Papers › Benchmarking Neural Network Training Algorithms

Benchmarking Neural Network Training Algorithms

12 Jun 2023arXiv:2306.07179archive 2025-07-28

George E. Dahl, Frank Schneider, Zachary Nado, Naman Agarwal, Chandramouli Shama Sastry, Philipp Hennig, Sourabh Medapati, Runa Eschenhagen, Priya Kasimbeg, Daniel Suo, Juhan Bae, Justin Gilmer, Abel L. Peirson, Bilal Khan, Rohan Anil, Mike Rabbat, Shankar Krishnan, Daniel Snider, Ehsan Amid, Kongtao Chen, Chris J. Maddison, Rakshith Vasudev, Michal Badura, Ankush Garg, Peter Mattson

Training algorithms, broadly construed, are an essential part of every deep learning pipeline. Training algorithm improvements that speed up training across a wide variety of workloads (e.g., better update rules, tuning protocols, learning rate schedules, or data selection schemes) could save time, save computational resources, and lead to better, more accurate, models. Unfortunately, as a community, we are currently unable to reliably identify training algorithm improvements, or even determine the state-of-the-art training algorithm. In this work, using concrete experiments, we argue that real progress in speeding up training requires new benchmarks that resolve three basic challenges faced by empirical comparisons of training algorithms: (1) how to decide when training is complete and precisely measure training time, (2) how to handle the sensitivity of measurements to exact workload details, and (3) how to fairly compare algorithms that require hyperparameter tuning. In order to address these challenges, we introduce a new, competitive, time-to-result benchmark using multiple workloads running on fixed hardware, the AlgoPerf: Training Algorithms benchmark. Our benchmark includes a set of workload variants that make it possible to detect benchmark submissions that are more robust to workload changes than current widely-used methods. Finally, we evaluate baseline submissions constructed using various optimizers that represent current practice, as well as other optimizers that have recently received attention in the literature. These baseline results collectively demonstrate the feasibility of our benchmark, show that non-trivial gaps between methods exist, and set a provisional state-of-the-art for future benchmark submissions to try and surpass.

PaperPDFCodeCode Syntology ran

In Syntology View this paper on Syntology, its page in Syntology's graph. That page adds which of its functions are also present in other papers' code Syntology holds, with dates, and its dated run record as JSON and a README badge. It also lists the repositories linked to the paper, the abstract and the calls for agents.

Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

For agents, on Syntology's MCP service (how to connect):

Code

No sample from the archive's official repository (mlcommons/algorithmic-efficiency) is linked to this paper; the 22 samples below come from 1 community repository. Syntology harvested code from mlcommons/algorithmic-efficiency for another paper.

mlcommons/algorithmic-efficiency: Syntology's graph links these samples to Accelerating Neural Network Training: An Analysis of the AlgoPerf Competition (arXiv:2502.15015; that paper's own run record: 16 ran (of which 0 constructed an object rather than computing a result; 16 with no instrument failure: 0 honoured, 0 violated, 16 with no contract checked; 0 where Syntology's instrument failed) · 3 unverified); nothing checks that they implement this paper's method.

Syntology 22 code samples harvested from 1 repository linked to this paper: 21 ran (of which 0 constructed an object rather than computing a result; 21 with no instrument failure: 0 honoured, 0 violated, 21 with no contract checked; 0 where Syntology's instrument failed) · 1 unverified. “Where Syntology's instrument failed” counts failures of Syntology's instrument, not of the code; here there are none.

By repository: community (archive-listed): 22 samples from 1 repository, 21 ran (21 with no instrument failure; 0 where Syntology's instrument failed). The run record, sample by sample. “Ran” means executed on a synthesized input, not that the code is correct or reproduces the paper.

No sample is linked to this paper from 2 of its repositories that Syntology harvested for other papers:

Levels written on 2026-09-24 (UTC): the dates Syntology's graph wrote each sample's verification level, not when the code executed.

mlcommons/algorithmic-efficiency officialmentioned in papermentioned on GitHubjaxApache-2.0Syntology: no sample linked to this paper (harvested for another paper). Syntology's graph links these samples to Accelerating Neural Network Training: An Analysis of the AlgoPerf Competition (arXiv:2502.15015; that paper's own run record: 16 ran (of which 0 constructed an object rather than computing a result; 16 with no instrument failure: 0 honoured, 0 violated, 16 with no contract checked; 0 where Syntology's instrument failed) · 3 unverified); nothing checks that they implement this paper's method. report
deepmind/optax mentioned on GitHubjaxApache-2.0Syntology: 22 samples, 21 ran (0 instrument-failed) · 1 unverified report
google-deepmind/optax mentioned on GitHubjaxApache-2.0Syntology: not harvested report
mlcommons/submissions_algorithms mentioned on GitHubjaxApache-2.0Syntology: no sample linked to this paper (harvested for another paper). Syntology's graph links these samples to Accelerating Neural Network Training: An Analysis of the AlgoPerf Competition (arXiv:2502.15015; that paper's own run record: 16 ran (of which 0 constructed an object rather than computing a result; 16 with no instrument failure: 0 honoured, 0 violated, 16 with no contract checked; 0 where Syntology's instrument failed) · 3 unverified); nothing checks that they implement this paper's method. report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

A paper named beside a repository with no sample linked to this paper is shown with that paper's own run record, not this paper's: “ran” means executed on a synthesized input, not that the code is correct, and “where Syntology's instrument failed” counts failures of Syntology's instrument, not of the code. Papers are listed in arXiv-id order, at most three per repository.

Code Syntology ran Syntology

22 samples harvested: 21 ran (of which 0 constructed an object rather than computing a result; 21 with no instrument failure: 0 honoured, 0 violated, 21 with no contract checked; 0 where Syntology's instrument failed) · 1 unverified. “Where Syntology's instrument failed” counts failures of Syntology's instrument, not of the code. Syntology's record for this page has not changed since , the first build that kept a record date for it; when this build read Syntology's graph is in the build record. That date is not when the samples ran.

21ran
1unverified

Licence: 0 of the 22 samples are pointer only, meaning Syntology does not serve that copy's text. This page shows no code text for any sample; each one links to its file in the repository.

Harvested from deepmind/optax. “Ran” means the sample executed on a synthesized input. It does not mean the output is correct, and nothing here reproduces the paper's results. A line beginning “fixture:” under a sample that ran gives the arguments Syntology's fixture called the sample with, each as name:type/rank/element kind (? where none was recorded), and the type of what the fixture's run returned, exactly as the harness recorded them; the arguments and the returned type were recorded by two separate runs of the same fixture. No output value is shown, and a returned type says nothing about whether the output was right. “Level recorded” gives the UTC date when Syntology's graph wrote the sample's verification level, not when the code executed; where the level was written before Syntology's graph kept that date, the row says “level recorded, date not recorded”. “Honoured” and “violated” refer to a contract Syntology drafted from the code itself. “Our draft was wrong” and “the contract check could not run” are failures of Syntology's instrument, not of the code: the sample ran on Syntology's synthesized input, and then the drafted contract did not describe it, or the check itself raised an error (RAISES), lacked a dependency (DEP_MISSING) or ran out of time (TIMEOUT); hover the label for which. A metamorphic tier, where shown, is evidence of the output's shape, not that the sample implements the method.

Each sample ends with its code_sha256. A sample's code_sha256 is Syntology's identity for that exact code. It is printed in full (16 hexadecimal characters for every sample in this build), and get_code(code_sha256="…") takes it exactly as printed. An agent calls it on Syntology's MCP service (how to connect); click an identity to copy that call.

Repository labels, per sample. official repository: The archive marks this repository official for the paper. named in the paper: The archive records that the paper mentions this repository; it is not marked official. community (archive-listed): In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper. found in paper text by Syntology: Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted. community: Not in the archive's code links for this paper; a community repository Syntology harvested. Samples are grouped by repository: a repository marked official first, then by how directly the repository is tied to the paper, then in the archive's order; never by how many ran. Licence labels name the repository's licence as recorded at harvest. “Pointer only” means Syntology does not serve that copy's text, for one of six reasons: no licence file was found; the licence was not identified; the licence is recorded as permissive but that copy's record is not marked cleared; the licence is outside the permissive list Syntology serves text under (MIT, Apache-2.0, BSD and similar); this paper's copy was not located, so its licence is not recorded; or this paper's copy is cleared, but the stored copy of the same code that Syntology serves is not. Some licences outside that list permit redistribution, such as WTFPL, and GPL-3.0 under its conditions; they are simply not on the list. Hover a licence label for the reason. File links open the file on GitHub at the default branch, which may have changed since the harvest.

A function or class name links to its page when the name occurs in the harvested code of at least 5 papers; the name is a string, not an identity.

deepmind/optax community (archive-listed) 22 samples, 21 ran (0 instrument-failed) · 1 unverified
abs_sq optax/_src/numerics.py ran fingerprinted Apache-2.0 (permissive) · 8c1ff079e1c251cf · report fixture: signature x:ndarray/1/float · returned ndarray · level recorded 2026-09-24
adabelief optax/_src/alias.py ran Apache-2.0 (permissive) · 52002c9d4813e1ec · report fixture: signature learning_rate:float/0/float, b1:float/0/float, b2:float/0/float, eps:float/0/float, eps_root:float/0/float, weight_decay:NoneType/0/?, weight_decay_mask:callable/0/?, nesterov:bool/0/bool · returned GradientTransformationExtraArgs · level recorded 2026-09-24
adadelta optax/_src/alias.py ran Apache-2.0 (permissive) · c1bf0f66f8ead4ab · report fixture: signature learning_rate:float/0/float, rho:float/0/float, eps:float/0/float, weight_decay:float/0/float, weight_decay_mask:NoneType/0/? · returned GradientTransformationExtraArgs · level recorded 2026-09-24
adafactor optax/_src/alias.py ran Apache-2.0 (permissive) · be540cbd764c7d2a · report fixture: signature learning_rate:float/0/float, min_dim_size_to_factor:int/0/int, decay_rate:float/0/float, decay_offset:int/0/int, multiply_by_parameter_scale:bool/0/bool, clipping_threshold:float/0/float, momentum:NoneType/0/?, dtype_momentum:callable/0/?, weight_decay_rate:NoneType/0/?, eps:float/0/float, factored:bool/0/bool, weight_decay_mask:NoneType/0/? · returned GradientTransformationExtraArgs · level recorded 2026-09-24
deprecation_getattr optax/_src/deprecations.py ran Apache-2.0 (permissive) · 46fa6da2a2e26515 · report fixture: signature module:str/0/str, deprecations:dict/0/? · returned function · level recorded 2026-09-24
global_norm optax/_src/linear_algebra.py ran Apache-2.0 (permissive) · 34f71acccac437e4 · report fixture: signature updates:dict/0/? · returned ArrayImpl · level recorded 2026-09-24
init_empty_state optax/_src/base.py ran Apache-2.0 (permissive) · 6cb62843159b4e5e · report fixture: signature params:ArrayImpl/1/float · returned EmptyState · level recorded 2026-09-24
lookahead optax/_src/lookahead.py ran Apache-2.0 (permissive) · f69cdcfdf348b5ba · report fixture: signature fast_optimizer:GradientTransformationExtraArgs/1/callable, sync_period:int/0/int, slow_step_size:float/0/float, reset_state:bool/0/bool · returned GradientTransformation · level recorded 2026-09-24
matrix_inverse_pth_root optax/_src/linear_algebra.py ran Apache-2.0 (permissive) · 86e6aca17da03e77 · report fixture: signature matrix:ArrayImpl/2/float, p:int/0/int, num_iters:int/0/int, ridge_epsilon:float/0/float, error_tolerance:float/0/float, precision:Precision/0/? · returned tuple · level recorded 2026-09-24
power_iteration in 30 papers optax/_src/linear_algebra.py ran Apache-2.0 (permissive) · ed4ddeade9eb2eee · report fixture: signature matrix:ArrayImpl/2/float, v0:NoneType/0/?, num_iters:int/0/int, error_tolerance:float/0/float, precision:Precision/0/?, key:ArrayImpl/1/int · returned tuple · level recorded 2026-09-24
safe_norm in 9 papers optax/_src/numerics.py ran Apache-2.0 (permissive) · 8446f6b71ea9fde1 · report fixture: signature x:ArrayImpl/1/float, min_norm:ArrayImpl/0/float, ord:NoneType/0/?, axis:NoneType/0/?, keepdims:bool/0/bool · returned ArrayImpl · level recorded 2026-09-24
safe_root_mean_squares optax/_src/numerics.py ran Apache-2.0 (permissive) · b8071ff27a01c329 · report fixture: signature x:ArrayImpl/2/float, min_rms:ArrayImpl/0/float · returned ArrayImpl · level recorded 2026-09-24
scale_by_backtracking_linesearch optax/_src/linesearch.py ran Apache-2.0 (permissive) · e41730968a3fffcb · report fixture: signature max_backtracking_steps:int/0/int, slope_rtol:float/0/float, decrease_factor:float/0/float, increase_factor:float/0/float, max_learning_rate:float/0/float, atol:float/0/float, rtol:float/0/float, store_grad:bool/0/bool, verbose:bool/0/bool · returned GradientTransformationExtraArgs · level recorded 2026-09-24
scale_by_factored_rms optax/_src/factorized.py ran Apache-2.0 (permissive) · 32fce6c5f1050dd1 · report fixture: signature factored:bool/0/bool, decay_rate:float/0/float, step_offset:int/0/int, min_dim_size_to_factor:int/0/int, epsilon:float/0/float, decay_rate_fn:callable/0/? · returned GradientTransformation · level recorded 2026-09-24
scale_by_rms optax/_src/transform.py ran Apache-2.0 (permissive) · 0c22d7c2cc2a90d0 · report fixture: signature decay:float/0/float, eps:float/0/float, initial_scale:float/0/float, eps_in_sqrt:bool/0/bool, bias_correction:bool/0/bool · returned GradientTransformation · level recorded 2026-09-24
scale_by_rss optax/_src/transform.py ran Apache-2.0 (permissive) · 49a32735e1eb328a · report fixture: signature initial_accumulator_value:float/0/float, eps:float/0/float · returned GradientTransformation · level recorded 2026-09-24
scale_by_stddev optax/_src/transform.py ran Apache-2.0 (permissive) · 670803584425b73a · report fixture: signature decay:float/0/float, eps:float/0/float, initial_scale:float/0/float, eps_in_sqrt:bool/0/bool, bias_correction:bool/0/bool · returned GradientTransformation · level recorded 2026-09-24
stateless optax/_src/base.py ran Apache-2.0 (permissive) · 08a168538fed09eb · report fixture: signature f:callable/0/? · returned GradientTransformation · level recorded 2026-09-24
stateless_with_tree_map optax/_src/base.py ran Apache-2.0 (permissive) · 8dcf52b4ebe33a4a · report fixture: signature f:callable/0/? · returned GradientTransformation · level recorded 2026-09-24
warn_deprecated_function optax/_src/deprecations.py ran Apache-2.0 (permissive) · db7af5450cd8ac59 · report fixture: signature fun:callable/0/?, replacement:str/0/str, version_removed:str/0/str · returned function · level recorded 2026-09-24
zoom_linesearch optax/_src/linesearch.py ran Apache-2.0 (permissive) · 5760b69d788a57d7 · report fixture: signature max_linesearch_steps:int/0/int, max_stepsize:float/0/float, tol:float/0/float, increase_factor:float/0/float, slope_rtol:float/0/float, curv_rtol:float/0/float, approx_dec_rtol:float/0/float, interval_threshold:float/0/float, verbose:bool/0/bool · returned tuple · level recorded 2026-09-24
scale_by_zoom_linesearch optax/_src/linesearch.py unverified Apache-2.0 (permissive) · 3fcd57975548149b · report

Tasks

Benchmarking

Results from the paper archive 2025-07-28

No leaderboard rows for this paper in the archive.

Methods

SPEED

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections