Papers › Masked Structural Growth for 2x Faster Language Model Pre-training

Masked Structural Growth for 2x Faster Language Model Pre-training

4 May 2023arXiv:2305.02869archive 2025-07-28

Yiqun Yao, Zheng Zhang, Jing Li, Yequan Wang

Accelerating large language model pre-training is a critical issue in present research. In this paper, we focus on speeding up pre-training by progressively growing from a small Transformer structure to a large one. There are two main research problems associated with progressive growth: determining the optimal growth schedule, and designing efficient growth operators. In terms of growth schedule, the impact of each single dimension on a schedule's efficiency is under-explored by existing work. Regarding the growth operators, existing methods rely on the initialization of new weights to inherit knowledge, and achieve only non-strict function preservation, limiting further improvements on training dynamics. To address these issues, we propose Masked Structural Growth (MSG), including (i) growth schedules involving all possible dimensions and (ii) strictly function-preserving growth operators that is independent of the initialization of new weights. Experiments show that MSG is significantly faster than related work: we achieve up to 2.2x speedup in pre-training different types of language models while maintaining comparable or better downstream performances. Code is publicly available at https://github.com/cofe-ai/MSG.

PaperPDFCodeCode Syntology ranMethod as code

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

Syntology 34 code samples harvested from 1 repository linked to this paper: 16 ran (of which 6 constructed an object rather than computing a result; 10 with no instrument failure: 0 honoured, 0 violated, 10 with no contract checked; 6 where Syntology's instrument failed) · 18 unverified. Where Syntology's instrument failed: 5 where our contract draft was wrong; 1 where the contract check could not run; failures of Syntology's instrument, not of the code.

By repository: official repository: 34 samples from 1 repository, 16 ran (10 with no instrument failure; 6 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.

Levels written between 2026-09-22 and 2026-09-24 (UTC): the dates Syntology's graph wrote each sample's verification level, not when the code executed. 2 of the 16 samples that ran have no recorded date.

cofe-ai/msg officialmentioned in papermentioned on GitHubpytorchSyntology: 34 samples, 16 ran (6 instrument-failed) · 18 unverified 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.

Code Syntology ran Syntology

34 samples harvested: 16 ran (of which 6 constructed an object rather than computing a result; 10 with no instrument failure: 0 honoured, 0 violated, 10 with no contract checked; 6 where Syntology's instrument failed) · 18 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.

5ran · our draft was wrong
1ran; the contract check could not run
10ran
18unverified

Licence: 0 of the 34 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 cofe-ai/msg. “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”. For 4 samples that ran, the graph records a signature this page does not show: for a class the fixture built without calling, the harness called the object with inputs it made up itself, so the recorded signature is the harness's probe, not the fixture's arguments. “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.

cofe-ai/msg official repository 34 samples, 16 ran (6 instrument-failed) · 18 unverified
BertEmbeddingsEx model_ex/grow_ops_v2.py ran Apache-2.0 (permissive) · 12c02fdc7a3ef516 · report fixture: signature input_ids:Tensor/2/int, token_type_ids:NoneType/0/?, position_ids:NoneType/0/?, inputs_embeds:NoneType/0/?, past_key_values_length:int/0/int · returned Tensor · level recorded 2026-09-24
BertIntermediateEx model_ex/grow_ops_v2.py ran fingerprinted Apache-2.0 (permissive) · abc2a5159b60c88e · report fixture: returned BertIntermediateEx · level recorded 2026-09-22
BertLMPredictionHeadEx model_ex/grow_ops_v2.py ran fingerprinted Apache-2.0 (permissive) · 8f34abe0617b5f97 · report fixture: returned BertLMPredictionHeadEx · level recorded 2026-09-22
BertOutputEx model_ex/grow_ops_v2.py ran Apache-2.0 (permissive) · ba319340870887ba · report fixture: signature hidden_states:Tensor/2/float, input_tensor:Tensor/2/float · returned Tensor · level recorded 2026-09-24
BertPoolerEx model_ex/grow_ops_v2.py ran Apache-2.0 (permissive) · f76bc6941b23b53c · report fixture: returned BertPoolerEx · level recorded 2026-09-22
BertPreTrainingHeadsEx model_ex/grow_ops_v2.py ran fingerprinted Apache-2.0 (permissive) · b3a39bfd6444ea7c · report fixture: returned BertPreTrainingHeadsEx · level recorded 2026-09-22
BertPredictionHeadTransformEx model_ex/grow_ops_v2.py ran fingerprinted Apache-2.0 (permissive) · dcb1a915dcc1578a · report fixture: returned BertPredictionHeadTransformEx · level recorded 2026-09-22
BertSelfAttentionEx model_ex/grow_ops_v2.py ran Apache-2.0 (permissive) · dc33918b0246bfa2 · report fixture: returned BertSelfAttentionEx · level recorded 2026-09-22
BertSelfOutputEx model_ex/grow_ops_v2.py ran Apache-2.0 (permissive) · f3837d1329cfa0a9 · report fixture: signature hidden_states:Tensor/3/float, input_tensor:Tensor/3/float · returned Tensor · level recorded 2026-09-24
LayerNormEx model_ex/grow_ops_v2.py ran fingerprinted Apache-2.0 (permissive) · a2517173045b652b · report fixture: signature input_vec:Tensor/3/float, ln_mask:Tensor/3/float, in_growth:bool/0/bool, debug_print:bool/0/bool · returned Tensor · level recorded 2026-09-24
count_parameters in 278 papers model_ex/grow_ops_v2.py ran; the contract check could not run (RAISES) Apache-2.0 (permissive) · c03c6d119d2ac973 · report fixture: signature model:callable/0/? · returned int · level recorded 2026-09-22
expand_embedding model_ex/grow_ops_v2.py ran · our draft was wrong Apache-2.0 (permissive) · a0b27279fea1a125 · report fixture: signature old_embedding:callable/0/?, new_in_num:int/0/int, vocab_size:int/0/int, args:Args/0/? · returned Embedding · level recorded 2026-09-22
expand_linear model_ex/grow_ops_v2.py ran · our draft was wrong Apache-2.0 (permissive) · 8162611121d247e7 · report fixture: signature old_linear:callable/0/?, new_out_num:int/0/int, new_in_num:int/0/int, args:dict/0/?, extra:dict/0/? · returned Linear · level recorded 2026-09-22
expand_norm model_ex/grow_ops_v2.py ran · our draft was wrong Apache-2.0 (permissive) · 519d9bcfb16dcf6c · report fixture: signature old_ln:callable/0/?, new_out_num:int/0/int, args:dict/0/? · returned LayerNormEx · level recorded 2026-09-22
prepare_eval_per_device run_grow_bert.py ran · our draft was wrong fingerprinted Apache-2.0 (permissive) · ee96c6045e2257e2 · report fixture: signature mlm_preds:Tensor/3/float, nsp_preds:Tensor/2/float, mlm_labels:Tensor/2/int, nsp_labels:Tensor/1/int · returned tuple · level recorded, date not recorded
static_mask_single preprocess_bert_data.py ran · our draft was wrong Apache-2.0 (permissive) · 681bc51b62ed3855 · report fixture: signature example:ndarray/1/int, tokenizer:MockTokenizer/0/? · returned tuple · level recorded, date not recorded
BertAttentionEx model_ex/grow_ops_v2.py unverified Apache-2.0 (permissive) · ae85ec1873dd1703 · report
BertEncoderEx model_ex/grow_ops_v2.py unverified Apache-2.0 (permissive) · 1ccc2565fa59af1d · report
BertLayerEx model_ex/grow_ops_v2.py unverified Apache-2.0 (permissive) · 86592733e07a8723 · report
grow_block model_ex/grow_ops_v2.py unverified Apache-2.0 (permissive) · c969b61fcdd66216 · report
grow_dim_intermediate model_ex/grow_ops_v2.py unverified Apache-2.0 (permissive) · 2c3575cfa3435b09 · report
grow_dim_nsp model_ex/grow_ops_v2.py unverified Apache-2.0 (permissive) · 596cf53dd41ddb02 · report
grow_dim_output model_ex/grow_ops_v2.py unverified Apache-2.0 (permissive) · fb20ae8f1b2151e9 · report
grow_dim_pooler model_ex/grow_ops_v2.py unverified Apache-2.0 (permissive) · ac46a64ea2608233 · report
grow_dim_self_att model_ex/grow_ops_v2.py unverified Apache-2.0 (permissive) · baaac4ac9de4e27a · report
grow_dim_self_output model_ex/grow_ops_v2.py unverified Apache-2.0 (permissive) · 88e9fafdee923a80 · report
grow_dim_transform model_ex/grow_ops_v2.py unverified Apache-2.0 (permissive) · 06fa339e87bd096d · report
grow_embedding model_ex/grow_ops_v2.py unverified Apache-2.0 (permissive) · d1ada439198d07ca · report
grow_head_num model_ex/grow_ops_v2.py unverified Apache-2.0 (permissive) · 42d5058f41beae5b · report
grow_ops model_ex/grow_ops_v2.py unverified Apache-2.0 (permissive) · 109797f27864fd02 · report
init_weight_ex model_ex/grow_ops_v2.py unverified Apache-2.0 (permissive) · ea16063f081619fc · report
stack_block model_ex/grow_ops_v2.py unverified Apache-2.0 (permissive) · 18d7d645c49b39b9 · report
transfer_states model_ex/grow_ops_v2.py unverified Apache-2.0 (permissive) · ef5ff778ebe66fa5 · report
vanilla_copy model_ex/grow_ops_v2.py unverified Apache-2.0 (permissive) · 8179d1035ab4a2dd · report

Method as code Syntology

What Syntology's graph records as this paper's claimed method, lifted into code: each assembly names its bricks (functions and classes), each with its own role and its own verification level. An assembly is not a check that the code implements the method correctly or reproduces the paper's results. Levels are each brick's own verification level in Syntology's graph (level 0: no recorded run), listed brick by brick and never combined into one level; “ran” means the brick executed on a synthesized input, not that its output is correct, and the instrument figure counts failures of Syntology's instrument, not of the code. Where a model read the code against the paper's claim, each assembly's label verdict is one model's judgement, kept in this page's JSON twin; it is not calibrated, so no card prints it.

1 assembly on this paper: 1 harvested, 0 generated.

harvested Assembly for Masked Structural Growth (MSG) (graph method: Masked Structural Growth (MSG)). 32 bricks; individually on a fixture: 14 ran (of which 6 constructed an object rather than computing a result; 10 with no instrument failure: 0 honoured, 0 violated, 10 with no contract checked; 4 where Syntology's instrument failed) · 18 unverified. The assembly has not been run end-to-end.

  1. grow_ops entry · level 0 · unverified · 109797f27864fd02
  2. count_parameters helper · level 1 · ran; the contract check could not run (RAISES) · c03c6d119d2ac973
  3. grow_block helper · level 0 · unverified · c969b61fcdd66216
  4. grow_dim_intermediate helper · level 0 · unverified · 2c3575cfa3435b09
  5. grow_dim_nsp helper · level 0 · unverified · 596cf53dd41ddb02
  6. grow_dim_output helper · level 0 · unverified · fb20ae8f1b2151e9
  7. grow_dim_pooler helper · level 0 · unverified · ac46a64ea2608233
  8. grow_dim_self_att helper · level 0 · unverified · baaac4ac9de4e27a
  9. grow_dim_self_output helper · level 0 · unverified · 88e9fafdee923a80
  10. grow_dim_transform helper · level 0 · unverified · 06fa339e87bd096d
  11. grow_embedding helper · level 0 · unverified · d1ada439198d07ca
  12. grow_head_num helper · level 0 · unverified · 42d5058f41beae5b
  13. stack_block helper · level 0 · unverified · 18d7d645c49b39b9
  14. transfer_states helper · level 0 · unverified · ef5ff778ebe66fa5
  15. vanilla_copy helper · level 0 · unverified · 8179d1035ab4a2dd
  16. BertEncoderEx helper · level 0 · unverified · 1ccc2565fa59af1d
  17. BertLayerEx helper · level 0 · unverified · 86592733e07a8723
  18. init_weight_ex helper · level 0 · unverified · ea16063f081619fc
  19. BertIntermediateEx helper · level 1 · ran · abc2a5159b60c88e
  20. expand_linear helper · level 1 · ran · our draft was wrong · 8162611121d247e7
  21. BertPreTrainingHeadsEx helper · level 1 · ran · b3a39bfd6444ea7c
  22. BertOutputEx helper · level 1 · ran · ba319340870887ba
  23. expand_norm helper · level 1 · ran · our draft was wrong · 519d9bcfb16dcf6c
  24. BertPoolerEx helper · level 1 · ran · f76bc6941b23b53c
  25. BertSelfAttentionEx helper · level 1 · ran · dc33918b0246bfa2
  26. BertSelfOutputEx helper · level 1 · ran · f3837d1329cfa0a9
  27. BertPredictionHeadTransformEx helper · level 1 · ran · dcb1a915dcc1578a
  28. BertEmbeddingsEx helper · level 1 · ran · 12c02fdc7a3ef516
  29. expand_embedding helper · level 1 · ran · our draft was wrong · a0b27279fea1a125
  30. BertAttentionEx helper · level 0 · unverified · ae85ec1873dd1703
  31. LayerNormEx helper · level 1 · ran · a2517173045b652b
  32. BertLMPredictionHeadEx helper · level 1 · ran · 8f34abe0617b5f97

Each brick ends with its code_sha256, Syntology's identity for that exact code; click one to copy the get_code call that returns it. For agents, on Syntology's MCP service: assembly(arxiv_id="2305.02869") returns these assemblies with each brick's role and wiring (how to connect).

Tasks

Language ModelingLanguage ModellingLarge Language ModelOpen-Ended Question Answering

Results from the paper archive 2025-07-28

No leaderboard rows for this paper in the archive.

Methods

Absolute Position EncodingsAdamAttentionBPEDense ConnectionsDropoutLabel SmoothingLayer NormalizationLinear LayerMulti-Head AttentionPosition-Wise Feed-Forward LayerResidual ConnectionSoftmaxTransformer

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