KOps/s	25.7345 KOps/s	$\color{#35bf28}+2.53\%$
test_step_mdp_speed[False-True-False-True-False]	41.1310μs	25.0729μs	39.8837 KOps/s	38.2875 KOps/s	$\color{#35bf28}+4.17\%$
test_step_mdp_speed[False-True-False-False-True]	41.2710μs	25.7931μs	38.7700 KOps/s	37.7910 KOps/s	$\color{#35bf28}+2.59\%$
test_step_mdp_speed[False-True-False-False-False]	33.3300μs	16.6132μs	60.1931 KOps/s	58.6200 KOps/s	$\color{#35bf28}+2.68\%$
test_step_mdp_speed[False-False-True-True-True]	65.7110μs	40.2592μs	24.8390 KOps/s	24.7378 KOps/s	$\color{#35bf28}+0.41\%$
test_step_mdp_speed[False-False-True-True-False]	53.7410μs	26.9172μs	37.1509 KOps/s	36.2873 KOps/s	$\color{#35bf28}+2.38\%$
test_step_mdp_speed[False-False-True-False-True]	92.2810μs	25.8202μs	38.7294 KOps/s	37.9929 KOps/s	$\color{#35bf28}+1.94\%$
test_step_mdp_speed[False-False-True-False-False]	33.8210μs	16.5543μs	60.4071 KOps/s	59.1055 KOps/s	$\color{#35bf28}+2.20\%$
test_step_mdp_speed[False-False-False-True-True]	58.4100μs	40.6822μs	24.5808 KOps/s	23.6408 KOps/s	$\color{#35bf28}+3.98\%$
test_step_mdp_speed[False-False-False-True-False]	51.0000μs	28.4764μs	35.1168 KOps/s	33.9722 KOps/s	$\color{#35bf28}+3.37\%$
test_step_mdp_speed[False-False-False-False-True]	46.3910μs	27.3161μs	36.6085 KOps/s	35.3213 KOps/s	$\color{#35bf28}+3.64\%$
test_step_mdp_speed[False-False-False-False-False]	34.2400μs	18.5225μs	53.9883 KOps/s	53.4061 KOps/s	$\color{#35bf28}+1.09\%$
test_values[generalized_advantage_estimate-True-True]	22.9465ms	22.6116ms	44.2251 Ops/s	43.0435 Ops/s	$\color{#35bf28}+2.75\%$
test_values[vec_generalized_advantage_estimate-True-True]	83.7874ms	3.2079ms	311.7277 Ops/s	304.1930 Ops/s	$\color{#35bf28}+2.48\%$
test_values[td0_return_estimate-False-False]	86.2010μs	60.7794μs	16.4529 KOps/s	15.7910 KOps/s	$\color{#35bf28}+4.19\%$
test_values[td1_return_estimate-False-False]	49.2283ms	48.6248ms	20.5656 Ops/s	19.9354 Ops/s	$\color{#35bf28}+3.16\%$
test_values[vec_td1_return_estimate-False-False]	2.0336ms	1.7116ms	584.2618 Ops/s	579.7887 Ops/s	$\color{#35bf28}+0.77\%$
test_values[td_lambda_return_estimate-True-False]	82.9479ms	81.2637ms	12.3056 Ops/s	12.4587 Ops/s	$\color{#d91a1a}-1.23\%$
test_values[vec_td_lambda_return_estimate-True-False]	2.0806ms	1.7208ms	581.1133 Ops/s	578.5999 Ops/s	$\color{#35bf28}+0.43\%$
test_gae_speed[generalized_advantage_estimate-False-1-512]	22.9709ms	22.1209ms	45.2062 Ops/s	46.5883 Ops/s	$\color{#d91a1a}-2.97\%$
test_gae_speed[vec_generalized_advantage_estimate-True-1-512]	0.8616ms	0.6521ms	1.5334 KOps/s	1.4902 KOps/s	$\color{#35bf28}+2.90\%$
test_gae_speed[vec_generalized_advantage_estimate-False-1-512]	0.6793ms	0.6076ms	1.6458 KOps/s	1.6110 KOps/s	$\color{#35bf28}+2.16\%$
test_gae_speed[vec_generalized_advantage_estimate-True-32-512]	1.5060ms	1.4184ms	705.0085 Ops/s	700.4333 Ops/s	$\color{#35bf28}+0.65\%$
test_gae_speed[vec_generalized_advantage_estimate-False-32-512]	0.9378ms	0.6199ms	1.6131 KOps/s	1.5415 KOps/s	$\color{#35bf28}+4.64\%$
test_dqn_speed	8.0564ms	1.4167ms	705.8470 Ops/s	710.0952 Ops/s	$\color{#d91a1a}-0.60\%$
test_ddpg_speed	2.8392ms	2.6670ms	374.9578 Ops/s	371.4999 Ops/s	$\color{#35bf28}+0.93\%$
test_sac_speed	8.4142ms	7.7995ms	128.2140 Ops/s	126.8708 Ops/s	$\color{#35bf28}+1.06\%$
test_redq_speed	11.7780ms	10.4265ms	95.9091 Ops/s	94.9090 Ops/s	$\color{#35bf28}+1.05\%$
test_redq_deprec_speed	11.7469ms	10.9656ms	91.1946 Ops/s	88.6271 Ops/s	$\color{#35bf28}+2.90\%$
test_td3_speed	8.0616ms	7.7992ms	128.2178 Ops/s	128.2228 Ops/s	$-0.00\%$
test_cql_speed	26.0683ms	25.2926ms	39.5372 Ops/s	39.1353 Ops/s	$\color{#35bf28}+1.03\%$
test_a2c_speed	6.0259ms	5.6002ms	178.5643 Ops/s	181.6196 Ops/s	$\color{#d91a1a}-1.68\%$
test_ppo_speed	6.2873ms	5.9761ms	167.3345 Ops/s	170.2161 Ops/s	$\color{#d91a1a}-1.69\%$
test_reinforce_speed	5.4095ms	4.5331ms	220.6016 Ops/s	219.2982 Ops/s	$\color{#35bf28}+0.59\%$
test_iql_speed	20.4522ms	19.6837ms	50.8035 Ops/s	50.3173 Ops/s	$\color{#35bf28}+0.97\%$
test_rb_sample[TensorDictReplayBuffer-ListStorage-RandomSampler-4000]	3.1392ms	2.8778ms	347.4923 Ops/s	347.7342 Ops/s	$\color{#d91a1a}-0.07\%$
test_rb_sample[TensorDictReplayBuffer-LazyMemmapStorage-RandomSampler-10000]	1.2408ms	0.5462ms	1.8309 KOps/s	1.8050 KOps/s	$\color{#35bf28}+1.43\%$
test_rb_sample[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-10000]	0.7325ms	0.5258ms	1.9017 KOps/s	1.8570 KOps/s	$\color{#35bf28}+2.41\%$
test_rb_sample[TensorDictReplayBuffer-ListStorage-SamplerWithoutReplacement-4000]	3.0943ms	2.8991ms	344.9310 Ops/s	342.8397 Ops/s	$\color{#35bf28}+0.61\%$
test_rb_sample[TensorDictReplayBuffer-LazyMemmapStorage-SamplerWithoutReplacement-10000]	1.3471ms	0.5426ms	1.8428 KOps/s	1.8268 KOps/s	$\color{#35bf28}+0.88\%$
test_rb_sample[TensorDictReplayBuffer-LazyTensorStorage-SamplerWithoutReplacement-10000]	0.8125ms	0.5237ms	1.9097 KOps/s	1.8965 KOps/s	$\color{#35bf28}+0.69\%$
test_rb_sample[TensorDictReplayBuffer-LazyMemmapStorage-sampler6-10000]	3.9098ms	1.4612ms	684.3508 Ops/s	657.9487 Ops/s	$\color{#35bf28}+4.01\%$
test_rb_sample[TensorDictReplayBuffer-LazyTensorStorage-sampler7-10000]	1.6256ms	1.4152ms	706.6100 Ops/s	687.4975 Ops/s	$\color{#35bf28}+2.78\%$
test_rb_sample[TensorDictPrioritizedReplayBuffer-ListStorage-None-4000]	3.1949ms	3.0009ms	333.2360 Ops/s	333.9926 Ops/s	$\color{#d91a1a}-0.23\%$
test_rb_sample[TensorDictPrioritizedReplayBuffer-LazyMemmapStorage-None-10000]	0.9300ms	0.6681ms	1.4968 KOps/s	1.4744 KOps/s	$\color{#35bf28}+1.52\%$
test_rb_sample[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-10000]	4.4668ms	0.6539ms	1.5292 KOps/s	1.5445 KOps/s	$\color{#d91a1a}-0.99\%$
test_rb_iterate[TensorDictReplayBuffer-ListStorage-RandomSampler-4000]	2.9668ms	2.8590ms	349.7754 Ops/s	346.3722 Ops/s	$\color{#35bf28}+0.98\%$
test_rb_iterate[TensorDictReplayBuffer-LazyMemmapStorage-RandomSampler-10000]	0.7978ms	0.5430ms	1.8415 KOps/s	1.7937 KOps/s	$\color{#35bf28}+2.66\%$
test_rb_iterate[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-10000]	4.2654ms	0.5363ms	1.8645 KOps/s	1.8448 KOps/s	$\color{#35bf28}+1.07\%$
test_rb_iterate[TensorDictReplayBuffer-ListStorage-SamplerWithoutReplacement-4000]	3.1183ms	2.9096ms	343.6874 Ops/s	342.0067 Ops/s	$\color{#35bf28}+0.49\%$
test_rb_iterate[TensorDictReplayBuffer-LazyMemmapStorage-SamplerWithoutReplacement-10000]	1.2687ms	0.5423ms	1.8441 KOps/s	1.8367 KOps/s	$\color{#35bf28}+0.41\%$
test_rb_iterate[TensorDictReplayBuffer-LazyTensorStorage-SamplerWithoutReplacement-10000]	0.6665ms	0.5254ms	1.9032 KOps/s	1.8860 KOps/s	$\color{#35bf28}+0.91\%$
test_rb_iterate[TensorDictPrioritizedReplayBuffer-ListStorage-None-4000]	3.1426ms	3.0017ms	333.1396 Ops/s	330.9529 Ops/s	$\color{#35bf28}+0.66\%$
test_rb_iterate[TensorDictPrioritizedReplayBuffer-LazyMemmapStorage-None-10000]	0.8267ms	0.6713ms	1.4897 KOps/s	1.4702 KOps/s	$\color{#35bf28}+1.33\%$
test_rb_iterate[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-10000]	4.6274ms	0.6588ms	1.5178 KOps/s	1.5137 KOps/s	$\color{#35bf28}+0.27\%$
test_rb_populate[TensorDictReplayBuffer-ListStorage-RandomSampler-400]	0.1148s	8.9723ms	111.4541 Ops/s	114.9524 Ops/s	$\color{#d91a1a}-3.04\%$
test_rb_populate[TensorDictReplayBuffer-LazyMemmapStorage-RandomSampler-400]	16.3224ms	14.0385ms	71.2328 Ops/s	69.6984 Ops/s	$\color{#35bf28}+2.20\%$
test_rb_populate[TensorDictReplayBuffer-LazyTensorStorage-RandomSampler-400]	1.1576ms	1.0645ms	939.4067 Ops/s	880.1653 Ops/s	$\textbf{\color{#35bf28}+6.73\%}$
test_rb_populate[TensorDictReplayBuffer-ListStorage-SamplerWithoutReplacement-400]	0.1117s	8.8931ms	112.4473 Ops/s	148.3898 Ops/s	$\textbf{\color{#d91a1a}-24.22\%}$
test_rb_populate[TensorDictReplayBuffer-LazyMemmapStorage-SamplerWithoutReplacement-400]	16.2913ms	14.0095ms	71.3802 Ops/s	70.0752 Ops/s	$\color{#35bf28}+1.86\%$
test_rb_populate[TensorDictReplayBuffer-LazyTensorStorage-SamplerWithoutReplacement-400]	2.3436ms	1.1956ms	836.3700 Ops/s	869.4117 Ops/s	$\color{#d91a1a}-3.80\%$
test_rb_populate[TensorDictPrioritizedReplayBuffer-ListStorage-None-400]	0.1096s	7.1626ms	139.6148 Ops/s	107.1246 Ops/s	$\textbf{\color{#35bf28}+30.33\%}$
test_rb_populate[TensorDictPrioritizedReplayBuffer-LazyMemmapStorage-None-400]	16.6940ms	14.4449ms	69.2284 Ops/s	68.7534 Ops/s	$\color{#35bf28}+0.69\%$
test_rb_populate[TensorDictPrioritizedReplayBuffer-LazyTensorStorage-None-400]	7.8783ms	1.6527ms	605.0576 Ops/s	693.1828 Ops/s	$\textbf{\color{#d91a1a}-12.71\%}$

vmoens · 2024-03-07T18:47:36Z

@albertbou92 I also took this opportunity to refactor preemption. Now you can stack (with padding) or cat (with masking) the values coming from the collectors.
I noticed we preempt based on the proportion of collectors that have reached the end, not on the proportion of frames collected. IMO that would be more sensible.
We could use a mp.Value() shared between processes that tracks how many frames have been collected, and as soon as this goes above the threshold we break. Would you like to give it a shot once this PR lands? No worry if you don't have the time!
I suspect it can bring some decent speed-up over what we're doing.

In a nutshell:

# in SyncDataCollector.rollout
def rollout(self):
    ...
    if self.interruptor is not None:
        self._collected_frames += frame_count

# in MultiSyncDataCollector.iterator
def iterator(self):
    if self.preemptive_threshold and ...:
        if self._collected_frames > self.frames_per_batch * self.preemptive_threshold:
            break # or similar

albertbou92 · 2024-03-08T15:04:45Z

@albertbou92 I also took this opportunity to refactor preemption. Now you can stack (with padding) or cat (with masking) the values coming from the collectors. I noticed we preempt based on the proportion of collectors that have reached the end, not on the proportion of frames collected. IMO that would be more sensible. We could use a mp.Value() shared between processes that tracks how many frames have been collected, and as soon as this goes above the threshold we break. Would you like to give it a shot once this PR lands? No worry if you don't have the time! I suspect it can bring some decent speed-up over what we're doing.

In a nutshell:
# in SyncDataCollector.rollout
def rollout(self):
    ...
    if self.interruptor is not None:
        self._collected_frames += frame_count

# in MultiSyncDataCollector.iterator
def iterator(self):
    if self.preemptive_threshold and ...:
        if self._collected_frames > self.frames_per_batch * self.preemptive_threshold:
            break # or similar

Yes the preemption was based on the idea that all workers collect a fixed number of frames and can not communicate during collection (it was also the assumption in DDPPO), but if we can have a shared global value that tracks how many frames have been collected globally and that does not impact speed that would be much better. Yes, I can give it shot!

vmoens · 2024-03-19T15:06:55Z

@matteobettini This PR addresses an issue that you pointed a long time ago RE the stacking of results in MultiSync
Do you think it solves it in an adequate way?

The idea is to build the collector with cat_results="stack" if you want to stack or cat_results=-1 if you want to cat along the time dimension. cat_result=None results in cat_results=0 + warning since this is the previous behaviour but we want to discourage it.
We recommend "stack" > -1 > 0

Sorry it took so long to solve this!

matteobettini · 2024-03-19T15:14:27Z

@matteobettini This PR addresses an issue that you pointed a long time ago RE the stacking of results in MultiSync Do you think it solves it in an adequate way?

The idea is to build the collector with cat_results="stack" if you want to stack or cat_results=-1 if you want to cat along the time dimension. cat_result=None results in cat_results=0 + warning since this is the previous behaviour but we want to discourage it. We recommend "stack" > -1 > 0

Sorry it took so long to solve this!

Amamzing!

So stack_results=True now is exactly like a single collector with the number of collectors B preappended to the batch size which feels so smooth!

This is what i wanted yes!

docs/source/reference/collectors.rst

matteobettini · 2024-03-19T15:20:23Z

docs/source/reference/collectors.rst

+--------------------+---------------------+-------------+---------------+------------------------------+
+|     Single env     |         [T]         |   `[B, T]`  |  `[B*(T//B)`  |              [T]             |
+--------------------+---------------------+-------------+---------------+------------------------------+
+| Batched env (n=P)  |       [P, T]        | `[B, P, T]` |  `[B * P, T]` |            [P, T]            |


[B * P, T], this is given that cat_results=0

my personal preference, to avoid having cat_results be both int and str is to have 2 args

cat_results true or false

collectors_dim dimension where to cat or stack (or similar nicer name)

but then what is collector_dim=0 and cat_results=False? That should not be allowed. So we need a complicated doc that lists the available configs
If we allow them we also need to test every single combination...

Note that the goal is to get rid of this arg in v0.6 or v0.7 so adding 2 args is twice as bad as adding 1 given that we want to break things on the long run

collector_dim=0 and cat_results=False

isn't this what happens currently when cat_results="stack"?

Anyway yes how you prefer

Yes sorry, I meant collector_dim=-1 or 1 or anything that isn't 0

you can't stack along any dim (and should not be able to do so), that's what I mean. Stack should be along dim 0

i don't super see this as in the example table we are referring to, in the row for stacking in multisync i can see both

B, P, T (collector_dim=0 or -3, cat_results=False) (currently the only available option)

P, B, T (collector_dim=1 or -2, cat_results=False)

P, T, B (collector_dim=2 or -1, cat_results=False)

There is no cat_results=False, it can only be a string ("stack") or an int
cat_results="stack" => stack B collectors along 0 => [B, P, T]
cat_results=0 => cat B collectors along 0 => [B * P, T]
cat_results=-1 => cat B collectors along -1 => [B, P * T]

I think I didn't get what you were trying to say though

I know, that is the point of this thread, I am discussing why for me it makes sense to separate them.

P, B, T (collector_dim=1 or -2, cat_results=False)
P, T, B (collector_dim=2 or -1, cat_results=False)

I don't think we should support these. If you want that you can transpose your resulting tensordict. Implementing it would require a lot of expensive tests for a feature we will deprecate in 2 releases.

docs/source/reference/collectors.rst

Co-authored-by: Matteo Bettini <55539777+matteobettini@users.noreply.github.com>

…-shapes

init

52d9e4d

facebook-github-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 Mar 6, 2024

vmoens added bug Something isn't working enhancement New feature or request labels Mar 6, 2024

vmoens and others added 9 commits March 7, 2024 09:24

Merge remote-tracking branch 'origin/main' into collector-shapes

6b5bc27

remove stack_results incompatibility

922b3f9

amend

49d5164

Merge remote-tracking branch 'origin/main' into collector-shapes

8527b10

amend

b8add98

amend

5020e72

amend

587846b

amend

5b140bb

amend

10594fc

vmoens added 5 commits March 7, 2024 18:49

amend

27fed55

amend

08ac7c0

amend

b458f63

amend

054c1f8

amend

dae9900

Vincent Moens added 4 commits March 12, 2024 18:43

amend

eb6f120

amend

0d6d845

Merge remote-tracking branch 'origin/main' into collector-shapes

96b2abf

empty

27d0f6e

vmoens marked this pull request as ready for review March 18, 2024 14:05

Vincent Moens added 2 commits March 18, 2024 14:05

Merge remote-tracking branch 'origin/main' into collector-shapes

91e4c28

Merge remote-tracking branch 'origin/main' into collector-shapes

396ca7f

Merge remote-tracking branch 'origin/main' into collector-shapes

8cf75fd

matteobettini approved these changes Mar 19, 2024

View reviewed changes

Vincent Moens added 2 commits March 19, 2024 16:29

amend

bcfa94f

amend

902d1bd

matteobettini reviewed Mar 19, 2024

View reviewed changes

docs/source/reference/collectors.rst Outdated Show resolved Hide resolved

Vincent Moens and others added 7 commits March 19, 2024 16:32

amend

bbacfa5

Update docs/source/reference/collectors.rst

72b1039

Co-authored-by: Matteo Bettini <55539777+matteobettini@users.noreply.github.com>

amend

26178c2

Merge remote-tracking branch 'origin/collector-shapes' into collector…

1a7ec7b

…-shapes

amend

b015b3b

amend

0b41ff9

amend

c742a43

vmoens merged commit f6fbc44 into main Mar 19, 2024

vmoens deleted the collector-shapes branch March 19, 2024 21:48

[Refactor,Feature] Refactor collector shapes and stack_result in sync collector #1994

[Refactor,Feature] Refactor collector shapes and stack_result in sync collector #1994

Uh oh!

Conversation

vmoens commented Mar 6, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pytorch-bot bot commented Mar 6, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/rl/1994

❗ 1 Active SEVs

❌ 13 New Failures

Uh oh!

github-actions bot commented Mar 6, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

$\color{#D29922}\textsf{\Large&amp;#x26A0;\kern{0.2cm}\normalsize Warning}$ Result of CPU Benchmark Tests

Total Benchmarks: 91. Improved: $\large\color{#35bf28}4$. Worsened: $\large\color{#d91a1a}5$.

Uh oh!

github-actions bot commented Mar 6, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

$\color{#D29922}\textsf{\Large&amp;#x26A0;\kern{0.2cm}\normalsize Warning}$ Result of GPU Benchmark Tests

Total Benchmarks: 94. Improved: $\large\color{#35bf28}4$. Worsened: $\large\color{#d91a1a}2$.

Uh oh!

vmoens commented Mar 7, 2024

Uh oh!

albertbou92 commented Mar 8, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

vmoens commented Mar 19, 2024

Uh oh!

matteobettini commented Mar 19, 2024

Uh oh!

Uh oh!

matteobettini Mar 19, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

vmoens commented Mar 6, 2024 •

edited

Loading

pytorch-bot bot commented Mar 6, 2024 •

edited

Loading

github-actions bot commented Mar 6, 2024 •

edited

Loading

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

github-actions bot commented Mar 6, 2024 •

edited

Loading

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

albertbou92 commented Mar 8, 2024 •

edited

Loading

matteobettini Mar 19, 2024 •

edited

Loading