Papers › On the Stability of Fine-tuning BERT: Misconceptions, Explanations, and Strong Baselines
On the Stability of Fine-tuning BERT: Misconceptions, Explanations, and Strong Baselines
Marius Mosbach, Maksym Andriushchenko, Dietrich Klakow
Fine-tuning pre-trained transformer-based language models such as BERT has become a common practice dominating leaderboards across various NLP benchmarks. Despite the strong empirical performance of fine-tuned models, fine-tuning is an unstable process: training the same model with multiple random seeds can result in a large variance of the task performance. Previous literature (Devlin et al., 2019; Lee et al., 2020; Dodge et al., 2020) identified two potential reasons for the observed instability: catastrophic forgetting and small size of the fine-tuning datasets. In this paper, we show that both hypotheses fail to explain the fine-tuning instability. We analyze BERT, RoBERTa, and ALBERT, fine-tuned on commonly used datasets from the GLUE benchmark, and show that the observed instability is caused by optimization difficulties that lead to vanishing gradients. Additionally, we show that the remaining variance of the downstream task performance can be attributed to differences in generalization where fine-tuned models with the same training loss exhibit noticeably different test performance. Based on our analysis, we present a simple but strong baseline that makes fine-tuning BERT-based models significantly more stable than the previously proposed approaches. Code to reproduce our results is available online: https://github.com/uds-lsv/bert-stable-fine-tuning.
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):
get_harvested_code_for_paper(arxiv_id="2006.04884")get_code(code_sha256="…")with a sample'scode_sha256from the list belowget_citation_path(paper_1="2006.04884", paper_2="…")with another paper's arXiv id or titleget_concepts_for_paper(arxiv_id="2006.04884")
Code
Syntology 20 code samples harvested from 2 repositories linked to this paper: 13 ran (of which 0 constructed an object rather than computing a result; 9 with no instrument failure: 3 honoured, 0 violated, 6 with no contract checked; 4 where Syntology's instrument failed) · 7 unverified. Where Syntology's instrument failed: 4 where our contract draft was wrong; failures of Syntology's instrument, not of the code.
By repository: official repository: 10 samples from 1 repository, 8 ran (4 with no instrument failure; 4 where Syntology's instrument failed); community (archive-listed): 10 samples from 1 repository, 5 ran (5 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.
Levels written between 2026-09-22 and 2026-09-25 (UTC): the dates Syntology's graph wrote each sample's verification level, not when the code executed. 3 of the 13 samples that ran have no recorded date.
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
20 samples harvested: 13 ran (of which 0 constructed an object rather than computing a result; 9 with no instrument failure: 3 honoured, 0 violated, 6 with no contract checked; 4 where Syntology's instrument failed) · 7 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.
Licence: 3 of the 20 samples are pointer only, meaning Syntology does not serve that copy's text. For 3 of them, this paper's copy is under a permissive licence and cleared, but the stored copy of the same code that Syntology serves is not; the licence label's tooltip gives Syntology's recorded reason. This page shows no code text for any sample; each one links to its file in the repository.
Harvested from 2 repositories linked to this paper, official or community; the list below is grouped by repository, each group headed by its name and what it is to the paper. “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. A census tag appears when Syntology's census ran implementations with that name on one shared input per bucket: it gives how many were compared, from how many papers, and how many distinct outputs they produced, and names this sample's bucket when this sample was one of them.
census: 14 implementations from 37 papers, 2 distinct outputs on one shared input this sample: bucket 1 of 1, arg 1: rank 2, kind float, dtype float32 src/transformers/activations.py ran · our draft was wrong fingerprinted Apache-2.0 (permissive) ·
0f786c407fb1ee4c · report
fixture: signature x:Tensor/2/float · returned Tensor · level recorded, date not recorded cf9ffa02a42184af · report
fixture: signature text:str/0/str · returned list · level recorded, date not recorded census: 43 implementations from 74 papers, 6 distinct outputs on one shared input src/transformers/modeling_tf_bert.py ran · honoured contract fingerprinted Apache-2.0 (permissive) ·
9bc31d06256e1691 · report
fixture: signature x:EagerTensor/2/float · returned EagerTensor · level recorded 2026-09-22 census: 3 implementations from 7 papers, 1 distinct output on one shared input this sample: bucket 1 of 1, arg 1: rank 2, kind float, dtype float32 src/transformers/activations.py ran · honoured contract fingerprinted Apache-2.0 (permissive) here; the stored copy is pointer only ·
77601724cad03f95 · report
fixture: signature x:Tensor/2/float · returned Tensor · level recorded 2026-09-22 census: 3 implementations from 7 papers, 1 distinct output on one shared input src/transformers/modeling_tf_bert.py ran fingerprinted Apache-2.0 (permissive) ·
83257a2b015273fa · report
fixture: signature x:EagerTensor/2/float · returned EagerTensor · level recorded 2026-09-25 e7fbc7a74a3457c7 · report
fixture: signature vocab_file:str/0/str · returned OrderedDict · level recorded, date not recorded census: 5 implementations from 8 papers, 1 distinct output on one shared input this sample: bucket 1 of 1, arg 1: rank 2, kind float, dtype float32 src/transformers/modeling_bert.py ran · our draft was wrong fingerprinted Apache-2.0 (permissive) here; the stored copy is pointer only ·
060527bb34794b59 · report
fixture: signature x:Tensor/2/float · returned Tensor · level recorded 2026-09-22 census: 14 implementations from 37 papers, 2 distinct outputs on one shared input src/transformers/modeling_tf_bert.py ran · honoured contract fingerprinted Apache-2.0 (permissive) ·
86d13b9f67274f19 · report
fixture: signature x:EagerTensor/2/float · returned EagerTensor · level recorded 2026-09-22 d772ba25f4de22b0 · report
0766f48ad6e01b56 · report
a6f1e11c54df3de5 · report
fixture: signature data:dict/0/?, dataset_name:str/0/str, name:str/0/str, description:str/0/str, tasks:list/1/str, url:str/0/str · returned dict · level recorded 2026-09-24 d87861334682866a · report
fixture: signature package_name:str/0/str · returned bool · level recorded 2026-09-24 6e1f4e02403886e3 · report
fixture: signature repo_path:str/0/str · returned list · level recorded 2026-09-24 e0c0f0759208e777 · report
fixture: signature file_path:str/0/str · returned bool · level recorded 2026-09-24 03c4a8263ac6e7a1 · report
fixture: signature data:dict/0/? · returned bool · level recorded 2026-09-24 33aeaf51efd2501f · report
ef9ba0bf81d816aa · report
02907c137bc813d1 · report
653c7bc811a7f863 · report
8db96d079a008116 · report
Tasks
Results from the paper archive 2025-07-28
No leaderboard rows for this paper in the archive.
Methods
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