Papers › Non-Vacuous Generalization Bounds for Large Language Models

Non-Vacuous Generalization Bounds for Large Language Models

28 Dec 2023arXiv:2312.17173archive 2025-07-28

Sanae Lotfi, Marc Finzi, Yilun Kuang, Tim G. J. Rudner, Micah Goldblum, Andrew Gordon Wilson

Modern language models can contain billions of parameters, raising the question of whether they can generalize beyond the training data or simply parrot their training corpora. We provide the first non-vacuous generalization bounds for pretrained large language models (LLMs), indicating that language models are capable of discovering regularities that generalize to unseen data. In particular, we derive a compression bound that is valid for the unbounded log-likelihood loss using prediction smoothing, and we extend the bound to handle subsampling, accelerating bound computation by orders of magnitude on massive datasets. To achieve the extreme level of compression required for non-vacuous bounds, we devise SubLoRA, a simple low-dimensional nonlinear parameterization that leads to non-vacuous generalization bounds for models with nearly a billion parameters. Finally, we use our bounds to understand LLM generalization and find that larger models have better generalization bounds and are more compressible than smaller models.

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 38 code samples harvested from 1 repository linked to this paper: 13 ran (of which 7 constructed an object rather than computing a result; 10 with no instrument failure: 1 honoured, 2 violated, 7 with no contract checked; 3 where Syntology's instrument failed) · 25 unverified. Where Syntology's instrument failed: 3 where our contract draft was wrong; failures of Syntology's instrument, not of the code.

By repository: official repository: 38 samples from 1 repository, 13 ran (10 with no instrument failure; 3 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 on 2026-09-22 (UTC): the dates Syntology's graph wrote each sample's verification level, not when the code executed.

sanaelotfi/sublora-bounds-for-llms officialmentioned in papermentioned on GitHubpytorchSyntology: 38 samples, 13 ran (3 instrument-failed) · 25 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

38 samples harvested: 13 ran (of which 7 constructed an object rather than computing a result; 10 with no instrument failure: 1 honoured, 2 violated, 7 with no contract checked; 3 where Syntology's instrument failed) · 25 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.

1ran · honoured contract
2ran · violated contract
3ran · our draft was wrong
7ran
25unverified

Licence: 0 of the 38 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 sanaelotfi/sublora-bounds-for-llms. “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.

sanaelotfi/sublora-bounds-for-llms official repository 38 samples, 13 ran (3 instrument-failed) · 25 unverified
FixedNumpySeed sublora/nn/projectors.py ran Apache-2.0 (permissive) · 42854ae0f684da62 · report fixture: returned FixedNumpySeed · level recorded 2026-09-22
FixedPytorchSeed sublora/nn/projectors.py ran Apache-2.0 (permissive) · 8e279895b6069a86 · report fixture: returned FixedPytorchSeed · level recorded 2026-09-22
IDModule sublora/nn/projectors.py ran · metamorphic tier: deterministic Apache-2.0 (permissive) · 7358581cd466c4e1 · report fixture: returned IDModule · level recorded 2026-09-22
RandomMultiply sublora/nn/projectors.py ran Apache-2.0 (permissive) · b066b14f9c37e272 · report fixture: returned RandomMultiply · level recorded 2026-09-22
_AdjointLinearOperator sublora/nn/projectors.py ran · metamorphic tier: deterministic Apache-2.0 (permissive) · 481607e48039c489 · report fixture: returned _AdjointLinearOperator · level recorded 2026-09-22
_CustomLinearOperator sublora/nn/projectors.py ran · metamorphic tier: deterministic Apache-2.0 (permissive) · 2defed82b8724ef8 · report fixture: returned _CustomLinearOperator · level recorded 2026-09-22
_TransposedLinearOperator sublora/nn/projectors.py ran · metamorphic tier: deterministic Apache-2.0 (permissive) · b8df645779d22d40 · report fixture: returned _TransposedLinearOperator · level recorded 2026-09-22
_get_dtype sublora/nn/projectors.py ran · our draft was wrong Apache-2.0 (permissive) · b2835d9d07c679de · report fixture: signature operators:list/1/NoneType, dtypes:NoneType/0/? · returned dtype · level recorded 2026-09-22
find_locations_from_condition sublora/nn/projectors.py ran · honoured contract Apache-2.0 (permissive) · 5b9d2e9ed76b6318 · report fixture: signature names:list/1/str, params:list/1/MockParam, condition_fn:callable/0/? · returned ndarray · level recorded 2026-09-22
isintlike sublora/nn/projectors.py ran · violated contract fingerprinted Apache-2.0 (permissive) · 0a7da0d18ce0c2fc · report fixture: signature x:int/0/int · returned bool · level recorded 2026-09-22
isshape sublora/nn/projectors.py ran · violated contract fingerprinted Apache-2.0 (permissive) · 73c2f05a719caea9 · report fixture: signature x:tuple/1/int, nonneg:bool/0/bool · returned bool · level recorded 2026-09-22
lazy_direct_matmat sublora/nn/projectors.py ran · our draft was wrong Apache-2.0 (permissive) · 0a857f86a88fb98e · report fixture: signature v:Tensor/2/float, Ms:list/1/Tensor, mults:list/1/int · returned Tensor · level recorded 2026-09-22
unflatten_like in 7 papers sublora/nn/projectors.py ran · our draft was wrong Apache-2.0 (permissive) · 6d8daee3396e21d5 · report fixture: signature vector:Tensor/1/float, likeTensorList:list/1/Tensor · returned list · level recorded 2026-09-22
CombinedRDKronFiLM sublora/nn/projectors.py unverified Apache-2.0 (permissive) · efb07aa0e63ef262 · report
CombinedRDKronQRFiLM sublora/nn/projectors.py unverified Apache-2.0 (permissive) · 51125dd47186c8c0 · report
ConcatLazy sublora/nn/projectors.py unverified Apache-2.0 (permissive) · 52a62393f3317699 · report
FastfoodOperator sublora/nn/projectors.py unverified Apache-2.0 (permissive) · 84ad10dc772efc21 · report
FiLMLazyRandom sublora/nn/projectors.py unverified Apache-2.0 (permissive) · 612604e0a472123c · report
FilterLazyRandom sublora/nn/projectors.py unverified Apache-2.0 (permissive) · 5a5b19e77dff59f3 · report
Lazy sublora/nn/projectors.py unverified Apache-2.0 (permissive) · 03f63f2c2e1eb7bf · report
LazyDirectSum sublora/nn/projectors.py unverified Apache-2.0 (permissive) · d032cccffcb09566 · report
LazyKron sublora/nn/projectors.py unverified Apache-2.0 (permissive) · 6d76cc862148b017 · report
LazyOneSidedKron sublora/nn/projectors.py unverified Apache-2.0 (permissive) · 0a1f5eeb87655050 · report
LazyPerm sublora/nn/projectors.py unverified Apache-2.0 (permissive) · 2da9befbc32d12d6 · report
LazyRandom sublora/nn/projectors.py unverified Apache-2.0 (permissive) · cbbd3e437d786806 · report
LazyRandomQR sublora/nn/projectors.py unverified Apache-2.0 (permissive) · 39aba651b25adab7 · report
LazySTFiLMRDKronQR sublora/nn/projectors.py unverified Apache-2.0 (permissive) · 0df6eef43ed7804a · report
RoundedDoubleKron sublora/nn/projectors.py unverified Apache-2.0 (permissive) · 03dfe475ed6ca9b6 · report
RoundedDoubleKronQR sublora/nn/projectors.py unverified Apache-2.0 (permissive) · ef5b20cc2c4467b6 · report
RoundedKron sublora/nn/projectors.py unverified Apache-2.0 (permissive) · 5d5797a72d977dae · report
SparseOperator sublora/nn/projectors.py unverified Apache-2.0 (permissive) · c55ed0bc145c418e · report
_PowerLinearOperator sublora/nn/projectors.py unverified Apache-2.0 (permissive) · 37d7ee47abf3dfc2 · report
_ProductLinearOperator sublora/nn/projectors.py unverified Apache-2.0 (permissive) · 5bba23ce771d6766 · report
_ScaledLinearOperator sublora/nn/projectors.py unverified Apache-2.0 (permissive) · 2c0692af09e2ca02 · report
_SumLinearOperator sublora/nn/projectors.py unverified Apache-2.0 (permissive) · 84b3ddf47d348630 · report
_delchainattr sublora/nn/projectors.py unverified Apache-2.0 (permissive) · cd1b260fe7f1ec38 · report
_setchainattr sublora/nn/projectors.py unverified Apache-2.0 (permissive) · d4ac8e1ac6ef5038 · report
create_intrinsic_model sublora/nn/projectors.py unverified Apache-2.0 (permissive) · d63815907fd184b0 · 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 SubLoRA (graph method: SubLoRA). 38 bricks; individually on a fixture: 13 ran (of which 7 constructed an object rather than computing a result; 10 with no instrument failure: 1 honoured, 2 violated, 7 with no contract checked; 3 where Syntology's instrument failed) · 25 unverified. The assembly has not been run end-to-end.

  1. create_intrinsic_model entry · level 0 · unverified · d63815907fd184b0
  2. CombinedRDKronFiLM helper · level 0 · unverified · efb07aa0e63ef262
  3. CombinedRDKronQRFiLM helper · level 0 · unverified · 51125dd47186c8c0
  4. FastfoodOperator helper · level 0 · unverified · 84ad10dc772efc21
  5. FiLMLazyRandom helper · level 0 · unverified · 612604e0a472123c
  6. IDModule model · level 1 · ran · 7358581cd466c4e1
  7. LazyRandom helper · level 0 · unverified · cbbd3e437d786806
  8. LazySTFiLMRDKronQR helper · level 0 · unverified · 0df6eef43ed7804a
  9. RoundedDoubleKron helper · level 0 · unverified · 03dfe475ed6ca9b6
  10. RoundedDoubleKronQR helper · level 0 · unverified · ef5b20cc2c4467b6
  11. RoundedKron helper · level 0 · unverified · 5d5797a72d977dae
  12. SparseOperator helper · level 0 · unverified · c55ed0bc145c418e
  13. FixedPytorchSeed helper · level 1 · ran · 8e279895b6069a86
  14. FilterLazyRandom helper · level 0 · unverified · 5a5b19e77dff59f3
  15. _delchainattr helper · level 0 · unverified · cd1b260fe7f1ec38
  16. _setchainattr helper · level 0 · unverified · d4ac8e1ac6ef5038
  17. unflatten_like helper · level 1 · ran · our draft was wrong · 6d8daee3396e21d5
  18. RandomMultiply helper · level 1 · ran · b066b14f9c37e272
  19. find_locations_from_condition helper · level 1 · ran · honoured contract · 5b9d2e9ed76b6318
  20. Lazy helper · level 0 · unverified · 03f63f2c2e1eb7bf
  21. LazyDirectSum helper · level 0 · unverified · d032cccffcb09566
  22. LazyKron helper · level 0 · unverified · 6d76cc862148b017
  23. LazyPerm helper · level 0 · unverified · 2da9befbc32d12d6
  24. LazyRandomQR helper · level 0 · unverified · 39aba651b25adab7
  25. ConcatLazy helper · level 0 · unverified · 52a62393f3317699
  26. LazyOneSidedKron helper · level 0 · unverified · 0a1f5eeb87655050
  27. FixedNumpySeed helper · level 1 · ran · 42854ae0f684da62
  28. lazy_direct_matmat helper · level 1 · ran · our draft was wrong · 0a857f86a88fb98e
  29. _AdjointLinearOperator helper · level 1 · ran · 481607e48039c489
  30. _CustomLinearOperator helper · level 1 · ran · 2defed82b8724ef8
  31. _PowerLinearOperator helper · level 0 · unverified · 37d7ee47abf3dfc2
  32. _ProductLinearOperator helper · level 0 · unverified · 5bba23ce771d6766
  33. _ScaledLinearOperator helper · level 0 · unverified · 2c0692af09e2ca02
  34. _SumLinearOperator helper · level 0 · unverified · 84b3ddf47d348630
  35. _TransposedLinearOperator helper · level 1 · ran · b8df645779d22d40
  36. _get_dtype helper · level 1 · ran · our draft was wrong · b2835d9d07c679de
  37. isintlike helper · level 1 · ran · violated contract · 0a7da0d18ce0c2fc
  38. isshape helper · level 1 · ran · violated contract · 73c2f05a719caea9

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="2312.17173") returns these assemblies with each brick's role and wiring (how to connect).

Tasks

Generalization Bounds

1 archive task tag without a task page not shown.

Results from the paper archive 2025-07-28

No leaderboard rows for this paper in the archive.

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